Handbook of Health Inequalities Across the Life Course 1800888155, 9781800888159

Citation preview

HANDBOOK OF HEALTH INEQUALITIES ACROSS THE LIFE COURSE

RESEARCH HANDBOOKS IN SOCIOLOGY Series Editor: Hans-Peter Blossfeld, Professor of Sociology, University of Bamberg, Germany The Research Handbooks in Sociology series provides an up-to-date overview on the frontier developments in current sociological research fields. The series takes a theoretical, methodological and comparative perspective to the study of social phenomena. This includes different analytical approaches, competing theoretical views and methodological innovations leading to new insights in relevant sociological research areas. Each Research Handbook in this series provides timely, influential works of lasting significance. These volumes will be edited by one or more outstanding academics with a high international reputation in the respective research field, under the overall guidance of series editor Hans-Peter Blossfeld, Professor of Sociology at the University of Bamberg. The Research Handbooks feature a wide range of original contributions by well-known authors, carefully selected to ensure a thorough coverage of current research. The Research Handbooks will serve as vital reference guides for undergraduate students, doctoral students, postdoctorate students and research practitioners in sociology, aiming to expand current debates, and to discern the likely research agendas of the future. Titles in the series include: Research Handbook on the Sociology of Education Edited by Rolf Becker Research Handbook on the Sociology of the Family Edited by Norbert F. Schneider and Michaela Kreyenfeld Research Handbook on Environmental Sociology Edited by Axel Franzen and Sebastian Mader Research Handbook on Analytical Sociology Edited by Gianluca Manzo Handbook of Sociological Science Contributions to Rigorous Sociology Edited by Klarita Gërxhani, Nan Dirk de Graaf and Werner Raub Research Handbook on the Sociology of Organizations Edited by Mary Godwyn Research Handbook on Digital Sociology Edited by Jan Skopek Research Handbook on Intersectionality Edited by Mary Romero Research Handbook on Public Sociology Edited by Lavinia Bifulco and Vando Borghi Handbook of Health Inequalities Across the Life Course Edited by Rasmus Hoffmann

Handbook of Health Inequalities Across the Life Course Edited by

Rasmus Hoffmann Professor of Sociology, Faculty of Social and Economic Sciences, University of Bamberg, Germany

RESEARCH HANDBOOKS IN SOCIOLOGY

Cheltenham, UK • Northampton, MA, USA

© The Editor and Contributors Severally 2023

All rights reserved. No part of this publication may be reproduced, stored in a retrieval system or transmitted in any form or by any means, electronic, mechanical or photocopying, recording, or otherwise without the prior permission of the publisher. Published by Edward Elgar Publishing Limited The Lypiatts 15 Lansdown Road Cheltenham Glos GL50 2JA UK Edward Elgar Publishing, Inc. William Pratt House 9 Dewey Court Northampton Massachusetts 01060 USA A catalogue record for this book is available from the British Library Library of Congress Control Number: 2023931368 This book is available electronically in the Sociology, Social Policy and Education subject collection http://dx.doi.org/10.4337/9781800888166

ISBN 978 1 80088 815 9 (cased) ISBN 978 1 80088 816 6 (eBook)

EEP BoX

Contents

List of contributorsviii Preface xi 1

Introduction to the Handbook of Health Inequalities Across the Life Course: a societal problem and a source for interdisciplinary research Rasmus Hoffmann

PART I

THEORETICAL APPROACHES TO HEALTH INEQUALITIES ACROSS THE LIFE COURSE

2

Sociology of the life course and its implications for health inequalities Karl Ulrich Mayer

3

Cumulative dis/advantage processes, nutrition transition, and global metabolic disparities: interrogating the cohort–policy linkage Jessica A. Kelley, Abolade Oladimeji and Dale Dannefer

4

Economic theories of health inequality across the life course Titus J. Galama and Hans van Kippersluis

5

Health as a consequence of genetic variation, gene transcription and life course experiences Martin Diewald

PART II

1

15

32 46

59

METHODOLOGICAL ISSUES FOR THE LONGITUDINAL ANALYSIS OF HEALTH INEQUALITIES

6

Methods for studying life course health inequalities Scott M. Lynch and Christina Kamis

75

7

Causal inference based on non-experimental data in health inequality research Michael Gebel

93

8

Predictive machine learning approaches – possibilities and limitations for the future of life course research Hannes Kröger

9

Instrumental variables in studies of health and health inequalities Rasmus Hoffmann and Gabriele Doblhammer

v

112 128

vi  Handbook of health inequalities across the life course PART III MECHANISMS AND EMPIRICAL EVIDENCE FOR HEALTH INEQUALITIES AT STAGES OF THE LIFE COURSE 10

Health inequalities in adolescence and their consequences for (emerging) adulthood Marie Bernard, Kristina Winter, and Irene Moor

146

11

Social inequalities, social capital, and health inequalities in the process of growing up Andreas Klocke and Sven Stadtmüller

160

12

Work and health inequalities Johannes Siegrist

172

13

Family relations and health inequalities: grandparents and grandchildren Valeria Bordone, Giorgio Di Gessa and Karsten Hank

188

14

The effects of retirement on health and mortality by socio-economic group Matthias Giesecke

203

15

Health inequalities in older age: the role of socioeconomic resources and social networks in context Martina Brandt, Nekehia T. Quashie and Alina Schmitz

215

PART IV LONG-TERM PERSPECTIVES ON THE DEVELOPMENT OF HEALTH INEQUALITIES ACROSS LIFE COURSE STAGES 16

Early childhood origins of modern social class health disparities Alberto Palloni, Daniel Ramirez and Sebastian Daza

233

17

The long arm hypothesis: childhood poverty, epigenetic ageing, and late-life health in America, Britain, and Europe Gindo Tampubolon

18

Childhood conditions and health later in life: examples from Sweden Serhiy Dekhtyar and Stefan Fors

275

19

The influence of early health on educational and socioeconomic outcomes Marco Cozzani and Juho Härkönen

292

252

20 Divergence and convergence: how health inequalities evolve as we age Johan Fritzell and Johan Rehnberg

307

21

Environmental inequality and health outcomes over the life course Christian König and Jan Paul Heisig

327

22

Infectious diseases across the life course: an inequalities perspective Nico Dragano

349

Contents  vii PART V

POLICY PERSPECTIVES AND EMPIRICAL EVALUATIONS OF INTERVENTIONS AGAINST HEALTH INEQUALITIES

23

Policy, inequity, and the life course in the US Sarah Petry

368

24

The role of Social Protection Policies in reducing health inequalities Amanda Aronsson, Hande Tugrul, Clare Bambra and Terje Andreas Eikemo

384

Index403

Contributors

Amanda Aronsson, Centre for Global Health Inequalities Research (CHAIN), Department of Sociology and Political Science, Norwegian University of Science and Technology (NTNU), Trondheim, Norway. Clare Bambra, Population Health Sciences Institute, Newcastle University, UK. Marie Bernard, Institute of Medical Sociology (IMS), Medical Faculty, Martin Luther University Halle-Wittenberg, Germany. Valeria Bordone, Department of Sociology, University of Vienna, Austria. Martina Brandt, TU Dortmund University, Germany. Marco Cozzani, Department of Social and Political Science, European University Institute, San Domenico di Fiesole, Italy. Dale Dannefer, Department of Sociology, Case Western Reserve University, Cleveland, USA. Sebastian Daza, Consejo Superior de Investigaciones Científicas (CSIC), Spain. Serhiy Dekhtyar, Aging Research Center, Karolinska Institute & Stockholm University, Stockholm, Sweden. Martin Diewald, Faculty of Sociology, Bielefeld University, Germany. Giorgio Di Gessa, Epidemiology & Public Health, University College London, UK. Gabriele Doblhammer, University of Rostock, Germany. Nico Dragano, Institute of Medical Sociology, Centre of Health and Society, Medical Faculty and University Hospital Düsseldorf, Germany. Terje Andreas Eikemo, Centre for Global Health Inequalities Research (CHAIN), Department of Sociology and Political Science, Norwegian University of Science and Technology (NTNU), Trondheim, Norway. Stefan Fors, Aging Research Center, Karolinska Institutet and Stockholm University; Centre for Epidemiology and Community Medicine, Region Stockholm, Stockholm, Sweden. Johan Fritzell, Aging Research Center (ARC), Karolinska Institute and Stockholm University. Titus J. Galama, University of Southern California, Dornsife College Center for Economic and Social Research, Los Angeles, USA; VU University Amsterdam, The Netherlands; Erasmus School of Economics, Erasmus University Rotterdam, The Netherlands; Tinbergen Institute, The Netherlands. Michael Gebel, University of Bamberg, Germany. viii

Contributors  ix Matthias Giesecke, Leibniz-Institut für Wirtschaftsforschung (RWI) and Institute of Labor Economics (IZA), Germany. Karsten Hank, Institute of Sociology & Social Psychology, University of Cologne, Germany. Juho Härkönen, Department of Social and Political Science, European University Institute, San Domenico di Fiesole, Italy. Jan Paul Heisig, WZB Berlin Social Science Center, Freie Universität Berlin, Germany. Rasmus Hoffmann, University of Bamberg, Germany. Christina Kamis, Center for Demography of Health and Aging, University of Wisconsin-Madison, USA. Jessica A. Kelley, Department of Sociology, Case Western Reserve University, Cleveland, USA. Andreas Klocke, Frankfurt University of Applied Sciences, Germany. Christian König, WZB Berlin Social Science Center. Hannes Kröger, German Institute for Economic Research, Germany. Scott M. Lynch, Department of Sociology, Duke University, USA. Karl Ulrich Mayer, Max Planck Institute for Human Development, Germany; Department of Sociology, Yale University, USA. Irene Moor, Institute of Medical Sociology (IMS), Medical Faculty Martin Luther University Halle-Wittenberg, Germany. Abolade Oladimeji, Department of Sociology, Case Western Reserve University, Cleveland, USA. Alberto Palloni, Center for Demography and Health of Aging, University of WisconsinMadison, USA and Consejo Superior de Investigaciones Científicas (CSIC), Spain. Sarah Petry, Sanford School of Public Policy and International Max Planck Research School for Population, Health and Data Science, Duke University, USA. Nekehia T. Quashie, University of Rhode Island, Kingston, USA. Daniel Ramirez, Consejo Superior de Investigaciones Científicas (CSIC), Spain. Johan Rehnberg, Aging Research Center (ARC), Karolinska Institute and Stockholm University, Stockholm, Sweden. Alina Schmitz, TU Dortmund University, Germany. Johannes Siegrist, Institute of Medical Sociology, Centre for Health and Society, Faculty of Medicine, Heinrich-Heine-University Düsseldorf, Germany. Sven Stadtmüller, Frankfurt University of Applied Sciences, Germany. Gindo Tampubolon, Global Development Institute, University of Manchester, UK.

x  Handbook of health inequalities across the life course Hande Tugrul, Department of Social and Political Science, Bocconi University, Milano, Italy. Hans van Kippersluis, Erasmus University Rotterdam, The Netherlands; Tinbergen Institute, The Netherlands. Kristina Winter, Institute of Medical Sociology (IMS), Medical Faculty Martin Luther University Halle-Wittenberg, Germany.

Preface

Dear reader, This Handbook will give you an overview of research on health inequalities across the life course. It may be useful for newcomers to this fascinating field of research and for experts who look for inspiration and insights into the current state of the art. A total of 44 researchers from eight countries took part in this project and contributed their perspective, experience, and expert knowledge. This variety entails many different topics with a very interesting heterogeneity in datasets and methodological approaches. Together with the differences in disciplinary backgrounds and personal styles of writing, this made the present Handbook a pleasure to assemble and, I hope, to read. First and foremost, I want to thank all authors for their time and energy devoted to this project, for their results and ideas, and their reliable cooperation over a long period. In a time when time pressure becomes a problem for many of us (not least because of Covid-19), it is laudable that so many excellent researchers volunteered to cooperate. The writing of this book included a peer-review process, in which all chapters where reviewed by two anonymous other authors of this volume and myself. This in itself was a stimulating endeavour. Further thanks go to Will Tayler (the Bamberg Translation Co.) for his professional language editing of some chapters of this Handbook and to Daniel Mather, who was our main contact person at Edward Elgar and helped us a lot during the first and most important stages of the project. Finally, I would like to thank Hans-Peter Blossfeld, who invited me to edit this volume and without whom the book would not have been written. And now, enjoy reading … Rasmus Hoffmann

xi

1. Introduction to the Handbook of Health Inequalities Across the Life Course: a societal problem and a source for interdisciplinary research Rasmus Hoffmann

Examining health inequalities across the life course is a specific perspective on the well-known issue of health inequalities. This perspective entails, simultaneously, a theoretical approach, a descriptive approach, an explanatory approach, and a policy approach designed to reduce health inequalities. All four approaches have evolved conjointly alongside the growing availability of longitudinal data and the development of analytical methods and tools that enable the disentangling of complex long-term relationships. These relationships, in this specific context, are the long-term processes at work between social factors in the life course, between aspects of health in the life course, and also between these two realms; this last relationship creates health inequalities. Whereas in 2009 research on ‘health across the life course’ was still being referred to as an ‘emerging topic’ (Mayer 2009), it has since become a flourishing, well established, and yet still advancing approach to the explanation of health inequalities. In this introductory chapter, I will ● review some important features of and developments in research on health inequalities across the life course; ● describe its place in, and in relation to, sociology, as well as some prominent research topics, in brief sections gathered under the concepts of complexity, causality, biology, and policy; ● motivate and describe the structure and content of the book as a whole; and ● summarise each chapter to give an overview and orientation for the reader. Sociological work on the life course as an explicit framework within which individual lives can be studied in interaction with social structures and historical contexts dates back at least to the seminal study of the cohort of the Great Depression by Elder (2018 [1974]). Since then, several related processes have contributed to the development of life course research. First, the improvement of available data through the establishment of longitudinal cohort studies, such as several British cohort studies (some dating back to the 1930s), but also more recent large-scale internationally comparative surveys such as the Health and Retirement Study (HRS), the Survey of Health, Ageing, and Retirement in Europe (SHARE), and the English Longitudinal Study of Ageing (ELSA). These data sources are complemented, for example, by national registers, mostly in Scandinavian countries, and national surveys such as the National Educational Panel Study (NEPS) that follow cohorts over longer periods of time. As survey studies cannot endlessly increase their costs and the volume (as the product of sample size,

1

2  Handbook of health inequalities across the life course follow-up period, and depth of the data), decisive progress has also been made by linking survey data to administrative data and by international collaboration and comparability. Second, the development of statistical methods to analyse such complex data has kept pace with data availability and growing interest in the long-term consequences of risk factors and living conditions, for example in childhood. Some of these methods are reviewed and presented in this Handbook, but the immense progress in statistical models and computational power can only briefly be mentioned here. Third, the theoretical understanding of fundamental principles that shape social inequalities and health inequalities in the life course has improved, for example by developing comprehensive theoretical models such as Cumulative Dis/Advantage (CDA) (Dannefer 2003, 2018; DiPrete & Eirich 2006). Mechanisms that contribute to risk accumulation can be considered in theoretical and empirical opposition to the idea of specific critical or sensitive periods (Kröger et al. 2016). Some have critically argued that the theory of accumulation is one of the rare explanatory theories in life course research for the effect of exposure on later outcomes (Mayer 2009). However, on the one hand, it is hard to say what kind of theoretical reasoning or assumptions exactly fulfil the criteria of a theory, for example, as opposed to a model. On the other hand, there have been a number of theoretical, or at least conceptual, developments in life course research in the past decades, for example the three assumptions of life course research formulated by Huinink (1995), the ‘life course paradigm’, with its five principles of the life course (Elder et al. 2003), or the ‘life course cube’ (Bernardi et al. 2019) (for more details, see Mayer, Chapter 2 in this volume). This progress and consolidation of life course research is also reflected in new journals (e.g. Longitudinal and Life Course Research, Advances in Life Course Research) and publications such as A Life Course Approach to Healthy Ageing (Kuh et al. 2014) or the Handbook of Life Course Health Development (Halfon et al. 2018), which address the aspect of health in the life course (as the present volume does), but without the focus on social inequalities that is fundamental to sociology. Finally, the scientific and public interest in inequalities and the life course can also be seen in large innovative projects, such as TwinLife (Hahn et al. 2016) and specific funding initiatives, such as the funding scheme ‘Dynamics of Inequality Across the Life-course: structures and processes (DIAL)’ launched by Norface and funded by the European Union’s Horizon 2020 research and innovation programme. Sociology as a discipline was not a major player in these developments, but life course research as a whole has always been a very interdisciplinary endeavour. The interest in health across the life course developed more in (social) epidemiology, demography, psychology, and economics, which all contributed substantially to the development of the theoretical and empirical analysis of the life course. Only recently has sociology taken an interest in a rigorous health research agenda, having realised that health is an important dimension of social inequality; at the same time, it belongs largely to the realm of biology and the human body, and as such is necessarily interdisciplinary and cannot be understood and studied with sociological perspectives alone. Perhaps the most important contributions of sociology to this cooperative approach are (1) a deeper understanding of the multiple dimensions of the socioeconomic status of a person not just as determinants of health or risk factors, but as processes in their own right that influence and are influenced by health in several ways, and (2) an explicit acknowledgement of population health as opposed to health on the individual level, which needs a structural sociological approach. The complexity resulting from these interdisciplinary interactions will be briefly outlined in the following section.

Introduction  3

COMPLEXITY The life course perspective as such is well-integrated in sociology, but the study of health inequalities through this specific lens developed more in other fields, such as social epidemiology (‘life course epidemiology’), health economics, and – when it comes to policy interventions – also public health. The life course approach has served as a point of intersection for the synthesis of sociological, biological, and psychological perspectives and enabled the fruitful analysis of turning points, transitions, and trajectories (Bynner 2016). One important feature of the life course approach is its complexity. This complexity is the result of ever more abundant and more detailed data on individual life courses, longer follow-up periods, and the well-justified but ambitious aim of following a multidimensional and multilevel approach to the life course, in which individuals and their characteristics develop in social and historical contexts. Beyond the usual set of social variables derived from a life course interview, there is a large amount of information in the human body itself, in the form of biomarkers and genetic information, as well as potentially macro-level variables from the social and natural environment that, considered altogether, may influence later (health) outcomes. Given this huge and increasing quantity of relevant complex information, one important challenge of future life course research will be to find ways to handle this complexity. The choice between different social variables (e.g. income, education, occupational status, social capital) on the one hand and an increasing number of available health variables (morbidity versus mortality, chronic diseases, infectious diseases, biomarkers etc.) may in some cases be dominated by pragmatic rationales. But this choice is not trivial and is relevant for theoretical and substantial differences in the relationship between specific aspects of socioeconomic status (SES) and health. Life course research encounters the additional problem that measures of the same variable at different ages should be comparable, but it cannot be taken for granted that an identical measure is equally significant at different ages. Advanced modern life course analysis may need more modern computer-based strategies to find patterns in the data (see Kröger, Chapter 8 in this volume, on machine learning approaches), but to create a similar kind of ‘boost’ for theoretical development is more difficult, although this is needed to the same extent. I hope to have assembled in this Handbook positive examples of a weighted use of advanced statistics and advanced theory (see e.g. Palloni et al., Chapter 16 in this volume, with their attempt to model long-term complexity). This Handbook also illustrates the enormous progress that has been achieved in handling life course complexity to produce significant research findings and advance the field. However, the following two topics may serve as examples of questions that have occupied life course epidemiologists for decades and still remain unsolved in principle: first, health across the life course and its relation to the ageing process and the problem of disentangling these two dimensions (Ben-Shlomo et al. 2016; Hoffmann 2011) and, second, the question as to whether health inequalities converge or diverge in old age. This may seem a simple empirical question (which I studied in my doctoral dissertation, see Hoffmann 2008), but whatever you measure, it is hard to analyse the data and come to a final conclusion, because both divergence and convergence can be interpreted as social dynamics of accumulation, biological effects, and policy effects, with the issues of heterogeneity and selective survival only adding to this complexity (see Kratz & Patzina 2020 and Fritzell & Rehnberg, Chapter 20 in this volume, as two recent analyses of this issue).

4  Handbook of health inequalities across the life course

CAUSALITY One important implication of the complexity outlined above is that it becomes both more difficult and more important to address fundamental issues in the definition and empirical study of causal effects. While it has always been a central task of sociology to explain social phenomena, be it individual behaviour or social structures (which means answering why-questions), it has not been a primary concern of sociology to discuss or solve problems related to causal analysis. Instead, other disciplines working on health-related issues (e.g. epidemiology or (health) economics) together with statisticians, have done pioneering work in this area (Hernán et al. 2004; Holland 1986; Rubin 1974, 2005; VanderWeele 2015). But later on, social sciences, demography, and sociology also redefined their approach to causal research (Engelhardt et al. 2009; Gangl 2010; Goldthorpe 2001; Hedström & Ylikoski 2010; Morgan 2013). Useful overviews of available causal methods for non-experimental data and in-depth discussions of specific methods have recently been published (Hoffmann & Doblhammer 2021, also see Chapters 7 and 9 in this volume by Gebel and Hoffmann & Doblhammer). One as yet unanswered causal question in life course research is, for example, how we should interpret the numerous studies that find a correlation between early life factors and later life outcomes (e.g. Pakpahan et al. 2017). Without theoretical and empirical progress in understanding (1) the complexity of factors leading from a point of origin to a distant outcome, and (2) which necessary simplifications of empirical models bias the results and which do not, one of the many pitfalls is the risk of interpreting early life factors as simple determinants of later outcomes. A so-called ‘direct effect’ between a childhood factor and an outcome later in life, i.e. one that cannot be ‘explained away’ by intermediate control variables, may just be another indirect effect running through a different, as yet unknown channel. Thus, research on health (inequalities) across the life course is often confronted with definitions of causality that are blurred by long, indirect, and multifactorial effects and partly unknown chains of risk (Dekhtyar & Fors, Chapter 18 in this volume). In such complex situations it is particularly laudable that researchers still set their sights high and try to establish, for example, when in the life course an omnipresent factor such as education has the largest effect on social inequality (Kratz et al. 2022). A second example of an unsolved causal problem is bidirectional causality, where there are good reasons to assume that X causes Y, but also that Y causes X. A prominent case is the association between socioeconomic status (SES) and health across the life course. This association is fundamental to the present Handbook, and yet it is largely unknown whether the main causal pathway is social causation, health selection, or common background factors that influence both SES and health. There are research findings supporting all three causal models (Hoffmann et al. 2018; Cesarini et al. 2016; Foverskov & Holm 2016) and also plausible arguments for the assumption that the importance of the latter model may increase over time (Mackenbach 2012). The answer to the question of SES and health causality seems to depend on the causal method used, which in turn partly reflects the research tradition of the respective discipline and the preferences and beliefs regarding the requirements for causality. This is understandable, but unsatisfactory from a scientific point of view and thus calls for continued and increased cooperation and more ‘synthesis and reconciliation studies that are based on a variety of different approaches’ (Moffitt 2005, p. 92; Kröger et al. 2015, see also Hoffmann & Doblhammer, Chapter 9 in this volume).

Introduction  5

BIOLOGY The topic of this Handbook is a showcase for a fundamental principle of the life course, namely that it is a lengthy process where individual characteristics interact with social and structural factors, and both may have lagged effects. For the study of health inequalities across the life course, one has to consider that many of the relevant individual factors are biological, or even genetic, and that human bodies develop good or bad health in a social and physical environment, the latter including climate change or biological threats such as COVID-19. Inequalities exist on both the individual and the social side, because genetic and physical endowment and predispositions are unequally distributed, and also because society has many mechanisms that create inequality. The challenge is to understand how these completely different spheres interact so closely that a human body can fall ill or die because of social influences. At some point, social influences must get ‘under our skin’ and become biologically relevant, because a serious disease or death is a biological fact, valid beyond the socially constructed social world. To examine this open and fluid border between society and biology, several concepts, approaches, and methods have been developed, some of which I will outline in the following. Blane et al. (2013) have coined the term ‘social-biological transitions’ and much related work and progress has been done by the interdisciplinary Society for Longitudinal and Life Course Studies (SLLS). Here and elsewhere, biomarkers have been increasingly used in surveys and analyses to complement our understanding of how social and biological factors interact to produce health inequalities (Hoffmann & Kröger 2021). Well established in demography and epidemiology, twin and sibling designs have increasingly been used in research on the development of inequalities across the life course and through generations (e.g. Kröger et al. 2018). The project TwinLife is one famous example of how sociology tries to accurately define its basic concepts of social origin and social inequality, exactly by defining biological origin and biological inequality, too (see Diewald, Chapter 5 in this volume). Genetics is a conflict-prone topic, and one might justifiably suspect that statements according to which genes influence health and SES will be misused by conservative minds in order to justify existing inequalities as natural. But here, too, progress in the biological and social sciences has opened new perspectives: the discovery of epigenetics has abolished the view that genes are fixed, and their influence on SES and health straightforward. Instead, gene-expression can be turned on and off and depends on social influences. In the debate about genetic endowment and educational achievement, it is now plausible to assume that it is not genes directly influencing education and SES, but that SES is a moderator of the effect of genes on education: high SES children can realize their genetic potential, while low SES children cannot. This means nothing less than a reset of the gene–environment discussion and many related debates on social inequality and justice. Many traditional assumptions are now open for future research and discussion, and this new perspective is acknowledged in four chapters of this volume (Diewald (Chapter 5), Palloni et al. (Chapter 16), Tampubolon (Chapter 17), and Dekhtyar & Fors (Chapter 18)). At the end of this section on biology, I want to make two remarks on pandemics and what they have to do with this Handbook. First, the creation of this book fell entirely within the first years of the COVID-19 pandemic. Only one chapter explicitly addresses how COVID-19 and other infectious diseases contribute to health inequalities across the life course (Dragano, Chapter 22 in this volume), but earlier publications have convincingly presented life course aspects of COVID-19 (Settersten et al. 2020). Future research will certainly find numer-

6  Handbook of health inequalities across the life course ous examples of how the present pandemic shapes cohorts and their life courses. Distinct COVID-19 cohorts will be found, depending on the age when this disaster affected the individual’s life course; consequently, numerous social and health consequences will emerge, presumably also for health inequalities much later on (also see Mayer 2022). Second, in terms of biological threats for future cohorts and life courses, there is more to come than COVID-19. Around the world, an obesity pandemic is in full swing, created by global social and economic structures and dynamics of inequality that impose obesogenic food and lifestyles on many people (see Kelley et al., Chapter 3 in this volume), while others are unaffected, and again others are starving to death. Finally, climate change has not been framed as a pandemic, but will soon become the most important global health threat; from the very beginning, the causation and the consequences of climate change have been and continue to be distributed hugely unequally.

POLICY I want to conclude this introduction with a short remark on the policy perspective in life course research on health inequalities. Policy relevance is not the most important focus of sociological research on health, but is addressed instead in public health research. Nevertheless, the important aspect of using science to solve problems should also briefly be reflected in this sociological Handbook (see Petry and Aronsson et al., Chapters 23 and 24 in this volume). As mentioned above, life course research has a genuine relation to policy, because – among other reasons – it was developed in order not only to better describe long-term processes, but also to explain them, i.e. identifying distant causes and effects. Knowing which factor influences health development across the life course, and when this happens, is a necessary precondition for interventions against health inequalities that do not just address superficial proximate causes. The life course framework also represents a toolbox to evaluate long-term effects of life course policies, at least in principle. However, this depends on available data, resources, and the political will to establish and openly communicate whether a policy was effective or not (quite apart from the level of the political enthusiasm to reduce inequality in the first place). What science can do for more evidence-based interventions is to use synergies between experience and capabilities, for example there is too little exchange between empirical policy evaluation research on the one hand and life course research on the development of (health) inequalities on the other (Herd 2016). Still, there are interesting initiatives to establish life course intervention research (Russ et al. 2022). Moreover, it is encouraging to see outstanding life course sociologists openly reflect on the relevance of sociological research for the reduction of inequalities and how to increase this relevance (e.g. DiPrete & Fox-Williams 2021), and that sociology cares about its relationship with social problems and policy (De Graaf & Wiertz 2019).

STRUCTURE AND CONTENT OF THIS BOOK The structure of this Handbook partly reflects the view on the successful ingredients of life course research on health inequalities outlined above, and partly the perspective of the readers

Introduction  7 of this Handbook, and what they are likely to find intuitive and practical to further their understanding of both the basics of and the latest advances in this field. In Part I of the Handbook, there are four chapters on theoretical aspects of the life course and health inequalities, comprising a general, a global, an economic, and a genetic perspective on the topic. Part II includes four chapters that give an overview of general and causal methods and also examine more deeply two specific approaches, discussing their methodological implications. Part III is an overview of empirical findings on health inequalities, namely from specific life course stages, that present stage-specific mechanisms between social factors and health. The life course stages addressed in Part III are adolescence, work, family relations, retirement, and old age. Part IV is devoted to the core of life course analysis, which is the analysis of large parts of the life course (if not the entire life span), thereby addressing the difficult questions and challenges mentioned above and also covering environmental risk factors and infectious diseases. The final part, Part V, addresses the important question of what, on the basis of existing evidence from two examples, can be done to reduce health inequalities over the life course? Chapter Summaries Part I Chapter 2 by Karl Ulrich Mayer gives an overview of traditions, concepts, and theories in the sociology of the life course and describes how this field of research fits with the problem of health inequalities across the life course. To some extent, the principles and mechanisms by which social inequalities are created and expressed in the life course are similar and related to the development of health inequalities. From this perspective, the author gives a comprehensive account of topics that have already been covered by life course research on health inequalities, but also where more could be achieved by better implementation and combination of existing data and by exploiting innovative theories and methods. Chapter 3 by Jessica A. Kelley, Abolade Oladimeji, and Dale Dannefer introduces an unusual, but very important perspective on health inequalities across the life course, namely the differential obesity risks between cohorts and global regions in nutrition transition. The authors show that the global increase of obesity and metabolic morbidity is structured by the timing of national economic development, influences of richer on poorer economies with regard to food markets and consumption styles, and cohorts that incorporate nutrition transitions and metabolic transitions towards obesogenic food in specific phases of the life course. Chapter 4 by Titus J. Galama and Hans van Kippersluis offers a short introduction to Grossman’s seminal health-capital theory and its assumptions, before extending it in order to address several criticisms of this theory and to address health inequalities in the life course in particular. The authors point out that empirical research on health inequalities has made major progress by applying quasi-experimental methods to the question of whether the association between health and socioeconomic status (SES) is causal, while noting that it is equally important to advance theories on the mechanisms between health and SES, on which any causal approach should be based. Chapter 5 by Martin Diewald addresses the important question regarding the extent to which health inequalities are the result of genetic differences. This relatively new and complex perspective covers both the genetic influence of the fixed DNA sequence as well as the epigenetic influence of activated (methylated) parts of the DNA. This activation happens in

8  Handbook of health inequalities across the life course close interaction with the (social) environment and can, for example, be the result of stress. Therefore, an unbiased definition of ‘social origin’ as a starting point for health development through the life course is only possible when genetic variation and ‘genetic origin’ are considered too. Genetic influences are not deterministic, but define a range of developmental opportunities and risks. The chapter cites empirical evidence that socioeconomic status is linked to the level of activation of genes and advocates a mutual enrichment of life course studies and genetic studies of health. Part II Chapter 6 by Scott M. Lynch and Christina Kamis presents the most important methods for studying health inequalities in the life course. They distinguish between methods for analysing the occurrence of events (hazard models and life table methods) and methods for analysing repeated measures (fixed effects, random effects, ‘mixed’ models, and growth and latent class models). For each method, the authors provide a short introduction on how to implement them and discuss some advantages and disadvantages for the study of health inequalities. Chapter 7 by Michael Gebel reviews a number of causal methods that can be used for health inequality research. First, the nature of causal hypotheses, the counterfactual model of causality, and directed acyclic graphs are explained. Then regression adjustment, matching, inverse probability weighting, instrumental variables, the control function approach, and regression discontinuity design are briefly explained as cross-sectional methods, before the difference-in-differences and the before–after estimator are presented as longitudinal estimators. The chapter closes with guidance for causal inference in applied research. Chapter 8 by Hannes Kröger introduces predictive machine learning as a relatively new and promising approach to improving life course research. Statistical life course modelling faces a number of typical problems, including the number and complexity of interacting predictors and outcomes, linked by a complex timing with often unknown latency times, or non-response or missing data. This chapter introduces readers to the specific terminology of the machine learning approach and describes similarities with and differences from more classic analytic methods. It specifies when these methods have advantages and gives examples from the literature. It also offers a warning that machine learning cannot solve the problem of theoretical and empirical complexities, but is complementary to more traditional approaches. Chapter 9 by Rasmus Hoffmann and Gabriele Doblhammer gives an overview of the use of instrumental variables (IV) in the study of health and health inequalities. The authors introduce the idea of IV and review their definition and related assumptions. They discuss what makes a good instrument, the strengths and weaknesses of the approach, in particular for the study of health inequalities, and outline how IV can be combined with other study designs of causality. Finally, they provide a scoping review of publications that use IV to study health inequalities. Part III Chapter 10 by Marie Bernard, Kristina Winter and Irene Moor discusses health inequalities in adolescence and integrates this life stage into the broader perspective on health inequalities in the life course. The authors argue that adolescence is an important period between childhood and young adulthood that matters for health inequalities because, for example, the dominance of the family as a social context may be gradually being replaced by the school and peer groups that have their own and often different health-related role models and incentives. This may entail experiences of autonomy and independence, including access to new health threats such

Introduction  9 as drugs and dangerous behaviour, which may or may not be carried on into adulthood. Thus, although disease prevalence may still be low in adolescence, it matters for the development of health and health inequalities across the life course. Chapter 11 by Andreas Klocke and Sven Stadtmüller studies the effects of families’ socioeconomic position (SEP) and social capital on health in adolescence (age 11–16). The authors use data from the German GUS study (Health Behavior and Injuries in School Age) and also test interactions between SEP and social capital, as well as the interactions between these two determinants of health and age. The findings suggest that social capital can compensate for low SEP and has a growing effect on health between the ages of 11 and 16. Chapter 12 by Johannes Siegrist deals with labour market participation as an important source of social inequalities and health inequalities. By first investigating the levels of social inequality in access to (and quality of) work and employment, and then the health consequences of work and employment, the author establishes the decisive role of work in explaining health inequities, both as mediator and as moderator. Summarized empirical findings and briefly discussed methodological issues are followed by the policy implications of this large body of literature. Finally, we are reminded that the focus on advanced research findings from economically developed societies ignores probably the worst and most unhealthy working conditions (outsourced) in other parts of the world. Chapter 13 by Valeria Bordone, Giorgio Di Gessa and Karsten Hank establishes the link between intergenerational family relations and health inequalities. The phases of being a child, becoming a parent and grandparent, and the resulting family relations structure the life course in an important way. Especially the health effects of grandparenthood and grandparenting, which are not uniformly positive, become increasingly important for health inequalities, because the likelihood of becoming a grandparent, the age at which this occurs, the number of children, and the kind of involvement in grandparenting that is undertaken are not equally distributed across socioeconomic groups. This also holds for grandchildren, for whom the authors also discuss theories and empirical findings of positive and negative health effects of intergenerational relations and transfers. Chapter 14 by Matthias Giesecke reviews the existing evidence on the effects of retirement on health and mortality by socioeconomic group. The differentiation by socioeconomic group is important, because the health effect of such an important event in the life course as the transition to retirement is not homogeneous. The effects may vary, for example, depending on whether work is a threat for health or whether it is a way to maintain social integration and physical and mental activity, and depending on whether the transition to retirement is voluntary or not. The author provides a good overview of findings for different health outcomes and various methods applied in empirical studies in order to address endogeneity issues. Finally, he presents different findings from the literature and links them to specific subgroups and mechanisms. Chapter 15 by Martina Brandt, Nekehia T. Quashie and Alina Schmitz deals with health inequalities in older age and demonstrates that this is not just another age range for studying the same problem, but that specific theoretical and methodological questions are still waiting to be answered. For example, it is still unclear whether health inequalities decrease, increase, or remain stable in old age, and how much of any measured decrease is due to mortality selection. For health inequalities in general, but also for the assumed process of accumulation, the relative importance of social causation and health selection is still unclear. In this chapter, the

10  Handbook of health inequalities across the life course authors put a special focus on social networks as a determinant of health in old age and how they interact with family relationships, gender, and socioeconomic status. Part IV Chapter 16 by Alberto Palloni, Daniel Ramirez and Sebastian Daza provides an interesting theoretical and empirical approach to the question of how early childhood origins contribute to growing social class health disparities over time. The self-replication of health inequalities across successive generations may depend on historical changes of epidemiological regimes and their interaction with class allocation in modern societies. Substantial increases in survival chances have increased the relative importance of long-term effects over the life course, and individual preferences, behaviour, and choices may become more important in modern societies. This chapter also addresses the role of epigenetics and calculates examples for the contribution of obesity and smoking to health inequalities with data from the Health and Retirement Study and the 1958 British Cohort Study. Chapter 17 by Gindo Tampubolon tests the ‘long arm of childhood’ hypothesis by using an international collection of comparable datasets from 28 countries (HRS, ELSA, SHARE). This test is embedded in an important dimension of life course theory, namely the historical setting of life courses and its effects on individual and cohort experiences and later life outcomes. The effect of childhood on old age health outcomes is studied with three health outcomes (depression, episodic memory, probable sarcopenia) and epigenetic ageing, measured by methylation rates. Overall, the long arm of childhood hypothesis is confirmed, and the results are discussed in the light of recent literature and possible future research. Chapter 18 by Serhiy Dekhtyar and Stefan Fors gives an account of the life course research, which is trying to explore the effect of childhood conditions on health in later life. In their comprehensive theoretical and methodological discussion of the various questions and issues involved in a seemingly simple research question, the authors focus on the two health outcomes: cognition and mortality. The chapter gives a very helpful overview of the questions and findings related to genetics as a confounder of associations between childhood and later health. Finally, they present empirical evidence from the literature on the two health outcomes and two main life course models (sensitive period and chain of risk model), based on the generally very good data from Sweden. They conclude with reflections and recommendations for future research in this area. Chapter 19 by Marco Cozzani and Juho Härkönen reviews the literature and research findings on the influence of early health on educational and socioeconomic outcomes. In doing so, they establish that, on balance, the majority of perspectives and research in social epidemiology focus on the effect of social determinants on health, rather than the effect of health on social outcomes. Starting with Barker’s famous hypothesis that perinatal health conditions influence health later in life, they also discuss health in other early periods of the life course such as childhood and adolescence, and how these might be related by the life course models of critical/sensitive period or chain of risk. Finally, they identify important areas for future research. Chapter 20 by Johan Fritzell and Johan Rehnberg addresses a long-standing question that is central to this volume, namely whether health inequalities converge or diverge with increasing age. They introduce the two fundamental and opposing hypotheses of age – as a leveller and as an accumulator – and the respective arguments and perspectives. Subsequently, they relate their central question to another important ongoing discussion, namely whether health inequal-

Introduction  11 ity is better described by relative or absolute measures of inequality. This chapter further provides a short literature review of the measures used and main results with regard to divergence or convergence, and then offers the authors’ own analyses of Swedish data comparing absolute and relative measures of health inequality across age groups. Chapter 21 by Christian König and Jan Paul Heisig explores a dimension of inequality that is rarely covered in standard social science research, namely environmental inequality and its effect on health inequalities over the life course. After explaining the life course relevance of environmental inequalities and addressing the challenges and approaches to causal identification in this area of research, they provide an extensive review of literature and research findings. This review concentrates on air pollution as an environmental factor, but also considers research findings on green space, extreme heat, and noise in the environment. Chapter 22 by Nico Dragano explores how infectious diseases contribute to health inequalities across the life course. The fact that infectious diseases also follow the typical socio-economic gradient is linked to common concepts and models of life course epidemiology, such as critical/sensitive period and accumulation. Many infectious diseases have a specific age pattern, for example of exposure and susceptibility. The author shows that there are also life-course-related reasons for the unequal social distribution of infectious diseases. There is good reason to do more longitudinal research on this particular group of diseases; they continue to be very significant for public health, especially in low-income countries, and have only grown in importance globally through COVID-19. Part V Chapter 23 by Sarah Petry provides an interesting account of how US welfare state policies developed historically, and how they influence health and health disparities at different stages of the life course. Examples are given of empirical evaluation studies that study specific policy effects and pathways. Policy-relevant empirical findings with regard to health inequalities are still rare, partly because of a separation of the field into research on social disparities in health and research on the impact of policies on health, resulting in too few studies on the impact of policies on health inequalities. Research could and should also determine whether, when (at which age), and how policies affect the development of health inequalities in the life course. Chapter 24 by Amanda Aronsson, Hande Tugrul, Clare Bambra and Terje Andreas Eikemo focuses on the role of Social Protection Policies in reducing health inequalities. After showing the principal pathways by which such policies can reduce health inequalities in the life course, the authors choose the example of parental leave policies to examine in more detail. Evaluation studies show that paid parental leave can improve children’s health and well-being and reduces health inequalities among them, which in turn positively influences their education and other social parameters later in life. This illustrates that social policies have the potential to reduce the inter- and intragenerational transmission of social and health inequalities.

REFERENCES Ben-Shlomo, Y., R. Cooper and D. Kuhn (2016), ‘Life course epidemiology: The last two decades and future directions, and its relevance for research on ageing’, International Journal of Epidemiology, 45(4), 973–987. Bernardi, L., J. Huinink and R. A. J. Settersten (2019), ‘The life course cube: A tool for studying lives’, Advances in Life Course Research, 41(3).

12  Handbook of health inequalities across the life course Blane, D., M. Kelly-Irving, A. d’Errico, M. Bartley and S. Montgomery (2013), ‘Social-biological transitions: How does the social become biological?’ Longitudinal and Life Course Studies, 4 (2), 136–146. Bynner, J. (2016), ‘Institutionalization of life course studies’, in Shanahan, Michael J. et al. (Eds), Handbook of the Life Course, Volume 2. Cham: Springer, 27–58. Cesarini, D., E. Lindqvist, R. Östling and B. Wallace (2016), ‘Wealth, health, and child development: Evidence from administrative data on Swedish lottery players’, The Quarterly Journal of Economics, 131(2), 687–738. Dannefer, D. (2003), ‘Cumulative advantage/disadvantage and the life course: Cross-fertilizing age and social science theory’, The Journals of Gerontology Series B, Psychological Sciences and Social Sciences, 58(6), 327–337. Dannefer, D. (2018), ‘Systemic and reflexive: Foundations of cumulative dis/advantage and life-course processes’, The Journals of Gerontology: Series B, 75(6), 1249–1263. De Graaf, N. D. and D. Wiertz (2019), Societal Problems as Public Bads. London, UK: Routledge. DiPrete, T. A. and G. M. Eirich (2006), ‘Cumulative advantage as a mechanism for inequality: A review of theoretical and empirical developments’, Annual Review of Sociology, 32, 271–297. DiPrete, T. A. and N. B. Fox-Williams (2021), ‘The relevance of inequality research in sociology for inequality reduction’, Sociological Research for a Dynamic World, 7, 1–30. Elder, G. H. (2018 [1974]), Children of the Great Depression: Social Change in Life Experience. London, UK: Routledge. Elder, G. H. Jr., M. K. Johnson and R. Crosnoe (2003), ‘The emergence and development of life course theory’, in Mortimer, J. T. and Shanahan, M. J. (Eds), Handbook of the Life Course. New York: Kluwer/Plenum, 3–19. Engelhardt H., H.-P. Kohler and A. Fürnkranz-Prskawetz (Eds) (2009), Causal Analysis in Population Studies. Cham: Springer. Foverskov, E. and A. Holm (2016), ‘Socioeconomic inequality in health in the British household panel: Tests of the social causation, health selection and the indirect selection hypothesis using dynamic fixed effects panel models’, Social Science & Medicine, 150, 172–183. Gangl, M. (2010), ‘Causal inference in sociological research’, Annual Review of Sociology, 36, 21–47. Goldthorpe, J. H. (2001), ‘Causation, statistics, and sociology’, European Sociological Review, 17(1), 1–20. Hahn, E., J. Gottschling, W. Bleidorn, C. Kandler, M. Spengler, A. E. Kornadt, W. Schulz, R. Schunck, T. Baier, K. Krell, V. Lang, F. Lenau, A. L. Peters, M. Diewald, R. Riemann and M. F. Spinath (2016), ‘What drives the development of social inequality over the life course? The German TwinLife Study’, Twin Research and Human Genetics, 19(6), 659–672. Halfon, N., B. C. Forrest, M. R. Lerner, M. E. Faustman (Eds) (2018), Handbook of Life Course Health Development. Cham: Springer. Hedström, P. and P. Ylikoski (2010), ‘Causal mechanisms in the social sciences’, Annual Review of Sociology, 36, 49–67. Herd, P. (2016), ‘The influence of social welfare policies on health disparities across the life course’, in Shanahan, Michael J. et al. (Eds), Handbook of the Life Course, Vol. 2. Cham: Springer, 661–676. Hernán, M. A., S. Hernández-Díaz, and J. M. Robins (2004), ‘A structural approach to selection bias’, Epidemiology, 15(5), 615–625. Hoffmann, R. (2008), Socioeconomic Differences in Old Age Mortality. Dordrecht: Springer. Hoffmann, R. (2011), ‘Illness, not age, is the leveler of social mortality differences in old age’, Journal of Gerontology: Social Sciences, 66(5), 374–379. Hoffmann, R. and G. Doblhammer (2021), ‘Approaches and methods for causal analysis of panel data in the area of morbidity and mortality’, in Comparative Population Studies, 46, Special Issue: Identification of causal mechanisms in demographic research: The contribution of panel data, J. Brüderl and J. Huinink (Eds), April 2021. Hoffmann, R. and H. Kröger (2021), ‘Comparing the educational gradients in three cardiovascular disease-specific health measures’, Longitudinal and Life Course Studies, 12(4), 591–607. Hoffmann, R., H. Kröger and E. Pakpahan (2018), ‘Pathways between socioeconomic status and health: Does health selection or social causation dominate in Europe?’, Advances in Life Course Research, 36, 23–36. Holland, P. (1986), ‘Statistics and causal inference’, J Am Stats Assoc, 81, 945–960.

Introduction  13 Huinink, J. (1995), Warum noch Familie? Zur Attraktivität von Partnerschaft und Elternschaft in unserer Gesellschaft. Frankfurt am M: Campus Verlag. Kratz, F. and A. Patzina (2020), ‘Endogenous selection bias and cumulative inequality over the life course: Evidence from educational inequality in subjective well-being’, European Sociological Review, 36(3), 333–350. Kratz, F., B. Pettinger and M. Grätz (2022), ‘At which age is education the great equalizer? A causal mediation analysis of the (in-)direct effects of social origin over the life course’, European Sociological Review, online first. Kröger, H., J. Fritzell and R. Hoffmann (2016), ‘The association of levels of and decline in grip strength in old age with trajectories of life course occupational position’, PlosOne, 11(5). Kröger, H., R. Hoffmann, L. Tarkiainen and P. Martikainen (2018), ‘Comparing observed and unobserved components of childhood: Evidence from Finnish register data on midlife mortality from siblings and their parents’, Demography, 55(1), 295–318. Kröger, H., E. Pakpahan and R. Hoffmann (2015), ‘What causes health inequality? A systematic review on the relative importance of social causation and health selection’, European Journal of Public Health, 25(6), 951–960. Kuh, D., R. Cooper, R. Hardy, et al. (Eds) (2014), A Life Course Approach to Healthy Ageing. Oxford: Oxford University Press. Mackenbach, J. P. (2012), ‘The persistence of health inequalities in modern welfare states: The explanation of a paradox’, Social Science & Medicine, 75(4), 761–769. Mayer, K. U. (2009), ‘New directions in life course research’, Annual Review of Sociology, 35, 413–433. Mayer, K. U. (2022), ‘Aspects of a sociology of the pandemic: Inequalities and the life course’, Vienna Yearbook of Population Research, 20. Moffitt, R. (2005), ‘Remarks on the analysis of causal relationships in population research’, Demography, 42(1), 91–108. Morgan, S. L. (2013), Handbook of Causal Analysis for Social Research. Cham: Springer. Pakpahan, E., R. Hoffmann and H. Kröger (2017), ‘The long arm of childhood circumstances on health in old age: Evidence from SHARELIFE’, Advances in Life Course Research, 31, 1–10. Rubin, D. B. (1974), ‘Estimating causal effects of treatments in randomized and non-randomized studies’, J Educational Psychology, 66, 688–701. Rubin, D. B. (2005), ‘Causal inference using potential outcomes’, Journal of the American Statistical Association, 100, 322–331. Russ, S. A., E. Hotez, M. Berghaus, C. Hoover, S. Verbiest, E. L. Schor, and N. Halfon (2022), ‘Building a life course intervention research framework’, Pediatrics, 149 (Supplement 5). Settersten, R. A., L. Bernardi and J. Härkönen (2020), ‘Understanding the effects of Covid-19 through a life course lens’, Advances in Life Course Research, online first. VanderWeele, T. (2015), Explanation in Causal Inference: Methods for Mediation and Interaction. Oxford: Oxford University Press.

PART I THEORETICAL APPROACHES TO HEALTH INEQUALITIES ACROSS THE LIFE COURSE

2. Sociology of the life course and its implications for health inequalities Karl Ulrich Mayer

INTRODUCTION The purpose of this chapter is to present and discuss traditions, concepts, and theorems of the sociology of the life course as a potential toolbox for research on health inequalities across the life course. First, I will look at the historical origins of and developments in the sociology of the life course. Secondly, I will review an ensemble of analytical concepts and ideas, which have at various stages informed life course research. Thirdly, I will discuss the few more rigorous theories of the life cycle. Fourthly, I will report on the interrelations between the study of social inequalities and social stratification and the life course as a specific avenue to analyze health inequalities across the lifespan. In conclusion, I will comment on the chartered and unchartered intersections between the fields of life course studies and health inequalities.

A SHORT HISTORY OF THE SOCIOLOGY OF THE LIFE COURSE What now has emerged as something like a distinctive subdiscipline of a sociology of the life course has many and quite varied precursors (Mayer 2005; Alwin 2016). In the beginnings, which might be dated in the period before and after the Second World War, different facets of human development – especially psychological and sociological ones – were still more or less meshed together. Examples are the studies on adolescence by Charlotte Bühler (1928) and others and the classic of a collective migration biography of the “Polish Peasant” by Thomas and Znaniecki (1918). An important theoretical contribution was Karl Mannheim’s essay on “generation”. In this essay, Mannheim (1928) develops two fundamental ideas. One idea is the universal fact that the finite life courses of individuals produce constant social and cultural change via the succession of birth cohorts (“social metabolism”) because new cohorts have each to be socialized anew and also have to embrace existing culture, which they might do selectively. At the same time, the “leaving” generations not only represent the potential for continuity but also the potential for innovation. The other idea is the one of “generations” – derived from cultural styles in the arts. Mannheim constructs the concept of “generation” in direct analogy and contrast to Marx’s concept of social class. Generation as a location (“Generation an sich”) is a mere statistical collectivity, defined by year of birth. Generation as actuality (“Generation für sich”) exposes self-consciously specific values and standards of behavior, while generation as unit (“Generation an und für sich”) does not only form its own identity but also is organized at least in its active parts. In the 1940s and 1950s the more psychological traditions of human development (Erikson 1980; Clausen 1986) focusing on internal personal dynamics became more clearly distinguished from the sociological concept of age roles, groups, and age differentiation (Linton 15

16  Handbook of health inequalities across the life course 1939; Parsons 1942; Eisenstadt 1964) as categories of social structure. Instead of a sociology of the life course, various sociologies of distinct life phases, especially a sociology of youth and a sociology of old age emerged. Nonetheless the close link between psychological, sociological and historical perspectives persisted, most clearly in the early and until now continuing work of the “father” of life course sociology, Glen Elder (1974, 2001). During the 1960s and the 1970s the idea of age differentiation and life phases branched out into various concepts and related areas of research. Matilda Riley’s concept of age stratification now focused on the interrelations between age groups and stressed not only functional and cultural specificity, but also inequalities in resources and power (Riley et al. 1972). The subjective experience was highlighted in biographical research, which combined both a specific qualitative method and an epistemic view of the significance of cultural narratives (Bertaux 1981; Kohli 1981). Mannheim’s lead was followed not only in ambitious reconstructions of historical generations (Bude 1995; Assmann 2007), but also in continuous surveys and repeated journalistic proclamations of ever “new” generations (X, Y, Z, millennials). Another off-shoot of Mannheim’s work was the proliferation of the narrower and more precise concept of “cohorts”, well-embedded both in demographic theory (Ryder 1965, 1980) and method (Mason and Fienberg 1985; Yang and Land 2013). In the 1980s, the idea of the life course as a major element of social structure and as an institution gained more and more traction (Mayer 1986). But here we still find different nuances: on the one hand the old tradition of age differentiation and the life course as a separate element of social structure postulating a “tripartite structure” of the life course before, during and after gainful employment (Kohli 1985), and on the other hand the notion that other dimensions of social structure like education, work, and welfare roles are institutionalized as longitudinal sequences like educational pathways, occupational careers or age-graded welfare entitlements (Mayer and Müller 1986). In contrast to such “structural” views of the life course more widely influential sociological currents were also applied to the life course: the life course as a domain of rational action (Huinink 1995) and the life course as mechanism and product of “human agency” (Hitlin and Kwon 2016). Finally, the life course was seen as a socio-cultural phenomenon. Such aspects were already prominent in the study of age norms or cultural concepts of life phases like old age (Ehmer 2019), but the overall life course as a cultural construct also received attention (Meyer 1986). What we see then as a more current outcome of this divergence and greater precision of theoretical traditions could be seen as the emergence of a “differential” sociology of the life course. One major thrust in this regard is the idea that the modern individual life course is not only a fairly recent historical phenomenon, dating to the Protestant notion of the self-responsible individual and related early autobiographies, but that one could clearly distinguish between different historical forms of life course “regimes” like the pre-industrial, agricultural, the industrial and the post-industrial life course regime (Myles 1993; Mayer 2004). Likewise more and more work was invested in developing and searching for national variants of life course regimes tied to fundamental institutional differences (DiPrete 2002; Mayer 2005). These attempts were inspired by the work on varieties of welfare states (Esping-Andersen 1990, 1999) and varieties of capitalism (Hall and Soskice 2001).

Sociology of the life course and its implications for health inequalities  17

CONCEPTS OF THE LIFE COURSE Analytically we can distinguish at least four formal visions of the life course: (a) (b) (c) (d) (a)

The life course as a series of life phases or age roles; The life course as a series of events and transitions; The life course as a sequence of states; The life course as trajectories. The Life Course as a Series of (Discrete) Life Phases or Age Roles

Initially closely tied to the developmental idea of functional organismic and psychological maturation and decline, the earliest sociological conception of the life course built on the notion of a sequence of discrete life phases: infancy and childhood, adulthood, and old age. Tied to these life phases were the social constructs of often gendered age roles. While at the outset this appeared to be little more than a “before and not yet mature and responsible” and “after and declining” an ever more differentiated sequence of life phases and age roles was observed or postulated: childhood as a special category, adolescence as transition to full adulthood, young adults, midlife, and – most recently – the distinction between young old and old old (Baltes and Mayer 1999). (b)

The Life Course as a Series of Events and Transitions

The organismic, but even more so the institutional nature of the life course allows us to construct and measure it as a series of singular or repeatable events: birth and death, entering school, age at menarche, leaving school, entering training and employment, job shifts, firm shift and occupational changes, marriage and divorce, birth of children, retirement. Such events provide easy demographic measures and functional forms of the corresponding transitions for respective populations, for example median ages of the event, variances of ages at transition, duration times till the event, proportions of a population experiencing or never experiencing an event. Events imply the processes of transitions and allow for turning points. Besides the standard demographic analysis of survival curves and life expectancy, the focus of events became initially most fruitful in the research tradition of transitions to adulthood variously defined as a series or sequence of leaving school, entering the labor market, marrying, and having a first child (Modell et al. 1976; Buchmann and Kriesi 2011). Life course studies in the1970s and 1980s received a huge push through the combination of the new availability of microanalytic, individual level longitudinal data and the statistical method of survival analysis (aka hazard analysis or event history analysis) (Blossfeld et al. 1989; Mayer and Tuma 1990) where applications moved beyond demography to the study of educational transitions (Mare 1980), employment shifts (Carroll and Mayer 1986), residential moves and migration (Wagner 1985) and retirement (Fasang 2012). More rarely were other kind of events measured or analyzed, like party voting in elections, or illnesses and accidents.

18  Handbook of health inequalities across the life course (c)

The Life Course as a Sequence of States

Once one moves beyond few age categories and a few selected events life courses can be mapped as very detailed sequences of social positions or “states”, each of which can be dated by beginning and ending times. Such sequences occur “naturally” in the labor market as states of employment, unemployment, and non-employment, jobs within and between firms and occupations. Similarly, institutionally predefined states can be mapped for educational trajectories (school classes and educational levels) and socio-demographic biographies (single, married, parents, divorced). The measurement of “states” can then be extended, for example in regard to residence and location, car ownership, membership in parties and voluntary associations, etc. The first step in treating “life courses as sequences” was the systematic collection of empirical data via life calendars or state sequences in as many life dimensions as possible. Pioneered by statistical censuses (Tegtmeyer 1976), it was then developed for large-scale retrospective and prospective surveys (Mayer 2015a; Leopoldina 2016). States always imply events and durations and therefore such data was initially predominantly analyzed via methods of event history analysis. Further steps in the representation of sequences were taken via cumulative survival graphs and illustrative mappings of a limited number of individual sequences. But the recent development of sequence analysis adapted from DNA-genetics has now led to a veritable boom of cross-cohort and cross-national sequence analysis (Abbott and Tsai 2000; Piccareta 2017; Piccareta and Studer 2019). Basic to such analyses is the systematic ordering and clustering of groups of sequences for large populations. The conjoint analysis of socio-demographic and socio-economic sequences opens up an entirely new kind of reconstruction of social stratification (Aisenbrey and Fasang 2017). (d)

The Life Course as Trajectories

Metaphorically, a view of life courses as continuous lifelines of growth, improvement, accumulation, and decline seems particularly adequate and natural, but the lack of data on such long time trajectories proved to be also an obstacle to concept formation, theory building and research. Exceptions were the study of life income and – in psychology – of cognitive development (Lindenberger 2014). It is again – as in the case of events and sequences – the innovations in statistical methodology which triggered developments in this area. Recently many studies applying latent growth analysis have been published, besides income mostly relying on measures of occupational status. Questions which then could be raised and answered were those regarding the shape of the status curves, and the differences between women and men and between birth cohorts (Manzoni et al. 2014; Härkönen et al. 2016; Stawarz 2018) as well as questions about the variance of status across the life span (Lersch et al. 2020). Also sociologists could entertain notions like lifetime status or lifetime “stock” analogous to the economists’ lifetime income and wealth. Furthermore, the continuous measurement of trajectories allows us to capture degrees of volatility (Fasang and Mayer 2020).

Sociology of the life course and its implications for health inequalities  19

PARADIGMS AND THEORIES OF THE LIFE COURSE By far the most influential program for a sociology of the life course was for decades Glen Elder’s “life course paradigm” (Elder et al. 2003: 11–13). It outlined five principles: (1) Lifetime development: human development and aging are to be considered as lifelong processes. (2) Agency: individuals construct their live course through the choices they make and the actions they take within the opportunities and constraints of history and social circumstance. (3) Time and place: the life course of individuals is embedded and shaped by the historical times and places they experience over their lifetime. (4) Timing: the developmental antecedents and consequences of life course transitions vary according to the timing in a person’s life. (5) Linked lives: lives are lived interdependently within families and social networks, leading to – among other things – social “convoys”. A further step in the refinement of such a paradigm was made by Huinink (1995: 154–155) who formulated three assumptions of a life course perspective: (1) The life course of an individual is a self-referential process. The individual acts or behaves on the basis of accumulated experiences and resources. The life course is an “endogenous causal mechanism”. (2) The life course is a multi-dimensional process in different consecutive and parallel life domains. (3) The life course is part of a multi-level process (social networks, social institutions, regional contexts). A major breakthrough in this way of conceptualizing the life course was achieved recently by the seminal paper by Bernardi, Huinink and Settersten (2019) on the “life course cube”. Bernardi and co-authors differentiate three axes. First, individual life courses are shaped by and acted out on three levels: a crucial middle personal action level, a lower organismic-psychological level and an upper collective socio-institutional and cultural level. When studying life course all three levels have to be taken into account and all three levels imply different kinds of causal mechanisms (endogenous and cross-level). Second, on each level life courses take place in different domains: different domains of psycho-physical functioning like cognition and motivation; different domains of action like work, family, and leisure; and different domains of institutional and cultural spheres. Third, life courses take place in different time dimensions, like developmental time, aging, institutional clocks, and historical time. The life course cube must be seen as a major analytical achievement and is likely to supersede and replace Elder’s five principles as major guidelines in the field. However, it is clear and probably advantageous that the life course cube is heavily individual and (rational) action centered. Both historical context and changes of the life course as well as cross-national, institutional differences are underplayed (Mayer 2004, 2005). Also, like Elder’s life course paradigm the life course cube is proto-theoretical. It turns our attention to different aspects of the life course rather than postulating, explaining, and predicting outcomes. For instance, it alerts us to changes across the life course and to causa-

20  Handbook of health inequalities across the life course tion within the life course. But one might as well search for stability across the life course or a high degree of independence and specify under which conditions various degrees of change or independence occur. Indeed, very few explicit theories actually exist in this area, and they tend to have been developed in fields other than sociology. Four such examples are the life cycle theory of reproduction and longevity; the theory of selective optimization with compensation (SOC) (Baltes and Carstensen 2003); the theory of primary and secondary control (Heckhausen 2003; Heckhausen and Buchmann 2019); and the life cycle theory of savings and consumption (Modigliani 1966). The biological life history theory views the life course as a trade-off between reproduction and longevity, i.e., between investment in one’s own growth and development (somatic effort) and investment in reproduction, comprising the functional system of mating and parenting. Resources and energy have to be invested among the processes of maintenance, growth, and reproduction. Trade-offs are inevitable, because investing in one process may be detrimental to the other ones (Keller 2001; Stulp and Sear 2019). The applicability of the theory to the human life course is unproven and questionable. While the higher rate of motorcycle deaths of young men might be taken as an indication in this direction, both marriage and parenthood seem to favor the life expectation of men. However, the more general idea of a trade-off might be very useful. Bernardi et al. (2019) remind us of “Gossen’s Law”, according to which investments in different life domains is crucial for life course action strategies. But the essential point here is that competing claims between activities, investments, and life domains in the life course context can be managed and resolved by investing differential amounts of the lifetime to various pursuits and, last but not least, by sequencing investments. Recently, there has been a lot of discussion about the “crowding” and “overburdening” of tasks in early adulthood: education and training, family formation and career entry and development. And it has been repeatedly suggested that the lengthening of the life span/increased life expectation could make it possible to order and space activities in a less stressful manner (Riley et al. 1994). But “sequencing” also has side-effects. Thus, the massive postponement of marriage and first births empirically leads to a lower number of children born and to an increase in childless adults. The psychological theory of selective optimization with compensation views human development as a process of investments in reachable goal attainment (Baltes and Carstensen 2003). Successful development is defined as the maximization of gains and the minimization of losses. The background of this theory relates to the interaction of evolutionary and cultural impacts on the aging process. It is usually assumed that evolutionary selection benefits decrease with age and that therefore the need for cultural supports increases with age, and finally that the efficacy of cultural benefits decrease with age. Gains tend to increase with age up to a certain point and then losses accumulate. If development strives after goals of functioning, then under conditions of limited capacity selection and optimization, the concentration on certain goals with higher efficacy and higher levels of functioning becomes mandatory. With increasing losses compensation can be employed. Good examples are glasses for better seeing and hearing aids that compensate for hearing loss. A somewhat analogous psychological theory sees life-span development as the outcome of two processes of active control efforts. Coping with obstacles to goal attainment can be achieved by two means: “Assimilation refers to active and intentional attempts to change the environment to the self, whereas accommodation addresses nonintentional changes in

Sociology of the life course and its implications for health inequalities  21 variations of goals … or cognitions about goal striving … that reflect adjustments to objective goal blocking” (Heckhausen 2003: 385, original emphasis). The aging process is then viewed as a gradual succession of assimilation and accommodation. Correspondingly, assimilation is primary control of the environment by the individual while accommodation is secondary control of the inner world of the individual and serves to optimize motivational and emotional resources. “During childhood, adolescence, and young adulthood, primary control potential increases substantially, reaches a maximum during midlife, and declines with the loss of social roles and physical fitness with old age” (Heckhausen 2003: 386). Behrman (2003) has noted the close affinities between the above psychological theories and the economic theories of human capital. The more private resources are devoted to human-resource investments the greater the marginal gains from those investments. Early investments have to be weighed against their costs and against their expected returns. Since the duration of expected benefits of human capital declines with age, one should expect that investments in human capital should concentrate in earlier life phases and decline rapidly with age. This is exactly what we observe, for instance, in regard to the rate of further training (Becker and Schömann 1996) and what makes investments in occupational skills more and more unlikely with advancing working life. Another explicit economic theory of the life course is Modigliani’s life cycle theory of savings, wealth, and consumption (Modigliani 1966). According to this theory, it is rational for economic actors to save and build up wealth in the early stages in life to be able to spread consumption opportunities more evenly across the life course. This should apply especially if in later life old age pensions are expected to be lower than earnings. It follows that older persons should decrease and eventually give up saving altogether and disinvest in accumulated wealth. Not least for this theory did Modigliani receive the Nobel Prize in economics in the year 1985. Remarkably, the theory turns out to be false, since very often older people continue to save and accumulate wealth, presumably with the motive of providing an inheritance for their children and grandchildren. Obviously, the empirical validity of the theory also depends on how old age pensions are organized after retirement. In places like the Scandinavian countries where old age pensions are paid out of general taxes one should expect less saving than in countries where old age income heavily depends on one’s own saving and wealth. Breen and Goldthorpe (1997) have developed a prominent rational choice theory on educational decisions and transitions. Educational decisions by parents for their children and those of young adults are based on goals (which questionably Breen and Goldthorpe assume to be “status maintenance” rather than upward social mobility), perceived costs of educational tracks, expected educational success and expected (wage) returns. It is noteworthy that these authors then claim to be able to explain educational differentials between social classes without having to resort to class differential norms and values. Another application of human capital theory to the life course can be found in Sicherman and Galor’s (1990) theory of occupational careers. If persons expect that lower status entry positions will increase their later career chances, this explains why they would choose such entry positions rather than better paying jobs.

22  Handbook of health inequalities across the life course

THE LIFE COURSE AS AN EXPRESSION AND MECHANISM OF SOCIAL INEQUALITIES If we want to assess the potential fruitfulness of life course theory and research for the study of health inequalities across the life course, we should be especially interested in the more general ways in which social inequalities and the life course can be linked. In the following, I will discuss two major aspects of this linkage: description – time related dimensions of social inequality; explanation – mechanisms in the generation and reproduction of inequalities (Fasang and Mayer 2020). Social inequalities refer to different aspects of power and valued resources: material resources, physical or political power, and symbolic power or cultural capital. Correspondingly, we can distinguish between distributional, relational, and symbolic aspects of social stratification, and between class structures, hierarchies, and status systems as different forms of stratification orders (Weber 1956: 223–226). Stratification orders are intrinsically bound up with lifetime in the sense that class positions and status membership are often thought to be lifelong. Max Weber defined “social classes” as exactly those social categories between which mobility either between or within generations is low (Weber 1956: 223). If we assume membership in social classes or status groups to be permanent, the major process of interest is the transition period between the family of origin and class of destination. Within life course research this corresponds to research in transitions to adulthood; in stratification research it is the interest in intergenerational social mobility and the process of status attainment. However, once we think of the allocation of resources and positions as a lifelong rather than transitory process, a broad scope of the interlinkages between the life course and social inequalities opens up. The following temporal dynamics of social inequality can be considered: (1) Timing: at which point in life do which type and extents of inequalities apply with what consequences? (2) Duration: how long do adversities and advantages last and how do they add up? (3) Sequence: in which order do states and positions, for example, of social class occur? (4) Direction and trajectories: do status or other inequality outcomes change over time and in which direction? How do trajectories cluster into typical socially stratified life course patterns? Together these aspects mark overall stability/volatility: how unstable, regular or irregular are markers of social inequality? (Fasang and Mayer 2020: 26–27). Indeed, stability or security and its counterpart, volatility, can be viewed as an additional aspect of inequality. Even more important than such descriptive aspects is the significance of the life course for the explanation of social inequalities. The life course can in this context be seen as the arena reflecting the mechanisms in which social inequalities are causally attained, allocated, accumulated, and redistributed. As a lifetime line these processes occur in early and later childhood in the family of origin, via pre-school, primary and secondary education, training, entry into the labor market, employment and work and retirement. In the stratification literature, one of the predominant paradigms is the so-called O-E-D (origin/education/destination) triangle, concerning the causal relationships between origin, education, and class of destination. Here the questions are how much in the destination outcome can we explain by social origin and education, and, crucially, are all the influences of the family of origin mediated via educational achievement or and to what extent can we also observe direct effects of origin on destination. More generally, we can ask whether causal influences across the life course conform to Markov chains, i.e., whether transmission occurs

Sociology of the life course and its implications for health inequalities  23 step by step by “absorbing” events and states, or whether there are “latent” influences which are not or not fully mediated by intervening states. As to the nature of causal influences across the life course regarding social inequalities, a lot of recent attention has been paid to processes of positive feedback – accumulation and cycles of advantage (Dannefer 2003, 2020; DiPrete and Eirich 2006) and less so to processes of negative feedback – cycles of disadvantage and deprivation. Organizational careers are a form of orderly progression, which might hit a plateau. Another functional form is that of diminishing returns, for example the wage gains resulting from early investment in skills. Positive and negative feedback would result in widening population inequalities across the life course, as in the example of processes of educational attainment during schooling. While – especially in a regression logic – we tend to assume that good initial conditions lead to good outcomes and bad conditions lead to bad outcomes, there is also increasing attention to the two deviating trajectories: “resilience”, meaning that people succeed despite adverse early conditions and “skidding”, the idea that downward mobility occurs despite advantageous early conditions (Schoon 2006). Fruitful here is the distinction between “disadvantage” and “handicap”. Disadvantage refers to conditions which are bad in a given life period while handicap refers to conditions which might not be experienced as bad at a given time but have negative consequences in the future. The conjoint study of socio-economic life courses and socio-demographic life courses has furthermore turned the focus on mutually reinforcing positive and negative events and processes. Here the idea of the life course as unfolding “exposure to risks” is crucial. Early childbirth in many countries leads to lifelong adverse employment trajectories and income poverty for mothers (Vandencasteele 2011; Aisenbrey and Fasang 2017). The idea of “path dependency” refers to crucial turning points as opportunities or risks, which may lead to either long-term positive or negative trajectories (Heckhausen and Buchmann 2019).

LIFE COURSE INEQUALITIES AND HEALTH INEQUALITIES Health as an aspect of social inequality is an ambiguous concept. In the analytical terms of social stratification, health like age or beauty seems to be a property of natural heterogeneity, i.e. an “ascribed” characteristic rather than an “achieved” characteristic. In this sense health may be important as a cause of social inequalities. But health is not only a given condition, but also to some extent malleable. And in this sense health has to be viewed as a consequence of social inequalities. Finally, could health even be seen as a dimension of inequality like education, income, or occupational status? If we appreciate health as a resource, as a capacity, then the distribution of this specific resource could also be taken as a dimension of social stratification. This analytical complexity is likewise mirrored in the interrelationships between the life course and health inequalities. At stake are not only the mutually causal impacts of social inequalities and health inequalities, but also their endogenous temporal dynamics and interconnections. Both the topics of the life course in health (Kuh and Ben-Shlomo 2004; Davey-Smith 2008; Ben-Shlomo et al. 2016; Hoffmann et al. 2017; Siegrist and Staudinger 2019; Staudinger et al. 2020) and the topic of health stratification (Mackenbach 2019) are relatively recent areas of research. While the literature is growing rapidly, I want to highlight

24  Handbook of health inequalities across the life course here what I see as topics where already the life course has gained significance in the study of health inequalities and topics which appear to be still unchartered. Some prominent topics of the sociology of the life course have already be taken up in the literature on health inequalities across the life course, as described below. Age gradient/age roles. That different ages and life phases are associated with differential rates of health risks, morbidity and mortality is well established. This is not least recognized by the fact that pediatrics and geriatrics are medical specialties. Within life course epidemiology three life phases have been demonstrated to have particular salience for subsequent unequal health outcomes: the prenatal period and early childhood; middle adulthood; and, more recently, adolescence (Wahrendorf 2015; Alwin 2016; Ben-Shlomo et al. 2016). A recent study traces increasing working age mortality in the U.S. to the impact of automation on workers (O’Brien et al., forthcoming). Normative age roles define sickness among adults as a temporary condition where both the subjects and the health system are obliged to restore the healthy condition (Parsons 1975). In contrast, chronic disease is a socially acceptable condition in old age. However, quite recently “young olds” are now normatively defined as healthy and physically active. In the ageing literature it has increasingly been recognized that chronological age only very poorly predicts functional age (Baltes and Mayer 1999). But at the same time chronological age has become increasingly important as a criterion of exclusion and allocation in the realm of welfare policies (Behrens 2006; Mayer and Müller 1986). Dimensions of social stratification. Although many epidemiological studies work with relatively crude proxy measures of social inequality, like years of education, exemplary theory building and research have demonstrated the very specific mechanisms of particular aspects of social stratification. This applies particularly to studies of occupational stress (Siegrist 2021 and Chapter 12 in this volume) and to the salience of the non-economic impact of status and hierarchy on mortality and cardiovascular disease (Marmot 2015). Both latter areas of the sociology of health have significantly contributed to the study of social inequality in general. Lifetime as a metric. The penultimate measure of health is life expectation and its social differentials, their cross-national differences and their changes are well documented and the object of intense public interest and debate. More recently, the “expectation of healthy or disability-free life” and thus the duration of years in chronic conditions have become prominent as descriptive uses of lifetime (Mackenbach 2019). Life course etiology. Among the many possible interconnections of health and the life course three areas have received most analytical and empirical attention: critical periods; cumulative exposure; and pathways (Kuh and Ben-Shlomo 2004). Analogous to the discussion of “sensitive periods” and corresponding cohort effects the topic of critical periods in epidemiology relates primarily to the prenatal period and early infancy (Barker 1998). A close association between adverse prenatal or childhood health deprivation and socio-economic conditions has frequently been established (Modin 2002; Doblhammer 2004; Coneus and Spiess 2012; Solis et al. 2015). Early health impairments, like low birth weight, have been postulated and shown to be significantly correlated with higher rates of heart disease, obesity or diabetes in later life. The causal interrelationships between early and later adverse socio-economic conditions and early and later health outcomes are objects of an intensive methodological reflection and debate (Kuh and Ben-Shlomo 2004; Ben-Shlomo et al. 2016). Less controversial are the assumptions and evidence about the effects of cumulative adverse exposure, for example, to smoking and pollution.

Sociology of the life course and its implications for health inequalities  25 Selection. Stratification is not only a process of exclusion, but also a process of selection – in education according to ability and performance; in labor markets according to skills; and as allocation to positions and in marriage markets according to resources and personal attractiveness. Health is stratified as strongly demonstrated in the occupational stress literature (Siegrist in this volume) and as effects of hierarchy in the Whitehall studies (Marmot 2015). But class and social status can also be the outcomes of health selection (Hoffmann et al. 2018), whereby less healthy and disabled persons do less well in education, labor and marriage markets. Chains of risk, latent transmission. A final aspect of the relationship between the life course and health (inequalities) relates to the mechanisms through which health impairments are transmitted across the life course (Kuh and Ben-Shlomo 2004; Ben-Shlomo et al. 2016). As indicated above in the life course as event histories tradition, causal influences can be transmitted via direct effects serially, like bad health leading to reduced educational attainment leading to lower occupational success. And health impairments can be transmitted in a latent manner and then emerge at later periods in life as an indirect effect. Organic inflammation has been shown to act as such a latent transmitter (Berger et al. 2019). Finally, what then are topics of stratification and the life course which have not or have been less taken up regarding health inequalities and the life course? What is the potential still to be uncovered? Measurement in biomedical and socio-economic studies. A persistent problem in longitudinal studies is that biomedical studies lack adequate socio-economic measures while socio-economic studies lack adequate biomarkers (Leopoldina 2016). Partial exceptions are the British Cohort Studies (Wadsworth et al. 2006; McMunn 2020) and the SHARE – Survey of Health, Ageing and Retirement in Europe (Börsch-Supan et al. 2013; Börsch-Supan 2020). But also the meaning of well-established indicators might change. Morbidity for a long time has been taken for granted as a significant indicator of health. But if WHO and the UN define self-determined full participation in society as a major characteristic of health, chronic disease and disability under optimal support conditions need not infringe such participation (Behrens 2002). Lifelong observations. Most studies on the impact of life course antecedents and later life outcomes have concentrated on relatively short periods (like prenatal conditions and childhood or transitions to adulthood). This is especially true for the analysis of health outcomes which have, for instance, related early childhood health to schooling outcomes or started observations in later adulthood (Wahrendorf 2015; Börsch-Supan 2020) or old age (Baltes and Mayer 1999). Rare exceptions are again the early British Cohort Studies (Wadsworth et al. 2006). With “the age” progression of prospective panels like the Household Income Studies and new cohort studies like the German National Cohort Study or the adult cohort of the German National Educational Panel this situation is likely to improve. One area where such data have increasingly become available are subjective indicators of health (Wahl et al. 2021) which show the expected hyperbolic age gradient. However, recent studies of cognitive development in older age have shown the average decline with age to be an artefact of selective survival which can be more adequately explained by distance to death (Gerstorf et al. 2013). Turning points. More continuous and lifelong observations would also allow us not only to study the onset of chronic diseases, but also the impact of accidents and illnesses as “turning points” and potentially resulting path dependencies. Fractures of the femur in old age often lead to transitions to nursing homes and downward spiraling health. Also, the onset of

26  Handbook of health inequalities across the life course Alzheimer’s and old age dementia may completely redirect the lives of the chronically ill and their caretakers. Linked lives. One major topic in the life course literature has been the consideration of close family members and other social relations as life course “convoys”. This has been given less attention so far in the sociology of health but is partially echoed in the studies on the salience of social support. Institutional contexts. There are wide differences between countries in mortality, the prevalence of diseases and in corresponding health inequalities. Also, the share of expenses on health in the gross national product varies hugely between countries, as do the organization and benefits of health insurance systems and health care. An emergence of cross-national comparisons has documented many of these variations (Mackenbach 2019; Sieber et al. 2020). Yet, there are still many open issues which could benefit from sociological studies on national “life course regimes” (DiPrete 2002; Mayer 2005; Van Winkle and Fasang 2017). Sequence analysis. Only recently, methods for analyzing long sequences of “states” across the life course have become available (Abbott and Tsai 2000). Multi-channel sequence analysis has proven to be particularly fruitful in studying the mutual impacts of sequences in several life domains like work and family (Aisenbrey and Fasang 2017; Piccareta and Studer 2019). The application to health outcomes has so far been exceptional. McMunn and co-authors have used the data from the British Cohort Studies to test the hypothesis of whether women combining work and family roles (“role enhancement”) have better health outcomes in later life than women with little or no attachment to the labor market (McMunn 2020). So far, however, no research has been published which constructs and analyzes health sequences. While this might be difficult based on survey data, administrative data – for instance from health insurances – might make this possible. Life course cube (Bernardi et al. 2019). As noted above, the proposal of the life course cube constitutes a major advance in the theory building on the life course. Its major improvement is the combination of multiple levels (intra-individual organic-psychological functioning, individual action and behavior, and inter-individual collective population, institutional and cultural) with respective multiple domains and multiple time clocks. This opens a wide scope for the study of socio-economic and health inequalities across the life course. Life course observatory. Within recent decades has data become available which contains observations not only on ever longer stretches of the lifetime, ever longer historical periods and birth cohorts, but also across ever more countries (Mayer 2015b; Van Winkle and Fasang 2017). This allows us to assess the relative universality and stability of life course dynamics and the interrelationships between the socio-economic, socio-demographic life course and health by means of cross-national and cross-cohort comparisons. It thereby allows to assess the explanatory power of institutional and cultural contexts and determinants of health, as for instance the salience of health insurance systems and other social policies. A striking example is a study of the impact of different national retirement policies on cognitive abilities in old age (Rohwedder and Willis 2010). Stability and independence. Famously, Kuh and Ben-Shlomo (2004) highlighted one dominant and one ascending view of health determinants. The dominant one is that concurrent environmental conditions and personal behaviors are responsible for the bulk of chronic diseases and that socio-economic factors play an important role. The ascending view is that early “critical periods” and lifelong processes of cumulative exposure must be taken into account in explaining chronic diseases. Regarding both the socio-economic determinants of health

Sociology of the life course and its implications for health inequalities  27 and their life course dynamics, the usual research strategy was to demonstrate the importance of such factors. One open question in this regard is also the issue of effect size and impact (Mackenbach 2019). How important is life course change in comparison with life course stability and what is the relative explanatory power of various dimensions and magnitudes of inequalities? The next step in demonstrating the importance of life course factors is to measure their relative weight.

CONCLUSIONS The fields of the sociology of social stratification, the sociology of the life course, and the sociology of health inequalities and of life course epidemiology have for many years developed independently from each other. Only in recent decades have more and more intersections been achieved regarding conceptual approaches, empirical research, and methodological tools. As the attempted overview in this chapter documents, this synthesis and cooperation has been enormously fruitful. But I would also claim that it is far from complete and still offers many opportunities.

REFERENCES Abbott, A. and Tsai, A. (2000) “Sequence analysis and optimal matching methods in sociology: Review and prospect”, Sociological Methods & Research, 29. https://​doi​.org​.10​.1177/​0049124100029001001 Alwin, D. (2016) “It takes more than one to tango: Life course epidemiology and related approaches”, International Journal of Epidemiology, 45(4), pp. 988–993. Aisenbrey, S. and Fasang, A. (2017) “The interplay of work and family trajectories over the life course: Germany and the United States in comparison”, American Journal of Sociology, 122(5), pp.  1448–1484. https://​doi​.org/​10​.1086/​691128 Assmann, A. (2007) Geschichte im Gedächtnis. München: Beck. Baltes, M. and Carstensen, L. (2003) “The process of successful aging: Selection, optimization, and compensation”. In Staudinger, U. M. and Lindenberger, U. (eds.), Understanding Human Development. Dialogues with Lifespan Psychology. Boston: Kluwer Academic Publishers, pp. 81–104. Baltes, P. B. and Mayer, K. U. (eds.) (1999) The Berlin Aging Study: Aging from 70 to 100. Cambridge: Cambridge University Press. Barker, D. J. (1998) Mothers, Babies and Health in Later Life. Edinburgh: Churchill Livingstone. Becker, R. and Schömann, K. (1996) “Berufliche Weiterbildung und Einkommensdynamik”, Kölner Zeitschrift für Soziologie und Sozialpsychologie, 48, pp. 426–461. Behrens, J. (2002) “Inklusion durch Anerkennung. Chronische Krankheit, das Veralten der Indikatoren sozialer Ungleichheit und die Herausforderungen an die Pflege und andere Gesundheitsberufe”, Österreichische Zeitschrift für Soziologie, 27(4), pp. 23–41. Behrens, J. (2006) “Altern und Alterung: Sozialdemografische Folgen betrieblicher Strategien”, Schweizerische Zeitschrift für Soziologie, 32(3), pp. 429–455. Behrman, J. (2003) “An economic perspective on selection, optimization, and compensation (SOC)”. In Staudinger, U. M. and Lindenberger, U. (eds.), Understanding Human Development. Dialogues with Life Span Psychology. Boston: Kluwer Academic Publishers, pp. 125–156. Ben-Shlomo, Y., Cooper, R. and Kuh, D. (2016) “Life course epidemiology: The last two decades and future directions”, International Journal of Epidemiology, 45(4), pp. 973–987. Berger, E., Castagné, R., Chadeau-Hyam, M. et al. (2019) “Multi-cohort study identifies social determinants of systemic inflammation over the life course”, Nature Communications, 10(1), p. 773. Bernardi, L., Huinink, J. and Settersten, R. A. J. (2019) “The life course cube: A tool for studying lives”, Advances in Life Course Research, 41(3). https://​doi​.org​./​10​.101016/​j​.alcr​.2018​.11​.004

28  Handbook of health inequalities across the life course Bertaux, D. (ed.) (1981) Biography and Society: The Life History Approach in the Social Sciences. Beverly Hills, CA: Sage. Blossfeld, H.-P., Hamerle, A. and Mayer, K. U. (1989) Event History Analysis: Statistical Theory and Application in the Social Sciences. Hillsdale, NJ: Erlbaum. Blossfeld, H.-P. Roßbach,H.-G. and von Maurice, J. (eds.) (2011) “Education as a lifelong process. The German National Educational Panel Study (NEPS)”, Zeitschrift für Erziehungswissenschaft, Special Issue 14. Börsch-Supan, A. (2020) “Life history analyses with SHARE”. In Falkingham, J., Evandrou, M. and Vlachantoni, A. (eds.), Handbook on Demographic Change and the Lifecourse. Cheltenham, UK and Northampton, MA, USA: Edward Elgar Publishing, pp. 69–81. Börsch-Supan, A., Brandt, M., Hunkler, Ch., Kneip, Th., Korbmacher, J., Malter, F., Schaan, B., Stuck, S. and Zuber, S. (2013) “Data resource profile: The Survey of Health, Ageing and Retirement in Europe (SHARE)”, International Journal of Epidemiology, 42, pp. 992–1001. Breen, R. and Goldthorpe, J. H. (1997) “Explaining educational differentials: Towards a formal rational action theory”, Rationality and Society, 9, pp. 275–305. Buchmann, M. C. and Kriesi, I. (2011) “Transition to adulthood in Europe”, Annual Review of Sociology, 7(1), pp.  481–503. https://​doi​.org/​10​.1146/​annurev​-soc​-081309​-150212 Bude, H. (1995) Das Altern einer Generation. Die Jahrgänge 1938 bis 1948. Frankfurt am Main: Suhrkamp. Bühler, Ch. (1928) Kindheit und Jugend. Leipzig: Hirzel. Carroll, G. R. and Mayer, K. U. (1986) “Job-shift patterns in the Federal Republic of Germany: The effects of social class, industrial sector, and organizational size”, American Sociological Review, 51(3), pp. 323–41. Clausen, J. A. (1986) The Life Course: A Sociological Perspective. Englewood Cliffs, NJ: Prentice Hall. Coneus, K. and Spiess, C. K. (2012) “The intergenerational transmission of health in early childhood: Evidence from the German Socio-Economic Panel study”, Economics & Human Biology, 10, pp. 89–97. Dannefer, D. (2003) “Cumulative advantage/disadvantage and the life course: Cross-fertilizing age and social science theory”. The Journals of Gerontology. Series B, Psychological Sciences and Social Sciences, 58(6), pp.  327–337. http://​www​.ncbi​.nlm​.nih​.gov/​pubmed/​14614120 Dannefer, D. (2020) “Systemic and reflexive: Foundations of cumulative advantage/disadvantage and life course processes: Cross-fertilizing age and social science theory”, The Journals of Gerontology. Series B, Psychological Sciences and Social Sciences, 75(6), pp. 1249–1273. doi:10.1093/geronb/ gby118 Davey-Smith, G. (2008) “Die Bedeutung der Lebenslaufsperspektive für die Erklärung gesundheitlicher Ungleichheit”. In Bauer, U., Bittlingmayer, U. H., and Richter, M. (eds.), Health Inequalities. Determinanten und Mechanismen gesundheitlicher Ungleichheit. Wiesbaden: VS Verlag für Sozialwissenschaften, pp. 291–330. DiPrete, T. A. (2002) “Life course risks, mobility regimes, and mobility consequences: A comparison of Sweden, Germany, and the United States”, American Journal of Sociology, 108, pp. 267–309. DiPrete, T. A. and Eirich, G. M. (2006) “Cumulative advantage as a mechanism for inequality: A review of theoretical and empirical developments”, Annual Review of Sociology, 32, pp. 271–297. Doblhammer, G. (2004) The Late Life Legacy of Very Early Life. Berlin and Heidelberg: Springer. Ehmer, J. (2019) “Altersbilder und Konzeptionen des Alter(n)s im historisch-kulturellen Vergleich”. In Hank, K., Schulz-Nieswandt, F., Wagner, M., and Zank, S. (eds.), Alternsforschung: Handbuch für Wissenschaft und Studium. Baden-Baden: Nomos Verlag, pp. 19–45. Eisenstadt, S. N. (1964) From Generation to Generation: Age Groups and Social Structure. New York: Free Press of Glencoe. Elder, G. H. Jr. (1974) Children of the Great Depression. Chicago, IL: University of Chicago Press. Elder, G. H., Jr. (2001) “Life course: Social aspects”. In Smelser, N. and Baltes, P. B. (eds.), International Encyclopedia of the Social and Behavioral Sciences (Vol. 13). Cambridge, UK: Elsevier, pp. 8817–8821. Elder, G. H. Jr., Johnson, M. K., and Crosnoe, R. (2003) “The emergence and development of life course theory”. In Mortimer, J. T. and Shanahan, M. J. (eds.), Handbook of the Life Course. New York: Kluwer/Plenum, pp. 3–19.

Sociology of the life course and its implications for health inequalities  29 Erikson, E. H. (1980) Identity and the Life Cycle. New York: Norton. Esping-Andersen, G. (1990) The Three Worlds of Welfare Capitalism. Princeton, NJ: Princeton University Press. Esping-Andersen, G. (1999) Social Foundations of Postindustrial Economies. Oxford: Oxford University Press. Fasang, A. E. (2012) “Retirement patterns and income inequality”, Social Forces, 90, pp. 685–711. Fasang, A. E. and Mayer, K. U. (2020) “Lifecourse and social Inequality”. In Falkingham, J., Evandrou, M. and Vlachantoni, A. (eds.), Handbook on Demographic Change and the Lifecourse. Cheltenham, UK and Northampton, MA, USA: Edward Elgar Publishing, pp. 22–39. Gerstorf, D., Ram, N., Lindenberger, U. and Smith, J. (2013) “Age and time-to-death trajectories of change in indicators of cognitive, sensory, physical, health, social, and self-related functions”, Developmental Psychology, 49, pp.  1805–1821. https://​doi​.org/​10​.1037/​a0031340 Hall, P. A. and Soskice, D. (eds.) (2001) Varieties of Capitalism. The Institutional Foundations of Comparative Advantage. Oxford: Oxford University Press. Härkönen, J., Manzoni, A. and Bihagen, E. (2016) “Gender inequalities in occupational prestige across the working life: An analysis of the careers of West Germans and Swedes born from the 1920s to the 1970s”, Advances in Life Course Research, 29, pp. 41–51. Heckhausen, J. (2003) “The future of lifespan developmental psychology: Perspectives from control theory”, in Staudinger, U. M. and Lindenberger, U. (eds.), Understanding Human Development. Dialogues with Life Span Psychology. Boston: Kluwer Academic Publishers, pp. 383–400. Heckhausen, J. and Buchmann, M. (2019) “A multi-disciplinary perspective on canalization and agency”, Advances in Life Course Research, 41(3). Heckman, J. J. (2006) “Skill formation and the economics of investing in disadvantaged children”, Science, 312, pp. 1900–1902. Hitlin, S. and Kwon, H. W. (2016) “Agency across the life course”. In Shanahan, M. J., Mortimer, J. T. and Johnson, M. K. (eds.) Handbook of the Life Course, Volume II. Heidelberg, Germany: Springer International Publishing, pp. 431–449. Hoffmann, R., Kröger, H. and Pakpahan, E. (2017) “Health inequalities and the interplay of socioeconomic factors and health in the life course”, in Meloni, M. et al. (eds.), Handbook of Biology and Society. London: Palgrave Macmillan. Hoffmann R., Kröger, H. and Pakpahan, E. (2018) “Pathways between socioeconomic status and health: Does health selection or social causation dominate in Europe?”, Advances in Life Course Research, 36, pp. 23–36. Huinink, J. (1995) Warum noch Familie? Zur Attraktivität von Partnerschaft und Elternschaft in unserer Gesellschaft. Frankfurt am M: Campus Verlag. Keller, H. (2001) “Lifespan development: Evolutionary perspectives”. In Smelser, N. and Baltes, P. B. (eds.) International Encyclopedia of the Social and Behavioral Sciences, (Vol. 13). Cambridge, UK: Elsevier, pp. 8840–8844. Kohli, M. (1981) “Biography: Account, text, method”. In Bertaux, D. (ed.), Biography and Society: The Life History Approach in the Social Sciences. Beverly Hills, CA: Sage, pp. 61–75. Kohli, M. (1985) “Die Institutionalisierung des Lebenslaufs. Historische Befunde und theoretische Argumente”, Kölner Zeitschrift für Soziologie und Sozialpsychologie, 37, pp. 1–29. Kuh, D. and Ben-Shlomo, Y. (2004) “Introduction”. In Kuh, D. and Ben-Shlomo, Y. (eds.), A Life Course Approach to Chronic Disease Epidemiology (2nd edn). Oxford: Oxford University Press. Leopoldina (2016) The Relevance of Population-based Longitudinal Studies for Science and Social Policies. Halle: Leopoldina-German National Academy of Sciences Lersch, Ph. M., Schulz, W. and Leckie, G. (2020) “The variability of occupational attainment: How prestige trajectories diversified within birth cohorts over the twentieth century”, American Sociological Review, 85, pp. 1084–1116. Lindenberger, U. (2014) “Human cognitive aging: Corriger la fortune?” Science, 346(6209), pp. 572–578. https://​doi​.org/​10​.1126/​science​.1254403 Linton, R. (1939) “A neglected aspect of social organization”, American Journal of Sociology, 65, pp. 870–886. Mackenbach, J. (2019) Health Inequalities. Persistence and Change in Modern Welfare States. Oxford: Oxford University Press.

30  Handbook of health inequalities across the life course Mannheim, K. (1928), “The problem of generations”, reprinted (1952) in P. Kecskemeti (ed.), Essays on the Sociology of Knowledge: Collected Works, Vol. 5. New York: Routledge, pp. 276–322. Manzoni, A., Härkönen, J. and Mayer, K. U. (2014) “Moving on? A growth-curve analysis of occupational attainment and career progression patterns in West Germany”, Social Forces, 92, 1285–1312. Mare, R. (1980) “Social background and school continuation transitions”, Journal of the American Statistical Association, 75, pp. 295–305. Marmot, M. (2004) The Status Syndrome: How Your Social Standing Affects Your Health and Longevity. New York: Holt. Marmot, M. (2015) The Health Gap: The Challenge of an Unequal World. London: Bloomsbury. Mason, W. M. and Fienberg, S. (eds.) (1985) Cohort Analysis in Social Research. New York: Springer Publishers. Mayer, K. U. (1986) “Structural constraints on the life course”, Human Development, 29, pp. 163–170. Mayer, K. U. (2004) “Whose lives? How history, societies, and institutions define and shape life courses”, Research in Human Development, 1, pp. 161–187. Mayer, K. U. (2005) “Life courses and life chances in a comparative perspective”. In Svallfors, S. (ed.), Analyzing Inequality. Stanford: Stanford University Press, pp. 17–55. Mayer, K. U. (2015a) “The German Life History Study – an introduction”, European Sociological Review, 31(2), pp.  137–43. https://​doi​.org/​10​.1093/​esr/​jcv011 Mayer, K. U. (2015b) “An observatory for life courses: Populations, countries, institutions, and history”, Research in Human Development, 12(3–4), pp. 196–201. Mayer, K. U. and Müller, W. (1986) “The state and the structure of the life course”. In Sorensen, A. B., Weinert, F. E. and Sherrod, L. R. (eds.), Human Development and the Life Course: Multidisciplinary Perspectives. Hillsdale, NJ: Lawrence Erlbaum Associates, Inc, pp. 217–245. Mayer, K. U. and Tuma, N. B. (1990) Event History Analysis in Life Course Research. Madison, WI: University of Wisconsin Press. McMunn, A. (2020) “Work, family, and the life course: Evidence from the British Cohort Studies”, in Falkingham, J., Evandrou, M. and Vlachantoni, A. (eds.) Handbook on Demographic Change and the Lifecourse. Cheltenham, UK and Northampton, MA, USA: Edward Elgar Publishing, pp 180–199. Meyer, J. W. (1986) “The self and the life course: Institutionalization and its effects”. In Sorensen, A. B., Weinert, F. E. and Sherrod L. R. (eds.) Human Development and the Life Course: Multidisciplinary Perspectives. Hillsdale, NJ: Lawrence Erlbaum, pp. 199–216. Modell, J., Furstenberg, F. F., Jr. and Hershberg, T. (1976) “Social change and transition to adulthood in historical perspective”, Journal of Family History, 1(1, Autumn), pp. 7–32. Modigliani, F. (1966) “The life cycle hypothesis of saving, the demand for wealth and the supply of capital”, Social Research, 33, pp. 160–217. Modin, B. (2002) Setting the Scene for Life. Longitudinal Studies of Early Social Disadvantage and Later Life Chances. Stockholm: Stockholm University/Karolinska Institute, Health Equity Series No 1. Myles, J. (1993) “Is there a post-Fordist life course?”. In Heinz, W. R. (ed.), Institutions and Gatekeeping in the Life Course. Weinheim, Germany: Deutscher Studien-Verlag, pp. 171–185. O’Brien, R., Bair, E. F. and Venkataramani, A. S. (forthcoming) “Death by Robots? Automation and working age mortality in the United States”, Demography. Parsons, T. (1942) “Age and sex in the social structure of the United States”, American Sociological Review, 7, pp. 604–616. Parsons, T. (1975) “The sick role and the role of the physician reconsidered”, The Milbank Memorial Fund Quarterly. Health and Society, pp. 257–278. Piccareta, R. (2017) “Joint sequence analysis. Association and clustering”, Sociological Methods & Research, 46(2), pp. 252–287. Piccareta, R. and Studer, M. (2019) “Holistic analysis in the life course: Methodological challenges and new perspectives”, Advances in Life Course Research, 41(3). https://​doi​.org/​10​.0161/​j​.alcr 2018.10.004 Riley, M. W., Johnson, M. and Foner, A. (1972) “Elements in a model of age stratification”. In Riley, M. W., Johnson, M. and Foner, A. (eds.) Aging and Society, Volume 3: A Sociology of Age Stratification. New York: Russell Sage Foundation, pp. 3–26. Riley, M. W., Kahn, R. L. and Foner, A. (eds.) (1994) Age and Structural Lag: Society’s Failure to Provide Meaningful Opportunities in Work, Family, and Leisure. New York: Wiley.

Sociology of the life course and its implications for health inequalities  31 Rohwedder, S. and Willis, R. J. (2010) “Mental retirement”, Journal of Economic Perspectives, 24 (1), pp. 119–138. Ryder, N. B. (1965) “The cohort as a concept in the study of social change”, American Sociological Review, 30, pp. 843–861. Ryder, N. B. (1980) The Cohort Approach: Essays in the Measurement of Temporal Variations in Demographic Behavior. New York: Arno. Schoon, I. (2006) Risk and Resilience: Adaptations to Changing Times. Cambridge, UK: Cambridge University Press. Sicherman, N. and Galor, O. (1990) “A theory of career mobility”, Journal of Political Economy, 98(1), pp. 169–192. Sieber, S., Cheval, B., Orsholits, D., van der Linden, B. W. A., Guessous, I., Gabriel, R., Kliegel, M., von Arx, M., Kelly-Irving, M., Aartsen, M. J., Boisgontier, M. P., Courvoisier, D., Burton-Jeangros, C. and Cullati, S. (2020) “Do welfare regimes moderate cumulative dis/advantages over the life course? Cross-national evidence from longitudinal SHARE data”, The Journals of Gerontology: Series B, 75(6), pp.  1312–1325. https://​doi​.org/​10​.1093/​geronb/​gbaa036 Siegrist, J. (2021) Anerkennung und Gesundheit. Ein Dialog zwischen Soziologie und Medizin. Stuttgart: Alfred Kröner. Siegrist, J. and Staudinger, U. (eds.) (2019) Lebensverlauf und gesundheitliche Ungleichheit. Halle: Leopoldina-German National Academy of Sciences, Leopoldina Forum, Nr. 2 Solis, C. B., Kelly-Irving, M., Fantin, R. et al. (2015) “Adverse childhood experiences and physiological wear-and-tear in midlife: Findings from the 1958 British birth cohort”, Proceedings of the National Academy of Sciences of the United States of America, 112(7), pp. E738–E746. Staudinger, U., Kempermann, G., Patzwaldt, K., Ehmer, J., Fürnkranz-Prskawetz, A., Mayer, K. U., Sieber, C. and Siegrist, J. (2020) Aging and the Life Course: Research for Longer Lives. Halle: Leopoldina-German National Academy of Sciences. Stawarz, N. (2018) “Patterns of intragenerational social mobility: An analysis of heterogeneity of occupational careers”, Advances in Life Course Research, 38, pp. 1–11. Stulp, G. and Sear, R. (2019) “How might life history theory contribute to life course theory”, Advances in Life Course Research, https://​doi​.org/​10​.1016/​j​.alcr​.2019​.04​.011 Tegtmeyer, H. (1976) “Berufliche und soziale Umschichtung der Bevölkerung”, Zeitschrift für Bevölkerungswissenschaft, 2(1) pp. 4–33. Thomas, W. I. and Znaniecki, F. (1918) The Polish Peasant in Europe and in America: Monograph of an Immigrant Group. Chicago: University of Chicago Press. Vandencasteele, L. (2011) “Life course risks or cumulative disadvantage? The structuring effect of social stratification determinants and life course events on poverty transitions in Europe”, European Sociological Review, 27(2), pp.  246–263. https://​doi​.org/​10​.1093/​esr/​jcq005 Van Winkle, Z. and Fasang, A. E. (2017) “Life course variations across time and space. Complexity in employment careers in Europe between 1924 and 2009”, Social Forces, 96(1), pp.1–30. Wadsworth M., Kuh D., Richards M. and Hardy R. (2006), “Cohort profile: The 1946 national birth cohort (MRC national survey of health and development)”, International Journal of Epidemiology, 35(1), pp. 49–54. Wagner, M. (1985) Räumliche Mobilität im Lebensverlauf. Stuttgart: Enke. Wahl, H.-W., Drewelies, J., Duezel, S., Lachman, M. E., Smith, J., Eibich, P., Steinhagen-Thiessen, E., Demuth, I., Lindenberger, U., Wagner, G. G., Ram, N. and Gerstorf, D. (2021) “Subjective age and attitudes toward own aging across two decades of historical time”. Psychology and Aging. Advance online publication. https://​doi​.org/​10​.1037/​pag0000649 Wahrendorf, M. (2015) “Previous employment histories and quality of life in older ages: Sequence analyses using SHARELIFE”, Ageing and Society, 35(9), pp. 1928–1959. doi:10.1017/S0144686X14000713 Weber, M. (1956) Wirtschaft und Gesellschaft. Köln/Berlin: Kiepenheuer & Witsch. Yang, Y. and Land, K. C. (2013) Age–Period–Cohort Analysis: New Models, Methods, and Empirical Applications. Boca Raton, FL: CRC Press.

3. Cumulative dis/advantage processes, nutrition transition, and global metabolic disparities: interrogating the cohort–policy linkage Jessica A. Kelley, Abolade Oladimeji and Dale Dannefer

THE NUTRITION TRANSITION IN GLOBAL PERSPECTIVE The nutrition transition is a global phenomenon marked by the rising consumption of energy-dense, nutrient-poor foods that are high in fats, sugars, and salt. Such diets are typically marked by low levels of fiber and macronutrients, largely due to significant processing (Popkin 1993, 2002) and have been linked to widespread shifts in physical health. Global rates of obesity tripled between 1975 and 2016, increasing from 3.2 percent to 10.8 percent for men and from 6.4 percent to 14.9 percent for women (Di Cesare et al., 2016; WHO 2021). An estimated 39 percent of adults around the world are overweight (WHO 2021). As overweight and obesity have risen, we observe commensurate rises in metabolic morbidities of diabetes, hypertension, and other non-communicable diseases (Anauati et al., 2015; NCD Risk Factor Collaboration, 2017). Most high-income countries made this transition between the 1970s and the 1990s, with three key markers: (1) rise in average energy intake at all ages resulting in population overweight and obesity; (2) sharp rise in childhood obesity as younger cohorts were born into the new food order; and (3) transition of obesity from upper socioeconomic strata to a marker of poverty (Flegal et al., 2016; Kelley and Thorpe, 2021). These high-income countries, over a period of 100 years, transitioned from a focus on health problems associated with undernutrition to those associated with overnutrition. Low- to middle-income countries initially appeared to follow high-income countries in the nutrition transition, indicated by increases in both the availability of cheaper processed foods and in rates of population overweight and obesity (Gómez and Ricketts, 2013). However, scholars warn strongly that much of what we know about population obesity and metabolic morbidity burden has been drawn from primarily Western (and Global North) contexts where economic stability and increasing standards of living preceded the nutrition transition. Much less is known about the nutrition transition in low- to middle-income countries, which are more likely to be undergoing simultaneous advances in economic stability and population health (Defo, 2014). Unique factors in these societal contexts such as conflict, continuing struggles with communicable and infectious diseases, and fragile food supply chains point to the need for a more nuanced model of the nutrition transition, one that can account for the dual burden of undernutrition and overnutrition in the same population (Crush and Battersby, 2016), and one that allows for diverse patterns of obesity and metabolic risk within low- to middle-income countries. For low-income and rising middle-income countries, one of the major sources of metabolic morbidities continues to be severe undernutrition, which especially impacts young children. 32

Interrogating the cohort–policy linkage  33 In Sub-Saharan Africa (SSA), for instance, the annual number of deaths from protein-energy malnutrition, resulting from insufficient caloric and protein intake, has improved in the past decades, but children under the age of five continue to represent well over half of the deaths (66 percent in 2019). At the same time, rates of overweight and obesity in children and adults in this region have grown substantially (WHO, 2021). However, rather than being a problem of simply too much available food, overweight and obesity in these populations are caused by the markers of food insecurity, such as a lack of dietary diversity, protein-based sources, and macronutrients. These forces are often obscured in public health discourse that has largely focused on the obesogenic effects of unregulated consumption of processed foods that characterize the obesity problem in high-income countries (Ngaruiya et al., 2017). Despite clear linkages between the economy and the population’s nutrition, Popkin and Reardon (2018) argue that literatures on the nutrition transition and the economics of food system transformation have largely been “two ships passing in the night.” As a consequence, we have no framework for explaining population changes in metabolic morbidity burden in these developing societal contexts, including, but not limited to: families with overweight adults and stunted children (Kimani-Murage et al., 2010); overweight middle-aged adults who experienced severe undernutrition as children (Hoffman et al., 2000); and high rates of insulin-resistant diabetes in normal weight adults (Fischer et al., 2002). In an effort to move toward a framework for understanding the nutrition transition in low- to middle-income countries, we employ comparative cohort analysis and a consideration of the role of policy to examine the cumulative dis/advantage processes involved in shaping global metabolic disparities over the life course. Recent developmental and life-course research has made clear that the health impact of such changes is not “age-neutral” but falls differently on individuals depending on their place in the life course. If it is the case that the nutrition transition can be characterized as a period effect, as we argue here, then we can expect it to have differential effects on sequential cohorts as well. Thus, it is essential to consider how the nutrition transition relates to both cohort dynamics and life-course processes.

COMPARATIVE COHORT ANALYSIS AND SOCIAL CHANGE Evidence for the impact of social conditions on inequality within and between cohorts has been recently shown in both historical and comparative studies. An exemplar of this can be found in a study analyzing the impact of shifts in the US economy on both intracohort and intercohort inequality in recent decades, by comparing trajectories of inequality across cohorts (Zewde and Crystal, 2022). Centering their inquiry on a specific financial shock, the 2008 financial crisis in the United States, Zewde and Crystal demonstrate that the age at which a cohort experienced the shock differentiates how much inequality grew within the cohort during the financial recovery. For example, earlier cohorts such as the Baby Boomers initially lost a large amount of collective wealth but the uneven recovery following the crisis led to a greater concentration of wealth in the top 10 percent than prior to the crisis. By contrast, Millennials – who had less overall wealth to lose – experienced a much more equitable recovery. This is a clear example showing that the age at which a cohort encounters a significant shock or rapid social change has implications for the accumulative processes of inequality and, we would posit, subsequent health. Given a shock that is significant and timely enough to radically alter parts of everyday life, adjacent cohorts can exhibit very different profiles of

34  Handbook of health inequalities across the life course inequality, accumulative processes, and health impacts. This is particularly relevant in the case of metabolic-related health conditions, as timing, duration, and severity of exposure shape risk over the life course (Ben-Shlomo and Kuh, 2002). This cohort-comparative example illustrates the potential importance of integrating the age at which cohorts encounter the nutrition transition in their country when studying population patterns in metabolic disparities. Metabolic morbidities cannot be reduced to simply undernutrition versus overnutrition, but also clearly depend on the timing and degree of impact of these circumstances (Elder and Shanahan, 2006). In the remaining sections of this chapter, we develop a framework that integrates cohort dynamics, and global economic food policy, and the operation of CDA processes within and between countries, to inform our understanding of distinct aspects of metabolic disparities in low- to middle-income countries. Cohorts become a critically important organizing frame for understanding patterns of metabolic disparities within societies (Kelley and Thorpe, 2021). In nations with ongoing economic development, food availability and nutritional profiles can change within a single cohort’s collective lifetime. Thus, the age at which a cohort encounters severe undernutrition or overnutrition matters for explaining patterns in metabolic morbidities. For example, one cohort may begin with undernutrition in early life and encounter a calorie-dense, nutrient-poor context of processed and convenience foods during adulthood. A later cohort may be born into a nutritional context of excess calories with low nutrient content, shaping early life development and metabolic risks differently than the previous cohort. In Figure 3.1, we present a conceptual model demonstrating the relative impact of the nutrition transition on metabolic disparities based on the timing of a cohort’s exposure. For our purposes here, we focus on lower- and middle-income countries who have undergone the nutritional transition most recently. We first set a theoretical period when a country would undergo the bulk of the nutrition transition. We then array three successive birth cohorts who would have encountered the nutrition transition at different ages. Cohort 1 is well into adulthood when they encounter rapid changes to food availability and composition. Their pre-transition early life, particularly those in rural areas, would be marked by low birth weight, severe undernutrition, potentially leading to wasting and vulnerability to infectious diseases. Child mortality prior to the age of five would be very high. Central

Figure 3.1

Conceptual figure of nutrition transition and timing of impact on three sequential birth cohorts

Interrogating the cohort–policy linkage  35 to understanding the implications of these conditions is the Thrifty Phenotype hypothesis, proposed by Barker and associates (Hales and Barker, 2001). This hypothesis proposes that metabolic rates are set based on early-life nutritional scarcity, which then increases the risk for diabetes, cardiovascular disease, and obesity in later life. In sum, epigenetic adaptations necessary to survive in a food-scarce environment trigger functional changes to the metabolic and cardiovascular disease systems (Gluckman and Hanson, 2006), which manifest by adulthood in a set of characteristics that have been termed metabolic syndrome. Characteristics of the metabolic syndrome include hypersensitivity in calorie intake, elevated endocrine sensitivity to stress, and lower energy expenditure (Byrne and Phillips, 2006; Gluckman et al., 2009). These are directly related to overweight and elevated diabetes risk in adulthood, without an actual change in dietary intake (McMillen and Robinson, 2005). Thus, for this cohort, overweight is a symptom of scarcity, rather than one of abundance. Cohort 2, which we label the “Mismatch Cohort,” is marked by widespread early life undernutrition just as Cohort 1, thus would share many of the early-life risks and midlife metabolic morbidity risks. With gestation, childhood, and early adolescence predating the rapid nutrition transition for this cohort, we can expect rates of wasting and stunting to be equivalent to the previous cohort. However, this cohort would experience the nutrition transition in early adulthood, marked by a sharp rise in intake of nutrient-poor but calorie-dense foods. The “mismatch” thesis put forward by Gluckman and Hansen refers to the longer-term effects of the metabolic syndrome, as epigenetic changes developed in early life in response to nutritional deprivation can be viewed as adaptive for survival. Once the food environment shifts to a higher calorie environment, these epigenetic changes can be maladaptive for the new environment, leading to increased adiposity and insulin resistance. Compared to Cohort 1 with sustained undernutrition, Cohort 2’s “mismatch” between early undernutrition to midlife overnutrition is seen as complicating the development of metabolic morbidities. For example, some scholars suggest that the interaction between early-life deprivation and adult access to high-calorie foods is a key mechanism leading to the “dual metabolic burden” of simultaneous undernutrition and overnutrition in the population (Kinra et al., 2005). This is also consistent with research in low- to middle-income countries showing families with wasting and stunting in children yet overweight parents (Godfrey et al., 2017). Cohort 3 represents those who are born into or immediately following the nutrition transition. Early life diets (and maternal diets while in utero) are not marked as distinctly by an absolute deprivation of calories, as economic development brings nutritional diversity and stability to the country. However, such advancements many be uneven across all contexts, whereby some areas may have sufficient calories, but continue to lack protein-rich and macronutrient courses. A public health system in such a context would contend with the widest number of forces influencing metabolic risk of any previous cohort: scarcity or unstable food supplies, substitution of processed foods for whole foods, “mismatched” childhood and adult food environments, and early-life overnutrition. Notably, in Cohort 3, childhood overweight and obesity is much more common, but can be accompanied by stunting (Lobstein et al., 2015). In addition to the traditional metabolic risk factors associated with excess weight (e.g., insulin resistance, elevated blood pressure), overweight children appear to develop unique metabolic risks such as the way that fat accumulates in the liver (Boyraz et al., 2014). Long-term surveillance studies have not fully apprehended the adult health implications of these early-life biological changes, but the biologic mechanisms and resultant chronic disease are going to have many

36  Handbook of health inequalities across the life course more factors than in previous cohorts, which will challenge both prevention and intervention efforts (Wu et al., 2016).

NUTRITION TRANSITION, ECONOMIC DEVELOPMENT, AND CUMULATIVE DIS/ADVANTAGE ON A GLOBAL SCALE Since much of what we understand about the nutrition transition and metabolic disparities is based on a model generalized almost exclusively from high-income countries, it is important to explore how the nutrition transition itself, in terms of timing, changes in food availability, and concurrent economic development may differ in low- to middle-income countries. Such an expanded framework helps us to understand cohort differences in metabolic risks, as well as increasing inequality within cohorts as they age. To this end, we provide some brief context on the policy and economic forces that shaped – and continue to shape – the nutrition transition in these regions. Following World War II, which had decimated food production in Asia, Europe, and the Soviet Union, a new order was established to stabilize trade in global markets and to provide economic assistance to poorer countries. The International Monetary Fund and World Bank were created to facilitate global trade, which included food production and distribution. From the outset, these organizations were driven heavily by the interests of the United States and Western Europe, resulting in capitalist-friendly and neoliberal policies that imposed strict economic conditions on poorer countries, including draconian fiscal requirements in exchange for loans, reduction in public expenditure, foreign direct investments, and trade liberalization (Peet, 2009). These measures echoed the imperial policies of earlier practices (e.g., of plantation and export agriculture), marked by unfavorable trade conditions for the poorer nations and heavy taxation. Early efforts to aid lower-income countries in the form of subsidies for food, medicine, and other human necessities eroded steadily as profit-driven pressures to make the food economy a free market rose (Coburn, 2010). The consequence of globalization regulated by laissez-faire economic policies is that richer countries gained increasing access to drivers of the economy such as raw materials, markets, and consumers for their goods, and low-income countries experienced the extraction of resources and an ever-decreasing access to drivers of their own economies (Wells, 2016). For our purposes here, we point to three key features of IMF and World Bank policies that directly shaped nutrition and economic development in low- to middle-income countries following World War II: structural adjustment programs/austerity policies, large-scale land acquisitions by agribusinesses, and foreign direct investment. We select these three because they are illustrative of key features of cumulative dis/advantage processes on a macro scale. Accumulation is a fundamental driving impulse of capitalism (O’Connor, 1986), and neoliberalism can be seen, in part, as a manifestation of that impulse, as capital seeks new ways to return profits to its owners in the context of saturated markets and global competition. Accumulation in this sense can be seen as a specific case, albeit a pervasive and very powerful one, of the more general sociological principle of cumulative dis/advantage, which has been recognized as a general tendency of social life (Dannefer, 2003, 2020; Merton, 1988; Myrdal, 1944; Rigney, 2010). Cumulative dis/advantage can be defined as “… the systemic tendency for inter-individual divergence in a given characteristic (e.g., wealth, health or status) with the passage of time” in

Interrogating the cohort–policy linkage  37 which a social position itself has a valence in producing or precluding further gains (Dannefer, 2020: 120; see also Dannefer, 2003; DiPrete and Eirich, 2006). While progressive policy can arguably be a vehicle for reversing tendencies toward increasing inequality, neoliberal economic restructuring accomplishes just the opposite: it weakens social institutions which might moderate the negative effects on the health of weakened social structures – and often accelerates inequality (Coburn, 2004). Thus, on a global scale, such a lens can help us see how policy structures increase inequality between nations. First, structural adjustment programs and austerity policies are directly linked to the increased poverty in the 1980s and 1990s for many African and South Asian countries (Peet, 2009). By the 1980s, the per capita income of African countries operating under the IMF policies had declined by 25 percent since World War II. These policies removed food and agricultural subsidies from these countries, leading to escalating food prices within the country and concomitant population food insecurity (Peet, 2009). The consequences of these policies, for example in Ethiopia, meant a transition from sustainable agriculture and biodiversity to mono-production for global export. Biotechnology and agribusiness reshaped the entire landscape, by creating dependence on genetically modified seeds and fertilizer. Thus, the irony is that increases in crop production in the most fertile parts of the country caused these regions to be the most susceptible to famine (Chossudovsky, 2003). Further, the IMF and the World Bank failed to address protectionist policies that the high-income countries employed for themselves, leading to an increasing gap between rich and poor countries. According to Wells (2016), “subsidies benefitting low- and middle-income countries are labeled ‘trade-distorting’, and prohibited by the World Trade Organization, those benefitting high-income countries are considered ‘non-trade-distorting’ and are unrestricted. This double standard in the rules of the game ensures that wealth is systematically drawn from poorer countries to the wealthier” (Wells, 2016, p. 353). Second, large-scale land acquisitions in low- to middle-income countries by agribusinesses – or their own government for the purpose of selling products to agribusinesses – is another major cause of food insecurity in Latin America and Sub-Saharan Africa (Broughton, 2013; Margulis et al., 2013). Colloquially known as “land grabbing,” it is defined as the “acquisition or lease of large tracts of land in developing countries by wealthy states and corporate investors, usually for engaging in agribusiness and securing reliable sources of food and biofuel” (Özsu, 2019). Desperate for access to the global economy, many low- to middle-income countries actively solicit foreign investors to develop their land for biofuel export crops, such as sugarcane, corn, and palm oil. The short-term gains of corporate investment undermine sustainable farming practices that would feed their own people, or displace farmers and agrarian indigenous groups, thereby increasing food insecurity (Borras et al., 2012; Chossudovsky, 2003). Third, the movement toward privatized economies in low- to middle-income countries in the 1980s and 1990s opened the door to foreign direct investment (FDI) in food manufacturing in these countries (Hawkes, 2005). In contrast to the global commerce in agricultural commodities, largely defined by trade, processed foods are produced and distributed primarily through foreign direct investment (FDI) of transnational food companies. Prior to the 1980s, most middle-income countries owned the means of food production in their country (Lawrence, 2017). Beginning in the early 1980s, dual global policy pressures toward increased privatization and loosening regulations of foreign investment led to a near-complete takeover of the global food processing market by transnational food companies (predominantly located in

38  Handbook of health inequalities across the life course the US and Europe) in less than 20 years (Bolling et al., 1998). Foreign firms, such as Nestlé, Kraft Foods, and PepsiCo were able to enter another country’s food markets by acquiring or merging with small to medium companies already operating in that country. The foreign firm would then own the means of production, the technology, and the rights to distribute the food items. Generally, production then shifted to goods produced for the FDI’s home market, in the form of processed foods like beer, soft drinks, crackers, and breakfast cereals (Bolling et al., 1998; Wells, 2016).

NEOLIBERALISM AND INCREASING INEQUALITY IN GLOBAL METABOLIC CONDITIONS Rates of overweight and obesity have risen steadily in low- to middle-income countries, with 62 percent of obese persons globally now residing in these regions (GBD Obesity Collaboration, 2014). More than half of the 671 million obese persons live in ten nations, eight of which are low- to middle-income countries: Brazil, China, Egypt, India, Indonesia, Mexico, Pakistan, and Russia (Ford et al., 2017). The prevalence of childhood obesity in low- to middle-income countries steadily increased over a 25-year period, doubling – and sometimes quadrupling – from 1991 to 2016 (WHO, 2021). The socioeconomic gradient of obesity is now steepest in countries with moderate GDPs, compared to low and high GDPs, and is exacerbated further in countries with greater income inequality (Alaba and Chola, 2014; Jones-Smith et al., 2011). These obesity and metabolic patterns underscore the complex relationship between economic development and the nutrition transition across country contexts, as well as the uneven nutrition transition within countries. Public health scholars identify several drivers of population obesity that low- to middle-income countries share with high-income countries: changes in diet, reduced physical activity (especially in urban areas), and built environment. However, low- to middle-income countries have unique risk factors for population obesity and metabolic conditions, associated with their uneven economic development and global trade policies. These include environmental pollutants, inconsistent food supply, and epigenetic adaptations to scarce food environments in early life (Ford et al., 2017). Below, we explore three obesogenic implications of neoliberal food policy and their implications for population overweight/obesity in low-to-middle income countries. First, the restructured world food market shifted paradigmatically from “feeding the world” to “selling to the world,” thus tying population nutrition to the ability to compete by the rules of capitalism. Second, the removal of subsidies and price guarantees for export crops prevented disadvantaged countries from being economically independent. Third, the glut of cheap grains and processed foods on the global food market and exportation of the Western diet made the available food for disadvantaged countries obesogenic. 1.

From “Feeding the World” to “Selling to the World”

The paradigmatic restructuring of the world food market around global trade and foreign direct investment of transnational food corporations essentially tied population nutrition to a country’s ability to compete in the open global marketplace. A classic example of the infusion of capitalism in population nutrition has been the predatory marketing scheme of and other producers of artificial milk and baby formula in poor countries (Swift, 1982). In the absence

Interrogating the cohort–policy linkage  39 of a regulated market, the corporations were free to employ strategies directly on the new potential consumers. These included the following: (1) Marketers played on mothers’ fears of starving their children, coupled with racist tropes that breastfeeding was backward and base. (2) Artificial milk producers paid quasi-professional women to make home visits to new mothers to directly promote their products. (3) Largely White-run hospitals gave women free samples under the guise of official medical care (Piwoz and Huffman, 2015). Research shows that formula-fed infants have a higher risk of chronic diseases and obesity in adulthood compared to breast-fed infants (Abul-Fadl et al., 2021; Smith and Harvey, 2011). The mechanisms seem to operate in the digestive systems of the infants, as formula-fed infants have a less diverse gut microbiome profile than breast-fed infants, and exhibit signs of “metabolic stress” that lead to permanent changes in insulin resistance (O’Sullivan et al., 2013). In other words, capitalist interests created a market in low- to middle-income countries for a product that was not necessary and has been directly linked to metabolic diseases in adulthood. The World Health Organization (WHO) and the United Nations Children’s Fund (UNICEF) cooperatively developed the International Code of Marketing of Breast Milk Substitutes in 1981, but, ironically, it had no means to enforce it. The Code relied on the transnational food corporations to voluntarily implement it. Nestlé, the world’s largest producer of infant formula, signed on to the Code in 1984, though it was largely motivated by the desire to end a seven-year global boycott of its products (Sasson, 2016; Sethi, 2012). A second example is the exportation of fast food to the world. The ubiquity of US fast food restaurants, such as Kentucky Fried Chicken and McDonalds, in middle-income countries has long been the starting point for discussions of the global rise in obesity. Indeed, the correlation between population overweight and number of McDonalds restaurants per capita is robust and positive (Alheritiere et al., 2013). While an easy target to vilify, the presence of US fast food restaurants in countries such as Thailand, China, and Nigeria is merely symbolic of other changes related to rapid economic growth: rural migration to urban areas, increasing labor force participation of women, and increased disposable income (Traill et al., 2014). As such, most scholars believe that the rising prevalence of overweight and obesity in urban areas reflects the forces of industrialization and urbanization more broadly. Availability, convenience, and aggressive marketing make processed and fast foods attractive choices in these areas. As such, the socioeconomic gradient in obesity and metabolic disparities tends to be much steeper in urban areas, with higher rates among those with money to spend. Rural areas, however, do not exhibit strong socioeconomic gradients in obesity and metabolic conditions, largely due to the lack of availability of fast and processed food, and ubiquitously high rates of poverty (Alaba and Chola 2014). 2.

Undermining Economic Independence

Complex formal and informal arrangements between the host country governments and multinational corporations (MNCs) are frequently lopsided negotiations in favor of MNCs. Host countries who are relatively poor, sometimes newly independent, have great need for the resources that can accompany deals with multinational corporations, despite the exploitative nature of the relationship (Sethi, 2012). Global agribusiness corporations impose capitalist-friendly conditions on the poorer countries as a price of doing business. One example has been the prioritization of cash crops for

40  Handbook of health inequalities across the life course export (e.g., coffee, biofuels) over food crops for citizen consumption. Pressures to compete in the global market by transitioning to mono-production for export has destabilized local sustainable food systems, driving risk of undernutrition even higher (Wells, 2016). Concomitantly, this has entailed the rise of industrial farming by multinational food corporations, rendering local farmers unable to compete. A large portion of the population in SSA are farmers, with an average farm size of 3 hectares, and earn less than $1USD per day (FAO, 2017). Young adults often migrate to urban areas to find employment to support their family members who remain in the rural area, but such massive migration has created a significant mismatch in labor supply and availability of non-farm work in the urban areas. These migrants often end up in slums, new consumers of less healthy food grown and produced outside of their country (Battersby and Crush, 2016). Governments of poorer countries themselves have reshaped their agricultural economies as a way of generating money. For example, the proportion of land dedicated to growing sugar cane in Guatemala increased 400 percent from 1980 to 2008 (from less than 4 percent to 14 percent), with the government actively seeking foreign investment and foreign buyers (Niezen, 2013). In a widely publicized example of land grabbing, the Guatemalan government forcibly removed 769 indigenous families from the Polochic Valley so that the government could sell the land to a biofuel company to produce palm oil (Haddok, 2012). In sum, the lack of economic independence and sufficient per capita income in poorer nations severely restricts the amount of food available, coupled with the displacement of persons from land that they could farm. 3.

Cheap Availability of Obesogenic Foods and the Reversal of the Socioeconomic Gradient

The glut of cheap grains on the global food market makes the available food for disadvantaged countries obesogenic. One of the hallmarks of the nutrition transition has been the shift of overweight and obesity from affluent members of a society to the poor and middle-income members of the society (Monteiro et al., 2007). For countries in Africa, Latin America, and Southeast Asia who have been subject to foreign direct investment, this transition has a direct through line from the marketing practices of transnational food companies who first gained access to the means of production of processed foods in these poorer countries and then controlled the marketing. Initially, in developing or middle-income countries, the processed food products sold by the transnational food companies had a higher price point and were targeted to higher-income customers. Epidemiologists have documented that in the early phases of infiltration into these countries, obesity became concentrated in the upper-socioeconomic strata, while underweight continued to dominate the rest of the population (Reyes Matos et al., 2020). However, with the desire to expand market share, the companies began dropping their price point on processed foods through the 1990s (Hawkes, 2002; Rosenheck, 2008). With the greater accessibility and affordability of the processed foods, overweight and obesity began to concentrate among the lowest economic strata. Using Sub-Saharan Africa as an example, rates of overweight have increased substantially in the past 30 years (Wariri et al., 2021), quickly transitioning from a symbol of wealth to a symbol of the middle class (Daran and Levasseur, 2022). Some research shows that the transition for most SSA countries can be dated to the late 1990s, when the distribution of obesity in the population changed to an upside down “U” shape, with middle-class women exhibiting

Interrogating the cohort–policy linkage  41 the highest rates relative to their poorer and more affluent counterparts (Daran and Levasseur, 2022). The timing of this shift directly coincides with the fruition of policies allowing for privatized and unregulated food markets which spurred changes in marketing to lower prices of processed food. Urban areas experienced the greatest increase in rates of obesity due to the more readily available Western fast food and global commercial food products (Wariri et al., 2021). Ironically, many urban areas in SSA now grapple with the dual burdens of both undernutrition and overnutrition (Wojcicki, 2014). Surveillance studies reveal a rising trend of mothers in poor to middle-class families being overweight while their children show evidence of chronic undernutrition via stunting and wasting (Garrett and Ruel, 2005; Popkin et al., 2012). This is largely due to chronic or intermittent scarcity of protein and macronutrient food sources coupled with the rising availability of cheap processed foods lacking such nutrients (Popkin et al., 2012). These findings are consistent across middle-income contexts in Latin America and Southeast Asia (Drewnowski and Popkin, 1997), providing significant evidence that the nutrition transition and its metabolic consequences are due to global economic policy rather than state-specific policies or individual “choices,” as the advocates of the neoliberal paradigm argue. Heterogeneity within SSA countries is evident as well, driven largely by variable histories of conflict, infectious diseases such as HIV, successes toward economic development, and historic ties to non-African countries in the colonial period (Defo, 2014). These structural factors help to explain the steeper socioeconomic gradient of obesity in southern Africa, particularly the country of South Africa, relative to its counterparts in upper sub-Saharan Africa (Alaba and Chola, 2014; Defo 2014). It also helps to explain a less robust association of socioeconomic status and overweight/obesity in low-income nations in SSA that are not as far along in both economic development and the nutrition transition (Daran and Levasseur, 2022).

CONCLUSIONS In this chapter, we have focused on nutrition transition and associated economic developments as related to processes of cumulative dis/advantage in health and their cohort-based intersection with life-course timing. Focusing on low- to middle-income countries in the past several decades, we provide a fresh lens for understanding the role of accumulative processes in generating health disparities. Accumulation is a dynamic that cannot be understood without reference to the power of large-scale economic forces, and in particular the economic interests of dominant states, international organizations, and corporate actors to impose their programs on small and relatively powerless indigenous societies and interests. Such dynamics arguably represent an example of cumulation and cumulative dis/advantage operating at the level of nation-states rather than individuals, and in historical time rather than biographical time. Our arguments herein also challenge the standing model of the nutrition transition that has been based largely on Western and Global North countries who attained economic stability and abatement of infectious diseases prior to undergoing the nutrition transition. With low- to middle-income countries experiencing these transitions simultaneously, one must consider skeptically the assumption of a preceding “golden age,” present in the framing of the nutrition transition in Western and Global North countries (Defo, 2014). Globalization in trade and economies beginning in the mid-20th century have imposed conditions upon many

42  Handbook of health inequalities across the life course lower-income societies that have reshaped their nutritional challenges, often coupled with a loss of economic and political independence. This transition has accompanied a shift in the structure of local political control and economic production, often to the detriment of citizens in many domains of experience, including nutrition. While such adverse developments are unlikely to be eliminated under current political and economic arrangements, it is important to document their consequences for individual health and well-being. The impact of such shifts in available food and nutrition upon individual health over the life course inevitably will depend on a range of factors operating both within and across social contexts. We have suggested that one such set of such factors that warrants added attention includes those associated with the timing of the transition. The issue of timing has from the beginning been recognized as a key component in the analysis of life-course effects (e.g., Elder and Shanahan, 2006). We have suggested how the nutrition transition can be expected to have predictably different effects depending on the time of life at which it occurs. Despite the likely importance of such timing-based differences, it is important also to keep in focus the more encompassing reality that the general impact of this transition has not been adverse for those of every age and cohort.

REFERENCES Abul-Fadl, A. A., Mourad, M. M., Arafa, O. S., and Al-Jawaldeh, H. A. 2021, “Cardiovascular disease risk factors: Role of legislations that control marketing of breastmilk substitutes”, Indian Journal of Child Health, 8(9), 308–313. Alaba, O., and Chola, L. 2014. “Socioeconomic inequalities in adult obesity prevalence in South Africa: A decomposition analysis”, International Journal of Environmental Research and Public Health, 11(3), 3387–3406. Alheritiere, A., Montois, S, Galinski, M., Tazaroute, K., and Lapostolle, F. 2013, “Worldwide relation between the number of McDonald’s restaurants and the prevalence of obesity”, Journal of Internal Medicine, 274(6), 610–611. Anauati, M. V., Galiani, S., and Weinschelbaum, F. 2015, “The rise of noncommunicable diseases in Latin America and the Caribbean: Challenges for public health policies”, Latin American Economic Review, 24(11). Battersby, J., and Crush, J. 2016, “The making of urban food deserts”, in Rapid Urbanisation, Urban Food Deserts and Food Security in Africa, Springer, Cham, pp. 1–18. Ben-Shlomo, Y., and Kuh, D. 2002, “A life course approach to chronic disease epidemiology: Conceptual models, empirical challenges and interdisciplinary perspectives”, International Journal of Epidemiology, 31(2), 285–293. Bolling, C., Neff, S., and Handy, C. 1998, “U.S. foreign direct investment in the western hemisphere processed food industry”, Agricultural Economics Report No, 760, US Department of Agriculture, Economic Research Service. https://​www​.ers​.usda​.gov/​webdocs/​publications/​40729/​33239​_aer742c​ .pdf​?v​=​0 Borras Jr, S. M., Kay, C., Gómez, S., and Wilkinson, J. 2012, “Land grabbing and global capitalist accumulation: Key features in Latin America”, Canadian Journal of Development Studies/Revue canadienne d’études du développement, 33(4), 402–416. Boyraz, M., Hatipoglu, N., Sari, E., Akcay, A., Taskin, N., Ulucan, K., and Akcay, T. 2014, “Non-alcoholic fatty liver disease in obese children and the relationship between metabolic syndrome criteria”, Obesity Research and Clinical Practice, 8, e356–e363. DOI: 10.1016/j.orcp.2013.08.003. Broughton, A. 2013, “Land grabbing: A new colonialism”, Green Social Thought, 61, 25–29. Byrne, C. D., and Phillips, D. I. W. 2006, “The developmental environment and its role in the metabolic syndrome”, in P. Gluckman and M. Hanon (Eds.), Developmental Origins of Health and Disease, Cambridge, UK: Cambridge University Press, pp. 265–274.

Interrogating the cohort–policy linkage  43 Chossudovsky, M. 2003, The Globalization of Poverty and the New World Order (Vol. 2), Quebec: Global Research. Coburn, D. 2004, “Beyond the income inequality hypothesis: Globalization, neo-liberalism and health inequalities”, Social Science and Medicine, 58(1), 41–56. Coburn, D. 2010 “Inequality and health”, in C. Leys and L. Panitch (Eds.), The Socialist Register, Halifax, NS: Fernwood Publishing. Crush, J., and Battersby, J. (Eds.) 2016, Rapid Urbanisation, Urban Food Deserts and Food Security in Africa, Cham: Springer. Crystal, S., Shea, D. G., and Reyes, A. M. 2016, “Cumulative advantage, cumulative disadvantage, and evolving patterns of late-life inequality”, Gerontologist, 57, 910–920. DOI:10.1093/geront/ gnw056. Dannefer, D. 2003, “Cumulative advantage/disadvantage and the life course: Cross-fertilizing age and social science theory”, The Journals of Gerontology Series B: Psychological Sciences and Social Sciences, 58(6), S327–S337. Dannefer, D. 2020, “Systemic and reflexive: Foundations of cumulative dis/advantage and life-course processes”, The Journals of Gerontology Series B: Psychological Sciences and Social Sciences, 75(6), 1249–1263. Daran, B., and Levasseur, P. 2022, “Is overweight still a problem of rich in sub-Saharan Africa? Insights based on female-oriented demographic and health surveys”, World Development Perspectives, 25, 100388. Defo, B. K. 2014, “Demographic, epidemiological, and health transitions: Are they relevant to population health patterns in Africa?”, Global Health Action, 7(1), 22443. Di Cesare, M., Bentham, J., Stevens, G. A., et al. 2016, “Trends in adult body-mass index in 200 countries from 1975 to 2014: A pooled analysis of 1698 population-based measurement studies with 19.2 million participants”, Lancet, 387(10026), 1377–1396. DiPrete, T. A., and Eirich, G. M. 2006, “Cumulative advantage as a mechanism for inequality: A review of theoretical and empirical developments”, Annual Review of Sociology, 32, 271–297. DOI:10.1146/ annurev.soc.32.061604.123127. Drewnowski, A., and Popkin, B. M. 1997, “The nutrition transition: New trends in the global diet”, Nutrition Reviews, 55(2), 31–43. Elder Jr., G. H., and Shanahan, M. J. 2006, “The life course and human development”, in R. Lerner (Ed.), Handbook of Child Psychology, Volume 1, Theory, Hoboken, NJ: John Wiley & Sons, 665–715. Fischer, S., Hanefeld, M., Haffner, S.M., Fusch, C., Schwanebeck, U., Köhler, C., … and Julius, U. 2002, “Insulin-resistant patients with type 2 diabetes mellitus have higher serum leptin levels independently of body fat mass”, Acta diabetologica, 39(3), 105–110. Flegal, K. M., Kruszon-Moran, D., Carroll, M. D., Fryar, C. D., and Ogden, C. L. 2016, “Trends in obesity among adults in the United States, 2005 to 2014”, Jama, 315(21), 2284–2291. Food and Agriculture Organization of the United Nations (FAO) 2017, “Rural Africa in motion”, https://​ www​.fao​.org/​3/​i7951en/​I7951EN​.pdf Ford, N. D., Patel, S. A., and Venkat Narayan, K. M. 2017, “Obesity in low- and middle-income countries: Burden, drivers, and emerging challenges”, Annual Review of Public Health, 38, 145–164. https://​doi​.org/​10​.1146/​annurev​-publhealth​-031816​-044604 Francis, R. C. 2011, Epigenetics: How Environment Shapes Our Genes, New York: WW Norton & Company. Garrett, J. L., and Ruel, M. T. 2005, “Stunted child–overweight mother pairs: Prevalence and association with economic development and urbanization”, Food and Nutrition Bulletin, 26(2), 209–221. GBD Obesity Collaboration. 2014, “Global, regional and national prevalence of overweight and obesity in children and adults 1980–2013: A systematic analysis”, Lancet, 384(9945), 766–781. doi: 10.1016/ S0140-6736(14)60460-8 Gluckman, P., and Hanson, M. 2006, Mismatch: Why Our World No Longer Fits Our Bodies, Oxford, UK: Oxford University Press. Gluckman, P. D., Hanson, M. A., Buklijas, T., Low, F. M. and Beedle, A. S. 2009, “Epigenetic mechanisms that underpin metabolic and cardiovascular diseases”, Nature Reviews Endocrinology, 5(7), 401–408.

44  Handbook of health inequalities across the life course Godfrey, K. M., Reynolds, R. M., Prescott, S. L., Nyirenda, M., Jaddoe, V. W., Eriksson, J. G., and Broekman, B. F. 2017, “Influence of maternal obesity on the long-term health of offspring”, The Lancet Diabetes & Endocrinology, 5(1), 53–64. Gómez, M. I., and Ricketts, K. D. 2013, “Food value chain transformations in developing countries: Selected hypotheses on nutritional implications”, Food Policy, 42, 139–150. Haddok, E. 2012, “Biofuels land grab: Guatemala’s farmers lose plots and prosperity to ‘energy independence’”, Scientific American, https://​www​.scientificamerican​.com/​article/​biofuels​-land​-grab​ -guatemala/​ Hales, C. N., and Barker, D. J. 2001, “The thrifty phenotype hypothesis: Type 2 diabetes”, British Medical Bulletin, 60(1), 5–20. Hawkes, C. 2002, “Marketing activities of global soft drink and fast food companies in emerging markets: A review”, in Globalization, Diets and Noncommunicable Diseases. Geneva: World Health Organization, 2002. http://​whqlibdoc​.who​.int/​publications/​9241590416​.pdf Hawkes, C. 2005, “The role of foreign direct investment in the nutrition transition”, Public Health Nutrition, 8(4), 357–365. DOI: 10.1079/PHN2004706. Hoffman, D. J., Sawaya, A. L., Verreschi, I., Tucker, K. L., and Roberts, S. B. 2000, “Why are nutritionally stunted children at increased risk of obesity? Studies of metabolic rate and fat oxidation in shantytown children from Sao Paulo, Brazil”, American Journal of Clinical Nutrition, 72(3), 702–707. Jones-Smith, J. C., Gordon-Larsen, P., Siddiqi, A., and Popkin, B. M. 2011, “Cross-national comparisons of time trends in overweight inequality by socioeconomic status among women using repeated cross-sectional surveys from 37 developing countries, 1989–2007”, American Journal of Epidemiology, 173(6), 667–675. Kelley, J. A., and Thorpe Jr, R. J. 2021, “The interplay of age, period, and cohort effects on obesity and metabolic diseases in later life”, in Ferraro, K. and Carr, D. (Eds.), Handbook of Aging and the Social Sciences, London: Academic Press, 165–177. Kimani-Murage, E. W., Kahn, K., Pettifor, J. M., Tollman, S. M., Dunger, D. B., Gómez-Olivé, X. F., and Norris, S. A. 2010, “The prevalence of stunting, overweight and obesity, and metabolic disease risk in rural South African children”, BMC Public Health, 10(1), 1–13. Kinra, S., Baumer, J. H., and Smith, G. D. 2005, “Early growth and childhood obesity: A historical cohort study”, Archives of Disease in Childhood, 90(11), 1122–1127. Lawrence, G. 2017, “Re-evaluating food systems and food security: A global perspective”, Journal of Sociology, 53(4), 774–796. Lobstein, T., Jackson-Leach, R., Moodie, M. L., Hall, K. D., Gortmaker, S. L., Swinburn, B. A., and McPherson, K. 2015, “Child and adolescent obesity: Part of a bigger picture”, The Lancet, 385(9986), 2510–2520. Margulis, M. E., McKeon, N., and Borras Jr, S. M. 2013, “Land grabbing and global governance: Critical perspectives”, Globalizations, 10(1), 1–23. McMillen, I. C., and Robinson, J. S. 2005. “Developmental origins of the metabolic syndrome: Prediction, plasticity, and programming”, Physiological Reviews, 85(2), 571–633. Merton, R. K. 1988, “The Matthew Effect in science, II: Cumulative advantage and the symbolism of intellectual property”, Isis, 79, 606–623. DOI:10.1086/354848. Monteiro, C. A., Conde, W. L., and Popkin, B. M. 2007, “Income-specific trends in obesity in Brazil: 1975–2003”, American Journal of Public Health, 97(10), 1808–1812. Morrison, J. A., Friedman, L. A., and Gray-McGuire, C. 2007, “Metabolic syndrome in childhood predicts adult cardiovascular disease 25 years later: The Princeton Lipid Research Clinics follow-up study”, Pediatrics, 120340–120345. DOI: 10.1542/peds.2006-1699. Morrison, J. A., Friedman, L. A., Wang, P., and Glueck, C. J. 2008, “Metabolic syndrome in childhood predicts adult metabolic syndrome and type 2 diabetes mellitus 25 to 30 years later”, Journal of Pediatrics, 152, 201–206. DOI: 10.1016/j.jpeds.2007.09.010. Myrdal, G. 1944, An American Dilemma: The Negro Problem and Modern Democracy (Vol. I), New York: Harper and Row. NCD Risk Factor Collaboration (NCD-RisC) – Africa Working Group 2017, “Trends in obesity and diabetes across Africa from 1980 to 2014: An analysis of pooled population-based studies”, International Journal of Epidemiology, 46(5), 1421–1432. DOI:10.1093/ije/dyx078.

Interrogating the cohort–policy linkage  45 Ngaruiya, C., Hayward, A., Post, L., and Mowafi, H. 2017, “Obesity as a form of malnutrition: Over-nutrition on the Uganda ‘malnutrition’ agenda”, Pan African Medical Journal, 28(1). Niezen, C. 2013, “Farmers struggle for land in Guatemala”, Oxfam America. https://​www​.oxfamamerica​ .org/​explore/​stories/​farmers​-struggle​-for​-land​-in​-guatemala/​ O’Connor, J. 1986, Capital Accumulation, New York: Basil Blackwell. O’Sullivan, A., He, X., McNiven, E. M. S., Haggarty, N. W., Lönnerdal, B., and Slupsky, C. M. 2013, “Early diet impacts infant rhesus gut microbiome, immunity, and metabolism”, Journal of Proteome Research, 130523132331005 DOI:10.1021/pr4001702. Özsu, U. 2019, “Grabbing land legally: A Marxist analysis”, Leiden Journal of International Law, 32(2), 215–233. Peet, R. 2009, Unholy Trinity: The IMF, World Bank and WTO, London: Zed Books Ltd. Piketty, T., and Saez, E. 2014, “Inequality in the long run”, Science, 344(6186), 838–843. Piwoz, E. G., and Huffman, S. L. 2015, “The impact of marketing of breast-milk substitutes on WHO-recommended breastfeeding practices”, Food and Nutrition Bulletin, 36(4), 373–386. Popkin, B. M. 1993, “Nutritional patterns and transitions”, Population and Development Review, 19(1), 138–157. Popkin, B. M. 2002, “An overview of the nutrition transition and its health implications: The Bellagio Meeting”, Public Health Nutrition, 5(1A), 93–103. Popkin, B. M., and Reardon, T. 2018, “Obesity and the food system transformation in Latin America”, Obesity Reviews, 19(8), 1028–1064. Popkin, B. M., Adair, L. S., and Ng, S. W. 2012, “Global nutrition transition and the pandemic of obesity in developing countries”, Nutrition Reviews, 70(1), 3–21. Reyes Matos, U., Mesenburg, M. A., and Victora, C. G. 2020, “Socioeconomic inequalities in the prevalence of underweight, overweight, and obesity among women aged 20–49 in low- and middle-income countries”, International Journal of Obesity, 44(3), 609–616. Rigney, D. 2010, The Matthew Effect: How Advantage Begets Further Advantage, New York: Columbia. Rosenheck, R. 2008, “Fast food consumption and increased caloric intake: A systematic review of a trajectory towards weight gain and obesity risk”, Obesity Review, 9, 535–547. Sasson, T. 2016, “Milking the third world? Humanitarianism, capitalism, and the moral economy of the Nestlé boycott”, The American Historical Review, 121(4), 1196–1224. Sethi, S. P. 2012, Multinational Corporations and the Impact of Public Advocacy on Corporate Strategy: Nestle and the Infant Formula Controversy (Vol. 6), New York: Springer Science & Business Media. Smith, J. P., and Harvey, P. J. 2011, “Chronic disease and infant nutrition: Is it significant to public health?”, Public Health Nutrition, 14(2), 279–289. Swift, R. 1982, “Babies mean business”, The New Internationalist, 110(4). https://​newint​.org/​features/​ 1982/​04/​01/​babies/​ Traill, W. B., Mazzocchi, M., Shankar, B., and Hallam, D. 2014, “Importance of government policies and other influences in transforming global diets”, Nutrition Reviews, 72(9), 591–604. Wariri, O., Alhassan, J. A. K., Mark, G., Adesiyan, O., and Hanson, L. 2021, “Trends in obesity by socioeconomic status among non-pregnant women aged 15–49 years: A cross-sectional, multi-dimensional equity analysis of demographic and health surveys in 11 sub-Saharan Africa countries 1994–2015”, International Health, 13(5), 436–445. Wells, J. C. 2016, The Metabolic Ghetto: An Evolutionary Perspective on Nutrition, Power Relations and Chronic Disease, Cambridge: Cambridge University Press. Wingfield, A. H. 2019, Flatlining: Race, Work, and Health Care in the New Economy, Oakland, CA: University of California Press. Wojcicki, J. M. 2014, “The double burden household in sub-Saharan Africa: Maternal overweight and obesity and childhood undernutrition from the year 2000: Results from World Health Organization Data (WHO) and Demographic Health Surveys (DHS)”, BMC Public Health, 14(1), 1–12. World Health Organization. 2021, “Obesity and overweight”, https://​ www​ .who​ .int/​ news​ -room/​ fact​ -sheets/​detail/​obesity​-and​-overweight Wu, Y. E., Zhang, C. L., and Zhen, Q. 2016, “Metabolic syndrome in children”, Experimental and Therapeutic Medicine, 12(4), 2390–2394. Zewde, N., and Crystal, S. 2022, “Impact of the 2008 Recession on wealth adjusted income and inequality for U.S. cohorts”, Journal of Gerontology Social Sciences, 77(4). https://doi.org/10.1093/geronb/ gbab141

4. Economic theories of health inequality across the life course Titus J. Galama and Hans van Kippersluis

INTRODUCTION Socially and economically disadvantaged individuals generally experience poorer health outcomes (e.g., Glymour et al., 2014), and the Covid-19 pandemic risks aggravating these inequalities (e.g., Bambra et al., 2020). Health inequality is not only an infringement of equity (Woodward and Kawachi, 2000; Anand, 2002), but the higher prevalence of mortality and morbidity among lower socioeconomic groups also impedes productivity and threatens to undermine economic growth and prosperity (WHO, 2001; Bloom et al., 2019). Reducing health inequality is therefore high on the policy agenda in many countries. For example, the US Centers for Disease Control and Prevention (CDC) launches a Healthy People initiative every decade with the goal of improving the health of all Americans. However, despite numerous policy efforts, health inequalities – like income inequalities – have not decreased over time, and if anything, are increasing (Meara et al., 2008; Cantu et al., 2021; Case and Deaton, 2021). Apparently, we still lack a full understanding of how health inequalities arise, and the tools policymakers have to reduce disadvantage in socioeconomic and health outcomes appear to be insufficiently effective. Anne Case and Nobel-laureate Angus Deaton famously wrote: “it is extremely difficult to untangle the links between work, earnings, health, and education, without some sort of guiding framework” (Case and Deaton, 2005, p. 187). In this spirit, this chapter will explore what economic theory can offer in terms of improving our understanding of the potential causes of health inequalities over the life course. We will focus on health-capital theory based on the seminal work of Michael Grossman (1972a, 1972b) and later extensions by Galama (2015), Galama and van Kippersluis (2015, 2019) and Galama et al. (2018). The late Adam Wagstaff sketched the basic intuition of the economic approach to understanding health inequalities (Wagstaff, 1986). The basic pillars of the economic approach are: (i) people care about utility or “well-being”, which depends on both consumption as well as health, resulting in trade-offs that can help explain why people sometimes engage in unhealthy types of consumption even if this is detrimental to their health; (ii) health is produced on the basis of a production function where medical care is just one of its inputs; and (iii) there are time and budget constraints such that choices are restricted by the time and money available. Whereas the basic intuition of the economic approach is still highly relevant, empirical evidence over the last decades has challenged many key assumptions underlying the conventional economic model due to Grossman (1972a, 1972b). We highlight four examples here. First, while Wagstaff (1986) illustrates how income differences lead to health differences in the traditional theory, recent quasi-experimental studies provide evidence that changes in income do not appear to improve health (e.g., O’Donnell et al., 2015; Cesarini et al., 2016).1 Second, whereas educational attainment is assumed to increase the efficiency of medical care 46

Economic theories of health inequality across the life course  47 and thereby health (e.g., Grossman, 1972a, 1972b, 2000), and there exists ample empirical evidence for a strong association between education and health (e.g., Freedman and Martin, 1999; Cutler and Lleras-Muney, 2010), recent quasi-experimental studies have established that the causal effect of education on health is much smaller than the raw association suggests and seems to exist only in certain contexts and at certain levels of education (e.g., Galama et al., 2018; Savelyev, 2020; Barcellos et al., 2021; Fletcher and Noghanibehambari, 2021; Xue et al., 2021). A third example is the growing body of evidence on the developmental origins of health and disease (e.g., Barker, 1995; Almond et al., 2018), which the conventional model is silent about. In fact, Dalgaard et al. (2021) show how the traditional health capital model implicitly assumes that initial differences are depreciated away as individuals grow older, whereas empirical evidence suggests that disparities in health start very early in life and continue to widen till around age 60 (e.g., Case et al., 2002). Finally, ample evidence from psychology and behavioural economics has convincingly demonstrated that human rationality is bounded. For example, people overweight small probabilities and exhibit present bias and loss aversion (e.g., DellaVigna, 2009; Loewenstein et al., 2013). Moreover, preferences are partly shaped by social forces (the so-called “habitus”; Bourdieu, 1986). Hence, the rigid assumption of a rational, forward-looking decision-maker who makes choices in a social vacuum is obsolete. Empirical research in economics, and increasingly also in other social sciences and epidemiology, has moved away from traditional “controlling for observables” approaches – e.g., regression adjustment or matching on observables. This has many advantages. Most importantly, it avoids the very bold and unconvincing assumption that the variation in the exposure of interest is fully random after adjusting for a limited set of confounders. In these traditional approaches it is unclear where the variation in the exposure of interest comes from. Consider the example where one is interested in the effect of years of education on health. After controlling for a few confounders, is it really just random that one person has more years of education, or are we simply missing a key unobserved confounder (such as having highquality teachers, parental stimulation, etc.) that made this person go to school one year longer? From such methods, we will never know where the variation across individuals in years of education is coming from, making controlling-for-observables approaches fundamentally unreliable in establishing causal effects. The empirical revolution in health and economic research essentially launched a shift towards (natural) experiments where the source of variation in the exposure is clear, with recent Nobel prizes in economics awarded to Angrist, Card and Imbens (in 2021) for the use of natural experiments, and to Banerjee, Duflo and Kremer (in 2019) for the use of randomized controlled trials. For example, staying with the example of the causal effect of years of education on health, Clark and Royer (2013) exploited an arbitrary birth date cut-off of the Raising of the School Leaving Age (ROSLA) education reform that induced otherwise similar groups, born on either side of the cut-off, to attain different years of schooling (e.g., Clark and Royer, 2013). They found that the reform increased years of education, but did not affect health or health behaviour. The main advantage of this approach is that – given the apparently random nature of the difference in education – the likelihood of residual confounding is much lower. However, these methods are not without critique either (e.g., Deaton, 2010, 2020; Heckman and Urzua, 2010; Imbens, 2010, 2018; Deaton and Cartwright, 2018). Perhaps most importantly, there are major doubts about the external validity of the local treatment effects: “A result

48  Handbook of health inequalities across the life course that is true in one place, at one time, and under one set of circumstances, will typically not be true in another place, another time, or under different circumstances” (Deaton, 2020, p. 9). We believe the focus on natural experiments has brought more transparency to empirical research, and got us closer to establishing causal effects of specific marginal additions to education, or income. However, given the sometimes highly specific nature of natural experiments, in order to make substantial progress in our understanding of the causal effects of socioeconomic status (SES) on health, we need to triangulate various empirical research methods, not just limiting ourselves to an opportunistic quest for natural experiments. Moreover, in a world where economics is increasingly dominated by empirical research, it is more important than ever to have theory assist in explaining heterogeneity across the diverse study contexts that the natural experiments span. Theory can help researchers test specific mechanisms to better understand and enhance the external validity of specific empirical studies. In this review chapter, we will first sketch the model by Galama and van Kippersluis (2019), which is based on the seminal health-capital theory of Grossman (1972a, 1972b), and discuss its most important insights relevant for health inequality research. Next, we revisit some of the most important sources of critique of the model and sketch directions of how the theory could be extended to do justice to the recent empirical literature refuting some of the theoretical assumptions. We believe advancements in these directions are feasible and expect them to occur in the next few years.

HEALTH CAPITAL THEORY In Galama and van Kippersluis (2015, 2019) we developed a theory of disparities in health between SES groups, based on the seminal framework that Grossman (1972a, 1972b, 2000) developed. The theory is developed as a mathematical constrained optimization problem, where a so-called “utility” function is optimized subject to several (dynamic) constraints. By analysing the theory, one can derive “optimal” trajectories for certain choice variables, where optimal means that they provide the highest lifetime utility for a given set of (dynamic) constraints. In the theory, there are three periods of life, a schooling period up to age S, working life up to the retirement age R, and a retirement phase that runs until age T. In all phases of life, individuals maximize a so-called utility function S

 t 0

U Cth , Ctu , H t , pS 

1   t



R

 t S

U Cth , Ctu , H t 

1   t



T

 t R

U Cth , Ctu , H t 

1   t

(4.1)

where utility U  is provided by healthy consumption � Cth , unhealthy consumption Ctu and health H t . In simple terms, people care about their consumption and would like to enjoy good health. During schooling years, there is an additional element pS that captures the (dis)utility of schooling, where some people enjoy the school experience and others do not. β is the rate at which individuals discount future utility (capturing that we tend to care more about today than about some far away future). The optimal (in the sense of providing the highest lifetime utility) school-leaving age S � and retirement age R� are obviously influenced by prevailing policies

Economic theories of health inequality across the life course  49 like minimum school-leaving ages and statutory retirement ages, but are like length-of-life T � assumed to be choices made by the individual, subject to a set of constraints. The first of these is that health depreciates with age at the aging rate dt H t 1  H t   I I t  dt Cth , Ctu , zt , H t  (4.2)

The aging process dt can be countered through health investments I t at efficiency µ I , and the health investment production process is subject to decreasing returns to scale  0    1 , which addresses the degeneracy of linear investment models (Ehrlich and Chuma, 1990; Galama, 2015). It also captures an important economic concept, namely that of diminishing returns. In simple terms, people can invest in their health to counteract aging, but higher levels of investment, while still better, are less effective than are smaller levels: exercising 150 minutes a week is good for one’s health, while the additional gains from exercising >500 or >1000 minutes become smaller and smaller. Lifestyles and consumption patterns may affect the biological aging rate (Case and Deaton, 2005; see also Forster, 2001). We distinguish healthy consumption Cth (such as the consumption of healthy foods, sports and exercise) from unhealthy consumption Ctu (such as smoking, excessive alcohol consumption). Healthy consumption provides utility In the model, there are three additional constraints. One is the budget constraint, the second is the time constraint, and finally there is the longevity constraint. The budget constraint stipulates that financial assets increase with the interest rate and earned income (or a pension benefit in the retirement phase), which is a positive function of health. Assets decrease with expenditures on healthy and unhealthy consumption, as well as medical care (or other types of health investments, such as healthy behaviour). Over the life cycle one cannot spend more on consumption and health than one earns and inherits. The time constraint simply states that the total time in a day is divided into leisure and time inputs into health investment and consumption, as well as schooling or work depending on the phase of life. Finally, the longevity constraint imposes that health cannot be below a certain minimum threshold, below which life is no longer sustainable, and death is defined as the first moment health reaches this minimum threshold.

KEY INSIGHTS In Galama and van Kippersluis (2019), we perform comparative dynamic analyses to assess the characteristics of the model and generate empirically testable predictions. Here we summarize the key conceptual insights in an intuitive manner. (1) Health as one of the inputs into utility, leading to trade-offs. An important feature of economic theories of health is that the utility function (equation 4.1), which can loosely be interpreted as life satisfaction (e.g., Kahneman and Krueger, 2006), depends not just on health but also on consumption. While health and consumption preferences are often enhancing, in some cases they lead to trade-offs. Many people enjoy a glass of wine, chocolate and other types of consumption goods and services that are detrimental

50  Handbook of health inequalities across the life course to health (at least when consumed in larger quantities). Economic theories generally assume that individuals can perfectly navigate these trade-offs, leading to an optimal bundle of healthy and unhealthy consumption and an associated health level. A consequence of the assumption that any person decides on her optimal bundle of consumption and health is that economists are typically wary of government intervention. In fact, in contrast to the health sciences, economists are not necessarily interested in improving public health at all costs. Only when externalities exist (e.g., second-hand smoke), when individuals are expected to be misinformed or biased (as evidence from psychology and behavioural economics suggests), or when poverty traps exist, such as food deserts and poverty leading to unhealthy lifestyles because of lack of affordability, government intervention is recommended by economists. Obviously, the assumption that individuals make rational and optimal decisions is clearly wrong (see also below), but we argue that the assumption of rationality still provides a useful normative benchmark. It is also, at least to some extent, correct. When faced with new information about, say, the detrimental effects of smoking (e.g., the Surgeon General’s report of 1964), people do respond by quitting or reducing consumption, in particular those of higher SES (something economic theory would predict [see below point 2]). (2) The health cost of unhealthy consumption. The theory predicts a central role for a “health cost” of unhealthy behaviours. The choice as to whether to engage in unhealthy consumption or not is not only a function of the direct monetary cost (e.g., the price of a pack of cigarettes) but also of an indirect health cost. This health cost is the marginal value (in terms of life-time utility) of health lost due to detrimental health behaviours. This cost considers all future consequences of a current health behaviour. In Galama and van Kippersluis (2019) we find that the health cost increases with wealth (as well as with permanent income2 and education) and with the degree of unhealthiness of the good. This leads to the prediction that higher wealth increases demand for healthy and moderately unhealthy consumption goods, but decreases demand for severely unhealthy goods. Our theory may thus provide an economic rationale for the observation that wealthy, high-income and educated (permanent income) individuals are more likely to drink moderately, but less likely to drink heavily and smoke (Cutler and Lleras-Muney, 2010; van Kippersluis and Galama, 2014). Thus, apart from well-established cultural and social determinants of health behaviour, the concept of a health cost has further potential for explaining variation in health behaviours over the life cycle and across SES groups. (3) Wealth and the difference between absolute and marginal utility. The theory predicts that greater wealth, higher earnings and a higher level of education induce individuals to invest more in health, shift consumption towards healthy consumption, and enable individuals to afford healthier working environments. As a result, they are healthier and live longer. Intuitively, at high levels of wealth individuals enjoy a large basket of consumption. As a result, only limited marginal utility (U / Cth  0) or U / Ctu  0) is gained from yet more consumption. In simple terms, when you own three cars and you have a swimming pool in your backyard, the pleasure from yet another car or another swimming pool is limited. Thus, consuming more when consumption is already high provides relatively

Economic theories of health inequality across the life course  51 small gains in utility (  2U /  2 Cth  0 and  2U /  2 Ctu  0) . Once again, this is the economic notion (and assumption) of diminishing returns, in this case in the utility function. Still, the absolute level of utility U � is very high for high levels of consumption and a wealthy individual is thus very interested in prolonging the period over which this high level of utility is experienced. The difference between health and consumption is that health extends length of life, providing additional time during which consumption (two cars and a swimming pool) can be enjoyed. This leads wealthy individuals (and those with high SES more generally) to place a higher value on their health and invest more in it (Becker, 2007; Hall and Jones, 2007). The flipside of this statement is that lower SES individuals place a lower value on their health. The gains from more consumption are relatively higher for them than are the gains from life extension (i.e., compared to higher SES individuals). Note that the use of the term “value” is not a normative judgement, it is the result of lower SES individuals facing more stringent constraints (lower wealth, lower income, and often poorer health) within the model. The theory thus provides an explanation for the observation that higher SES individuals tend to lead healthier lives (as they can afford it and benefit more from it; see point 4 below). (4) The value of health and health inequalities over the life cycle. As high SES individuals place a higher value on their health, this increases the marginal benefits of healthy consumption, as well as the marginal costs of unhealthy working environments and unhealthy consumption. As a result, high SES individuals lead healthier lives, and this gradually leads to a health advantage with age. The more rapidly worsening health of low SES individuals (who engage more in unhealthy behaviour and in more physically/ psycho-socially demanding work) may lead to early withdrawal from the labour force and associated lost earnings, further widening the gradient in early- and mid-age. Jointly these behavioural choices gradually lead to growing health advantages for higher SES groups with age, similar to a process of cumulative advantage (e.g., Beckett, 2000; Dannefer, 2003; Lynch, 2003; DiPrete and Eirich, 2006). The model allows for a subsequent narrowing of the SES–health gradient, due to mortality selection but also because the marginal value of health grows when health declines. In simple terms, when health is low, it becomes a primary concern and the value of health increases. With low levels of health people care more about it. Hence, the theory predicts that low SES individuals after a certain age increase their health investment and improve their health behaviour as a result of their rapidly worsening health. This prediction relies on the assumption that health depreciation can be remedied through curative care and/or improved health behaviours, or at least deteriorate at a slower pace. What is interesting is that the prediction provides an economic behavioural interpretation of the age-as-leveller hypothesis (e.g., Dupre, 2007; see also Hoffmann, 2011; and Fritzell and Rehnberg, Chapter 20 in this volume). The theory thus allows for a number of stylized facts about the observed life-cycle patterns of the SES–health gradient, although empirical evidence will be needed to discriminate between competing explanations for the same phenomenon. (5) The importance of being able to influence longevity. Finally, a central prediction of the theory is that the ability of individuals to influence their longevity is crucial in explaining observed associations between SES and health. If life expectancy is fixed and exogenously given, associations between SES and health are small. If, however, life can be extended, SES and health are positively associated and the greater the degree

52  Handbook of health inequalities across the life course of life extension, the greater is their association. The intuition behind this result is that the horizon (life expectancy) is a crucial determinant of the return to investments in health. Investments in health lead to higher utility and make one more productive, but also importantly can boost life expectancy since mortality is defined as the first instance where health hits a minimum threshold. Now if life expectancy is (perceived to be) fixed, then investments in health will have much lower returns, and high SES individuals will be less tempted to invest in health. This suggests that in settings where it is (perceived as) difficult for wealthier, higher income and higher educated individuals to increase life expectancy (e.g., due to a high disease burden, competing risks, low efficiency of health investment, etc.), health disparities across socioeconomic groups would be smaller. Instead, in contexts where life expectancy is (perceived to be) to a large degree under the control of an individual (e.g., low disease burden, state-of-the-art medical technology) one would expect stronger disparities in health across SES groups (Galama and van Kippersluis, 2022). (6) Interpreting policy changes such as changes in compulsory schooling. As the utility function in equation (1) illustrates, economic models posit that individuals optimally choose their years of schooling, retirement age, and even length of life. While the latter is obviously a drastic simplification of reality, the idea that individuals consciously choose a certain level of education is less controversial. When accepting the notion that – without any government intervention – individuals would optimally select the years or level of education that is optimal for them (again, in the sense of deriving most lifetime utility from this choice), Galama et al. (2018) derive from this that imposing a minimum school-leaving age will therefore reduce utility levels for some. After all, if it is optimal to drop out of school at, say, age 16 for a certain individual who really dislikes school, has bad-quality teachers and would be better off entering the labour force, yet the government increases the minimum school-leaving age to 17, then this individual will be worse off in terms of lifetime utility. Linking it to the empirics, it is actually this group of people that would have liked to drop out of school but were forced to stay in school under the new rules (sometimes referred to as “compliers”) on the basis of which the “treatment effect” of raising a minimum school-leaving age is estimated. From a theoretical perspective, it is therefore perhaps not so surprising that most empirical studies of minimum school-leaving ages fail to detect meaningful effects on health. In fact, in line with this reasoning, Avendano et al. (2020) show limited and, if anything, negative effects on mental health between the ages of 16 and 70. This point applies more broadly than just to minimum school-leaving ages. In fact, any policy reform may interfere with optimal decisions by individuals, and the treatment effects identified on the basis of these, so-called, compliers affected by the policy reform may be quite different from the average treatment effects that we typically are interested in (see also Hoffmann and Doblhammer, Chapter 9 in this volume).

FUTURE EXTENSIONS The theory of health disparities we have sketched includes health investment, healthy and unhealthy consumption, job choices and longevity, and is capable of replicating a number of stylized facts regarding socioeconomic inequalities in health over the life cycle. Economic

Economic theories of health inequality across the life course  53 theory is typically stated as a mathematical optimization problem where individuals optimize a certain utility function under constraints, and is therefore essentially an analysis of the benefits and costs of a certain decision. We feel this systematic and formal exposition of a theory is a great tool to generate predictions and hypotheses about real-world behaviour. However, traditionally, too stringent – and frankly sometimes implausible – assumptions were historically imposed for mathematical convenience. As touched upon in the introduction, there are a number of empirical findings and regularities that call for further extensions and the loosening of the assumptions of the theory. In our view, these can be grouped into three broad directions: (i) an early childhood phase; (ii) bounded rationality; and (iii) social and contextual influences. First, the theory predicts that traditional socioeconomic advantages, like higher income, educational attainment and wealth, will lead to more health investment and eventually better health. However, despite a very strong association between socioeconomic status and health, recent studies that seek to estimate causal effects of income or education on health outcomes show only limited or no evidence for these theoretical hypotheses (e.g., Clark and Royer, 2013; Cesarini et al., 2016; Avendano et al., 2020). While sometimes methodological arguments are brought up,3 it seems fair to say that at least some, and arguably a substantial, portion of the socioeconomic health gradient is due to third factors influencing both SES and health. Examples of such third factors include time preferences (e.g., Fuchs, 1982) and cognitive and non-cognitive skills (e.g., Conti et al., 2010; Bijwaard et al., 2015; Strulik, 2018), characteristics which themselves are shaped by pre-natal and early-life factors (e.g., Barker, 1995), family socioeconomic status (e.g., Hart and Risley, 1995; Currie, 2009; Almond et al., 2018), and also genetic variants (e.g., Boardman et al., 2015; Diewald, Chapter 5 in this volume). The traditional Grossman model treats education as exogenously given. Galama and van Kippersluis (2022) allow individuals to invest in both education as well as health, but their model starts around age 16, where wealth and health endowments at that age are given. An important extension of the theory is therefore the inclusion of a distinct childhood phase of life. One might treat the child as passive, in the sense that parents make decisions regarding time and financial investments in their children, but the children themselves do not. The child’s own adult phase, where he/she makes conscious decisions, then is shaped by preferences and constraints, both of which are shaped by conditions at conception (e.g., genes, family SES), in the pre-natal period (e.g., maternal behaviour) and the early childhood phase (e.g., parental attention). Such a model would do justice to the overwhelming evidence for an influence of genetic and early-life factors implicated in both health and SES, and potentially account for inequalities in health across SES groups, which arise at very young ages (e.g., Case et al., 2002). A second promising direction is to move away from the assumption of rationality and allow for systematic and predictable irrationality in decision-making. There is ample evidence from psychology and behavioural economics that individuals are at most boundedly rational, and deviate from rationality in predictable ways. Examples include individuals disproportionately overweighting the present (a so-called present bias, e.g., Laibson, 1997; O’Donoghue and Rabin, 1999), overweighting small probabilities, disproportionately valuing certain over uncertain outcomes (e.g., Tversky and Kahneman, 1992; Prelec, 1998), and being subject to temptations, especially in stressful situations (e.g., Loewenstein and O’Donoghue, 2004). Unhealthy behaviours are a case in point. The benefits are immediate and certain: lighting up a cigarette or late-night snacking provides immediate gratification. But the costs are mostly in

54  Handbook of health inequalities across the life course the future and highly uncertain: a higher risk of developing disease and premature mortality some time in older age. Moreover, recent evidence suggests that “scarcity” in terms of money (or time) may impede the capacity to rationally calculate and weigh these future costs (e.g., Mani et al., 2013) and education and cognition influence the accuracy of longevity expectations (Bago d’Uva et al., 2020). Hence, SES clearly influences the extent to which individuals make rational decisions and there may be substantial pay-offs to incorporating deviations from rational risk and time preferences into the theory. A final promising extension is incorporating the role of social contexts, peer groups, and social and cultural capital into the theory. The traditional model is one of individual decision-making, where an individual optimally makes decisions in a social vacuum. However, empirically, it has been well established that decisions are shaped by social contexts, in particular peer effects and social expectations (e.g., Nakajima, 2007; Heckman et al., 2008; Cawley and Ruhm, 2011). There is therefore a need to incorporate social and cultural capital, along the lines of Bourdieu (1986), into health capital theory.

DISCUSSION While the scientific evidence on the drivers of the socioeconomic gradient in health has rapidly expanded, the mechanisms through which health differences across socioeconomic groups emerge and persist are still relatively poorly understood. Economic theory can help guide empirical studies in identifying mechanisms through which specific socioeconomic indicators and health interact. We believe that the frontier needs to be pushed both theoretically as well as empirically. Theoretically, the model could incorporate a childhood phase in which parents invest in the cognitive skills, non-cognitive skills and health of children, and where later-life preferences and constraints are shaped by this childhood phase and genetic and environmental differences. Empirically, we have to find a middle ground between internal and external validity, observational and (quasi-)experimental studies, where theory could help in bridging this gap.

ACKNOWLEDGEMENTS We would like to thank the volume editor Rasmus Hoffmann and anonymous reviewers for helpful comments. Research reported in this publication was supported the National Institute on Aging of the National Institutes of Health under Award Numbers RF1AG055654 and R56AG058726. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institute of Health. Hans van Kippersluis thanks the Erasmus Initiative Smarter Choices for Better Health for financial support. Titus Galama also thanks the Netherlands Organization of Scientific Research for financial support (NWO Vidi grant 016.Vidi.185.044) and is grateful to the School of Economics of Erasmus University Rotterdam for a Visiting Professorship in the Economics of Human Capital.

Economic theories of health inequality across the life course  55

NOTES 1. There is, however, a literature showing the beneficial effects of parental income on the child’s health (e.g., Akee et al., 2018), and most quasi-experimental studies on own income rely on one-off windfall shocks that are not necessarily representative for earned income effects on health (see discussion below on generalizability of quasi-experimental studies). 2. Permanent income is a measure of “life-time income”. If earnings are high over the duration of life, permanent income is high. 3. For example, some argue this could be due to a difference between the Local Average Treatment Effect (LATE) that natural experiments typically uncover versus the Average Treatment Effects on the Treated (ATT) that OLS seeks to estimate.

REFERENCES Akee, R., Copeland, W., Costello, E. J., & Simeonova, E. (2018). How does household income affect child personality traits and behaviors? American Economic Review, 108(3), 775–827. Almond, D., Currie, J., & Duque, V. (2018). Childhood circumstances and adult outcomes: Act II. Journal of Economic Literature, 56(4), 1360–1446. Anand, S. (2002). The concern for equity in health. Journal of Epidemiology and Community Health, 56(7), 485. Avendano, M., De Coulon, A., & Nafilyan, V. (2020). Does longer compulsory schooling affect mental health? Evidence from a British reform. Journal of Public Economics, 183, 104137. Bago d’Uva, T., O’Donnell, O., & Van Doorslaer, E. (2020). Who can predict their own demise? Heterogeneity in the accuracy and value of longevity expectations. Journal of the Economics of Ageing, 17, 100135. Bambra, C., Riordan, R., Ford, J., & Matthews, F. (2020). The COVID-19 pandemic and health inequalities. J Epidemiol Community Health, 74(11), 964–968. Barcellos, S. H., Carvalho, L. S., & Turley, P. (2021). Distributional effects of education on health. Journal of Human Resources, 0720-11064R1. Barker, D. J. P. (1995). Fetal origins of coronary heart disease. British Medical Journal, 311, 171–4. Becker, G. S. (2007). Health as human capital: synthesis and extensions. Oxford Economic Papers, 59(3), 379–410. Beckett, M. (2000). Converging health inequalities in later life: an artifact of mortality selection? Journal of Health and Social Behavior, 41, 106–119. Bijwaard, G. E., van Kippersluis, H., & Veenman, J. (2015). Education and health: the role of cognitive ability. Journal of Health Economics, 42, 29–43. Bloom, D. E., Canning, D., Kotschy, R., Prettner, K., & Schünemann, J. J. (2019). Health and Economic Growth: Reconciling the Micro and Macro Evidence (No. w26003). National Bureau of Economic Research. Boardman, J. D., Domingue, B. W., & Daw, J. (2015). What can genes tell us about the relationship between education and health? Social Science & Medicine, 127, 171–180. Bourdieu, P. (1986). The forms of capital. Cultural Theory: An Anthology, 1, 81–93. Cantu, P. A., Sheehan, C. M., Sasson, I., & Hayward, M. D. (2021). Increasing education-based disparities in healthy life expectancy among US non-Hispanic whites, 2000–2010. Journals of Gerontology: Series B, 76(2), 319–329. Case, A., & Deaton, A. (2005). Broken down by work and sex: how our health declines. In: David A. Wise (ed.), Analyses in the Economics of Aging, The University of Chicago Press, Chicago, 185–212. Case, A., & Deaton, A. (2021). The Great Divide: Education, Despair and Death (No. w29241). National Bureau of Economic Research. Case, A., Lubotsky, D., & Paxson, C. (2002). Economic status and health in childhood: the origins of the gradient. American Economic Review, 92, 1308–1334. Cawley, J., & Ruhm, C. J. (2011). The economics of risky health behaviors. In Handbook of Health Economics, Elsevier, Amsterdam, 95–199.

56  Handbook of health inequalities across the life course Cesarini, D., Lindqvist, E., Östling, R., & Wallace, B. (2016). Wealth, health, and child development: evidence from administrative data on Swedish lottery players. Quarterly Journal of Economics, 131(2), 687–738. Clark, D., & H. Royer (2013). The effect of education on adult mortality and health: evidence from Britain. American Economic Review, 103(6), 2087–2120. Conti, G., Heckman, J. J., & Urzua, S. (2010), The education-health gradient. American Economic Review Papers and Proceedings, 100(2), 234–238. Currie, J. (2009). Healthy, wealthy, and wise: socioeconomic status, poor health in childhood, and human capital development. Journal of Economic Literature, 47(1), 87–122. Cutler, D. M., & Lleras-Muney, A. (2010), Understanding differences in health behaviors by education. Journal of Health Economics, 29(1), 1–28. Dalgaard, C. J., & Strulik, H. (2014). Optimal aging and death: understanding the Preston curve. Journal of the European Economic Association, 12(3), 672–701. Dalgaard, C. J., Hansen, C. W., & Strulik, H. (2021). Fetal origins – a life cycle model of health and aging from conception to death. Health Economics, 30(6), 1276–1290. Dannefer, D. (2003). Cumulative advantage/disadvantage and the life course: cross-fertilizing age and social science theory. Journals of Gerontology Series B: Psychological Sciences and Social Sciences, 58(6), S327–S337. Deaton, A. (2010). Instruments, randomization, and learning about development. Journal of Economic Literature, 48(2), 424–455. Deaton, A. (2020). Randomization in the Tropics Revisited: A Theme and Eleven Variations (No. w27600). National Bureau of Economic Research. Deaton, A., & Cartwright, N. (2018). Understanding and misunderstanding randomized controlled trials. Social Science & Medicine, 210, 2–21. DellaVigna, S. (2009). Psychology and economics: evidence from the field. Journal of Economic literature, 47(2), 315–372. DiPrete, T. A., & Eirich, G. M. (2006). Cumulative advantage as a mechanism for inequality: a review of theoretical and empirical developments. Annu. Rev. Sociol., 32, 271–297. Dupre, M. E. (2007). Educational differences in age-related patterns of disease: reconsidering the cumulative disadvantage and age-as-leveler hypotheses. Journal of Health and Social behavior, 48(1), 1–15. Ehrlich, I., & Chuma, H. (1990), A model of the demand for longevity and the value of life extension. Journal of Political Economy, 98(4), 761–782. Fletcher, J., & Noghanibehambari, H. (2021). The Effects of Education on Mortality: Evidence Using College Expansions (No. w29423). National Bureau of Economic Research. Forster, M. (2001). The meaning of death: some simulations of a model of healthy and unhealthy consumption. Journal of Health Economics, 20(4), 613–638. Freedman, V. A., & Martin, L. G. (1999). The role of education in explaining and forecasting trends in functional limitations among older Americans. Demography, 36(4), 461–473. Fuchs, V. R. (1982). Time preference and health: an exploratory study. Economic Aspects of Health, 93, 120. Galama, T. J. (2015). A contribution to health capital theory. Human Capital and Economic Opportunity Global Working Group Working Paper No. 2015-008. Galama, T., & van Kippersluis, H. (2015). A Theory of Education and Health. Tinbergen Institute Discussion Paper (No. TI 15-031/V). Galama, T. J., & van Kippersluis, H. (2019). A theory of socioeconomic disparities in health. Economic Journal, 129(617), 338–374. Galama, T. J., & van Kippersluis, H. (2022). Human capital formation: the importance of endogenous longevity. Human Capital and Economic Opportunity Global Working Group Working Paper No. 2022-009. Galama, T. J., Lleras-Muney, A. and van Kippersluis, H. (2018). The effect of education on health and mortality: a review of experimental and quasi-experimental evidence. In Oxford Research Encyclopedia of Economics and Finance. Glymour, M. M., Avendano, M., & Kawachi, I. (2014). Socioeconomic status and health. Social Epidemiology, 2, 17–63.

Economic theories of health inequality across the life course  57 Grossman, M. (1972a). The demand for health – a theoretical and empirical investigation’. National Bureau of Economic Research, New York. Grossman, M. (1972b). On the concept of health capital and the demand for health. Journal of Political Economy, 80(2), 223–255. Grossman, M. (2000). The human capital model. In: A. J. Culyer and J. P. Newhouse (eds.), Handbook of Health Economics, Volume 1, Amsterdam, Elsevier Science, 347–408. Hall, R. E., & Jones, C. I. (2007). The value of life and the rise in health spending. Quarterly Journal of Economics, 122(1), 39–72. Hart, B., & Risley, T. R. (1995). Meaningful Differences in the Everyday Experience of Young American Children, Paul H Brookes Publishing, Baltimore, MD. Healthy People 2020, Secretary’s Advisory Committee on Health Promotion and Disease Prevention Objectives for 2020. Healthy People 2020: An Opportunity to Address the Societal Determinants of Health in the United States, July 26, 2010. Heckman, J. J., Flyer, F., & Loughlin, C. (2008). An assessment of causal inference in smoking initiation research and a framework for future research. Economic Inquiry, 46(1), 37–44. Heckman, J. J., & Urzua, S. (2010). Comparing IV with structural models: what simple IV can and cannot identify. Journal of Econometrics, 156(1), 27–37. Hoffmann, R. (2011). Illness, not age, is the leveler of social mortality differences in old age. Journals of Gerontology Series B: Psychological Sciences and Social Sciences, 66(3), 374–379. Imbens, G. W. (2010). Better LATE than nothing: some comments on Deaton (2009) and Heckman and Urzua (2009). Journal of Economic Literature, 48(2), 399–423. Imbens, G. W. (2018). Comments On: Understanding and Misunderstanding Randomized Controlled Trails by Cartwright and Deaton (No. 3648). Kahneman, D., & Krueger, A. B. (2006). Developments in the measurement of subjective well-being. Journal of Economic Perspectives, 20(1), 3–24. Laibson, D. (1997). Golden eggs and hyperbolic discounting. Quarterly Journal of Economics, 112(2), 443–478. Loewenstein, G., Asch, D. A., & Volpp, K. G. (2013). Behavioral economics holds potential to deliver better results for patients, insurers, and employers. Health Affairs, 32(7), 1244–1250. Loewenstein, G. F. & O’Donoghue, T. (2004). Animal spirits: affective and deliberative processes in economic behavior. Available at SSRN: https://​ssrn​.com/​abstract​=​539843 Lynch, S.M. (2003). Cohort and life-course patterns in the relationship between education and health: a hierarchical approach. Demography, 40, 309–331. Mani, A., Mullainathan, S., Shafir, E., & Zhao, J. (2013). Poverty impedes cognitive function. Science, 341(6149), 976–980. Meara, E. R., Richards, S. & Cutler, D. (2008). The gap gets bigger: changes in mortality and life expectancy, by education, 1981–2000. Health Affairs 27(2), 350–360. Nakajima, R. (2007). Measuring peer effects on youth smoking behaviour. Review of Economic Studies, 74(3), 897–935. O’Donnell, O., Van Doorslaer, E., & Van Ourti, T. (2015). Health and inequality. In: A. B. Atkinson & F. J. Bourguigson (eds), Handbook of Income Distribution (Vol. 2). Elsevier, North Holland, 1419–1533. O’Donoghue, T., & Rabin, M. (1999). Doing it now or later. American Economic Review, 89(1), 103–124. Prelec, D. (1998). The probability weighting function. Econometrica, 497–527. Savelyev, P. A. (2020). Conscientiousness, extraversion, college education, and longevity of high-ability individuals. Journal of Human Resources, 0918-9720R2. Strulik, H. (2018). The return to education in terms of wealth and health. Journal of the Economics of Ageing, 12, 1–14. Tversky, A., & Kahneman, D. (1992). Advances in prospect theory: cumulative representation of uncertainty. Journal of Risk and Uncertainty, 5(4), 297–323. Van Kippersluis, H. and Galama, T. J. (2014). Why the rich drink more but smoke less: the impact of wealth on health behaviors. European Economic Review, 72, 197–220. Wagstaff, A. (1986). The demand for health: theory and applications. Journal of Epidemiology & Community Health, 40(1), 1–11.

58  Handbook of health inequalities across the life course Woodward, A., & Kawachi, I. (2000). Why reduce health inequalities? Journal of Epidemiology & Community Health, 54(12), 923–929. World Health Organization (WHO) (2001). Macroeconomics and Health: Investing in Health for Economic Development. Report of the Commission on Macroeconomics and Health, World Health Organization, Geneva. Xue, X., Cheng, M., & Zhang, W. (2021). Does education really improve health? A meta-analysis. Journal of Economic Surveys, 35(1), 71–105.

5. Health as a consequence of genetic variation, gene transcription and life course experiences Martin Diewald

INTRODUCTION Two strands in health-related research have proliferated in recent years but developed largely independently from each other. The first of these concerns the increased interest in health development as part of the life course and the role of life course theoretical concepts to explain the relevance of specific life phases and the accumulation of life experiences for health development (Kuh and Ben Shlomo, 2004; Beckfield, Olafsdottir and Bakhtiari, 2013). Longitudinal perspectives are especially important for the investigation of the social determinants of health and how these translate into social inequalities for health, as they can account for both selection and causation processes over the life course (Hoffmann, Kröger and Pakpahan, 2018). Theoretical concepts of how experiences accumulate over the life course have been increasingly adopted in health research (Halfon, Forrest and Lerner, 2018). The second strand of public health research that has proliferated in recent years concerns the relevance of genes, which has been largely driven by the rapid development of molecular genetics and epigenetics. This, in turn, led to increased optimism, which provided hopes that we could predict health outcomes simply by deciphering the human genome. Although these hopes were largely exaggerated, genetic information derived from twin-based studies, molecular genetics, and gene transcription have helped disentangle health determinants related to genetic and social origin respectively. This provides us with two starting points, both being the legacies inherited from our parents, for the understanding of lifelong health trajectories. Common for both genetic and social determinants is that they should not be understood as deterministic, rather they define the upper and lower bounds of our health potentials. Exceptions only concern very few rare diseases, like Tay-Sachs disease, which is inevitable if an individual carries two copies of the mutation in the HEXA gene. For the vast majority of health-related outcomes, our genes contribute with their effects in addition to their interaction with environmental conditions and social experiences. Next to the development of molecular genetics, which ended with the completion of the Human Genome Project in 2003, the so-called post-genomic era emerged (Perbal, 2015). In this new era, genes were conceptualized as DNA that remains more or less fixed over the individual life course. Identifications of genetic influences were mainly based on evidence stemming from twin-based research. This research was able to separate the overall variance of a phenotypic trait, such as health, in latent black-box components representing the whole of additive genetic as well as the whole of environmental sources, some of which make twins more similar to each other (so-called shared environment), while others drive them apart in their development (so-called nonshared environment; Kohler, Behrman and Schnittker, 2011). Molecular genetic research relies on Mendelian segregation of genes to predict disease risks and health states mostly in the form of polygenic risk scores, which in turn are based on 59

60  Handbook of health inequalities across the life course associations between a certain phenotype and many single nucleotide polymorphisms (SNPs) (Choi, Mak and O’Reilly, 2020). The landmark of the so-called post-genomic era is that genes are no longer understood as fixed but are reconceptualized as “fluid entities of genome expression” (Turner et al., 2020, p. 4). This means less dependence on the underlying sequence and increased dependence on the environment experienced over the life course, resulting in epigenetic changes. In the following, I review how the fixed and fluid concept of genes can be fruitfully linked to explaining the etiology of health and illness over the life course. It is beyond the scope of this contribution to touch upon genetic contributions to all facets of health and all single diseases. Rather, I will focus on general approaches to studying the mechanisms over which genetic variation influences health, and how genetics can enrich a life course view on health development. First, in the following section, I will distinguish different pathways as to how genes may influence health and disease related to specific health outcomes. This will include: (i) direct effects of genes; (ii) effects through environmental susceptibility; and (iii) effects through healthy behaviors. Second, I will provide an overview of the heritability of health outcomes by discussing the relevance of how the genetic origin of health and disease is operationalized. I will then focus on the question of how research on genetic influences and research on health development over the life course can profit from each other. The fourth section starts from the biological side of the interplay between genes and the environment by focusing on changes in the epigenome and asks how stability and change in the epigenome can enrich life course studies on health. Vice versa the following section refers to several theoretical concepts of how gene–environment correlation and interaction play a role over the life course. The concluding section summarizes the state-of-the-art and discusses prospects for future research.

PATHWAYS OF HOW GENES INFLUENCE HEALTH In twin-based methodology, the contributions of genetic variation to explaining a health state is per se a black box. We only estimate an overall additive influence of genetic variation. With molecular genetic methods it was shown that with very few exceptions for rare diseases, there is no single gene accounting for somatic or mental health or specific diseases. Rather, there have been dozens, hundreds, or even more genetic variants identified which are somehow associated with health outcomes. Also in molecular genetics, with polygenic scores (PGS) capturing the overall genetic influence on a trait, it is widely unknown what exact mechanisms create an association of such a score with the health outcome under investigation, since we have been able to identify the biological mechanisms for only parts of the genetic variants underpinning these scores. Generally, genetic variants relevant for health outcomes may include: (i) direct effects of genes on specific health outcomes via influencing specific biological mechanisms that create health and disease; (ii) effects through genetically influenced environmental susceptibility; and (iii) effects through genetically influenced healthy behaviors. In the first case, genetic variants influence malfunctions of the organism leading to specific diseases, for example in brain functioning, blood composition, hormone production, or inflammatory processes. Knowledge of such genetic variants allows for individual precision medicine based on the identification of genetic risk for a disease as well as an appropriate, specific medication (Alzu’bi, Zhou and Watzlaf, 2019).

Genetic variation, gene transcription and life course experiences  61 In the second case, genetic variants are relevant to how our organism and our behavior react to environmental opportunities and challenges. In other words, there are genes that can influence how our organism and our behavior react to environmental opportunities and challenges. Regarding the organism, a genetic propensity to develop specific mental disorders or somatic diseases may be triggered by environmental strains more often if there is at the same time a general neurobiological vulnerability to strain exposure, e.g., traumatic experiences, stressful life events, or daily hassles (Nudel et al., 2019). Such a genetic susceptibility was also identified for a Covid-19 infection (SeyedAlinaghi et al., 2021). This is the genetic contribution to the well-known vulnerability concept. In other words, whereas some people are resilient against negative effects in highly stressful circumstances, for others such circumstances may lead to a breakout of a hitherto latent disease or an exacerbation of an already existing one with less severe symptoms. However, there is growing evidence for a general neurobiological sensitivity to the social and physical environment, i.e. there are people who are generally more environmentally susceptible, for better or worse. On the one hand, they exhibit under adverse conditions more maladaptive outcomes, and, on the other hand, they profit disproportionally from supportive, predictable and safe environments (Ellis et al., 2011; Mitchell et al., 2014). Thus, susceptible genetic profiles are an important component for the identification of genetic risk and genetic potential in the interplay of genetic variation with the social environment over the life course (Kim-Cohen et al., 2004). A third genetic contribution to health and disease is the genetically influenced ability to exhibit healthy behaviors. Especially for health states and diseases that require compliance with healthy behaviors, cognitive ability and personality traits can prevent both morbidity and premature mortality. Among these, cognitive ability seems to be the most important one (Gottfredson and Deary, 2004), and it is at the same time the most heritable characteristic, ranging from 20 to 80 percent depending on age (Polderman et al., 2015). Concerning personality, conscientiousness is most helpful, whereas neuroticism is a risk factor for healthy behaviors (Bogg and Roberts, 2004; Weiss and Costa, 2005; Shanahan et al., 2014). These characteristics may underlie the often observed phenotypic correlation of educational attainment with healthy behaviors as well as genetic correlations of educational attainment with health outcomes (Harden and Koellinger, 2020, p. 3). In sum, one has to be aware of all three pathways when talking about genetic sources of health and disease. The overall genetic component in predicting health-related phenotypes may consist of different genetic variants related to only one of these pathways, or the same genetic variants may be relevant for several pathways, due to the usually pleiotropic character of genes, i.e., genes influence two or more unrelated phenotypes. To understand the interdependencies between different phenotypic as well as underlying genetic pathways, twin-based studies have developed models where heritability is estimated not only for a single trait but also regarding the extent to which the same genes (unobserved, as black boxes) influence different pathways of how genes become relevant for health-related outcomes. The same is possible with GWAS data, and in this case, the single genetic variants are in principle identifiable. However, as Harden and Koellinger (2020, p. 569) warn, … it is important to remember that genetic correlations are not, by themselves, informative about causal mechanisms, nor do they necessarily imply direct, “inside the skin” pleiotropic effects of genes on two traits. They may also reflect indirect, possibly environmentally mediated pathways (for

62  Handbook of health inequalities across the life course example, high childhood IQ → higher educational attainment → less smoking → reduced risk for lung cancer).

It is apparent from the distinction between these pathways that genes and environment do influence health not only separately, in an additive way, but that there may also be an interplay at work between them. Generally, two forms of gene–environment interplay can be distinguished: gene–environment interaction (GxE) and gene–environment correlation (rGE). GxE means that a genetic predisposition to health-related outcomes is moderated by environmental influences (e.g., Flowers, Froelicher and Aouizerat, 2012), e.g., the present positioning in the social inequality structure (education, occupation, money), or the position during earlier sensitive phases. Or the other way around, that environmental influences are moderated by a genetic predisposition that contributes to individual characteristics influencing how the same situation might be perceived and evaluated differently, e.g. that the detrimental effects of disadvantaged life course circumstances for health and aging may be further exacerbated among persons with a genetic predisposition to coronary artery disease (Liu et al., 2019). The analysis of GxE is convincing only if it is evident that both are exogenous to each other, or in other words, that there is no rGE. However, environments may be conflated with genetic predispositions. For example, the availability of helpful networks may be conflated by genes that are relevant for the ability to construct support networks and seek help (active rGE), or by genes that are relevant for evoking helpful behaviors from others (passive rGE).

HERITABILITY ESTIMATES OF HEALTH: AN OVERVIEW Despite the variety of ways in which genes and the environment interact in the genesis of health and disease, the identification of the relative contribution of genetic variation to the variation in a phenotype remains theoretically and methodologically an important point of departure. Health-related characteristics are overall more heritable than most other complex traits. This applies to somatic as well as mental health characteristics (MacGregor et al., 2000; Harden, 2021). Whereas we find in twin-based studies an average of about 50 percent for genes and environment for all traits ever investigated, among health-related traits we find a high heritability for ophthalmological traits (heritability of 71%), followed by diseases of ear, nose and throat (64%) and dermatological (60%), skeletal (59%), metabolic (58%) and respiratory (55%) symptoms. Around average are immunological (50%), hematological (50%) and neurological (50%) symptoms, with psychiatric (46%), nutritional (44%), cardiovascular (44%), endocrinological (40%) and gastrointestinal (35%) symptoms below average (Polderman et al., 2015; supplementary table 16). When reading such heritabilities, it is important to understand what such across-populations averages mean and what they do not mean. First, it has to be noted that due to methodological reasons these are all upper-bound estimates since GxE and rGE are mostly attributed to the genetic component, if not explicitly taken into consideration (see also below). Second, these estimates are not all-time, all-place transcendent truths. Instead, they are highly dependent on the populations studied and on the selectivity of the samples used. Heritabilities can be different for men and women, ethnicities, ages and societies across time and space. This is not a weakness but a strength in estimating genetic influences, since it highlights the importance of social conditions for the relevance of genetic variation. Third, genes are almost never deterministic. Rather, they indicate a certain

Genetic variation, gene transcription and life course experiences  63 proneness to develop a health outcome. Fourth, heritability estimates based on variance decomposition into a genetic and a shared and unshared environmental component assume that there is no rGE or GxE. Of course, this conflicts with the attention paid to gene–environment interplay. Especially in older age, heritability comprises to some degree also environmental influences due to active and passive rGE, in complex transactions between genetic variation and environmental experiences over time (Briley et al., 2019). Nevertheless, twin-based heritability estimates have proved to be more realistic upper bound estimates than numerous critics believe (e.g., Harden, 2021), whereas molecular genetic analyses are still far away from capturing the overall relevance of genetic variation. Social origin as a traditional concept for the starting point of individual life courses receives a lot of attention in health-related research. For a life course perspective on health development, heritability estimates of health and diseases have a place in adding genetic origin to social origin, and in correcting social origin by genetic origin to the degree to which social and genetic origin overlap (Diewald et al., 2015). Thus, if the genetic origin is not included in studies on the relevance of social origin for health development, estimates of social origin influences are biased because conflated by genetic variation to an unknown degree.

ENRICHMENT OF LIFE COURSE STUDIES ON HEALTH BY EPIGENETIC INFORMATION Genes do not only add to social conditions in explaining health development. Their influence may also be dependent on living conditions experienced from pregnancy through all life periods thereafter. Whereas genomic research deals with genes in the form of the DNA being fixed since conception, gene expression is a more “fluid” or dynamic concept “where segments of the DNA function differently depending on physiological or environmental contexts” (Turner et al., 2020 p. 3). Such epigenetic regulation converts genomic data into exploitable information for the organism. In other words, gene transcription is temporarily or permanently upregulated or downregulated, with the genomic sequence remaining unchanged. Epigenetic regulation leads to a structural adaptation of chromosomal regions altering the functioning of the genome as a reaction to social experiences (ibid.). Such encoding of experiences in the epigenome starts with conception, with early developmental experiences and exposures becoming neurobiologically instantiated in the brain, thereby affecting trajectories of health and other life domains for the remainder of life (Boyce and Hertzman, 2018), e.g. for the comorbidity of cardiovascular disease and cognitive impairment (Vidrascu et al., 2019). Epigenetic modifications comprise very different patterns, ranging from non-functional or unclarified modifications to the immediate outbreak of diseases. They may also stay latent for many years until environmental stimulation leads to the outbreak of the formerly encoded disease risk to an actual disease (Barker, 2004). Changes in the epigenome are reversible but can (if stabilized) last for one’s whole life and even be inherited into the next generation. They emerge most likely in the sensitive periods of life, in which the organism is most susceptible to environmental exposure, implying the very early development from conception to early childhood, and puberty. The emergence of epigenetic changes as well as possible consequences comprise both health-related processes and outcomes, and they are closely linked to social inequality. The emergence of epigenetic changes can be understood as experiences of stress that get embedded “under the skin”.

64  Handbook of health inequalities across the life course Epigenetic marks can thereby be considered as a biological “memory” of such experiences. Traumatic experiences and mental problems (Vinkers et al., 2015), malnutrition (Choi and Friso, 2010), smoking (Zakarya, Adcock and Oliver, 2019), own and mother’s alcohol consumption in the womb (Simpkin et al., 2016; Pandey, Kyzar and Zhang, 2017), and severe illness belong to experiences than can trigger epigenetic changes and in the following years can alter mental states as well as somatic processes with lifelong consequences for health. These experiences, as well as the probability that they lead to changes in the epigenome, are closely linked to social inequality. It has long been recognized in life course research as well as in medical sociology that people belonging to higher socioeconomic groups are not necessarily less exposed (Schieman, Whitestone and Van Gundy, 2006) but are better equipped to deal with stress and risks, whereas more deprived people (for example unemployed) are more exposed to poverty, stress, stigma and social deprivation. Consequently, stressful life experiences lead less often to epigenetic changes when buffer mechanisms provided by supportive resources are in play (Miller and Chen, 2010; Miller et al., 2018). The embodiment of early stress in the epigenome is crucial for understanding why stressful experiences, especially at early ages, make individuals vulnerable to adult adversity. This includes a higher probability of diseases, but at the same time social adversity, even if no additional meaningful stressors are experienced after this initial period of life. Predominantly immune-mediated proinflammatory processes seem responsible for this linkage (McGuinness et al., 2012; Ziol-Guest et al., 2012). For the study of health over the life course four lessons follow from epigenetic mechanisms: (1) The embodiment of stressful experiences in the epigenome is closely linked to social inequality over the life course, including social advantage by high material resources, social security systems, social respect and helpful social networks, as well as social disadvantages like unemployment, welfare state dependency or low income. Thus, epigenetic changes provide important explanations for the health trajectory over the life course, including how health takes different pathways between social groups. (2) Epigenetic changes happen especially when stress experiences are rooted in sensitive phases, i.e. in utero, in early childhood or puberty. If encoded in the epigenome, experiences can result in health problems and diminished life chances decades later in adulthood, even if similar life experiences in between have no such effect. However, the epigenetic encoding of psychic and social stress is not limited to early sensitive phases but can happen also in other life phases including adulthood (Meaney, 2010; van Dongen et al., 2016; Vick and Burris, 2017). (3) Because the experience of stress, as well as the possibilities to buffer against stress, depend largely on socioeconomic status, the genesis of epigenetic changes plays a role in explaining the link between social inequality and health over the life course. Socioeconomic status is closely linked to different levels of epigenetic imprint (McGuinness et al., 2012). This is not only due to the unequal distribution of skill, education, money, occupation, and social recognition mentioned above. Moreover, several risks relevant to epigenetic changes mentioned above, like malnutrition, drug use, alcohol, smoking, drug use, smoking of the mother during pregnancy, and also pollution are unequally distributed across socioeconomic strata (Vick and Burris, 2017; Turner, 2018). (4) In consequence, health-related consequences of social inequality need to be understood not simply as a result of unequal amounts of material and non-material resources alone,

Genetic variation, gene transcription and life course experiences  65 as it is often operationalized, but as resulting from different bundles of resources and demands, the latter being more closely linked to stressful experiences. This approach is well established in the inducements and contributions theory in economics (March and Simon, 1958), the job resources and demands model in research on work–life balance (Bakker and Demerouti, 2007), and the “stress of higher status” hypothesis (Schieman, Whitestone and Van Gundy, 2006) challenging the unquestioned equalization of low status with high pressures in the sociological literature on labor market inequalities. Only in this way can we fully comprehend the linkages between epigenetics and social inequality. From a life course perspective, intragenerational mobility, as well as intergenerational mobility, are part of the role of epigenetics in the health and social inequality puzzle. Biological aging is one of the major risk factors for many chronic diseases and therefore a key concept for assessing the role of epigenetic encoding for health development as a whole, over and above single diseases. One of the most widespread examples in epigenetics related to the development of individual health relates to the so-called “epigenetic clock”. All in all, about 40 percent of the variation between individuals in the speed of their epigenetic clock seems genetically determined (Bell et al., 2019; Ryan, 2021). This is why trajectories of epigenetic aging tend to emerge early in life and that they remain relatively stable across the life course (Li et al., 2020). At the same time, this implies that about 60 percent of the variance in the pace of biological aging can be traced back to environmental conditions and individual life course experiences over a lifetime. There is more than one epigenetic clock in the literature. These emphasize partly different drivers and regulators of age-related changes in single-cell, tissue- and disease-specific models in the measurement of biological aging (Bell et al., 2019). Although they differ in many such aspects, a comparative analysis of 11 existing epigenetic clocks has shown that they share core signals that are relevant to the biology of aging (Liu et al., 2020), which provides support for epigenetic clocks serving as a promising biomarker of aging (Ryan, 2021). Their power to estimate an individual’s age is remarkable (Horvath and Raj, 2018). The most refined measures of the epigenetic clock are “multi-tissue age estimators”, which apply to all tissues and cell types across the entire duration of the human lifespan to predict the functional capability of a person or specific organs (ibid.). Horvath and Raj (2018) list many diseases and disease-related conditions that are all linked to epigenetic age acceleration, among which are BMI, cardiovascular disease, coronary heart disease, cellular senescence, Parkinson’s, dementia and neuropathology. In addition to health and disease, the epigenetic clock also mirrors inequality-related life course outcomes. In the overview provided by Horvath and Raj (2018), education and income were linked to the epigenetic clock as well. The same was found for household income, in addition to lifestyle characteristics (Levine et al., 2018; Zhao et al., 2019). Other studies have identified a trade-off between resilience to stress at the cost of accelerated epigenetic age, which suggests that an equation in the form of resilience protecting from accelerated aging may be too simplistic (Miller et al., 2015; Mehta et al., 2018). Overall these examples illustrate how complex and comprehensive the linkage of epigenetic encodings with life course experiences is, including social and genetic origins as points of departure, and patterns of shared genetic influences as well as gene–environment correlation. Adding to this, macrosocial characteristics of societies impinge on the biological embedding of individual social conditions

66  Handbook of health inequalities across the life course and experiences, including labor market characteristics, social inequality regimes and physical conditions (Simons et al., 2021).

GENE-HEALTH STUDIES THROUGH THE LENS OF THE LIFE COURSE Studies on epigenetic differences and changes highlight the role of strains and resources across the life course, with early conditions and experiences sometimes still influencing life course outcomes in later life stages (see Dekhtyar and Fors as well as Cozzani and Härkönen, Chapters 18 and 19 in this volume). Yet this mostly happens in a reductionist way in terms of focusing on single experiences or differentiating only between more or less sensitive periods when stressors occur. Though epigenetic encoding also changes during less sensitive life phases, sensitive life periods have gained the lion’s share of the attention devoted to the study of how the social environment has triggered epigenetic and health effects. Stability and change in epigenetic encoding call for systematic accounts of relevant experiences on the social side of gene–environment interactions in a longitudinal perspective. The prevailing reductionism is in stark contrast to complex epigenetic patterns like the epigenetic clocks covering a multitude of markers at the biological side of gene–environment interactions, and having an impact on a multitude of health-related and other inequality-related life course outcomes. In this section, I take a closer look at how life course concepts can systematically enrich this limited understanding of the interplay between social and biological processes. Types of Gene–Environment Interaction The first step in this direction is to distinguish basic types of gene–environment interaction known from behavioral genetic research on the one hand, and life course theoretical ideas about interdependencies between life domains on the other hand. Perhaps the most important type is what is called stress-diathesis in behavioral genetics or triggering in a taxonomy proposed by Shanahan and Hofer (2005). In this case, a latent genetic vulnerability (or diathesis) exists that is only expressed in the presence of a triggering agent (“strong triggering”) or is expressed markedly more so in the presence of the agent (“weak triggering”). This idea builds on the stress paradigm, which holds that the experience of a stressor may result in some form of distress. But as Shanahan and Hofer note, also normative factors that are not perceived as stressors may trigger a diathesis, e.g., if peer groups may encourage behaviors that have negative consequences for health, as is the case for binge drinking or drug use. A second type is referred to as the bioecological model or enhancement in the taxonomy of Shanahan and Hofer. This type comprises resource-rich, favorable environments that for desirable characteristics are helpful to push a genetic potential towards upper capacity. An example of this type is the so-called Scarr-Rowe hypothesis, which assumes that a given genetic potential for cognitive ability is best developed in resource-rich parental homes (Tucker-Drob and Bates, 2016). A third type mentioned by Shanahan and Hofer is compensation, which means the opposite of triggering, i.e. that a positive, enriched setting prevents the expression of a genetic diathesis, or turns a negative health development, or epigenetic encoding, back into a positive direction. A fourth type is social control, which refers to norms and behaviors that are placed upon people to prevent them from exhibiting health-threatening behaviors, like living as

Genetic variation, gene transcription and life course experiences  67 “couch potatoes” or smoking (e.g., Kendler, Thornton and Pederson, 2000). This type is linked to the fact that social capital has often been proved to be a major resource for healthy lifestyles and recovery from diseases (Ehsan et al., 2019; see also Klocke and Stadtmüller, Chapter 11 in this volume). Beyond Traumatic Experiences: What Other Kinds Of Stressors? Traumatic experiences (e.g., parental abuse or neglect, or being sexually assaulted) draw a lot of attention, and rightly so, when it comes to the analysis of life course experiences on epigenetic encoding and health trajectories more generally. However, we also need to focus attention on other and more common experiences, by leaning towards the classical distinction of events (e.g. stressful, time-limited life events like the loss of significant others or a job loss) and states (e.g. daily hassles) (Udayar et al., 2021). Another distinction in the literature on life stress distinguished between a “lumping” and a “splitting” approach (Smith and Pollak, 2020). Whereas in the first case various types of stressors are treated as a heterogeneous, broad category, in the latter the premise is that different types of adversity each confer specific effects. Similarly, an “independent risk” approach, singling out the specific influence of a specific stressor independent of other stressors, can be distinguished from a “Dimensional Model of Adversity and Psychopathology” where different types of stressors are distinguished, e.g. deprivation versus threat (McLaughlin, Sheridan and Lambert, 2014; Henry et al., 2021), or between “major live events and trauma” versus “chronic social and environmental stress” and “physiological stressors” (McEwen and Akil, 2020), combined in a processual model including social structure, genetic and epigenetic inheritance and epigenetic change (McEwen and McEwen, 2017). Though the more common focus is on stressful events, studies find increasingly a stronger impact of more enduring daily hassles (Zannas et al., 2015), which has received less attention. More attention should be devoted to experiences that specifically threaten an individual’s social well-being, which in terms of the social production function theory (Ormel et al., 1999) can be distinguished in status or respect (Honneth, 2007), behavioral confirmation and affection. Experiences of discrimination, stigmatization and social exclusion or affront may result in bitterness and hurt. Such experiences proved to be more robust sources of chronic inflammation than any lifestyle indicators in ethnic minorities and low SES families (Simons et al., 2018; Cuevas et al., 2020). Moreover, events and states are not completely independent from one another: Singular events can alter daily life for quite a long time, e.g. if the loss of a close relationship severely limits the availability of social support thereafter. It is also important to look at the duration of states and to locate this duration in different life phases. An important distinction here is whether life events are predictable and expected, or non-normal and unexpected at a certain age, e.g. becoming widowed before old age. It is especially uncommon events that challenge resilience. Longitudinal Perspective: Stressors over the Life Course An important step away from single, isolated events and states is to locate them within a longitudinal series of experiences accumulating over a lifetime. This can result in cumulative advantage (a favorable relative position becomes a resource that produces further

68  Handbook of health inequalities across the life course relative gains) or disadvantage (adversity becomes a risk for further adversity; see also Kelley, Oladimeji and Dannefer, Chapter 3 in this volume). The idea seems so plausible that it makes sense to distinguish between different mechanisms as to how the accumulation comes into being (DiPrete and Eirich, 2006). Among them is a social multiplier effect proposed by Dickens and Flynn (2001) for the development of cognitive ability. It explains how a genetic propensity interacts with the environment to produce cumulative advantage. This mechanism starts with individuals’ self-selection into different social environments based on their IQ. For example, a high IQ leads to a higher propensity to select stimulating rather than comforting contexts and activities. This self-selection induces a correlation between genetic and environmental factors, which over time leads to “further cognitive gains in the subsequent period, and then further environmental selection, and then further IQ gains, and so forth into the future … [which allows] … relatively small exogenous changes in the environment to cause surprisingly large changes over time” (DiPrete and Eirich, 2006, p. 284). Similarly, in epigenetics, early encoding of a health risk or health-threatening behaviors would repeatedly lead to increased health risks under adverse circumstances – if not followed by compensatory, resource-rich conditions interrupting this chain of risk and adversity. Against the widespread assumptions of cumulative advantage and disadvantage, Seery et al. (2013) found evidence for a moderate accumulation of adversity in the form of stressors being most beneficial to cope with new stressors. A moderate number of adverse life events was associated with less negative responses to new stressors compared to a history of low or no adversity as well as a history of high adversity. A kind of training effect contributes to developing resilience, a propensity for managing well in the face of stressors. This propensity seems to generalize to new domains and does not depend on other potential coping resources and liabilities (ibid., p. 1188).

CONCLUSIONS Research on genetic variation and epigenetics has reached impressive advancements during the last two decades. This applies to many subjects, but especially health-related research profits considerably in two respects. First, the strong interest in the role of social origin in the form of socioeconomic status – or better the single components of socioeconomic status – gets complemented by genetic origin. Both together do allow us to disentangle more accurately whether social origin influences are factual social influences or confounded to some degree by genetic variation. Second, the transition from genetic variation in the DNA to a dynamic view on gene transcription as the development of the epigenome opens avenues for a comprehensive interdisciplinary view on health development over the life course. The genetic side is then represented by elaborated, multidimensional concepts like the epigenetic clock. The social side would cover multidimensional, longitudinal concepts of social inequality in the combination of resources and stressful demands, and different experiences of social well-being. In combining both sides, there is no clear boundary between social science genetics and medical genetics (Harden and Koellinger, 2020, p. 573). In other words, to take the term “individual” in its original, Latin meaning as “indivisible”, we see that the biological organism, the mental system, and the social embedding do not work independently from each other. Instead, they influence each other in both directions.

Genetic variation, gene transcription and life course experiences  69 The more promising new possibilities to elaborate on the complex interplay between genes and social forces become, the more we become aware of their complexity. The challenge for future research is to find a way between the extremes of “everything depends on everything” and simplifying concepts by negating the multidimensional character of biological and social influences over the life course. This will presumably not succeed without the help of new approaches to cope with abundant information adopted from machine learning (Kröger, Chapter 8 in this volume) or the use of established methods to cope with unobserved heterogeneities like sampling monozygotic twins in a case co-twin design (Turner et al., 2020). Despite these challenges, advances in molecular genetic and epigenetic methodology allow for individualized analyses over and above population-related statements based on twin-based methodology. This opens up new avenues for new businesses, including recent collaborations on the biological reprogramming of cells to put the epigenetic clock back in the direction of eternal youth (Regalado, 2021).

REFERENCES Alzu’bi, A. A., Zhou, L. and Watzlaf, V. (2019) ‘Genetic Variations and Precision Medicine’, Perspectives in Health Information Management, 16(Spring), p. 1a. Bakker, A. B. and Demerouti, E. (2007) ‘The Job Demands–Resources Model: State of the Art’, Journal of Managerial Psychology, 22, pp. 309–328. doi:10.1108/02683940710733115. Barker, D. J. (2004) ‘The Developmental Origins of Adult Disease’, Journal of the American College of Nutrition, 23(6 Suppl), pp.  588S–595S. https://​doi​.org/​10​.1080/​07315724​.2004​.10719428 Beckfield, J., Olafsdottir, S. and Bakhtiari, E. (2013) ‘Health Inequalities in Global Context’, American Behavioral Scientist, 57, pp.  1014–1039. https://​doi​.org/​10​.1177/​0002764213487343 Bell, C. G., Lowe, R., Adams, P. D., Baccarelli, A. A., Beck, S., Bell, J. T., Christensen, B. C., Gladyshev, V. N., Heijmans, B. T., Horvath, S., Ideker, T., Issa, J. J., Kelsey, K. T., Marioni, R. E., Reik, W., Relton, C. L., Schalkwyk, L. C., Teschendorff, A. E., Wagner, W., Zhang, K., … Rakyan, V. K. (2019) ‘DNA Methylation Aging Clocks: Challenges and Recommendations’, Genome Biology 20, p.  249. https://​doi​.org/​10​.1186/​s13059​-019​-1824​-y Bogg, T. and Roberts, B. W. (2004) ‘Conscientiousness and Health-related Behaviors: A Meta-analysis of the Leading Behavioral Contributors to Mortality’, Psychological Bulletin, 130, pp. 887–919. Boyce, W. T. and Hertzman, C. (2018) ‘Early Childhood Health and the Life Course: The State of the Science and Proposed Research Priorities: A Background Paper for the MCH Life Course Research Network’, in: Halfon, N., Forrest, C. B., and Lerner R. M. (eds.), Handbook of Life Course Health Development. Cham: Springer. Available from: https://​www​.ncbi​.nlm​.nih​.gov/​books/​NBK543727/​ doi: 10.1007/978-3-319-47143-3_4 Briley, D. A., Livengood, J., Derringer, J., Tucker-Drob, E. M., Fraley, R. C. and Roberts, B. W. (2019) ‘Interpreting Behavior Genetic Models: Seven Developmental Processes to Understand’, Behavior Genetics, 49, pp. 196–210. Choi, S. W. and Friso, S. (2010) ‘Epigenetics: A New Bridge between Nutrition and Health’, Advances in Nutrition, 1, pp. 8–16. Choi, S. W., Mak, T. S. and O’Reilly, P. F. (2020) ‘Tutorial: A Guide to Performing Polygenic Risk Score Analyses’, Nat Protoc, 15, pp. 2759–2772. https://​doi​.org/​10​.1038/​s41596​-020​-0353​-1 Cuevas, A. G., Ong, A. D., Carvalho, K., Ho, T., Chan, S. W. C., Allen, J. D., Chen, R., Rodgers, J., Biba, U. and Williams, D. R. (2020) ‘Discrimination and Systemic Inflammation: A Critical Review and Synthesis’, Brain Behavior Immunity, 89, pp. 465–479. doi: 10.1016/j.bbi.2020.07.017. Dickens, W. T. and Flynn, J. R. (2001) ‘Heritability Estimates Versus Large Environmental Effects: The IQ Paradox Resolved’, Psychological Review, 108, pp.  346–369. https://​doi​.org/​10​.1037/​0033​-295X​ .108​.2​.346

70  Handbook of health inequalities across the life course Diewald, M., Baier, T., Schulz, W. and Schunck, R. (2015) ‘Status Attainment and Social Mobility: How Can Genetics Contribute to an Understanding of their Causes?’‚ Kölner Zeitschrift für Soziologie und Sozialpsychologie, 67(1 Supplement), pp.  371–395. https://​doi​.org/​10​.1007/​s11577​-015​-0317​-6 DiPrete, T.A. and Eirich, G.M. (2006) ‘Cumulative Advantage as a Mechanism for Inequality: A Review of Theoretical and Empirical Developments’, Annual Review of Sociology, 32, pp. 271–297. Ehsan, A., Klaas, H. S., Bastianen, A. and Spini, D. (2019) ‘Social Capital and Health: A Systematic Review of Systematic Reviews’, SSM – Population Health, 8, p.  100425. https://​doi​.org/​10​.1016/​j​ .ssmph​.2019​.100425 Ellis, B., Boyce, W., Belsky, J., Bakermans-Kranenburg, M. and van Ijzendoorn, M. (2011) ‘Differential Susceptibility to the Environment: A Neurodevelopmental Theory’, Development and Psychopathology, 23, pp. 7–28. Flowers, E., Froelicher, E. S. and Aouizerat, B. E. (2012) ‘Gene–Environment Interactions in Cardiovascular Disease’, European Journal of Cardiovascular Nursing, 11, pp. 472–478. https://​doi​ .org/​10​.1016/​j​.ejcnurse​.2011​.06​.001. Gottfredson, L. S. and Deary, I. J. (2004) ‘Intelligence Predicts Health and Longevity, but Why?’, Current Directions in Psychological Science, 13, pp. 1–4. Halfon, N., Forrest C. B. and Lerner, R. M. (eds.) (2018), Handbook of Life Course Health Development. Cham: Springer. Harden, K. P. (2021) ‘“Reports of My Death Were Greatly Exaggerated”: Behavior Genetics in the Postgenomic Era’, Annual Review of Psychology, 72, pp. 37–60. Harden, K. P. and Koellinger, P. D. (2020) ‘Using Genetics for Social Science’, Nature Human Behavior, 4, pp. 567–576. https://​doi​.org/​10​.1038/​s41562​-020​-0862​-5 Henry, L. M., Gracey, K., Shaffer, A., Ebert, J., Kuhn, T., Watson, K. H., Gruhn, M., Vreeland, A., Siciliano, R., Dickey, L., Lawson, V., Broll, C., Cole, D. A. and Compas, B. E. (2021) ‘Comparison of Three Models of Adverse Childhood Experiences: Associations with Child and Adolescent Internalizing and Externalizing Symptoms’, Journal of Abnormal Psychology, 130(1), pp. 9–25. Hoffmann, R., Kröger, H. and Pakpahan, E. (2018) ‘Pathways between Socioeconomic Status and Health: Does Health Selection or Social Causation Dominate in Europe?’, Advances in Life Course Research, 36, pp. 23–36. Honneth, A. (2007) Disrespect. The Normative Foundations of Critical Theory. Cambridge: Polity Press. Horvath, S. and Raj, K. (2018) ‘DNA Methylation-based Biomarkers and the Epigenetic Clock Theory of Ageing’, Nature Review Genetics, 19, pp. 371–84. Kendler, K. S., Thornton, L. M. and Pedersen, N. L. (2000) ‘Tobacco Consumption in Swedish Twins Reared Apart and Reared Together’, Archives of General Psychiatry, 57, pp. 886–892. https://​doi​.org/​ 10​.1001/​archpsyc​.57​.9​.886 Kim-Cohen, J., Moffitt, T. E., Caspi, A. and Taylor, A. (2004) ‘Genetic and Environmental Processes in Young Children’s Resilience and Vulnerability to Socioeconomic Deprivation’, Child Development, 75, pp. 651–668. Kohler, H. P., Behrman, J. R. and Schnittker, J. (2011) ‘Social Science Methods for Twins Data: Integrating Causality, Endowments and Heritability’, Biodemography and Social Biology, 57, pp. 88–141. doi:10.1080/19485565.2011.580619. Kuh, D. and Ben Shlomo, Y. (eds) (2004) A Life Course Approach to Chronic Diseases Epidemiology, 2nd edition. Oxford: Oxford University Press. Levine, M. E., Lu, A. T., Quach, A., Chen, B. H., Assimes, T. L., Hou, L. and Ferrucci, L. (2018) ‘An Epigenetic Biomarker of Aging for Lifespan and Healthspan’, Aging, 10, p. 19. Li, X., Ploner, A., Wang, Y., Magnusson, P. K. E., Reynolds, C., Finkel, D., Pedersen, N. L., Jylhävä, J. and Hägg, S. (2020) ‘Longitudinal Trajectories, Correlations and Mortality Associations of Nine Biological Ages across 20-Years Follow-Up’, eLife, 9, e51507. Liu, Z., Chen, X., Gill, T. M., Ma, C., Crimmins, E. M., and Levine, M. E. (2019) ‘Associations of Genetics, Behaviors, and Life Course Circumstances with a Novel Aging and Healthspan Measure: Evidence from the Health and Retirement Study’, PLoS Medicine, 16, e1002827. doi: 10.1371/ journal.pmed.1002827. Liu, Z., Leung, D., Thrush, K., Zhao, W., Ratliff, S., Tanaka, T., Schmitz, L. L., Smith, J. A., Ferrucci, L. and Levine, M. E. (2020) ‘Underlying Features of Epigenetic Aging Clocks In Vivo and In Vitro’, Aging Cell, 19(10), e13229. https://​doi​.org/​10​.1111/​acel​.13229

Genetic variation, gene transcription and life course experiences  71 MacGregor, A. J., Snieder, H., Schork, N. J., and Spector, T. D. (2000) ‘Twins: Novel Uses to Study Complex Traits and Genetic Diseases’, Trends in Genetics, 16, pp. 131–134. March, J. G. and Simon, H. A. (1958) Organizations. New York: John Wiley & Sons. McEwen, B. S. and Akil, H. (2020) ‘Revisiting the Stress Concept: Implications for Affective Disorders’, Journal of Neuroscience, 40, p. 12. McEwen, C. A. and McEwen, B. S. (2017) ‘Social Structure, Adversity, Toxic Stress, and Intergenerational Poverty: An Early Childhood Model’, Annual Review of Sociology, 43(1), pp. 445–472. McGuinness, D., McGlynn, L. M., Johnson, P. C., MacIntyre, A., Batty, G. D., Burns, H., Cavanagh, J., Deans, K. A., Ford, I., McConnachie, A., McGinty, A., McLean, J. S., Millar, K., Packard, C. J., Sattar, N. A., Tannahill, C., Velupillai, Y. N. and Shiels, P. G. (2012) ‘Socio-economic Status is Associated with Epigenetic Differences in the pSoBid Cohort’, International Journal of Epidemiology, 41, pp. 151–160. McLaughlin, K. A., Sheridan, M. and Lambert, H.K. (2014) ‘Childhood Adversity and Neural Development: Deprivation and Threat as Distinct Dimensions of Early Experience’, Neuroscience & Biobehavioral Reviews, 47, pp. 578–591. Meaney, M. J. (2010) ‘Epigenetics and the Biological Definition of Gene × Environment Interactions’, Child Development, 81, pp. 41–79. doi:10.1111/j.1467-8624.2009.01381.x Mehta, D., Bruenig, D., Lawford, B., Harvey, W., Carrillo-Roa, T., Morris, C. P. and Voisey, J. (2018) ‘Accelerated DNA Methylation Aging and Increased Resilience in Veterans: The Biological Cost for Soldiering On’, Neurobiology of Stress, 8, pp. 112–119. Miller, G. E. and Chen, E. (2010) ‘Harsh Family Climate in Early Life Presages the Emergence of a Proinflammatory Phenotype in Adolescence’, Psychological Science, 21, pp. 848–856. https://​doi​ .org/​10​.1177/​0956797610370161 Miller, G. E., Yu, T., Chen, E. and Brody, G. H. (2015) ‘Self-control Forecasts Better Psychosocial Outcomes but Faster Epigenetic Aging in low-SES Youth’, Proceedings of the National Academy of Sciences, 112, pp.  10325–10330. https://​doi​.org/​10​.1073/​pnas​.1505063112 Miller, G. E., Chen, E., Shalowitz, M. U., Story, R. E., Leigh, A. K. K., Ham, P., Arevalo, J. M. G. and Cole, S. W. (2018) ‘Divergent Transcriptional Profiles in Pediatric Asthma Patients of Low and High Socioeconomic Status’, Pediatric Pulmonology, 53, pp. 710–719. Mitchell, C., Hobcraft, J., McLanahan, S. S., Siegel, S. R., Berg, A., Brooks-Gunn, J., Garfinkel, I. and Notterman, D. (2014) ‘Social Disadvantage, Genetic Sensitivity, and Children’s Telomere Length’, Proceedings of the National Academy of Sciences, 111, pp. 5944–5949. Nudel, R., Wang, Y., Appadurai, V. et al. (2019) ‘A Large-scale Genomic Investigation of Susceptibility to Infection and its Association with Mental Disorders in the Danish Population’, Transl Psychiatry, 9, p. 283. https://​doi​.org/​10​.1038/​s41398​-019​-0622​-3 Ormel, J., Lindenberg, S., Steverink, N. and Verbrugge, L. M (1999) ‘Subjective Well-Being and Social Production Functions’, Social Indicators Research, 46, pp. 61–90. Pandey, S. C., Kyzar, E. J. and Zhang, H. (2017) ‘Epigenetic Basis of the Dark Side of Alcohol Addiction’, Neuropharmacology, 122, pp. 74–84. Perbal, L. (2015) ‘The Case of the Gene: Postgenomics between Modernity and Postmodernity’, EMBO Reports, 16, pp. 777–781. doi:10.15252/embr.201540179 Polderman, T. J., Benyamin, B., de Leeuw, C. A., Sullivan, P. F., van Bochoven, A., Visscher, P. M. and Posthuma, D. (2015) ‘Meta-analysis of the Heritability of Human Traits based on Fifty Years of Twin Studies’, Nature Genetics, 47, pp.  702–709. https://​doi​.org/​10​.1038/​ng​.3285 Regalado, A. (2021) ‘Meet Altos Labs, Silicon Valley’s latest Wild Bet on Living Forever’, MIT Technology Review, Sept. 4. https://​www​.technologyreview​.com/​2021/​09/​04/​1034364/​altos​-labs​ -silicon​-valleys​-jeff​-bezos​-milner​-bet​-living​-forever/​ Ryan, C. P. (2021) ‘Epigenetic Clocks: Theory and Applications in Human Biology’, American Journal of Human Biology, 33, e23488. https://​doi​.org/​10​.1002/​ajhb​.23488 Schieman, S., Whitestone, Y. and Van Gundy, K. (2006) ‘The Nature of Work and the Stress of Higher Status’, Journal of Health and Social Behavior, 47, pp. 242–257. Seery, M. D., Leo, R. J., Lupien, S. P., Kondrak, C. L. and Almonte J. L. (2013) ‘An Upside to Adversity? Moderate Cumulative Lifetime Adversity is Associated with Resilient Responses in the Face of Controlled Stressors’, Psychological Science, 24, pp. 1181–1189. doi: 10.1177/0956797612469210.

72  Handbook of health inequalities across the life course SeyedAlinaghi, S., Mehrtak, M., MohsseniPour, M. et al. (2021) ‘Genetic Susceptibility of COVID-19: A Systematic Review of Current Evidence’, Eur J Med Res, 26, p.  46. https://​doi​.org/​10​.1186/​s40001​ -021​-00516​-8 Shanahan, M. J. and Hofer, S. M. (2005) ‘Social Context in Gene-Environment Interactions: Retrospect and Prospect’, Journals of Gerontology: Series B, Psychological Sciences and Social Sciences, 60, Spec No 1, pp.  65–76. https://​doi​.org/​10​.1093/​geronb/​60​.special​_issue​_1​.65 Shanahan, M. J., Hill, P. L., Roberts, B. W., Eccles, J. and Friedman, H. S. (2014) ‘Conscientiousness, Health, and Aging: The Life Course of Personality Model’, Developmental Psychology, 50, pp. 1407–1425. Simons, R. L., Lei, M. K., Beach, S. R. H., Barr, A. B., Simons, L. G., Gibbons, F. X. and Philibert, R. A. (2018) ‘Discrimination, Segregation, and Chronic Inflammation: Testing the Weathering Explanation for the Poor Health of Black Americans’, Developmental Psychology, 54, pp. 1993–2006. doi: 10.1037/dev0000511. Simons, R. L., Lei, M. K., Klopach, E, Berg, M., Zhang, Y. and Beach, S. S. R. (2021) ‘(Re) Setting Epigenetic Clocks: An Important Avenue Whereby Social Conditions Become Biologically Embedded across the Life Course’, Journal of Health and Social Behavior, 62, pp. 436–453. Simpkin, A. J., Hemani, G., Suderman, M., Gaunt, T. R., Lyttleton, O., McArdle, W. L., Ring, S. M., Sharp, G. C., Tilling, K., Horvath, S., Kunze, S., Peters, A., Waldenberger, M., Ward-Caviness, C., Nohr, E. A., Sørensen, T. I., Relton, C. L. and Smith, G. D. (2016) ‘Prenatal and Early Life Influences on Epigenetic Age in Children: A Study of Mother-Offspring Pairs from Two Cohort Studies’, Human Molecular Genetics, 25, pp. 191–201. Smith, K. E. and Pollak, S. D. (2020) ‘Early Life Stress and Development: Potential Mechanisms for Adverse Outcomes’, Journal of Neurodevelopmental Disorders, 12(1), p. 34. Tucker-Drob, E. M. and Bates, T. C. (2016) ‘Large Cross-National Differences in Gene × Socioeconomic Status Interaction on Intelligence’, Psychological Science, 27, pp.  138–149. https://​doi​.org/​10​.1177/​ 0956797615612727 Turner, J. D. (2018) ‘Holistic, Personalized, Immunology? The Effects of Socioeconomic Status on the Transcriptional Milieu of Immune Cells’, Pediatric Pulmonology, 53, pp. 696–697. doi: 10.1002/ ppul.23986. Turner, J. D., D’Ambrosio, C., Vögele, C. and Diewald, M. (2020) ‘Twin Research in the Post-Genomic Era: Dissecting the Pathophysiological Effects of Adversity and the Social Environment’, International Journal of Molecular Sciences, 21(9), p. 3142. Udayar, S., Urbanaviciute, I., Morselli, D., Bollmann, G., Rossier, J. and Spini, D. (2021) ‘The LIVES Daily Hassles Scale and its Relation to Life Satisfaction’, Assessment, 0(0). https://​doi​.org/​10​.1177/​ 10731911211047894 van Dongen, J., Nivard, M., Willemsen, G. et al. (2016) ‘Genetic and Environmental Influences Interact with Age and Sex in Shaping the Human Methylome’, Nature Communications, 7, p. 11115. https://​ doi​.org/​10​.1038/​ncomms11115 Vick, A. D. and Burris, H. H. (2017) ‘Epigenetics and Health Disparities’, Current Epidemiological Reports, 4, pp. 31–37. Vidrascu, E. M., Bashore, A. C., Howard, T. D. and Moore, J. B. (2019) ‘Effects of Early- and Mid-Life Stress on DNA Methylation of Genes Associated with Subclinical Cardiovascular Disease and Cognitive Impairment: A Systematic Review’, BMC Medical Genetics, 20, p. 39. https://​doi​.org/​10​ .1186 Vinkers, C. H., Kalafateli, A. L., Rutten, B. P., Kas, M. J., Kaminsky, Z., Turner, J. D. and Boks, M. P. (2015) ‘Traumatic Stress and Human DNA Methylation: A Critical Review’, Epigenomics, 7, pp. 593–608. Weiss, A. and Costa, P. T. (2005) ‘Domain and Facet Personality Predictors of All-cause Mortality among Medicare Patients aged 65 to 100’, Psychosomatic Medicine, 67, pp. 724–733. Zakarya, R., Adcock, I. and Oliver, B. G. (2019) ‘Epigenetic Impacts of Maternal Tobacco and e-Vapour Exposure on the Lung’, Clinical Epigenetics, 11, p. 32. Zannas, A. S., Arloth, J., Carrillo-Roa, T., Iurato, S., Röh, S., Ressler, K. J., Nemero, C. B., Smith, A. K., Bradley, B. and Heim, C. J. G. (2015) ‘Lifetime Stress Accelerates Epigenetic Aging in an Urban, African American Cohort: Relevance of Glucocorticoid Signaling’, Genome Biology, 16, 266.

Genetic variation, gene transcription and life course experiences  73 Zhao, W., Ammous, F., Ratliff, S., Liu, J., Yu, M., Mosley, T. H. and Smith, J. A. (2019) ‘Education and Lifestyle Factors are Associated with DNA Methylation Clocks in Older African Americans’, International Journal of Environmental Research and Public Health, 16, p. 3141. https://​doi​.org/​10​ .3390/​ijerph16173141 Ziol-Guest, K. M., Duncan, G. J., Kalil, A., and Boyce, W. T. (2012) ‘Early Childhood Poverty, Immune-mediated Disease Processes, and Adult Productivity’, Proceedings of the National Academy of Sciences, 109 (Supplement 2), pp. 17289–17293.

PART II METHODOLOGICAL ISSUES FOR THE LONGITUDINAL ANALYSIS OF HEALTH INEQUALITIES

6. Methods for studying life course health inequalities Scott M. Lynch and Christina Kamis

INTRODUCTION The key goal of life course analyses of health is to understand the emergence of, and changes in, health inequalities between social groupings over the lives of individuals and across sociohistorical time. Contemporary social science work in this area uses a variety of statistical methods mostly applied to large-scale longitudinal (i.e., panel) data sets, period mortality files, or cross-sectional survey data sets often linked to mortality files. The most commonly used methods applied to such data can be divided into two broad domains: analyses of discrete events and analyses of repeated measures. This chapter provides a detailed overview of some of the most common and general methods employed in the social sciences within these two domains.

METHODS FOR ANALYZING THE OCCURRENCE OF EVENTS The life course is marked by sequences of events that occur at different (st)ages, such as a baby’s first steps, a child’s first day of school, graduation from various educational programs, entry into the workforce, marriage, childbearing, retirement, acquisition of a chronic health condition, and death, among others. Not all individuals in a society experience all events that may characterize the insitutionalized life course; some experience some events more than once; and some experience events but are delayed in their experience of them relative to others. Variation across a population in the timing of events is natural, but systemic inequalities – that is, consistent differences between socially defined groups – in the timing of normative events are not necessarily “natural,” and they may beget further inequalities. Statistical methods used in the life course study of inequalities are therefore concerned with the differential experience of, and especially the timing of, events. The most basic type of analysis of events is the familiar logit (or probit) model. Most often, scholars using this model will simply estimate the effects of covariates representing social groupings on whether individuals experience a given event within a specified time period, such as first marriage by age 25, death prior to age 75, or acquisition of a chronic condition prior to the next survey wave. We will not review the logit (or probit) model here for two reasons. First, the model is well known. Second, modeling the occurrence of events within a broad time period is limiting for the study of inequality. The key limitation of modeling simply whether an event occurs within a particular time period is that it ignores information about the differential timing of the event, and that information may be crucial to understanding inequalities. As an important example, everyone dies: there is no inequality in that fact. However, age at death is highly variable, and these differ75

76  Handbook of health inequalities across the life course ences can be systemic. Even if one constrains the measurement of death within a somewhat narrow time interval (e.g., five years), considerable inequality may be missed if the specific timing of death is ignored. Thus, rather than a simple logit/probit model, a better approach should incorporate the timing of the event. A seemingly straightforward approach to modeling the timing of events might be to use a linear regression model for the time-to-event. There are two problems with such an approach, however, when using typical social science data sets. First (and this pertains to any data set), a distribution of times to events is bounded at zero. Event times are strictly nonnegative, so any time-to-event distribution is necessarily right skewed, making a linear model inappropriate. Second, and more importantly, in large-scale survey data sets, many individuals do not experience the event of interest within a survey window. Thus, the time-to-event (TTE) distribution is censored for many if not most sample members, posing a missing data problem for a large portion of any sample. Importantly, the missing data is almost assuredly directly related to predictors of inequality in timing, so that the censoring cannot simply be ignored. Hazard Models Remedying these problems requires the development of an alternative set of methods to usual linear and generalized linear modeling. Central to these methods is the notion of hazard functions, survival functions, and time-to-event distributions. The hazard function represents the probability of the occurrence of an event in an infinitely narrow window of time, conditional on survival up to the start of the window. It is often expressed as: h t   lim

t 0

p (t  T  t  t T t ) , (1) t

where ∆t is a small increment of time, and T is the event occurrence time. The hazard function is related to the TTE distribution, f t  , which is simply a probablity density function representing the relative frequencies of event times, and the survival function, S t  , which is

simply the area under the TTE distribution beyond t . Indeed, h t  , S t  , and f t  are perfectly related to one another: if one knows one function, one necessarily knows the others. Specifically, the survival function is simply the complement of the cumulative distribution function for f t  : S t   1 



t 0

f t   1  F t . (2)

Further, the hazard function can be expressed as a function of the time to event distribution and the survival function: h t  

f t  . (3) S t 

Methods for studying life course health inequalities  77 This result follows from the fact that the density function provides the frequency of event occurrences at an instant in time, while the survival function conditions this occurrence on survival to that instant. Thus, if one knows f t  , S t  and h t  are easily derivable using integral calculus to obtain S t  from f t  and then forming the ratio above. Similarly, if one knows S t  , one

can derive f t  using differential calculus, and, again, forming the ratio above. If one knows

h t  , deriving f t  and S t  takes a little more effort. The ratio in the above equation for the

hazard can be viewed as a differential equation, such that S t  and f t  can be derived: h t   

f t  (4) 1  F t 

du (5) 1 u t

 h t   ln 1  F t  (6)



0 t

e 0

 h t 

 S t . (7)

Given S t  , derivation of f t  is straightforward, since f t   h t  S t  from Equation 3.

Because of the one-to-one relationships between f t  , S t  , and h t  , models for events can be derived for any of them. In biological/zoological and clinical sciences involving either short-lived species or human survival following treatments for often fatal diseases, methods for exploring the survival function (e.g., Kaplan-Meier and related methods) are of considerable interest, because there is relatively little censoring. We will not discuss those methods here, but see Kalbfleisch and Prentice (2002) for details. In demography and the social sciences more broadly, which tend to involve cross-sectional and short-term longitudinal data, especially relative to the length of the human lifespan, models for events have developed around the hazard function. Such models can easily handle the censoring that is common in human data. For example, in demography, age-specific estimates of hazards are often obtained by forming occurrence-exposure rates from the count of events (usually deaths) divided by the mid-year population, where the mid-year population is an estimate of the person-years lived within an age interval (Preston, Heuveline, and Guillot, 2001). These estimates allow for the production of period life tables (discussed below) as well as macro-level models of mortality rates (e.g., see Lynch and Brown, 2001); that is, models in which the unit of analysis is larger than the individual (in this case, population level rates). Although macro-level models are somewhat common in demographic studies of human mortality, micro-level (i.e., individual level) models of mortality or other events are more common across the social sciences. Whereas studying inequality using macro-level models of events relies on disaggregation of data, limiting the ability to pinpoint controlled group differences, micro-level models using survey data can easily incorporate multiple covariates. Typically, one decides on a specific pattern for the baseline hazard, that is, the general shape of

78  Handbook of health inequalities across the life course the hazard across time. For human populations, a common form is the Gompertz model, which implies a hazard that increases exponentially with age (t): h(t )i   i e  ti , (8) where i references the ith person in the sample, ti is the time of the event or censoring for person i, and β is a rate of “senescence” – or the pace at which risk increases across age. Individual level covariates typically enter the equation as an exponential function of covariates, X, including an intercept:  i  exp  T X i . It is common to use an exponential link function for α in





this model, because α must be nonnegative. Parameters can be estimated using maximum likelihood estimation after establishing a likelihood function based on the implied TTE distribution (obtained from Equation 3): L( ,  | T , D) 

n

  e i

i 1

 i   ti    exp   1  e   (9)

 ti  di

In this likelihood function, the bracketed term on the left is the hazard function, and the exponential function on the right is the survival function. di is an indicator for whether sample member i experienced the event of interest (e.g., mortality); if the respondent did not experience the event – i.e., the respondent’s event time is censored – this indicator takes a value of 0 so that the respondent only contributes to the survival function portion of the likelihood. This model is called a proportional hazards model because the exponential link function for the effects of covariates implies a multiplicative relationship between them: exp 0   1 xi1   2 xi 2   k xik   e 0 e 1xi1 e 2 xi 2 e k xik . (10) Thus, a one unit increase in x1, holding all else constant, multiplies the hazard by eγ 1 :  x 1 e 1 1   1 x1 1 1 x1   1 e  e . (11) e 1x1

Although the Gompertz model is routinely used in studies of human mortality, not all hazards increase exponentially across age/time. Some hazards may remain flat over time, while other hazards may increase with time and plateau or even decrease over time. Other common parametric hazard models include the exponential model and the Weibull model, which can capture varying age patterns for events (see Allison, 1984 for detailed discussion). However, we often do not care about the shape of the baseline hazard. In clinical sciences and non-demographic social sciences, the Cox Regression model is more often used. The Cox model, often ambiguously called “the proportional hazards model,” looks similar to a parametric model, but without the functional form of the hazard specified. Using the logic shown in Equation 10, so long as hazards are assumed to be proportional across age, the shape of the hazard is irrelevant because, whatever underlying parameters would be required to capture the shape, they would

Methods for studying life course health inequalities  79 cancel out if included, much like the effect of other stable covariates cancel in Equation 11. The real “trick” is estimating the coefficients in the model. In 1972, Cox published a seminal paper showing how a “partial likelihood function” could be constructed from sorted event times, and the parameters of the model estimated without concern for the age pattern in the hazard (Kalbfleisch and Prentice, 2002). In the 1980s, the “discrete time” logit model was popularized as an alternative for estimating risks of events among non-demographers. The discrete time logit model is a regular dichotomous logistic regression model but is applied to a specially constructed data set. Specifically, the data set is constructed so that each record in it is a person-year (or other time-interval) record, with individuals contributing varying numbers of records based on their exposure prior to experiencing an event. Suppose, for example, an individual dies four years after the beginning of a panel study. S/he would contribute four records to the person-year data set with his/her outcome set to 0 in the first three records and 1 in the final record. A key predictor variable in the model would most likely be age, which would be incremented in each of his/ her records. Thus, the model predicts the age-specific probability of death over the course of a year (assuming the time interval in the survey were years). Inequalities in event times can be studied by including additional covariates, such as gender or race/ethnicity, allowing for similar conclusions about characteristic-specific probabilities. Both parametric hazard models and the Cox regression model can be approximated using the discrete time hazard model, and the approximation becomes increasingly close as the time intervals get smaller. This should be apparent, given the definition of the hazard shown above as the limit of the probability an event occurs in an infinitely small interval of time. In particular, if age (or time) is entered as a linear predictor in the logit model so that the effect of age/time increases the log-odds of experiencing the event of interest linearly, the model is approximately equivalent to the Gompertz model. If, instead, time is measured with a set of dummy variables for each time interval, the time pattern is non-parametric so that the model approximates the Cox model. Over the past half-century, parametric hazard models, the Cox model, and discrete time logit models have been used in countless studies concerned with inequalities in event timing. In the last several decades, studies of inequality in event times have moved in two additional directions. First, parametric hazard models have been developed to capture “unobserved heterogeneity.” The basic parametric hazard model described above is unlike a typical linear model that contains an error term that captures factors that influence the outcome that are not included among the covariates. Yet, surely the covariates we include in our models are not an exhaustive collection of potential confounders. The basic hazard model can be modified to include an “error” term, which can then be extracted from the exponential for easier interpretation: h(t )i  zi i e

T

Xi

. (12)

z is often called a “frailty” term, and models that incorporate such a term are generically called “frailty models.” z is assumed to be non-negative and have a mean of 1. If zi > 1, person i has a greater expected hazard rate at all ages than others with an identical covariate profile, indicating s/he is frailer than his/her peers, while if z < 1, s/he has a lower expected hazard and is thus considered more robust than his/her peers.

80  Handbook of health inequalities across the life course z is assumed to follow a prespecified distribution. In early frailty models, z was assumed to follow a gamma distribution parameterized by a mean fixed at 1 and a variance to be estimated. The variance parameter reflected the extent of heterogeneity (or inequality) in mortality experience in a population beyond that captured by the covariates. Importantly, in early work on this “gamma-Gompertz” model, it was shown that, while individual hazards under this model follow a Gompertz pattern as described above, the hazard at the population level is logistic in shape (Vaupel and Yashin, 1985). This result fits nicely with research from the late 1970s forward that has found that mortality rates tend to decelerate at the oldest ages; that is, mortality rates continue to increase with age but they do not do so exponentially. While much research has argued that this is evidence of heterogeneity in mortality within the population (e.g., Horiuchi and Wilmoth, 1998; Lynch and Brown 2001), recent research has argued that deceleration is attributable to a different form of heterogeneity, namely, the use of period mortality data, which consists of rates from birth cohorts that vary in their mortality experience due to changing mortality over time (Gavrilov and Gavrilova, 2011). Nonetheless, frailty models remain common for investigating heterogeneity in mortality in micro-level analyses. Life Table Methods A second area of growth and development in studying inequalities in the timing of events has been in the development of multistate life table methods for sample data. The straightforward, single decrement life table has been a staple in demography for centuries. The basic period life table is constructed from a collection of age-specific death rates computed from vital statistics counts of deaths in a year and age-specific population estimates obtained or extrapolated from census counts (Preston, Heuveline, and Guillot, 2001). The key output of the life table is life expectancy, which is an estimate of the mean age of death in a population. This quantity could be obtained analytically from the TTE distribution if we assume a particular age pattern for mortality across age: t 



 0

xf  x  dx. (13)

However, real mortality rates are not perfectly smooth, nor do they necessarily follow a clear parametric pattern across age (see discussion above). The life table is thus a non-parametric approach that approximates the TTE distribution in a piecewise fashion; that is, age-group by age-group. The key information required for a standard life table is a collection of age-specific mortality rates, m  x  , x  0. The life table generally consists of at least 6 columns, including: (1) the number of survivors at the start of each age interval, l  x ; (2) the proportion dying over

each age interval, q  x  ; (3) the number of deaths occuring in an age interval, d  x   l  x   q  x ;

(4) the number of person-years lived in an age interval, L  x ; (5) the total number of

person-years lived by persons age x to age ω , T  x ; and (6) life expectancy at each age,

e  x . The key assumption that underlies the production of the period life table is stationarity: that mortality rates are constant over time. This assumption is required to justify combining

Methods for studying life course health inequalities  81 age-specific mortality rates in a given year – thus, the mortality experience of multiple birth cohorts at different ages – and applying the life table calculations to a “synthetic” cohort whose mortality rates follow the observed, period rates. Two mathematical decisions must also be made to produce the life table: (1) the relationship between mortality rates and probabilities of death over an age interval; and (2) the number of person years lived by persons dying within an age interval. These decisions are ultimately related, because occurrence exposure rates can be represented as m  x   d  x  / L  x  : the number of events that occur over an age interval divided by the exposure, which is a combination of the time length of the interval and the number of persons at risk. The number of persons at risk decreases throughout an interval as deaths occur, so that the denominator is not simply the number of survivors at the start of the interval (i.e., l  x  ). d  x  is also related to l  x  and q  x  as shown above, and so q  x  and

L  x  are necessarily related. A common assumption is that deaths are uniformly distributed over an age interval, which is akin to assuming that all deaths, on average, occur in the middle of the interval. In that case, all survivors contribute (in the case of single year intervals) one person-year to L  x  , while those who die contribute one-half a person-year of exposure. Thus: L  x   l  x   .5d  x   l  x   .5l  x  q  x . The latter expression for L  x  can be substi-

tuted into the equation for m  x  above, and after some algebraic manipulation, we find that q  x   m  x  / 1  .5m  x  (see Preston, Heuveline, and Guillot, 2001).

Thus, given a set of m  x  , one can compute a set of q  x  and then construct the life table using an iterative process across the age range: 1. Start with a “radix” population, l  0  .

2. Compute d  x   l  x  q  x  .

3. Compute l  x  1  l  x   d  x  .

4. Compute L  x   l  x   .5d  x  . 5. Increment x and return to step 2 until the last age interval. The last age interval, ω , is open-ended, because there is no upper bound on age. q   is

necessarily 1 (all persons die), but m  x  is not. m  x  can be used to estimate L  x  as

L    l   / m   from the above formula for m  x  , since d    l   . The life table can

be completed by computing T  x  





L  x  for every age and then computing

i x

e  x   T  x  / l  x  for every age. e  x  is simply the number of person years remaining to be

lived by those alive at age x , divided by the number alive at that age. e  0  is life expectancy at birth and is the most common quantity of interest. Inequalities in life expectancy (at any age) can be measured by obtaining mortality rates for specific subpopulations (e.g., male and female, white and nonwhite, etc.), computing separate life tables for each group, computing standard errors for life expectancy (see Chiang, 1960) and conducting t-tests or related statistical tests comparing life expectancy at a given age. Although the single decrement life table is an incredibly useful tool, it is quite limited for use in contemporary studies of inequality for several reasons. First, there is a limit to how

82  Handbook of health inequalities across the life course specific one can be in examining inequality in life expectancy using basic life table construction methods. To compare populations while controlling for confounders, one must be able to obtain highly disaggregated mortality rate data, and such data are rarely available. Second, in many developed countries, interest has shifted from simply studying inequalities in years of life to be lived to inequalities in quality of life, and, in particular, years of disability-free life or years of life free from particular diseases and/or conditions. Both of these limitations of the single decrement life table have been remedied over the last several decades with the development of regression-based methods for producing “multistate” life tables. The central idea behind regression-based approaches to constructing life tables is that we use micro-level hazard regression models, such as the Gompertz model, first to estimate the influence of covariates on the risk of death and then produce smoothed age-specific hazard rates for a specified set of covariate values (a covariate profile). These rates can then be used to directly construct a life table for the given covariate profile. Values of covariates not of interest can be fixed at means or other meaningful values to produce marginalized estimates of life expectancy by variables of interest free from confounding (or compositional variation across populations). This strategy can be extended to construct multistate life tables, which extend the “state space” of the basic life table from simply “alive” and “dead” to additional, meaningful states such as “disabled” and “healthy.” The goal of analyses using multistate life tables is to estimate the number of remaining years an average individual can expect to live in each of the living states prior to death. In studies of life course inequalities in health, the most common state space investigated includes two living states – healthy and unhealthy – and one “absorbing state,” death, with “unhealthy” often defined by the presence of at least one ADL disability. In recent years, larger, more complex state spaces have been investigated (e.g., see Zang, Lynch, and West, 2020). A full discussion of multistate life table methods is beyond the scope of this chapter, but we will provide a brief discussion here (and see Preston, Heuveline, and Guillot, 2001). Whereas the basic life table only involves a single transition (from alive to dead), multistate life tables involve multiple possible transitions, probabilities for which are collected in transition matrices. In general, a matrix is simply an array of values collected into a single object with a fixed number of rows and columns. A transition probability matrix contains values that reflect the risk of transitioning between all possible states over a time interval. For example, a transition probability matrix for a state space that includes healthy (h), unhealthy (u), and deceased (d) states would look like:  phh P  x   P  x, x  k    puh  pdh  0

phu puu pdu  0

phd  pud  , (14) pdd  1

where P  x  is the transition probability matrix for transitioning between states in the age interval x to x + k , and pij is the probability of transitioning from state i to state j over the time interval. For example, pud would represent the probability of transitioning from unhealthy (u) to deceased (d). In this specification, the rows sum to 1, and, as indicated in the last row, the probability of transitioning from the deceased state to any living state is 0.

Methods for studying life course health inequalities  83 A multistate life table would be constructed from a collection of age-specific transition probability matrices much the same way that a basic life table is constructed: 1. Start with a “radix” population, l  0  . 2. Compute l  x  k   l  x  P  x  .

3. Compute L  x   1 / 2  l  x   l  x  k  .

4. Increment x by k and return to step 2 until the last age interval. The key difference between the single decrement and multistate computations is that the computations are matrix computations: l  x  is a vector – a matrix with a single column – indicat-

ing the number of individuals in each state at exact age x , and L  x  is a vector containing the

number of person-years spent in each state over a time interval, so that, when e  x  is com-

puted, e  x  is a vector that contains the number of years to be spent in each living state prior to death. While, as discussed above, investigating life course inequalities using single decrement life tables is difficult because of the limited ability to disaggregate mortality rates to a degree sufficient to control for a variety of confounders, this issue is even more problematic for the construction of multistate life tables because of the need to have multiple rates of transfers between states and not simply a single transition probability from living to dead. Further, detailed data on transitions is generally only available in panel survey data. Thus, the last three decades have seen the development of a variety of methods for generating multistate life tables from survey data. Some of the earliest of these methods were concerned simply with developing multivariate hazard models for creating smoothed transition probability matrices (e.g., Land, Guralnki, and Blazer, 1994). Efforts over the past two decades have focused on obtaining standard errors for the multistate estimates, allowing for the construction of interval estimates for statistical comparisons (e.g., see Lynch and Zang, 2022, for a review). In a recent application, for example, Zang, Lynch, and West (2020) examined regional disparities in years of life to be lived with diabetes, with diabetes and chronic conditions such as heart disease, and with diabetes, chronic conditions, and disabilities, using data on transitions between healthy, diabetic, conditions, and disability states obtained from the Health and Retirement Study (a long-term panel study of persons over age 50 in the U.S.). In most cases, estimation of parameters for developing smoothed transition probabilities has used discrete time multinomial logistic regression models or similar models that allow for the simultaneous estimation of multiple outcome transitions, where the discrete time multinomial logit model is simply an extension of the discrete time logit model described above.

METHODS FOR ANALYZING REPEATED MEASURES While many studies of life course inequality investigate disparities in the experience and timing of events, perhaps even more studies investigate patterns of change over time in repeated measures. Prior to the 1970s, there were relatively few panel data sets available to study actual within-person change associated with individual aging. Instead, many studies used cross-sectional data, often collected on only one occasion (i.e., single cross-sections).

84  Handbook of health inequalities across the life course In such data, patterns observed across age can only be assumed to reflect true age (i.e., maturation) patterns, because no individual change is actually observed: only persons at different ages at a single point in time are observed. Such data therefore cannot help us determine whether differences in health observed among persons at different ages reflect the consequences of individual maturation or simply represent static differences between birth cohorts With the collection of repeated cross-sectional data, more can be learned about life course inequalities using age-period-cohort (APC) models (see Yang and Land, 2013). However, the limitations of APC models are well known, and although their merits are often debated (see Yang and Land, 2013), it is clear that APC models cannot model true within-person change. At a minimum, selective survival is generally ignored in APC models because only survivors are surveyed at each cross-sectional occasion. For true models of individual change, one must use panel data. The collection of panel data began in earnest in the 1980s and continues today. In many developed countries, there are numerous panels studies that have collected repeated measures on individuals over extended periods of the life course, such as a decade or more. Studies of life course inequalities in social science that leverage panel data most often use three types of modeling strategies: fixed effects models, random effects (“mixed”) models, and latent class models. Fixed Effects Models Most panel data sets contain both time-varying measures, X, and time-invariant measures, Z. To simplify exposition here (to avoid triple subscripting), assume we have only one of each type of measure. Thus, xit is the time-varying variable for individual i at time point t, and zi is the time-invariant measure for individual i. Suppose we are interested in predicting some outcome, y, as a function of x, z, and possibly time t, so that: yit  b0  b1 xit  b2 zi  b3tit  eit , (15) where e follows the usual OLS assumptions of homoscedasticity and independence: e  N 0, e2 I . Notice that the time variable t is double subscripted – this is because the time





measure in the data is actually a time-varying variable (e.g., age). In this equation, there are two levels of measures represented by x, t, and z – two “within-person” measures ( x and t) and a “between-person” measure (z). We might therefore call this a “multilevel model,” although most analysts restrict that terminology to other types of equations in which there is nesting in the model parameters, as we discuss below. A second thing to notice is that the independence of errors assumption may be unwarranted. By the same token that z is time-invariant, it may be reasonable to believe that there are other, unobserved (and possibly unobservable) time-invariant variables that induce dependence in the error term, so that: eit  ui  wit , (16)

Methods for studying life course health inequalities  85 where wit follows the OLS assumption, but uit is a stable portion of eit that induces dependence. In that case, we could rewrite Equation 15 as: yit  b0  b1 xit  b2 zi  b3tit  ui  wit . (17) The problem with estimating the model in Equation 15 (or 17) using OLS is that ui is unobserved and becomes part of the error term, biasing standard errors downward as a consequence of violating the independence assumption. We can remedy this problem in a few ways. First, we can simply estimate the model using OLS regression but with “robust, clustered” standard errors. Second, we can incorporate the error dependence into the model and/or estimation process by adding some structure to the error term; this approach is the basis of random effects models, as discussed below. Both of these approaches require an extension of the exogeneity assumption that ui, and therefore eit, is uncorrelated with x. A third approach is to “sweep” ui out of the model altogether; this is the approach used in fixed effects modeling (as defined in econometrics; see Allison, 2009). Reconsider Equation 17, but assume for simplicity that we only observe each individual on two occasions (i.e., t=1,2). b2 ultimately only tells us how individuals differ from one another, since z does not change within individuals. In contrast, b1 potentially tells us how y changes within units across time AND how y varies between units that vary on x. Ideally, to make a causal argument about x’s influence on y, we need to be able to extract the within-unit change in y for a within-unit change in x from the between-group differences. We also need to rule out spuriousness due to unobserved factors that might explain the relationship between x and y (i.e., confounders, or sources of endogeneity). Note that we are assuming, as is common, that a static variable like z cannot be a cause: causes must be manipulable. One way to perform both tasks is to take the difference between Equation 17 at the two time points for each individual: yi1    yi 2 1 14 yi

 b0  b0  0

 b1 xi1  b1 xi 2  b1xi

 b2 zi  b2 zi  0

 b3ti1  b3ti 2  b3ti

 ui  ui  0

 wi1  wi 2 (18)  wi

We can rewrite the result of this subtraction as: yi  b0  b1xi  eit , (19) where b3∆ti has become a new intercept term ( b0 ) representing (average) change in y for everyone over time (e.g., aging), and eit is the change in the wit, which, because wit was assumed to be normal, eit is also normal. In this equation, all time-invariant variables have been eliminated, including both observed (e.g., z) and unobserved (e.g., u) variables. Thus time-invariant variables cannot be included in fixed effects models: they drop out. In the more general case in which individuals are observed on more than two occasions, we can deviate individual values of all variables from individual means, or we can include a dummy variable for each individual in the model. Both accomplish the same task, which is

86  Handbook of health inequalities across the life course to account for differences between individuals, so that the model reduces to calculating how differences in x within individuals across time relate to differences in y within individuals across time (Allison, 2009). The fixed effects model is often preferred over the random effects models discussed below and other models, because sweeping out all between-individual variation – observed, unobserved, and unobservable – makes causal claims regarding the association between x and y more plausible. That is, if all potential time-invariant confounders are eliminated, meeting the exogeneity assumption of the OLS regression model is much more likely than if we simply control on all observed variables because we almost certainly never measure all possible confounders of the relationship between x and y. This property makes the fixed effects model a good choice for studying causal effects of inequality. For example, Avendano (2012) used fixed effects when assessing whether the relationship between income inequality and infant mortality was causal or simply correlational. That said, there are several drawbacks to the fixed effects model that limit its usefulness in studying many questions of interest in life course studies of inequality. The first is mainly technical: the fixed effects model is less statistically efficient than the random effects models discussed in the next section because it discards much of the information (variation) in the data. Thus, there is a tradeoff between the fixed effect model’s ability to more plausibly meet the exogeneity assumption of regression, and therefore allow for causal interpretations of coefficient estimates, and its weakened power to do so. The Hausman specification test is therefore commonly used to determine whether coefficient estimates between the fixed and an equivalent random effects model differ enough to warrant the use of the more restrictive fixed effects model (Allison, 2009; Halaby, 2004). A second limitation of the fixed effects model is that it cannot, without modifications, allow us to investigate between-group differences, which is a staple of studies of inequality. Many of the sources of inequality scholars investigate concern fixed characteristics such as sex, race, and socioeconomic status, the latter of which is often measured by educational attainment or occupational status, both of which are often fixed after age 25 or so. To be sure, many scholars would argue that, while we often examine disparities across fixed characteristics, our ultimate concern is about the modifiable characteristics that these fixed characteristics proxy for, such as sexism or racism (Allison, 2009; Halaby, 2004). However, at the same time, we are generally limited in our ability to measure the underlying concepts these fixed characteristics represent. Thus, random effects models are more commonly used in studies of inequality. Random Effecs (or “Mixed”) Models The dummy variable approach to estimating the fixed effects model discussed above essentially allows each individual to have his/her own intercept, and is akin to rewriting Equation 17 as: yit  b0  ui   b1 xit  b2 zi  b3tit  wit (20)  bi 0  b1 xit  b2 zi  b3tit  wit . (21) Although we have included zi in this equation, its effect cannot be uniquely estimated: its effect is not identified if there is no restriction on ui. That is, we can adjust everyone’s value of bi0

Methods for studying life course health inequalities  87 by some amount, and adjust the coefficient b2 by an appropriate amount, and obtain the same model fit as without the adjustment. If we wish to include time-invariant variables, we can impose some structure on ui that helps distinguish it from z so that the effects of each can be uniquely estimated. In particular, we generally assume that ui  N  0,  and is part of the error term so that cor ui , x   0 and

cor ui , z   0 . That is, u has a normal distribution (rather than the unspecified distribution in the fixed effects model), and u is uncorrelated with other variables in the model (rather than being allowed to correlate with other variables as in the fixed effects model). This latter assumption is consistent with the exogeneity assumption in our usual regression model: we assume the error term is uncorrelated with observed variables in the model. While the upside to the random effects model is that we can include time-invariant variables in the model, there is a key downside: it does not sweep out sources of unobserved heterogeneity like the fixed effects model does. Thus, it is not helpful for establishing that any relationship between x and y is causal. Specifically, it does not help us reduce or eliminate the correlation between x and e any better than the OLS regression model. A random effects model can be written in several ways. Although ui is part of the error term, we can think of it as giving each individual a unique intercept, as in Equation 21, and we can “move” the time-invariant variables and the random effects error assumption to a “level 2” equation: yit  bi 0  b1 xit  b3tit  wit (22) bi 0  b.0  b.1 zi  ui (23) ui  N  0,  (24)





wit  N 0, 2 . (25) In this equation, b.0 is a “grand mean” around which the individual specific random effects, ui, are centered. Often, b.0 is written as b00 . When written this way, the model is often called a “hierarchical model,” because there are variables (and variances) at two different levels. At level 1, we have time-varying variables; at level 2, we have time-invariant variables (Raudenbush and Bryk, 2002). We also have variance components at each level: τ at level 2 and σ 2 at level 1. The “intraclass correlation coefficient” is a measure based on these two variance components that gives us the proportion of the total variance in y that is between groups vs. within groups. It is:  /    2 .





In addition to being called a hierarchical model, this model is also sometimes called a “random intercept” model, because the intercept is allowed to vary across individuals via the random effect, u. The model is also sometimes called a “mixed model,” because the model contains random effects, like u, but also fixed effects, like b.0 and b.1 in level 2 and b3 in level 1. Basically, under this terminology, a parameter that maintains a single value for the entire sample is a fixed effect, while any parameter that takes multiple values (like one per group) is a random effect.

88  Handbook of health inequalities across the life course The model can also be extended to handle more than two levels of hierarchy (like repeated observations of students nested in classes nested in schools nested in cities, etc.; see Raudenbush and Bryk, 2002). With the addition of nesting structures, there are additional “grand means” around which the unit-specific random effects are centered. In the case of repeated observations of students nested in classes, there would be a grand mean of the outcome value for each student, and a grand mean of the outcome value for each class. The model’s hierarchy is not limited to the intercept. For example, in a full-blown hierarchical model, b1 and b3 may be allowed to vary across groups. If b3 – that is, the effect of time – is allowed to vary across individuals, then the model is called a “growth model.” This model is sometimes referred to as a “random slope” model, because the slope b3 varies across individuals. Just as time-invariant variables may be included to predict differences in intercepts, time-invariant variables may be included to predict differences in slopes (see Lynch and Taylor, 2016). Such a model can be represented as: yit  bi 0  b1 xit  bi 3tit  wit (26) bi 0  b.0  b.1 zi  u0i (27) bi 3  b.3  b.3 zi  u3i . (28) In this equation it should be noted that b3 in Equation 22 has now become bi3 . This slope is then the summation of the “grand mean” slope for time and unit-specific random effects centered around this “grand mean” slope. The time-invariant variable zi now predicts between-person differences in both the intercept bi0 and the slope bi3 . Growth models estimate an average trajectory for the sample, but, more importantly, variation around the starting points (intercepts), and the change of y with time (slopes). These models are commonly used in contemporary studies that focus on life course inequalities (Pakpahan, Hoffman, and Kröger, 2017). Some studies, for example, focus on growth (or decline) in health disparities across age by social groupings captured by variables represented here as z (see Haas 2008, for example, examining the growth of functional limitations predicted by childhood socioeconomic experiences). The model presented here represents a linear relationship between time and the outcome y, but these models may be extended to include additional functional forms of the relationship between time and the outcome y with additional random slopes for these parameters. Latent Class Methods A growth model assumes that units all follow the same basic pattern in y across time, with unit-specific variation captured by the random effects, which follow a smooth (typically normal) distribution. Conversely, one may assume that unit-specific deviations are not distributed smoothly around a mean, but instead cluster in discrete groupings, and that patterns in y are not inherently similar in shape across the groupings. Latent class analysis (LCA) is a mixture modeling approach that identifies subpopulations (“classes”) that are constituted by discrete, clustered groupings that share similar patterns of outcomes (Lanza and Cooper, 2016; Lynch and Taylor, 2016).

Methods for studying life course health inequalities  89 The basic latent class model defines the disribution of responses, y (e.g., repeated or multiple measures for individuals), as a mixture of component distributions for y conditional on membership in each component (class), c. A generic likelihood function for a latent class model can be expressed as: L(Y | c; ) 

n

 K  f ( yi | ck ; k ) p i  ck ; k   , (29)    i 1  k 1 



where f ( yi | ck ;θ k ) is a distribution for y under each of the K classes in the population, ck (k=1…K), with unique parameter values, θ k , and p i  ck ; k  is a discrete probability reflecting the probability that individual i is in class k. Classes are defined by their unique values of θ k , so that ck and θ k are essentially interchangeable in meaning. It should be apparent that if there is only one class in the population (so K =1 and therefore p i  c1   1 ), this likelihood reduces to the generic likelihood for any common model. The general goals of latent class analysis are (1) to identify the number of ( K ), and characteristics of ( θ ), classes that exist in a population, and (2) identify factors that predict class membership. The strategy for determining the number of latent classes in a population is to first establish an assumed distribution f ( y |.) for the outcome variable(s) and then to estimate a series of models with increasing numbers of classes and select the model with the lowest Bayesian Information Criterion (BIC) value that still retains interpretability of classes. The BIC is a measure of model fit that is appropriate for comparing non-nested models and that provides a measure of how well a model fits the observed data while adjusting (penalizing) for the number of estimated parameters. BIC is commonly used in latent class models because models with differing numbers of latent classes are not nested, rendering likelihood ratio testing inappropriate, and increasing the number of latent classes increases the number of parameters estimated potentiating overfitting. The component distributions, f ( y | ck ;θ k ) , are chosen based on the data available and the assumed model structure, and the particular type of latent class model is defined by such. For example, if y consists of multiple cross-sectional measures of the presence or absence of a list of health conditions and one is interested in patterns of them, one may treat y as multinomially distributed and conduct a “latent profile analysis.” Thus, f ( yi | ck ; k )  where θ k is the vector of response probabilities p jk



 j  1J 



J



,

I y 1 p ij j 1 jk

in the kth class for the J health



condition measures, and I yij  1 is an indicator function indicating whether the respondent has health condition j. As another example, if y were a series of repeated, continuous measures for individuals, and we assumed individual-level linear growth, as described in the previous section, but we assume that there are distinct clusters of linear trajectories, such as low and flat, high and flat, low but increasing rapidly, etc., we might estimate a “latent class growth model” or a “growth mixture model,” each of which is akin to a regular growth model but they assume individual-level intercepts and slopes cluster instead of being smoothly, normally distributed (see Lynch and Taylor, 2016).

90  Handbook of health inequalities across the life course Once f ( y | ck ;θ k ) is selected, models with increasing numbers of classes are estimated until the BIC statistic indicates that gains in model fit are not justified by increasing the number of parameters further and the classes are substantively interpretable and distinct. The analyst then names the classes (based on values of θ k ) and examines the proportion of the population in each class. Although above p i  ck ; k  is clearly defined at the individual level, in many cases no information is included in p that distinguishes individuals from one another, so that p i  ck ; k  is identical for all units, thus representing a population proportion belonging to each class rather than individual probabilities. The membership of individuals in each class is unknown, but “posterior probabilities of class membership” can be computed for each individual based on their values of y using Bayes’ Theorem and individuals can be assigned to latent classes (Lynch and Taylor, 2006). Once individuals are assigned to classes in some fashion using the posterior probabilities, analysts usually estimate a multinomial logit or similar model using individual covariates to predict class membership and/or use class membership as a predictor of a distal outcome. Such an approach allows a researcher to examine whether characteristics, such as race/ ethnicity, gender, socioeconomic status, etc., pattern membership in latent classes, and thus pattern experiences with health outcomes/behaviors. For example, Pais (2014) estimated how early life exposures and occupational attainment predicted membership in latent classes of health impairment trajectories for Black and White adults. We note, however, that there are several methods for conducting these second-stage analyses, including some approaches that involve a single-step process that combines estimation of latent classes (and membership) and prediction of class membership/outcomes of latent class membership. Discussion of these approaches is beyond the scope of this chapter, but see Bakk and Kuha, 2021; Lynch and Taylor, 2016; van de Schoot et al., 2017 for more in-depth discussion. Latent Class Analysis is useful for producing meaningful and interpretable typologies of health outcomes over time. Furthermore, these typologies are derived from the data and not via decisions made by the reseracher. However, LCA may be still be biased by the researcher, particularly in the naming and interpretation of these derived classes. Yet, whether a researcher uses a mixed model approach or a latent class approach should not only depend on weighing the strengths and weaknesses of each method. Rather, the selection between these methods should depend on whether theory suggests that the population follows a singular trajectory with some variation (mixed model approach) or whether the population is made up of several subpopulation who are distinct in their trajectories (latent class approach) (see Lynch and Tyalor, 2016 for greater discussion).

CONCLUSIONS In this chapter, we have provided a broad overview of methods commonly used in studies of life course inequalities. Sources and outcomes of inequality are highly varied; thus, so are methods for studying them. We distinguished between studies of events and studies of repeated measures and discussed several specific methods used for each type of outcome. For events, we discussed hazard modeling methods and life table methods, while for repeated measures, we discussed fixed effects models, random effects models – including growth curve models, and latent class analsysis. For each of these more specific classes of methods, we described

Methods for studying life course health inequalities  91 some of the most common flavors. However, our overview barely scratched the surface of the full repertoire of models available. This chapter also does not address some of the well-known limitations of life course research, such as sample attrition or selective mortality that may downwardly bias estimation of health inequalities, particularly later in the life course. This chapter should therefore serve merely as a starting point for more in-depth study.

REFERENCES Allison, Paul D. 1984. Event history analysis: Regression for longitudinal event data. Sage University Paper Series on Quantitative Applications in the Social Sciences, 46. Newbury Park, CA: Sage. Allison, Paul D. 2009. Fixed effects regression models. Sage University Paper Series on Quantitative Applications in the Social Sciences, 160. Newbury Park, CA: Sage. Avendano, M. 2012. Correlation or causation? Income inequality and infant mortality in fixed effects models in the period 1960–2008 in 34 OECD countries. Social Science & Medicine 75, 754–760. https://​doi​.org/​10​.1016/​j​.socscimed​.2012​.04​.017 Bakk, Zsuzsa and Kuha, Jouni 2021. Relating latent class membership to external variables: An overview. Br J Math Stat Psychol 74, 340–362. https://​doi​.org/​10​.1111/​bmsp​.12227 Chiang, Chin L. 1960. A stochastic study of the life table and its applications: I. Probability distributions of the biometric functions. Biometrics 16, 618. https://​doi​.org/​10​.2307/​2527766 Cox, D. R. 1972. Regression model and life-tables. Journal of the Royal Statistical Society: Series B 34(2), 187–202. https://​rss​.onlinelibrary​.wiley​.com/​doi/​10​.1111/​j​.2517​-6161​.1972​.tb00899​.x Gavrilov, Leonid A. and Gavrilova, Natalia S. 2011. Mortality measurement at advanced ages: A study of the Social Security Administration death master file. North American Actuarial Journal 15(3), 432–447. Haas, S. 2008. Trajectories of functional health: The ‘long arm’ of childhood health and socioeconomic factors. Social Science & Medicine 66, 849–861. https://​doi​.org/​10​.1016/​j​.socscimed​.2007​.11​.004 Halaby, Charles N. 2004. Panel models in sociological research: Theory into practice. Annual Review of Sociology 30:507–544. Horiuchi, Shiro and Wilmoth, John R. 1998. Deceleration in the age pattern of mortality at older ages. Demography 35, 391–412. https://​doi​.org/​10​.2307/​3004009 Kalbfleisch, John D. and Prentice, Ross L. 2002. The Statistical Analysis of Failure Time Data. (2nd Ed.). Hoboken, NJ: Wiley. Land, Kenneth C., Guralnik, Jack M., and Blazer, Dan G. 1994. Estimating increment–decrement life tables with multiple covariates from panel data. Demography 31, 297–319. Lanza, Stephanie T. and Cooper, Brittany R. 2016. Latent Class Analysis for developmental research. Child Development Perspectives 10, 59–64. https://​doi​.org/​10​.1111/​cdep​.12163 Lynch, Scott M. and Brown, J. S. 2001. Reconsidering mortality compression and deceleration: An alternative model of mortality rates. Demography 38, 79–95. https://​doi​.org/​10​.1353/​dem​.2001​.0007 Lynch, Scott M. and Taylor, Miles G. 2016. Chapter 2 – Trajectory models for aging research, in: George, L. K. and Ferraro, K. F. (Eds.), Handbook of Aging and the Social Sciences (8th Ed.). San Diego: Academic Press, pp.  23–51. https://​doi​.org/​10​.1016/​B978​-0​-12​-417235​-7​.00002​-0 Lynch, S. M. and Zang, E. 2022. Bayesian multistate life table methods for large and complex state spaces: Development and illustration of a new method. Sociological Methodology 52(2), 254–286. https://​doi​.org/​10​.1177/​00811750221112398 Pais, J. 2014. Cumulative structural disadvantage and racial health disparities: The pathways of childhood socioeconomic influence. Demography 51, 1729–1753. https://​doi​.org/​10​.1007/​s13524​-014​-0330​-9 Pakpahan, E., Hoffmann, R. and Kröger, H. 2017. Statistical methods for causal analysis in life course research: An illustration of a cross-lagged structural equation model, a latent growth model, and an autoregressive latent trajectories model. International Journal of Social Research Methodology 20, 1–19. https://​doi​.org/​10​.1080/​13645579​.2015​.1091641 Preston, Samuel H., Heuveline, Patrick, and Guillot, Michel 2001. Demography: Measuring and Modeling Population Processes. Malden, MA: Blackwell Publishers.

92  Handbook of health inequalities across the life course Raudenbush, S. W. and Bryk, A. S. 2002. Hierarchical Linear Models: Applications and Data Analysis Methods (2nd Ed.). Thousand Oaks, CA: Sage. van de Schoot, Rens, Sijbrandij, Marit, Winter, Sonja D., Depaoli, Sarah, and Vermunt, Jeroen K. 2017. The GRoLTS-Checklist: Guidelines for Reporting on Latent Trajectory Studies. Structural Equation Modeling: A Multidisciplinary Journal 24, 451–467. https://​doi​.org/​10​.1080/​10705511​.2016​.1247646 Vaupel, James W. and Yashin, Anatoli I. 1985. Heterogeneity’s ruses: Some surprising effects of selection on population dynamics. The American Statistician 39, 176–185. https://​doi​.org/​10​.2307/​ 2683925 Yang, Yang and Land, Kenneth C. 2013. Age-Period Cohort Analysis: New Models, Methods, and Empirical Applications. Boca Raton, FL: CRC. Zang, Emma, Lynch, Scott M., and West, Jessie 2020. Regional differences in the impact of diabetes on population health in the United States. Journal of Epidemiology and Community Health doi: 10.1136/ jech-2020-213267

7. Causal inference based on non-experimental data in health inequality research Michael Gebel

INTRODUCTION Causal inference is one of the major aims of empirical research on health inequalities. Researchers often want to test causal hypotheses. In its simplest form, a causal hypothesis states that a specific variable causally affects another one, e.g. unemployment is detrimental to health, bad early health conditions negatively affect later health, or a specific health policy intervention improves health. The aim of causal inference must be distinguished from other aims, which are equally important, such as description or prediction. Description aims at describing distributions and associations of variables. It is implemented with uni-/bivariate analyses for a clearly defined target population or subgroups. There is no need to adjust for confounders because it is not the aim to separate the causal parts of an association from its non-causal parts (Hernán, 2018; Lundberg et al., 2021). Prediction aims at predicting the outcome variable. It is usually based on multivariate analyses, in which multiple explanatory variables are introduced in order to reach the best model fit of the outcome. Like description, prediction is agnostic about association having causal or non-causal origins. Good predictors do not need to be causes of the outcome; they can also stand in a non-causal relationship with the outcome. Hence, a predictor cannot be causally interpreted. There are three common problems of mixing up the different aims in applied research. First, by not making their goal explicit and remaining ambiguous, researchers often protect themselves from criticism, e.g. about the justification of selection of covariates. Second, researchers often refrain from explicitly stating their aim of causal inference and try to escape into the world of description and prediction (Hernán, 2018). However, as soon as causal statements are made, which often happens when researchers interpret their findings, appropriate methodological tools of causal inference are needed. This also includes bringing in expert knowledge in terms of clarifying the causal effect of interest and the overall causal structure (Hernán et al., 2019). Third, methodological tools for description and prediction are often misused to draw causal inferences (Moffitt, 2005). However, causal inference requires different methodological approaches, which will be the focus of this chapter. Randomized experiments are often seen as the “gold standard” for causal inference. However, the aim of causal inference can be taken regardless of whether the study is randomized or non-randomized (Hernán, 2018). Causal inference based on randomized experiment also rests on assumptions that can be violated. In addition, the external validity is usually low and there are often ethical and practical problems (Burtless, 1995). This applies particularly to research on health inequalities, when the causal variable of interest, such as social origin, education, work or certain life course events, such as marriage or childbirth, cannot be manipulated by researchers for practical or ethical reasons. Hence, this review focuses on causal 93

94  Handbook of health inequalities across the life course inference based on non-experimental data. It follows previous reviews on this topic in the research field of health (inequalities) (Hu et al., 2017; Matthay et al., 2020). Due to limited space, this chapter does not provide in-depth reviews of each method that can be found in the specialized literature. Instead, the aim is to give a simplified overview of conceptual frameworks and selected methods. After introducing conceptual frameworks of causality, different cross-sectional and longitudinal methods of causal analysis are covered in the following sections before, finally, a brief guidance for applied research is given.

CAUSAL HYPOTHESES, COUNTERFACTUALS AND CAUSAL GRAPHS Causal Hypotheses A proper causal study requires the clear ex-ante formulation of each causal hypothesis. This is necessary in order to evaluate whether the empirical estimands match the effect as stated in the hypothesis (Lundberg et al., 2021). Here, we focus on a simple causal hypothesis, which is about the total causal effect of a variable (D) on another variable (Y). “Simple” means that the hypothesis states only one causal effect, for which identification is already a complex task, as will be shown later. There are complex hypotheses, such as hypotheses on mediation, direct effects, interactions, and effect comparisons, that can be seen as combinations of simple hypotheses. Precise definitions of D and Y must be given, including their level of reference and reference group. For example, just stating a hypothesis on the causal effect of unemployment on individual mental health does not sufficiently clarify whether the interest is in the effect of national/ regional unemployment rate (= macro–micro hypothesis) or individual unemployment (= micro–micro hypothesis), and whether the comparison is to the reference of being employed or the reference of being inactive. Both variables D and Y must be distinct theoretical construct to avoid tautologies. For example, it does not make sense to state the causal effect between two items of an item battery of mental health if they aim at measuring the same overarching concept. Following a deductive approach, the causal hypotheses should be derived from theory. Particularly, the causal mechanisms of how D causally affects Y should be clarified, though they enter neither the explicit formulation of the hypothesis nor the empirical model. In addition, there is a need for theoretical knowledge on causal relationships in the overall system under study (see the section on ‘Causal Graphs’ below) (Hernán et al., 2019). This theory-guided causal approach stands in contrast to data-driven predictive approaches of searching empirical regularities, such as machine learning (see Kröger, Chapter 8 in this Handbook). In the following, two conceptual frameworks are introduced that provide clear definitions of a causal effect and potential biases. These frameworks can also be used to explain the logic of different methods of causal analysis. The Counterfactual Model of Causality Adopting experimental language, the counterfactual model of causality or potential outcome (PO) approach starts with the premise that almost each non-experimental setting can be

Causal inference based on non-experimental data in health inequality research  95 described in terms of a thought experiment (Rubin, 1974). In its simplest form, it defines the causal effect of a binary treatment1 D, on an outcome Y, which will be exemplified for the individual-level effect of higher education (Di = 1) versus not having higher education (Di = 0) on health (Y) for individual i. An individual faces two states of the world: having a higher education degree or not. Both states are equal except for the differences in the treatment status and its consequences. As they are defined at the same moment in time, one state of the world is observed and the other is not (= counterfactual). Irrespectively of the factual treatment status, potential outcomes are defined for both states. Yi1 measures health when having higher

education, whereas Yi0 measures health when not having higher education. The individual





causal effect is theoretically defined as Yi1  Yi0 but not observed, as one person cannot be observed simultaneously in both states (Holland, 1986). In applied research, the focus is often on average causal effects.2 In our example, the average treatment effect ATE  E Yi1  Yi0 is defined as the average causal effect of higher education





on health for a randomly chosen person from the target population. Alternatively, the average treatment effect on the treated ATT  E Yi1  Yi0|Di  1 measures the average effect of higher





education on health for the highly educated, whereas the average treatment effect on the non-treated ATNT  E Yi1  Yi0|Di  0 measures the (hypothetical) average effect of higher





education on health for the low educated. The ATT is the most relevant effect because it refers to those who are affected by the treatment (Gangl, 2010). While its first part E Yi1|Di  1 is





observable in terms of the health of high educated when having higher education, its second part E Yi0|Di  1 is counterfactual, as we do not know the health of a high-educated person if





she were low educated at the same moment in time. Research designs differ in their strategy to approximate the unknown counterfactual. Whereas longitudinal designs draw on observed outcomes of the treatment group before the treatment, cross-sectional designs use the observed outcome of the control group. In successful randomized experiments, the observed outcome of the control group equates with the unknown counterfactual, i.e. E Yi0|Di  0  E Yi0|Di  1 . Due to randomization the independence



 







assumption (IA) of the potential outcomes holds with regard to the treatment, i.e. Y 0 ,Y 1  D, such that the causal effects equate with the naive estimator NE  E Yi1|Di  1  E Yi0|Di  0  ATE  ATT  ATNT (Abadie and Cattaneo, 2018). In



 



non-experimental data, the NE produces biased results for the ATT and ATE (Winship and Morgan, 1999):

 

 



NE  ATT  E Yi0|Di  1  E Yi0|Di  0  

 

baseline bias

 



NE  ATE  E Yi |Di  1  E Yi0|Di  0  1      ATT  ATNT        0

baseline bias

causal effect heterogeneity bias

96  Handbook of health inequalities across the life course with π as the share of the treated. The baseline bias is the difference in the average outcome in the absence of treatment between treated and controls. It results from selection into the treatment based on observed and unobserved pre-treatment characteristics, which also affect the outcome in the absence of outcomes in the absence of the treatment. Baseline bias occurs in our example if being female or having good genetic health conditions increases the chances of obtaining higher education and if being female or having good genetic health conditions improves health also in the absence of higher education. The causal effect heterogeneity bias refers to differences in causal effects between the treatment and control group, represented by  ATT  ATNT  , multiplied by the proportion of

the control group which is represented in the multiplicative factor 1  . It results from observed and unobserved pre-treatment characteristics that causally affect the treatment and that interact with the treatment producing different causal effects for those treated or not. In our example, this happens if gender and genetic health condition causally affect the level of education and interact with education in their effects on health. Causal Graphs

Directed acyclic graphs (DAGs) are equivalent to the PO framework. They are also rooted in complex math but have the advantage of offering graphical tools (Pearl, 2009). Tennant et al. (2021) recently reviewed the use of DAGs in applied health research. Causal graphs give a visual representation of all aspects of the assumed data generation process that are relevant to the identification of the causal effect of interest. DAGs cannot be drawn based on empirical results of description and prediction because these tools cannot distinguish causal from non-causal relationships and directions of relationships. Instead, DAGs are built on theory and they can only be empirically supported by trustworthy results of previous empirical causal analysis. Basic elements of DAGs are nodes that are labeled by letters representing observed (solid circle) and unobserved (hollow circle) variables of any kind of measurement scale and distribution. A directed arrow between two variables visualizes the assumption of a causal effect between these variables. Effects are non-parametrically defined, i.e. they can take any functional form, and they are heterogeneous, i.e. they can vary with the levels of other variables. A missing arrow between two variables means that there is no causal effect between the two variables. Descendants are variables (in-)directly caused by a given variable, whereas ancestors are the (in-)direct causes of a variable. Another key concept is conditioning, which is the introduction of information about a variable by statistically controlling for a variable, stratifying on a variable or sample selection based on a variable (Elwert and Winship, 2014). Table 7.1 shows three basic configurations between three variables C, D, and Y. It helps in illustrating the implications of conditioning or not conditioning on a third variable C, when the interest is in the total causal effect of D on Y (Elwert and Winship, 2014). There is an unconditional association between D and Y if C is a mediator  D  C  Y  or a confounder  D  C  Y . Only “mediation” represents a total causal effect of D on Y and conditioning on C creates overcontrol bias. If C is a confounder, not conditioning on C induces confounding bias. If C is a collider  D  C  Y  there is a zero unconditional association and zero total causal effect of D on Y. In this configuration, conditioning on C creates collider bias, which is also called endogenous selection bias.

Causal inference based on non-experimental data in health inequality research  97 Table 7.1

Basic configurations between three variables C, D and Y Mediation/ Chain

D  C Y 

Common cause/ Fork

D  C Y 

Common outcome/ Inverted fork

D  C Y 

mediator

confounder

collider

Total causal effect of D on Y

≠0

=0

=0

Unconditional association (D,Y)

≠0

≠0

=0

Is the unconditional association causal?

Yes

No

Yes

Conditional association(A,B|C)

=0

=0

≠0

Is the conditional association causal?

No

Yes

No

Overcontrol bias

No

Collider bias

No

Confounding bias

No

C

Is there a bias when conditioning on C? Is there a bias when not conditioning on C?

Source: Own illustration.

In DAGs one can define different types of paths, as will be illustrated based on Figure 7.1, in which D represents the treatment and Y the outcome variable of interest. A path is defined as any sequence of arrows pointing in any (also changing) direction that connects D and Y (Elwert and Winship, 2014). A causal path is a path in which all arrows point in the same direction from D to Y, i.e. D → X 7 → Y and D → X 8 → Y . A non-causal path is a path between D and Y that is not a causal path, i.e. D  X 3  X 4  X 5  Y , D  X 4  X 5  Y and D  X 6  Y . Among the non-causal paths, a backdoor path (BDP) is defined as a non-causal path between D and Y that begins with a directed arrow that points to D, i.e. D  X 3  X 4  X 5  Y and D  X 4  X 5  Y . DAGs do not allow bi-directional causation or causal paths emanating from a variable that also terminate at the same variable. Causal identification in DAGs is defined as purging an observed association of all non-causal components such that only the causal effect of interest remains (Elwert and Winship, 2014). In our example, the observed association between D and Y has to be purged of the non-causal components (e.g. D  X 4  X 5  Y ) to isolate the causal effect of interest

Source: Own illustration.

Figure 7.1

Example of a DAG

98  Handbook of health inequalities across the life course (D → X 7 → Y and D → X 8 → Y ). Causal identification is discussed in the abstract ideal world with ideal data and separated from the estimation of causal effects. The process of estimation requires adequate measures for the theoretically defined variables. It also often involves making further assumptions, such as assumptions on the functional form of the relationship between variables. In the following, different methods of causal inference and their strategies and potential problems of identification and estimation are introduced.

CROSS-SECTIONAL METHODS DEALING WITH SELECTION ON OBSERVABLES Several cross-sectional non-experimental methods of causal inference rely on the identification assumption of selection on observables. In the counterfactual framework, it can be stated as the Conditional Independence Assumption (CIA):

Y

0



,Y 1  D | X 2 , X 3 , X k

Whereas the IA holds in a successful randomized experiment, the CIA postulates that independence can be created ex-post by conditioning on a set of observed variables X 2 , X 3 ,… X k . The PO approach is not very helpful in guiding researchers in choosing appropriate conditioning variables but DAGs are. The usual recommendation in the PO literature is limited to condition only on pre-treatment variables and variables that are not affected by the treatment. Here, the DAG framework can be a better guidance. The equivalence of the CIA is the backdoor criterion in the DAG framework. For the identification of the total causal effect of D on Y it requires conditioning on a set of observed variables X (which may be empty) that (i) are no descendants of D and that (ii) block all backdoor paths from D to Y (Elwert, 2013; Pearl, 2009). A path is already blocked if there is a collider on the path, and neither the collider nor its descendants have been conditioned on. Or, a path can be blocked by conditioning on a non-collider on the path (Elwert, 2013; Pearl, 2009). Returning to Figure 7.1, the BDP D  X 3  X 4  X 5  Y is already blocked because X4 is a collider. The BDP D  X 4  X 5  Y can be blocked by conditioning on X5. So, the total causal effect of D on Y can be identified just by conditioning on X5. One could also close the second BDP by conditioning on X4 but, without simultaneously conditioning on X5, this would reopen the first BDP because conditioning on a collider induces a correlation. The causal identification totally rests on X5 being observable. Otherwise, there is confounding bias due to selection on unobservables. In studies on health inequality researchers often include those variables as controls that are correlated with the outcome, i.e. X4 to X9. This logic is adequate if the aim is to predict the outcome variable but not for our aim of causal inference. Including X7 and X8 induces overcontrol bias, as our interest is in the total causal effect of D on Y. Including X6 induces collider bias. It is not a problem that X9 is an unobserved determinant of Y because there is no open BDP from D to Y. Cinelli et al. (2022) provide a helpful guide on the appropriate choice of control variables. A key challenge in applied research is the knowledge of the exact DAG. Though some testable implications can be derived (Elwert, 2013), the validity of the DAG cannot be fully

Causal inference based on non-experimental data in health inequality research  99 empirically tested (Hernán et al., 2019).3 Knowledge from theory and previous causal studies must be used instead. Time structure can ease arguments on causal order. One may also rule out constellations that are unlikely. Sometimes we do not need full knowledge. For example, conditioning on X5 would be still adequate if there were the additional arrows X 2 → X 3, X 3 → X 5 or X 7 → X 6 in Figure 7.1. Even if the DAG is not perfect, it makes researchers’ assumptions more explicit and open to criticism (Tennant et al., 2021). The methods of regression, matching and inverse probability weighting equally rely on the CIA and backdoor criterion and, hence, one should use the same conditioning set. They only differ in their estimation approach, which will be explained in the following. Regression and Regression Adjustment It is often denied that regression4 is a method of causal inference because it is mostly applied for predictive aims. However, combining it with causal knowledge can turn it into a method of causal inference (Hernán et al., 2019). It deserves the same attention as matching and weighting as they all rest on the same CIA. As regression textbooks mainly follow the logic of prediction, the key challenge is to learn how to perform regression analysis in the causal logic and to avoid confusions with the logic of prediction. Textbooks are full of discussions and tests of assumptions on statistical inference such as homoscedasticity, normal distribution or problems of imperfect or strong multicollinearity. In contrast, there is little information on the untestable exogeneity assumption on the error term u, E u|X1 , X 2 ,, X k   0 that is key to causal identification and equivalent to the CIA and the

BDP criterion. Misunderstandings also arise as u is usually defined as a residual Y  E Y |X  in the world of prediction, which is incorrect from a causal perspective (Chen and Pearl, 2013). Some textbooks interpret u as the unobserved variables other than X that affect Y but still miss the important requirement that the unobserved variables representing u are unaffected by X (Pearl et al., 2016). This clarifies that omitting mediators would not violate the exogeneity assumption and adding them would induce overcontrol bias. Textbooks often guide specification choices with the aim of getting the best model fit but, as explained above, choosing variables correlating with Y is wrong for causal inference. There is often an insufficient distinction between the roles of and attention paid to the interpretation of treatment and the control variables. Instead, there is the tendency to give equal attention to each variable as regression as a “multi-parameter approach” produces estimates for all X variables (Keele et al., 2020). However, the estimation of several causal effects in one model is usually infeasible (Gangl, 2010). Different sets of control variables might be needed for each causal effect of interest. For example, in Figure 7.2, we control for social origin when estimating the causal effect of education on health to tackle confounding bias due to the open BDP via social origin. In contrast, we do not control for education when estimating the causal effect of social origin on health because this would induce overcontrol bias as education is a mediator for the causal effect of social origin on health. A prime example is Bartram (2021) explaining the proper choice of controls in regression analysis on subjective well-being. Regression analysis usually makes the restrictive assumption of a homogenous causal effect, which enforces equality between the ATE, ATT and ATNT. However, there is the less-known version of Regression Adjustment (RA), which allows for heterogeneous causal

100  Handbook of health inequalities across the life course

Source: Own illustration.

Figure 7.2

Simplified DAG for the causal effect of education on health

effects building on the PO framework. Whereas the PO framework is nonparametric, regression imposes linearity and separability of the observed controls X and the error term u: Yi0   0 X i  ui0 � � � � if � � � Di  0 Yi1   1 X i  u1i � � � � if � � � Di  1 Making use of the relationship between the observed outcome and the potential outcomes Yi  DiYi1  1  Di Yi0 and rearranging terms yields the RA model









Yi   0 X i   1   0 X i Di  ui0  ui1  ui0 Di As in the standard regression model, the first term controls for observed baseline differences. The innovation is the second term that controls for observed causal effect heterogeneity by including interactions of the treatment with each control variable. The error term has two components requiring E ui0|X , D  0, which is violated if there is baseline bias due to unob-

  servables and E u  u  D |X , D   0, which is violated if there is causal effect heterogeneity 1 i

0 i

i

bias due to unobservables. Returning to our example above in the section on “The Counterfactual Model of Causality” we would be able to control for gender to account for observed baseline bias due to gender and for an interaction term between gender and the treatment of higher education to account for observed causal effect heterogeneity bias due to gender. However, our estimates of the causal effect of education on health would still suffer from unobserved baseline bias and unobserved causal effect heterogeneity bias due to genetic health condition because (in most datasets) genetic health condition is an unobserved variable.

Causal inference based on non-experimental data in health inequality research  101 Matching The most popular matching approach is Propensity Score Matching (PSM). It is widely applied in health research such as Eyjólfsdóttir et al. (2019)’s study on the effect of prolongation of working life on mortality and health in older adults. Rosenbaum and Rubin (1983) showed that if the CIA holds conditional on X , conditioning





on the propensity score (PS) P  X  is sufficient, i.e. Y 0 ,Y 1  D | P  X . The PS is defined as the conditional probability of receiving the treatment, i.e. P  X  = P  D  1 X . It reduces the multidimensional X into one dimension measuring the similarity of units. The PS is usually estimated using a parametric model in the first stage. A common support condition can be imposed, which excludes treated and controls in areas where the opposite group is not or only weakly represented. This improves the comparability of treated and controls but may also reduce external validity. The second stage involves a non-parametric outcome estimation. To estimate the ATT for each unit i of the treatment group (TG) an estimate of the counterfactual outcome is subtracted from the observed outcome:

 ATT 



iTG

Y   1 i

W Y0 j CG ij j



The counterfactual outcome is constructed based on observed outcomes of units j of the control group (CG) which are weighted with the matching weight Wij that are defined by the respective matching algorithm (Gangl, 2015). For example, in the case of 1-nearest-neighbor matching the control group unit with the most similar PS is chosen as a matching partner, acting as the “statistical twin”. The number of neighbors can be increased. There are also algorithms such a kernel matching that use all control units. The performance of different algorithms can be assessed by checking the balancing of control variables between the treatment and matched control group in their means, higher moments, or distributions. Statistical inference based balancing tests such as t-tests or Pseudo-R² tests, which have been propagated in the initial matching literature (Caliendo and Kopeinig, 2008), are not appropriate because balancing is a sample property (Gangl, 2015). Balancing problems can be solved by using different matching algorithms, adapting the common support condition, changing the functional form of control variables, or changing the PS link function. Balancing tests should never be misused for choosing control variables, as this must be done using the backdoor criterion guided by theory. There are several matching alternatives to PSM. For example, exact matching builds pairs based on identical values of X. However, this is infeasible if there are many, especially continuous, variables. Hence, exact matching is often just used complementary to PSM to guarantee perfect balancing of one or more important control variables. A related alternative is coarsened exact matching (CEM), which theoretically or empirically categorizes variables along strata and performs comparisons within strata. CEM has been advertised by King and Nielsen (2019) by demonizing PSM. However, these arguments have been critically reflected by Jann (2017) and CEM has become criticized as well (Black et al., 2020).

102  Handbook of health inequalities across the life course Regression and matching both rest on the CIA and only differ in estimation. So, what are the (dis-)advantages of each method? Even if using a model fully saturated in all values of the adequate control variables, parametric regression produces biased estimates (Abadie and Cattaneo, 2018). Matching overcomes this estimation problem with its non-parametric outcome equation. Matching implicitly accounts for individual causal effect heterogeneity, whereas standard regression assumes homogeneous effects and only RA partially accounts for effect heterogeneity. The nonparametric outcome equation of matching simplifies estimation and interpretations with respect to categorical outcomes, which is often a challenge in regression. In contrast, regression has clear practical advantages in dealing with multi-categorical and continuous treatments, which are more demanding in matching. Whereas matching as a “nuisance-parameter approach” gives only the treatment effect, regression as a “multi-parameter approach” provides effect estimates for the treatment effect and the effects of control variables. Being confronted with so many effect estimates in regression analysis, researchers often do not clearly distinguish between treatment and control variables, give incorrect causal interpretations of control variable effects, and do inadequate effect comparisons (Keele et al., 2020). Inverse Probability Weighting Similarly to PSM, Inverse Probability Weighting (IPW) is a two-step procedure that uses the PS in a weighting formula. The first stage estimation of the PS is identical to PSM. The difference occurs in the second stage, where the PS enters as a weighting factor in a mean comparison, which can be estimated in a weighted bivariate regression. For example, ATT is estimated as 1  ATT  n



n

DiYi 

i 1

1 n



X  P i i 1  Di Yi X  i 11  P n

i

i

An IPW research example is Popham and Iannelli (2021)’s study on the effect of comprehensive versus selective education systems on health inequalities. Adopting the idea of RA, there is also an Inverse Probability Weighting Regression Adjustment (IPWRA) version of IPW. It adds the control variables once again to the two potential outcomes equation, which turns into a weighted multiple linear regression with control variables and interactions of each control variable with the treatment. There is mixed evidence on how IPW compares to matching. Frölich (2004) finds that IPW had worse performance compared to all matching estimators considered in the simulation. In contrast, Busso et al. (2014) show that IPW is much more effective than suggested by Frölich (2004). They conclude that IPW is competitive with the most effective matching estimators when overlap is good, and that matching is only more effective when overlap is sufficiently poor.

Causal inference based on non-experimental data in health inequality research  103

CROSS-SECTIONAL METHODS DEALING WITH SELECTION ON UNOBSERVABLES The cross-sectional methods of the previous section suffer from confounding bias due to unobservables if there is selection on unobservables. This happens if there is an unblocked BDP that cannot be blocked because the respective variable is unobserved. This section gives a brief introduction to cross-sectional methods that can deal with selection on unobservables. Instrumental Variable Estimator In contrast to many textbooks, the identifying assumptions of the instrumental variables (IV) estimator are illustrated here based on DAGs. In the example illustrated in Figure 7.3 the methods of the previous section fail because the BDP D  X  Y can be blocked by conditioning on X but not the BDP D  U  Y because U is unobserved. However, the IV estimator can be implemented because Z is an IV if the DAG is correct. The first testable assumption (“IV relevance”) is that Z has an effect on D. It is usually assumed that Z has a causal effect on D but a “surrogate”/“proxy” IV is also sufficient, which is non-causally associated with Z (Hernán and Robins, 2020; Morgan and Winship, 2015). The second untestable assumption (“IV exogeneity”) is that there is no path from Z to Y that does not run via D. Testing whether Z has an effect on Y after controlling for D fails because D is a collider. The IV estimator is implemented in a two-stage estimation. First, the treatment is regressed on the IV. Second, the outcome variable is regressed on the predicted values of the first stage. A research example is the study of Kemptner et al. (2011) that uses changes in compulsory schooling laws as an IV (Z) in the estimation of the effect of years of education (D) on health-related behavior (Y). DAGs can also be used to illustrate how an invalid IV can be transformed into a valid IV. Figure 7.4 gives an example, where the second IV assumption is violated. There are two paths between Z and Y that do not run via D: Z → H 1 → Y and Z  H 2  Y . However, by additionally conditioning on H1 and H2 in both the selection and outcome equation, the invalid IV

Source: Own illustration.

Figure 7.3

DAG with a valid IV

104  Handbook of health inequalities across the life course

Source: Own illustration.

Figure 7.4

DAG with an invalid IV

Z can be transformed into a valid one. This is called the “conditional” IV estimator (Brito and Pearl, 2002). In the PO framework it can be shown that the classical IV estimator is biased if there is unobserved causal effect heterogeneity bias. Angrist et al. (1996) introduced the modern IV interpretation as the Local Average Treatment Effect (LATE). The LATE only gives the ATE for the specific subgroup of units who changed their treatment status due to a change in the instrument. Thus, the gain in internal validity comes at the costs of lower level of external validity and statistical power (Hu et al., 2017; Matthay et al., 2020). For more details on IV see Hoffmann and Doblhammer, Chapter 9 in this Handbook. Regression Discontinuity Design The Regression Discontinuity Design (RDD) requires an assignment variable A that is related to treatment D in a strongly discontinuous deterministic manner. All variation in this variable must be exogenous. In its simplest form, it is assumed that the relationship between the outcome Y and A would be continuous without treatment and if there is a jump in the relationship between Y and A at the discontinuity threshold, it is attributed to D. The identification focuses on comparing units around the cut-off value, which are assumed to be indistinguishable except for the treatment as in the case of an experiment. If the treatment assignment is imperfect, the RDD becomes fuzzy and there is an analogy of A to an IV (Abadie and Cattaneo, 2018). A research example is the study of Zhong (2016) who uses the Chinese Cultural Revolution as a negative educational shock in an RDD design when estimating the effect of education on health. The RDD identifies the local causal effect at the threshold only, which limits the external validity of this method. The RDD is biased if there is another (omitted) variable that also jumps discontinuously at the same threshold. For a detailed introduction, see Cattaneo et al. (2019).

Causal inference based on non-experimental data in health inequality research  105 Table 7.2

Observed outcomes in a longitudinal design

D = 0 (control group) D = 1 (treatment group)

t0

t1

Y0t0

Y0t1

Y0t0

Y1t1

Source: Own illustration.

LONGITUDINAL METHODS A longitudinal design is often seen as a panacea for causal inference in non-randomized settings. The broad definition of a longitudinal design means the retrospective or prospective repeated measurement of the outcome. The strict definition of a “within”-longitudinal design additionally requires that there is a time-varying treatment variable and one unaffected period sample prior to the treatment, in which the treatment has no causal effect, i.e. ruling out anticipation effects.5 Then, within-estimators can be used bringing true advantages in causal inference. For example, if the health outcome is repeatedly observed over time but the treatment is time-constant (e.g. social origin) this brings no advantages for the causal identification compared to cross-sectional methods. Only if the treatment is time-varying, e.g. employment status, and if the health outcome is measured in at least one period prior to the treatment having an effect, i.e. ruling out an anticipation effect in at least one period, the within-estimators can be applied that will be introduced here. However, even in a within-longitudinal design the problem of causal inference is not automatically solved. Table 7.2 illustrates this in the PO framework, which will be exemplified with the example of estimating the causal effect of becoming unemployed (vs. remaining employed) on health. The treatment group is employed at t0 and unemployed at t1. Thus, we can observe their health in the status of employment at t0, which is the baseline outcome Yi0,t 0 , and their health in the status of unemployment at t1, which is the outcome with the treatment Yi0,t1. The causal effect is still defined as the difference in potential outcomes for the same person at the same post-treatment time Yi1,t1 − Yi0,t1 . There is still the fundamental problem of causal inference as we do not know the health of unemployed persons at t1 if they were employed at t1. For the control group of persons who were employed at t0 and t1 we can observe their baseline health Yi0 both at t0 and t1. The Before–After Estimator Two basic estimators are introduced that use the within-logic in causal longitudinal designs. They are also applicable to time-series cross-sectional (TSCS) data when the treatment and within-comparison is at an aggregated level. However, we focus on the design where the treatment is defined at the same level as in our example. The Before–After (BA) estimator ignores the control group and takes only the difference in the observed outcomes of the treatment group:





BA  E Yi1,t1  Yi0,t 0|D  1

106  Handbook of health inequalities across the life course Thus, it measures the observed change in health between t0 and t1 for the treated who were employed at t0 and unemployed at t1. The strength of the BA estimator is that differencing eliminates group-specific fixed effects in TSCS designs and even individual-specific fixed effects in panel designs. A fixed effect subsumes all time-constant observed and unobserved confounding variables. Hence, confounding bias, which is due to open BDP that are based on time-constant observed and unobserved confounding variables, is eliminated. The causal identification assumption postulates a zero baseline trend for the treatment group





E Yi0,t1  Yi0,t 0|D  1  0 . Applied to our example, we had to assume that the health of treated persons becoming unemployed at t1 would not have changed if they had remained in employment. This assumption is untestable because it contains the counterfactual E Yi0,t1|D  1 , i.e.





the health of an unemployed person at t1 if she were employed at t1. This counterfactual is approximated by the observed baseline outcome E Yi0,t 0|D  1 , i.e. the health of the unem-





ployed person in t0 when she was employed. This assumption is represented by the grey dotted line in Figure 7.5a. This is a strong assumption as there might be a trend in baseline health between t0 and t1 for the treatment group due to period effects (e.g. a pandemic deteriorating health in the state of employment) or life course effects (e.g. aging deteriorating health in the state of employment). Thus, any kind of confounding bias due to time-varying observed and unobserved confounders will remain. The Difference-in-Differences Estimator The Difference-in-Differences (DID) estimator partly overcomes the shortcomings of the BA estimator by using the observed outcomes from the control group (Lechner, 2011). It compares the observed outcome change in the treatment group to the observed outcome change in the control group:



 

DID  E Yi1,t1  Yi0,t 0 | D  1  E Yi0,t1  Yi0,t 0|D  0



The first component is the BA estimator. Thus, the DID estimator also eliminates group-specific fixed effects in TSCS designs and individual-specific fixed effects in panel designs. The double-differencing additionally eliminates time trends in the baseline outcome that are equal between the treatment and the control group, e.g. a pandemic or ageing affecting both groups equally. The causal identification assumption postulates a common baseline trend for the treatment and the control group E Yi0,t1  Yi0,t 0|D  1  E Yi0,t1  Yi0,t 0|D  0 . Applied to our example, we



 



had to assume that the health of treated persons becoming unemployed at t1 would have developed in the same way as the health of the control group if they had remained in employment. Thus, the counterfactual is approximated by extrapolating the trend in baseline outcomes observed for the control group to the treatment group. This is represented by the dashed grey line having the same slope as the dashed black line in Figure 7.5b. Again, this assumption is untestable because it contains the counterfactual E Yi0,t1|D  1 . The common baseline trend





Causal inference based on non-experimental data in health inequality research  107

Source: Own illustration.

Figure 7.5

Graphical illustration of (a) BA estimator and (b) DID estimator

assumption between t0 and t1 can be violated if period and life course effects affected the baseline health trends of treated and controls differently between t0 and t1. For example, a pandemic or aging may deteriorate the health of employed persons and unemployed persons if they were employed differently. Empirical tests of the common baseline trend assumption are widespread. They are usually performed by using information of another pre-treatment period t −1, which allows observing the health trends between t −1 and t0 for controls and treated who are all employed at both dates. Although a validation of this test may increase trust in the identification assumption, it is not a perfect test because it looks at trends at different dates (t −1, vs. t0 instead of t1 vs. t0). The common baseline trend assumption can be made more plausible in a DID-Matching approach by using matching to form a control group. For example, see Gebel and Voßemer

108  Handbook of health inequalities across the life course (2014)’s study on the effect of employment transitions on health. Matching is performed based on variables to block BDP between the treatment variable and the baseline trend in outcomes. In practice, DID matching is usually done based on control variables measured in pre-treatment periods in order to reduce potential risks of overcontrol and collider bias. When matching on time-varying variables there must be good arguments that they represent time-varying confounders and not time-varying mediators or colliders. In our example of the effect of unemployment on health one may argue that death of loved ones is a time-varying confounder. Unless we assume unemployed people were paying medical bills for a terminally ill partner, it is likely that death of a loved one is a rather a cause than a consequence of own unemployment. Another misinterpretation of DID is that it is sometimes claimed that DID rest on the assumption that the initial outcome levels of the treatment and control group are identical. However, this is not needed. There are even situations in which conditioning on pre-treatment outcomes induces collider bias (Lechner, 2011). The BA and DID(-Matching) estimators can be repeatedly applied to cover other period comparisons, e.g. t2 vs t0 or t3 vs t0. This allows the estimation of an impact function, i.e. how the effect varies over time since the treatment. The periods may also refer to pre-treatment periods to generate estimates of anticipation effects. There are also extensions of the DID estimator, such as applying a triple differencing, which are, however, beyond the scope of this review (Olden and Møen, 2022). This also applies to recent advanced debates on DID (Roth et al., 2022). The BA and DID estimators have an analogy to the first difference (FD) but also the fixed effect (FE) panel estimators (Brüderl and Ludwig, 2015). Both approaches require a within-longitudinal design. The discussion of the equivalence of panel estimators and BA and DID estimators is beyond the scope of this overview chapter.

GUIDANCE FOR APPLIED RESEARCH This chapter introduced different methods of causal inference. All methods rely on critical untestable identification assumptions (Matthay et al., 2020). In practice, each assumption is probably never fulfilled perfectly. Hence, there is no panacea for causal inference. This often let researchers give up their aims of causal inference and stick to the aims and tools of description and prediction. However, as explained in the introductory section, this would also require not making any causal statement anymore, which is never practiced. The guidance for applied research is to try to get as close as possible to the causal effect of interest and try to discuss the necessary assumptions and strategies to increase their plausibility as transparently as possible. The first crucial step is to clearly formulate the causal hypotheses of interest ex ante and to choose an identification and estimation strategy that targets the causal effect of interest (Lundberg et al., 2021). Depending on the causal hypothesis and the available data, the assumptions of one method might be more convincing than those of others, which may guide the method choice. Potential trade-offs with regard to external validity should also be considered (Moffitt, 2005). It is essential that the assumptions of causal identification and estimation of the chosen method are clearly formulated, critically discussed and that potential biases are highlighted.

Causal inference based on non-experimental data in health inequality research  109 This chapter emphasized the use of DAGs as graphical tools for applied research to illustrate assumptions of causal identification and potential biases of different methods. DAGs are often misunderstood as a specific method and criticized, although they just represent a common conceptual framework for different methods. For example, failures to fulfill the backdoor criterion in regression analysis or to find an instrumental variable do not represent a limitation of DAGs but of the respective method. Similarly, the inability to draw a DAG for the causal effect of interest is not a limitation of DAGs. It just signals the researcher’s inability to clearly formulate the assumptions of causal identification and potential biases of a specific method given the causal hypothesis of interest and the data. It also reveals the lack of theory and trustworthy previous causal studies in the research field. Improvements can also be reached via data collection informed by causal inference. Data collection efforts should not only focus on generating appropriate measures of the treatment, moderator, mediator, and outcome variables of interest but particularly of confounding variables. In health research, this especially refers to confounders of health that are often not collected in surveys such as health pre-conditions. Introducing the time dimension via retrospective and prospective designs can be helpful to justify the causal order of variables as long as anticipation effects can be ruled out. If such data is collected, careful choices between retrospective and prospective designs must be made. Retrospective reports might be biased by current views of the respondent. Prospective designs are subject to attrition and mortality bias, which may induce endogenous sample selection bias. Data collection projects can also improve the applicability of IV and RDD designs by directly collecting instrumental and scoring variables.

NOTES 1. Starting from Holland (1986) there is a discussion on which variables can be a treatment. In its strictest form, the manipulability of the treatment is required, which excludes defining gender or ethnicity as treatments. 2. Comparing distributions of potential outcomes quantile treatment effects can also be identified (Abadie and Cattaneo 2018). 3. However, once the conditioning set is chosen, the adequate functional form of variables can be chosen empirically. 4. The following remarks also apply to structural equation models, which add measurement models to the regression framework. 5. Just measuring the treatment prior to the outcome, e.g. when estimating the effect of early careers on later life health that is only measured once, is still a cross-sectional design.

REFERENCES Abadie A. and Cattaneo M.D. (2018) Econometric methods for program evaluation. Annual Review of Economics 10(1): 465–503. Angrist J.D., Imbens G.W. and Rubin D.B. (1996) Identification of causal effects using instrumental variables. Journal of the American Statistical Association 91(434): 444–455. Bartram D. (2021) Cross-sectional model-building for research on subjective well-being: Gaining clarity on control variables. Social Indicators Research 155: 725–743.

110  Handbook of health inequalities across the life course Black B.S., Lalkiya P. and Lerner J.Y. (2020) The trouble with coarsened exact matching. Northwestern Law & Econ Research Paper. Forthcoming, available as preprint at SSRN (last accessed on 7 January 2022): http://​dx​.doi​.org/​10​.2139/​ssrn​.3694749 Brito C. and Pearl J. (2002) Generalized instrumental variables. Proceedings of the Eighteenth Conference on Uncertainty in Artificial Intelligence (UAI2002): 85–93. Brüderl J. and Ludwig V. (2015) Fixed-effects panel regression. In: Best H. and Wolf C. (eds) The SAGE handbook of regression analysis and causal inference. Los Angeles: Sage, pp. 327–357. Burtless G. (1995) The case for randomized field trials in economic and policy research. Journal of Economic Perspectives 9(2): 63–84. Busso M., DiNardo J. and McCrary J. (2014) New evidence on the finite sample properties of propensity score reweighting and matching estimators. Review of Economics and Statistics 96(5): 885–897. Caliendo M. and Kopeinig S. (2008) Some practical guidance for the implementation of propensity score matching. Journal of Economic Surveys 22(1): 31–72. Cattaneo M.D., Idrobo N. and Titiunik R. (2019) A practical introduction to regression discontinuity designs: Foundations. Cambridge, New York, NY, Port Melbourne, VIC, New Delhi, Singapore: Cambridge University Press. Chen B. and Pearl J. (2013) Regression and causation: A critical examination of econometrics textbooks. Real-World Economics Review 65: 2–20. Cinelli C., Forney A. and Pearl J. (2022) A crash course in good and bad controls. Sociological Methods and Research. Elwert F. (2013) Graphical causal models. In: Morgan S.L. (ed.) Handbook of causal analysis for social research: Dordrecht: Springer, pp. 245–273. Elwert F. and Winship C. (2014) Endogenous selection bias: The problem of conditioning on a collider variable. Annual Review of Sociology 40: 31–53. Eyjólfsdóttir H.S., Baumann I., Agahi N., et al. (2019) Prolongation of working life and its effect on mortality and health in older adults: Propensity score matching. Social Science & Medicine 226: 77–86. Frölich M. (2004) Finite-sample properties of propensity-score matching and weighting estimators. Review of Economics and Statistics 86(1): 77–90. Gangl M. (2010) Causal inference in sociological research. Annual Review of Sociology 36(1): 21–47. Gangl M. (2015) Matching estimators for treatment effects. In: Best H. and Wolf C. (eds) The SAGE handbook of regression analysis and causal inference. Los Angeles: Sage, pp. 251–276. Gebel M. and Voßemer J. (2014) The impact of employment transitions on health in Germany: A difference-in-differences propensity score matching approach. Social Science & Medicine 108: 128–136. Hernán M.A. (2018) The C-word: Scientific euphemisms do not improve causal inference from observational data. American Journal of Public Health 108(5): 616–619. Hernán M.A., Hsu J. and Healy B. (2019) A second chance to get causal inference right: A classification of data science tasks. CHANCE 32(1): 42–49. Hernán M.A. and Robins J.M. (2020) Causal inference: What if. Boca Raton: Chapman & Hall/CRC. Holland P.W. (1986) Statistics and causal inference. Journal of the American Statistical Association 81(396): 945–960. Hu Y., van Lenthe F.J., Hoffmann R., et al. (2017) Assessing the impact of natural policy experiments on socioeconomic inequalities in health: How to apply commonly used quantitative analytical methods? BMC Medical Research Methodology 17(1): 68. Jann B. (2017) Why propensity scores should be used for matching: Presentation at Stata Meeting 2017. https://​www​.stata​.com/​meeting/​germany17/​slides/​Germany17​_Jann​.pdf. Keele L., Stevenson R.T. and Elwert F. (2020) The causal interpretation of estimated associations in regression models. Political Science Research and Methods 8(1): 1–13. Kemptner D., Jürges H. and Reinhold S. (2011) Changes in compulsory schooling and the causal effect of education on health: Evidence from Germany. Journal of Health Economics 30(2): 340–354. King G. and Nielsen R. (2019) Why propensity scores should not be used for matching. Political Analysis 27(4): 435–454. Lechner M. (2011) The estimation of causal effects by difference-in-difference methods. Foundations and Trends in Econometrics 4(3): 165–224.

Causal inference based on non-experimental data in health inequality research  111 Lundberg I., Johnson R. and Stewart B.M. (2021) What is your estimand? Defining the target quantity connects statistical evidence to theory. American Sociological Review 86(3): 532–565. Matthay E.C., Hagan E., Gottlieb L.M., et al. (2020) Alternative causal inference methods in population health research: Evaluating tradeoffs and triangulating evidence. SSM - Population Health 10: 100526. Moffitt R. (2005) Remarks on the analysis of causal relationships in population research. Demography 42(1): 91–108. Morgan S.L. and Winship C. (2015) Counterfactuals and causal inference: Methods and principles for social research. Cambridge: Cambridge University Press. Olden A. and Møen J. (2022) The triple difference estimator. The Econometrics Journal 00: 1–23. Pearl J. (2009) Causality. Cambridge: Cambridge University Press. Pearl J., Glymour M. and Jewell N.P. (2016) Causal inference in statistics: A primer. Chichester, West Sussex: Wiley. Popham F. and Iannelli C. (2021) Does comprehensive education reduce health inequalities? SSM – Population Health 15: 100834. Rosenbaum P.R. and Rubin D.B. (1983) The central role of the propensity score in observational studies for causal effects. Biometrika 70(1): 41–55. Roth J., Sant’Anna P.H.C., Bilinski A., et al. (2022) What’s trending in difference-in-differences? A synthesis of the recent econometrics literature. Version v2 from 14 January 2022. https://​arxiv​.org/​abs/​ 2201​.01194 Rubin D.B. (1974) Estimating causal effects of treatments in randomized and nonrandomized studies. Journal of Educational Psychology 66(5): 688–701. Tennant P.W.G., Murray E.J., Arnold K.F., et al. (2021) Use of directed acyclic graphs (DAGs) to identify confounders in applied health research: Review and recommendations. International Journal of Epidemiology 50(2): 620–632. Winship C. and Morgan S.L. (1999) The estimation of causal effects from observational data. Annual Review of Sociology 25: 659–706. Zhong H. (2016) Effects of quantity of education on health: A regression discontinuity design approach based on the Chinese Cultural Revolution. China Economic Review 41: 62–74.

8. Predictive machine learning approaches – possibilities and limitations for the future of life course research Hannes Kröger

INTRODUCTION TO ML ALGORITHMS The application of machine learning (ML) in social sciences and life course research has increased over the last decade. Still, most studies in the area of health inequalities in the life course do not use this portfolio of analysis techniques (at first sight). I will describe the basics of the ML approach and evaluate how and to what degree it can be made helpful for life course research and the analysis of health inequalities. Throughout the manuscript, I will use the terms “traditional” and conventional” statistics for statistical methods applied from a non-ML framework, knowing that this is not a clearly defined set of methods but a different approach. Throughout the manuscript, I will use several terms, highlighted in italic, which might not be familiar from a traditional statistics perspective. Table 8.1 gives a non-exhaustive overview of such terms and highlights more well-known equivalents. It’s All About Prediction In this chapter I will focus solely on predictive machine learning methods, and when I refer to ML, I mean predictive ML (and ML for short). However, it should be noted that these methods can be integrated into traditional causal analytical models as well, depending on the overall research design.1 As the name suggests, the main aim of predictive ML is prediction. In particular, models are built which should be able to predict outcome values of new, unseen cases. This constitutes the first significant difference from the conventional approaches we apply in life course research. Here, we usually have a fixed data set that we analyze and do not aim to predict values of cases from newly collected data. The consequence of the focus on prediction is that validation of models on new data is a central part of any analysis applying ML techniques. This will be discussed in detail below. The Eternal Struggle of Bias vs. Variance A critical discussion when applying ML is the trade-off between bias and variance of the estimator. This is also sometimes referred to as the trade-off between in-sample and out-of-sample error (Wolpert, 1992; Inoue and Kilian, 2007). Bias means how closely predictions of the model match the observed values, while variance characterizes how strongly the predictions vary when the model is applied to different data sets. The problem of optimizing the model 112

Predictive machine learning approaches  113 Table 8.1

Translation of some terms from ML to traditional statistics

Term in ML framework

Meaning or corresponding terms typically used in life course research

(Learning, training) algorithm

The process of finding optimal statistical parameters for the model based on the data. Optimal means minimizing an objective loss function (e.g., likelihood). Often referred to as “estimation” in traditional life course research.

Bagging

Bootstrap aggregation is an ensemble method for combining predictions from many trees through majority voting (the prediction which most trees do is accepted).

Bias

The deviation of the expectation of the estimator or model from the actual values

Black-box model

A model for which it is hard to follow how we derived the predictions. We do not know precisely why we get the prediction we see.

Boosting

Models are sequentially optimized to reduce error and boost best performing models.

Data mining

Explorative analysis

Ensemble

Algorithm to combine many models to improve overall prediction. Usually used to combine many weak learners instead of using one strong learning algorithm.

Feature

Variable

Gradient descent

The term refers to an iterative optimization algorithm for finding a local minimum of a differentiable function. We can use it for different kinds of estimators like maximum likelihood.

Hyperparameters

Parameters that determine how a model learns from the data. We can use the tuning process to choose these parameters.

Label

Observed value of a variable of interest (usually the outcome)

Loss function

The mathematical function is minimized to find optimal parameters in the training of a model. For example, OLS minimizes the sum of squares of residuals.

Noisy labels, misclassification

Measurement error

Overfitting

Learning patterns that are part of random noise instead of associations that are generalizable to other data.

Regularization

All (hyper)parameters in a model that lead to the penalization of more complex models are regularization parameters. Regularization is used to reduce out-of-sample error and overfitting.

Sample, Instance

A single observation in the data set

Supervised machine learning (SML)

Methods for data with labeled outcomes

Test data

The part of the data used only to evaluate a trained model, NOT for training

Training data

The part of the data used for training the model, NOT for evaluation

Decision tree

A flow-chart-like way to illustrate an algorithm that contains only conditional control statements (“if-statements”).

Underfitting

Training a model which is under-complex and does not perform well in terms of prediction. Essential aspects of the data-generating process are not modeled.

Unsupervised machine learning (UML)

Methods for data without labeled outcomes; finding labels is the aim of these methods

Variance

The degree to which the estimate of the model will change if it is given different data

estimation on data such that it yields excellent predictions on this specific data set is called overfitting. This means that the model incorrectly attributes random variations as systematical patterns. If this happens, predictions on new data – which have different random variations than the training data – become very unstable. This is referred to as high variance. High variance implies low levels of generalizability of the patterns found in the training data, while low variance implies a high degree of generalizability. Given the same data and algorithm,

114  Handbook of health inequalities across the life course we usually get higher variance if we put too much emphasis on reducing bias. We get high bias if we focus too much on reducing variance. This trade-off is important to note because, in much of applied social science life course research, the discussion of this trade-off is almost entirely absent. The focus of model evaluation is put on minimizing bias, but not on variance (for an exception, see the use of multilevel models for regularization (Gelman and Hill, 2006; Gelman, Hill and Yajima, 2012)). Including hyperparameters in models that lead to a reduction of variance is often referred to as regularization. How Valid Are My Predictions? The trade-off between bias and variance necessitates the evaluation of trained models on new data. Using new or unseen data allows answering the question of how valid predictions are beyond the data on which they were trained. It leads to the distinction between training data and test data. Training and test data An essential principle in ML is that models are trained on a training data set and evaluated on a test data set. Training data can be the whole data that is available at the moment, or (more often) it can be a random sample of the entire data set. The other part of the data (or newly collected) data functions as the test data. A fundamental principle of machine learning is that the model is never allowed to “see” the test data set directly or indirectly. It can only be trained on the training data. Else we run into the risk of learning the random variation of the test data. We lose our opportunity to validate the predictions made on independent data. We then no longer know if we identified generalizable patterns or mostly random noise. If we cannot collect new data, in practice, the larger part of the data is used as training data and a smaller part as the test data set. Cross-validation While we train our model in ML, we often need to learn (or tune) the optimal setting for a set of hyperparameters in a model. If we want to know which combination of hyperparameters works best, we need some way to evaluate the performance of the model under different hyperparameters. We already know that we cannot use the test data for this purpose because else we would be training the model partially on test data, learning its idiosyncrasies instead of systematic regularities. Therefore, we apply a similar approach to the training data and divide them into k parts of equal size called “folds”. Then the hyperparameter tuning is conducted on the whole training data set minus the kth fold and evaluated on the kth fold to evaluate the chosen set of hyperparameters. This is done k-times and then averaged across the k results. In this way, the logic of trying to avoid overfitting is maintained without misusing the test data for training. What’s in the Box? One aspect of many tools used in ML is that the algorithms applied are so complex and often built on aggregation (bagging, ensemble methods) of many “weak learners” such that we no longer easily understand how predictions are made by the model. Such tools are often referred to as black-box models. While these tools can yield impressive results in terms of prediction,

Predictive machine learning approaches  115 it makes interpretation in the sense of explanation much more challenging. Sometimes it is argued that this is a fundamental problem of ML. However, there are very many ways that are currently being explored to make ML more explainable (Molnar, 2020; Masís, 2021). In any case, this already shows us that it will usually be easier to integrate ML methods into life course research if the purpose they are applied for can work well with a black-box model. If we want to know what’s in the box, the effort will always be greater. Do You Need Supervision? A common classification of ML tools is by the way they are allowed to learn from the data. The question is whether we want to supervise our algorithm or not. A major distinction in machine learning algorithms between supervised (SML) and unsupervised (UML) methods is often used, but like many heuristics, it is not always exact (and there are hybrids like semi-supervised approaches). It is noteworthy that this is a target-oriented informal term, and certain algorithms can be used in UML and SML settings (Goodfellow, Bengio and Courville, 2016). The distinction remains useful from a perspective of life course research, however. Supervised methods work with labeled data. This means that we observe the outcome (y) directly in our data for each input X (if it is partially observed, it is semi-supervised). In contrast to SML, unsupervised methods search for patterns in data that are not labeled and not directly observable. For example, we can look for typical types of health behavior among a set of health-related variables. Below, I will give more detail, first on SML then on UML. Supervised methods Typical examples of SML are the classification of data into persons who will develop a certain disease within a time period versus those who remain healthy, but also continuous outcomes like labor income can be analyzed with these methods. Very generally speaking, one tries to find a function of input X that minimizes an objective loss function (e.g., log-likelihood or OLS) with the purpose of attaining optimal explanatory power of outcome y: y  f  X  . One key difference from classical statistical approaches is that researchers do not always need to specify the functional relationship between y and X (f(X)) explicitly. Instead, it is learned from the data. The researcher instead specifies so-called hyperparameters that determine how the actual learning process is shaped, which can be different according to the algorithm. Evaluating the predictive power is then done on a metric that should not be method-specific but comparable across different methods. This ensures that different algorithms can be compared in their predictive power using the same data. For the purpose of this chapter, I will introduce three commonly used SML methods. These are LASSO and Ridge, Random Forest (RF), and extreme gradient boosting (XGBoost). With LASSO (L1) and Ridge (L2), we use a normal regression model like we are used to, but we apply regularization parameters. L1 regularization adds a shrinkage factor which gives weights to the slopes (coefficients) of the variables. High weights imply that these variables contribute strongly to the prediction, while low weights mean that the variable only makes small contributions to the prediction. If the weight goes to zero, the variable is effectively eliminated from the model (coefficient set to zero). This is also referred to as “feature selection”. These weights are hyperparameters and are usually determined in the tuning process. Excluding variables or reducing their influence simplifies the model and increases the prob-

116  Handbook of health inequalities across the life course ability that predictions generalize to new data at the cost of better prediction in the training data (see the section on bias vs. variance above). L2 regularization adds a shrinkage factor to all parameters (coefficients) estimated, pushing them towards zero. However, it does not eliminate any variables completely from the model. Both L1 and L2 regularization can be combined in models. Random forests (Breiman, 2001) are a generalization of classification and regression trees (CART) (Breiman et al., 2017). The purpose of CART is to classify observations according to their outcome value in the dependent variable y based on sub-groups in the data defined by combinations of values in a set of predictors. For this purpose, the whole data set is split into smaller and smaller subgroups (defined by the predictors) so that the observations within a subgroup are as similar in their values of y as possible (based on a certain metric). The endpoints of such a splitting process are so-called “leaves”. The resulting division of the data set into small groups based on the predictors is called a (classification or regression) tree. Random forests use this approach as well but apply two significant changes to the procedure to achieve a better balance between over- and underfitting (or bias and variance). First, trees in a random forest are not completely grown on all predictors. Instead, for each tree, only a random subset of all predictors is used. The division is done up until a certain depth of the tree (which is a hyperparameter of the model). This is meant to guard against overfitting the training data. Second, to counter underfitting, this process of growing small trees (“stumps”) is repeated very often. On their own, these “stumps” are regarded as weak learners because, on their own, they would give very poor predictions. These trees (often hundreds or thousands) are combined (ensemble) into a random forest. The final model then gives a prediction that averages over all predictions of single trees within the random forest. Extreme gradient boosting (XGBoost, Chen and Guestrin, 2016) is a specific development of gradient boosting techniques. Gradient boosting uses gradient descent to minimize the error of prediction (or the loss function). XGBoost is similar to the random forest algorithm in the sense that it also grows trees and then combines them (ensemble) to obtain predictions from many individual weak learners. It also uses randomized variables selection like random forest. Random forest starts each tree without using the result of prior trees. In contrast, XGBoost optimizes each tree sequentially to account for the degree of error that has been made by previous trees. In the end, each tree is given a weight according to how much it contributes to the improvement of error reduction so that not all trees contribute equally to prediction. XGBoost also applies LASSO (L1) and Ridge (L2) kind of regularization to prevent overfitting the training data. Unsupervised methods Two methods that can be regarded as UML methods or closely related, which have been implemented in life course research, are cluster analysis and latent class analysis. Cluster analysis seeks to find hidden clusters or groupings in the data based on the set of variables used as input. Are there typical profiles of health behavior or social conditions in the data? Closely related is latent class analysis (LCA) which allows more probabilistic versions of group membership (Barban and Billari, 2012). Factor analysis and principal component analysis (PCA) reduce several variables to fewer underlying factors, which represent shared variance among the observed variables. These tools are also well known, especially in psychology and when working with larger item-batteries in

Predictive machine learning approaches  117 surveys that represent concepts that are not directly measurable (like personality traits, values, attitudes, etc.). Sequence analysis has also found regular application in life course research. It allows the description and classification of life course trajectories based on sequences of different states and allows us to cluster these into distinct types of life course trajectories (Abbott, 1995; Aisenbrey and Fasang, 2010; Stone, Netuveli and Blane, 2014). We can therefore see that life course research already uses several methods, which are called unsupervised machine learning methods in other fields. All these UML approaches have in common that they try to identify heterogeneity in the data (or population) with respect to certain outcomes and classify this heterogeneity. This can also relate not only to levels of variables but also to trends over time (e.g., growth curve analysis) or for causal effect estimations. In these cases, the question is whether there are systemically different types of trajectories (Østbye, Malhotra and Landerman, 2011; Sterba and Bauer, 2014) or heterogeneity in the causal effect, which might cluster in certain value regions or might be related to other sets of predictors. Whether it is with SML or UML, there is no perfect algorithm. Each method has certain advantages and drawbacks, and their appropriateness and performance will depend on the research question which is to be answered.

HOW DO WE MAKE USE OF ML METHODS? After this short introduction, I will turn to the question of why we deviate from some standard practices in ML and how and when we can make the best use of ML. Why Don’t We Do Cross-validation? The question of validating the results of an analysis in life course research often does not even come up. We are used to having data sets that are cohorts or quasi-random samples of a certain population. We, therefore, want to use the whole data set for maximal statistical power and usually do not collect new data which could be used for validation. Further, a deductive hypothesis testing approach using large data usually gives us relatively small standard errors and relatively high confidence in the stability of the results. Even more important is the fact that most models are rather simple (compared to what can be done with ML models). This leads us to the reasonable expectation that a standard analysis in life course research has low variance, which makes validation on a test or hold-out data set a less pressing issue than in ML. If we turn from the deductive approach to more explorative analysis in life course research, we get a different picture. It is worth noting that tools classified as UML have been used in social science and life course research often without doing any kind of cross-validation or using test data to evaluate their validity outside of the data they were trained on. Here we can have similarly complex aggregation patterns, e.g., cluster analysis or latent class analysis, which might capture mostly random fluctuations of our original data set and might not reproduce well on new data. Systematically using validation on unseen test data seems a useful and important step in this area. It can and should be adapted from the realm of ML into statistical

118  Handbook of health inequalities across the life course analysis in life course research. A wide variety of techniques exist to balance the trade-off between bias and variance here already. Another example where we should be cautious of the risk of overfitting (even if validation might not be possible) is when choosing the functional form of predictors. It might look like the question is deductive, but the choice of the functional form of some predictors might be data driven (or arbitrarily complex). A prominent example comes from the analysis of the effect of pollution on life expectancy (Chen et al., 2013). This is discussed in detail in Gelman and Imbens (2018). The key issue here is that a high polynomial is used to fit the relationship between latitude and life expectancy. Very high order polynomials can fit almost any arbitrary function, which might be the result of random noise or influences of other unobserved factors. This very close fit might therefore be an artifact of the specific data set. If such a polynomial is used in regression discontinuity design, the discontinuity which is found might be a statistical artifact of the complex polynomial modeling. As the discontinuity is the central estimator for causal effects, this might lead to biases of unknown direction and magnitude. In this case, new data might also not be available, but the question of potential overfitting is not discussed in sufficient detail. Under Which Circumstances can ML Show its Strengths? The question that we can ask now is whether there are some general conditions that we can identify under which ML methods might perform quite well, and we might benefit from their use. While it is not an exhaustive list, I argue that the following aspects of the data and the modeling make ML show its strengths: (1) The number of observations is large (2) The size of the set of predictors X, after relevant information reduction techniques (factor analysis, cluster analysis, etc.) are applied, is still large (3) The functional form of X and y a. is not known and b. might have non-linearities (4) There might be relevant interactions among the variables in the set of predictors X (5) The noise in the outcome and the predictors is small If the data structure is complex and the outcome has many determinants, a large number of observations helps the algorithms learn the relevant systematical pattern. Further, if we have large sets of potential predictors, automated algorithms for variable inclusion, exclusion, or weighting down their influence help to sort relevant predictors from those who are (conditionally) unrelated to the outcome. If we do not know how X and y are related and suspect that the relationship might be non-linear but do not know how, using ML to determine the approximate functional form without any prior assumptions might be helpful. This is more of an issue if we want to relate continuous predictors to an outcome than if we have categorical data where we often use single values of the variable as predictors with one reference and also do not make assumptions about functional form. Closely related is the issue of interactions among predictors. Again, using methods like XGBoost or Random Forest allows the exploration of any possible interaction among the predictors, so we do not have to specify each possibility by hand. In combination with the

Predictive machine learning approaches  119 search for non-linearities, this can be a great help. Note, however, that not all ML algorithms have this capacity. LASSO regression, for example, penalizes weak predictors but only allows interactions and non-linearities which are explicitly specified. One could argue at this point that searching for interaction patterns lacks any theoretical base. This is true, but the opposite perspective usually also holds true. Assuming no interactions is in most applied cases also not clearly backed by theoretical predictions but simply a more convenient modeling choice of the researcher. When we search for interactive patterns, the analysis will at least be backed by data. Lastly, very noisy (mislabeled) data is always problematic but becomes more so if complex patterns in the data are to be explored. It vastly increases the risk of finding randomness in the training data instead of structure that replicates in new data. With simpler, traditional models, we get inflated standard errors from noisy data and reduced certainty on our estimates as well. We do not learn incorrect patterns from the data unless we pair high noise with small number of observations (and null-hypothesis-significance-testing (Gelman and Stern, 2006; Gelman and Carlin, 2014)). When Do Traditional Methods Serve Us Better? From the criteria above, we can deduce that with a theory-driven model building and deductive hypotheses testing or descriptive analyses, we are well within the analytical territory in which the main analyses often do not directly benefit from ML approaches. As soon as we want to search for overall patterns in the data or want to explore interactions and non-linearities in certain relationships, we might begin to struggle with traditional approaches because especially social theories are often not precise enough, e.g., to predict non-linearities. Particular Notes for Research on Health Inequalities and Life Course Research When we take a look at stylized life course processes of critical period, accumulation, and social mobility (Hallqvist et al., 2004; Mishra et al., 2009), we see that both critical (or sensitive) period and accumulation are relatively simple models without inherent interactions or non-linearities. In fact, the concept of allostatic load is built on the idea of an accumulation process in which we do not need to know each specific exposure but can work with the sum of proxies of exposures over the life course to get a good estimate of risk for health in older age (Seeman et al., 2001; Dowd, Simanek and Aiello, 2009; Turner, Thomas and Brown, 2016). When more complex models are tested against simple linear summation models, the resulting gain in predictive power is real but often small (Kröger and Hoffmann, 2018). This means that additional complexities, interactions, and non-linearities might exist when doing research on health inequalities in the life course, but often we find simple linear cumulative modeling of risk a sufficient simplification that does not warrant more complex ML models. However, we should note that without employing such more complex models as robustness checks, we would only have our assumptions to rely on and do not have data and analysis to back up claims of the usefulness of simple cumulative modeling strategies. The idea of social mobility lends more importance to the notion that complex patterns or changes throughout the life course might be relevant risk factors. If we only use three to five points of measurement (which is often the case in some life course studies), we probably have not reached the degree of complexity in which ML methods have a clear edge over traditional

120  Handbook of health inequalities across the life course approaches. Sequence analytic methods can be an important tool here to account for heterogeneity in the life course when, e.g., transition to retirement is evaluated in its effects on health after retirement (e.g., Barban et al., 2017). Research on health inequalities in the life course will probably make more use of new types of data, which are more and more becoming available in longitudinal studies like the measurement of electroencephalograms (EEG) to detect early signs of cognitive decline (Anstey, 2012; Simpraga et al., 2017) or early differences in the cognitive development of children or the composition of the microbiome as a determinant for health throughout the life course (Findley et al., 2016; Herd et al., 2018; Cammarota et al., 2020; Hanson et al., 2021). Here we might reasonably expect high degrees of non-linearity and complex interactions. At the current stage, we might also accept a black-box view on processes happening in the brain and the gut microbiome as long as we have reliable ways to establish if there are relevant differences between social or demographic groups in these processes. This points to the fact that the more we integrate new forms of complex data, what is often called Big Data, in life course research, the more we will be able to use the advantages of complex ML methods as complementary tools for our data analysis. Without new data, the limitations of the data will put stricter bounds on the usefulness of ML approaches for life course research.

EXAMPLES OF THE USE OF ML METHODS After some general discussion, I will give some examples of the application of ML in the life course literature and present an applied example where ML methods can be used for estimating the inverse probability of drop-out weights for a longitudinal survey. From the Literature Fritze et al. (2020) use recursive partitioning methods (SML, similar to a CART approach) to find profiles that describe similar levels of risk of diabetes, coronary heart disease, and cerebrovascular disease in later life among birth cohorts, finding four empirically based historical birth cohorts. Clark et al. (2021) use support vector machines, random forests, and regularized logistic regression, which are all SML algorithms, to predict self-rated health across the life course. They find that socio-economic predictors are particularly relevant in mid-life while comorbidities become more relevant in old age. One example of identifying heterogeneity in life course sequences for comparison between two countries comes from Billari, Fürnkranz and Prskawetz (2006). They find that the sequencing of events (e.g., education, marriage, entering the labor market) distinguishes best between Austrians and Italians, while the timing of these events comes second (an example of UML). Particularly synchronous leaving home and marrying for the first time is much more common in Italy than in Austria. Porto Chiavegatto Filho et al. (2018) use 16 different SML algorithms (including Random Forest, support vector machines, LASSO) to identify Brazilian communities who are over- and underachieving with respect to life expectancy based on a set of predictors for life expectancy. They then average over the results (ensemble super-learning) of these 16 algorithms to get a single (super-)model which predicts over- and underachieving. Then they investigate

Predictive machine learning approaches  121 what characteristics are associated with over- and underachieving. Underachievers showed more caesarean deliveries and mammographies and had more life-support health equipment. Overachievers showed better results regarding primary health care and higher coverage of the family support programs. Will They Stay? An Application for Inverse Probability Weighting for Survey Drop-out One common applied problem in which a black-box approach might be useful is dealing with drop-out from longitudinal studies. It is well established that drop-out from panel and cohort studies is related to both socio-economic situation and health conditions (Eaton et al., 1992; Fitzgerald, Gottschalk and Moffitt, 1998; Jones, Koolman and Rice, 2006; Watson and Wooden, 2009; Littman et al., 2010) and can therefore affect any estimation of health inequalities in a longitudinal or life course perspective. Inverse probability weighting (IPW) is a commonly used approach to account for selective drop-out in longitudinal studies. In this case, I want to estimate the probability of drop-out (P(DO)) given a set of covariates (X) which we assume might predict the drop-out. The easiest and most common approach to do this is to estimate a logistic regression model, predict drop-out status and take the inverse of this prediction as a weight in the final analyses of interest. p  DO | X, 

1  X 1 e  

. (1)

This is a well-known equation. However, it requires us to specify the functional form in which the covariates X are related to the probability of drop-out (or to the log-odds of drop-out). In many cases, we might have an idea about the set of variables influencing drop-out, but it is less clear that we know what the functional form looks like. This is the point at which we might want to employ several machine learning algorithms which make fewer assumptions about the functional form than logistic regression and can also be used to filter relevant variables out of the set of predictors X. p  DO | X,  f  X, (2) The advantage of taking a more agnostic approach to the functional form is that if we misspecify the model, IPW might no longer accurately correct for biases due to selective drop-out. Instead, we can use algorithms like random-forest (RF) or XGBoost to estimate the probability of drop-out. The main difference from the classic approach is that, as explained above, we would fit the model on a training data set and then validate it on a test data set to pick the model which best predicts drop-out. From the discussion on RF or XGBoost, we can already expect that we should have an advantage over a simple logistic regression model if we have a large set of X, if X contains the many continuous variables, and if we can expect many non-linearities and interactions in the “true” model for drop-out. To illustrate such an approach, I show the results from predicting drop-out from the SOEP-CoV study from wave 1 to wave 2 (Liebig et al., 2019; Entringer et al., 2020; Kühne et al., 2020). This is an add-on study from the Socio-economic

122  Handbook of health inequalities across the life course Panel Study (SOEP) during the first (April–July 2020) and second (January/February 2021) Corona-pandemic-related lockdown in Germany. It allows the integration of pre-pandemic predictors of participation in the second wave as it is linked to the main long-running panel study. The set of predictors includes demographic, social, and economic factors as well as participation behavior in earlier waves. Figure 8.1 shows the predictions from logistic regression, XGBoost, RF, and LASSO for the probability of staying in the survey.

Figure 8.1

Probability of staying in the SOEP-CoV survey

We can infer two main conclusions from the figure. First, the overall distribution of predicted probabilities is not identical but similar between the chosen methods. Second, we can see that standard logistic regression has a shift towards the right of the distribution, indicating that it predicts very low probabilities of staying in the survey (even close to one) much more often than the ML algorithms do. Third, RF and XGboost predict more “moderate” values of probability and do not produce strong outliers. Table 8.2 shows the distribution of the predicted probabilities by the method used. We see that conventional logistic regression predicts much more extreme values (sometimes close to one and zero) than the ML algorithms. The reason for the higher spread of logistic regression results is that continuous predictors tend to generate extreme values, while the algorithm of RF and XGBoost implicitly categorizes continuous variables and therefore does not interpolate a linear trend on the log-odds as logistic regression does. For inverse probability, weighting truncating weights at the 1% and 99% percentile is often recommended (Cole and Hernán, 2008) for exactly this reason. This does not seem to be necessary for RF and XGBoost here. This form of regularization is built into the methods themselves. Table 8.3 shows the correction between the same probabilities predicted by different methods.

Predictive machine learning approaches  123 Table 8.2

Distribution of probabilities predicted by different methods

Method

Minimum

Maximum

Mean

SD

1st percentile

99th percentile

Logistic Regression

0.000444

1

0.904

0.0986

0.467

0.993

LASSO

0.201

0.978

0.904

0.0638

0.662

0.965

Random Forest

0.381

0.975

0.903

0.0684

0.631

0.967

XGBoost

0.508

0.975

0.895

0.0622

0.657

0.961

Table 8.3

Correlations between predictions Logistic Regression

LASSO

Random Forest

1

0.8459265

0.6660789

0.648379

LASSO

0.8459265

1

0.7287948

0.7335977

Random Forest

0.6660789

0.7287948

1

0.8174472

XGBoost

0.648379

0.7335977

0.8174472

1

Logistic Regression

Table 8.4

XGBoost

Predictive power AUC-ROC

Logistic Regression

0.747

LASSO

0.720

Random Forest

0.989

XGBoost

0.870

As surmised from Figure 8.1, we can see that those correlations are quite high, ranging between 0.64 and 0.84. We will, therefore, not expect to see completely different results when using these IPW in the final analyses. Table 8.4 shows the predictive power measured by the area under the curve of the receiver-operator-characteristic (AUC-ROC). This is a common metric for predictive power in classification tasks. It is lowest for logistic regression and LASSO, which are similar and clearly higher for XGBoost and Random Forest, with Random Forest giving the best prediction on the test data. We can therefore see in this example that some of the characteristics which can make the application of supervised ML methods relevant apply. For example, extreme values are avoided, but the overall similarity is high, and added predictive power depends on the algorithm and does not always need to be higher, although in this case, XGBoost and Random Forest perform much better. However, it still makes sense to check the final results for robustness against the use of different IPW specifications in the form of specification curve analysis (Steegen et al., 2016; Simonsohn, Simmons and Nelson, 2019), because we cannot know whether certain interactions and non-linearities are (not) relevant to our final analyses. Similar to inverse probability weighting, another strategy for dealing with missing data, multiple imputation (MI), has also found application in many life course studies and can be implemented using different ML algorithms as well. In certain software packages, methods like regression and classification trees, random forests, and nearest-neighbor are already implemented (Royston and White, 2011).

124  Handbook of health inequalities across the life course

THE FUTURE. TOGETHER. NOW. I have discussed the very basic criteria for distinguishing ML approaches from more traditional approaches used in quantitative life course research. I have looked at conditions under which ML methods become useful and identified four major areas in which we are already using methods that can be categorized as ML and where we might do this more often in the future or take up other practices from the realm of ML. (1) Explorative approaches in the first stage of the analyses can either be used to replace arbitrary definitions of certain variables based on cut-offs (e.g., cohorts (Fritze et al., 2020), income groups, health behavior types through cluster analysis) or validate theory-driven definitions on an empirical basis. Both in cases of empirically derived classifications via ML methods or in a set of different classifications based on theoretical considerations, it seems prudent to run different specifications and see how sensitive the results are to the choices made. (2) A more technical application can be in model or variable selection, which can be especially important in life course research where we often deal with observational data, including large numbers of variables potentially relevant for a given research question. (3) Investigating heterogeneity in trends over the life course or populations more generally can also be furthered by ML. Predictions derived from theory might hold under some context factors but not others. This problem is well known, but ML offers a data-driven, explorative approach instead of testing particular sub-hypotheses, which refer explicitly to the moderation of certain effects by circumstances. (4) Validating our model on new kinds of data is an important practice we can also adapt to applied life course research. This holds especially for the application of explorative methods with which we want to draw inference beyond the data we do analysis on (which is the most common case in applied research). These areas imply that it seems likely that life course research will take up ML methods more often in the future, not to replace traditional methods but to supplement them. In this sense, we should neither fight the rapid ascent of ML tools nor embrace it blindly, but extend our everyday research toolbox to include many of these methods to address research questions about inequalities in the life course and perhaps develop new types of research questions in the process.

NOTE 1. One field which receives less attention from social science, but also shows a rapid development in the last decades is machine learning (ML) for causal inference (Mooney and Pejaver, 2018; Molina and Garip, 2019). This part of machine learning follows a different purpose than predictive machine learning. It would be beyond the scope of this article to cover both areas.

REFERENCES Abbott, A. (1995) ‘Sequence Analysis: New Methods for Old Ideas’, Annual Review of Sociology, 21, pp. 93–113. doi:10.2307/2083405.

Predictive machine learning approaches  125 Aisenbrey, S. and Fasang, A.E. (2010) ‘New Life for Old Ideas: The “Second Wave” of Sequence Analysis Bringing the “Course” Back into the Life Course’, Sociological Methods & Research, 38(3), pp. 420–462. doi:10.1177/0049124109357532. Anstey, K.J. (2012) ‘Biomarkers and Memory Aging: A Life-course Perspective’, in Naveh-Benjamin, M. and Ohta, N. (eds), Memory and Aging: Current Issues and Future Directions. Psychology Press, pp. 349–372. Barban, N. and Billari, F.C. (2012) ‘Classifying Life Course Trajectories: A Comparison of Latent Class and Sequence Analysis’, Journal of the Royal Statistical Society: Series C (Applied Statistics), 61(5), pp. 765–784. doi:10.1111/j.1467-9876.2012.01047.x. Barban, N. et al. (2017) ‘Causal Effects of the Timing of Life-course Events: Age at Retirement and Subsequent Health’, Sociological Methods & Research, 49(1), pp. 216–249. doi:10.1177/0049124117729697. Billari, F.C., Fürnkranz, J. and Prskawetz, A. (2006) ‘Timing, Sequencing, and Quantum of Life Course Events: A Machine Learning Approach’, European Journal of Population / Revue européenne de Démographie, 22(1), pp. 37–65. doi:10.1007/S10680-005-5549-0. Breiman, L. (2001) ‘Random Forests’, Machine Learning, 45(1), pp. 5–32. doi:10.1023/A:1010933404324. Breiman, L. (2017) Classification and Regression Trees. Routledge. Cammarota, G. et al. (2020) ‘Gut Microbiome, Big Data and machine Learning to Promote Precision Medicine for Cancer’, Nature Reviews Gastroenterology & Hepatology, 17(10), pp. 635–648. doi:10.1038/s41575-020-0327-3. Chen, T. and Guestrin, C. (2016) ‘XGBoost: A Scalable Tree Boosting System’, in Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. New York, NY, USA: ACM, pp. 785–794. doi:10.1145/2939672. Chen, Y. et al. (2013) ‘Evidence on the Impact of Sustained Exposure to Air Pollution on Life Expectancy from China’s Huai River Policy’, Proceedings of the National Academy of Sciences of the United States of America, 110(32), pp. 12936–12941. doi:10.1073/PNAS.1300018110/-/ DCSUPPLEMENTAL. Clark, C.R. et al. (2021) ‘Predicting Self-Rated Health Across the Life Course: Health Equity Insights from Machine Learning Models’, Journal of General Internal Medicine, 36(5), pp. 1181–1188. doi:10.1007/S11606-020-06438-1. Cole, S.R. and Hernán, M.A. (2008) ‘Constructing Inverse Probability Weights for Marginal Structural Models’, American Journal of Epidemiology, 168(6), pp. 656–664. doi:10.1093/aje/kwn164. Dowd, J.B., Simanek, A.M. and Aiello, A.E. (2009) ‘Socio-economic Status, Cortisol and Allostatic Load: A Review of the Literature’, International Journal of Epidemiology, 38(5), pp. 1297–1309. doi:10.1093/ije/dyp277. Eaton, W.W. et al. (1992) ‘Psychopathology and Attrition in the Epidemiologic Catchment Area Surveys’, American Journal of Epidemiology, 135(9), pp. 1051–1059. Entringer, T. et al. (2020) ‘Psychische Krise durch Covid-19? Sorgen sinken, Einsamkeit steigt, Lebenszufriedenheit bleibt stabil’, SOEP Papers on Multidisciplinary Panel Data Research, 1087. Findley, K. et al. (2016) ‘Health Disparities and the Microbiome’, Trends in Microbiology, 24(11), pp. 847–850. doi:10.1016/J.TIM.2016.08.001. Fitzgerald, J., Gottschalk, P. and Moffitt, R. (1998) ‘An Analysis of Sample Attrition in Panel Data: The Michigan Panel Study of Income Dynamics’, Journal of Human Resources, 33(2), p. 251. doi:10.2307/146433. Fritze, T. et al. (2020) ‘World War II Cohorts and Diabetes Mellitus, Coronary Heart Disease and Cerebrovascular Disease Later in Life: An Observational Cohort Study Based on German Claims Data’, medRxiv. doi:10.1101/2020.11.11.20227660. Gelman, A. and Carlin, J. (2014) ‘Beyond Power Calculations’, Perspectives on Psychological Science, 9(6), pp. 641–651. doi:10.1177/1745691614551642. Gelman, A. and Hill, J. (2006) Data Analysis Using Regression and Multilevel/Hierarchical Models. Cambridge University Press. Gelman, A., Hill, J. and Yajima, M. (2012) ‘Why We (Usually) Don’t Have to Worry About Multiple Comparisons’, Journal of Research on Educational Effectiveness, 5(2), pp. 189–211. doi:10.1080/19 345747.2011.618213.

126  Handbook of health inequalities across the life course Gelman, A. and Imbens, G. (2018) ‘Why High-Order Polynomials Should Not Be Used in Regression Discontinuity Designs’, Journal of Business & Economic Statistics, 37(3), pp. 447–456. doi:10.1080 /07350015.2017.1366909. Gelman, A. and Stern, H. (2006) ‘The Difference Between “Significant” and “Not Significant” Is Not Itself Statistically Significant’, The American Statistician, 60(4), pp. 328–331. Available at: http://​ www​.tandfonline​.com/​doi/​pdf/​10​.1198/​000313006X152649. Goodfellow, I., Bengio, Y. and Courville, A. (2016) Deep Learning. MIT Press. Hallqvist, J. et al. (2004) ‘Can We Disentangle Life Course Processes of Accumulation, Critical Period and Social Mobility? An Analysis of Disadvantaged Socio-economic Positions and Myocardial Infarction in the Stockholm Heart Epidemiology Program’, Social Science & Medicine, 58(8), pp. 1555–1562. doi:10.1016/S0277-9536(03)00344-7. Hanson, H.A. et al. (2021) ‘Charting the Life Course: Emerging Opportunities to Advance Scientific Approaches using Life Course Research’, Journal of Clinical and Translational Science, 5(1), pp. 9–10. doi:10.1017/CTS.2020.492. Herd, P. et al. (2018) ‘Social and Population Health Science Approaches to Understand the Human Microbiome’, Nature Human Behaviour, 2(11), pp. 808–815. doi:10.1038/s41562-018-0452-y. Inoue, A. and Kilian, L. (2005) ‘In-Sample or Out-of-Sample Tests of Predictability: Which One Should We Use?’, Econometric Reviews, 23(4), pp. 371–402. doi:10.1081/ETC-200040785. Jones, A.M., Koolman, X. and Rice, N. (2006) ‘Health-Related Nonresponse in the British Household Panel Survey and European Community Household Panel: Using Inverse-probability-weighted Estimators in Non-linear Models’, Journal of the Royal Statistical Society: Series A – Statistics in Society, 169(3), pp. 543–569. doi:10.1111/j.1467-985X.2006.00399.x. Kröger, H. and Hoffmann, R. (2018) ‘The Association between CVD-related Biomarkers and Mortality in the Health and Retirement Survey’, Demographic Research, 38, pp. 1933–2002. doi:10.4054/ DemRes.2018.38.62. Kühne, S. et al. (2020) ‘The Need for Household Panel Surveys in Times of Crisis: The case of SOEP-CoV’, Survey Research Methods, 14(2), pp. 195–203. doi:10.18148/srm/2020.v14i2.7748. Liebig, S. et al. (2019) ‘Socio-Economic Panel (SOEP), Data from 1984–2018, Version 35’. doi:10.5684/ soep-core.v35. Littman, A.J. et al. (2010) ‘Assessing Nonresponse Bias at Follow-up in a Large Prospective Cohort of Relatively Young and Mobile Military Service Members’, BMC Medical Research Methodology, 99, p. 11. doi:10.1186/1471-2288-10-99. Masís, S. (2021) ‘Interpretable Machine Learning with Python’, Packt, p. 737. Available at: https://​ www​.packtpub​.com/​product/​interpretable​-machine​-learning​-with​-python/​9781800203907 (accessed: 1 January 2022). Mishra, G. et al. (2009) ‘A Structured Approach to Modelling the Effects of Binary Exposure Variables over the Life Course’, International Journal of Epidemiology, 38(2), pp. 528–537. doi:10.1093/ije/ dyn229. Molina, M. and Garip, F. (2019) ‘Machine Learning for Sociology’, Annual Review of Sociology, 45, pp. 27–45. doi:10.1146/ANNUREV-SOC-073117-041106. Molnar, C. (2020) Interpretable Machine Learning: A Guide for Making Black Box Models Interpretable. lulu.com. Available at: https://​christophm​.github​.io/​interpretable​-ml​-book/​ (accessed: 1 January 2022). Mooney, S.J. and Pejaver, V. (2018) ‘Big Data in Public Health: Terminology, Machine Learning, and Privacy’, Annual Review of Public Health, 39, pp. 95–112. doi:10.1146/ANNURE V-PUBLHEALTH-040617-014208. Østbye, T., Malhotra, R. and Landerman, L.R. (2011) ‘Body Mass Trajectories through Adulthood: Results from the National Longitudinal Survey of Youth 1979 Cohort (1981–2006)’, International Journal of Epidemiology, 40(1), pp. 240–250. doi:10.1093/ije/dyq142. Porto Chiavegatto Filho, A.D. et al. (2018) ‘Overachieving Municipalities in Public Health: A Machine-learning Approach’, Epidemiology (Cambridge, Mass.), 29(6), pp. 836–840. doi:10.1097/ EDE.0000000000000919. Royston, P. and White, I.R. (2011) ‘Multiple Imputation by Chained Equations (MICE): Implementation in Stata’, Journal of Statistical Software, 45(4), pp. 1–20.

Predictive machine learning approaches  127 Seeman, T.E. et al. (2001) ‘Allostatic Load as a Marker of Cumulative Biological Risk: MacArthur Studies of Successful Aging’, Proceedings of the National Academy of Sciences, 98(8), pp. 4770–4775. doi:10.1073/pnas.081072698. Simonsohn, U., Simmons, J.P. and Nelson, L.D. (2019) ‘Specification Curve: Descriptive and Inferential Statistics on All Reasonable Specifications’, Available at SSRN: https://​ssrn​.com/​abstract​=​2694998 or http://​dx​.doi​.org/​10​.2139/​ssrn​.2694998. Simpraga, S. et al. (2017) ‘EEG Machine Learning for Accurate Detection of Cholinergic Intervention and Alzheimer’s Disease’, Scientific Reports, 7(1), pp. 1–11. doi:10.1038/s41598-017-06165-4. Steegen, S. et al. (2016) ‘Increasing Transparency Through a Multiverse Analysis’, Perspectives on Psychological Science, 11(5), pp. 702–712. doi:10.1177/1745691616658637. Sterba, S.K. and Bauer, D.J. (2014) ‘Predictions of Individual Change Recovered with Latent Class or Random Coefficient Growth Models’, Structural Equation Modeling: A Multidisciplinary Journal, 21(3), pp. 342–360. doi:10.1080/10705511.2014.915189. Stone, J., Netuveli, G. and Blane, D. (2014) ‘Life-course Occupational Social Class and Health in Later Life: The Importance of Frequency and Timing of Measures’, European Journal of Ageing, 11(3), pp. 273–284. doi:10.1007/s10433-014-0307-y. Turner, R.J., Thomas, C.S. and Brown, T.H. (2016) ‘Childhood Adversity and Adult Health: Evaluating Intervening Mechanisms’, Social Science & Medicine, 156, pp. 114–24. doi:10.1016/j. socscimed.2016.02.026. Watson, N. and Wooden, M. (2009) ‘Identifying Factors Affecting Longitudinal Survey Response’, in Lynn, P. (ed.), Methodology of Longitudinal Surveys. Chichester, UK: John Wiley & Sons, Ltd, pp. 157–181. doi:10.1002/9780470743874.ch10. Wolpert, D.H. (1992) ‘On the Connection between In-sample Testing and Generalization Error’, Complex Systems, 6(1), p. 47.

9. Instrumental variables in studies of health and health inequalities Rasmus Hoffmann and Gabriele Doblhammer

INTRODUCTION The application of instrumental variables (IVs) is almost 100 years old (Wright 1928) and has long been a core method in econometrics; it is currently experiencing something of a renaissance and has spread to various health-related disciplines. It is used to address causality issues caused by confounding, measurement error, or simultaneity bias (bidirectional causality) in observational studies (as opposed to in experiments). This chapter briefly introduces the idea of instrumental variables and reviews the definition of the term, the accompanying (theoretical) assumptions, and the types of IV. It provides an overview of the study questions and instruments used in health and health inequality research by reviewing the literature. We discuss what makes a good instrument, the advantages and disadvantages of the approach, and offer our conclusion on the future potential of this method for longitudinal causal analysis in the field of health and health inequalities. Early examples of IV include birth month, which was used as early as 1938 to study effects of early life on health outcomes later in life (Huntington 1938) and has since been applied to contemporary elderly populations in Austria, Denmark, the United States and Australia (Doblhammer and Vaupel 2001); natural “experiments” such as famines, e.g. the Great Finnish Famine 1866–68 (Doblhammer, van den Berg, Lumey 2013; Kannisto et al. 1997) and the Chinese Famine in 1959–61 (Gørgens et al. 2012; Song 2010); business cycles (Doblhammer, van den Berg, Fritze 2013; van den Berg et al. 2006); the process of German reunification (Vogt and Vaupel 2015); and most recently genes (Davey Smith and Ebrahim 2003; DiPrete et al. 2018). A classic example is Angrist’s study of income differentials using the U.S. draft lotteries during the Vietnam War (Angrist 1990). The instrumental variable approach has been widely applied in health literature in past decades, from which we will present and discuss more examples below. Methods employing IVs can adjust for unobserved confounders in observational studies. Other methods of adjusting for confounding effects, which include stratification, propensity score matching, and regression adjustment, can only adjust for observed confounders. Observational studies are often implemented as a substitute for or complement to an experiment, although experiments are the gold standard for making causal inferences. The main concern with using observational data to make causal inferences is that an individual may be more likely to receive a treatment because that individual already has a health condition. To “receive a treatment”, e.g. through a public policy, means the same as being “exposed to a risk factor”, e.g. smoking. The treatment effect may be biased because some individuals received the treatment precisely due to their personal or health characteristics, which means that the treatment is endogenous (endogeneity bias). An instrumental variable is a variable that influences the treatment, but does not directly influence the outcome. As an example, more edu128

Instrumental variables in studies of health and health inequalities  129 cation is the treatment and health the outcome. In this example, an often-used instrument for education is policy change, such as increasing the number of compulsory years of education. The IV approach quantifies how variation in the treatment induced by the instrument changes the outcome. The treatment may be affected by selection into treatment (e.g. health affecting education on the individual level), while the instrument is not. In this chapter, we focus on IVs, but it is worth mentioning the similarity to difference-in-differences (DiD) and regression discontinuity (RD) approaches. For example, a DiD analysis can be identical to one using IV, where the interaction between the policy and timing of implementation is the IV (Matthay et al. 2020). In other words, when the IV is a policy change, the same analysis can be described as DiD and RD, as in Banks and Mazzonna (2012). Because of this similarity, Matthay et al. (2020) group non-randomized study designs into (1) “confounder-control” approaches that control for observable confounders, such as regression adjustment or propensity score matching and (2) “instrument-based” approaches that can also be called “quasi-experimental”. This is because the main strategy of the IV approach for estimating causal effects is to find a (conditionally) random source of variation, i.e. a random process that influences the probability of exposure. Such situations can sometimes be created by “natural experiments”, such as policy changes, natural disasters, weather events and the like.

ASSUMPTIONS OF THE IV APPROACH The two main assumptions have been mentioned already above, but will be presented more formally here. First, the IV should affect the treatment. For example, a policy which introduces more years of compulsory schooling should improve educational outcomes. In another example, cigarette prices may be the instrument for smoking and the first assumption would imply that the cigarette price affects the prevalence of smoking. This assumption, which is also called “relevance”, is testable, e.g. by checking the correlation between the IV and the treatment or by an F-test for the significance of excluded instruments. If the relevance of an IV is low, i.e. the correlation with the treatment is low, we speak of a weak instrument. More about checking whether and what to do if the instrument is weak can be found elsewhere (Gangl 2010; Martens et al. 2006; Moffitt 2009). Second, the IV should not affect the outcome, except via the treatment. In our examples, it means that the change in education policy as such should not affect health, but only through increased education. Likewise, the cigarette price alone should not affect health, but only via the prevalence of smoking. This criterion is called the validity of an instrument or the “exclusion restriction” and cannot be tested. A third assumption is that the IV should not have shared unmeasured causes with the outcome (Matthay et al. 2020). This criterion is called exchangeability and is violated if e.g. public policy in a country or region has integrated sets of legislation that improve both education and health, or when cigarette price and health have common background factors such as cultural contexts on the macro level. The graphs in Figure 9.1 (A, B, C, D) illustrate these assumptions (taken from Glymour et al. 2017):

130  Handbook of health inequalities across the life course

Source: Glymour et al. 2017.

Figure 9.1

Causal diagrams showing one valid (A) and three invalid instruments (B, C, D)

A valid IV influences the outcome, but only through the exposure or treatment. Biases due to unmeasured confounders are avoided (Figure 9.1A). Invalid IVs have an effect on the outcome (Figure 9.1B) or on unmeasured confounders (Figure 9.1C). IVs are also invalid if there are common background factors for the IV and the outcome (Figure 9.1D), which would be a violation of the third assumption above. The assumptions for a valid IV are similar to those for a randomized control trial (RCT), which underlines again the relations between the IV approach and an experiment. The treatment group assignment in an RCT would be analogous to the instrument: it should strongly predict having a treatment, but otherwise not be related to the outcome. The IV assumptions may also be met conditional on a set of control variables (“conditionally valid instruments”) (Glymour et al. 2017). As already mentioned, it is not possible to test or even prove that assumptions two or three are fulfilled. However, they sometimes can be disproven by fal-

Instrumental variables in studies of health and health inequalities  131 sification tests, which depend on both theoretical knowledge and statistical evidence (van Kippersluis and Rietveld 2018). For example, if an IV is correlated with known, measured risk factors for Y, it may also be correlated with unmeasured risk factors for Y (Glymour et al. 2017; Swanson and Hernán 2013).

TYPES AND EXAMPLES OF IV There are several types of instrumental variables; Moffitt (2009) distinguishes between: 1. Cross-sectional ecological variables, such as differences in policies, laws, and social structure by geographic area which are independent of an individual’s own choices. 2. Population-segment fixed effects instruments, where the segments are defined by social or demographic groups (instead of geographic groups) and the instruments pertain to groups, such as welfare reforms for certain income or demographic groups (e.g. single mothers). 3. Siblings and related instruments such as genetic groups, where the instrument is the deviation of each individual’s treatment from the average group-specific treatment, e.g. one sibling has children, the other does not, and the instrument is the deviation in number of children from the family’s average. 4. Natural experiments, which are often defined as a residual category of instruments which appear to be random, such as the month of birth, the birth of (naturally conceived) twins, pandemics such as the Spanish Flu in 1918/1919, wars, German reunification, etc. So what makes a good instrument? Let us discuss this using the month of birth and its association with health later in life as an example, where the month of birth is an IV for nutrition and infectious diseases at the time of birth (treatment) and health later in life is the outcome. First, the month of birth must be correlated with seasonal changes in nutrition and infectious diseases to which the infant is exposed (relevance). Clearly, nutrition and infectious diseases vary seasonally. However, with respect to nutrition, seasonality might be weaker in contemporary populations than in historical ones. Therefore, relevance might decrease for younger birth cohorts. Second, month of birth needs to be correlated with health outcomes later in life only through changes in diet and infectious diseases (validity). Theoretical knowledge is needed here. Fetal development is strongly influenced by maternal nutrition, while infant health is strongly influenced by breastfeeding and the infant’s susceptibility to waterborne and airborne diseases. If this is the only pathway, the exclusion criterion is met. While this is largely true for historical populations, this assumption may be violated in contemporary ones, where month of birth affects age at school entry through enrolment periods at a particular time of year, which in turn affects age at school completion, which in turn could affect income and other demographic outcomes related to health later in life (Skirbekk et al. 2004). Third, month of birth does not simultaneously affect nutrition or susceptibility to infectious disease at the time of birth and health later in life (exchangeability), so this condition appears to be met. A good instrument is often characterized by the fact that its connection with the outcome is somehow surprising, because only then are the assumptions of exclusion and exchangeability fulfilled. Only when the connection of the IV with the treatment is theoretically understood does the connection with the outcome become understandable. Another example, puzzling at first glance, is the relationship between particularly long commute times and sex distribution at birth, which can be implemented to examine the effects of early life environment on health.

132  Handbook of health inequalities across the life course Excessive commute times have been associated with skipping breakfast, which is possibly related to a lack of glucose availability during early fetal development. Male fetuses, which are generally more susceptible to adverse conditions in the womb, are therefore at higher risk of spontaneous abortion (Mazumder and Seeskin 2015). Examining determinants of health and health inequalities requires a life course approach that explores exposures at critical and sensitive life stages, as well as interrelated exposures in a risk chain (Kuh et al. 2003). As Palloni et al. (Chapter 16, in this volume) argue, health inequalities “can only manifest over long periods of time, across the life cycle of individuals”. In the following, we distinguish between IV approaches that focus on critical or sensitive life stages, such as the time of birth, and studies that look at different exposures throughout adult life and use the IV approach to distinguish between direct and indirect pathways. We discuss exemplary studies without claiming to be exhaustive. IV Approaches of Critical and Sensitive Periods Early in Life The cohort approach, which uses an individual’s date of birth at a specific time and place as the IV, and IV approaches that utilize exogenous shocks are widely used in studying the effects of critical or vulnerable periods in early life on health later in life. An important reason for this is that longitudinal data on health from birth to old age are scarcely available. The earliest studies explored successive birth cohorts (Doblhammer 2004; Kermack et al. 1934) to examine changes in mortality later in life, and were generally able to establish a cohort effect on mortality in later life. Later studies examined specific cohorts such as those who were born during famines or wars. Studies of the Great Finnish Famine of 1860–1868 (Doblhammer-Reiter et al. 2011; Kannisto et al. 1997), the Chinese Famine 1959–1961 (Song 2009, 2010), the Siege of Leningrad (Sparén et al. 2004; Stanner et al. 1997), the Dutch Potato Famine (Potrafke 2010), and the Dutch Famine at the end of World War II (Portrait et al. 2011) belong to this group. While some of these studies find long-term effects of exposure to famine and war in early life, they often suffer from unobserved heterogeneity due to selective survival, in which only the fittest survive to older ages and effect sizes tend toward zero. A recent study goes beyond simply using exposure to famine and war as IVs and examines the effects of prenatal loss of the father during World War I on life expectancy (Todd et al. 2017). The IV month of birth, which was discussed above, as well as other exceptional events such as influenza pandemics (Almond 2006) and radioactive fall-out (Almond et al. 2009), have also been used in recent studies. Another type of IV often used in the context of historical parish data is the annual fluctuation in infant mortality, which measures the burden of disease faced by an infant in the first year of life (e.g., Bengtsson and Lindström 2003). Alternative IVs are rye prices, which measure short-term economic stress affecting maternal nutrition during pregnancy and breastfeeding, and mortality rates at ages 20–50, which measure the burden of disease on mothers during pregnancy (e.g., Bengtsson and Lindström 2003). In a similar vein, short-term economic recession and boom periods in the first part of the twentieth century have been explored in connection with e.g. mortality (van den Berg et al. 2011) and cognition at old age (Doblhammer, van den Berg, Fritze 2013; Fritze et al. 2014). More recently, the reform of primary care in rural Sweden between 1890 and 1917 has been used to study the long-term effects of environmental conditions in early childhood on mortality in later life (Lazuka 2019), which is an example of the combination of an IV with the difference-in-differences approach

Instrumental variables in studies of health and health inequalities  133 and propensity score matching. Another prominent type of instrumental variable approach is Mendelian randomization, where the instrument is genes (Mills et al. 2020) or genetic IV regression (DiPrete et al. 2018) For a number of reasons, one can argue that natural experiments such as birth month or recession and boom periods are not (valid) instruments. First, the exclusion criteria do not usually apply, since natural experiments can affect the outcome through different pathways. Second, they are usually estimated in reduced form (see “IV estimation” section below) rather than using a two-stage least squares estimator. Finally, the issue of external validity has often been raised, especially when the natural experiments apply only to small groups and generalizability to larger populations is questionable. However, it is important to remember that indicators such as birth weight (Barker 1995), child poverty (Palloni et al., Chapter 16 in this volume; Tampubolon, Chapter 17 in this volume), and education (Dekhtyar and Fors, Chapter 18 in this volume) do not meet the criteria of an IV, but rather are indicators that operate through chains of risk over the life course. Thus, natural experiments used in the sense of IVs are often the only solution to address the endogeneity problem in the study of critical and sensitive periods in the early years of life. IV Approaches Used at Adult Ages Using fixed effect (FE) panel analysis alone without controlling for endogeneity with the help of instrumental variables is now considered insufficient, since an individual’s changes in behavior might be related to reasons that also affect changes in the outcome. Thus, studying changes alone does not solve the endogeneity problem (Moffitt 2009). FE panel analysis with instrumental variables has also been used e.g. by Calvo et al. (2013) with data from the Health and Retirement Study (HRS), combining fixed and random effects with two instrumental variables (changes in statutory full retirement age and in offers made during unexpected early retirement windows); they find that early retirement decreases health. Several studies have used the statutory retirement age as an IV that influences retirement without affecting health. For example, Hessel (2016) uses data from the European Union Statistics of Income and Living Conditions (EU-SILC) and finds that retirement improves health. Hanemann (2017) uses data from HRS, the English Longitudinal Study of Ageing (ELSA), and the Survey of Health, Ageing and Retirement in Europe (SHARE), and finds that retirement improves physical health and deteriorates cognitive health. A potential problem with this method is that it identifies the local average treatment effect (LATE), i.e. the effect of retirement in the population for which the instrument applies. In this case, it is the population that retires because the statutory retirement age is reached. This effect can differ from the overall effect of retirement for all people who retire (see section on “Estimating complier-specific treatment effects” below). There are indications that studies using IVs tend to find positive health effects of retirement, while studies that control for confounding factors find negative effects (Behncke 2012). Combinations of a fixed effects model and an IV are also used to answer another important question of causality, namely the effect of income on health. This approach was applied to a natural experiment by Frijters et al. (2005), examining a causal effect of changes in income on health after German reunification using the German Socio-Economic Panel. They find a statistically significant but small effect. In the same vein as the study above, and again reflecting the great interest of economists in this question, a large number of studies analyze

134  Handbook of health inequalities across the life course the effects between material wealth and health with instrumental variables. It is interesting to see how in the following three studies, sources of external variation are exploited to establish causal evidence. Michaud and van Soest (2008) use inheritances and find no evidence that wealth affects health, but strong evidence of effects from both spouses’ health on household wealth in the Health and Retirement Study. Lindahl (2005) uses the Swedish Level of Living Survey and lottery prizes as an exogenous source of variation in income and finds causal effects from income to health; see also Cesarini et al. (2016) for a more recent application of this approach. Finally, in several publications based on the Health and Retirement Study and the Panel Study of Income Dynamics, Smith (e.g. 2004) uses stock market changes as an instrument for changes in income and concludes that income as such has no effect on health, but socio-economic status does, namely through education. As conceded by Smith himself, it is questionable whether the instruments used in his study or by the other authors above represent the causal effects of material wealth in the general population. IV Estimation As expressed in the assumptions for an IV above, it must be relevant (correlated with the probability of receiving the treatment) and mean-independent of the unobservables directly influencing the outcome. These relationships may be specified in two equations, resulting in the two-stage least square estimator (2SLS). The first equation estimates the relationship between the IV (Z) and the treatment (X) by ordinary least squares (OLS) regression. The predicted values for X are used as independent variables in a second equation, estimating the  (the predicted treatment from the first equation) and the outcome (Y), relationship between X also by OLS regression (Glymour et al. 2017).      Z   C (1) X 0 1 k k    C   (2) Y  0  1X k k Due to use of an estimated variable in the second equation, standard errors need to be adjusted. The 2SLS estimator can be applied to exposures and IVs with multiple values, including continuous variables, and allows simultaneous adjustment for a set of covariates, which should be the same in both stages. For the reduced form IV estimation, where these equations are combined into one (integrating the second equation into the first), for the Wald IV estimate for binary variables, for multiple instruments, as well as for the definition and calculation of weak instruments, see Moffitt (2009) and Glymour et al. (2017). If an IV is not only used to examine the effect of a risk factor or a policy on health in general, but to estimate the effect on health inequality, which means health differences between two groups, one cannot simply apply the textbook procedure from above, but needs to go one step further. Essentially, this implies that the treatment effect is heterogeneous among different groups or that not all groups comply with the treatment. If the differential effect of a treatment on two social groups needs to be analyzed, there are two different approaches: First, the IV analysis can be stratified by the social variable. Different effects in the two groups would indicate that the treatment increases or decreases

Instrumental variables in studies of health and health inequalities  135 health inequality. Secondly, an interaction effect between the social stratifier and the treatment can reveal a differential effect that changes health inequality. While the approach above can easily be stratified by social group, e.g. educational groups, it is more complicated to use an interaction between the stratifier (e.g. education) and the treatment. Because exposure to the treatment is endogenous, the interaction between education and treatment exposure is endogenous as well. This requires an additional instrument for the interaction term. In the first step of the two-stage regression, both instruments predict exposure to the treatment and the interaction between education and exposure to the treatment. The predicted values are then used in the second stage of the regression, resulting in unbiased effects of (1) exposure to the treatment and (2) the interaction between exposure to the treatment and education. The latter serves to assess the effect of the treatment on health inequalities (Hu et al. 2017).

ESTIMATING COMPLIER-SPECIFIC TREATMENT EFFECTS Besides the important property of the IV approach, namely that it can solve the problem of unmeasured confounding explained above, there is another perspective on and potential reason to use IVs, which we discuss in the following. In a standard RCT, the outcome among those assigned to the treatment group is compared to the control group. Such an analysis reveals the effect of being assigned to receive the treatment or the intent to treat (ITT). In reality, where policies or other treatments are implemented or offered to the population or a social group, not every individual will comply with the treatment, which creates the problem of non-compliance. The compliers include two groups: Those who were assigned to the treatment and actually get or take it, and those who were not assigned to the treatment and accordingly do not get or take it. This is shown in Figure 9.2 (taken from Glymour et al. 2017). On the other hand, defiers are those who always do the opposite of what they are assigned to: they take the treatment, even if not assigned to it, and do not take it when assigned it. For the never-takers and the always-takers, the assignment (= the IV) does not change what they would do anyway. Figure 9.2 shows a simplified situation where, first, the treatment is binary and can be “taken”. The logic of the table, however, can be applied to any kind of exposure or (quasi-) experiment. An example may further illustrate the meaning of the categories. Let us assume the policy or treatment consists of a free medical care service in a city. As our IV, we take the distance between the residence of an individual and this service point, because it predicts the usage of the service point, but does not directly influence health, doing so only via the use of the service (Hu et al. 2017). There will be (1) compliers, who use the service because they live close by, or do not use it because they live far away, (2) defiers, who do not use it although they live close by, or use it although they live far away, (3) never-takers, who do not use it, independent of their residence, and (4) always-takers, who use it independent of their residence. Without any further assumptions, the IV estimate is simply the effect among those who live close to the medical service point. Probably, this result will already be more valid than a naive regression estimate comparing those who use the service to those who do not, because the group of people who use it most likely have below-average health from the onset. Extensions of the traditional IV theory to the potential outcome model (Angrist et al. 1996) and the more specific assumption of monotonicity (i.e. the IV only affects treatment in one direction; there are no defiers) allow for the estimation of the more specific effect among the compliers. This

136  Handbook of health inequalities across the life course

Source: Glymour et al. 2017.

Figure 9.2

Categorization of individuals based on how the IV (random assignment) affects the actual exposure (treatment)

is the group whose treatment status changes due to the instrument or, in our example, the group that uses the medical service, because e.g. it is close by (see discussion of LATE below). With regard to a general interest in the effect of such medical services, it may be a too small and too selected group, but with regard to the policy of offering a close-by medical service, it may be exactly the group for which the policy effect is of interest, perhaps of greater interest than the overall effect (Smith 2013) (see discussion below). The assessment of the additional assumptions for the group-specific interpretation of IV estimates and the related terminology can get quite complex. Therefore, we can only give an overview of this discussion in the following and refer to previous literature for more detailed information (e.g. Gangl 2010; Glymour et al. 2017; Klungel et al. 2015; Matthay et al. 2020; Moffitt 2009). Under the assumption that the effect of the treatment is the same for everybody in the population, one can estimate the population average treatment effect (ATE), which is the effect if everyone in the population is treated compared to the situation where nobody is treated. Instruments such as historical, population-wide famines, wars, and economic depressions might largely satisfy this assumption. However, this assumption is often not realistic, particularly when using policies as instrumental variables, because not everybody will comply with the treatment. Another possibility is to estimate the average intention-to-treat effect (ITT), which compares the outcome among the treated to the outcome among the control group. This measure, however, does not give the effect of the treatment itself, because it neglects the defiers. A third option is to estimate the local average treatment effect among the treated

Instrumental variables in studies of health and health inequalities  137 (LATE), which is similar to an instrumental variable estimate, and is the average effect on those that actually comply with their assignment to treatment. The LATE is estimated by dividing the ITT estimate by the estimated share of compliers; one gets exactly the same result when running a 2SLS regression. The LATE might not be very different from the ATE if the share of compliers is large and the treatment effects for the different types in the population are similar enough. The more group-specific an IV estimate is, the more IV estimates can be reconciled with the assumption of effect heterogeneity. In principle, different instruments or different kinds of policies that apply to different population subgroups and their specific effects may be analyzed separately and compared thereafter (Gangl 2010).

SCOPING REVIEW OF IV STUDIES ON HEALTH AND HEALTH INEQUALITIES In this section we provide a brief overview of the use of IV in studies of health and health inequalities in different journals that represent different disciplines, and the trend over time. We performed a literature search in PubMed for the period 2000 to 2021 with the search terms instrumental variables, health inequalities, inequalities in health, disparities in health, health disparities, social differences in health, and social differences in mortality. This yielded 47 studies using IV analysis to study health inequalities. We performed two more searches in ScienceDirect and SocINDEX/EBSCO with the same strategy to cover studies outside the disciplinary fields of PubMed and deleted double findings already included in the PubMed search. These databases yielded 23 and six additional publications, respectively, which makes 76 altogether. One limitation for this literature review was that it was not always easy to categorize the studies as studies on health inequalities, specifically. Although our search terms were designed to look for health inequality studies, in many cases this turned out to be a fuzzy criterion, with gradual differences regarding the question as to whether health inequalities are addressed or just health as an outcome. This fuzziness is due to the fact that studies can empirically measure health in different social groups and then examine the effect of a treatment on the difference in health between these groups (the core of what we were looking for). Alternatively, however, a study can focus on a disadvantaged subgroup of the population only and measure some effect there, arguing that if we know what policy improves the effect in this group, we can also reduce health inequalities as a whole. To illustrate our difficulties in categorizing such studies, the following cases may serve as examples. We classified health differences between a rural and an urban population as social differences in health, because the dimension “rural–urban” constitutes part of the overall concept of social structure in the sociological sense. We did not classify publications as studies on health inequalities that looked at the health effect of home care, insurance coverage, or happiness. This is because the treatment and control groups who get or do not get this “treatment” are not socio-structural groups in a narrow sense, although these three “treatments” may well be relevant factors influencing health inequalities. We admit that these decisions are theoretically complex and to some extent arbitrary, but we think that reporting the results from the literature review still serves the useful purpose of showing the time trend of IV analyses and their distribution across journals and disciplines. The graph in Figure 9.3 shows the publication dates of our 76 review results as a time trend that increases sharply between 2000 and 2021.

138  Handbook of health inequalities across the life course

Figure 9.3

Time trend of IV publications on health inequalities

Table 9.1 shows the journals that appeared in our literature search with the number of IV studies for each journal. The interdisciplinary journal Social Science & Medicine and its related journal SSM Population Health make up about one third of all review findings, followed by the Journal of Health Economics.

DISCUSSION OF STRENGTHS AND WEAKNESSES OF IV The IV method has particular strengths: 1. It can avoid bias from unmeasured confounders of the association between treatment and outcome. 2. It can provide an effect estimate specific to the subgroup that is most affected by the instrument and for which it is most relevant to know the effect of a treatment or a policy. But IV approaches also come with some weaknesses: 1. Valid instruments are difficult to identify and it is difficult to show that they are valid. The search for a valid IV requires a comprehensive knowledge of the subject matter and an intelligent, creative, and indeed fortuitous search for a valid IV. Especially in life course research, it may be easy to show that the IV is related to the treatment, but it is much harder to claim that the IV is not correlated with unobserved confounders (Moore and Brand 2016). 2. Even if the fundamental and additional assumptions of IV are met, a highly specific IV may generate only little variation in the treatment variable (Moore and Brand 2016). 3. This specificity also produces only low external validity for generalizations outside the context of the IV.

Instrumental variables in studies of health and health inequalities  139 Table 9.1

Journals publishing IV studies on health inequalities

Journal

Number of IV publications

Social Science & Medicine

20

Journal of Health Economics

10

SSM Population Health

5

World Development

4

Conference Papers – American Sociological Association

3

Demography

2

Economics and Human Biology

2

Health & Place

2

Health Services Research

2

International Journal of Environmental Research and Public Health

2

Journal of Epidemiology and Community Health

2

African Journal of AIDS Research

1

American Journal of Epidemiology

1

BMC Health Services Research

1

BMC Public Health

1

Demographic Research

1

Economics of Education Review

1

Global Public Health

1

Health Policy and Planning

1

International Journal of Migration, Health & Social Care

1

Journal of Clinical Periodontology

1

Journal of General Internal Medicine

1

Journal of the American Heart Association

1

Medical Care

1

Neurology

1

PLoS One

1

Preventive Medicine

1

Salud Publica de Mexico

1

Social Forces

1

Social Science Research

1

Socio-Economic Review

1

The European Journal of Health Economics HEPAC

1

The Lancet HIV

1

Transport Policy

1

4. Compared to other approaches, an IV analysis often has lower statistical power (or it needs larger samples), because the population group affected by the IV may be small (Muller et al. 2015). Strength no. 2 and weakness no. 3 are the two sides of the same coin. On the one hand, it may be good to know the treatment effect in a specific (disadvantaged) subgroup of the population, especially for the analysis of health inequalities. This may also be the case when the marginal cases are the ones we are interested in. For example, in a study on whether disability insurance discouragements work (Maestas et al. 2013), we do not focus on severely disabled or slightly disabled persons (the cases without any doubt), but on persons who may or may not receive payments, depending on softer or stricter criteria of the insurance scheme. On the other hand, it may be desirable to know the policy effect on the population level, especially when we are concerned with health inequalities, which is by definition an issue on the aggregated level.

140  Handbook of health inequalities across the life course This trade-off can also be illustrated by the famous studies by Lleras-Muney (2002), who used changes in compulsory schooling laws as an instrument to study the effect of education on health (see also Lleras-Muney 2005). This is taken as a positive example of a valid IV in life course research on health inequalities, and the results are interpreted on a rather general population level. But even in this study, Lleras-Muney notes that compulsory schooling laws were enforced much more among white than among black people, which means that the results cannot be generalized to all population subgroups (Matthay et al. 2020). In a comparison of seven quantitative methods for assessing the impact of natural policy experiments on socioeconomic inequalities in health, Hu et al. (2017) used simulated data and come to the conclusion that the low external validity makes IVs less desirable for assessing policy effects on the population level.

CONCLUSION First, we need to acknowledge one limitation of our study that we discovered while writing – that it was not possible to focus our review and discussion of the use of IV on studies on health inequalities, instead of just health. As described in the section “Scoping review of IV studies on health and health inequalities”, there are clear-cut examples of health studies and there are clear-cut examples of health inequality studies, but in the middle there is an area with only gradual differences where it is very hard to decide whether an application of the IV methodology is concerned with health or with health inequalities. In the section on “IV estimation” we outlined two ways to apply an IV analysis to health differences. The most important conclusion should be drawn with regard to the relative advantages or disadvantages of IVs that we tried to illustrate and discuss in this chapter. The trade-off between valid estimates for a small specific group (which may even be the group that we are most interested in) and estimates that can be generalized to more social groups or the whole population is representative of a very general trade-off in causal research between internal and external validity. As many have done before, we argue that it is impossible to aspire solely to one or the other. Smith (2013) advocates a balance of randomization, representation, and realism as integrated aspects of a meaningful causal analysis. When there are two different strategies to addressing causality, the best way is to use them in a complementary way, through a mix of methods and interdisciplinary cooperation (Hoffmann and Doblhammer 2021). We think it is fair to say that social science research on health has underused randomization in the past, as if there were no random processes that matter for health and could be used in the analysis of the social world. That is why IVs may have future potential for sociological analyses of health inequalities, which can also be seen in the increasing numbers of IV analyses and their spread from economics to other disciplines, shown in our literature review. The assumptions of an IV approach are not better or worse than the assumptions of confounder control approaches. What is good about them is that they are different from the assumptions usually used in sociological research. This difference allows for the triangulation of results from different methods and different kinds of data (Matthay et al. 2020). In this endeavor, one can make use of meta-reviews and comparisons of the results from different approaches. As Glymour et al. (2017) note, there are several examples of such comparisons in the literature that find both inconsistent findings from RCTs as compared to observational

Instrumental variables in studies of health and health inequalities  141 studies (Ioannidis 2005; Omenn et al. 1996; Rossouw et al. 2002), but also consistent findings from both approaches (Anglemyer et al. 2014; Benson and Hartz 2000; Concato et al. 2000). There are also interesting reflections on the factors upon which these differences may depend (Anglemyer et al. 2014; Cook et al. 2008; Deaton and Cartwright 2018; Gelman 2018; Ioannidis 2018). In addition to thought-provoking single studies, meta-perspectives such as these may bring about interdisciplinary progress in the long-standing problem of causality. As Moffitt (2005) points out: To reach conclusions about causality we need “synthesis and reconciliation studies that are based on a variety of different approaches”.

REFERENCES Almond, D. (2006), “Is the 1918 influenza pandemic over? Long-term effects of in utero influenza exposure in the post-1940 US population”, Journal of Political Economy, 114(4), 672–712. Almond, D., L. Edlund and M. Palme (2009), “Chernobyl’s subclinical legacy: prenatal exposure to radioactive fallout and school outcomes in Sweden”, Quarterly Journal of Economics, 124(4), 1729–72. Anglemyer, A., H.T. Horvath and L. Bero (2014), “Healthcare outcomes assessed with observational study designs compared with those assessed in randomized trials”, Cochrane Database of Systematic Reviews (4). Angrist, J.D. (1990), “Lifetime earnings and the Vietnam era draft lottery: evidence from social security administrative records”, American Economic Review, 80(3), 313–36. Angrist, J.D., G.W. Imbens and D.B. Rubin (1996), “Identification of causal effects using instrumental variables”, Journal of the American Statistical Association, 91(434), 444–55. Banks, J. and F. Mazzonna (2012), “The effect of education on old age cognitive abilities: evidence from a regression discontinuity design”, Economic Journal, 122(560), 418–48. Barker, D.J.P. (1995), “Fetal origins of coronary heart disease”, BMJ, 311(6998), 171–74. Behncke, S. (2012), “Does retirement trigger ill health?”, Health Economics, 21(3), 282–300. Bengtsson, T. and M. Lindström (2003), “Airborne infectious diseases during infancy and mortality in later life in southern Sweden, 1766–1894”, International Journal of Epidemiology, 32(2), 286–94. Benson, K. and A.J. Hartz (2000), “A comparison of observational studies and randomized, controlled trials”, New England Journal of Medicine, 342(25), 1878–86. Calvo, E., N. Sarkisian and C.R. Tamborini (2013), “Causal effects of retirement timing on subjective physical and emotional health”, Journals of Gerontology Series B: Psychological Sciences and Social Sciences, 68(1), 73–84. Cesarini, D., E. Lindqvist, R. Östling and B. Wallace (2016), “Wealth, health, and child development: Evidence from administrative data on Swedish lottery players”, Quarterly Journal of Economics, 131(2), 687–738. Concato, J., N. Shah and R.I. Horwitz (2000), “Randomized, controlled trials, observational studies, and the hierarchy of research designs”, New England Journal of Medicine, 342(25), 1887–92. Cook, T.D., W.R. Shadish and V.C. Wong (2008), “Three conditions under which experiments and observational studies produce comparable causal estimates: New findings from within‐study comparisons”, Journal of Policy Analysis and Management: The Journal of the Association for Public Policy Analysis and Management, 27(4), 724–50. Davey Smith, G. and S. Ebrahim (2003), “‘Mendelian randomization’: can genetic epidemiology contribute to understanding environmental determinants of disease?”, International Journal of Epidemiology, 32(1), 1–22. Deaton, A. and N. Cartwright (2018), “Understanding and misunderstanding randomized controlled trials”, Social Science & Medicine, 210, 2–21. DiPrete, T.A., C.A.P. Burik and P.D. Koellinger (2018), “Genetic instrumental variable regression: explaining socioeconomic and health outcomes in nonexperimental data”, Proceedings of the National Academy of Sciences, 115(22), E4970-E4979. Doblhammer, G. (2004), The late life legacy of very early life. Springer Science & Business Media.

142  Handbook of health inequalities across the life course Doblhammer, G. and J.W. Vaupel (2001), “Lifespan depends on month of birth”, Proceedings of the National Academy of Sciences, 98(5), 2934–39. Doblhammer, G., G.J. van den Berg and T. Fritze (2013), “Economic conditions at the time of birth and cognitive abilities late in life: evidence from ten European countries”, PloS one, 8(9), e74915. Doblhammer, G., G.J. van den Berg and L.H. Lumey (2013), “A re-analysis of the long-term effects on life expectancy of the Great Finnish Famine of 1866–68”, Population Studies, 67(3), 309–22. Doblhammer-Reiter, G., G.J. van den Berg and L. Lumey (eds) (2011), Long-term effects of famine on life expectancy: A re-analysis of the Great Finnish Famine of 1866–1868. IZA Discussion Paper No. 5534. Frijters, P., J.P. Haisken-DeNew and M.A. Shields (2005), “The causal effect of income on health: evidence from German reunification”, Journal of Health Economics, 24(5), 997–1017. Fritze, T., G. Doblhammer and G.J. van den Berg (2014), “Can individual conditions during childhood mediate or moderate the long-term cognitive effects of poor economic environments at birth?”, Social Science & Medicine, 119, 240–48. Gangl, M. (2010), “Causal inference in sociological research”, Annual Review of Sociology, 36, 21–47. Gelman, A. (2018), “Benefits and limitations of randomized controlled trials: a commentary on Deaton and Cartwright”, Social Science & Medicine (1982), 210, 48–49. Glymour, M.M., S. Walter and E.J. Tchetgen Tchetgen (2017), “Natural experiments and instrumental variables analyses in social epidemiology”, in J. Michael Oakes and Jay Kaufman (eds), Methods in social epidemiology, San Francisco: Jossey-Bass, pp. 493–537. Gørgens, T., X. Meng and R. Vaithianathan (2012), “Stunting and selection effects of famine: a case study of the Great Chinese Famine”, Journal of Development Economics, 97(1), 99–111. Hanemann, F. (ed.) (2017), Working Conditions, retirement and health – longitudinal evidence from Europe and the US. MEA Discussion Paper 07-2017, Munich. Hessel, P. (2016), “Does retirement (really) lead to worse health among European men and women across all educational levels?”, Social Science & Medicine, 151, 19–26. Hoffmann, R. and G. Doblhammer (2021), “Approaches and methods for causal analysis of panel data in the area of morbidity and mortality”, Comparative Population Studies, 46. Hu, Y., F.J. van Lenthe, R. Hoffmann, K. van Hedel and J.P. Mackenbach (2017), “Assessing the impact of natural policy experiments on socioeconomic inequalities in health: how to apply commonly used quantitative analytical methods?”, BMC Medical Research Methodology, 17(1), 1–17. Huntington, E. (1938), Season of birth – its relation to human abilities. Wiley. Ioannidis, J.P.A. (2005), “Contradicted and initially stronger effects in highly cited clinical research”, Jama, 294(2), 218–28. Ioannidis, J.P.A. (2018), “Randomized controlled trials: often flawed, mostly useless, clearly indispensable: a commentary on Deaton and Cartwright”, Social Science & Medicine (1982), 210, 53–56. Kannisto, V., K. Christensen and J.W. Vaupel (1997), “No increased mortality in later life for cohorts born during famine”, American Journal of Epidemiology, 145(11), 987–94. Kermack, W.O., A.G. McKendrick and P.L. McKinlay (1934), “Death-rates in Great Britain and Sweden. Some general regularities and their significance”, Lancet, 698–703. Klungel, O.H., J.M. Uddin, A. de Boer, S.V. Belitser, R.H. Groenwold and K. Roes (2015), “Instrumental variable analysis in epidemiologic studies: an overview of the estimation methods”, Pharmaceutica Analytica Acta, 6(4), 2. Kuh, D., Y. Ben-Shlomo, J. Lynch, J. Hallqvist and C. Power (2003), “Life course epidemiology”, Journal of Epidemiology and Community Health, 57(10), 778–83. Lazuka, V. (2019), “Early-life assets in oldest-old age: evidence from primary care reform in early twentieth century Sweden”, Demography, 56(2), 679–706. Lindahl, M. (2005), “Estimating the effect of income on health and mortality using lottery prizes as an exogenous source of variation in income”, Journal of Human Resources, 40(1), 144–68. Lleras-Muney, A. (2002), “Were compulsory attendance and child labor laws effective? An analysis from 1915 to 1939”, The Journal of Law and Economics, 45(2), 401–35. Lleras-Muney, A. (2005), “The relationship between education and adult mortality in the United States”, Review of Economic Studies, 72(1), 189–221.

Instrumental variables in studies of health and health inequalities  143 Maestas, N., K.J. Mullen and A. Strand (2013), “Does disability insurance receipt discourage work? Using examiner assignment to estimate causal effects of SSDI receipt”, American Economic Review, 103(5), 1797–1829. Martens, E.P., W.R. Pestman, A. de Boer, S.V. Belitser and O.H. Klungel (2006), “Instrumental variables: application and limitations”, Epidemiology, 17(3), 260–67. Matthay, E.C., E. Hagan, L.M. Gottlieb, M.L. Tan, D. Vlahov, N.E. Adler and M.M. Glymour (2020), “Alternative causal inference methods in population health research: evaluating tradeoffs and triangulating evidence”, SSM-Population Health, 10, 100526. Mazumder, B. and Z. Seeskin (2015), “Breakfast skipping, extreme commutes, and the sex composition at birth”, Biodemography and Social Biology, 61(2), 187–208. Michaud, P.-C. and A. van Soest (2008), “Health and wealth of elderly couples: causality tests using dynamic panel data models”, Journal of Health Economics, 27(5), 1312–25. Mills, M.C., N. Barban and F.C. Tropf (2020), An introduction to statistical genetic data analysis. MIT Press. Moffitt, R. (2005), “Remarks on the analysis of causal relationships in population research”, Demography, 42(1), 91–108. Moffitt, R.A. (2009), “Issues in the estimation of causal effects in population research, with an application to the effects of teenage childbearing”, in H. Engelhardt, H.-P. Kohler and A. Fürnkranz-Prskawetz (eds), Causal analysis in population studies. Springer, pp. 9–29. Moore, R. and J.E. Brand (2016), “Causality in life course studies”, in M.J. Shanahan, J.T. Mortimer and M. Johnson (eds), Handbook of the life course. Springer, pp. 515–39. Muller, C., C. Winship and L. Stephen (2015), “Instrumental variables regression”, in H. Best and C. Wolf (eds), The SAGE handbook of regression analysis and causal inference. Sage, pp. 277–99. Omenn, G.S., G.E. Goodman, M.D. Thornquist, J. Balmes, M.R. Cullen, A. Glass, J.P. Keogh, F.L. Meyskens Jr, B. Valanis and J.H. Williams Jr (1996), “Risk factors for lung cancer and for intervention effects in CARET, the Beta-Carotene and Retinol Efficacy Trial”, JNCI: Journal of the National Cancer Institute, 88(21), 1550–59. Portrait, F., E. Teeuwiszen and D. Deeg (2011), “Early life undernutrition and chronic diseases at older ages: the effects of the Dutch famine on cardiovascular diseases and diabetes”, Social Science & Medicine, 73(5), 711–18. Potrafke, N. (2010), “The growth of public health expenditures in OECD countries: do government ideology and electoral motives matter?”, Journal of Health Economics, 29(6), 797–810. Rossouw, J.E., G.L. Anderson, R.L. Prentice, A.Z. LaCroix, C. Kooperberg, M.L. Stefanick, R.D. Jackson, S.A.A. Beresford, B.V. Howard and K.C. Johnson (2002), “Risks and benefits of estrogen plus progestin in healthy postmenopausal women: principal results from the Women’s Health Initiative randomized controlled trial”, Jama, 288(3), 321–33. Skirbekk, V., H.-P. Kohler and A. Prskawetz (2004), “Birth month, school graduation, and the timing of births and marriages”, Demography, 41(3), 547–68. Smith, H.L. (2013), “Research design: toward a realistic role for causal analysis”, in S.L. Morgan (ed.), Handbook of Causal Analysis for Social Research. New York: Springer, pp. 45–73. Smith, J.P. (2004), “Unraveling the SES: health connection”, Population and Development Review, 30, 108–32. Song, S. (2009), “Does famine have a long-term effect on cohort mortality? Evidence from the 1959–1961 Great Leap Forward Famine in China”, Journal of Biosocial Science, 41(4), 469–91. Song, S. (2010), “Mortality consequences of the 1959–1961 Great Leap Forward famine in China: debilitation, selection, and mortality crossovers”, Social Science & Medicine, 71(3), 551–58. Sparén, P., D. Vågerö, D.B. Shestov, S. Plavinskaja, N. Parfenova, V. Hoptiar, D. Paturot and M.R. Galanti (2004), “Long term mortality after severe starvation during the siege of Leningrad: prospective cohort study”, BMJ, 328(7430), 11. Stanner, S.A., K. Bulmer, C. Andres, O.E. Lantseva, V. Borodina, V.V. Poteen and J.S. Yudkin (1997), “Does malnutrition in utero determine diabetes and coronary heart disease in adulthood? Results from the Leningrad siege study, a cross sectional study”, BMJ, 315(7119), 1342–48. Swanson, S.A. and M.A. Hernán (2013), “Commentary: how to report instrumental variable analyses (suggestions welcome)”, Epidemiology, 24(3), 370–74.

144  Handbook of health inequalities across the life course Todd, N., A.-J. Valleron and P. Bougnères (2017), “Prenatal loss of father during World War One is predictive of a reduced lifespan in adulthood”, Proceedings of the National Academy of Sciences, 114(16), 4201–6. van den Berg, G.J., G. Doblhammer-Reiter and K. Christensen (2011), “Being born under adverse economic conditions leads to a higher cardiovascular mortality rate later in life: Evidence based on individuals born at different stages of the business cycle”, Demography, 48(2), 507–30. van den Berg, G.J., M. Lindeboom and F. Portrait (2006), “Economic conditions early in life and individual mortality”, American Economic Review, 96(1), 290–302. van Kippersluis, H. and C.A. Rietveld (2018), “Beyond plausibly exogenous”, Econometrics Journal, 21(3), 316–31. Vogt, T.C. and J.W. Vaupel (2015), “The importance of regional availability of health care for old age survival – findings from German reunification”, Population Health Metrics, 13(1), 1–8. Wright, P.G. (1928), Tariff on animal and vegetable oils. Macmillan Company, New York.

PART III MECHANISMS AND EMPIRICAL EVIDENCE FOR HEALTH INEQUALITIES AT STAGES OF THE LIFE COURSE

10. Health inequalities in adolescence and their consequences for (emerging) adulthood Marie Bernard, Kristina Winter, and Irene Moor

THE IMPORTANCE OF THE ADOLESCENCE DEVELOPMENT STAGE Childhood and adolescence are very vulnerable phases of life in which important “courses” are set for current health, but also for health in later life. Adolescence is established as a distinct phase of life between childhood and adulthood (Hurrelmann & Quenzel, 2016; Kuhn & King, 2021; Lampert, 2010; Patton et al., 2016; Sawyer et al., 2012). Similar to childhood, adolescence is also characterised by dramatic physical, cognitive, social, and emotional developments and changes, including physical growth, as well as the emergence of health-related and risk behaviours (Pearce et al., 2020). Moreover, the onset of puberty is an important typical marker in adolescence and “is part of an often prolonged life course process connecting childhood to adolescence and adolescence to young adulthood” (Johnson et al. 2011, p. 276). In the wake of this, the relationships with significant others, the environment, and oneself are transformed in crucial ways. The qualities of the “adolescent possibility space” (King, 2013) are particularly significant for shaping adolescent self-identity and adolescent re-creation in the transition between childhood and adulthood. Considering the heterogeneous and thus unequal opportunities and conditions of growing up, the quality of adolescent possibility spaces can differ significantly (Kuhn & King, 2021). The developmental tasks of adolescents, the importance of the social environment, and possible resources are the focus of socialisation theory approaches. “Protected” developmental spaces for adolescents are rather rare due to, for example, the risk of poverty, the pluralisation of family lifestyles, or increasing demands for successful high educational attainment (Bengel et al., 2009). Hurrelmann and Bauer (2015) identified three main instances of socialisation: (1) the family as the primary, (2) school and peers as the secondary, and (3) the vocational training institution and workplace as the tertiary instance of socialisation. While the latter plays an important role in (emerging) adulthood, the primary and secondary socialisation instances are the most important socialisation contexts for young people. In childhood, socialisation takes place predominantly in the family, the initial social context. Within this context, individuals undergo their first socialisation experiences and developmental steps. Thus, the family provides the primary conditions of socialisation on which everything else is built on (Corsten, 2020; Hurrelmann & Bauer, 2015). The adolescent phase is primarily characterised by the transition from the guided and emotional relationships of the close social environment – the family. But also, the transition to a more and more independently selectable diverse circle of relationships is typical for the adolescent phase. The school and peers as secondary socialisation instances thus represent important developmental spaces (Hurrelmann & Quenzel, 2016). As adolescents spend more time with their peers, they adapt themselves strongly to the predominant attitudes and 146

Health inequalities in adolescence and their consequences for (emerging) adulthood  147 behavioural patterns. Besides family, school represents the main institutional context shaping their physical, psychological, and social development. The organisation of school and school attendance is closely related to how life phases, such as childhood, adolescence, and emerging adulthood are structured. At school, adolescents encounter an environment that places new demands and spheres of action on them that differ from the family environment (Corsten, 2020; Hurrelmann & Bauer, 2015). For a long time, research regarding the health of children and adolescents was neglected because it is a relatively healthy period of life (Patton et al., 2016). However, indicators of mental health and risky behaviours particularly show health impairments and increased risks for adolescents’ current and also future health. Commonly, these behaviours are explored during adolescence and often continued into adulthood. During adolescence, young people are distancing themselves from the parental home, and the secondary socialisation instance gains importance in adolescents’ lives. Therefore, research on adolescent health has particularly focused on potential resources and risks from the mentioned social contexts. Relevant studies show correlations between the (mental) health and well-being of adolescents and, for example, social support from classmates, teachers, and friends, school satisfaction and connectedness, as well as school stress and pressure (Cosma et al., 2020; Joyce & Early, 2014; Markkanen et al., 2019; Ringdal et al., 2020; Vaičiūnas & Šmigelskas, 2019). In addition, associations are found between social characteristics and health behaviours, such as substance use and bullying (Kim & Chun, 2018; Picoito et al., 2019; Ringdal et al., 2020; Simetin et al., 2011; Vogel et al., 2015). Further, lack of friendships, social isolation, or poor relationships with friends have a negative impact on (mental) health and well-being. However, the circle of friends can also have negative influences. Depending on predominant attitudes and health behaviour among peers, negative habits can taint one’s own health, for example, by leading to risky health behaviours, or even posing an increased risk for depression and suicide (Barman-Adhikari et al., 2016; Fulginiti et al., 2016; Laursen et al., 2007; Moor et al., 2020a). In this respect, socialisation phases always represent specific life course phases (Corsten, 2020). These can be associated with both opportunities and risks, depending on with whom adolescents are confronted on their way to self-determination and (healthy) personality development and whose individual coping patterns are continued into adulthood. These challenges, as well as associated resources and risks, encompass the adolescent lifeworld and are therefore of particular importance for the health-related and psychosocial development of young people (Fröhlich-Gildhoff & Rönnau-Böse, 2021; Hurrelmann & Quenzel, 2016; Patton et al., 2016; Sawyer et al., 2012).

HEALTH INEQUALITIES IN ADOLESCENCE AND HOW THESE INEQUALITIES ARISE Although most adolescents have a good health status, there are some at particular health risk. These are those who grow up under disadvantageous socioeconomic conditions or poverty, which are often linked to several challenges and adverse childhood experiences (e.g., violence, child maltreatment, sexual abuse, or losses). These adolescents are more vulnerable, as such experiences can have serious and lasting consequences for their (mental) health and conditions in future life stages (Sethi, 2013; Weber Ku et al., 2020). Thus, not every adolescent has equal

148  Handbook of health inequalities across the life course life and health chances due to unfavourable conditions as they grow up, depending on the family they were born into (Elgar et al., 2015; Inchley et al., 2020; Viner et al., 2012). In the following, we will focus on adolescents who have lower health chances because of their socioeconomic position (SEP) compared to young people with a higher socioeconomic background. These socioeconomic inequalities were identified for many health outcomes in young people (Inchley et al., 2020; Poulain et al., 2019). For example, adolescents with a lower SEP were two or three times more likely to develop mental health problems. Those who persisted over time in socioeconomically disadvantageous circumstances report higher rates of low mental health, and a decrease in SEP is associated with increased mental health problems as a systematic review showed (Reiss 2013). Further, higher rates of psychosocial health complaints as well as lower subjective levels of health and well-being were also more often reported by young people with low SEP (Moor et al., 2019; Torsheim et al., 2018; Weinberg et al., 2019). Health inequalities were also observed in health and risk behaviours in adolescence. Those with lower SEP appear, for example, to be less physically active and have a less healthy diet (Chzhen et al., 2018; Kalman et al., 2015), are more frequently overweight or obese (Frederick et al., 2014; Inchley et al., 2017, 2020) and are more likely to smoke (Lorant et al., 2017; Pförtner et al., 2015). For example, adolescents with a lower educational level have about a two times higher risk of smoking compared to those at the highest educational level (gymnasium). However, results for alcohol consumption or cannabis use are heterogeneous (Moor et al., 2020b). The strength of the association between health and health behaviours with SEP varies depending on age, gender, health outcomes and indicators of socioeconomic position (Inchley et al., 2020; Moor et al., 2019; Reiss, 2013). Despite this, trend analyses found that the existing health inequalities in young people mostly did not decrease but rather have persisted over decades or even increased in some health outcomes (Elgar et al., 2015; Moor et al., 2015). However, there are also studies showing heterogeneous findings according to health inequalities in adolescence, indicating no or weak associations between SEP and health (Inchley et al., 2020). As an underlying reason for these results, different measures of socioeconomic inequalities in adolescence need to be considered. There is evidence that the impact of parental indicators on adolescent social position and health inequalities decreases at some point. In contrast, adolescent-specific indicators (e.g., school type or academic performance, subjective social status) gain more importance (Ahlborg et al., 2017; Elgar et al., 2016; Moor et al., 2019). Another reason for a weak(er) association between SEP in youth and health was stated by West (1997) in his thesis on “equalization in youth” (West, 1997). The assumption is that in adolescence, the effects of socioeconomic inequalities flatten and increase again in the life period of emerging adulthood. It is argued that school, peers, and youth culture are relevant to all adolescents, thus levelling the effects out. Adolescents with lower SEP would benefit from integration in school and peer-group activities. In contrast, adolescents from highly affluent families would perceive high pressure regarding the educational aspirations of their parents, leading to a higher burden of school stress (Klocke et al., 2020). As the evidence predominantly shows socioeconomic inequalities in adolescent health, it is rather a period where the inequalities might be less visible or pronounced but are still factual. So why does this inequity in young people’s health already exist, even before they reach their own SEP? Although there are many relevant factors for health and health inequalities, research on the explanation of health inequalities in adolescence is still incomplete. In the following, we will focus on the mechanisms and pathways which are important to explain socio-

Health inequalities in adolescence and their consequences for (emerging) adulthood  149 economic inequalities in adolescent health based on the different levels of health determinants – the individual (micro) level, the organisational/institutional (meso) level and the macro level. On the micro level, three pathways are seen as key mechanisms to explain health inequalities: the materialist/structural, the psychosocial, and the behavioural (Moor et al., 2017; Scambler, 2012). The material/structural pathway describes the importance of material and structural living conditions, which include financial resources (income or wealth) as well as living (housing) and working circumstances. It is shown that those with greater access to material resources have more chances of experiencing healthy environments such as higher standards of living or a less health-detrimental job, as well as the possibility to buy the essentials for a healthy lifestyle (e.g., healthy food that is generally more expensive). Thus, they often have better health compared to those with less financial resources and less favourable structural environments of living and working conditions. Further, it is recognised that it is not only the absolute material/structural deprivation that matters but also how people react and perceive their situation and the (psychosocial) resources they have to handle their circumstances (Marmot & Wilkinson, 2001). This assumption is covered by the psychosocial pathway. Studies have found that a low SEP is often linked with unequal exposure and vulnerability to psychosocial factors (Denton et al., 2004). For example, people with lower SEP are faced with more hazards and burdens in their life compared to those who are socioeconomically better off, and, in addition, have less psychosocial resources such as coping strategies and self-esteem to handle them in a healthy way. They also compare themselves negatively (negative upward social comparison) to better-off people, which is why they often feel frustrated and ashamed, as well as perceiving higher stress, resulting in more mental burdens (Kondo et al., 2008). The third mechanism describes the behavioural pathway, in which it is stated that higher risk behaviour and less healthy behaviour are more often found in those with lower SEP, for example higher prevalence of smoking and alcohol misuse, more sedentary behaviour, less physical activity, etc. In comparison, those with higher SEP seem to adhere to a healthier lifestyle (Laaksonen et al., 2005; van Oort et al., 2005). A systematic review could show that these three pathways can explain a large part of the inequalities in self-rated health (Moor et al., 2017). It is important to state that they are related to each other and that one pathway cannot fully explain the link between SEP and health; thus, a large part of these inequalities is the result of those pathways with shared contributions (Moor et al., 2017; Stronks et al., 1996; van Oort et al., 2005). This fact underlines the importance of analysing these three mechanisms simultaneously for explaining health inequalities. However, most of these studies focused on the adult population (Aldabe et al., 2011; Granström et al., 2021; Laaksonen et al., 2005; Pförtner & Moor, 2017; Stronks et al., 1996; van Oort et al., 2005), and only a few studies analysed the contribution of these pathways for adolescents. However, they could confirm the importance of material, psychosocial and behavioural factors and their shared contribution to explaining health inequalities in adolescence (Moor et al., 2014a, 2014b; Richter et al., 2012). Still, a large part of the inequalities could not be explained by these pathways on the individual level, leading to the assumption that there are other factors on the meso or macro level that also play an important role. Compared to micro- or macro-level determinants for explaining health inequalities in adolescent health, the meso-level determinants have been analysed less comprehensively (Richter & Dragano, 2018), but they seem to have an additional explaining power that merits further investigation. The meso-level can be understood as smaller-scale entities with institutional structures in which individuals live, grow and work (e.g., kindergarten, school, workplaces,

150  Handbook of health inequalities across the life course etc.) (Richter & Dragano, 2018). As individuals are socialised within social or rather institutional contexts with differing habits and norms, social interaction and living environments are formed by meso-level structures. When looking at individual factors solely, structural mechanisms that have the potential to compensate for the risks facing young people from low SEP are neglected. The socio-environmental approach addresses this shortcoming by understanding individual risk factors as embedded in social environments that, in turn, shape health-related behaviour (Raphael, 2013). Instead of apprehending poorer health behaviours such as substance consumption, poorer diet, and lack of physical activity among low SEP groups as lifestyle choices, they should rather be acknowledged as coping mechanisms shaped by disadvantageous structural, social, and economic circumstances. As adolescents spend most of their time at school with their classmates with whom they (need to) interact, the school can be considered the most important institution on the meso-level. School and the classroom are important educational settings with different meso-level characteristics such as different learning environments, participation possibilities, and relationships with teachers and classmates (Eccles & Roeser, 2011), which can influence their health and are responsible for health inequalities. On the meso level, it is necessary to differentiate between contextual and compositional factors. Contextual factors describe the characteristics of schools or classes taking into consideration the organisational, structural, cultural, and physical characteristics. In contrast, compositional characteristics refer to the (social) composition of the school and classroom, based on aggregating information about their sociodemographic composition or school-rated factors (e.g., teacher–student ratio or school/class climate). While these meso-level characteristics have been analysed regarding their impact on students’ academic outcomes and partly on health, the importance of explaining health inequalities has rarely been studied, as Herke et al. (2022) found in a scoping review. The authors summarised the current evidence on the meso-level characteristics of the school and their role in health inequalities in adolescence. They found 26 studies, of which 12 studies identified a moderating and three studies a mediation effect of the meso-level characteristic of the school between SEP and adolescent health. In particular, the school composition had a strong moderating impact, such as that a low average SEP at school reinforced the negative impact of a low SEP on students’ mental health and well-being. However, advantageous meso-level characteristics of the school were able to partly cushion the negative effects of low individual SEP on students’ health (Herke et al. 2022). In explaining health inequalities, researchers need to consider different mechanisms and pathways on different levels. Although the evidence has increased in the last years, the link between SEP and health is not yet fully understood, particularly during adolescence, but also in terms of how these inequalities in adolescence evolve into later life.

THE LONG-TERM IMPACT OF SOCIOECONOMIC HEALTH INEQUALITIES IN ADOLESCENCE ON LATER LIFE Although health inequalities in adolescence are less pronounced or less visible compared to earlier and later life stages, it can be assumed that these health inequalities continue with increasing age. However, there are only a few long-term studies that investigate how socioeconomic inequalities affect the health of adolescents and their future health.

Health inequalities in adolescence and their consequences for (emerging) adulthood  151 When considering the long-term effects of adolescents’ health inequalities on later life, it is important to take a closer look at the transition between adolescence and adulthood. This phase can be characterised as highly dynamic because of the numerous environmental, social, and lifestyle changes young adults face. Although most adolescents detach from their parents (or at least try), they still depend on them financially. At the beginning of emerging adulthood, there is another shift regarding their (in-)dependencies. For example, during this phase, young adults often leave their parental home and the tightly controlling educational institutions (i.e., school). Thus, they experience greater independence and less social control (Stone et al., 2012), which in turn has the potential to amplify high-risk health behaviour, such as poorer diet (Winpenny et al., 2018) or substance use (Stone et al., 2012). Moreover, health-related habits such as smoking and physical activity, which are established in adolescence, are likely to be carried over into adulthood. Nevertheless, unhealthy habits might also diminish over time depending on individual determinants, circumstances, and lifestyle choices. The previous section showed what kind of factors determine the mental and physical health inequalities among adolescents and which driving forces and barriers need to be considered. But to what extent are SEP-based advantages and disadvantages, as well as health inequalities in adolescence, transferred into (emerging) adulthood? As mentioned before, there is a lack of longitudinal or rather panel data on this subject. The examples given should thus not be considered as exhaustive in terms of the state of research but will rather give a brief insight into current findings that somehow underline the persistent character of health inequalities. There are divergent lines of thoughts on how health (inequalities) trajectories are manifested throughout the life course and particularly in emerging adulthood. The most prominent approach is the theory of cumulative advantages. This implies that favourable resources, such as economic, educational, and other advantages at a younger age are compounded in future adult life (DiPrete & Eirich, 2006). According to this hypothesis, the health of individuals diverges systematically over their lifetime based on their personal resources (e.g., (health) behaviours, social relations) and particularly based on their SEP (DiPrete & Eirich, 2006; Due et al., 2011). This approach argues that people with higher and lower SEP experience divergent health trajectories. Young people with better resources, that is, advantageous life circumstances (e.g., high familial SEP and support system), are provided with a good foundation for favourable health trajectories. In turn, individuals with unfavourable backgrounds often have fewer resources and experience more stressful life events (Wickrama et al., 2009), leading to poorer health over the life course. During emerging adulthood, parental/familial SEP-based influences emerge, for example when deciding on a career or educational path. Literature has shown that parental encouragement to attain higher educational or occupational success differs between SEP groups, which may subsequently contribute to a resurgence of SEP-based health inequalities. Thus, the long arm of childhood and the process of cumulative (dis)advantages regain momentum in emerging adulthood (Wickrama et al., 2009). Since resources and health (dis)advantages cumulate with age, health inequalities will strengthen over time (Crystal et al., 2017; Prus, 2007). Following this approach, health inequalities over the life course do not only maintain but also increase relative inequalities between divergent social groups. Another approach regarding health trajectories argues that young adults can modify their supposedly predetermined health-related trajectories despite their socioeconomic background by changing their life circumstances in early adulthood (i.e., theory of social mobility). However, although some adolescents can rewrite their (health) trajectories by making good choices, it can be assumed that most young adults cannot (Wickrama et al., 2009). Either

152  Handbook of health inequalities across the life course way, the phase of emerging adulthood must be considered an important stage in life in which the direction of health trajectories is manifested. This time of transition can thus not only be acknowledged as a critical period but can also be seen as a window of opportunity. Pampel et al., for example, argue that both approaches (i.e., cumulative (dis)advantages and social mobility) can be applied to explain health trajectories during the life course (Pampel et al., 2014). By investigating the tobacco use of young adults and its link to (parental) SEP and changing life circumstances during the transition into emerging adulthood, they provided evidence that both approaches exist. On the one hand, they showed that parental SEP does not only translate into smoking disparities among young adults but also widens these disparities in emerging adulthood (Pampel et al., 2014). On the other hand, they found that young adults can redirect their health trajectories by achieving socioeconomic attainment or by taking social roles with a high level of commitment in early adulthood. However, since these lifestyle changes are somewhat related to socioeconomic background, a stronger indication for the cumulative approach is assumed. The negative impact of smoking on other health outcomes is undisputed. It has been shown that smoking is a strong mediator between SEP and cardiovascular and metabolic diseases. Moreover, tobacco use was found to contribute the most (compared to alcohol consumption, physical activity, and diet) to socioeconomic health inequalities (Petrovic et al., 2018). Such long-term effects of adolescent SEP can also be expected regarding physical activity in adult life. Although to date there are no longitudinal data assessing the long-term effects of SEP in younger years and engagement in physical activity in adult life, an association can be assumed when looking at previous findings. Previous literature indicates a positive association between SEP and physical activity in adolescence, that is, more physical activity in higher SEP individuals (Stalsberg & Pedersen, 2010). For example, Batista and colleagues found at least a moderate association of physical activity among adults who reported early life engagement in physical activity (Batista et al., 2019). In contrast, others detected a general decrease in physical activity in young adulthood (Corder et al., 2019). Thus, it might be assumed that socioeconomic advantages in younger years are correlated with more physical activity, even though this somewhat decreases in adult life (Stalsberg & Pedersen, 2010). These findings are highly relevant considering that physical activity is a modifiable risk factor for non-communicable diseases, such as obesity. Obesity is one of the most severe public health issues worldwide that is tenacious throughout the life course and comes along with a wide range of negative physical and mental health outcomes (Burki, 2021; Newton et al., 2017). In their systematic review and meta-analysis, Simmonds et al. (2016) investigated whether obesity and overweight in adolescents predict an excessive BMI in adulthood. Although obesity is not necessarily carried from childhood to adolescence, the authors found that obesity persisted from adolescence to adulthood. Obesity remained unchanged in 80% of adolescents with obesity until the age of 40 (no data available beyond that age). This is concerning since excessive overweight is found more often in lower SEP groups (Frederick et al., 2014; Inchley et al., 2017, 2020) throughout the life course, particularly in women (Newton et al., 2017). It can, therefore, be concluded that SEP-based health inequalities in the form of (excessive) overweight in adolescence are carried over into adulthood. SEP-based health inequalities are also pronounced in cardiovascular diseases (CVDs), such as ischemic heart disease and cerebrovascular disease. In high-income countries, low SEP is linked to an increased risk for CVDs. This association can be traced back to various risk factors

Health inequalities in adolescence and their consequences for (emerging) adulthood  153 varying between SEP groups (e.g., behavioural, psychosocial, and environmental factors, as well as health services use). The long-lasting impact of familial socioeconomic background on health was also shown: there is some evidence that individuals who grew up with fewer socioeconomic resources are at higher risk of having cardiovascular disease, irrespective of their socioeconomic achievements in adulthood (Clark et al., 2009; Galobardes et al., 2006). Previous research also demonstrates that (parental) SEP is not only linked to physical (Link et al., 2017) but also to mental health outcomes (Wickrama et al., 2009) in adulthood. Depressive symptoms in adulthood were found to be strongly associated with family SEP in early adolescence. While in middle and late adolescence the association seems to diminish (i.e., “equalization in youth” (West, 1997)), there is some evidence that the process of health equalisation is only temporary since SEP-based disparities in depressive symptoms resurface in emerging adulthood (Wickrama et al., 2009). These findings back up the approach of cumulative (dis)advantages. In their systematic review, Johnson et al. (2018) found strong evidence that depression in adolescence is strongly associated with depression and anxiety disorders in adulthood, and even some evidence for a link between depression in adolescence and suicidality in adulthood. Considering how SEP determines health-related behaviour, such as tobacco use, lack of physical activity, and having (excessive) overweight, and how SEP-based health inequalities widen with age, SEP can be seen as a predictor for developing comorbidities throughout the life course (Katikireddi et al., 2017). There is some evidence that not only adulthood SEP but also SEP in younger years is independently associated with a higher risk of developing multi-morbidity at all ages (Khanolkar et al., 2021).

CONCLUSION Adolescence is a sensitive stage of life between childhood and adulthood that includes several physical, emotional, and social developmental changes. The socialisation context, such as the family, but especially the social environment (school and peers), is of particular importance. Adolescents are engaged to find their social roles and embark on their (social and educational) paths. This life period lays thus the foundation for their future lives and consequently their health. Although most adolescents assess their health as good, their life and health chances are unequally distributed, to the disadvantage of those who grow up with lower socioeconomic backgrounds. These inequalities can be explained to a large part by considering micro-level mechanisms such as material/structural, psychosocial, and behavioural factors or meso-level characteristics that shape the institutional contexts in which they grow up (e.g., school characteristics). However, macro-level determinants are also important, as they refer to macroeconomic, social, and public health policy. It is argued that many of the key determinants of health can be best addressed at a population level as they have a tremendous impact on our living and working circumstances and our choices regarding our lifestyle behaviours (Thomson et al., 2016; Walls et al., 2018). Macro-level structures are of particular importance when considering macrosocial events such as the COVID-19 pandemic, which emphasised SEP-based differences for health risks. Although the longitudinal database is rather small, individual SEP and parental/familiar socioeconomic background must be acknowledged as important determinants of health trajectories. The current state of research underlines how long-lasting circumstances in younger years translate into health trajectories in later life. Health inequali-

154  Handbook of health inequalities across the life course ties have widened significantly over the past decades. These findings are rather alarming and emphasise the urgent need for early life intervention as “investments in adolescent health and wellbeing bring benefits today, for decades to come, and for the next generation” (Patton et al. 2016, p. 2423).

REFERENCES Ahlborg, M., Svedberg, P., Nyholm, M., Morgan, A., & Nygren, J. M. (2017). Socioeconomic inequalities in health among Swedish adolescents – adding the subjective perspective. BMC Public Health, 17(1), 838. https://​doi​.org/​10​.1186/​s12889​-017​-4863​-x Aldabe, B., Anderson, R., Lyly-Yrjänäinen, M., Parent-Thirion, A., Vermeylen, G., Kelleher, C. C., & Niedhammer, I. (2011). Contribution of material, occupational, and psychosocial factors in the explanation of social inequalities in health in 28 countries in Europe. Journal of Epidemiology and Community Health, 65(12), 1123–1131. https://​doi​.org/​10​.1136/​jech​.2009​.102517 Barman-Adhikari, A., Begun, S., Rice, E., Yoshioka-Maxwell, A., & Perez-Portillo, A. (2016). Sociometric network structure and its association with methamphetamine use norms among homeless youth. Social Science Research, 58, 292–308. https://​doi​.org/​10​.1016/​j​.ssresearch​.2016​.01​.004 Batista, M. B., Romanzini, C. L. P., Barbosa, C. C. L., Blasquez Shigaki, G., Romanzini, M., & Ronque, E. R. V. (2019). Participation in sports in childhood and adolescence and physical activity in adulthood: A systematic review. Journal of Sports Sciences, 37(19), 2253–2262. https://​doi​.org/​10​.1080/​ 02640414​.2019​.1627696 Bengel, J., Meinders-Lücking, F., & Rottmann, N. (2009). Schutzfaktoren bei Kindern und Jugendlichen: Stand der Forschung zu psychosozialen Schutzfaktoren für Gesundheit. Forschung und Praxis der Gesundheitsförderung: Vol. 35. Bundeszentrale für Gesundheitliche Aufklärung BZgA. http://​www​ .bzga​.de/​pdf​.php​?id​=​62​7a25580459​b7f8ce4095​cea966dca9 Burki, T. (2021). European Commission classifies obesity as a chronic disease. The Lancet Diabetes & Endocrinology, 9(7), 418. https://​doi​.org/​10​.1016/​S2213​-8587(21)00145​-5 Chzhen, Y., Moor, I., Pickett, W., Toczydlowska, E., & Stevens, G. W. J. M. (2018). International trends in ‘bottom-end’ inequality in adolescent physical activity and nutrition: HBSC study 2002–2014. European Journal of Public Health, 28(4), 624–630. https://​doi​.org/​10​.1093/​eurpub/​ckx237 Clark, A. M., DesMeules, M., Luo, W., Duncan, A. S., & Wielgosz, A. (2009). Socioeconomic status and cardiovascular disease: Risks and implications for care. Nature Reviews: Cardiology, 6(11), 712–722. https://​doi​.org/​10​.1038/​nrcardio​.2009​.163 Corder, K., Winpenny, E., Love, R., Brown, H. E., White, M., & van Sluijs, E. (2019). Change in physical activity from adolescence to early adulthood: A systematic review and meta-analysis of longitudinal cohort studies. British Journal of Sports Medicine, 53(8), 496–503. https://​doi​.org/​10​ .1136/​bjsports​-2016​-097330 Corsten, M. (2020). Lebenslauf und Sozialisation – ein Paar Schuhe? In M. Corsten (Ed.), Lebenslauf und Sozialisation (pp.  109–129). Springer Fachmedien Wiesbaden. https://​doi​.org/​10​.1007/​978​-3​-658​ -30397​-6​_5 Cosma, A., Stevens, G., Martin, G., Duinhof, E. L., Walsh, S. D., Garcia-Moya, I., Költő, A., Gobina, I., Canale, N., Catunda, C., Inchley, J., & Looze, M. de (2020). Cross-national time trends in adolescent mental well-being from 2002 to 2018 and the explanatory role of schoolwork pressure. The Journal of Adolescent Health: Official Publication of the Society for Adolescent Medicine, 66(6S), S50–S58. https://​doi​.org/​10​.1016/​j​.jadohealth​.2020​.02​.010 Crystal, S., Shea, D. G., & Reyes, A. M. (2017). Cumulative advantage, cumulative disadvantage, and evolving patterns of late-life inequality. The Gerontologist, 57(5), 910–920. https://​doi​.org/​10​.1093/​ geront/​gnw056 Denton, M., Prus, S., & Walters, V. (2004). Gender differences in health: A Canadian study of the psychosocial, structural and behavioural determinants of health. Social Science & Medicine, 58(12), 2585–2600. https://​doi​.org/​10​.1016/​j​.socscimed​.2003​.09​.008

Health inequalities in adolescence and their consequences for (emerging) adulthood  155 DiPrete, T. A., & Eirich, G. M. (2006). Cumulative advantage as a mechanism for inequality: A review of theoretical and empirical developments. Annual Review of Sociology, 32(1), 271–297. https://​doi​ .org/​10​.1146/​annurev​.soc​.32​.061604​.123127 Due, P., Krølner, R., Rasmussen, M., Andersen, A., Trab Damsgaard, M., Graham, H., & Holstein, B. E. (2011). Pathways and mechanisms in adolescence contribute to adult health inequalities. Scandinavian Journal of Public Health, 39(6_suppl), 62–78. https://​doi​.org/​10​.1177/​1403494810395989 Eccles, J. S., & Roeser, R. W. (2011). Schools as developmental contexts during adolescence. Journal of Research on Adolescence, 21(1), 225–241. https://​doi​.org/​10​.1111/​j​.1532​-7795​.2010​.00725​.x Elgar, F. J., McKinnon, B., Torsheim, T., Schnohr, C. W., Mazur, J., Cavallo, F., & Currie, C. (2016). Patterns of socioeconomic inequality in adolescent health differ according to the measure of socioeconomic position. Social Indicators Research, 127(3), 1169–1180. https://​doi​.org/​10​.1007/​s11205​ -015​-0994​-6 Elgar, F. J., Pförtner, T.-K., Moor, I., Clercq, B. de, Stevens, G. W. J. M., & Currie, C. (2015). Socioeconomic inequalities in adolescent health 2002–2010: A time-series analysis of 34 countries participating in the Health Behaviour in School-aged Children study. The Lancet, 385(9982), 2088–2095. https://​doi​.org/​10​.1016/​S0140​-6736(14)61460​-4 Frederick, C. B., Snellman, K., & Putnam, R. D. (2014). Increasing socioeconomic disparities in adolescent obesity. Proceedings of the National Academy of Sciences, 111(4), 1338–1342. https://​doi​.org/​ 10​.1073/​pnas​.1321355110 Fröhlich-Gildhoff, K., & Rönnau-Böse, M. (2021). Resilienz und Resilienzförderung im Jugendalter (Adoleszenz). In K. Fröhlich-Gildhoff & M. Rönnau-Böse (Eds.), Studien zur Resilienzforschung. Menschen stärken (pp.  175–184). Springer Fachmedien Wiesbaden. https://​doi​.org/​10​.1007/​978​-3​ -658​-32259​-5​_8 Fulginiti, A., Rice, E., Hsu, H.-T., Rhoades, H., & Winetrobe, H. (2016). Risky integration. Crisis, 37(3), 184–193. https://​doi​.org/​10​.1027/​0227​-5910/​a000374 Galobardes, B., Smith, G. D., & Lynch, J. W. (2006). Systematic review of the influence of childhood socioeconomic circumstances on risk for cardiovascular disease in adulthood. Annals of Epidemiology, 16(2), 91–104. https://​doi​.org/​10​.1016/​j​.annepidem​.2005​.06​.053 Granström, F., Garvin, P., Molarius, A., & Kristenson, M. (2021). Distinguishing independent and shared effects of material/structural conditions and psychosocial resources on educational inequalities in self-rated health: Results from structural equation modelling. Public Health, 196, 10–17. https://​doi​ .org/​10​.1016/​j​.puhe​.2021​.04​.032 Herke, M., Moor, I., Winter, K., Hack, M, .Hoffmann, S., Spalleck, J., Hilger-Kolb, J., Herr, R., Pischke, C., Dragano, N., Novelli, A., & Richter, M. (2022). Role of contextual and compositional characteristics of schools for health inequalities in childhood and adolescence: A scoping review. BMJ Open. 2022; 12:e052925. https://​doi​.org/​10​.1136/​bmjopen​-2021​-052925 Hurrelmann, K., & Bauer, U. (2015). Einführung in die Sozialisationstheorie: Das Modell der produktiven Realitätsverarbeitung (11., vollst. überarb. Aufl.). Beltz-Studium. Beltz. http://​content​-select​ .com/​index​.php​?id​=​bib​_view​&​ean​=​9783407294333 Hurrelmann, K., & Quenzel, G. (2016). Lebensphase Jugend: Eine Einführung in die sozialwissenschaftliche Jugendforschung (13th edn). Beltz Juventa. Inchley, J., Currie, D., Budisavljevic, S., Torsheim, T., Jastad, A., Cosma, A., & editors. (2020). Spotlight on adolescent health and well-being. Findings from the 2017/2018 Health Behaviour in School-aged Children (HBSC) survey in Europe and Canada: International report. Volume 2. Key data. Copenhagen, WHO Regional Office for Europe. Inchley, J., Currie, D., Jewell, J., Breda, J., & Barnekow, V. (2017). Adolescent obesity and related behaviours: Trends and inequalities in the WHO European Region, 2002–2014. Observations from the Health Behaviour in School-aged Children (HBSC) WHO collaborative cross-national study. World Health Organization. Johnson, D., Dupuis, G., Piche, J., Clayborne, Z., & Colman, I. (2018). Adult mental health outcomes of adolescent depression: A systematic review. Depression and Anxiety, 35(8), 700–716. https://​doi​ .org/​10​.1002/​da​.22777 Johnson, M. K., Crosnoe, R., & Elder Jr, G. H. (2011). Insights on adolescence from a life course perspective. Journal of Research on Adolescence, 21(1), 273–280. https://​doi​.org/​10​.1111/​j​.1532​-7795​ .2010​.00728​.x

156  Handbook of health inequalities across the life course Joyce, H. D., & Early, T. J. (2014). The impact of school connectedness and teacher support on depressive symptoms in adolescents: A multilevel analysis. Children and Youth Services Review, 39, 101–107. https://​doi​.org/​10​.1016/​j​.childyouth​.2014​.02​.005 Kalman, M., Inchley, J., Sigmundova, D., Iannotti, R. J., Tynjälä, J. A., Hamrik, Z., Haug, E., & Bucksch, J. (2015). Secular trends in moderate-to-vigorous physical activity in 32 countries from 2002 to 2010: A cross-national perspective. European Journal of Public Health, 25 Suppl 2, 37–40. https://​ doi​.org/​10​.1093/​eurpub/​ckv024 Katikireddi, S. V., Skivington, K., Leyland, A. H., Hunt, K., & Mercer, S. W. (2017). The contribution of risk factors to socioeconomic inequalities in multimorbidity across the lifecourse: A longitudinal analysis of the Twenty-07 cohort. BMC Medicine, 15(1). https://​doi​.org/​10​.1186/​s12916​-017​-0913​-6 Khanolkar, A. R., Chaturvedi, N., Kuan, V., Davis, D., Hughes, A., Richards, M., Bann, D., & Patalay, P. (2021). Socioeconomic inequalities in prevalence and development of multimorbidity across adulthood: A longitudinal analysis of the MRC 1946 National Survey of Health and Development in the UK. PLOS Medicine, 18(9), e1003775. https://​doi​.org/​10​.1371/​journal​.pmed​.1003775 Kim, H. H.-S., & Chun, J. (2018). Analyzing multilevel factors underlying adolescent smoking behaviors: The roles of friendship network, family relations, and school environment. Journal of School Health, 88(6), 434–443. https://​doi​.org/​10​.1111/​josh​.12630 King, V. (Ed.). (2013). Adoleszenzforschung Ser: v.1. Die Entstehung des Neuen in der Adoleszenz: Individuation, Generativität und Geschlecht in modernisierten Gesellschaften (2. Auflage). Springer VS. https://​ebookcentral​.proquest​.com/​lib/​kxp/​detail​.action​?docID​=​1592391 Klocke, A., Stadtmüller, S., & Giersiefen, A. (2020). Lebensverlaufsperspektive und soziale Ungleichheit. In P. Kriwy & M. Jungbauer-Gans (Eds.), Handbuch Gesundheitssoziologie (pp. 397–418). Springer Fachmedien Wiesbaden. https://​doi​.org/​10​.1007/​978​-3​-658​-06392​-4​_22 Kondo, N., Kawachi, I., Subramanian, S. V., Takeda, Y., & Yamagata, Z. (2008). Do social comparisons explain the association between income inequality and health? Relative deprivation and perceived health among male and female Japanese individuals. Social Science & Medicine, 67(6), 982–987. https://​doi​.org/​10​.1016/​j​.socscimed​.2008​.06​.002 Kuhn, M., & King, V. (2021). Adoleszenz. In U. Deinet, B. Sturzenhecker, L. von Schwanenflügel, & M. Schwerthelm (Eds.), Handbuch Offene Kinder- und Jugendarbeit (pp. 1303–1313). Springer Fachmedien Wiesbaden. https://​doi​.org/​10​.1007/​978​-3​-658​-22563​-6​_99 Laaksonen, M., Roos, E., Rahkonen, O., Martikainen, P., & Lahelma, E. (2005). Influence of material and behavioural factors on occupational class differences in health. Journal of Epidemiology and Community Health, 59(2), 163–169. https://​doi​.org/​10​.1136/​jech​.2003​.019323 Lampert, T. (2010). Frühe Weichenstellung: Zur Bedeutung der Kindheit und Jugend für die Gesundheit im späteren Leben [Fruhe Weichenstellung: Zur Bedeutung der Kindheit und Jugend fur die Gesundheit im spateren Leben]. Bundesgesundheitsblatt-Gesundheitsforschung-Gesundheitsschutz, 53(5), 486–497. https://​doi​.org/​10​.1007/​s00103​-010​-1055​-6 Laursen, B., Bukowski, W. M., Aunola, K., & Nurmi, J.-E. (2007). Friendship moderates prospective associations between social isolation and adjustment problems in young children. Child Development, 78(4), 1395–1404. https://​doi​.org/​10​.1111/​j​.1467​-8624​.2007​.01072​.x Link, B. G., Susser, E. S., Factor-Litvak, P., March, D., Kezios, K. L., Lovasi, G. S., Rundle, A. G., Suglia, S. F., Fader, K. M., Andrews, H. F., Johnson, E., Cirillo, P. M., & Cohn, B. A. (2017). Disparities in self-rated health across generations and through the life course. Social Science & Medicine (1982), 174, 17–25. https://​doi​.org/​10​.1016/​j​.socscimed​.2016​.11​.035 Lorant, V., Rojas, V. S., Robert, P.-O., Kinnunen, J. M., Kuipers, M. A. G., Moor, I., Roscillo, G., Alves, J., Rimpelä, A., Federico, B., Richter, M., Perelman, J., & Kunst, A. E. (2017). Social network and inequalities in smoking amongst school-aged adolescents in six European countries. International Journal of Public Health, 62(1), 53–62. https://​doi​.org/​10​.1007/​s00038​-016​-0830​-z Markkanen, I., Välimaa, R., & Kannas, L. (2019). Associations between students’ perceptions of the psychosocial school environment and indicators of subjective health in Finnish comprehensive schools. Children & Society, 48(1), 168. https://​doi​.org/​10​.1111/​chso​.12334 Marmot, M., & Wilkinson, R. G. (2001). Psychosocial and material pathways in the relation between income and health: A response to Lynch et al. BMJ, 322(7296), 1233–1236. https://​doi​.org/​10​.1136/​ bmj​.322​.7296​.1233

Health inequalities in adolescence and their consequences for (emerging) adulthood  157 Moor, I., Lampert, T., Rathmann, K., Kuntz, B., Kolip, P., Spallek, J., & Richter, M. (2014a). Explaining educational inequalities in adolescent life satisfaction: Do health behaviour and gender matter? International Journal of Public Health, 59(2), 309–317. https://​doi​.org/​10​.1007/​s00038​-013​-0531​-9 Moor, I., Rathmann, K., Stronks, K., Levin, K., Spallek, J., & Richter, M. (2014b). Psychosocial and behavioural factors in the explanation of socioeconomic inequalities in adolescent health: A multilevel analysis in 28 European and North American countries. Journal of Epidemiology and Community Health, 68(10), 912–921. https://​doi​.org/​10​.1136/​jech​-2014​-203933 Moor, I., Richter, M., Ravens-Sieberer, U., Ottová-Jordan, V., Elgar, F. J., & Pförtner, T. K. (2015). Trends in social inequalities in adolescent health complaints from 1994 to 2010 in Europe, North America and Israel: The HBSC study. The European Journal of Public Health, 25(suppl_2), 57–60. https://​doi​.org/​10​.1093/​eurpub/​ckv028 Moor, I., Spallek, J., & Richter, M. (2017). Explaining socioeconomic inequalities in self-rated health: a systematic review of the relative contribution of material, psychosocial and behavioural factors. Journal of Epidemiology and Community Health, 71(6), 565–575. https://​doi​.org/​10​.1136/​jech​-2016​ -207589 Moor, I., Kuipers, M. A. G., Lorant, V., Pförtner, T.-K., Kinnunen, J. M., Rathmann, K., Perelman, J., Alves, J., Robert, P.-O., Rimpelä, A., Kunst, A. E., & Richter, M. (2019). Inequalities in adolescent self-rated health and smoking in Europe: Comparing different indicators of socioeconomic status. Journal of Epidemiology and Community Health, 73(10), 963–970. https://​doi​.org/​10​.1136/​jech​-2018​ -211794 Moor, I., Hoffmann, L., Mlinarić, M., & Richter, M. (2020a). Soziale Netzwerke, Gesundheit und gesundheitliche Ungleichheiten im Jugendalter. In A. Klärner, M. Gamper, S. Keim-Klärner, I. Moor, H. von der Lippe, & N. Vonneilich (Eds.), Soziale Netzwerke und gesundheitliche Ungleichheiten (pp. 163–192). Springer Fachmedien Wiesbaden. Moor, I., Winter, K., Rahtmann, K., Ravens-Sieberer, U., Richter, M. (2020b). Alcohol, tobacco and cannabis use in adolescence – Cross-sectional results of the 2017/18 HBSC study. Journal of Health Monitoring, 5(3). DOI 10.25646/6903 Newton, S., Braithwaite, D., & Akinyemiju, T. F. (2017). Socio-economic status over the life course and obesity: Systematic review and meta-analysis. PLoS ONE, 12(5), e0177151. https://​doi​.org/​10​.1371/​ journal​.pone​.0177151 Pampel, F. C., Mollborn, S., & Lawrence, E. M. (2014). Life course transitions in early adulthood and SES disparities in tobacco use. Social Science Research, 43, 45–59. https://​doi​.org/​10​.1016/​j​ .ssresearch​.2013​.08​.005 Patton, G. C., Sawyer, S. M., Santelli, J. S., Ross, D. A., Afifi, R., Allen, N. B., Arora, M., Azzopardi, P., Baldwin, W., Bonell, C., Kakuma, R., Kennedy, E., Mahon, J., McGovern, T., Mokdad, A. H., Patel, V., Petroni, S., Reavley, N., Taiwo, K., … Viner, R. M. (2016). Our future: A Lancet commission on adolescent health and wellbeing. The Lancet, 387(10036), 2423–2478. https://​doi​.org/​10​.1016/​S0140​ -6736(16)00579​-1 Pearce, A., Mason, K., Fleming, K., Taylor-Robinson, D., & Whitehead, M. (2020). Reducing inequities in health across the life-course: Early years, childhood and adolescence. Copenhagen. https://​www​ .euro​.who​.int/​en/​health​-topics/​health​-determinants/​social​-determinants/​publications/​2020/​reducing​ -inequities​-in​-health​-across​-the​-life​-course​.​-early​-years​,​-childhood​-and​-adolescence​-2020 Petrovic, D., Mestral, C. de, Bochud, M., Bartley, M., Kivimäki, M., Vineis, P., Mackenbach, J., & Stringhini, S. (2018). The contribution of health behaviors to socioeconomic inequalities in health: A systematic review. Preventive Medicine, 113, 15–31. https://​doi​.org/​10​.1016/​j​.ypmed​.2018​.05​.003 Pförtner, T.-K., & Moor, I. (2017). Wie kommt die Gesellschaft unter die Haut? Eine Mediatoranalyse gesundheitlicher Ungleichheit mit den Daten des Sozioökonomischen Panels 2011 [How does the social get under the skin? A mediator analysis with data from the German Socioeconomic Panel 2011]. Psychotherapie, Psychosomatik, medizinische Psychologie, 67(1), 9–18. https://​doi​.org/​10​.1055/​s​ -0042​-116153 Pförtner, T.-K., Moor, I., Rathmann, K., Hublet, A., Molcho, M., Kunst, A. E., & Richter, M. (2015). The association between family affluence and smoking among 15-year-old adolescents in 33 European countries, Israel and Canada: The role of national wealth. Addiction, 110(1), 162–173. https://​doi​.org/​ 10​.1111/​add​.12741

158  Handbook of health inequalities across the life course Picoito, J., Santos, C., Loureiro, I., Aguiar, P., & Nunes, C. (2019). Gender-specific substance use patterns and associations with individual, family, peer, and school factors in 15-year-old Portuguese adolescents: A latent class regression analysis. Child and Adolescent Psychiatry and Mental Health, 13, 21. https://​doi​.org/​10​.1186/​s13034​-019​-0281​-4 Poulain, T., Vogel, M., Sobek, C., Hilbert, A., Körner, A., & Kiess, W. (2019). Associations between socio-economic status and child health: Findings of a large German cohort study. International Journal of Environmental Research and Public Health, 16(5). https://​doi​.org/​10​.3390/​ijerph16050677 Prus, S. G. (2007). Age, SES, and health: A population level analysis of health inequalities over the lifecourse. Sociology of Health & Illness, 29(2), 275–296. https://​doi​.org/​10​.1111/​j​.1467​-9566​.2007​ .00547​.x Raphael, D. (2013). Adolescence as a gateway to adult health outcomes. Maturitas, 75(2), 137–141. https://​doi​.org/​10​.1016/​j​.maturitas​.2013​.03​.013 Reiss, F. (2013). Socioeconomic inequalities and mental health problems in children and adolescents: A systematic review. Social Science & Medicine (1982), 90, 24–31. https://​doi​.org/​10​.1016/​j​ .socscimed​.2013​.04​.026 Richter, M., & Dragano, N. (2018). Micro, macro, but what about meso? The institutional context of health inequalities. International Journal of Public Health, 63(2), 163–164. https://​doi​.org/​10​.1007/​ s00038​-017​-1064​-4 Richter, M., Moor, I., & van Lenthe, F. J. (2012). Explaining socioeconomic differences in adolescent self-rated health: The contribution of material, psychosocial and behavioural factors. Journal of Epidemiology and Community Health, 66(8), 691–697. https://​doi​.org/​10​.1136/​jech​.2010​.125500 Ringdal, R., Espnes, G. A., Eilertsen, M.-E. B., Bjørnsen, H. N., & Moksnes, U. K. (2020). Social support, bullying, school-related stress and mental health in adolescence. Nordic Psychology, 72(4), 313–330. https://​doi​.org/​10​.1080/​19012276​.2019​.1710240 Sawyer, S. M., Afifi, R. A., Bearinger, L. H., Blakemore, S.-J., Dick, B., Ezeh, A. C., & Patton, G. C. (2012). Adolescence: A foundation for future health. The Lancet, 379(9826), 1630–1640. https://​doi​ .org/​10​.1016/​S0140​-6736(12)60072​-5 Scambler, G. (2012). Health inequalities. Sociology of Health & Illness, 34(1), 130–146. https://​doi​.org/​ 10​.1111/​j​.1467​-9566​.2011​.01387​.x Sethi, D. (Ed.). (2013). European report on preventing child maltreatment. WHO Regional Office for Europe. Simetin, I. P., Kuzman, M., Franelic, I. P., Pristas, I., Benjak, T., & Dezeljin, J. D. (2011). Inequalities in Croatian pupils’ unhealthy behaviours and health outcomes: Role of school, peers and family affluence. European Journal of Public Health, 21(1), 122–128. https://​doi​.org/​10​.1093/​eurpub/​ckq002 Simmonds, M., Llewellyn, A., Owen, C. G., & Woolacott, N. (2016). Predicting adult obesity from childhood obesity: A systematic review and meta-analysis. Obesity Reviews, 17(2), 95–107. https://​ doi​.org/​10​.1111/​obr​.12334 Stalsberg, R., & Pedersen, A. V. (2010). Effects of socioeconomic status on the physical activity in adolescents: A systematic review of the evidence. Scandinavian Journal of Medicine & Science in Sports, 20(3), 368–383. https://​doi​.org/​10​.1111/​j​.1600​-0838​.2009​.01047​.x Stone, A. L., Becker, L. G., Huber, A. M., & Catalano, R. F. (2012). Review of risk and protective factors of substance use and problem use in emerging adulthood. Addictive Behaviors, 37(7), 747–775. https://​doi​.org/​10​.1016/​j​.addbeh​.2012​.02​.014 Stronks, K., van de Mheen, H. D., Looman, C. W. N., & Mackenbach, J. P. (1996). Behavioural and structural factors in the explanation of socio-economic inequalities in health: An empirical analysis. Sociology of Health & Illness, 18(5), 653–674. https://​doi​.org/​10​.1111/​1467​-9566​.ep10934524 Thomson, K., Bambra, C., McNamara, C., Huijts, T., & Todd, A. (2016). The effects of public health policies on population health and health inequalities in European welfare states: Protocol for an umbrella review. Systematic Reviews, 5, 57. https://​doi​.org/​10​.1186/​s13643​-016​-0235​-3 Torsheim, T., Nygren, J. M., Rasmussen, M., Arnarsson, A. M., Bendtsen, P., Schnohr, C. W., Nielsen, L., & Nyholm, M. (2018). Social inequalities in self-rated health: A comparative cross-national study among 32,560 Nordic adolescents. Scandinavian Journal of Public Health, 46(1), 150–156. https://​ doi​.org/​10​.1177/​1403494817734733

Health inequalities in adolescence and their consequences for (emerging) adulthood  159 Vaičiūnas, T., & Šmigelskas, K. (2019). The role of school-related well-being for adolescent subjective health complaints. International Journal of Environmental Research and Public Health, 16(9). https://​ doi​.org/​10​.3390/​ijerph16091577 van Oort, F. V. A., van Lenthe, F. J., & Mackenbach, J. P. (2005). Material, psychosocial, and behavioural factors in the explanation of educational inequalities in mortality in The Netherlands. Journal of Epidemiology and Community Health, 59(3), 214–220. https://​doi​.org/​10​.1136/​jech​.2003​.016493 Viner, R. M., Ozer, E. M., Denny, S., Marmot, M., Resnick, M., Fatusi, A., & Currie, C. (2012). Adolescence and the social determinants of health. The Lancet, 379(9826), 1641–1652. https://​doi​ .org/​10​.1016/​S0140​-6736(12)60149​-4 Vogel, M., Rees, C. E., McCuddy, T., & Carson, D. C. (2015). The highs that bind: School context, social status and marijuana use. Journal of Youth and Adolescence, 44(5), 1153–1164. https://​doi​.org/​ 10​.1007/​s10964​-015​-0254​-8 Walls, H., Baker, P., & Parkhurst, J. (2018). Addressing trade policy as a macro-structural determinant of health: The role of institutions and ideas. Global Social Policy, 18(1), 94–101. https://​doi​.org/​10​ .1177/​1468018117748700 Weber Ku, E. B., Hagler, M. A., Parnes, M. F., Schwartz, S. E. O., Rhodes, J. E., & Erickson, L. D. (2020). Natural mentoring relationships among survivors of caregiver childhood abuse: Findings from the Add Health Study. Annals of the New York Academy of Sciences. Advance online publication. https://​doi​.org/​10​.1111/​nyas​.14313 Weinberg, D., Stevens, G. W. J. M., Duinhof, E. L., & Finkenauer, C. (2019). Adolescent socioeconomic status and mental health inequalities in the Netherlands, 2001–2017. International Journal of Environmental Research and Public Health, 16(19). https://​doi​.org/​10​.3390/​ijerph16193605 West, P. (1997). Health inequalities in the early years: Is there equalisation in youth? Social Science & Medicine, 44(6), 833–858. https://​doi​.org/​10​.1016/​s0277​-9536(96)00188​-8 Wickrama, K. A. S., Noh, S., & Elder, G. H. (2009). An investigation of family SES-based inequalities in depressive symptoms from early adolescence to emerging adulthood. Advances in Life Course Research, 14(3). https://​doi​.org/​10​.1016/​j​.alcr​.2010​.04​.001 Winpenny, E. M., van Sluijs, E. M. F., White, M., Klepp, K.-I., Wold, B., & Lien, N. (2018). Changes in diet through adolescence and early adulthood: Longitudinal trajectories and association with key life transitions. International Journal of Behavioral Nutrition and Physical Activity, 15(1), 86. https://​doi​ .org/​10​.1186/​s12966​-018​-0719​-8

11. Social inequalities, social capital, and health inequalities in the process of growing up Andreas Klocke and Sven Stadtmüller

INTRODUCTION Health inequalities are frequently discussed, be it social inequalities that trigger health inequalities (causation), or, vice versa, health impairments that result in lower social positions (selection). With regard to young people, the selection thesis is supported when bullied pupils report mental disorders that, in turn, result very often in lower school performance (Klocke, Clair and Bradshaw, 2014). The same is true for differently abled young people, who are more prone to bullying, which results in problems including lower self-esteem, sleep difficulties, and even a lower school performance (Mundy et al., 2017). However, most studies conclude that social economic position (SEP) determines health (Kröger, Pakpahan and Hoffmann, 2015; Hoffmann, Kröger, Pakpahan, 2018), and this might be true for young people as their living conditions are determined by their parents’ SEP rather than their own personal social, physical, or mental condition. Thus, we start with the family background and assess whether family SEP affects the health of young people. Furthermore, the concept of social capital is introduced as a mediator that focuses on disposable networks and ties to identify its joint effects with family SEP on health inequalities in youth. Finally, we analyze whether the effects of family SEP and social capital on health change in the process of growing up by examining longitudinal survey data from German adolescents.

THEORETICAL FRAMEWORKS Adolescence Analyzing the life course effects in adolescence is a challenge, as adolescence represents a comparatively short period in the entire life course, and encompasses some analytical difficulties. The period of adolescence has expanded significantly in the last 70 years after World War II from five to 15 years today (Giddens and Sutton, 2021). A life course perspective can be taken more confidently over the entire life span of a person, which is about 80 years today, as opposed to one taken over the adolescent period. If observed exclusively, the youth phase in adolescence must first be delimited. The youth phase of life has, on the one hand, moved forward biographically, and, on the other hand, extended due to longer training periods. It now spans from about 12 years (the beginning of puberty) to the end of the third decade (the beginning of parenting), and is therefore quite heterogeneous from a developmental psychological and sociocultural point of view. Overall, adolescence is an inconsistent phase of life, where enormous differentiation in terms of lifestyles, youth-culture, and values is observed (see also 160

Social inequalities, social capital, and health inequalities in the process of growing up  161 Bernard et al., Chapter 10 in this book). Furthermore, frequent cultural differentiations based on regional origin, ethnic affiliation, or gender are hardly surprising. Growing Up and Social Inequality A life course perspective on social inequality and health in adolescence can be read differently. The first question that arises here is whether social inequality has changed during adolescence. Simply put, this question asks whether the macrostructures of society have undergone a change and whether, for example, an increase or decrease in general social inequality has been observed (period effects). In line with this, one can question whether these changes cause immediate and simultaneous consequences for the younger generation, whether these consequences affect the entire youth or only groups of young people, and whether different effects of social inequality on the group of early and late youth can be observed (cohort or age effects). On the other hand, one can question whether the effects of (constant) social inequalities on the health and health behavior of young people change over time or with age (age effects), e.g., younger adolescents (12–18 years) suffer more from social inequality than older adolescents (19–30 years) or vice versa. Although research on social inequality is by no means clear on this point, it is well documented that social inequality in living conditions affects the health and health behavior of young people in various ways (Inchley et al., 2020). This chapter aims to investigate the extent to which the effects of family SEP on health persist, weaken, or intensify over the course of adolescence. Child Health A central developmental task in adolescence is to build up a personal and social identity for which the comparison with the peer group plays an extremely important role. Therefore, for the age group of children and adolescents, it can be assumed that social action is highly socially determinable. Social behavior often follows clear group norms or represents coping strategies for age-specific developmental tasks (Havighurst, 1972). These group norms or coping patterns are not the same for all young people, but are socially segmented by the educational background of their parents’ home. Likewise, young people of different social origins fall back on the resources that are available to them. Health in childhood is to a large extent determined by emotions of fitness and well-being (Klocke, Clair and Bradshaw, 2014; Inchley et al., 2020), which, in turn, reflect social contexts. In the last decades much attention has been devoted to the theoretical elaboration of the relationship between living conditions and health behavior. If the state of health is understood as a balance or an equilibrium between illness and health, the more recent sociology of health indicates that health is to be understood as a process of permanent adaptation to external requirements. The health situation of children and adolescents is then very often interpreted with reference to the stress-coping paradigm of stress theory (Lazarus and Folkmann, 1987; Lazarus, 1991; Jessor, 1998), according to which stressful living conditions lead to specific coping reactions. The resources to cope with the situation can arise from personal abilities and talents, as well as from external resources such as social support or material wealth. The chances of health-compatible coping naturally increase with the available resources, and are therefore socially segmented (Furstenberg et al., 1999). The higher the level of cognitive and social skills, the more likely it is that long-term and health-compliant coping actions can be

162  Handbook of health inequalities across the life course achieved. Besides, recent studies suggest that, apart from extreme events, it is not the stresses that affect human health, but rather the available resources that determine the health consequences of the stresses. Accordingly, the health of adolescents is to be primarily understood through the individually available social capital. Social Capital The distinguishing feature of social capital is its focus on the relationships among individuals (Lin, 2001). Unlike human capital, which focuses on individuals’ abilities, and economic capital, which measures possession, social capital addresses the networks and ties that the individuals are involved in. Being a member of a network is advantageous, as one gains information, support, access, and trust. And this is true independent from personal characteristics, which might make it easier for some people to access the existing network resources. Through these aspects, social capital can improve one’s life satisfaction and well-being. As Lin (2001) pointed out, the concept of social capital starts on the micro-level, and can be extended to the meso- and macro-level (see also Halpern, 2005). The individualistic approach can be captured with Bourdieu’s conception of social capital, where social capital is simply another form of capital in addition to economic and cultural capital (Bourdieu, 1986). Thus, it is embedded in the structure of social inequality, and can moderate (deteriorate or improve) an individual’s position in the societal structure of social inequality. It is a multiplier; when applied positively, it functions as a support network. One’s economic resources or cultural competencies can have a greater effect on life if the “lever” of social capital (connectedness, support) can be applied. As Morrow argued, “the basic argument, then, is that the extent to which people are embedded within their family relationships, social networks, and communities, and their sense of belonging and civic identity, constitutes ‘social capital’. This stock of ‘social capital’, in turn, has an impact on health and well-being” (Morrow, 1999, p. 768). For children and young adolescents in particular, social capital can be described as a trust-based network, which can be accessed when social support is needed. This thought relates to the work of Coleman (1990), who defined social capital as something that is embodied in the relations among persons, and stated that “a group whose members manifest trustworthiness and place extensive trust in one another will be able to accomplish much more than a comparable group lacking that trustworthiness and trust” (Coleman, 1990, p. 321).1

HYPOTHESES Several health impairments in adolescence are socially segmented. Since adolescence represents a phase of life in which serious physical, emotional, and social changes take place, the effects of social living conditions on health cannot simply be separated from the coping patterns that arise from processing individual developmental tasks (Havighurst, 1972). In principle, an interlinking can be assumed between the external influences, which includes social environment, and internal developmental requirements, for which the young people are seeking an answer. The effects of family SEP and social capital on the health of adolescents is studied for a broader understanding of social inequalities, and to acknowledge that young people rely not only on material resources, but also on trust-based ties and networks when dealing with

Social inequalities, social capital, and health inequalities in the process of growing up  163 stressful tasks and events during adolescence. These resources determine how young people cope with those stressors, which, in turn, influences their health. In line with previous research, family SEP and social capital is expected to positively affect the health of young people (H1, H2). Moreover, since social capital is supposed to work as a mediator that buffers the effects of unfavorable living conditions, an interaction between social capital and family SEP is expected. More precisely, social capital is expected to compensate for the negative consequences of a low family SEP on health (H3). If this holds true, social capital can be considered a protective factor in the health development of young people. Finally, we question how the effects of family SEP on health evolve during adolescence. As we have no clear expectations from literature, this is treated as an open research question (RQ1). Data We rely on data from the German study Health Behavior and Injuries in School Age (GUS) to analyze the impact of family SEP and social capital on health in the process of growing up. The GUS study included 18,923 adolescents surveyed annually from 5th to 10th grade between the school years 2014/15 and 2019/20. The repeated measurement resulted in more than 52,000 observations from adolescents attending 173 secondary schools in 14 German federal states. For the classroom surveys, each participating student, who had a parental informed consent form, received a tablet device to answer their questionnaire during a (regular) school period (of 45 minutes). Trained interviewers were also present in the classroom. The main objective of the study was to elicit the antecedents of injuries in the school environment (Stadtmüller et al., 2021; Filser et al., 2022). However, the survey also enquired about physical and mental health, and included indicators to assess family affluence and social capital. Panel data allows us to isolate the age effects of family SEP on health-related variables while excluding cohort effects, as the same cohort is surveyed over time. Moreover, age effects are supposedly not impaired by period effects, as the macrostructures of society had hardly undergone fundamental changes in the comparatively short period of time between 2014 and 2020.2 However, our results cannot generalize for other birth cohorts, countries, or time periods. Measures The key dependent variable of our analysis is a health index that covers 16 health-related items, covering general and physical health, mental health, and health behavior. All items were measured on five-point scales and recoded to range from 0 to 4 with higher values indicating better health (see Table 11.1). The final index ranges from 0 to 64, and shows a high internal consistency with alpha values between 0.72 and 0.81 in the six survey waves. For all 47,752 observations with valid values, the mean value is 46.6, and the quartile boundaries are at 42 (25%), 48 (50%), and 52 (75%), respectively. For independent variables, we borrowed from the family affluence scale (FAS) proposed by Currie et al. (2012) to measure family SEP (see Table 11.2). To weight the four items equally, their scales were standardized and the four values were averaged for the final index ranging from 1 (lowest FAS) to 5 (highest FAS). For all 50,221

164  Handbook of health inequalities across the life course Table 11.1

The health index

Health domain

Item

Physical health

General health status (self-assessment) During the last week: Suffered from headaches? During the last week: Suffered from back pains?

Mental health

During the last week: Felt fit and comfortable? During the last week: Full of energy? During the last week: Irritated and in a bad mood? During the last week: Sleep difficulties? During the last week: Felt sad? During the last week: Felt lonely? During the last week: Had problems concentrating? During the last week: Felt unhappy and depressed?

Health behavior

How often during the week: Drinking coke or lemonade? How often during the week: Eating vegetables or salad? How often during the week: Eating fruits? How often during the week: Eating sweets? How often during the week: Had fast food?

Table 11.2

The family affluence scale (FAS)

Item

Responses

During the past 12 months, how many times did you go away

Not at all

on holiday with your family?

Once Twice More than twice

Does your family have a car?

No Yes, one Yes, two or more

Do you have your own bedroom?

No Yes

How many books does your family have at home?

None or only a few (0–10 books) About a book board (11–25 books) About a bookshelf (26–100 books) About two bookshelves (101–200 books) Three or more bookshelves (more than 200 books)

observations with valid values, the mean value is 4.0, and the quartile boundaries are at 3.6 (25%), 4.2 (50%), and 4.5 (75%), respectively. To measure social capital, the index proposed by Klocke and Stadtmüller (2019) was employed, which covers the students’ assessment of the quality of their relationship with their parents and classmates, and the quality of their neighborhood (see Table 11.3). The three domains of trust were standardized to range from 1 (lowest quality) to 5 (highest quality) and averaged for the final index (1 = low social capital, 5 = high social capital). For all 44,051 observations with valid values, the mean value was 3.9, and the quartile boundaries were at 3.4 (25%), 3.9 (50%), and 4.3 (75%), respectively. To analyze the effects of the FAS and social capital over time, the students’ age was calculated, referring to the date of the survey (stored by the survey application) and students’ month and year of birth, which were asked in each survey wave. The majority of the observations

Social inequalities, social capital, and health inequalities in the process of growing up  165 Table 11.3 Trust domain

The Social Capital Index Item

Responses

How easy is it to talk with your father/stepfather about things that really Parents

concern you? How easy is it to talk with your mother/stepmother about things that really

very difficult–very easy

concern you? Most students in my class like being together. School

Most students in my class are kind and helpful. Other students accept me as I am. People greet and speak with each other.

Neighborhood

Smaller children can play outside during the day. One can trust people.

do not agree at all–fully agree do not agree at all–fully agree

with valid values for the survey variables on the birthdate were aged between 10.8 years (5th percentile) and 16.1 years (95th percentile). Additionally, dummy variables were included for gender (1 = girls), region (1 = students attending schools in Eastern Germany), migration status (1 = at least one parent was not born in Germany), and school type (1 = students attending the Gymnasium, which is the highest school type in Germany). Findings Three multi-level regression models were estimated, including 38,722 observations from 15,322 students attending 173 secondary schools (see Table 11.4). The multi-level models account for the hierarchical structure of the survey data (observations are nested in students and students are nested in schools), as ignoring the clustering data would result in too optimistic standard errors (and tests for significance). Moreover, the multi-level estimation allows us to deduce the proportion of variance accounted for by the different levels. Accordingly, differences in health are only to a small extent (0.6%) linked to the school attended by the young people. Rather, the total variance of the health index score is almost equally divided between and within students (47% and 52%, respectively), suggesting a high intra-individual variance of health in youth. Moreover, the models reveal that girls show significantly lower levels of health (compared to boys). However, one must be aware that the difference is, on average, only one point on an index with 65 scale points. Most importantly, the models aim to uncover the effects of the FAS and social capital on the health index. As expected, both show significant and positive effects on the health index, thus supporting our first two hypotheses. However, as seen in Figure 11.1, the effect of social capital is more pronounced than family SEP. In the left half of the plot region, the predicted scores for the health index are compared for observations with a comparatively low (10th percentile) and high (90th percentile) FAS while holding all other variables at their observed values. Here, the difference in the health index is 1.3 scale points, with students who reported a higher FAS also scoring higher on the health index. For social capital (the right half), however, the respective difference is 8.3 scale points, and thus more than six times higher. Following this, we tested whether social capital could buffer inequalities in family SEP. If so, an interaction between the two variables is expected, more precisely, the effect of FAS on health is expected to decrease with higher levels of social capital (H3).

166  Handbook of health inequalities across the life course

Figure 11.1

Predicted values for the health index of students with low vs. high values of the FAS and social capital

Table 11.4

The effects of family SEP (FAS) and social capital on the health index Model 1 b

Model 2

se

p

b

Model 3

se

p

b

Fixed Effects Constant

34.97

.462

***

26.46

1.10

***

43.79

2.57

***

Sex (Girls)

–1.06

.099

***

–1.06

.097

***

–1.04

.099

***

Region (Eastern Germany)

.029

.181

.024

.181

.008

.180

–.019

.112

–.031

.112

–.029

.112

.170

.139

.193

.138

.199

.138

Age

–.792

.021

***

–.790

.021

***

–2.03

.161

***

FAS

.651

.061

***

2.82

.263

***

2.11

525

***

Social Capital

5.17

.062

***

7.46

.277

***

3.52

.532

***

–.583

.069

***

***

Migration background School type (Gymnasium)

FAS × Social Capital

–.525

.071

FAS × Age

.036

.030

Social Capital × Age

.282

.034

Variances of Random Effects Variance: Level 3 (School)

.308 (.079)

.307 (.079)

.302 (.078)

Variance: Level 2 (Student)

23.58 (.444)

23.57 (.443)

23.55 (.442) 26.76 (.249)

Variance: Level 1 (Observation)

26.89 (.251)

26.83 (.250)

Proportion of Level 3-Variance

0.6%

0.6%

0.6%

Proportion of Level 2-Variance

47.0%

47.1%

47.1%

Notes: All estimates are linear regression coefficients based on multi-level linear regression models with observations (level 1) nested in students (level 2) and students nested in schools (level 3); n(observations) = 38,722; n(students) = 15,322; n(schools) = 173. *p < 0.05; **p < 0.01; ***p < 0.001.

***

Social inequalities, social capital, and health inequalities in the process of growing up  167

Figure 11.2

Predicted values for the health index of the interaction between FAS and social capital

As Figure 11.2 suggests, the effect of family affluence on health is most pronounced when the amount of social capital is low (p10, see the point estimates on the left). However, for relatively high levels of social capital (p90), no significant effects of the FAS were observed on the health index, thus supporting hypothesis 3. Besides, the interaction effect is statistically significant as revealed in Table 11.4 (Model 2). Finally, the effects of FAS and social capital on health during adolescence was studied. For this, additional interaction terms for FAS and social capital with age were included in the model to estimate whether the effects of social inequalities on health change as students grow older.3 The results are displayed in Table 11.4 (Model 3) and Figures 11.3 and 11.4. Figures 11.3 and 11.4 reflect the main effects of FAS and social capital on health, as higher values on both variables coincide with higher scores on the health index. Similarly, the larger effect of social capital is obvious, as differences in the health index between the three groups are much more pronounced for social capital than for the FAS. Moreover, both figures indicate a general decline in health with increasing age, irrespective of the values for FAS and social capital. However, both variables differ remarkably with respect to their effect on health with increasing age: while the effect of FAS on health hardly changes during adolescence (as the lines almost run parallel to each other for any age group), Figure 11.4 suggests an increasing effect of social capital on health in the process of growing up. This is evident when the differences in the health index scores between the three groups of social capital are compared for students aged 11 and 16, respectively. Moreover, the positive and significant interaction effect of social capital and age in Table 11.4 (Model 3) confirms this finding, while the effect for FAS × age is also positive, but much weaker and not statistically significant.

168  Handbook of health inequalities across the life course

Figure 11.3

Predicted values for the health index of students with different ages and varying FAS

Figure 11.4

Predicted values for the health index of students with different ages and varying social capital

Social inequalities, social capital, and health inequalities in the process of growing up  169

DISCUSSION This chapter’s main question was whether the effect of social inequalities on health changes in the process of growing up. Family SEP and social capital were emphasized as two similar, yet different concepts that reflect social inequalities in youth. Based on the data from a large-scale panel survey in Germany with adolescents aged 11 to 16, the findings from previous studies were replicated, which suggested that both family SEP and social capital significantly influenced the health status of young people (Klocke and Stadtmüller, 2019; Inchley et al., 2020). Moreover, our results indicate that social capital acts as a buffer in unfavorable conditions, and is, thus, able to compensate for a low family SEP. In line with this, the analysis revealed no significant differences in health between students with a very low and very high family affluence when they shared a high volume of social capital. The importance of social capital in the health development of young people is finally highlighted by the results from our analysis on the effects of FAS and social capital over time. While the (weak) effect of the FAS on health remains constant, the effect of social capital increases as adolescents get older. In essence, social capital gains power over time, and the developments examined are statistically significant. This finding is crucial, as there are hints for a causal effect of social capital on health based on fixed effects regression models for panel data of the same age cohort (Klocke and Stadtmüller, 2019). If young people are considered to enter puberty today at ages 12 and 13, our observation period between the ages of 11 and 16 comprises the passage from childhood to youth, and monitors the early years of adolescence very well. We believe that young people become more sensitive about their integration into social networks as they get older. Hence, they respond with a more differentiated rating of health, depending on the resources they can rely on from their trust-based networks. The reported health, in turn, reflects their abilities to cope with their developmental tasks during adolescence. This might explain the strong effect of social capital, and its increase in the early years of adolescence. An open question still is whether this continues when young people become young adults. Our results of a “widening gap” in health during adolescence contradicts the observations of Patrick West (1997) and West and Sweeting (2004), who concluded that there is a social “equalization” in youth with regard to health inequalities. Their main argument is that the school and youth culture have a levelling effect that results in an “equalisation in youth”. However, in our study, social capital has a growing effect on health in young people. Even for family SEP, the equalization thesis can not be verified, as differences do not vanish, but slightly increase over time, albeit not to a statistically significant extent. Whether this apparently different finding is a pure German phenomenon or a reflection of the period effects (1990s vs. 2010s) on the youth culture and societal prospects is uncertain. In general, health in childhood can be argued to be largely determined by the factors of fitness and well-being (Klocke, Clair and Bradshaw, 2014; Inchley et al., 2020), which, in turn, reflect social capital. The Covid-19 pandemic may have even increased the importance of social capital as a source of resilience for young people. If so, our results, based on data collected prior to the pandemic, might underestimate the observed impact of social capital on adolescent health. As we carry on with a short sample of our survey, further analysis will probably answer this question.

170  Handbook of health inequalities across the life course

NOTES 1. Coleman (1990, p. 321) also pointed out that unlike economic capital, the “use” of social capital strengthens and increases social capital and does not “consume” it. The more I trust (mutual) other people, the more I increase in my social capital. It, thereby, reduces not only control costs but also creates further social capital. 2. The last panel wave includes only half of the targeted sample as data collection had to be stopped in March 2020 due to the school closures in the course of the first Covid-19 lockdown in Germany. 3. We also tested for a three-way interaction of FAS, social capital and age. However, as there is neither a theoretical argument nor empirical support for such an interaction, we did not include it in our final model.

REFERENCES Bourdieu, P. (1986) ‘The forms of capital’ in Richardson, J. (ed.) Handbook of theory and research for the sociology of education. New York: Greenwood, pp. 241–258. Coleman, J.S. (1990) Foundations of social theory. Cambridge: Harvard University Press. Currie, C., Zanotti, C., Morgan, A., Currie, D., de Looze, M., Roberts, C., Samdal, O., Smith, O.R.F. and Barnekow, V. (2012) Social determinants of health and well-being among young people. Health behaviour in school-aged children (HBSC) study: International report from the 2009/2010 survey. Copenhagen: WHO Regional Office for Europe. Filser, A., Stadtmüller, S., Lipp, R. and Preetz, R. (2022) ‘Adolescent school injuries and classroom sex compositions in German secondary schools’, BMC Public Health, 22, p. 62. Available at: 10.1186/ s12889-021-12370-8. Furstenberg, F.F. Jr., Cook, T.D., Eccles, J., Elder, G.H. Jr. and Sameroff, A. (1999) Managing to make it. Urban families and adolescent success. Chicago: University of Chicago Press. Giddens, A. and Sutton, P. (2021) Sociology. Cambridge: Polity Press. Halpern, D. (2005) Social capital. Cambridge: Polity Press. Havighurst, R.J. (1972) Developmental tasks and education. New York: McKay. Hoffmann R., Kröger H. and Pakpahan E. (2018) ‘Pathways between socioeconomic status and health: Does health selection or social causation dominate in Europe?’, Advances in Life Course Research, 36, pp. 23–36. Inchley, J., Currie, D., Budisavljevic, S., Torsheim, T., Jåstad, A., Cosma A., Kelly, C. and Arnarsson, A.M. (2020) Spotlight on adolescent health and well-being. Findings from the 2017/2018 Health Behaviour in School-aged Children (HBSC) survey in Europe and Canada. International report. Volume 1 & 2. Key findings. Copenhagen: WHO Regional Office for Europe. Jessor, R. (1998) New perspectives on adolescent risk behavior. New York: Cambridge University Press. Klocke, A., Clair, A. and Bradshaw, J. (2014) ‘International variation in child subjective well-being’, Child Indicators Research, 7(1), pp. 1–20. Available at: 10.1007/s12187-013-9213-7. Klocke, A. and Stadtmüller, S. (2019) ‘Social capital in the health development of children’, Child Indicators Research, 12(4), pp. 1167–1185. Available at: 10.1007/s12187-018-9583-y. Kröger, H., Pakpahan, E. and Hoffmann, R. (2015) ‘What causes health inequality? A systematic review on the relative importance of social causation and health selection’, European Journal of Public Health, 25(6), pp. 951–960. Available at: 10.1093/eurpub/ckv111. Lazarus, R.S. (1991) Emotion & Adaptation. Oxford: Oxford University Press. Lazarus, R.S. and Folkman, S. (1987) ‘Transactional theory and research on emotions and coping’, European Journal of Personality, 1(3), pp. 141–169. Available at: 10.1002/per.2410010304. Lin, N. (2001) Social capital: A theory of social structure and action. Cambridge: Cambridge University Press. Morrow, V. (1999) ‘Conceptualising social capital in relation to the well-being of children and young people: A critical review’, The Sociological Review, 47(4), pp. 744–765. Available at: 10.1111/1467-954X.00194.

Social inequalities, social capital, and health inequalities in the process of growing up  171 Mundy, L.K., Canterford, L., Tucker, D., Bayer, J., Romaniuk, H., Sawyer, S., Lietz, P., Redmond, G., Proimos, J., Allen, N. and Patton, G. (2017) ‘Academic performance in primary school children with common emotional and behavioral problems’, Journal of School Health, 87(8), pp. 565–637. Available at: 10.1111/josh.12531. Stadtmüller, S., Klocke, A., Giersiefen, A., Lipp, R. and Wacker, C. (2021) ‘Approaching the causes of unintentional injuries in the school environment: A panel analysis of survey data from Germany’, Journal of School Health, 92(2), pp. 148–156. Available at: 10.1111/josh.13112. West, P. (1997) ‘Health inequalities in early years: Is there equalisation in youth?’, Social Science and Medicine, 44(6), pp. 833–858. Available at: 10.1016/s0277-9536(96)00188-8. West, P. and Sweeting, H. (2004) ‘Evidence on equalization in health in youth from the west of Scotland’, Social Science and Medicine, 59(1), pp. 13–27. Available at: 10.1016/j.socscimed.2003.12.004.

12. Work and health inequalities Johannes Siegrist

INTRODUCTION In economically advanced societies, paid work continues to play a crucial role in adult life. Participation in the labour market offers a continuous income and basic social protection; it confers social status and extends social relations beyond family life. Under favourable conditions, work and employment meet important human needs, such as the development of capabilities and skills, the promotion of personal growth and career advancement, and the recurrent experience of control, mastery, recognition, and self-esteem. Preparing individuals for successful job acquisition and promotion is a major goal of primary socialization, education and training. Moreover, employed or self-employed activities absorb the largest amount of time and energy during several decades of the adult lifespan, thus exposing people recurrently to distinct work environments with direct impact on their health and well-being. Finally, quality of work and employment trajectories extend their influence on the life during retirement, affecting living standards and psychosocial resources, as well as healthy life expectancy. Given the centrality of work in a life course perspective, substantial social inequalities are expected to shape its development and its outcomes. It is well known that educational prerequisites of job acquisition and promotion, level of occupational standing, and quality of exposure to work environments are socially patterned. With each step one moves up on the ladder of a society’s opportunity structure, the better one’s work and employment conditions. This also holds true for main outcomes, such as health and life expectancy. In this chapter, these associations are explored in more detail, using updated evidence from research in social and occupational epidemiology and medical sociology. Its content is structured into four main parts. The first part addresses social inequalities in access to work and employment, and provides summary information on the social patterning of quality of work and employment. Subsequently, associations of quality of work and employment with significant health outcomes are outlined, using evidence from systematic literature reviews. Based on a substantial amount of evidence, the third part deals with the chapter’s core question of whether, and to what extent, the socially patterned quality of work and employment conditions can explain social inequalities in health. To this end, research findings from mediation analysis are incorporated. In the final part, I discuss the potential policy implications of this knowledge for attempts to reduce social inequalities in health.

SOCIAL INEQUALITIES IN ACCESS TO AND QUALITY OF WORK AND EMPLOYMENT Paternal and maternal socioeconomic background exerts strong and pervasive effects on a variety of children’s developmental outcomes. These outcomes include health functioning during childhood (Irwin et al., 2007; Power and Kuh, 2006), cognitive, affective and 172

Work and health inequalities  173 motivational abilities (Farah and Hackman, 2012), educational performance (OECD, 2018), and occupational attainment (Bond and Saunders, 1999), among others. Multiple pathways link mother’s and father’s socioeconomic disadvantage, as measured by low socioeconomic position (SEP), with children’s development. Pregnancy and the first few years after birth define a critical period. In addition to exposure to toxic material and nutritional conditions, psychosocial adversity contributes to the embodiment of social inequalities in children through impaired brain development. More specifically, chronic stress experienced during this sensitive period affects neural circuits involved in cognitive performance and in the self-regulation of emotion and behaviour. According to McEwen and McEwen (2017), stress-induced impact on neural circuits in the prefrontal cortex compromises several essential cognitive, emotional, and motivational brain actions, such as executive function, working memory, attention, delay of gratification, control of impulse and regulation of emotional expression. In the long run, these subtle alterations of functioning translate into developmental delays, and they reduce children’s cognitive performance and self-regulatory capacity. As an obvious consequence of these processes, children with low socioeconomic parental background often lag behind the educational success and school performance of children with a more privileged socioeconomic background (Evans and Kim, 2013). Educational achievement is a powerful determinant of access to, and quality of labour market participation. Higher levels of educational qualification, as documented in school certificates, degrees and examinations, offer better opportunities for entry into occupational life. In all advanced societies, though to a different extent, we observe social gradients in occupational attainment (OECD, 2018). Longitudinal investigations provided in-depth examinations of this relationship. For instance, father’s low SEP in combination with low parental support during school and unstable family relationships predicted children’s unemployment risk and poor working conditions at age 33 in a British birth cohort study (Power et al., 2002). A similar result was observed in an Australian study exploring youth unemployment (Caspi et al., 1998). In a Finnish investigation, adverse psychosocial working conditions at age 31 were predicted by low parental SEP, low school performance and attendance, and unhealthy behaviour during adolescence (Elovainio et al., 2007). Similar risk factors were reported focusing on poor school engagement, low self-rated abilities between age seven and 15 and their effect on adverse psychosocial work experience in early adulthood (Wang et al., 2018). In several of these studies, impaired health during childhood contributed to the explanation of poor occupational standing later on. These latter findings underline the bi-directional associations between job opportunities and health, adding to the complexity of causal analysis (Duncan et al. 2017). Additional support for a strong link between socioeconomic adversity during childhood and critical access to the labour market is provided by cohort studies with retrospective assessment of childhood conditions (Hintsa et al., 2010; see also Wahrendorf and Demakakos, 2020). It is important to mention that these socioeconomic disadvantages in labour market access and quality of employment are magnified by macrostructural conditions, specifically economic shocks, such as the global financial crisis in 2007/2008 (Eurofound, 2013) and the Covid-19 pandemic (Eurofound, 2021). In fact, several research findings reported links between economically induced growth of youth unemployment and poor mental health, including suicidal risk (WHO, 2011).

174  Handbook of health inequalities across the life course Social Gradients of Poor Working Conditions There is abundant evidence available on social gradients of exposure to physically adverse working conditions, using occupational job task groups, as defined by the International Standard Classification of Occupations (ISCO) (ILO, 2012), or using a theory-based classification of occupations, focusing on power and resources, as defined by the European Socio-economic Classification (ESeC) (d’Errico et al., 2017). Rather consistently, people working in less-skilled, manual, blue-collar or other elementary occupations are more often exposed to adverse physical working conditions, compared to those with higher occupational standing. This holds true for chemical substances with carcinogenic or mutagenic effects (Montano, 2020), physical exposures including noise, heat, cold or radiation (Toch et al., 2014), and ergonomic factors, in particular postural constraints, repetitive movements, heavy lifting or vibrations (Montano, 2014). In addition, risks of occupational injuries follow a social gradient (d’Errico et al., 2007). Long working hours and shift work are also established occupational health risks, but their social distribution is less obvious. These material working conditions persist even in economically advanced societies, as is obvious from recent comparative data across European countries (Eurofound, 2019). However, with a major shift from the industrial to the service sector of employment, alongside technological progress and economic globalization, psychosocial work environments became more prominent and received substantial attention in research and policy. Critical features of these environments include those employment conditions and organizational and interpersonal factors that affect workers’ mental and physical health via sensory input, cognitions, emotions, and their physiological consequences. For instance, with the expansion of service delivery- and communication technology-driven jobs, mental and socioemotional demands are increasing. Alongside economic globalization, a heightened amount of competition and work pressure was observed (Gallie, 2013). Moreover, economic pressure and ground-breaking technological advances resulted in profound changes of employment relations, such as a rise in nonstandard employment, increased flexibility of work-time arrangements and workplaces, and growing job instability, insecurity, involuntary part-time work and precarious employment (Kalleberg, 2009). Besides these highly prevalent conditions of work pressure and job insecurity, additional critical features of psychosocial work environments are emerging in advanced economies. Yet, due to their complexity and variation, it is difficult to identify and assess them in a consistent way. To this end, theoretical models need to be introduced. It is the aim of a theoretical model in this domain of research to define those distinct aspects of work and employment organization that expose workers to chronic stressful experience. These aspects are delineated at a level of generalization that allows for their identification in a wide range of occupations. Whereas several such theoretical models were developed and tested (for review Theorell, 2020), two concepts have received special attention in recent international occupational and social epidemiological research, the models termed ‘demand-control’ and ‘effort–reward imbalance’. The former model posits that jobs characterized by high (mainly psychological) demands and by low job task control or decision authority exert noxious effects on health. The effects of this combination, termed ‘job strain’, on health are even more pronounced if no social support at work is available (Karasek and Theorell, 1990). The latter model focuses on contractual features at work where high effort is often not met by adequate rewards in terms of pay, promotion, job security and esteem. A recurrent imbalance between high ‘cost’ and low

Work and health inequalities  175 ‘gain’ compromises workers’ coping abilities, triggering adaptive emotional and physiological breakdown (Siegrist, 1996). As these conditions continue to be highly prevalent in current labour markets, it is of interest to see whether they follow the social gradient documented in the case of adverse physical work. But how are these models measured? In either case, their core dimensions are assessed by standardized psychometrically validated scales containing answers to Likert-scaled items. For instance, one out of several items measuring a relevant component of job control in the ‘demand-control’ model is ‘Can you influence how to do your job?’ (Karasek and Theorell, 1990). In the ‘effort–reward imbalance’ model, the statement ‘My job promotion prospects are poor’ is one out of 11 items assessing low reward (Siegrist, 1996). By means of questionnaires these scales are administered in large-scale epidemiological studies providing self-reported data on critical psychosocial work environments. In these studies, scores of single scales or of their theory-based combination are used as predictors of incident disease risk (see below). In the former model, a standard approach of assessing ‘job strain’ is the construction of quartiles of the median-split two scales, with ‘high demand/low control’ as the main predictor. In the latter model, as an alternative to interaction analysis, a logarithmic ratio of the scales ‘effort’ and ‘reward’ is often used to quantify the imbalance between high cost and low gain at individual level (Siegrist and Wahrendorf, 2016). Several cross-national investigations offer opportunities to answer the question of a social gradient of these models of a health-adverse psychosocial work environment, specifically the European Working Conditions Survey (EWCS) and the Survey of Health, Aging and Retirement in Europe (SHARE). In this latter study, a pronounced social gradient of high effort and low reward, and low control at work, were documented according to three alternative measures of unequal occupational positions. For instance, the prevalence of effort–reward imbalance was 44.5 per cent among employees in the lowest occupational class compared to 21.1 per cent among those in the highest class, and the prevalence of low control was 32.3 in the lowest vs. 9.3 per cent in the highest occupational class (Wahrendorf et al., 2013). Similar findings were documented in the former data set, using education (Lunau et al., 2015) or occupational skill level as SEP indicators (Rigó et al., 2021). This latter study is of particular interest for two reasons. First, in addition to the two summary measures of high effort/low reward and job strain, the main components of these concepts were additionally analysed. This is important as high demand and high effort are often more prevalent among higher-skilled groups, thus producing inconsistent results (Dragano and Wahrendorf, 2016). This European study confirmed most consistent gradients for the two aspects of low control at work (‘decision authority’; ‘skill discretion’) and for the component ‘low reward at work’. As a second strength, the investigation used repetitive measures of work stress covering a period from 1995 to 2015. Remarkably, levels of work stress persisted at a rather high level during this time, with little change except a significant increase in job demands (Rigó et al., 2021). In conclusion, physically and psychosocially adverse working and employment conditions are more prevalent among employed populations with lower occupational standing. This also holds true if the composite scores of these two types of exposures are investigated (Toch et al., 2014). Thus, in addition to the burden of social disadvantage during childhood and adolescence, social inequalities related to work and employment persist into midlife and early old life. It is therefore important to demonstrate their adverse effects on health. Moreover, in a subsequent step, the contribution of adverse work to the explanation of social inequalities in morbidity and mortality needs to be analysed.

176  Handbook of health inequalities across the life course

HEALTH CONSEQUENCES OF POOR QUALITY OF WORK AND EMPLOYMENT For many decades, occupational medicine has dealt with the impact of toxic working conditions on workers’ health, predominantly in the frame of occupational diseases and injuries (Lance, 2019). An elaborated list of occupational diseases points to several disorders with a public health impact, due to their high prevalence. In a joint project, the World Health Organization (WHO) and the International Labour Organization (ILO) provided a new synthesis of knowledge, combined with original data analysis (WHO/ILO, 2021). Its aim was to assess the burden of deaths and of disability adjusted life years (DALYs) that is attributable to main occupational risk factors. With regard to mortality, risk factors with the highest prevalence were (1) long working hours, (2) particulate matters, gases and fumes, (3) injuries, and (4) asbestos. The main related causes of death were chronic obstructive pulmonary diseases (COPD), stroke, ischaemic heart disease (IHD) and lung cancer. If DALYs were considered, the main occupational risk factors were (1) injuries, (2) long working hours, (3) ergonomic factors, (4) particulate matters, and (5) noise. The order of health outcomes related to these risk factors differs if DALYs instead of deaths are analysed. Stroke is followed by back and neck pain, COPD and IHD, ranging before hearing loss and lung cancer (WHO/ILO, 2021). Two observations emerge from these facts. First, in a majority of cases, occupational risk factors and their related health outcomes are in line with established procedures of defining physical, chemical, biologic and ergonomic occupational hazards and occupational diseases. Yet, there is an important exception, as long working hours are not adequately defined as a material risk factor. Rather, exposure time to a variety of working conditions, including adverse psychosocial work environments, seems to matter. Secondly, data on the health burden do not include information on social inequalities. This gap in knowledge needs to be filled by data from other sources. Social gradients in the prevalence of lung cancer (Mackenbach, 2019), chronic obstructive pulmonary disease (Blanc and Torén, 2020) and musculoskeletal disorders (Montano, 2014) are well established, and occupational factors contribute significantly to their development. In addition, disability pensions are more frequent among socially less privileged employees. Part of an increased burden of disease preceding disability pension in these employment groups can be attributed to adverse working conditions (Haukenes et al., 2011). Importantly, a consistent association between occupational grade and risk of disability pension is obvious from a coordinated data analysis of seven European cohort studies, as documented in Figure 12.1. In the lowest of three occupational classes, disability risks are significantly elevated if compared to the highest group, both among men and women. Moreover, effects are stronger in the two occupational cohorts (Whitehall II; GAZEL) than in the five studies of general populations (Carr et al., 2018). As mentioned, the impact of work on health goes beyond traditional occupational diseases and injuries. In fact, a large number of prospective epidemiologic studies investigated the contribution of stressful psychosocial work environments towards explaining elevated disease risks. This rapid increase of knowledge over the past two decades has been documented in a recent meta-review of 72 literature reviews of studies, covering a range of psychosocial predictors and a variety of health outcomes (Niedhammer et al., 2021). The main evidence of this most comprehensive summary account of the current state of art is derived from systematic reviews of prospective cohort studies with a comparable design, where meta-analyses of findings provide a pooled quantitative estimate of incident disease risk among exposed vs.

Work and health inequalities  177

Notes: BHPS, British Household Panel; ELSA, English Longitudinal Study of Ageing; FPS, Finnish Public Sector Study; GAZEL, Electricité De France-Gaz De France; HRS, Health and Retirement Study; NSHD, National Survey of Health and Development. Source: Carr et al. (2018).

Figure 12.1

Association of low occupational grade with risk of health-related work exit (disability pension) in seven European cohort studies

non-exposed populations. Quantitative estimates and their 95 percent confidence intervals that reach beyond the threshold of 1.0 (the relative disease risk of the non-exposed population) are interpreted as evidence in favour of a causal effect of stressful work on disease development, in particular if they are adjusted for important confounding factors, and if there is supporting information on psychobiological or behavioural factors underlying these associations (Theorell, 2020). The study by Niedhammer et al. (2021) builds on several dozens of meta-analyses of prospective risk estimates, where ischaemic heart disease, depression, and type 2 diabetes were most often selected as health outcomes. Additional disorders include musculoskeletal disorders, sleep problems, and addictive disorders (psychotropic substances). Of the two theoretical models of a stressful psychosocial work environment described earlier, job strain was by far most often examined, based on 37 reviews, whereas effort–reward imbalance was tested by studies summarized in 12 reviews. Moreover, two single components were addressed in a larger number of reviews – long working hours (a potential proxy indicator of high demand or high effort), and job insecurity (a reward component within the latter model). In the next section, a short summary of main findings from recent systematic reviews with meta-analyses is given. Information on health outcomes related to the two theoretical models is provided in terms of risk estimates, indicated as odds ratios (OR), relative risks (RR), or hazard ratios (HR) among exposed vs. unexposed populations. Given the constraints of a short summary, statistically significant risk estimates only are reported (indicated as RR) without

178  Handbook of health inequalities across the life course additional data. When interpreting these numbers, differences in the risk estimates across reviews may reflect variations in the number of studies included or in the selection criteria applied and the quality assessments performed. Depression is the disorder associated with the relatively highest risk estimates of exposure-outcome associations. Among four partly overlapping systematic reviews, RRs for job strain varied from 1.22 to 1.77 (Niedhammer et al., 2021). In the remaining reviews, RRs were 1.14 (Mikkelsen et al., 2021), 1.47 (Duchaine et al., 2020) and 1.60 (‘high demand’ only; van der Molen et al., 2020). Respective RRs for effort–reward imbalance were 1.68 in the Niedhammer et al. (2021) review, and 1.53 (Mikkelsen et al., 2021), 1.66 (Duchaine et al., 2020) and 1.90 (van der Molen et al., 2020) in the remaining reviews. The first two reviews found additional support for elevated RRs for the component ‘job insecurity’. Overall, based on dozens of cohort studies including large working populations from a variety of economically advanced countries, we can conclude that the relative risk of developing clinically relevant depressive symptoms or a depressive disorder is elevated by about 60 per cent among those exposed to chronic psychosocial adversity at work, if compared to non-stressed groups. A similar conclusion can be drawn with regard to ischaemic heart disease (IHD) although at a somehow lower level of risk. Here, RRs for job strain varied from 1.17 to 1.45, with a majority of reviews reporting values beyond 1.30 (Niedhammer et al., 2021). In addition, a more recent systematic review found an elevated RR of 1.50 of fatal IHD for the component ‘low job control’ which was attenuated but remained statistically significant after multiple statistical adjustment (Taouk et al., 2021). Associations of effort–reward imbalance with IHD incidence were of comparable size, with RRs ranging from 1.18 to 1.58 (Niedhammer et al., 2021). Again, the majority of estimates were in the upper range, approaching the pooled estimate of RR = 1.42 that was obtained from another review (Siegrist and Li, 2020). Further support for the associations of these two conditions of stressful work with IHD risk comes from some longitudinal studies that documented an elevated risk of recurrent IHD among cardiac patients who returned to work and continued to be exposed to stressful work (Li et al., 2015). Type 2 diabetes was the third health outcome studied in two systematic reviews, with elevated RRs of 1.15 for job strain and 1.24 for effort–reward imbalance (Niedhammer et al., 2021; Pena-Gralle et al., 2021). For both outcomes – IHD and diabetes – comparable effects were observed for job insecurity. Other review findings (e.g. on musculoskeletal disorders; Taibi et al., 2021) complement this summary. Taken together, there is solid evidence of elevated risks of three chronic diseases with high relevance to public health due to exposure to a stressful psychosocial work environment, as measured by the two theoretical models or single components (depression, IHD, and type 2 diabetes; GBD, 2015). All three disorders tend to be more prevalent among lower socioeconomic population groups (Mackenbach, 2019), and, as was reported above, this also holds true for the exposures under study. Despite the rather modest risk elevations – they do not reach or exceed a doubling of relative risk – these risks deserve attention, given the fact that about a quarter of the total workforce reports job strain or effort–reward imbalance at work (Karasek and Theorell, 1990; Siegrist and Wahrendorf, 2016). Against this background, an answer is expected to the question of whether, and to what extent, psychosocial and material working conditions can explain the documented social inequalities in morbidity and mortality of working-age populations.

Work and health inequalities  179

THE ROLE OF WORK IN EXPLAINING SOCIAL INEQUALITIES IN HEALTH Mediation The main approach towards studying this question is termed ‘mediation analysis’. Three assumptions must be met in order to apply this statistical approach. First, a significant association of SEP with the disease under study is required: the lower people’s SEP, the higher their disease risk. Secondly, the explanatory construct proposed as a mediator is significantly associated with SEP. In this case, adverse physical and psychosocial working conditions are expected to be more prevalent among people with lower SEP. Third, there is a significant relationship between the mediator and the disease under study. If these preconditions are met, one can explore to what extent the mediator contributes to the explanation of the link between SEP and disease. From a policy perspective, this information is crucial, given the expectation that improving the mediating condition results in a reduction of social inequalities in health. However, to ensure that these associations represent a causal chain, additional requirements must be met. Working conditions need to exert their effect long before the occurrence of incident disease, and data on biological processes underlying the association are needed. Likewise, a reverse relationship between SEP and work must be excluded (low SEP as result of occupational downward mobility). Sophisticated statistical models were developed to tackle these requirements (see below). So far, in mainstream epidemiological research, a classical, methodologically weaker statistical approach towards conducting mediation analysis was applied, termed ‘difference method’ (Baron and Kenny, 1986). As many epidemiologic data sets are analysed by a stepwise multivariable regression approach, this method, in a first step, estimates the effect of SEP on a specific disease outcome. To take a famous example from the British Whitehall II study, Marmot et al. (1997) documented a social gradient of IHD incidence over a mean 5.3-year observation period according to the employment grade of civil servants. For those in the lowest grade, an odds ratio of IHD of 1.47 was observed, compared to those in the highest grade. In a second step, this crude estimate was adjusted for the mediator variable ‘low job control’ as this variable was associated with both employment grade and IHD. This adjustment reduced the odds ratio from 1.47 to 1.20, and this attenuation of risk was interpreted as an explanatory contribution of low job control to the relationship between SEP and IHD. In this case, low job control accounted for about half the social gradient of IHD (calculated as 100*(1.47 – 1.20)/(1.47 – 1.0) = 57.4 per cent; see Mackenbach, 2019, p. 55). Additional adjustment for confounding factors (specifically biomedical and behavioural cardiovascular risk factors) further reduced the social gradient. Importantly, when adjusting the association of low job control with IHD for SEP, no noticeable reduction of the effect size was observed (Marmot et al., 1997). A large part of currently available evidence on the role of work in explaining health inequalities is based on this ‘difference method’ (for a review see Hoven and Siegrist, 2013). On balance, results of these studies suggest that a combination of distinct material (physical, biomechanical) and psychosocial work factors is best suited to explain social inequalities in health. Among others, this conclusion is illustrated by the impressive example of a French prospective cohort study, where 4,118 working women and men were followed from 1996 to 2008. Questionnaire-based data on occupational factors collected at baseline were linked to register data on mortality during follow-up (Niedhammer et al., 2011). As a main finding,

180  Handbook of health inequalities across the life course manual workers as compared to managers had a significantly elevated hazard ratio of premature mortality of 1.88 (95% CI 3.01; 1.17). When adjusted for material and psychosocial occupational factors, the hazard ratio was reduced to 1.22 (95% CI 2.21; 0.74). Job insecurity, as measured by temporary employment, and biomechanical and physical work characteristics, contributed most to this reduction, in particular among men. Further adjustment for behavioural risk factors did not substantially change these associations. The usual interpretation of policy implications of such findings was expressed in the concluding sentences of this study as follows: ‘Preventive actions focusing on these factors and specific social groups may be useful to reduce social inequalities in mortality’ (Niedhammer et al., 2011, p. 10). More recently, the ‘difference method’ of mediation analysis was criticized on methodological grounds. First, unobserved confounders may interfere with the three-way associations under study as they can act on the predictor, the mediator and the outcome, thus causing biased results. Secondly, effects of the mediator on health may vary between socioeconomic groups, causing effect heterogeneity. Moreover, in many cases, more than one mediator needs to be studied, and associations between mediators add to the complexity of analysis. Therefore, new statistical approaches based on a counterfactual model were developed to tackle these challenges (Lange et al., 2017; VanderWeele, 2015). While first applications of this new approach seem promising (e.g., Laine et al., 2020), it is not yet clear whether, and to what extent, they invalidate the knowledge established by the conventional approach. To my knowledge, a comparative evaluation of both methods with regard to the mediating role of working conditions in the relationship between SEP and health has not yet been conducted. A further advantage of the new methodology is the combined analysis of mediation and moderation effects on health outcomes. For instance, a stronger effect of the mediator on health in a low vs. high SEP group points to a moderating role of low SEP (see above, effect heterogeneity). Moderation Mediation analysis offers the main approach to explaining the link between SEP, working conditions and health, based on the assumption that exposure to the working conditions varies along the social gradient. Yet, as was mentioned, the effect of this exposure on health may vary according to SEP, pointing to differential disease susceptibility. To this end, moderation analysis supplements mediation analysis, exploring the interaction of SEP and work in the estimation of unequal health effects. The main hypothesis in this context maintains that the effect of adverse work on health is particularly strong in the group of workers with low socioeconomic position. This reflects a pronounced vulnerability, or a lack of available resources for successfully coping with adversity at work. In policy terms, results derived from moderation analysis can inform decision-makers to prioritize intervention measures according to need. In the case of scarce resources, such measures will be offered primarily to the least privileged groups, i.e., the group with highest need. Among research on social inequalities in health, far fewer results testing the moderation hypothesis are currently available than is the case for mediation analysis (Mackenbach, 2019). One reason may be the large sample size required for a valid test of multiplicative interaction analysis. Another reason points to a lack of convincing knowledge about the causes of a pronounced vulnerability or susceptibility among more disadvantaged socioeconomic groups. Furthermore, similar to mediation analysis, the conventional moderation analysis depends on some uncontrolled assumptions that can only be addressed by a newly developed statistical approach, the one based on a counterfactual

Work and health inequalities  181 paradigm. This approach integrates mediation and moderation analysis within a unifying framework (VanderWeele, 2015). Despite these limitations, several epidemiologic studies confirm stronger effects of adverse work on health among manual, elementary or poorly skilled workers compared to the effects among more privileged workers, thus supporting the notion of their increased vulnerability to chronic psychosocial stress at work. This notion is illustrated by two examples. The first example complements the mediation analysis mentioned above on the role of low job control as a partial explanation of the link between SEP and IHD (Marmot et al., 1997). In a Swedish cross-sectional study, relationships between these three variables were explored in terms of moderation analysis. In the group of manual workers, the relative risk of myocardial infarction due to job strain, and specifically low job control, was substantially higher than the respective relative risk in the group of non-manual workers. This observation was interpreted by the authors as follows: ‘Manual workers are more susceptible when exposed to job strain, explaining about 20–50% of the relative risk among manual workers’ (Hallqvist et al., 1998, p. 1405). To summarize, in addition to a higher prevalence of exposure to stressful work (mediation hypothesis), the strength of the effect of this exposure on health (moderation hypothesis) contributes to explanations of social inequalities in health. With a second example, the joint effect of low SEP and stressful work on the risk of experiencing insomnia, as measured by effort–reward imbalance, was explored among several thousand middle-aged employed men in Japan. Taking the highest of three occupational groups that scored low on work stress as reference (OR = 1.0), the OR for insomnia in this occupational group scoring high on work stress (upper tertile) was 3.94. In comparison, in the lowest occupational group with a low level of work stress, the respective OR was 1.31, but in the low occupation group with high work stress, the OR of insomnia was 9.43. A statistically significant synergy index indicated that this combination exceeded the additive effects (Yoshioka et al., 2013). In conclusion, available evidence from mediation analysis supports the proposition that adverse physical and psychosocial working conditions contribute to the explanation of the social gradient in health and disease. In addition to differential exposure to adverse working conditions, differential vulnerability according to SEP, as documented by moderation analysis, may aggravate the burden of work on health. This knowledge can be useful in developing preventive activities that aim at improving working conditions and their effects on health.

POLICY IMPLICATIONS This chapter has documented scientific evidence on social inequalities in the quality of work and employment, and poor quality of work and employment was shown to increase the burden of work-related disease. Even if one can still question the causality of links between socioeconomic position, work and health, as documented by mediating pathways, there are convincing reasons to address the policy implications of available knowledge. Reducing substantial inequalities in the quality of work and employment between socioeconomic strata of the workforce seems mandatory from a moral as well as from an economic point of view. In line with the United Nations Sustainable Developmental Goals, societies are expected to develop measures towards increasing and promoting decent and sustainable working conditions (United Nations, 2015). Economically, the financial gain of investments into good quality of work and employment were demonstrated at company level in a variety of investigations (Johanson

182  Handbook of health inequalities across the life course and Aboagye, 2020). These moral and economic arguments can equally be applied to the aim of reducing the avoidable burden of work-related disease (Barnay, 2020). In consequence, policy measures are required at least in the following two interrelated areas. First, occupational risk factors deserve systematic monitoring and surveillance at company level, supported by national regulations. Importantly, this monitoring activity includes the newly established psychosocial occupational risk factors. Based on this information, prevention and intervention efforts need to be designed and implemented. To this end, qualified and well-equipped occupational health and safety professionals are required, and business leaders and managers are expected to prioritize worksite health promotion programs according to need. However, in view of the magnitude and scale of this challenge, preventive measures cannot be restricted to the level of companies and organizations. Given the intimate links between the micro-environment of enterprises and the macro-environment of influential labour-market developments and economic forces, a second area of engagement must be addressed, the area of national labour and social policies. These policies represent essential elements of modern welfare states offering basic safety and protection against major threats, such as unemployment, occupational injury, work-related disability, need for medical care, and poverty in old age. In many cases, they are extended to include measures of improving quality of work and engagement, of integrating people into the labour market, and of reconciling work and family life. Huge variations in these policies exist between countries, and it is well established that Scandinavian countries have achieved the relatively highest amount of welfare benefits. Well-developed labour and social policies improve the quality of work and employment, as evidenced by cross-national studies analysing these associations. One such investigation is of special interest in the context of this chapter as it examined the impact of policies that strengthen integration into the labour market (e.g., among unemployed, low-skilled, chronically ill people) on inequalities in quality of work (Lunau et al., 2015). The study combined data from the Survey of Health, Aging and Retirement in Europe with data from the English Longitudinal Study of Aging collected during 2010–2011, with 13,695 employed men and women aged 50 to 64 years from 16 European countries. As illustrated in Figure 12.2, the country’s integration efforts were assessed by two indicators, ‘active labour market policies’ (ALMP), measured by the number of public expenditures to promote reintegration into work (as a percentage of Gross Domestic Product), and ‘lifelong learning’. While the former indicator was obtained from administrative data, the latter indicator was computed from survey information. It represents the percentage of older workers (aged 55–64) who confirmed that they had received education or training during the past 12 months. In this study, quality of work was assessed by short versions of the two work stress models mentioned above, effort–reward imbalance and job strain (low job control only). Finally, the social inequality of employed participants was assessed by the highest educational degree (three categories). In a pooled dataset, the effects of these two policy indicators on social inequalities of work stress were analysed, using multilevel linear regression models. These effects are visualized in Figure 12.2 by the prediction of level of work stress for each educational group according to the extent of policy development for each policy indicator. Two findings from this figure are remarkable. First, levels of work stress are generally higher if integrated policies are less well developed. Second, the steepness of the social (educational) gradient of work stress varies according to the degree of policy investment. In countries with a highly developed integration policy, social inequalities are small (or even absent in the case of lifelong learning), whereas the opposite is obvious in countries with poor

Work and health inequalities  183

Source: Lunau et al. (2015).

Figure 12.2

Predicted levels of work stress (effort–reward imbalance; low control) by socioeconomic position (education) at different levels of national integrative policies (ALMP; lifelong learning) in 16 European countries

policy development. Of interest, these two findings are very similar for the two work stress models (Lunau et al., 2015). The results of this cross-national analysis lend some support to the notion that distinct well-developed national labour policies are associated with lower levels of psychosocial stress at work. Moreover, under these conditions, social inequalities of stressful work tend to be less pronounced. In this chapter, a limited scientific frame of analysis for the relationships between work, health and social inequalities was offered. In times of economic globalization, restricting the analysis to economically advanced countries seems questionable, given the strong global interconnections of trade, labour market, and financial developments. The global threats of a new pandemic and the accelerated climate crisis call for a comprehensive, world-wide analytical perspective. This view is further supported by new, rapidly expanding technologies that are likely to transform the world of work and employment in unprecedented ways. Nevertheless, even within the most advanced societies, substantial inequalities in the quality of work and employment, as well as in the health of working people, became obvious. Tackling these inequalities continues to be an urgent challenge to science and policy.

184  Handbook of health inequalities across the life course

REFERENCES Barnay, T. (2020) ‘Interactions of work and health: An economic perspective’, in Theorell, T. (ed.) Handbook of socioeconomic determinants of occupational health. Cham: Springer Nature, pp. 37–52. Baron, R.M. and Kenny, D.A. (1986) The moderator–mediator variable distinction in social psychological research: conceptual, strategic, and statistical considerations. Journal of Personality and Social Psychology 51, 1173–1182. Available at: https://​psycnet​.apa​.org/​buy/​1987​-13085​-001 (accessed: 14 December 2021). Blanc, P.D. and Torén, K. (2020) ‘Respiratory work disability in relation to occupational factors’, in Bültmann, U. and Siegrist, J. (eds.) Handbook of disability, work and health. Cham: Springer International Publications, pp. 153–167. Bond, R. and Saunders, P. (1999) Routes of success: influences on the occupational attainment of young British males. British Journal of Sociology 50, 217–249. doi: 10.1111/j.1468-4446.1999.00217.x. Carr, E., Fleischmann, M., Goldberg, M., Kuh, D., Murray, E.T., Stafford, M. et al. (2018) Occupational and educational inequalities in exit from employment at older ages: evidence from seven prospective cohorts. Occupational and Environmental Medicine 75, 369–377. doi: 10.1136/oemed-2017-104619. Caspi, A., Wright, B.R.E., Moffitt, T.E. and Silva, P.A. (1998) Early failure in the labor market: childhood and adolescent predictors of unemployment in the transition to adulthood. American Sociological Review 63(3), 424–451. doi: 10.2307/2657557. Dragano, N. and Wahrendorf, M. (2016) ‘A social inequalities perspective on effort–reward imbalance at work’, in Siegrist, J. and Wahrendorf, M. (eds.) Work stress and health in a globalized economy. Cham: Springer International, pp. 67–85. Duchaine, C.S., Aubé, K., Gilbert-Ouimet, M., Vézina, M., Ndjaboujé, R., Massamba, V. et al. (2020) Psychosocial stressors at work and the risk of sickness absence due to a diagnosed mental disorder. JAMA Psychiatry 77(8), 842–851. doi:10.1001/jamapsychiatry.2020.0322. Duncan, G.J., Magnuson, K. and Votruba-Drzal, E. (2017) Moving beyond correlations in assessing consequences of poverty. Annual Review of Psychology 68(1), 413–434. doi: 10.1146/ annurev-psych-010416-044224. d’Errico, A., Punnet, L., Cifuentes, M., Boyer, J., Tessler, J., Gore, R. et al. (2007) Hospital injury rates in relation to socioeconomic status and working conditions. Occupational and Environmental Medicine 64(5), 325–333. doi: 10.1136/oem.2006.027839. d’Errico, A., Riccari, F., Stringhini, S., Carmeli, C., Kivimäki, M., Bartley, M. et al. (2017) Socioeconomic indicators in epidemiologic research: a practical example from the LIFEPATH study. PLoS One 12(5), e0178071. doi: 10.1371/journal.pone.0178071. Elovainio, M., Kivimäki, M., Ek, E., Vahtera, J., Honkonen, T., Taanila, A. et al. (2007) The effect of pre-employment factors on job control, job strain, and psychological distress: a 31-year longitudinal study. Social Science & Medicine 65(2), 187–199. doi: 10.1016/j.socscimed.2007.02.052. Eurofound (2013) Impact of the crisis on working conditions in Europe. Dublin: European Foundation for the Improvement of Living and Working Conditions. Available at: https://​www​.eurofound​.europa​ .eu/​publications/​reports/​2013/​impact​-of​-the​-crisis​-on​-working​-conditions​-in​-europe (accessed: 14 December 2021). Eurofound (2019) Working conditions and workers’ health. Luxembourg: Publications Office of the European Union. Available at: https://​www​.eurofound​.europa​.eu/​publications/​report/​2019/​working​ -conditions​-and​-workers​-health (accessed: 14 December 2021). Eurofound (2021) Covid-19: Implications for employment and working life. Luxembourg: Publications Office of the European Union. Available at: https://​www​.eurofound​.europa​.eu/​publications/​report/​ 2021/​covid​-19​-implications​-for​-employment​-and​-working​-life (accessed: 14 December 2021). Evans, G.W. and Kim, P. (2013) Childhood poverty, chronic stress, self-regulation, and coping. Child Development Perspectives 7(1), 43–48. Available at: https://​srcd​.onlinelibrary​.wiley​.com/​doi/​abs/​10​ .1111/​cdep​.12013 (accessed: 14 December 2021). Farah, M.J. and Hackman, D.A. (2012) ‘SES, child development, and the neural bases of cognition’, in Maholmes, V. and King, R.B. (eds.) The Oxford handbook of poverty and child development. Oxford: Oxford University Press, pp. 307–318. Gallie, D. (2013) Economic crisis, quality of work, and social integration. The European experience. Oxford: Oxford University Press.

Work and health inequalities  185 GBD 2015 Mortality and Causes of Death Collaborators (2016) Global, regional, and national life expectancy, all-cause mortality and cause-specific mortality for 249 causes of death, 1980–2015: a systematic analysis for the Global Burden of Disease Study 2015. Lancet 388(10053), 1459–1544. doi: 10.1016/S0140-6736(16)31012-1; PMCID: PMC5388903. Hallqvist, J., Diderichsen, F., Theorell, T., Reuterwall, C., Ahlbom, A. and SHEEP Study Group (1998) Is the effect of job strain on myocardial infarction due to interaction between high psychological demands and low decision latitude? Results from the Stockholm Heart Epidemiology Program (SHEEP). Social Science & Medicine 46(11), 1405–1415. PII: S0277-9536(97)10084-3. Haukenes, I., Mykletun, A., Knudsen, A.K., Hansen, H.-T. and Gunnar Mæland, J. (2011) Disability pension by occupational class – the impact of work-related factors: the Hordaland Health Study Cohort. BMC Public Health 11, 406. doi: 10.1186/1471-2458-11-406. Hintsa, T., Shipley, M.J., Gimeno, D., Elovainio, M., Chandola, T., Jokela, M. et al. (2010) Do pre-employment influences explain the association between psychosocial factors at work and coronary heart disease? The Whitehall II study. Occupational and Environmental Medicine 67(5), 330–334. doi: 10.1136/oem.2009.048470. Hoven, H. and Siegrist, J. (2013) Work characteristics, socioeconomic position and health: a systematic review of mediation and moderation effects in prospective studies. Occupational and Environmental Medicine 70(9), 663–669. doi: 10.1136/oemed-2012-101331. ILO (2012) International Standard Classification of Occupations 2008 (ISCO-08): Structure, definitions and correspondence tables. Geneva: International Labour Organization. Available at: https://​www​.ilo​ .org/​public/​english/​bureau/​stat/​isco/​isco08/​index​.htm (Accessed: 14 December 2021). Irwin, L.G., Siddiqi, A. and Hertzman, C. (2007) Early child development: a powerful equalizer. Geneva: World Health Organization. Available at: https://​apps​.who​.int/​iris/​handle/​10665/​69729 (accessed: 14 December 2021). Johanson, U. and Aboagye, E. (2020) ‘Financial gains, possibilities, and limitations of improving occupational health at the company level’, in Theorell, T. (ed.) Handbook of socioeconomic determinants of occupational health. Cham: Springer Nature, pp. 537–553. Kalleberg, A. (2009) Precarious work, insecure workers: Employment relations in transition. American Sociological Review 74(1), 1–22. doi: 10.1177/000312240907400101. Karasek, R.A. and Theorell, T. (1990) Healthy work. New York: Basic Books. Laine, J.E., Baltar, V.T., Stringhini, S., Gandini, M., Chadeau-Hyam, M., Kivimäki, M. et al. (2020) Reducing socio-economic inequalities in all-cause mortality: a counterfactual mediation approach. International Journal of Epidemiology 49(2), 497–510. doi: 10.1093/ije/dyz248. Lance, C. (2019) Occupational and environmental medicine. Foster Academics. Lange, T., Hansen, K., Sorensen, R. and Galatius, S. (2017) Applied mediation analyses: a review and tutorial. Epidemiology and Health 39, e2017035. doi: 10.4178/epih.e2017035. Li, J., Zhang, M., Loerbroks, A., Angerer, P. and Siegrist, J. (2015) Work stress and the risk of recurrent coronary heart disease events: a systematic review and meta-analysis. International Journal of Occupational Medicine and Environmental Health 28, 8–19. doi: 10.2478/s13382-014-0303-7. Lunau, T., Siegrist, J., Dragano, N. and Wahrendorf, M. (2015) The association between education and work stress: does the policy context matter? PLoS One 10(3), e0121573. doi: 10.1371/journal. pone.0121573 Mackenbach, J.P. (2019) Health inequalities. Oxford: Oxford University Press. Marmot, M.G., Bosma, H., Hemingway, H., Brunner, E.J. and Stansfeld, S. (1997) Contribution of job control and other risk factors to social variations in coronary heart disease incidence. Lancet 350, 235–239. doi: 10.1016/S0140-6736(97)04244-X. McEwen, C.A. and McEwen, B.S. (2017) Social structure, adversity, toxic stress, and intergenerational poverty: an early childhood model. Annual Review of Sociology 43, 445–472. doi: 10.1146/ annurev-soc-060116-053252. Mikkelsen, S., Coggon, D., Andersen, J.H., Casey, P., Meulengracht Flachs, E., Albert, H. et al. (2021) Are depressive disorders caused by psychosocial stressors at work? European Journal of Epidemiology 36, 479–496. doi: 10.1007/s10654-021-00725-9. Montano, D. (2014) Upper body and lower limbs musculoskeletal symptoms and health inequalities in Europe: an analysis of cross-sectional data. BMC Musculoskeletal Disorders 15, 285. doi: 10.1186/1471-2474-15-285

186  Handbook of health inequalities across the life course Montano, D. (2020) ‘Social distribution of occupational hazards’, in Theorell, T. (ed.) Handbook of socioeconomic determinants of occupational health. Cham: Springer Nature, pp. 169–189. Niedhammer, I., Bertais, S. and Witt, K. (2021) Psychosocial work exposures and health outcomes: a meta-review of 72 literature reviews with meta-analysis. Scandinavian Journal of Work, Environment & Health 47(7), 489–508. doi: 10.5271/sjweh.3968 Niedhammer, I., Bourgkard, E., Chau, N. and The Lorhandicap Study Group (2011) Occupational and behavioural factors in the explanation of social inequalities in premature and total mortality: a 12.5-year follow-up in the Lorhandicap study. European Journal of Epidemiology 26(1), 1–12. doi: 10.1007/s10654-010-9506-9. OECD (2018) Equity in education. Paris: OECD Pena-Gralle, A.P.B., Talbot, D., Duchaine, C.S., Lavigne-Robichaud, M., Trudel, X., Aubé, K. et al. (2021) Job strain and effort–reward imbalance as risk factors for type 2 diabetes mellitus: a systematic review and meta-analysis of prospective studies. Scandinavian Journal of Work, Environment & Health 47(4), 249–257. doi: 10.5271/sjweh.3987. Power, C. and Kuh, D. (2006) ‘Life course development of unequal health’, in Siegrist, J. and Marmot, M. (eds.) Social inequalities in health. Oxford: Oxford University Press, pp. 27–53. Power, C., Stansfeld, S.A., Matthews, S., Manor, O. and Hope, S. (2002) Childhood and adulthood risk factors for socio-economic differentials in psychological distress: evidence from the 1958 British birth cohort. Social Science & Medicine 55(11), 1989–2004. doi: 10.1016/S0277-9536(01)00325-2 Rigó, M., Dragano, N., Wahrendorf, M., Siegrist, J. and Lunau, T. (2021) Work stress on rise? Comparative analysis of trends in work stressors using the European working conditions survey. International Archives of Occupational and Environmental Health 94(3), 459–474. doi: 10.1007/ s00420-020-01593-8. Siegrist, J. (1996) Adverse health effects of high effort-low reward conditions at work. Journal of Occupational Health Psychology 1(1), 27–43. doi: 10.1037/1076-8998.1.1.27. Siegrist, J. and Li, J. (2020) ‘Effort–reward imbalance and occupational health’, in Theorell, T. (ed.) Handbook of socioeconomic determinants of occupational health. Cham: Springer Nature, pp. 355–382. Siegrist, J. and Wahrendorf, M. (eds.) (2016) Work stress and health in a globalized economy: the model of effort–reward imbalance. Cham: Springer International Publications. Taibi, Y., Metzler, Y.A., Bellingrath, S. and Müller, A. (2021) A systematic overview on the risk effects of psychosocial work characteristics on musculoskeletal disorders, absenteeism and workplace accidents. Applied Ergonomics 95, 103434. doi: 10.1016/j.apergo.2021.103434. Taouk, Y., Spittal, M.J., LaMontagne, A.D. and Milner, A.J. (2021) Psychosocial work stressors and risk of all-cause and coronary heart disease mortality: a systematic review and meta-analysis. Scandinavian Journal of Work, Environment & Health 46(1), 19–31. doi: 10.5271/sjweh.3854. Theorell, T. (ed.) (2020) Handbook of socioeconomic determinants of occupational health. Cham: Springer Nature Toch, M., Bambra, C., Lunau, T., van der Wel, K.A., Witvliet, M.I., Dragano, N. et al. (2014). All part of the job? The contribution of the psychosocial and physical work environment to health inequalities in Europe and the European health divide. International Journal of Health Services 44(2), 285–305. doi: 10.2190/HS.44.2.g. United Nations (2015) Transforming our world: the 2030 Agenda for Sustainable Development. New York. van der Molen, H.F., Nieuwenhuijsen, K., Frings-Dresen, M.H.W. and de Groene, G. (2020) Work-related psychosocial risk factors for stress-related mental disorders: an updated systematic review and meta-analysis. BMJ Open 10, e034849. doi: 10.1136/bmjopen-2019-034849. VanderWeele, T. (2015) Explanation in causal inference: methods for mediation and interaction. Oxford: Oxford University Press. Wahrendorf, M. and Demakakos, P. (2020) ‘Childhood determinants of occupational health at older ages’ in Theorell, T. (ed.) Handbook of socioeconomic determinants of occupational health. Cham: Springer Nature, pp. 321–338. Wahrendorf, M., Dragano, N. and Siegrist, J. (2013) Social position, work stress, and retirement intentions: a study with older employees from 11 European countries. European Sociological Review 29(4), 792–802. doi: 10.1093/esr/jcs058.

Work and health inequalities  187 Wang, S., Sanderson, K., Venn, A., Dwyer, T. and Gall, S. (2018) Association between childhood health, socio-economic and school-related factors and effort–reward imbalance at work: a 25-year follow-up study. Occupational and Environmental Medicine 75(1), 37–45. WHO (2011) Impact of economic crisis on mental health. Copenhagen: World Health Organization. Regional Office for Europe. WHO/ILO (2021) WHO/ILO joint estimates of the work-related burden of disease and injury, 2000–2016. Geneva: WHO; ILO. doi: 10.1136/oemed-2017-104308. Available at: https://​www​.who​ .int/​publications/​i/​item/​9789240034945 (accessed: 14 December 2021). Yoshioka, E., Saijo, Y., Kita, T., Staoh, H., Kawaharada, M. and Kishi, R. (2013) Effect of the interaction between employment level and psychosocial work environment on insomnia in male Japanese public service workers. International Journal of Behavioral Medicine 20(3), 355–364. doi: 10.1007/ s12529-012-9230-9.

13. Family relations and health inequalities: grandparents and grandchildren Valeria Bordone, Giorgio Di Gessa and Karsten Hank

INTRODUCTION The ongoing pluralization of family forms, which was initially perceived as a threat for the institution of the family, now takes the role of opportunity, especially for grandparents who are seen as increasingly central for their families, often satisfying demands for childcare (Silverstein, Giarrusso and Bengtson, 2003). Social sciences have thus enlarged their focus of research to the grandparent–parent–child relationship and their health-related consequences. Indeed, family dynamics determine the existence, timing, and extent of critical intergenerational relationships, which in turn affect the health and well-being of the family members involved and do so in heterogeneous ways (Hünteler and Hank, 2021). Parents share 50% of the same genes with their biological children and 25% with their biological grandchildren. From an evolutionary perspective, parents and grandparents can thus boost their fitness by devoting resources to their children and grandchildren, helping to ensure the growth, development and reproduction of their kin. In answer to the pluralization of family forms, the shared genetic interest has been further developed to also involve in-laws or in general family members who are usually not closely genetically related but become “inversely” genetically related to each other through common descendants (see Tanskanen and Danielsbacka, 2019 for a review). Classic sociological theory and a growing body of empirical literature has shown that social integration affects well-being throughout the life course (see e.g. Mayer, Chapter 2 in this volume). Cohen (2004) described two general mechanisms through which social relationships can promote health and well-being. On the one hand, social relations provide resources that are helpful to cope with stress. On the other hand, social integration is conducive to well-being irrespective of the level of stress, as it promotes favorable psychological states. Intergenerational relationships are the main source of support and social integration in later life and are therefore central to older adults’ well-being. Moreover, roles invested with importance are assumed to have a greater impact on mental health than less important roles because they are enacted with more commitment (e.g., Antonucci, Jackson and Biggs, 2007; Koropeckyj-Cox, 2002). On the one hand, the family social network constitutes an important resource to protect individuals’ health by, for example, reducing psychosocial stress and increasing one’s overall well-being (e.g., Antonucci et al., 2007; Franks, Campbell and Shields, 1992; also see Brandt et al., Chapter 15 in this volume). On the other hand, low relationship quality between, for example, parents and adult children (e.g., An and Cooney, 2006; Koropeckyj-Cox, 2002) or burdens associated with providing care to kin (e.g., Sherwood et al., 2005) have been shown to result in health deterioration, especially if psychological well-being is considered. The individual’s trust in the family network as a potential source of support and the emotional stability of family relations appear to be particularly important here. Some evidence suggests 188

Family relations and health inequalities: grandparents and grandchildren  189 that the subjective perception of support might even be more relevant for individuals’ health than the actual support one has (or has not) received (e.g., Antonucci, 2001; also see McIlvane, Ajrouch and Antonucci, 2007). Research in this field has largely focused on aspects of the parent–child relation, and in light of the longer shared lives of grandparents and grandchildren, it has extended to consider grandparent–grandchild relationships to explain variation in well-being through later life.1 The likelihood of being a (grand)parent and the timing of these transitions depend on reproductive behaviors (e.g., the age at first birth and the number of children), which have been shown to be associated with health and mortality (Henretta, 2007; Högnäs et al., 2017). Parenthood and grandparenthood, in both their roles and functions, indeed tend to vary across population subgroups, particularly by sex, generating or even reinforcing inequalities in health and well-being. Indeed, women generally have children younger and live longer than men, making them more likely than men to experience grandparenthood for greater durations. The remainder of this chapter is structured as follows: the second section provides a brief overview of health inequalities in relationships between older parents and adult children. The third section subsequently discusses health inequalities from the grandparents’ perspective, distinguishing between grandparenthood and grandparenting. The fourth section adds to the debate by considering health inequalities among grandchildren. Each section highlights the mechanisms linking the intergenerational family role considered and health before reviewing the relevant empirical evidence from Europe and the USA. In the following sections, we will consider both health (with operationalizations ranging from self-rated general health to specific physician-diagnosed conditions) and the broader concept of well-being (including for example, life satisfaction). The final section then concludes with policy recommendations and perspectives for future research.

ADULT PARENT–CHILD RELATIONSHIP QUALITY AND HEALTH INEQUALITIES A growing number of social science studies investigate the intergenerational reproduction of health inequalities in families, stressing the role of education as an important transmission channel (e.g., Halliday, Mazumder and Wong, 2020; Willson and Shuey, 2019). Another relevant factor in the “production” of health within families and across individuals’ life course is the quality of intergenerational relationships between parents and their adult children (Thomas, Liu and Umberson, 2017). Whereas emotional closeness and relationship quality strongly influence the well-being of parents and adult children (e.g., Lai et al., 2019; Merz et al., 2009a, b), geographic proximity and frequency of contact appear to be unrelated to older parents’ life satisfaction – at least if the exchange of instrumental support is controlled (e.g., Lai et al., 2019; Lowenstein, Katz and Gur‐Yaish, 2007). Findings regarding the exchange of intergenerational instrumental support are, however, mixed. Whereas a cross-sectional study by Scodellaro, Khlat and Jusot (2012) indicates a positive association of having received large financial transfers on adult children’s health, a longitudinal study by Ong, Nguyen and Kendall (2018) fails to provide systematic evidence of a causal effect of the receipt of such transfers on the younger generation’s health and well-being. Giving financial support to adult children has been shown to be associated with better mental health for older parents (Roll and Litwin, 2010), whereas the receipt of upward instrumental support may even reduce parents’

190  Handbook of health inequalities across the life course well-being (e.g., Merz and Consedine, 2009; Merz et al., 2009a; also see Bordone, 2015). This kind of adverse effect appears most likely in situations characterized by a strong imbalance in the exchange of intergenerational support (e.g., Lowenstein et al., 2007; Pillemer et al., 2007). These findings underscore the importance of distinguishing between emotional support (including closeness and relationship quality) on the one hand and instrumental support on the other hand. Health, however, is not only affected by characteristics of intergenerational family relations but is also a determinant of the latter (e.g., Mao et al., 2020): good health may be an important resource to provide intergenerational support, whereas poor health might often trigger the need to establish an exchange of instrumental and/or financial support between generations in a family. It is therefore surprising that thus far, only relatively little empirical research has been conducted that addresses this causal direction of the intergenerational relations and health nexus (but see Choi et al., 2015; Gilligan et al., 2017). Health outcomes are often merely treated as control variables in multivariate models, without much theoretical or thorough empirical consideration of possible underlying mechanisms. Longitudinal analyses in particular are scarce – despite the obvious relevance of the question as to what extent changes in the individual’s health status might affect various dimensions of intergenerational relations. Cross-sectional findings suggest that good (poor) health in both the parents’ and the children’s generation is positively (negatively) associated with reports of some dimensions of relationship quality (e.g., Rossi and Rossi, 1990; Stepniak, Suitor and Gilligan, 2021). A longitudinal study by Merz et al. (2009b) supports the notion that the observed cross-sectional associations indeed reflect a causal effect of health on the intergenerational relationship. Stress has been put forward as a possible explanation for this finding. Health deterioration causes stress in parents and children, which eventually affects relationship quality in negative ways (also see Kaufman and Uhlenberg, 1998). Whereas changes in parents’ health status appear to be unrelated to the frequency of contact with adult children (Ward, Deane and Spitze, 2014), studies have found the expected changes in residential proximity (Choi et al., 2015) and instrumental support. Those in poorer health – especially parents – are more likely to receive and less likely to provide help (e.g., Chan and Ermisch, 2012; also see Gilligan et al., 2017).

HEALTH INEQUALITIES FROM THE GRANDPARENTS’ PERSPECTIVE Recently, an increasing number of studies have concentrated on the associations between grandparenthood or grandparenting and grandparents’ health. Grandparenthood refers to the transition that most individuals today will experience in later life from not having grandchildren to having at least one grandchild (Margolis and Verdery, 2019) but also to the status that follows such a transition, that is, being a grandparent. This status is a central role for older people, rated as highly important even before experiencing it (Mahne and Motel-Klingebiel, 2012) and encompassing expectations and meaning (Thiele and Whelan, 2006). Although the role of grandparent may be enacted in different ways (Silverstein and Marenco, 2001), in answer to the recognized increasingly vital support of grandparents to their families and society by looking after grandchildren, a major strand of research on grandparents has largely focused on behavioral aspects of grandparenthood, that is, grandparenting. We, therefore, discuss grandparenthood and grandparenting separately below. Concerning the latter, grand-

Family relations and health inequalities: grandparents and grandchildren  191 parental childcare arrangements can be further distinguished as primary or secondary. Primary carers are grandparents who assume primary responsibility for raising a grandchild for multiple reasons, including teenage pregnancy, drug addiction, incarceration, or health problems (Choi, Sprang and Eslinger, 2016; Glaser et al., 2018; Hayslip, Fruhauf and Dolbin-MacNab, 2019; Pilkauskas and Dunifon, 2016). Secondary care refers to the most common type of care, complementary to parental care. Support from grandparents taking care of grandchildren without replacing the parenting functions of the middle generation can vary from regular to occasional, and its intensity can range from a few hours per year to several per day (Glaser et al., 2013; Herlofson and Hagestad, 2012). Grandparenthood and Health Inequalities The majority of individuals aged 50 and older in Europe, Canada, and the United States have faced the transition to grandparenthood (Glaser et al., 2013; Leopold and Skopek, 2015; Margolis, 2016). Such a high prevalence of grandparenthood and the important implications that such an event might have for the experience itself and for its intersection with other life events and roles call for a better understanding of the actual life transition of becoming a grandparent and the relationship between grandparental status and health. Although grandparents are, on average, more satisfied with their lives than grandchildless people, such a “grandparenthood effect” is mainly driven by the provision of grandchild care. Indeed, grandparents who never look after their grandchildren are less satisfied with their lives compared to their grandchildless counterparts (Arpino, Bordone and Balbo, 2018). The lack of change in health measures for grandparents tends to be more prominent than change, implying that if a grandparent effect exists, it is mainly due to their caregiving role. From an evolutionary perspective, however, the birth of a grandchild per se tends to improve individuals’ fitness (see Tanskanen and Danielsbacka, 2019 for a review), in part due to the positive emotions brought by becoming a grandparent (Thiele and Whelan, 2006). Grandparents might also benefit from a grandchild birth in view of its subsequent opportunities, such as interaction with, nurturing of, and passing on their knowledge to the grandchild. Furthermore, the transition into grandparenthood could also be considered an avenue for improving parenting or as an extension of the provision of support to one’s own child that enhances older people’s sense of purpose in life (Ellwardt, Hank and Mendes de Leon, 2021). These theoretical arguments suggesting positive effects of entering grandparenthood on older people’s well-being are supported by empirical studies from analyses of longitudinal data. Using two waves of a representative sample of the Dutch population, Kalmijn and De Graaf (2012) found that parents whose adult children transitioned into parenthood had lower depression. However, Sheppard and Monden (2019), using a fixed-effects approach on SHARE data from 15 European countries, highlighted inequalities in the effect of becoming grandparents for the first time on (fewer) depressive symptoms, with a beneficial effect only among women. Drawing on the same dataset, Di Gessa, Bordone and Arpino (2020) analyzed the associations between becoming a grandparent and three indicators of well-being (life satisfaction, positive affect, and depression) and found that the first transition matters, but only among women who become grandmothers via their daughter. This finding is in line with the grandmother hypothesis stating that the long postmenopausal lifespan of human females might have evolved to enable post-reproductive women to contribute to the fertility of their adult children and the survival of their grandchildren (Hawkes, 2003). In practice, the birth of

192  Handbook of health inequalities across the life course a grandchild may contribute to the inclusive fitness of older women, whereas older men can potentially have children until they die (Coall and Hertwig, 2011). At the proximate level, this difference may translate into sex differences, with the birth of a grandchild having a greater impact on grandmothers than grandfathers. Grandmothers might also be more strongly affected by the transitions into grandparenthood due to the gendered social prescriptiveness of grandmotherhood (Reitzes and Mutran, 2004) and the different tasks and responsibilities associated with it (Kaufman and Elder, 2003; Winefield and Air, 2010). Moreover, the presence of grandchildren is known to increase the frequency of contact with children, especially among grandmothers (e.g., Bordone, 2009), reducing their risk of isolation while strengthening their kin-keeping role within the family, which in turn contributes to increasing grandmothers’ well-being. The empirical evidence in this respect not only suggests a matrilineal advantage in the quality of parent–child bonds (Chan and Elder, 2000; Jamieson, Ribe and Warner, 2018) but also hints at the importance of the event itself (i.e., the birth of the (first) grandchild) for (short-term) well-being rather than the role associated with grandparenthood. Becoming a grandparent might also be associated with negative stereotypes of aging that in turn negatively affect grandparents’ well-being. Indirect evidence is provided by studies showing that while in general older people tend to feel younger than their chronological age, grandparenthood may function as an “age reminder”, increasing subjective age for younger grandparents compared to entering this role “on time” (Kaufman and Elder, 2003). Chronological age has been identified as a moderator of the association between being a grandparent and well-being. Both men and women in the younger age group tend to feel older if they have grandchildren, while the significant effect found for the older age group of women highlights a reversed association, with grandmothers feeling younger than grandchildless women (Bordone and Arpino, 2016). The likelihood and timing of grandparenthood intersects with educational differences as well (Skopek and Leopold, 2017), which are paralleled by educational disparities in adult health. Whereas the subjective importance of the grandparent role does not vary by social class (Mahne and Motel-Klingebiel, 2012), education matters for the types of activities done by grandparents with their grandchildren (King and Elder, 1998) and for their ability to cope with the stress associated with grandparenthood (Mahne and Huxhold, 2015). The null finding about the moderating effect of education in the association between grandparenthood and life satisfaction (Mahne and Huxhold, 2015 on Germany and Arpino et al., 2018 on Europe) suggests, however, that two opposite mechanisms might be at work. On the one hand, highly educated grandparents may engage in a larger variety of social activities independently of their grandchildren, thereby reducing the relative importance of the role of grandparenthood on well-being or even increasing the costs associated with it. On the other hand, higher education may allow a better use of family ties as a barrier against negative life events or stressors. As the grandparent role often extends across several decades, well into grandparents’ later life and grandchildren’s adulthood, when the provision of grandchild care is no longer part of the grandparent role, recent studies have investigated the long-term health consequences of grandparenthood, i.e., whether there are any survival benefits. The work by Christiansen (2014), however, suggested a significant survival disadvantage among grandfathers in Norway. Grandmothers also exhibited elevated mortality risks, but only if they were married, had four or more grandchildren, or made the transition to grandparenthood early in the life course (i.e., before age 50). More recently, using longitudinal data from 12 waves of the Health

Family relations and health inequalities: grandparents and grandchildren  193 and Retirement Study, Ellwardt et al. (2021) did not find statistically significant survival differences by grandparental status or heterogeneity across gender, Whites and non-Whites, or across different levels of educational attainment and work status in the USA. However, elevated risks of mortality were confirmed for grandmothers living with a partner, below 65 years old, or having a larger number of grandchildren compared to non-grandmothers. The authors explain the general finding of grandparental excess mortality along the lines of parental depletion models (Barclay and Kolk, 2019), according to which emotional or social stress associated with the (grand-)parent role may result in adverse health effects. The related heterogeneities may derive from the role strain resulting from gendered role expectations and obligations, making partnered, younger, less-educated grandmothers face contemporaneous overlap with other roles (e.g., spousal caregiving or employment) more often. In contrast, there was a survival advantage for widowed grandmothers over widowed non-grandmothers, possibly reflecting the role of grandchildren as a buffer to adverse events such as loss of a spouse. Grandparenting and Health Inequalities Although the definition of grandchild care can vary across studies and countries (Hank et al., 2018), a considerable body of work shows that around the globe, grandparents are significant providers of secondary grandchild care (Di Gessa, Zaninotto and Glaser, 2020; Grundy et al., 2012; Ko and Hank, 2014; Ku et al., 2013; Laughlin, 2013). In Europe, for instance, 58% of grandmothers and 49% of grandfathers looked after at least one of their grandchildren under the age of 16 (Hank and Buber, 2009), with 12% providing care almost daily or at least 15 hours a week (Di Gessa, Glaser and Tinker, 2016a). Numerous studies have investigated the impact of grandchild care provision on grandparents’ health and well-being in different societal contexts (ranging from Chile to the U.S., Europe, and China) and have considered both primary and secondary grandchild care (see Danielsbacka, Křenková and Tanskanen, 2022, for a recent systematic review). The starting point of most of these studies is that caring for grandchildren may have both positive and negative health effects. According to role enhancement theory, which suggests that occupying multiple roles may provide individuals with a sense of usefulness and competence, enhancing control and reinforcing meaning in later life, grandparents caring for their grandchildren may benefit from the emotional rewards and gratification stemming from this activity, and a sense of belonging, attachment and usefulness, which in turn may enhance health and life satisfaction (Grinstead et al., 2003). Moreover, it is plausible that grandparents providing childcare have stronger social ties with both grandchildren and their parents and are, therefore, likely to benefit from greater emotional, instrumental, and social family support, which may act to buffer the potential negative effects of caregiving and have a direct positive impact on health by promoting healthy behaviors (Hayslip, Blumenthal and Garner, 2015). Looking after grandchildren may also lead to grandparents maintaining or increasing their levels of physical activity and health behaviors, which in turn are associated with better physical health and well-being (Holmes and Joseph, 2011). Providing grandchild care, however, might also be demanding both physically and emotionally. Role strain theory postulates that multiple roles are associated with poor health outcomes because of the psychological and physical stressors caused by demanding and potentially competing role responsibilities. For instance, if an individual’s obligations exceed their physical and psychological capacity to cope, this situation may cause an increase in stress and physical demands, which in turn may be detrimental

194  Handbook of health inequalities across the life course for health. This problem may exist for those grandparents who act as primary carers or who provide full-time care for their grandchildren (e.g., Ellwardt et al., 2021). The effect of grandchild care on grandparents’ health is quite complex and seems to depend on a number of factors, including the type of care provided (primary vs. secondary), the health measure considered, the intensity and hours of care provided, the regional/cultural context, and grandparents’ sociodemographic characteristics (Arpino and Bordone, 2014; Chen and Liu, 2012; Di Gessa, Glaser and Tinker, 2016a, b; Hank et al., 2018; Tsai, Motamed and Rougemont, 2013). Evidence (largely from the US) suggests that grandparents raising grandchildren and those coresiding with them tend to report a higher prevalence of health problems, including limitations in daily living activities, chronic conditions, depressive symptoms, and poorer self-rated health (Blustein, Chan and Guanais, 2004; Grinstead et al., 2003; Hughes et al., 2007; Minkler and Fuller-Thomson, 2001). However, in some Asian countries, the experience of coresiding with grandchildren or being a custodial grandparent is associated with better rather than worse health and well-being (Cong and Silverstein, 2012; Silverstein, Cong and Li, 2006). Ku and colleagues (2013), for instance, found that coresident grandparents in Taiwan were more likely to report better self-rated health and fewer problems with physical health, whereas Chen and Liu (2012), using the longitudinal China Health and Nutrition Survey, found no differences in health between coresiding and non-coresiding grandparents, but coresiding grandparents who provided more than 15 hours per week of grandchild care were more likely to report worse self-rated health. Studies that have focused on secondary grandchild caregivers have generally found that grandparents providing grandchild care are more likely to report either better health (including better self-rated health, lower levels of loneliness, fewer depressive symptoms and better cognition) compared to grandparents with primary care responsibility for a grandchild or no childcare at all or no major widespread health effects once previous characteristics (and prior health status in particular) are considered (Arpino and Bordone, 2014; Chen and Liu, 2012; Di Gessa et al., 2016a, b; Ku et al., 2013; Tsai et al., 2013). For non-coresiding grandparents, however, the association between grandchild care provision and health might depend on the outcomes and on the intensity of care considered. For instance, Di Gessa et al. (2016a) found that grandparents looking after grandchildren, whether intensively or non-intensively, experienced better self-rated health but no beneficial effects in terms of depressive symptoms or limitations in daily living activities. Many studies have found a positive impact of grandparental childcare on mental health and well-being, mostly among grandparents providing lower intensity levels of grandparental care (Chen and Liu, 2012; Grundy et al., 2012; Ku et al., 2013). However, if the cognitive abilities of grandparents were considered, it is the grandparents who provide more care who experience greater increases in their cognition (Ahn and Choi, 2019; Arpino and Bordone, 2014; Silverstein and Zuo, 2020; Sneed and Schultz, 2019), even if the positive effect might be relevant to the verbal fluency of grandparents but not to other measures of cognitive function (Arpino and Bordone, 2014). It should, however, be noted that the abovementioned positive associations between grandchild care provision and grandparents’ health and well-being were based on between-person variations. This finding means that the health differential between grandparents who provide and those who do not provide grandchild care may not be due to grandparenting per se. Recent studies using fixed-effects models, thus drawing on within-person variation (Ates, 2017; Bordone and Arpino, 2022; Danielsbacka et al., 2019; Sheppard and Monden, 2019), found no beneficial effects of secondary grandparenting, pointing to a strong selection bias (with

Family relations and health inequalities: grandparents and grandchildren  195 healthier grandparents more likely to provide grandchild care from the outset) and highlighting the presence of unobservable third factors that might explain the associations between grandchild care provision and better health (see Gebel, Chapter 7 in this volume for a more general methodological discussion).

HEALTH INEQUALITIES FROM THE GRANDCHILDREN’S PERSPECTIVE Grandparents can influence grandchildren’s health in different ways (e.g., Delgado-Angulo et al., 2020). One mechanism is through social class and the intergenerational transmission of material resources, where grandparents need not necessarily be in direct contact with grandchildren (see Anderson, Sheppard and Monden, 2018, for a related discussion of grandparent effects on educational outcomes). Other mechanisms require more direct intergenerational interaction, where grandparents act as role models and provide childcare or emotional support, for example. Sound empirical evidence on grandparents’ influence on children’s health is, however, scarce. This lack of research seems unfortunate against the background of, first, grandparents’ important role as providers of care (see Bordone et al., 2021) and, second, an accumulating body of research stressing the role of childhood family structure and living conditions for individuals’ well-being across the life course (e.g., Brandt, Deindl and Hank, 2012; Gaydosh and Harris, 2018; also see Dekhtyar and Fors, Chapter 18 in this volume). Moreover, the results of previous studies investigating health inequalities from the grandchild’s perspective are fairly mixed, suggesting that grandparent effects, if they exist at all, are neither unambiguously good nor bad (see Pulgaron et al., 2016; Sadruddin et al., 2019, for reviews). For example, grandmothers were shown to play a potentially important role in grandchildren’s nutrition (e.g., Rogers, Bell and Mehta, 2019; also see Aubel, 2012), but some studies found grandmaternal involvement to be associated with higher risks of childhood obesity (e.g., Tanskanen, 2013), whereas others reported the reverse association (e.g., Lindberg et al., 2016). Both salutary and health-damaging effects are likely to be driven by grandparental care practices (e.g., Chambers et al., 2021), which have also been proposed as an explanation of the overall adverse impact of grandparents on their grandchildren’s cancer risk factors (Chambers et al., 2017). Studies investigating the role of grandparents in grandchildren’s psychological well-being found that greater cohesion with grandparents reduced depressive symptoms in adolescent/adult grandchildren, whereas more frequent contact increased symptoms (e.g., Moorman and Stokes, 2016; Ruiz and Silverstein, 2007). Previous research provided no indication, however, of a grandparent effect on grandchildren’s self-esteem (Ruiz and Silverstein, 2007) or risky health behaviors (Dunifon and Bajracharya, 2012), but coresidence with a grandparent was found to be associated with an increased risk of child mental health problems (externalizing or internalizing problems) in a European sample (Masfety et al., 2019). Recently, Tanksanen and Danielsbacka (2018) suggested that previously proposed positive “grandparent effects” might actually reflect “grandchild effects” in the sense that grandparents invest more resources (time or money) in healthier grandchildren performing better. Conversely, one could argue that grandparents may be more involved in relationships with less healthy grandchildren who are in greater need of support. Accordingly, the authors’ “within-child” (longitudinal) analysis does not indicate a causal association between grand-

196  Handbook of health inequalities across the life course parental investment (contact frequency or financial support) and grandchildren’s cognitive development or emotional and behavioral problems. Importantly, studies assessing the physical and socioemotional health and cognitive development of children raised by custodial grandparents generally indicate poorer health outcomes (Sadruddin et al., 2019: Table 2; also see Hayslip et al., 2014; Smith, Cichy and Montoro-Rodriquez, 2015, for example). However, these findings do not necessarily reflect causal effects of grandparents’ involvement with their grandchildren who are the most economically vulnerable. Previous research in this field indeed suffers from a variety of more general methodological and conceptual shortcomings (such as lack of longitudinal data and representative samples, inconsistent measurement of predictors and outcomes, or poorly identified theoretical mechanisms). Thus, more work seems necessary to better understand the causal role of grandparents in children’s health outcomes (see Sadruddin et al., 2019). Specifically, it would be important to further investigate the extent to which the observed relationships might be specific to particular family constellations (e.g., Jappens and Van Bavel, 2019; Krueger et al., 2015) or welfare state contexts (see Bordone et al., 2021).

CONCLUSIONS In aging societies, family relationships across more than two generations are becoming increasingly important, driven by changes in terms of low fertility and increasing life expectancy that result in an age structure for families described increasingly as a “beanpole” rather than a “pyramid” (Bengtson, 2001). In this chapter, we, therefore, discussed health inequalities in view of intergenerational relationships by reviewing parent–child relationships first and giving particular emphasis to grandparents and grandchildren. The research we reviewed in this chapter shows that there is a link between intergenerational relationships and the health of the family members involved. However, the direction of such an association can be both ways, with better intergenerational relationships promoting well-being, but also poor health being an explanatory factor for less intergenerational exchange (e.g., Mao et al., 2020). The most recent research drawing on panel data and relying on within-individual estimations has also hinted at possible unobservable third factors playing a role in the association between intergenerational relationships and health heterogeneities, especially when considering grandparental outcomes from the grandparent–grandchild relationship. Future research in this area should thus aim at disentangling grandparent–parent–grandchild relations by considering a multigenerational perspective across different institutional and cultural environments (Price et al., 2018). A better understanding of how the sociodemographic characteristics of all actors involved simultaneously interact with each other and with contextual characteristics might shed further light on the complex intergenerational relationships between family members. Moreover, it is likely that grandparent–parent–grandchild relations change over time as family members age (e.g., Silverstein, 2019). It is thus important to monitor these dynamics over longer periods of time. Such studies could further help inform family policies on how the roles that individuals hold in family life can be best supported throughout the life course.

Family relations and health inequalities: grandparents and grandchildren  197

NOTE 1.

A further perspective is that of the “sandwich generation” facing competing demands from (elderly) parents and (adult) children (e.g., Fingerman et al., 2010; Wiemers and Bianchi, 2015), which might eventually affect this generation’s health outcomes (see, for example, Brenna, 2021; McGarrigle, Cronin, and Kenny, 2014).

REFERENCES Ahn, T. and Choi, K.D. (2019) ‘Grandparent caregiving and cognitive functioning among older people: Evidence from Korea’, Review of Economics of the Household, 17, pp. 553–586. An, J.S. and Cooney, T.M. (2006) ‘Psychological well-being in mid to late life: The role of generativity development and parent–child relationships across the lifespan’, International Journal of Behavioral Development, 30, pp. 410–421. Anderson, L.R., Sheppard, P. and Monden, C.W.S. (2018) ‘Grandparent effects on educational outcomes: A systematic review’, Sociological Science, 5, pp. 114–142. Antonucci, T.C. (2001) ‘Social relations: An examination of social networks, social support and sense of control’, in Birren, J.E. and Schaie, K.W. (eds.) Handbook of the psychology of aging (5th edn). New York: Academic Press, pp. 427–453. Antonucci, T.C., Jackson, J.S. and Biggs, S. (2007) ‘Intergenerational relations: Theory, research, and policy’, Journal of Social Issues, 63, pp. 679–693. Arpino, B. and Bordone, V. (2014) ‘Does grandparenting pay off? The effect of child care on grandparents’ cognitive functioning’, Journal of Marriage and Family, 76(2), pp. 337–351. Doi: 10.1111/ Jomf.12096. Arpino, B., Bordone, V. and Balbo, N. (2018) ‘Grandparenting, education and subjective wellbeing of older Europeans’, European Journal of Ageing, 15(3), pp. 251–263. Doi: 10.1007/s10433-018-0467-2. Ates, M. (2017) ‘Does grandchild care influence grandparents’ self-rated health? Evidence from a fixed-effects approach’, Social Science & Medicine, 190, pp. 67–74. Aubel, J. (2012) ‘The role and influence of grandmothers on child nutrition: Culturally designated advisors and caregivers’, Maternal & Child Nutrition, 8, pp. 19–35. Barclay, K. and Kolk, M. (2019) ‘Parity and mortality: An examination of different explanatory mechanisms using data on biological and adoptive parents’, European Journal of Population, 35(1), pp. 63–85. Doi: 10.1007/s10680-018-9469-1. Bengtson, V. (2001) ‘Beyond the nuclear family: The increasing importance of multigenerational bonds’, Journal of Marriage and Family, 63, pp. 1–16. Blustein, J., Chan, S. and Guanais, F.C. (2004) ‘Elevated depressive symptoms among caregiving grandparents’, Health Services Research, 39(6), pp. 1671–1689. Doi: 10.1111/j.1475-6773.2004.00312.x. Bordone, V. (2009) ‘Contact and proximity of older people to their adult children: A comparison between Italy and Sweden’, Population, Space and Place, 15(4), pp. 359–380. Bordone, V. (2015) ‘Est modus in rebus: A longitudinal study of intergenerational solidarity and Locus of Control’, Ageing & Society, 35, pp. 1242–1277. Bordone, V. and Arpino, B. (2016) ‘Do grandchildren influence how old you feel?’, Journal of Aging and Health, 28(6), pp. 1055–1072. Doi: 10.1177/0898264315618920. Bordone, V. and Arpino, B. (2022) ‘Is there a rejuvenating effect of (grand)childcare? A longitudinal study on German data’, The Journals of Gerontology: Series B, 77(2), pp. 446–455. Bordone, V., Hank, K., Arpino, B. and Tomassini, C. (2021) ‘Childcare by grandparents in the context of welfare state policies’, in Daly, M. et al. (eds.) International handbook of family policy. A life-course perspective. Oxford: Oxford University Press. Brandt, M., Deindl, C. and Hank, K. (2012) ‘Tracing the origins of successful aging: The role of childhood conditions and social inequality in explaining later life health’, Social Science & Medicine, 74, pp. 1418–1425. Brenna, E. (2021) ‘Should I care for my mum or for my kid? Sandwich generation and depression burden in Italy’, Health Policy, 125, pp. 415–423. Doi: 10.1016/j.healthpol.2020.11.014

198  Handbook of health inequalities across the life course Chambers, S.A., Dobbie, F., Radley, A. and Rowa-Dewar, N. (2021) ‘Grandmothers’ care practices in areas of high deprivation of Scotland: The potential for health promotion’, Health Promotion International [advance online access]. Chambers, S.A., Rowa-Dewar, N., Radley, A. and Dobbie, F. (2017) ‘A systematic review of grandparents’ influence on grandchildren’s cancer risk factors’, PLoS ONE, 12, e0185420. Chan, C.G. and Elder, G.H., Jr. (2000) ‘Matrilineal advantage in grandchild-grandparent relations’, The Gerontologist, 40, pp. 179–190. Doi: 10.1093/geront/40.2.179. Chan, T.W. and Ermisch, J. (2012) ‘Intergenerational exchange of instrumental support: Dynamic evidence from the British Household Panel Survey’, University of Oxford, Sociology Working Papers 2011-05. Chen, F. and Liu, G. (2012) ‘The health implications of grandparents caring for grandchildren in China’, The Journals of Gerontology: Series B, 67(1), pp. 99–112. Doi: 10.1093/geronb/gbr132. Choi, H., Schoeni, R.F., Langa, K.M. and Heisler, M.M. (2015) ‘Older adults’ residential proximity to their children: Changes after cardiovascular disease’, The Journals of Gerontology: Series B, 70(6), pp. 995–1004. Choi, M., Sprang, G. and Eslinger, J.G. (2016) ‘Grandparents raising grandchildren: A synthetic review and theoretical model for interventions’, Family & Community Health, 39(2), pp. 120–128. Doi: 10.1097/fch.0000000000000097. Christiansen, S.G. (2014) ‘The association between grandparenthood and mortality’, Social Science & Medicine, 118, pp. 89–96. Doi: 10.1016/j.socscimed.2014.07.061. Coall, D.A. and Hertwig, R. (2011) ‘Grandparental investment: A relic of the past or a resource for the future?’, Psychological Science, 20, pp. 93–98. Doi:10.1177/0963721411403269. Cohen, S. (2004) ‘Social relationships and health’, The American Psychologist, 59, pp. 676–684. Cong, Z. and Silverstein, M. (2012) ‘Caring for grandchildren and intergenerational support in rural China: A gendered extended family perspective’, Ageing & Society, 32(3), pp. 425–450. Danielsbacka, M., Křenková, L. and Tanskanen, A.O. (2022) ‘Grandparenting, health, and well-being: A systematic literature review’, European Journal of Ageing [advance online access]. Danielsbacka, M., Tanskanen, A.O., Coall, D.A. and Jokela, M. (2019) ‘Grandparental childcare, health and well-being in Europe: A within-individual investigation of longitudinal data’, Social Science & Medicine, 230, pp. 194–203. Delgado-Angulo, E.K., Alsheri, R., Scambler, S. and Bernabé, E. (2020) ‘Is grandparents’ social class associated with adult grandchildren’s oral health? Analysis of two British cohorts’, Community Dentistry and Oral Epidemiology, 48, pp. 402–408. Di Gessa, G., Bordone, V. and Arpino, B. (2020) ‘Becoming a grandparent and its effect on well-being: The role of order of transitions, time, and gender’, The Journals of Gerontology: Series B, 75(10), pp. 2250–2262. Doi: 10.1093/geronb/gbz135. Di Gessa, G., Glaser, K. and Tinker, A. (2016a) ‘The health impact of intensive and nonintensive grandchild care in Europe: New evidence from SHARE’, The Journals of Gerontology: Series B, 71(5), pp. 867–879. Doi: 10.1093/geronb/gbv055. Di Gessa, G., Glaser, K. and Tinker, A. (2016b) ‘The impact of caring for grandchildren on the health of grandparents in Europe: A lifecourse approach’, Social Science & Medicine, 152, pp. 166–175. Doi: 10.1016/j.socscimed.2016.01.041. Di Gessa, G., Zaninotto, P. and Glaser, K. (2020) ‘Looking after grandchildren: Gender differences in “when,” “what,” and “why”: Evidence from the English Longitudinal Study of Ageing’, Demographic Research, 43(53), pp. 1545–1562. Doi: 10.4054/DemRes.2020.43.53. Dunifon, R. and Bajracharya, A. (2012) ‘The role of grandparents in the lives of youth’, Journal of Family Issues, 33, pp. 1168–1194. Ellwardt, L., Hank, K. and Mendes de Leon, C.F. (2021) ‘Grandparenthood and risk of mortality: Findings from the Health and Retirement Study’, Social Science & Medicine, 268, 113371. Fingerman, K.L., Pitzer, L.M., Chan, W., Birditt, K., Franks, M.M., and Zarit, S. (2010) ‘Who gets what and why? Help middle-aged adults provide to parents and grown children’, Journal of Gerontology: Social Sciences, 66B, pp. 87–98. Doi: 10-1093/geronb/gbq009 Franks, P., Campbell, T.L. and Shields, C.G. (1992) ‘Social relationships and health: The relative roles of family functioning and social support’, Social Science & Medicine, 34, pp. 779–788.

Family relations and health inequalities: grandparents and grandchildren  199 Gaydosh, L. and Harris, K.M. (2018) ‘Childhood family instability and young adult health’, Journal of Health and Social Behavior, 59, pp. 371–390. Gilligan, M., Suitor, J.J., Rurka, M., Con, G. and Pillemer, K. (2017) ‘Adult children’s serious health conditions and the flow of support between the generations’, The Gerontologist, 57(2), pp. 179–190. Glaser, K., Price, D., Di Gessa, G., Ribe Montserrat, E., Stuchbury, R. and Tinker, A. (2013) Grandparenting in Europe: Family policy and grandparents’ role in providing childcare. London: Grandparents Plus. Glaser, K., Stuchbury, R., Price, D., Di Gessa, G., Ribe, E., Tinker, A. (2018) ‘Trends in the prevalence of grandparents living with grandchild(ren) in selected European countries and the United States’, European Journal of Ageing, 15(3), pp. 237–250. Grinstead, L.N., Leder, S., Jensen, S. and Bond, L. (2003) ‘Review of research on the health of caregiving grandparents’, Journal of Advanced Nursing, 44(3), pp. 318–326. Doi: 10.1046/j.1365-2648.2 003.02807.x. Grundy, E., Albala, C. Allen, E., Dangour, A.D., Elbourne, D. and Uauy, R. (2012) ‘Grandparenting and psychosocial health among older Chileans: A longitudinal analysis’, Aging & Mental Health, 16(8), pp. 1047–1057. Doi: 10.1080/13607863.2012.692766. Halliday, T.J., Mazumder, B. and Wong, A. (2020) ‘The intergenerational transmission of health in the United States: A latent variables analysis’, Health Economics, 29, pp. 367–381. Hank, K. and Buber, I. (2009) ‘Grandparents caring for their grandchildren. Findings from the 2004 Survey of Health, Ageing, and Retirement in Europe’, Journal of Family Issues, 30(1), pp. 53–73. Doi: 10.1177/0192513x08322627. Hank, K., Cavrini, G., Di Gessa, G. and Tomassini, C. (2018) ‘What do we know about grandparents? Insights from current quantitative data and identification of future data needs’, European Journal of Ageing, 15(3), pp. 225–235. Doi: 10.1007/s10433-018-0468-1. Hawkes, K. (2003) ‘Grandmothers and the evolution of human longevity’, American Journal of Human Biology, 15, pp. 380–400. Hayslip, B., Blumenthal, H. and Garner, A. (2015) ‘Social support and grandparent caregiver health: One-year longitudinal findings for grandparents raising their grandchildren’, The Journals of Gerontology: Series B, 70(5), pp. 804–812. Doi: 10.1093/geronb/gbu165. Hayslip, B., Heidemaire, B. and Garner, A. (2014) ‘Health and grandparent−grandchild wellbeing: One-year longitudinal findings for custodial grandfamilies’, Journal of Aging and Health, 26, pp. 559–582. Hayslip, B., Jr, Fruhauf, C.A. and Dolbin-MacNab, M.L. (2019) ‘Grandparents raising grandchildren: What have we learned over the past decade?’, Gerontologist, 59(3), pp. e152–e163. Doi: 10.1093/ geront/gnx106. Henretta, J.C. (2007) ‘Early childbearing, marital status, and women’s health and mortality after age 50’, Journal of Health and Social Behavior, 48(3), pp. 254–266. Herlofson, K. and Hagestad, G.O. (2012) ‘Transformations in the role of grandparents across welfare states’, in Arber, S. and Timonen, V. (eds.) Contemporary grandparenting: changing family relationships in global contexts. Bristol: Policy Press. Högnäs, R.S., Roelfs, D.J., Shor, E., Moore, C. and Reece, T. (2017) ‘J-Curve? A meta-analysis and meta-regression of parity and parental mortality’, Population Research and Policy Review, 36(2), pp. 273–308. Holmes, W. and Joseph, J. (2011) ‘Social participation and healthy ageing: A neglected, significant protective factor for chronic non communicable conditions’, Globalization and Health, 7(1), 43. Doi: 10.1186/1744-8603-7-43. Hughes, M.E., Waite, L.J., LaPierre, T.A. and Luo, Y. (2007) ‘All in the family: The impact of caring for grandchildren on grandparents’ health’, The Journals of Gerontology: Series B, 62(2), pp. S108– S119. Doi: 10.1093/geronb/62.2.s108. Hünteler, B. and Hank, K. (2021) ‘Life course generational placements in the family system and individuals’ well-being and health in later life’ Paper presented at the 15th Conference of the European Sociological Association, Barcelona. Jamieson, L., Ribe, E. and Warner, P. (2018) ‘Outdated assumptions about maternal grandmothers? Gender and lineage in grandparent–grandchild relationships’, Contemporary Social Science, 13, pp. 261–274. Doi: 10.1080/21582041.2018.1433869.

200  Handbook of health inequalities across the life course Jappens, M. and Van Bavel, J. (2019) ‘Relationships with grandparents and grandchildren’s well-being after divorce’, European Sociological Review, 35, pp. 757–771. Kalmijn, M. and De Graaf, P.M. (2012) ‘Life course changes of children and well-being of parents’, Journal of Marriage and Family, 74, pp. 269–280. Doi: 10.1111/j.1741-3737.2012.00961.x. Kaufman, G. and Elder, G.H., Jr. (2003) ‘Grandparenting and age identity’, Journal of Aging Studies, 17, pp. 269–282. Kaufman, G. and Uhlenberg, P. (1998) ‘Effects of life course transitions on the quality of relationships between adult children and their parents’, Journal of Marriage and Family, 60, pp. 924–938. King, V. and Elder, G.H. (1998) ‘Education and grandparenting roles’, Research on Aging, 20(4), pp. 450–474. Doi: 10.1177/0164027598204004. Ko, P.-C. and Hank, K. (2014) ‘Grandparents caring for grandchildren in China and Korea: Findings from CHARLS and KLoSA’, The Journals of Gerontology: Series B, 69(4), pp. 646–651. Doi: 10.1093/geronb/gbt129. Koropeckyj-Cox, T. (2002) ‘Beyond parental status: Psychological well-being in middle and old age’, Journal of Marriage and Family, 64, pp. 957–971. Krueger, P.M., Jutte, D.P., Franzini, L., Elo, I. and Hayward, M.D. (2015) ‘Family structure and multiple domains of child well-being in the United States: A cross-sectional study’, Population Health Metrics, 13, 6. Ku, L.-J.E., Stearns, S.C., Van Houtven, C.H., Lee, S.-Y.D., Dilworth-Anderson, P. and Konrad, T.R. (2013) ‘Impact of caring for grandchildren on the health of grandparents in Taiwan’, The Journals of Gerontology: Series B, 68(6), pp. 1009–1021. Doi: 10.1093/geronb/gbt090. Lai, D.W.L., Lee, V.W.P., Li, J. and Dong, X. (2019) ‘The impact of intergenerational relationship quality on health and well-being of older Chinese Americans’, Journal of the American Geriatrics Society, 67, pp. S557–S563. Laughlin, L. (2013) Who’s minding the kids? Child care arrangements. Washington, DC: U.S. Census Bureau. Leopold, T. and Skopek, J. (2015) ‘The demography of grandparenthood: An international profile’, Social Forces, 94(2), pp. 801–832. Doi: 10.1093/sf/sov066. Lindberg, L., Ek, A., Nyman, J., Marcus, C., Ulijaszek, S. and Nowicka, P. (2016) ‘Low grandparental social support combined with low parental socioeconomic status is closely associated with obesity in preschool‐aged children: a pilot study’, Pediatric Obesity, 11, pp. 313–316. Lowenstein, A., Katz, R. and Gur‐Yaish, N. (2007) ‘Reciprocity in parent–child exchange and life satisfaction among the elderly: A cross‐national perspective’, Journal of Social Issues, 63, pp. 865–883. Mahne, K. and Huxhold, O. (2015) ‘Grandparenthood and subjective well-being: Moderating effects of educational level’, The Journals of Gerontology: Series B, 70(5), pp. 782–792. Doi: 10.1093/geronb/ gbu147. Mahne, K. and Motel-Klingebiel, A. (2012) ‘The importance of the grandparent role – a class specific phenomenon? Evidence from Germany’, Advances in Life Course Research, 17(3), pp. 145–155. Mao, W., Silverstein, M., Prindle, J.J. and Chi, I. (2020) ‘The reciprocal relationship between instrumental support from children and self-rated health among older adults over time in rural China’, Journal of Aging and Health, 32(10), pp. 1528–1537. Margolis, R. (2016) ‘The changing demography of grandparenthood’, Journal of Marriage and Family, 78, pp. 610–622. Doi:10.1111/ jomf.12286. Margolis, R. and Verdery, A.M. (2019) ‘A cohort perspective on the demography of grandparenthood: Past, present, and future changes in race and sex disparities in the United States’, Demography, 56(4), pp. 1495–1518. Masfety, V.K., Aarnink, C., Otten, R., Bitfoi, A., Mihova, Z., Lesinskiene, S., Carta, M.G., Goelitz, D. and Husky, M. (2019) ‘Three-generation households and child mental health in European countries’, Social Psychiatry and Psychiatric Epidemiology, 54, pp. 427–436. McGarrigle, C.A., Cronin, H. and Kenny, R.A. (2014) ‘The impact of being the intermediate caring generation and intergenerational transfers on self-reported health of women in Ireland’, International Journal of Public Health, 59, pp. 301–308. Doi: 10.1007/s00038-013-0521-y McIlvane, J.M., Ajrouch, K.J. and Antonucci, T.C. (2007) ‘Generational structure and social resources in mid‐life: Influences on health and well‐being’, Journal of Social Issues, 63, pp. 759–773.

Family relations and health inequalities: grandparents and grandchildren  201 Merz, E.-M. and Consedine, N.S. (2009) ‘The association of family support and wellbeing in later life depends on adult attachment style’, Attachment & Human Development, 11, pp. 203–221. Merz, E.-M., Consedine, N.S., Schulze, H.-J. and Schuengel, C. (2009a) ‘Wellbeing of adult children and ageing parents: Associations with intergenerational support and relationship quality’, Ageing & Society, 29, pp. 783–802. Merz, E.-M., Schuengel, C. and Schulze, H.-J. (2009b) ‘Intergenerational relations across 4 years: Well-being is affected by quality, not by support exchange’, The Gerontologist, 49, pp. 536–548. Minkler, M. and Fuller-Thomson, E. (2001) ‘Physical and mental health status of American grandparents providing extensive child care to their grandchildren’, Journal of the American Medical Women’s Association, 56(4), pp. 199–205. Moorman, S.M. and Stokes, J.E. (2016) ‘Solidarity in the grandparent–adult grandchild relationship and trajectories of depressive symptoms’, The Gerontologist, 56, pp. 408–420. Ong, R., Nguyen, T. and Kendall, G. (2018) ‘The impact of intergenerational financial transfers on health and wellbeing outcomes: A longitudinal study’, Social Science & Medicine, 214, pp. 179–186. Pilkauskas, N.V. and Dunifon, R.E. (2016) ‘Understanding grandfamilies: Characteristics of grandparents, nonresident parents, and children’, Journal of Marriage and Family, 78(3), pp. 623–633. Doi: 10.1111/jomf.12291. Pillemer, K., Suitor, J.J., Mock, S.E., Sabir, M., Pardo, T.B. and Sechrist, J. (2007) ‘Capturing the complexity of intergenerational relations: Exploring ambivalence within later‐life families’, Journal of Social Issues, 63, pp. 775–791. Price, D., Ribe, E., Glaser, K. and Di Gessa, G. (2018) ‘Grandparental childcare: A re-conceptualisation of family policy regimes’, in Timonen, V. (ed.) Grandparenting practices around the world. Bristol: Policy Press. Pulgaron, E.R., Marchante, A.N., Agosto, Y., Lebron, C.N. and Delamater, A.M. (2016) ‘Grandparent involvement and children’s health outcomes: The current state of the literature’, Families, Systems, and Health, 34, pp. 260–269. Reitzes, D.C. and Mutran, E.J. (2004) ‘Grandparenthood: Factors influencing frequency of grandparent– grandchildren contact and grandparent role satisfaction’, The Journals of Gerontology: Series B, 59, pp. S9–S16. Rogers, E., Bell, L. and Mehta, K. (2019) ‘Exploring the role of grandparents in the feeding of grandchildren aged 1–5 years’, Journal of Nutrition Education and Behavior, 51, pp. 300–306. Roll, A. and Litwin, H. (2010) ‘Intergenerational financial transfers and mental health: An analysis using SHARE-Israel data’, Aging & Mental Health, 14, pp. 203–210. Rossi, A.S. and Rossi, P.H. (1990) Of human bounding: Parent–child relations across the life course. New York: Aldine de Gruyter. Ruiz, S.A. and Silverstein, M. (2007) ‘Relationships with grandparents and the emotional well-being of late adolescent and young adult grandchildren’, Journal of Social Issues, 63, pp. 793–808. Sadruddin, A.F.A., Ponguta, L.A., Zonderman, A.L., Wiley, K.S., Grimshaw, A. and Panter-Brick, C. (2019) ‘How do grandparents influence child health and development? A systematic review’, Social Science & Medicine, 239, 112476. Scodellaro, C., Khlat, M. and Jusot, F. (2012) ‘Intergenerational financial transfers and health in a national sample from France’, Social Science & Medicine, 75, pp. 1296–1302. Sheppard, P. and Monden, C. (2019) ‘Becoming a first-time grandparent and subjective wellbeing: A fixed-effects approach’, Journal of Marriage and the Family, 81(4), pp. 1016–1026. Sherwood, P.R., Given, C.W., Given, B.A. and Von Eye, A. (2005) ‘Caregiver burden and depressive symptoms analysis of common outcomes in caregivers of elderly patients’, Journal of Aging and Health, 17, pp. 125–147. Silverstein, M. (2019) ‘Growing old and growing up: Grandparents and their adult grandchildren in the context of multigenerational families’, in Hayslip, B. and Fruhauf, C.A. (eds.) Grandparenting: Influences on the dynamics of family relationships. New York: Springer, pp. 81–94. Doi: 10.1891/9780826149855.0005. Silverstein, M., Cong, Z. and Li, S. (2006) ‘Intergenerational transfers and living arrangements of older people in rural China: Consequences for psychological well-being’, The Journals of Gerontology: Series B, 61(5), pp. S256–S266.

202  Handbook of health inequalities across the life course Silverstein, M., Giarrusso, R. and Bengtson, V.L. (2003) ‘Grandparents and grandchildren in family systems: A social-developmental perspective’, in Bengtson, V.L. and Lowenstein, A. (eds.) Global aging and challenges to families. New York: Aldine de Gruyter, pp. 75–103. Silverstein, M. and Marenco, A. (2001) ‘How Americans enact the grandparent role across the family life course’, Journal of Family Issues, 22(4), pp. 493–522. Doi: 10.1177/019251301022004006. Silverstein, M. and Zuo, D. (2021) ‘Grandparents caring for grandchildren in rural China: Consequences for emotional and cognitive health in later life’, Aging & Mental Health, 25(11), pp. 2042–2052. Skopek, J. and Leopold, T. (2017) ‘Who becomes a grandparent – and when? Educational differences in the chances and timing of grandparenthood’, Demographic Research, 37(29), pp. 917–928. Doi: 10.4054/DemRes.2017.37.29. Smith, G., Cichy, K. and Montoro-Rodriquez, J. (2015) ‘Impact of coping resources on the wellbeing of custodial grandmothers and grandchildren’, Family Relations, 64, pp. 378–392. Sneed, R.S. and Schulz, R. (2019) ‘Grandparent caregiving, race, and cognitive functioning in a population-based sample of older adults’, Journal of Aging and Health, 31(3), pp. 415–438. Stepniak, C., Suitor, J.J. and Gilligan, M. (2021) ‘Mothers’ functional limitations and relationship quality with adult children: Exploring the moderating roles of race and gender’, Research on Aging [advance online access]. Tanskanen, A.O. (2013) ‘The association between grandmaternal investment and early years overweight in the UK’, Evolutionary Psychology, 11, pp. 417–425. Tanskanen, A.O. and Danielsbacka, M. (2018) ‘Multigenerational effects on children’s cognitive and socioemotional outcomes: A within-child investigation’, Child Development, 89, pp. 1856–1870. Tanskanen, A.O. and Danielsbacka, M. (2019) Intergenerational family relations: An evolutionary social science approach. London: Routledge Advances in Sociology. Thiele, D.M. and Whelan, T.A. (2006) ‘The nature and dimensions of the grandparent role’, Marriage & Family Review, 40, pp. 93–108. Doi: 10.1300/J002v40n01_06. Thomas, P.A., Liu, H. and Umberson, D. (2017) ‘Family relationships and well-being’, Innovation in Aging, 1, pp. 1–11. Tsai, F.-J., Motamed, S. and Rougemont, A. (2013) ‘The protective effect of taking care of grandchildren on elders’ mental health? Associations between changing patterns of intergenerational exchanges and the reduction of elders’ loneliness and depression between 1993 and 2007 in Taiwan’, BMC Public Health, 13, 567. Doi: 10.1186/1471-2458-13-567. Ward, R., Deane, G. and Spitze, G. (2014) ‘Life-course changes and parent–adult child contact’, Research on Aging, 36, pp. 568–602. Wiemers, E.E. and Bianchi, S.M. (2015) ‘Competing demands from aging parents and adult children in two cohorts of American women’, Population and Development Review, 41, pp. 127–146. Doi: 10.1111/j.1728-4457.2015.00029.x Willson, A.E. and Shuey, K.M. (2019) ‘A longitudinal analysis of the intergenerational transmission of health inequality’, Journals of Gerontology: Social Sciences, 74, pp. 181–191. Winefield, H. and Air, T. (2010) ‘Grandparenting: Diversity in grandparent experiences and needs for healthcare and support’, International Journal of Evidence Based Healthcare, 8, pp. 277–283.

14. The effects of retirement on health and mortality by socio-economic group Matthias Giesecke

INTRODUCTION Retirement is one of the most incisive events in the life course. The transition to retirement implies a fundamental change of living conditions, especially for those who retire from active work. Several consequences arise with retirement, including a reduction of individual and household income and a substantial increase in leisure time. The disruptive nature of these changes also points at the potential impact that retirement has on individual health and longevity. This chapter reviews recent evidence on the effects of retirement on health and mortality. A particular emphasis is on differential effects in socio-economic subgroups, aiming to reveal health inequalities in late stages of the life course. The discussion covers a wide range of health outcomes covered in the literature, concentrating both on subjective outcomes such as self-rated health but also on objective outcomes such as diseases or mortality. The chapter also dedicates a substantive part to mental health and cognitive functioning, and points at the timing of effects, distinguishing immediate effects after retirement from long-run effects at older ages. While particularly focusing on the economic literature, the chapter also includes a range of relevant aspects from other disciplines such as epidemiology, gerontology, psychology, and public health. In an early contribution on the relationship between retirement and health, Minkler (1981) has pointed out that “The contention that retirement may have an adverse effect on health has become increasingly popular with the recent categorization of this phenomenon as a stressful life event. The small number of empirical studies examining the health outcomes of retirement, however, appear neither to support nor to refute this hypothesis.” Although the number of empirical studies has grown over the past 40 years, the findings are still largely heterogeneous and the subject matter remains somewhat controversial. Nevertheless, some clear patterns have been emerged from the literature on the retirement– health nexus. Despite the clear evidence on retirement-related cognitive decline corroborated by the so-called “use-it-or-lose-it” hypothesis,1 more recent research has shown that there seems to be heterogeneity regarding the age patterns linked to early or late retirement choices. It has been shown that people who make use of early retirement pathways seem to exhibit improving or at least non-detrimental effects on cognitive capacities, while for people who retire later in their life (or at statutory retirement ages) the adverse effects in terms of cognitive losses prevail. Such results have been documented by Celidoni et al. (2017) and, based on a more sophisticated methodology, by Schmitz and Westphal (2021). A similar pattern has been documented for mortality outcomes around age-based eligibility thresholds, showing mortality-reducing effects for retirees that make use of early retirement options and mortality-increasing effects for people claiming pensions at normal (or statu203

204  Handbook of health inequalities across the life course tory) retirement ages (Giesecke, 2019).2 The mechanism behind this is closely attached to socio-economic background variables and seems very similar across different health outcomes (cognition and mortality). While early retirees tend to be men working in manual jobs for whom retirement rather coincides with a relief from job-related stress, late retirees tend to be people with high earnings for whom retirement implies a loss of job-related networks and prestige.3 Technically speaking, if people who are successful in their jobs rather lack skills that complement well with the retirement-induced value of leisure, retirement can be consequential in terms of – for example – social isolation, which has been shown to fairly well predict mortality (see Pantell et al., 2013). Health effects of retirement can in fact be mediated through increasing leisure time. More leisure correlates with increasing leisure activities that themselves are associated with improved health in older ages (Chang et al., 2014). Moreover, social participation (e.g. number of social contacts, neighborhood activities, participation in sports organizations) can mitigate the adverse health effects of retirement (Shiba et al., 2017). Further evidence points at volunteering to coincide with a slower decline of health during old age, including positive effects on mental health (reduced depressive symptoms) and reduced mortality rates (Lum and Lightfoot, 2005). Overall, the existing literature largely supports the view that old age activities, which significantly increase when people retire and substitute previous work activity, tend to be health improving. Important policy questions posed by the retirement–health nexus have attracted attention not only among researchers but also in the general public. One issue in this regard is that common pay-as-you-go pension systems redistribute towards people who live longer and spend more time in retirement before they die, thus accumulating higher total pensions. This can be a considerable source of inequality, such that studying the impact that retirement has on health and longevity also involves questions on redistributive aspects of pension systems. Another policy-relevant question arising in this context is whether reforms that shift the retirement age create health externalities, positive or negative, for older people. Increasing labor force participation of older workers usually aims at ensuring the functioning of social security systems and thus old-age incomes. However, this may have side effects, so that empirical evidence documenting adverse health effects of retirement points at the possibility that upward shifts of the retirement age, usually as a response to demographic pressure, could reduce public health care costs. Identification of the health effects of retirement is particularly challenged by “endogenous” retirement choices. Given that retirement decisions are usually non-random, establishing a robust causal relationship with respect to the health effects of retirement requires some type of exogenous variation. Typical sources of exogenous variation are institutional peculiarities such as an age threshold for pension eligibility or reforms that shift the retirement age. This chapter briefly summarizes the different types of natural experiments used in the literature to identify the health effects of retirement in order to provide insights into how causal relationships can be established despite methodological issues. Regarding country choice, the focus is on several OECD countries such as Austria, Germany, Israel, Italy, Norway, Sweden, and the U.S. These countries all have in common that they provide large-scale social security data sets that make it possible to study health and mortality outcomes with reference to retirement. For this reason, most of what we know on the retirement–health relationship is still centered around these countries.

The effects of retirement on health and mortality by socio-economic group  205

THE EFFECTS OF RETIREMENT ON COGNITIVE ABILITIES Over decades, social scientists have intensely studied the formation of human capital and cognitive abilities. The mere interest in skill accumulation is driven by the attempt to understand the underlying mechanisms for increasing individual and societal productivity and wealth accumulation. An aspect that has yet been somewhat understudied, however, is the depreciation of cognitive abilities as the flip side that usually appears when people grow older. Over the entire life-cycle, cognitive abilities roughly follow a hump-shaped pattern that peaks around age 50 and declines thereafter (for more details on this empirical observation, see, for example, Mazzonna and Peracchi, 2018).4 McFadden (2008) has raised some crucial questions regarding the nexus between cognitive abilities and aging. When looking at the effects of retirement on health and cognitive functioning, the relevant point is whether human capital components depreciate as an exogenous consequence of aging or whether they can be influenced by individual behavior and activities, including work and retirement decisions. This also relates to the question whether depreciation of cognitive abilities is fully random or (at least partially) predictable. A recent strand of the literature sheds more light on these aspects. Although these studies differ in many details, such as data sources, methods, country, and institutional background, they do agree that retirement usually coincides with reduced activity, which itself leads to a phenomenon that can be summarized as cognitive decline (Rohwedder and Willis, 2010; Bonsang et al., 2012; Coe et al., 2012; Mazzonna and Peracchi, 2012, 2017; Atalay et al., 2019; Schmitz and Westphal, 2021). This literature has predominantly documented adverse effects of retirement on cognition. The general mechanism emphasized in most of these studies for explaining the detrimental effects of retirement is based on the so-called “use-it-or-lose-it” hypothesis, meaning that a reduction in (cognitive) activity can have detrimental effects on cognitive functioning. Since retirement is a disruptive event involving activity changes that – for most people – are associated with reduced activity, it seems plausible that retirement then also coincides with cognitive decline. Such a correlation is evident, for example, in Rohwedder and Willis (2010), who examine the relationship between cognitive performance and employment rates, in terms of the relative differences observed in older (60–64) versus younger (50–54) men. They employ aggregated data with cross-country variation indicating that older men at a given age (say 65) perform more strongly in cognitive tests in countries with higher employment rates at that age. From this finding they argue there is a causal link between retirement (relative to staying employed) and decreasing cognitive abilities. Rohwedder and Willis (2010) defend their causal interpretation by the fact that country-level differences in employment rates and thus in retirement behavior are driven by country-specific pension or tax policies that are valid instruments to the extent that they affect the outcome “cognitive ability” only through the endogenous retirement age and are otherwise exogenous. Another advantage of the country level variation in retirement-related policies is that reverse causality can largely be ruled out since it is extremely unlikely that age-related cognitive decline motivated the implementation of these policies. A recent paper that deserves some attention is the one by Schmitz and Westphal (2021), who combine two distinct empirical methods to estimate the effects of retirement on cognition. In particular, they nest marginal treatment effects estimation into an event study design to decompose two sources of heterogeneity: (1) retirement duration as time elapsed since retirement;

206  Handbook of health inequalities across the life course and (2) preferences for retirement in terms of individual choices for earlier or later retirement. Based on data for 18 European countries, they find that ten years after retirement, people lose almost 10% of their cognitive capacity (measured in a well-established word recall test) relative to age 50. The effects broadly follow a linear pattern over time spent in retirement, meaning that losses in cognitive functioning are small just after retirement and then increase steadily. This result is consistent with the “use-it-or-lose-it” hypothesis, suggesting that cognitive decline gets more severe once people have reduced their activity level. Second, based on marginal treatment effects estimation, they find no adverse effects on cognitive functioning for those people with a strong desire to retire as early as possible, but large losses of cognitive capacity (up to 20%) for those who prefer to retire at high ages or not at all. In line with the heterogenous effects on cognition that depend on preferences for the timing of retirement, Celidoni et al. (2017) show that early retirement pathways can improve cognition. While they do find adverse effects on cognitive functioning for people retiring at the statutory retirement age, consistent with the detrimental effects reported in most papers, their results suggest that early retirement options seem to induce positive effects on cognitive functioning in the short run and no negative effects (thus zero effects) in the long run. The explanation for this result seems to be connected to the socio-economic background of the two groups of retirees. Those who retire early are likely to be men in low-skilled jobs, while those who retire later frequently report to be satisfied with their income and have higher degrees of freedom at their workplace. In summary, when looking at socio-economic groups it seems that older people appear to be impacted more strongly by losses in cognitive functioning relative to younger people, at least when looking at aggregated cross-country differentials (Rohwedder and Willis, 2010). This result is driven by the coincidence of old-age and retirement status such that – according to the “use-it-or-lose-it” hypothesis – older people exhibit on average lower levels of cognitive activity. The detrimental effects of retirement on cognition tend to be more adverse for those who are less willing to retire (retiring at higher ages) in contrast to people with a strong desire for retirement (Schmitz and Westphal, 2021). One explanation for the correlation between cognitive decline and retirement preferences is the complementarity between skills and the retirement-induced value of leisure. In contrast to age-related differences, most of the studies have not found clear-cut gender differentials regarding cognitive functioning. One exception is the paper by Atalay et al. (2019), which argues that smaller adverse effects on cognition among women can be explained by their finding that women increase the time spent in mental activities after retirement.

THE EFFECTS OF RETIREMENT ON MENTAL HEALTH Researchers have investigated the association between mental health and retirement over decades with numerous studies deepening our understanding of this relationship (see e.g. Lindeboom et al., 2002; Dave et al., 2008; Westerlund et al., 2009; Jokela et al., 2010). However, the focus here is on research that provides causal evidence on the effects of retirement on mental health, such as, for example, Coe and Zamarro (2011), Eibich (2015) and Heller-Sahlgren (2017). One recent study that is exclusively devoted to “mental health” is by Heller-Sahlgren (2017), who examined data for ten European countries (based on SHARE). The definition

The effects of retirement on health and mortality by socio-economic group  207 of mental health follows the Euro-D scale that classifies depressive tendencies ranging from 0–12. An important feature of the paper is that the timing dimension of mental health effects is distinguished between short-term effects (about two years after retirement) and long-term effects (about six years after retirement). As key findings, the study documents no short-run effects of retirement on mental health whatsoever. Instead, it is demonstrated that retirement has considerable adverse long-run effects on mental health, increasing the Euro-D score by 1.5–2.4 points, which corresponds to an increase by 0.8–1.3 standard deviations. Another important finding is that there is no effect heterogeneity across socio-economic characteristics regarding gender, educational background, or occupational strain. In contrast, Eibich (2015) uses German survey data (SOEP) to estimate mental health effects of retirement (among other physical health outcomes). His measurement of mental health is based on 12 questions that cover several dimensions of emotional functioning (SF12). Responses are then converted into an “as-if-continuous” scale, ranging from zero to 100 (mean 50 and standard deviation of 10). He finds that retirement affects mental health by increasing the SF12 score by 0.25 standard deviations. Eibich (2015) also studies differential effects along several socio-economic characteristics, documenting that the mental health improvements are driven by people who retire at age 65 (and insignificant for early retirees at age 60). When looking at differences between men and women, however, mental health effects do not differ. Moreover, the results are quite homogenous across educational degrees. In summary, the effects of retirement on mental health are mixed and contradictory to the extent that the literature has documented both detrimental effects (e.g. Heller-Sahlgren, 2017) and health-improving effects (e.g. Eibich, 2015). Although these studies also describe some relevant channels, future research is needed to reconcile their findings. So far, there is no evidence for differential effects between men and women or across educational groups.

THE EFFECTS OF RETIREMENT ON PHYSICAL HEALTH AND MORTALITY General Health Measures and Physical Health Measuring physical health as an objective outcome is difficult, especially in surveys due to common misreporting and incomplete answers. More subjective measures, such as self-reported health typically surveyed on a five-point scale from “(1) very poor” to “(5) very good”, are easier to record because people are usually willing to provide answers and such measures have turned out to perform well in predicting mortality. Self-reported health status generally covers anything that people consider when thinking about their own subjective health status, including physical health and diseases but also mental health or general well-being. Due to these ambiguities, subjective health measures involve several methodological drawbacks and thus studies on the health effects of retirement have only rarely looked at self-reported health outcomes. One exception is the study by Coe and Zamarro (2011) who find that the probability of reporting poor health (measured as very bad, bad or fair health on a five-point scale) decreases by 35% when people retire, but this effect seems to vanish after a while according to the authors. Constructing a health index that combines self-reported health with more objective health outcomes (like diseases), they still find health-improving effects of retirement by one

208  Handbook of health inequalities across the life course standard deviation in the index. Overall, this study is one of the few examples that report retirement-related health improvements. In line with the principle finding of the health-enhancing effects of retirement is more recent evidence by Hessel (2016), who examines health effects of retirement in more detail regarding socio-economic groups. Using three types of self-reported health (1 – general health, 2 – morbidity, 3 – activity limitations) that are included in the EU SILC data on 12 European countries, he finds that the probability of reporting poor health reduces significantly when people retire and further documents retirement-related health improvements equally for both men and women and in all educational subgroups. More explicit physical health outcomes have been studied by Behncke (2012). Based on UK ELSA survey data and using the state pension age as an instrumental variable, she finds that retirement significantly increases the risk of cardiovascular diseases and cancer, which is further corroborated by increased risk factors such as BMI, cholesterol, and blood pressure. The results of this study further point to increased problems in physical activity such as walking. Although she estimates the health effects of retirement for different sets of control variables (finding quite stable health effects across various specifications), Behncke (2012) does not explicitly report differential effects across socio-economic groups. In summary, studies that are based on self-reported health measures tend to find health-enhancing effects of retirement or document a reduced probability of reporting poor health. It must be kept in mind, however, that self-reported health is somewhat ambiguous regarding the health components addressed. Evidence on more specific physical health outcomes point to a significant increase in cardiovascular diseases and cancer, caused by retirement. None of the studies dealing with physical health outcomes report systematic differences in health effects across socio-economic groups. Mortality Although self-reported health is easy to document and widely available in surveys, using such measures as health outcome also involves some disadvantages. In particular, subjective conjectures of individuals about their own health are quite diverse and include a whole range of health components so that it remains somewhat unclear what individuals mean by their own health status. Using objective health outcomes, such as mortality, can circumvent some of the issues related to more subjective health measures. In the past one or two decades, administrative data and particularly death records have been made more easily available to researchers and hence a growing body of literature has examined the mortality effects of retirement. Measuring health in terms of mortality outcomes is interesting in its own right but also features some distinct economic insights. Probably the most important aspect is that mortality is closely linked to redistribution. In common pay-as-you-go pension systems, two persons who are identical with respect to retirement age and contributions (and thus pension claims) will receive different total pensions if they differ by age at death. This difference is simply due to the fact that lower total pensions accumulate when people receive pensions for a shorter period before they die. Consequently, pay-as-you-go pension systems redistribute from people with short lifetimes to people with long lifetimes. The evidence on the mortality effects of retirement is quite mixed. This makes it impossible to conclude one uniform direction of effects, positive or negative, that retirement has on mortality. Summarizing the existing studies allows us to contextualize the estimates and to classify

The effects of retirement on health and mortality by socio-economic group  209 these studies in a meaningful way. A natural starting point is to sort the previous literature by the direction of effects. In what follows, studies are thus grouped by mortality-increasing effects (positive effects), mortality-decreasing effects (negative effects), and zero effects. Mortality-reducing effects Mortality-reducing effects of retirement have been documented in several recent papers. Both the studies by Hallberg et al. (2015) and Bloemen et al. (2017) examine mortality outcomes based on early retirement programs for public sector workers. This is worthwhile mentioning to the extent that, compared to population averages, public sector workers tend to be a positively selected group. Hallberg et al. (2015) document that for Swedish army officers the mortality risk is reduced through an early retirement offer shifting the retirement age downward from 60 to 55. Very similarly, Bloemen et al. (2017) find that the probability of dying within five years is reduced considerably among Dutch civil servants who are offered the option of retiring early by up to ten years (55 relative to the normal retirement age of 65). While Bloemen et al. (2017) measure mortality within five years after retirement, Hallberg et al. (2015) take a much longer perspective and document mortality within up to 15 years after retirement. In both studies the actual retirement age is about 57 in the treatment group and thus relatively low compared to other studies. Looking at immediate mortality effects within 12 months after retirement, Giesecke (2019) also finds mortality reductions among men who take an early retirement option at the age of 63. Saporta-Eksten et al. (2021) also document mortality reductions as a consequence of retirement. Their empirical framework builds on the so-called “housewives reform” in Israel that shifted pension benefits from husbands to their non-working wives and thereby reduced the implicit tax on the employment of older husbands aged 65–70.5 This incentivized older husbands to delay retirement, meaning that the housewives reform induced a considerable drop in retirement probabilities of older husbands. Saporta-Eksten et al. (2021) then estimate the effects of employment (i.e. retirement) on health and – in particular – longevity by using the housewives reform to instrument for employment. Their results imply that delaying retirement by one year (one additional year of employment) reduces life expectancy of older husbands by up to one year. When relating reduced longevity to the annual earnings gains of older husbands from delayed retirement, the conclusion is that the value attributed to an additional year of life seems low. Since these life expectancy reducing effects are detectable only in the long run about 10–20 years after retirement, the result further indicates that people may not take into account the true costs of delaying retirement (here: losing one year of their lifetimes on average) at the time of making the retirement decision. Zero effects There are also several studies that document zero effects of retirement on mortality. Coe and Lindeboom (2008) use unintended early retirement offers in the U.S. to instrument retirement choices based on U.S. HRS data and find no mortality effects of retirement six years after retirement. Hernaes et al. (2013) examine a reduction of the early retirement age from 65 to 62 in Norway, measuring zero mortality effects for more than ten years after people retire. Early retirement eligibility is used as an instrument for the actual retirement age to isolate exogenous variation within retirement choices. The retirement program under study covers the entire Norwegian public sector and private sector firms that participate in the program voluntarily, with a treatment group that is more female, more educated and tends to work in white-collar

210  Handbook of health inequalities across the life course occupations more frequently than the control group. It is noteworthy here that the actual retirement age in their treatment group ranges between 63.7 and 66.3 and is relatively high compared to other studies. Another recent study that is consistent with these findings is the one by Boizo et al. (2021), who find that effectively raising the pension-claiming age in France (by raising the contribution length required to obtain a full pension) did not significantly impact the probability of dying among people above age 65. Mortality-increasing effects Mortality-increasing effects of retirement are documented by Fitzpatrick and Moore (2018), who use a regression discontinuity design at the age-based eligibility threshold for U.S. Social Security at the age of 62. They find mortality increases immediately after retirement (within 12 months in their baseline), and these effects are the strongest for white unmarried males who did not complete high school. Kuhn et al. (2020) investigate a specific population of older male workers who live in Austrian regions that are eligible for an early labor force exit, finding that these workers have a considerably higher risk of premature death compared to workers in non-eligible regions. The actual retirement age among men is between 57 and 58 and mortality is measured up to age 72, thus spanning a period of up to 15 years after retirement. Using a research design very similar to Fitzpatrick and Moore (2018), Giesecke (2019) also finds increasing mortality (around 2–3%) for older German workers who retire at the statutory retirement age of 65. These effects are very similar for men and women. Interestingly, these effects are measured in a group that tends to retire later than average and has above average earnings. There is, however, no clear-cut evidence that the mortality effects of retirement are driven by one specific socio-economic indicator or the time-span observed after retirement. The only exception is the actual retirement age, regarding which the mortality effects of retirement tend to be rather non-decreasing (i.e. zero or increasing) when retirees are older. Although Kuhn et al. (2020) mark an exception, mortality-increasing effects tend to be documented for older groups, such as in Hernaes et al. (2013) (actual retirement age 64–66), whose estimates are positive for men across specifications (though insignificant), Fitzpatrick and Moore (2018) (age 62), and Giesecke (2019) (age 65).

NATURAL EXPERIMENTS AND IDENTIFICATION ISSUES Researchers are often interested in the causal effects of “x” on “y”. Such causal relationships are, however, not always easy to identify or to distinguish from mere correlations. When looking at the health effects of retirement, the challenge is to separate the causal impact of retirement itself from numerous other sources of (adverse) health effects such as – for example – aging. This exercise is particularly difficult, because retirement naturally coincides with aging so that health impairments can emerge independently from retirement status as people grow older. Identifying health effects of retirement is almost impossible when the corresponding analysis is only based on non-experimental data, even if these are high-quality surveys or administrative data.6 The reason for this is that several endogeneity issues arise. One severe problem could be reverse causality in cases where bad health drives retirement, since people in poor health may retire earlier. This issue can also be described as selectivity in a sense that people

The effects of retirement on health and mortality by socio-economic group  211 select into retirement based on their individual preferences for consumption and leisure. For example, manual workers and people in physically demanding jobs tend to retire earlier to relieve themselves from harsh occupations and to enjoy leisure, but – at the same time – have a higher likelihood of suffering from job-related health impairments (Giesecke, 2018). If one were to ignore this, the researcher would falsely conclude that retirement causes poor health. Nevertheless, there exist approaches that provide convincing identification of causal relationships using different sources of exogenous variation from “natural experiments” (also termed “quasi-experiments”). These experiments appear in situations that involve some type of randomization by chance. Such “accidental” randomization can happen in settings when (1) actual reforms lead to policy changes that affect some but not others or (2) whenever institutional peculiarities induce randomization even in existing social security systems (for example at age-based eligibility thresholds). In what follows, I briefly summarize some examples of these two types of natural experiments that have been used in the literature to establish causality regarding the retirement–health nexus. The first type of natural experiments appear in new policies that involve changes inducing random variation, especially if they are largely unforeseen and if they cannot be influenced or manipulated by the individual. In principle, one could think of any policy that incentivizes changes in retirement behavior. The most direct variant are pension reforms that shift the retirement age and thus induce at least partially exogenous variation in the retirement age that would otherwise be chosen endogenously. One example in the literature on the retirement–health nexus is the study by Hernaes et al. (2013), which uses variation from a Norwegian pension reform that gradually reduced the retirement age for a specified group of workers from 65 to 62. They use early retirement eligibility as an instrumental variable for the actual retirement age. The instrumental variable framework is then embedded in a difference-in-differences setting, separating the treatment group eligible for the early retirement program from a control group of workers whose retirement age remained constant throughout the time under study. Then they compare mortality outcomes for these two groups before and after the changes to retirement ages were implemented.7 Another interesting policy change for identifying longevity effects of retirement is the study by Saporta-Eksten et al. (2021) that differs to some extent from typical age shifts in retirement policies. Their setting is based on a reform in Israel that shifted pension claims from husbands to their non-working wives and thereby created incentives for husbands to extend their working careers. The reason for this is that shifting pension claims from husbands to their wives reduces the implicit tax on working another year for the respective husband. Since the reform was implemented sharply between two birth cohorts (1931: affected, 1930: unaffected), it generates exogenous variation that can be exploited in a regression discontinuity design.8 The second type of natural experiments includes institutional peculiarities within existing social security systems that induce random assignment. This is the case, for example, when people can qualify for pension eligibility at existing age thresholds. Using age-based eligibility thresholds has been used extensively within instrumental variable (IV) approaches – for example in the literature on the effects of retirement on cognition that usually use variation in country-specific early and normal retirement ages to instrument otherwise endogenous choices of the retirement age (see e.g. Coe and Zamarro, 2011; Mazzonna and Peracchi, 2012, 2017; Celidoni et al., 2017; Schmitz and Westphal, 2021). When looking at only one specific age threshold, say within a specific country, eligibility thresholds are also suitable for implementing regression discontinuity designs (RD), as is done, for example, in Fitzpatrick and Moore

212  Handbook of health inequalities across the life course (2018) or Giesecke (2019).9 From the viewpoint of the (quasi-)experimental design, these two approaches (IV and RD) are very similar in a sense that random assignment comes from the fact that treatment or the probability of treatment (retirement) jumps discontinuously only because people cross a certain age threshold that – by chance – defines the early or normal retirement age.

CONCLUSIONS The relationship between retirement and health is complex and empirical research on related questions has produced a whole range of different results. When looking at the health and mortality effects of retirement, there is considerable heterogeneity of the results reported up to the present time. A superordinate mechanism to explain these mixed and partially controversial results is largely unavailable, but what dominates the health effects of retirement – positive or negative – seems to be driven by the labor market biography and career that lies behind the individual when retiring. This appears plausible to the extent that individual working histories largely determine how activities and social networks change in the moment of retirement. Based on that view, retirement tends to have adverse health effects on the elderly whenever working is good for them and avoids social isolation. At the same time, positive health effects of retirement dominate whenever working is bad or the joy of leisure is high. The literature has – so far – not documented important differentials of the impact that retirement has on health along the lines of socio-economic groups such as gender or education. Future research will have to provide robust empirical evidence on these relationships in order to better reconcile the diverse empirical findings.

NOTES 1. 2.

3. 4. 5.

6.

The term “use-it-or-lose-it” has been used, for example, by Rohwedder and Willis (2010). The idea behind this is that an individual reduction in cognitive activity can induce cognitive decline. Both directions of effects have been documented in numerous studies, including evidence on mortality-reducing effects of retirement (e.g. Hallberg et al., 2015, Bloemen et al., 2017; Giesecke, 2019) and on mortality-increasing effects of retirement (e.g. Fitzpatrick and Moore, 2018; Giesecke, 2019; Kuhn et al., 2020). This does not preclude that both early and late retirement may either be forced or voluntary. For an overview on changes in cognitive functioning over age, see Glisky (2007). For further discussion on the impact of age on cognition, see Deary et al. (2009). The reform intended to abolish discrimination against women in Israel. Before 1996, non-working wives were ineligible for social security retirement benefits. Instead, married men received a supplemental payment to their pension benefits that was meant to cover their wives needs if she was older than 45, not working, and not eligible for benefits. This principle did not hold the other way around (wives receiving supplemental payments for their non-eligible husbands) and therefore implied gender discrimination that was removed by the “housewives reform”. Some empirical strategies for establishing causal relationships that rely on non-experimental data without exogenous variation are rare in the literature on the health effects of retirement. One example is propensity score matching, which essentially compares “statistical twins” based on observable characteristics (say comparing the health of a person who is retired to an otherwise identical person who is not retired), but does not allow for tracing out unobserved heterogeneity.

The effects of retirement on health and mortality by socio-economic group  213 7. Hernaes et al. (2013) find a positive correlation between retirement and mortality but using early retirement eligibility as in instrument for actual retirement shows that this relationship is not a causal one. According to their findings, retirement does not have a direct effect on mortality. 8. Using this sharp discontinuity as an instrument for (otherwise endogenous) employment decisions, Saporta-Eksten et al. (2021) find that working an additional year reduces longevity and that the value of an additional year of life seems low when evaluated against the earnings gains from retiring later. 9. Using age-based eligibility thresholds to identify changes in health outcomes has been done in earlier seminal papers such as in Card et al. (2008, 2009), examining, for example, U.S. Medicaid.

REFERENCES Atalay, K., G. F. Barrett, and A. Staneva, (2019). The effect of retirement on elderly cognitive functioning. Journal of Health Economics 66, 37–53. Behncke, S. (2012). Does retirement trigger ill health? Health Economics 21, 282–300. Bloemen, H., S. Hochguertel, and J. Zweerink (2017). The causal effect of retirement on mortality: Evidence from targeted incentives to retire early. Health Economics 26(12), 204–218. Boizo, A., C. Garrouste, and E. Perdrix (2021). Impact of later retirement on mortality: Evidence from France. Health Economics 30(5), 1178–1199. Bonsang, E., S. Adam, and S. Perelman (2012). Does retirement affect cognitive functioning? Journal of Health Economics 31(3), 490–501. Card, D., C. Dobkin, and N. Maestas (2008). The impact of nearly universal insurance coverage on health care utilization: Evidence from Medicare. American Economic Review 98(5), 2242–58. Card, D., C. Dobkin, and N. Maestas (2009). Does Medicare save lives? Quarterly Journal of Economics 124(2), 597–636. Celidoni, M., C. D. Bianco, and G. Weber (2017). Retirement and cognitive decline: A longitudinal analysis using SHARE data. Journal of Health Economics 56, 113–125. Chang, P. J., L. Wray, & Y. Lin, (2014). Social relationships, leisure activity, and health in older adults. Health Psychology 33(6), 516–523. Coe, N. B., and Lindeboom, M. (2008). Does retirement kill you? Evidence from early retirement windows. IZA Discussion Paper No. 3817. Coe, N. B. and G. Zamarro (2011). Retirement effects on health in Europe. Journal of Health Economics 30(1), 77–86. Coe, N. B., H. von Gaudecker, M. Lindeboom, and J. Maurer (2012). The effect of retirement on cognitive functioning. Health Economics 21(8), 913–927. Dave, D., I. Rashad, and J. Spasojevic (2008). The effects of retirement on physical and mental health outcomes. Southern Economic Journal 75(2), 497–523. Deary, I. J., J. Corley, A. J. Gow, S. E. Harris, L. M. Houlihan, R. E. Marioni, L. Penke, S. B. Rafnsson, and J. M. Starr (2009). Age-associated cognitive decline, British Medical Bulletin 92 (1), 135–152. Eibich, P. (2015). Understanding the effect of retirement on health: Mechanisms and heterogeneity. Journal of Health Economics 43, 1–12. Fitzpatrick, M. D. and T. J. Moore (2018). The mortality effects of retirement: Evidence from social security eligibility at age 62. Journal of Public Economics 157, 121–137. Giesecke, M. (2018). The effect of benefit reductions on the retirement age: The heterogeneous response of manual and non-manual workers. Review of Income and Wealth 64(1), 213–238. Giesecke, M. (2019). The retirement mortality puzzle: Evidence from a Regression Discontinuity Design. Ruhr Economic Papers #800, RWI. Glisky, E. L. (2007). Changes in cognitive function in human aging. In: Brain Aging: Models, Methods, and Mechanisms (edited by D. R. Riddle). CRC Press, Taylor & Francis, Chapter 1. Hallberg, D., P. Johansson, and M. Josephson (2015). Is an early retirement offer good for your health? Quasi-experimental evidence from the army. Journal of Health Economics 32(3), 586–598. Heller-Sahlgren, G. (2017). Retirement blues. Journal of Health Economics 54, 66–78.

214  Handbook of health inequalities across the life course Hernaes, E., S. Markussen, J. Piggott, and O. L. Vestad (2013). Does retirement age impact mortality? Journal of Health Economics 32(3), 586–598. Hessel, P. (2016). Does retirement (really) lead to worse health among European men and women across all educational levels? Social Science & Medicine 151, 19–26. Jokela, M., J. E. Ferrie, D. Gimeno, T. Chandola, M. J. Shipley, J. Head, J. Vahtera, H. Westerlund, M. G. Marmot, and M. Kivimäki (2010). From midlife to early old age: Health trajectories associated with retirement. Epidemiology 21(3), 284–290. Kuhn, A., S. Staubli, J.-P. Wuellrich, and J. Zweimüller (2020). Fatal Attraction? Extended unemployment benefits, labor force exits, and mortality. Journal of Public Economics 191, 1–19. Lindeboom, M., F. Portrait, and G. J. van den Berg (2002). An econometric analysis of the mental-health effects of major events in the life of older individuals. Health Economics 11, 505–520. Lum, T. Y. and E. Lightfoot (2005). The effects of volunteering on the physical and mental health of older people. Research on Aging 27(1), 31–55. Mazzonna, F. and F. Peracchi (2012). Ageing, cognitive abilities and retirement. European Economic Review 56(4), 691–710. Mazzonna, F. and F. Peracchi (2017). Unhealthy retirement? Journal of Human Resources 52(1), 128–151. Mazzonna, F. and F. Peracchi (2018). The economics of cognitive aging. In Oxford Research Encyclopedia of Economics and Finance. Oxford University Press. McFadden, D. (2008). Human capital accumulation and depreciation. Applied Economic Perspectives and Policy 30(3), 379–385. Minkler, M. (1981). Research on the health effects of retirement: An uncertain legacy. Journal of Health and Social Behaviour 22(2), 117–130. Pantell, M., D. Rehkopf, D. Jutte, S. L. Syme, J. Balmes, and N. Adler (2013). Social isolation: A predictor of mortality comparable to traditional clinical risk factors. American Journal of Public Health 103(11), 2056–2062. Rohwedder, S. and R. J. Willis (2010). Mental retirement. Journal of Economic Perspectives 24(1), 119–38. Saporta-Eksten, I., I. Shurtz and S. Weisburd (2021). Social security, labor supply and health of older workers: Quasi-experimental evidence from a large reform. Journal of the European Economic Association 19(4), 2168–2208. Schmitz, H. and M. Westphal (2021). The Dynamic and Heterogeneous Effects of Retirement on Cognitive Decline. Unpublished Manuscript. Shiba, K., N. Kondo, and K. Kondo (2017). Retirement and mental health: Does social participation mitigate the association? A fixed-effects longitudinal analysis. BMC Public Health 17, 526 (2017). Westerlund, H., M. Kivimäki, A. Singh-Manoux, M. Melchoir, J. E. Ferrie, J. Pentti, M. Jokela, C. Leineweber, M. Goldberg, M. Zins, and J. Vahtera (2009). Self-rated health before and after retirement in France (GAZEL): A cohort study. The Lancet 374, 1889–1896.

15. Health inequalities in older age: the role of socioeconomic resources and social networks in context Martina Brandt, Nekehia T. Quashie and Alina Schmitz

INTRODUCTION Due to declining fertility rates and rising life expectancies, the share of persons aged 65 or over has nearly doubled across OECD countries: from less than 9 percent in 1960 to more than 17 percent in 2017. By 2050, the share of older persons is expected to increase to near 27 percent (OECD Health Statistics, 2019). While many individuals maintain good health and functioning until old age, the ageing process goes along with an increased risk of several chronic diseases, disability and mortality. Thus, the famous “compression of morbidity” (Fries, 1980) in “times of broken limits to life expectancy” (Oeppen and Vaupel, 2002) seems to hold only for specific morbidity dimensions in certain groups and contexts (Parker and Thorslund 2007; Crimmins and Beltrán-Sánchez, 2010). It is therefore crucial to identify conditions that promote “active”, “healthy” or even “productive” and “successful” ageing for all individuals (Rowe and Kahn, 1997). Still, comparative research on health inequalities among older adults remains sparse in comparison to younger and middle aged adults. This may be partly due to past assumptions of homogeneity in the older population and several methodological challenges in studying health inequalities across the life course in different contexts (Grundy and Holt, 2001; Brandt, Kaschowitz and Lazarevič, 2016; Schmitz, 2019). In this chapter, we describe the state of knowledge concerning health inequalities in the second half of life from a sociological perspective and focus on methodological challenges in comparative research on ageing. After describing general theoretical assumptions about the links between social and health inequality, we give special attention to the role of socioeconomic resources as well as social networks and family relationships for inequalities in the well-being of older adults in different contexts. We close with critical remarks about methodological limitations and open questions for future research on health inequalities in the second half of life.

SOCIAL INEQUALITY AND HEALTH INEQUALITY: CONCEPTUAL REMARKS The term “health inequalities” is often used to describe social gradients in health. This is, however, an abridged representation of the various links between differences in socioeconomic resources and health as, for example, represented in the model introduced by Mielck (2000) shown in Figure 15.1.

215

216  Handbook of health inequalities across the life course

Source: Jansßsen, Sauter and Kowalski (2012).

Figure 15.1

Links between social and health inequality

This model also includes pathways, representing different mechanisms explaining social inequalities in health. Broadly, materialist explanations highlight the importance of financial resources, employment and housing conditions, as well as access to goods, services and health care. According to the behavioural explanation, health inequalities are due to social differences in health behaviours, such as smoking, excessive alcohol consumption, physical inactivity and inadequate nutrition. Psychosocial explanations, in contrast, highlight the importance of social networks, chronic stressors and critical life events for health (Moor, Spallek and Richter, 2017). The overarching model (Figure 15.1) represents the close linkages across all these dimensions. Material conditions influence health burdens, resources, and individual behaviours as well as the access to health care, which all together shape health inequalities – which then, in turn, might again lead to different social inequalities. This cycle may result in an accumulation of social and health resources across the life course, also see Kelley, Oladimeji and Dannefer, Chapter 3 in this book. In the following, we present theoretical and empirical evidence for the different competing hypotheses on socially induced health inequalities in older age and their contextual embeddedness, with a focus on methodological challenges.

Health inequalities in older age  217

SOCIOECONOMIC RESOURCES AND HEALTH ACROSS THE LIFE COURSE Selection Processes and Social Causation Numerous studies document the existence of a social gradient in health: individuals with a lower socioeconomic status (SES) bear a higher risk of poor health and mortality relative to their counterparts from higher SES groups (Marmot, 2005). In most studies, SES is measured by income, occupational position or education, and SES-related health inequalities are evident for numerous indicators of physical and mental health including chronic illness, functional limitations (Dugravot et al., 2020), poor self-rated health (Schmitz and Pförtner, 2017), and depressive symptoms (Schmitz and Brandt, 2019). Pronounced health inequalities are also evident with respect to life expectancy. Across Europe, socioeconomic inequalities in overall life expectancy usually amount to between five and ten years, and between ten and twenty years difference in disability-free life expectancy (Mackenbach, 2006; Mackenbach et al., 2008). While the social gradient in health clearly persists into older ages, and even among the oldest-old (Fors and Thorslund, 2015), there are competing hypotheses on how health inequalities evolve across the life course. The health selection-hypothesis claims that differences in health status lead to inequalities in social positions, as measured by educational attainment, income or occupation. Healthier individuals tend to achieve favourable positions in societies, while less healthy individuals have lower chances for upward social mobility and will probably only achieve a lower SES (health affects SES). In contrast, the social causation hypothesis suggests that higher SES presents several resources that are beneficial for health. From this point of view, individuals with a high SES have higher chances of maintaining a good health status due to their access to social support, knowledge, and health-promoting behaviours (SES affects health; Kröger, Pakpahan and Hoffmann, 2015). It has been argued that health selection is less relevant than social causation, especially in middle and late adulthood, but a closer examination of the existing literature shows inconsistent empirical evidence. Kröger, Pakpahan and Hoffmann (2015) conducted a systematic literature review and the majority of the 34 studies included found evidence for both directions of causality. Still, the studies differed with respect to the relative importance of social causation or health selection. The mixed evidence may be due to several reasons. First, studies differ with respect to the operationalisation of SES. Generally, health selection and social causation are equally important in studies relying on labour-market-oriented SES indicators (e.g. wages, employment status or employment grade). In contrast, studies relying on broad indicators of SES, such as education or household income, tend to find that social causation is more important than health selection. Second, the relevance of the mechanisms differs across life stages. Hoffmann, Kröger and Geyer (2019) found that social causation and health selection are equally important for the emergence of health inequalities during childhood and middle adulthood. In the transition from adulthood to old age, however, health selection processes become less important and social causation becomes the dominant mechanism. Nevertheless, scholars argue that the most frequently used SES indicators in contemporary research (income, occupation, and education) do not adequately capture the living conditions of older people. Beyond the age of 65, most adults are retired, rendering classifications based on occupational status and position problematic. Again, selection processes play a role because

218  Handbook of health inequalities across the life course poor health may be associated with downward social mobility towards the end of working life. For the same reason, income is a problematic indicator for measuring SES in old age (Grundy and Holt, 2001). Aligned with this argument, a longitudinal study (Andreß and Hörstermann, 2012) showed that deprivation poverty (measured by whether a socially accepted standard of living can be achieved) is more strongly related to health than income poverty. As deprivation poverty only occurs after a prolonged period of income deprivation, it suggests an accumulation of disadvantages over the life course, especially for older people. In contrast to income and occupation, educational attainment is typically achieved early in life, so problems of social selection may be less evident (but still relevant, see Cozzani and Härkönen, Chapter 19 in this book). However, the majority of today’s older population attained a low level of formal education, thereby limiting educational differentiation and likely distinguishing the most privileged older adults (Grundy and Holt, 2001). Gender differences in life course trajectories present another concern regarding the measurement of SES at older ages. At least among current cohorts of older adults, there are pronounced gender differences in their employment histories – both in terms of labour market involvement and occupation types. Many women of current older cohorts did not engage in paid labour, but were responsible for care work and household chores, so indicators of individual SES do not adequately reflect the lived realities for many older women (Grundy and Holt, 2001). Against this background, researchers have suggested couple- or household-based indicators for measuring women’s SES among older cohorts. Empirical evidence suggests that, for married women, the differentiation of SES groups based on their husband’s occupation is associated with greater differentiation in mortality and health than SES measures based on women’s own occupation (Sacker et al., 2000). Accordingly, health selection processes combined with gender differences in social stratification compound the challenge of measuring SES in the older population. The Social Gradient in Health across the Life Course Another recurring research question concerns changes in the magnitude of health inequalities across the life course. According to the continuity hypothesis, socioeconomic resources acquired across the life course exert a stable influence over time, so that health inequalities in old age have a similar magnitude as in earlier stages of the life course (O’Rand and Henretta, 1999). The age-as-a-leveller hypothesis, in contrast, claims that health inequalities decrease in old age due to biological processes that affect all individuals irrespective of SES. In addition, the hypothesis refers to the buffering effect of social security measures as well as to the premature mortality in lower SES groups, so only relatively healthy individuals with low SES reach higher age groups (Lynch, 2003). Finally, the cumulation hypothesis postulates that health risks in socially disadvantaged groups accumulate over the lifespan, thereby widening health inequalities at older ages (Ross and Wu, 1996), also see Kelley, Oladimeji and Dannefer, Chapter 3 in this book. Empirical evidence supports all three hypotheses (see the overview in Asada et al., 2018 or Stolz et al., 2017). Many studies use cross-sectional data, which limits the examination of age-associated changes in health as the comparison of different age groups at a certain time might also reflect cohort-related differences in health status (Idler, 1993). Cross-sectional study designs are especially problematic due to selective study samples. Individuals with low SES and/or health limitations are less likely to participate in surveys and may die within the

Health inequalities in older age  219 observation period (Galea and Tracy, 2007). The sample of many surveys thus comprises a selection of relatively healthy individuals as compared to the actual population due to “survivorship bias” and “age related non-response”, so health inequalities at older ages are likely to be underestimated (Brandt, Kaschowitz and Lazarevič, 2016). Another reason for selective study samples is that many surveys, even explicit ageing surveys, typically sample individuals living in private households and exclude older adults living in nursing homes. Kelfve (2017), based on a Swedish sample including nursing homes, found that health inequalities were underestimated when nursing home inhabitants were not included in the analytical sample. Longitudinal studies also do not provide consistent results (Leopold and Engelhardt, 2013; Stolz et al., 2017; Asada et al., 2018), which might reflect differences in the measurement of health and SES, and systematic differences in measurement due to age-related reporting behaviours (Lazarevič and Brandt, 2020). Regarding the role of health measures, Leopold and Engelhardt (2013) found widening educational inequalities in objective health measures including functional limitations, grip strength, and mobility, while the educational gradient in overall self-rated health (subjective measure) remained more or less constant with increasing age. Again, also the measurement of SES is important. A study comparing different SES-indicators (Schöllgen, Huxhold and Tesch-Römer, 2010) showed that health inequalities across occupational groups decrease at older ages when individuals quit working life, so individuals with negative working conditions may profit from retirement. When SES is measured by educational attainment, in contrast, we do not observe a decrease in health inequalities at older age. Findings like these suggest that education-related resources, such as knowledge about health-promoting behaviour and strategies to cope with stressful life conditions, are relevant across the life course and exert a stable long-term influence on health inequalities. Finally, and importantly, empirical studies show that the country context matters. A comparative assessment shows that health inequalities increase more rapidly at older ages in the US, a country with a liberal welfare state model that provides little welfare services, as compared to Sweden, a country with a generous social security system (Leopold, 2018; but see also Hoffmann, 2011). Moreover, empirical studies support the assumption that the cutbacks in welfare services and health services that can be observed in many European countries during the last decade are associated with increasing health inequalities in more recent cohorts (Leopold, 2016, but see also Bambra et al., 2010). Thus, not surprisingly, the availability of a social security net, health care, and knowledge for different socioeconomic groups are crucial with respect to social inequalities in individual health trajectories across the life course and well-being outcomes in old age.

SOCIAL NETWORKS, FAMILY AND HEALTH ACROSS THE LIFE COURSE Social Networks and Pathways to Health Social “convoys” (Antonucci, Ajrouch and Birditt, 2014) in terms of social and family networks form another important determinant for individual health, which can be conceptualized as “social capital” (Kawachi and Berkman, 2000); also see Klocke and Stadtmüller, Chapter 11 in this book. Social networks, defined as the constellation of an individual’s social relationships including structural (size and composition), functional (social support exchanges) and

220  Handbook of health inequalities across the life course interactional (frequency of contact and quality of relationships) aspects, are critical to individuals’ health across the life course (Smith and Christakis, 2008; Umberson and Karas Montez, 2010; Thoits, 2011). Social networks change considerably across the life course, not only due to individual decisions but also related to different life events and social roles (Wrzus et al., 2013), including retirement and spousal loss (Ha, 2008; Kalmijn, 2012), with implications for one’s health. The literature provides two main hypothesized pathways to explain the link between social networks and health. Primarily, the direct effect hypothesis proposes that social networks can promote health independent of any presence of stress. Accordingly, social isolation (the lack of social relationships) has been associated with increased risks of poor physical and mental health (Coyle and Dugan, 2012) as well as mortality (Steptoe et al., 2013). Alternatively, the stress-buffering hypothesis proposes that social networks indirectly promote health under stressful circumstances whereby the availability of social ties may help improve one’s appraisal of stressors when they arise (Thoits, 2011). Social relationships present both resources (support) and costs (strains, e.g. conflict) that can promote or undermine individuals’ health (Cohen and Janicki-Deverts, 2009). Having strong meaningful relationships may provide psychological benefits such as belongingness and life purpose that, in turn, induce a sense of responsibility for one’s health and motivate individuals to adopt healthier lifestyles (Berkman et al., 2000; Umberson and Karas Montez, 2010). Relatedly, network members can also be agents of social control that influence, encourage, and regulate health behaviours through the exchange of information, health habits and health care utilisation (Umberson, Crosnoe and Reczek, 2010; Conklin et al., 2014). Taken together, social relationships are considered central to the process of cumulative (dis-) advantages across the life course (DiMaggio and Garip, 2012); also see Kelley, Oladimeji and Dannefer, Chapter 3 in this book. Family relationships are distinct from all other social relationships due to the depth of intimacy and interdependent bonds that shape social integration, connection, behaviour and overall health across the lifespan (Elder and Johnson, 2003), also see Bordone, Di Gessa and Hank, Chapter 13 in this book. As such, family relationships are key social determinants of health (Russell, Coleman and Ganong, 2018), exerting their influences from early childhood on (Brandt, Deindl and Hank, 2012), also see Tampubolon and Dekhtyar and Fors, Chapters 17 and 18 in this book. Of all social network members, family members, especially partners and adult children, are typically the main sources of social support in later life (Deindl and Brandt, 2016; Thomas, Liu and Umberson, 2017), comprise the core familial network ties among older adults (Cornwell et al., 2009), and may protect against poor health when faced with life stressors (Quashie and Andrade, 2020). Friendships, sometimes secondary to familial relationships, are also an important source of support, especially for older adults without traditional familial resources such as childless or unpartnered people (Ermer and Proulx, 2019). Additionally, friendship network size tends to shrink as individuals age whereas family network size remains more stable across the lifespan (Wrzus et al., 2013), thereby reinforcing the salience of familial networks for later life health. In the following, we therefore focus on the role of familial relationships for health inequalities in later life.

Health inequalities in older age  221 Family Relationships and Later Life Health From a resource perspective, older adults with more family members hypothetically have a wider pool of social resources relative to those with smaller family sizes. Shiovitz-Ezra and Litwin (2012) found that older adults embedded in family-based and restricted networks, which tend to have relatively fewer social support resources relative to diverse networks, were more likely to adopt health-damaging behaviours. Other research, however, has found that older adults with a larger share of kin within their personal networks had a lower risk of mortality (Roth, 2020). Here, likely selection effects again play a role: Individuals who are able to invest in social ties are more likely to be better off in other respects, too. Another line of research suggests that the number of key family ties, particularly adult children who are important sources of social support in later life (Hansen, 2012), is critical for health. The existing literature, however, does not provide a consensus on whether ageing without children or with few children is harmful to one’s health as there may be several offsetting mechanisms (e.g. health and social selection into parenthood; see Dykstra, 2009). For instance, studies based on samples of older adults from Europe have found that childless older adults and parents of many children have a higher risk of mortality (Grundy and Kravdal, 2010), and lower cognition (Read and Grundy, 2017) relative to those with two or fewer children. Other studies among US and European older adults suggest that childlessness or having only one child presents a risk for depression (Grundy, van den Broek and Keenan, 2019) and cognitive impairment (Zhang and Fletcher, 2021) relative to parents with many children. Still, comparative research using harmonised measures of health and parenthood across a wide range of middle- and high-income countries has shown that childlessness is generally not associated with older adults’ physical and mental health (Quashie et al., 2021). Beyond the sheer number of family ties, however, the composition of the family network is also linked to health. It has been shown that older adults with one key family tie (e.g. partner) but lacking another (e.g. child) have been shown to have elevated risks of mortality (Patterson, Margolis and Verdery, 2020). Although support from children might be generally beneficial for parents’ well-being (Merz, Schulze and Schuengel, 2010), some studies suggest that receiving support from adult children may undermine health. Apart from obvious selection processes (high levels of support reflect high levels of need), this may be, in part, due to older adults’ perceived loss of independence (Thomas, 2010) or the limited motivation to regain functional independence in the case of older adults with mobility limitations who co-reside with their children (Litwin and Stoeckel, 2013). Yet, the extent to which family relationships enhance older adults’ health also depends on the relationship quality (Uchino et al., 2015; Umberson and Thomeer, 2020). Whereas positive relations (support, understanding, intimacy) with partners and adult children are associated with health benefits, perceived poor-quality family relationships (e.g. ambivalence, criticism, high demands) are associated with higher health risks, including cardiovascular disease and depression (Uchino et al., 2015; Polenick et al., 2018). Importantly, partnership quality becomes more critical for health at older ages given that partners tend to live together and there is more exposure to opportunities for support and strain. As such, studies suggest that strained partnerships relative to both positive partnerships and being unpartnered, have stronger health impacts at older ages (Hank and Wagner, 2013; Umberson and Thomeer, 2020). Substantial and methodological issues are likely to play an important role in explaining the partly inconclusive or even contradictory results. First, there are also various downsides of

222  Handbook of health inequalities across the life course social capital (e.g. contagion with health damaging behaviours) for health (Villalonga-Olives and Kawachi, 2017), which are often not explicitly assessed. Especially within the family, informal care may be needed and expected, which can become a demanding form of support that induces various strains and undermines health, especially among older adults who care for their partners (Pinquart and Sörensen, 2011; Kaschowitz and Brandt, 2017). Second, different measurements of health and social networks (e.g. role relations, name generators, task-related assessments) may lead to different conclusions. Third, reverse causality (health influencing social networks and social support) and selection effects in terms of health and social networks (also due to unobserved influences earlier in life) cannot be ruled out in cross-sectional studies. Fourth, substantive or methodological contextual influences (e.g. availability and access to health services or cultural biases and reporting behaviours) are often not accounted for in single country or group studies. Finally, but importantly, survey samples often underrepresent older people in care homes and thus only include the “tip of the iceberg” regarding health impairment, small social networks and lack of informal support (see e.g. Abbott, Prvu Bettger and Hampton, 2015; Luppa et al., 2010). Studies on older populations based on survey data from private households are thus likely to underestimate the importance of social networks for health in older age. Not least, we know that individuals’ exposure to stress and their capacity to draw upon social relationships as a coping resource are shaped by macro-structural conditions including gender roles and socioeconomic positions (Berkman et al., 2000). Thus, the health benefits and costs of social relationships vary across social groups. Some important external factors for the links between social networks and health inequalities have already been studied in more detail as will be presented now. Differences between Women and Men Many studies have investigated gender differences in the role of social relationships for later life health. Due to the traditional gender division of labour, women tend to bear more responsibility for unpaid labour (providing care and support, maintaining social networks), while men are more likely to be in the paid labour market and maintain the financial security of the household (Spitze and Ward, 2000). Thus, heterosexual partnerships tend to confer economic benefits for women while men receive more health benefits, which may be indirectly linked to the social network gains from their partners (Umberson and Karas Montez, 2010). Women are more likely to provide care to their health-impaired partners without additional support from other network members and professional services, at higher intensities, and thus experience the strains of balancing family caregiving and other non-family responsibilities more than men (Carr and Utz, 2020). Much research suggests that female caregivers have more negative caregiving experiences and worse mental health relative to their male counterparts (Pinquart and Sörensen, 2006; Lin, Fee and Wu, 2012). These gendered health outcomes of care, however, again seem to vary between different care constellations and contexts, as comparative studies show (Kaschowitz and Brandt, 2017; Floridi et al., 2022). As men are more reliant on their female partners for care, social support, and health protection (e.g. monitoring health behaviours), some studies suggest that spousal loss is associated with worse health among men due to the loss of social and instrumental support from their partners (Carr and Bodnar-Deren, 2009). Nonetheless, recent longitudinal studies do not find gender differences

Health inequalities in older age  223 in depressive symptoms following widowhood (Schaan, 2013; Schmitz, 2021) suggesting that spousal loss is a stressor that has comparable health effects for men and women. The health impacts of ties to adult children may also vary by gender, see also Bordone et al., Chapter 13 in this book. Given traditional gender norms and women’s kin keeping social roles, the presence of children is arguably more critical for women’s health relative to men (Dykstra and Hagestad, 2007). Parenthood also presents strains in the allocations of time, economic and social resources that may accumulate to produce poor health over the life course, with potentially greater impact on women due to the traditional gender norms of parenthood. A growing body of research examines the links between fertility history and later-life health and tends to show similar patterns for women and men: having two (or three) children is associated with better health outcomes than having one child, many (four or more) or no children. Additionally age of entry to parenthood is salient for later life health, with early entry, compared to later ages, being associated with poor health in later life for both genders but with stronger associations among women (Grundy and Read, 2015; Sironi, Ploubidis and Grundy, 2020). It is not merely ageing with children but the interactions with children that can have gendered health effects. For instance, research among older adults in Eastern Europe has shown that frequent social contact with children is associated with lower risks of depression for mothers, regardless of their partnership status, whereas for fathers frequent contact with their children is associated with lower depression only when they do not have a partner (Tosi and Grundy, 2019). This may reflect differences in men’s reliance on their partners for support and women’s higher vulnerability to unsupportive/inadequate relationships with their children. Differences between Socioeconomic Groups Disparities in the structural, functional, and interactional aspects of social relationships also vary by SES. Higher educated older adults tend to have more diverse social networks, whereas those with lower education levels have more restricted kin-based networks (Cornwell et al., 2009). Moreover, individuals from lower SES groups tend to live in closer proximity to their family members compared to higher SES groups (Fors and Lennartsson, 2008). As health and care needs arise with advancing age, closer geographic proximity can facilitate easier access to familial support to maintain health, as well as delay or limit formal care, which can be costly (Choi et al., 2015). As such, socioeconomically disadvantaged older adults are more likely to receive informal care (Floridi, Carrino and Glaser, 2021) and provide care to family members than those with more socioeconomic resources (Quashie et al., 2022). Given the inverse relationship between SES and health, we may expect differential psychosocial vulnerabilities across socioeconomic groups such that lower SES individuals have a higher exposure to chronic stress and less control over their life circumstances that accompany financial hardship and material deprivation (Taylor and Seeman, 1999; Siegrist and Marmot, 2004). Few studies have specifically examined the role of the family networks to the social gradient in later life health, and the existing studies rely on cross-sectional data so that causal interpretation is not possible. Earlier research of US adults (Antonucci, Ajrouch and Janevic, 1999) pointed to the importance of the composition of the family network and life course stage in the moderating role of social support for SES-health disparities. Their finding suggests that being in a low resource social position and unable to rely on one’s partner exacerbates health risks in old age. Recent research among older Europeans, especially within Central Europe, further support the salience of partnership for the health of lower SES older

224  Handbook of health inequalities across the life course adults (Craveiro, 2017). When considering the importance of adult children to SES-health disparities among older adults, earlier research in Malaysia (Wu and Rudkin, 2000) identified that interactional aspects of family relations are also important, as limited contact with adult children was more detrimental to the health of lower SES compared to higher SES adults. Taken together, the findings suggest that having a partner and children may offset or buffer some of the psychosocial strains associated with lower SES by offering emotional support or other health-related resources. Given the intergenerational transmission of socioeconomic (dis)advantages and “linked lives” of families across the life course (Bengtson and Allen, 2009), a growing body of literature has been attentive to how adult children’s educational attainment influences their parents’ health at older ages. Studies across diverse country settings consistently show health advantages of adult children’s higher education for several parental health outcomes, including mortality (Elo, Martikainen and Aaltonen, 2018), cognition (Ma et al., 2021), physical health (Yahirun, Sheehan and Hayward, 2017) and mental health (Dennison and Lee, 2021) – also if the parents have low levels of education (Sabater, Graham and Marshall, 2020). The health benefits of children’s higher educational attainment may operate through higher or better social support (e.g. purchase access to better health care), exposure to health information, and positive health behaviour resources (Friedman and Mare, 2014). Nevertheless, methodological challenges, for example due to selection into parenthood, remain. For instance, childhood social disadvantage is associated with early parenthood (and larger family size), which is in turn associated with lower educational attainment, and poorer health (including risky health behaviours) in mid and later life (Lee and Ryff, 2016; Grundy and Read, 2015). Additionally, lower SES individuals have higher risks of premature mortality relative to higher SES individuals (Lewer et al., 2020), thus potentially being unable to reap the benefits of their children’s relatively higher educational attainment. Taken together, socioeconomic disparities in the health benefits of children’s higher educational attainment may be underestimated. Overall, however, positive intergenerational relations and partnership potentially buffer the stressors and minimize the health risks of lower SES adults as they age. Nevertheless, the health benefits of adult children also depend on the socioeconomic resources of their children and are potentially more salient for socioeconomically disadvantaged older adults.

CONCLUSION Health inequalities in old age are a result of unequal access to resources and exposure to risk factors across the entire life course due to a combination of individual decisions and contextual circumstances (e.g., social networks, societal context). Disentangling the various, partly cumulative and compensating mechanisms on different levels is a complex issue. Relatedly, whether health gaps in older age are (actually) widening over time between social groups and country contexts remains an open question (also see Kelley, Oladimeji and Dannefer; Fritzell and Rehnberg, Chapters 3 and 20 in this book). As we have pointed out, methodological issues, including systematic context, age and health related measurement biases and sample selectivity in studies based on survey data (also see Gebel, Chapter 7 in this book) potentially impede the correct assessment of (the development of) health inequalities among older adults. Given this complexity, our review only provides a glimpse into conceptual foundations, hypothetical pathways, and mechanisms, as well as empirical findings on older age health

Health inequalities in older age  225 inequalities. We focused on the role of socioeconomic resources and family networks in old age, leaving many other central dimensions (e.g. work and employment) aside (e.g. Hank and Brandt, 2013, also see Chapters 12 and 14 by Siegrist and Giesecke in this book). Although we cannot rule out reverse causality and endogeneity, we have shown that socioeconomic resources and social networks matter for health across the life course and lead to social gradients in old age health. There are, however, group and gender differences in the effects of different resource types, and contextual influences such as social cohesion and health (care) policies matter (Deindl, Brandt and Hank, 2016; Craveiro, 2017, also see Aronsson, Tugrul, Bambra and Eikemo, Chapter 24 in this book). These contextual factors cannot only compensate for some deprivation of individual resources, but also influence the different mechanisms (e.g. health behaviours, care) which likely lead to social inequalities in health. They are thus also closely linked to differential morbidity and mortality patterns within and between different contexts, as introduced at the beginning of this chapter: some groups seem to benefit from gains in (healthy) life expectancy while others do not (Gutin and Hummer, 2021). Not least, different historical events and their impacts on everyday lives (e.g., wars, hunger, persecution), still play an important role for inequalities in older age health today (Teerawichitchainan and Korinek, 2012; Kesternich et al., 2014). It will be a task for future researchers to assess how inequalities in health evolve, and whether and how the Covid-19 pandemic and other (environmental and societal) crises (also see König and Heisig, Chapter 21 in this book) impact health inequalities in older and younger generations, and across their life courses. Together with the adjacent chapters in this book we hope to provide a reliable foundation for doing so.

REFERENCES Abbott, K. M., Prvu Bettger, J. and Hampton, K. N. (2015) ‘The feasibility of measuring social networks among older adults in assisted living and dementia special care units’, Dementia, 14(2), pp. 199–219. doi:10.1177/1471301213494524. Andreß, H.-J. and Hörstermann, K. (2012) ‘Lebensstandard und Deprivation im Alter in Deutschland. Stand und Entwicklungsperspektiven’, Zeitschrift für Sozialreform, 58(2), pp. 209–234. doi:10.1515/ zsr-2012-0206. Antonucci, T. C., Ajrouch, K. J. and Birditt, K. S. (2014) ‘The convoy model: Explaining social relations from a multidisciplinary perspective’, The Gerontologist, 54(1), pp. 82–92. doi:10.1093/geront/ gnt118. Antonucci, T. C., Ajrouch, K. J. and Janevic, M. (1999) ‘Socioeconomic status, social support, age, and health. Annals of New York academic science’, Annals of the New York Academy of Sciences, 896(1), pp. 390–392. doi:10.1111/j.1749-6632.1999.tb08150.x. Asada, Y., Hurley, J., Grignon, M. and Kirkland, S. (2018) ‘Health inequalities and inequities by age: Stability for the Health Utilities Index and divergence for the Frailty Index’, SSM – Population Health, 5, pp. 17–32. doi:10.1016/j.ssmph.2018.04.002. Bambra, C., Gibson, M., Sowden, A., Wright, K., Whitehead, M. and Petticrew, M. (2010) ‘Tackling the wider social determinants of health and health inequalities: evidence from systematic reviews’, Journal of Epidemiology & Community Health, 64(4), pp. 284–291. doi:10.1136/jech.2008.082743. Bengtson, V. L. and Allen, K. R. (2009) ‘The life course perspective applied to families over time’, in Boss, P. et al. (eds) Sourcebook of Family Theories and Methods. Boston, MA: Springer US, pp. 469–504. doi:10.1007/978-0-387-85764-0_19. Berkman, L. F., Glass, T., Brissette, I. and Seeman, T. E. (2000) ‘From social integration to health: Durkheim in the new millennium’, Social Science & Medicine, 51(6), pp. 843–857. doi:10.1016/ S0277-9536(00)00065-4.

226  Handbook of health inequalities across the life course Brandt, M., Deindl, C. and Hank, K. (2012) ‘Tracing the origins of successful aging: The role of childhood conditions and social inequality in explaining later life health’, Social Science & Medicine, 74(9), pp. 1418–1425. doi: 10.1016/j.socscimed.2012.01.004. Brandt, M., Kaschowitz, J. and Lazarevič, P. (2016) ‘Gesundheit im Alter: Ein Überblick über den Stand der sozialwissenschaftlichen Forschung’, in Jungbauer-Gans, M. and Kriwy, P. (eds) Handbuch Gesundheitssoziologie. Wiesbaden: Springer Fachmedien Wiesbaden, pp. 1–18. doi:10.1007/978-3-658-06477-8_23-1. Carr, D. and Bodnar-Deren, S. (2009) ‘Gender, aging and widowhood’, in Uhlenberg, P. (ed.) International Handbook of Population Aging. Dordrecht: Springer Netherlands, pp. 705–728. doi:10.1007/978-1-4020-8356-3_32. Carr, D. and Utz, R. L. (2020) ‘Families in later life: A decade in review’, Journal of Marriage and Family, 82(1), pp. 346–363. doi:10.1111/jomf.12609. Choi, H., Schoeni, R. F., Langa, K. M. and Heisler, M. M. (2015) ‘Spouse and child availability for newly disabled older adults: Socioeconomic differences and potential role of residential proximity’, The Journals of Gerontology Series B: Psychological Sciences and Social Sciences, 70(3), pp. 462–469. doi:10.1093/geronb/gbu015. Cohen, S. and Janicki-Deverts, D. (2009) ‘Can we improve our physical health by altering our social networks?’, Perspectives on Psychological Science, 4(4), pp. 375–378. doi:10.1111/j.1745-6924.2009.01141.x. Conklin, A. I., Forouhi, N. G., Surtees, P., Khaw, K.-T., Warcham, N. J. and Monsivais, P. (2014) ‘Social relationships and healthful dietary behaviour: Evidence from over-50s in the EPIC cohort, UK’, Social Science & Medicine, 100, pp. 167–175. doi:10.1016/j.socscimed.2013.08.018. Cornwell, B., Schumm, L. P., Laumann, E. O. and Graber, J. (2009) ‘Social networks in the NSHAP study: Rationale, measurement, and preliminary findings’, The Journals of Gerontology Series B: Psychological Sciences and Social Sciences, 64B(Supplement 1), pp. i47–i55. doi:10.1093/geronb/ gbp042. Coyle, C. E. and Dugan, E. (2012) ‘Social isolation, loneliness and health among older adults’, Journal of Aging and Health, 24(8), pp. 1346–1363. doi:10.1177/0898264312460275. Craveiro, D. (2017) ‘The role of personal social networks on health inequalities across European regions’, Health & Place, 45, pp. 24–31. doi:10.1016/j.healthplace.2017.02.007. Crimmins, E. M. and Beltrán-Sánchez, H. (2010) ‘Mortality and morbidity trends: Is there compression of morbidity?’, The Journals of Gerontology Series B: Psychological Sciences and Social Sciences, 66B(1), pp. 75–86. doi:10.1093/geronb/gbq088. Deindl, C. and Brandt, M. (2016) ‘Support networks of childless older people: informal and formal support in Europe’, Ageing and Society, 37(8), pp. 1543–1567. doi:10.1017/S0144686X16000416. Deindl, C., Brandt, M. and Hank, K. (2016) ‘Social networks, social cohesion, and later-life health’, Social Indicators Research, 126(3), pp. 1175–1187. doi:10.1007/s11205-015-0926-5. Dennison, C. R. and Lee, K. S. (2021) ‘Adult children’s educational attainment and parent health in midand later-life’, The Journals of Gerontology: Series B. Edited by D. S. Carr, 76(9), pp. 1857–1869. doi:10.1093/geronb/gbab109. DiMaggio, P. and Garip, F. (2012) ‘Network effects and social inequality’, Annual Review of Sociology, 38(1), pp. 93–118. doi:10.1146/annurev.soc.012809.102545. Dugravot, A., Fayosse, A., Dumurgier, J., Bouillon, K., Rayana, T. B., Schnitzler, A., Kivimaki, M., Sabia, S. and Snigh-Manoux, A. (2020) ‘Social inequalities in multimorbidity, frailty, disability, and transitions to mortality: A 24-year follow-up of the Whitehall II cohort study’, The Lancet Public Health, 5(1), pp. e42–e50. doi:10.1016/S2468-2667(19)30226-9. Dykstra, P. A. (2009) ‘Childless old age’, in Uhlenberg, P. (ed.) International Handbook of Population Aging. Dordrecht: Springer Netherlands, pp. 671–690. doi:10.1007/978-1-4020-8356-3_30. Dykstra, P. A. and Hagestad, G. O. (2007) ‘Childlessness and parenthood in two centuries: Different roads— different maps?’, Journal of Family Issues, 28(11), pp. 1518–1532. doi:10.1177/0192513X07303881. Elder, G. H. and Johnson, M. K. (2003) ‘The life course and aging: Challenges, lessons, and new directions’, in Settersten, R. A., Invitation to the Life Course: Toward New Understandings of Later Life. Amityville, NY: Baywood, pp. 49–81.

Health inequalities in older age  227 Elo, I. T., Martikainen, P. and Aaltonen, M. (2018) ‘Children’s educational attainment, occupation, and income and their parents’ mortality’, Population Studies, 72(1), pp. 53–73. doi:10.1080/00324728.2 017.1367413. Ermer, A. E. and Proulx, C. M. (2019) ‘Associations between social connectedness, emotional well-being, and self-rated health among older adults: Difference by relationship status’, Research on Aging, 41(4), pp. 336–361. doi:10.1177/0164027518815260. Floridi, G., Carrino, L. and Glaser, K. (2021) ‘Socioeconomic inequalities in home-care use across regional long-term care systems in Europe’, The Journals of Gerontology: Series B, 76(1), pp. 121–132. doi:10.1093/geronb/gbaa139. Floridi, G., Quashie, N. T., Glaser, K. and Brandt, M. (2022) ‘Partner care arrangements and well-being in mid- and later life: The role of gender across care contexts’, The Journals of Gerontology: Series B, 77(2), pp. 435–455. doi:10.1093/geronb/gbab209. Fors, S. and Lennartsson, C. (2008) ‘Social mobility, geographical proximity and intergenerational family contact in Sweden’, Ageing and Society, 28(2), pp. 253–270. doi:10.1017/S0144686X07006617. Fors, S. and Thorslund, M. (2015) ‘Enduring inequality: educational disparities in health among the oldest old in Sweden 1992–2011’, International Journal of Public Health, 60(1), pp. 91–98. doi:10.1007/s00038-014-0621-3. Friedman, E. M. and Mare, R. D. (2014) ‘The schooling of offspring and the survival of parents’, Demography, 51(4), pp. 1271–1293. doi:10.1007/s13524-014-0303-z. Fries, J. F. (1980). ‘Aging, natural death, and the compression of morbidity’, New England Journal of Medicine, 303(3), pp. 130–135. doi:10.1056/NEJM198007173030304. Galea, S. and Tracy, M. (2007) ‘Participation rates in epidemiologic studies’, Annals of Epidemiology, 17(9), pp. 643–653. doi:10.1016/j.annepidem.2007.03.013. Grundy, E. and Holt, G. (2001) ‘The socioeconomic status of older adults: How should we measure it in studies of health inequalities?’, Journal of Epidemiology & Community Health, 55(12), pp. 895–904. doi:10.1136/jech.55.12.895. Grundy, E. and Kravdal, Ø. (2010) ‘Fertility history and cause-specific mortality: A register-based analysis of complete cohorts of Norwegian women and men’, Social Science & Medicine, 70(11), pp. 1847–1857. doi:10.1016/j.socscimed.2010.02.004. Grundy, E. and Read, S. (2015) ‘Pathways from fertility history to later life health: Results from analyses of the English Longitudinal Study of Ageing’, Demographic Research, 32, pp. 107–146. doi:10.4054/ DemRes.2015.32.4. Grundy, E., van den Broek, T. and Keenan, K. (2019) ‘Number of children, partnership status, and later-life depression in Eastern and Western Europe’, The Journals of Gerontology: Series B, 74(2), pp. 353–363. doi:10.1093/geronb/gbx050. Gutin, I. and Hummer, R. A. (2021) ‘Social inequality and the future of US life expectancy’, Annual Review of Sociology 47(1), pp. 501–520, doi: 10.1146/annurev-soc-072320-100249. Ha, J.-H. (2008) ‘Changes in support from confidants, children, and friends following widowhood’, Journal of Marriage and Family, 70(2), pp. 306–318. doi:10.1111/j.1741-3737.2008.00483.x. Hank, K. and Brandt, M. (2013) ‘Health, families, and work in later life: A review of current research and perspectives’, Analyse & Kritik, 35(2), pp. 303–320. doi:10.1515/auk-2013-0203. Hank, K. and Wagner, M., 2013. Parenthood, marital status, and well-being in later life: Evidence from SHARE’, Social Indicators Research, 114(2), pp. 639–653. doi:10.1007/s11205-012-0166-x. Hansen, T. (2012) ‘Parenthood and happiness: A review of folk theories versus empirical evidence’, Social Indicators Research, 108(1), pp. 29–64. doi:10.1007/s11205-011-9865-y. Hoffmann, R. (2011) ‘Socioeconomic inequalities in old-age mortality: A comparison of Denmark and the USA’, Social Science & Medicine, 72(12), pp. 1986– 1992. doi: 10.1016/j.socscimed.2011.04.019. Hoffmann, R., Kröger, H. and Geyer, S. (2019) ‘Social causation versus health selection in the life course: Does their relative importance differ by dimension of SES?’, Social Indicators Research, 141(3), pp. 1341–1367. doi:10.1007/s11205-018-1871-x. Idler, E. L. (1993) ‘Age differences in self-assessments of health: Age changes, cohort differences, or survivorship?’, Journal of Gerontology, 48(6), pp. S289–S300. doi:10.1093/geronj/48.6.S289. Janßen, C., Sauter, S. and Kowalski, C. (2012) ‘The influence of social determinants on the use of prevention and health promotion services: Results of a systematic literature review’, GMS Psycho-Social-Medicine; 9:Doc07; ISSN 1860-5214 [Preprint]. doi:10.3205/PSM000085.

228  Handbook of health inequalities across the life course Kalmijn, M. (2012) ‘Longitudinal analyses of the effects of age, marriage, and parenthood on social contacts and support’, Advances in Life Course Research, 17(4), pp. 177–190. doi:10.1016/j. alcr.2012.08.002. Kaschowitz, J. and Brandt, M. (2017) ‘Health effects of informal caregiving across Europe: A longitudinal approach’, Social Science & Medicine, 173, pp. 72–80. doi:10.1016/j.socscimed.2016.11.036. Kawachi, I. and Berkman, L. (2000) ‘Social cohesion, social capital, and health’, in Berkman, L. F. and Kawachi, I. (eds) Social Epidemiology. New York: Oxford University Press, pp. 174–190. Kelfve, S. (2017) ‘Underestimated health inequalities among older people—A consequence of excluding the most disabled and disadvantaged’, The Journals of Gerontology: Series B. Edited by D. Carr, 74(8), pp. e125–e134. doi:10.1093/geronb/gbx032. Kesternich, I., Siflinger, B., Smith, J. P. and Winter, J. K. (2014) ‘The effects of World War II on economic and health outcomes across Europe’, Review of Economics and Statistics, 96(1), pp. 103–118. doi:10.1162/REST_a_00353. Kröger, H., Pakpahan, E. and Hoffmann, R. (2015) ‘What causes health inequality? A systematic review on the relative importance of social causation and health selection’, European Journal of Public Health, 25(6), pp. 951–960. doi:10.1093/eurpub/ckv111. Lazarevič, P. and Brandt, M. (2020) ‘Diverging ideas of health? Comparing the basis of health ratings across gender, age, and country’, Social Science & Medicine, 267, p. 112913. doi:10.1016/j. socscimed.2020.112913. Lee, C. and Ryff, C. D. (2016) ‘Early parenthood as a link between childhood disadvantage and adult heart problems: A gender-based approach’, Social Science & Medicine, 171, pp. 58–66. doi:10.1016/j. socscimed.2016.10.028. Leopold, L. (2016) ‘Cumulative advantage in an egalitarian country? Socioeconomic health disparities over the life course in Sweden’, Journal of Health and Social Behavior, 57(2), pp. 257–273. doi:10.1177/0022146516645926. Leopold, L. (2018) ‘Education and physical health trajectories in later life: A comparative study’, Demography, 55(3), pp. 901–927. doi:10.1007/s13524-018-0674-7. Leopold, L. and Engelhardt, H. (2013) ‘Education and physical health trajectories in old age. Evidence from the Survey of Health, Ageing and Retirement in Europe (SHARE)’, International Journal of Public Health [Preprint]. doi:10.1007/s00038-012-0399-0. Lewer, D. et al. (2020) ‘Premature mortality attributable to socioeconomic inequality in England between 2003 and 2018: An observational study’, The Lancet Public Health, 5(1), pp. e33–e41. doi:10.1016/S2468-2667(19)30219-1. Lin, I.-F., Fee, H. R. and Wu, H.-S. (2012) ‘Negative and positive caregiving experiences: A closer look at the intersection of gender and relationship’, Family Relations, 61(2), pp. 343–358. doi:10.1111/j.1741-3729.2011.00692.x. Litwin, H. and Stoeckel, K. J. (2013) ‘Social network and mobility improvement among older Europeans: The ambiguous role of family ties’, European Journal of Ageing, 10(3), pp. 159–169. doi:10.1007/ s10433-013-0269-5. Luppa, M., Luck, T., Weyerer, S., König, H. H., Brähler, E. and Riedel-Heller S. G. (2010) ‘Prediction of institutionalization in the elderly. A systematic review’, Age and Ageing, 39(1), pp. 31–38, doi: 10.1093/ageing/afp202. Lynch, S. M. (2003) ‘Cohort and life-course patterns in the relationship between education and health: A hierarchical approach’, Demography, 40(2), pp. 309–331. doi:10.1353/dem.2003.0016. Ma, M., Yahirun, J., Saenz, J. and Sheehan, C. (2021) ‘Offspring educational attainment and older parents’ cognition in Mexico’, Demography, 58(1), pp. 75–109. doi:10.1215/00703370-8931725. Mackenbach, J. P. (2006) ‘Health inequalities: Europe in profile. An independent expert report commissioned by the UK Presidency of the EU’. Rotterdam: Department of Public Health Erasmus MC. Mackenbach, J. P., Stirbu, I., Roskam, A.-J. R., Schaap, M. M., Menvielle, G., Leinsalu, M. and Kunst, A. E. (2008) ‘Socioeconomic inequalities in health in 22 European countries’, New England Journal of Medicine, 358(23), pp. 2468–2481. doi:10.1056/NEJMsa0707519. Marmot, M. (2005) ‘Social determinants of health inequalities’, The Lancet, 365(9464), pp. 1099–1104. doi:10.1016/S0140-6736(05)71146-6.

Health inequalities in older age  229 Merz, E.-M., Schulze, H.-J. and Schuengel, C. (2010) ‘Consequences of filial support for two generations: A narrative and quantitative review’, Journal of Family Issues, 31(11), pp. 1530–1554. doi:10.1177/0192513X10365116. Mielck, A. (2000) Soziale Ungleichheit und Gesundheit: empirische Ergebnisse, Erklärungsansätze, Interventionsmöglichkeiten. Bern: Hans Huber. Moor, I., Spallek, J. and Richter, M. (2017) ‘Explaining socioeconomic inequalities in self-rated health: A systematic review of the relative contribution of material, psychosocial and behavioural factors’, Journal of Epidemiology and Community Health, 71(6), pp. 565–575. doi:10.1136/jech-2016-207589. OECD Health Statistics (2019) OECD historical population data and projections database. Available at: https://​www​.oecd​-ilibrary​.org/​sites/​c05578aa​-en/​index​.html​?itemId​=/​content/​component/​c05578aa​ -en. Oeppen, J. and Vaupel, J. W. (2002) ‘Broken limits to life expectancy’, Science, 296(5570), pp. 1029–1031. doi:10.1126/science.1069675. O’Rand, A. M. and Henretta, J. C. (1999) Age and Inequality: Diverse Pathways through Later Life. Boulder, CO: Westview Press (Social inequality series). Parker, M. G. and Thorslund, M. (2007) ‘Health trends in the elderly population: Getting better and getting worse’, The Gerontologist, 47(2), pp. 150–158. doi:10.1093/geront/47.2.150. Patterson, S. E., Margolis, R. and Verdery, A. M. (2020) ‘Family embeddedness and older adult mortality in the United States’, Population Studies, 74(3), pp. 415–435. doi:10.1080/00324728.2020.1817529. Pinquart, M. and Sörensen, S. (2006) ‘Gender differences in caregiver stressors, social resources, and health: An updated meta-analysis’, The Journals of Gerontology Series B: Psychological Sciences and Social Sciences, 61(1), pp. P33–P45. doi:10.1093/geronb/61.1.P33. Pinquart, M. and Sörensen, S. (2011) ‘Spouses, adult children, and children-in-law as caregivers of older adults: A meta-analytic comparison’, Psychology and Aging, 26(1), pp. 1–14. doi:10.1037/a0021863. Polenick, C. A., DePasquale, N., Eggebeen, D. J., Zarit, S. H. and Fingerman, K. L. (2018) ‘Relationship quality between older fathers and middle-aged children: Associations with both parties’ subjective well-being’, The Journals of Gerontology Series B: Psychological Sciences and Social Sciences, 73(7), pp. 1203–1213. doi:10.1093/geronb/gbw094. Quashie, N. T. and Andrade, F. C. D. (2020) ‘Family status and later-life depression among older adults in urban Latin America and the Caribbean’, Ageing and Society, 40(2), pp. 233–261. doi:10.1017/ S0144686X18000879. Quashie, N. T., Arpino, B., Antczak, R. and Mair, C. A. (2021) ‘Childlessness and health among older adults: Variation across five outcomes and 20 countries’, The Journals of Gerontology: Series B. Edited by D. Carr, 76(2), pp. 348–359. doi:10.1093/geronb/gbz153. Quashie, N. T., Wagner, M., Verbakel, E. and Deindl, C. (2022) ‘Socioeconomic differences in informal caregiving in Europe’, European Journal of Ageing, 19, pp. 621–633. doi:10.1007/ s10433-021-00666-y. Read, S. L. and Grundy, E. M. D. (2017) ‘Fertility history and cognition in later life’, The Journals of Gerontology: Series B, 72(6), pp. 1021–1031. doi:10.1093/geronb/gbw013. Ross, C. E. and Wu, C.-L. (1996) ‘Education, age, and the cumulative advantage in health’, Journal of Health and Social Behavior, 37(1), pp. 104–120. Roth, A. R. (2020) ‘Personal networks and mortality in later life: racial and ethnic differences’, Journal of Public Health, 43(4), pp. 824–832. doi:10.1093/pubmed/fdaa135. Rowe, J. W. and Kahn, R. L. (1997) ‘Successful aging’, The Gerontologist, 37(4), pp. 433–440. doi:10.1093/geront/37.4.433. Russell, L. T., Coleman, M. and Ganong, L. (2018) ‘Conceptualizing family structure in a social determinants of health framework: Conceptualizing family structure’, Journal of Family Theory & Review, 10(4), pp. 735–748. doi:10.1111/jftr.12296. Sabater, A., Graham, E. and Marshall, A. (2020) ‘Does having highly educated adult children reduce mortality risks for parents with low educational attainment in Europe?’, Ageing and Society, 40(12), pp. 2635–2670. doi:10.1017/S0144686X19000795. Sacker, A., Firth, D., Fitzpatrick, R., Lynch, K. and Bartley, M. (2000) ‘Comparing health inequality in men and women: Prospective study of mortality 1986–96’, BMJ, 320(7245), pp. 1303–1307. doi:10.1136/bmj.320.7245.1303.

230  Handbook of health inequalities across the life course Schaan, B. (2013) ‘Widowhood and depression among older Europeans – The Role of gender, caregiving, marital quality, and regional context’, The Journals of Gerontology Series B: Psychological Sciences and Social Sciences, 68(3), pp. 431–442. doi:10.1093/geronb/gbt015. Schmitz, A. (2019) ‘Gesundheitliche Ungleichheiten im Alter: Theoretische Perspektiven und methodische Herausforderungen’, Zeitschrift für Gerontologie und Geriatrie, 52(2), pp. 116–121. doi:10.1007/ s00391-019-01518-2. Schmitz, A. (2021) ‘Gendered experiences of widowhood and depression across Europe’, Journal of Affective Disorders, 280, pp. 114–120. doi:10.1016/j.jad.2020.11.091. Schmitz, A. and Brandt, M. (2019) ‘Gendered patterns of depression and its determinants in older Europeans’, Archives of Gerontology and Geriatrics, 82, pp. 207–216. doi:10.1016/j. archger.2019.02.015. Schmitz, A. L. and Pförtner, T.-K. (2017) ‘Health inequalities in old age: The relative contribution of material, behavioral and psychosocial factors in a German sample’, Journal of Public Health, 40(3), pp. e235–e243. doi:10.1093/pubmed/fdx180. Schöllgen, I., Huxhold, O. and Tesch-Römer, C. (2010) ‘Socioeconomic status and health in the second half of life: Findings from the German Ageing Survey’, European Journal of Ageing, 7(1), pp. 17–28. doi:10.1007/s10433-010-0140-x. Shiovitz-Ezra, S. and Litwin, H. (2012) ‘Social network type and health-related behaviors: Evidence from an American national survey’, Social Science & Medicine, 75(5), pp. 901–904. doi:10.1016/j. socscimed.2012.04.031. Siegrist, J. and Marmot, M. (2004) ‘Health inequalities and the psychosocial environment—two scientific challenges’, Social Science & Medicine, 58(8), pp. 1463–1473. doi:10.1016/S0277-9536(03)00349-6. Sironi, M., Ploubidis, G. B. and Grundy, E. M. (2020) ‘Fertility history and biomarkers using prospective data: Evidence from the 1958 national child development study’, Demography, 57(2), pp. 529–558. doi:10.1007/s13524-020-00855-x. Smith, K. P. and Christakis, N. A. (2008) ‘Social networks and health’, Annual Review of Sociology, 34(1), pp. 405–429. doi:10.1146/annurev.soc.34.040507.134601. Spitze, G. and Ward, R. (2000) ‘Gender, marriage, and expectations for personal care’, Research on Aging, 22(5), pp. 451–469. doi:10.1177/0164027500225001. Steptoe, A., Shankar, A., Demakakos, P. and Wardle, J. (2013) ‘Social isolation, loneliness, and all-cause mortality in older men and women’, Proceedings of the National Academy of Sciences, 110(15), pp. 5797–5801. doi:10.1073/pnas.1219686110. Stolz, E., Mayerl, H., Waxenegger, A., Rásky, É. and Freidl, W. (2017) ‘Impact of socioeconomic position on frailty trajectories in 10 European countries: Evidence from the Survey of Health, Ageing and Retirement in Europe (2004–2013)’, Journal of Epidemiology and Community Health, 71(1), pp. 73–80. doi:10.1136/jech-2016-207712. Taylor, S. E. and Seeman, T. E. (1999) ‘Psychosocial resources and the SES–health relationship’, in Adler, N. E. et al. (eds) Socioeconomic Status and Health in Industrial Nations: Social, Psychological, and Biological Pathways. New York Academy of Sciences, pp. 210–225. Teerawichitchainan, B. and Korinek, K. (2012) ‘The long-term impact of war on health and wellbeing in Northern Vietnam: Some glimpses from a recent survey’, Social Science & Medicine, 74(12), pp. 1995–2004. doi:10.1016/j.socscimed.2012.01.040. Thoits, P. A. (2011) ‘Mechanisms linking social ties and support to physical and mental health’, Journal of Health and Social Behavior, 52(2), pp. 145–161. doi:10.1177/0022146510395592. Thomas, P. A. (2010) ‘Is it better to give or to receive? Social support and the well-being of older adults’, The Journals of Gerontology Series B: Psychological Sciences and Social Sciences, 65B(3), pp. 351–357. doi:10.1093/geronb/gbp113. Thomas, P. A., Liu, H. and Umberson, D. (2017) ‘Family relationships and well-being’, Innovation in Aging, 1(3), p. igx025. doi:10.1093/geroni/igx025. Tosi, M. and Grundy, E. (2019) ‘Intergenerational contacts and depressive symptoms among older parents in eastern Europe’, Aging & Mental Health, 23(6), pp. 686–692. doi:10.1080/13607863.201 8.1442412. Uchino, B. N., Ruiz, J. M., Smith, T. W., Smyth, J. M., Taylor, D. J., Allison, M. and Ahn, C. (2015) ‘Erratum to: The strength of family ties: Perceptions of network relationship quality and levels of

Health inequalities in older age  231 c-reactive proteins in the north Texas heart study’, Annals of Behavioral Medicine, 49(5), p. 782. doi:10.1007/s12160-015-9720-5. Umberson, D., Crosnoe, R. and Reczek, C. (2010) ‘Social relationships and health behavior across the life course’, Annual Review of Sociology, 36(1), pp. 139–157. doi:10.1146/annurev-soc-070308-120011. Umberson, D. and Karas Montez, J. (2010) ‘Social relationships and health: A flashpoint for health policy’, Journal of Health and Social Behavior, 51(Suppl), pp. S54–S66. doi:10.1177/0022146510383501. Umberson, D. and Thomeer, M. B. (2020) ‘Family matters: Research on family ties and health, 2010 to 2020’, Journal of Marriage and Family, 82(1), pp. 404–419. doi:10.1111/jomf.12640. Villalonga-Olives, E. and Kawachi, I. (2017) ‘The dark side of social capital: A systematic review of the negative health effects of social capital’, Social Science & Medicine, 194, pp. 105–127. https://​doi​ .org/​10​.1016/​j​.socscimed​.2017​.10​.020. Wrzus, C., Hänel, M., Wagner, J. and Neyer, F. J. (2013) ‘Social network changes and life events across the life span: A meta-analysis’, Psychological Bulletin, 139(1), pp. 53–80. doi:10.1037/a0028601. Wu, Z. H. and Rudkin, L. (2000) ‘Social contact, socioeconomic status, and the health status of older Malaysians’, The Gerontologist, 40(2), pp. 228–234. doi:10.1093/geront/40.2.228. Yahirun, J. J., Sheehan, C. M. and Hayward, M. D. (2017) ‘Adult children’s education and changes to parents’ physical health in Mexico’, Social Science & Medicine, 181, pp. 93–101. doi:10.1016/j. socscimed.2017.03.034. Zhang, Y. and Fletcher, J. (2021) ‘Parental status in later life and parents’ risk of cognitive impairment’, SSM – Population Health, 16, p. 100968. doi:10.1016/j.ssmph.2021.100968.

PART IV LONG-TERM PERSPECTIVES ON THE DEVELOPMENT OF HEALTH INEQUALITIES ACROSS LIFE COURSE STAGES

16. Early childhood origins of modern social class health disparities Alberto Palloni, Daniel Ramirez and Sebastian Daza1

BACKGROUND There is a widespread consensus among historical demographers that socioeconomic differentials in health and mortality are a rather recent phenomenon in the history of homo sapiens. In a classic study, Hollingsworth (1977) argued persuasively that, as early as the 17th century, mortality differentials between the wealthiest and the destitute in England were either minuscule or the reverse of what they are now. More recent historical accounts of the demography of pre-industrial populations suggest that the emergence of mortality inequalities as a fixture of modern populations dates back at the earliest to the period when mortality began a secular decline (Antonovsky, 1967; Bengtsson and Dribe, 2011). Their initial manifestations roughly coincide with the beginning of a period of sustained gains in human height after 1800 (Fogel, 2012). Necessary conditions to support disparities surge throughout the 19th and first half of the 20th century, because of inequities in the social distribution of improvements in nutritional status and dissemination of new scientific knowledge. They were strengthened in the last quarter of the 19th century as a result of unequally allocated benefits from the diffusion of germ theory and massive public health works for sewage water purification and distribution. A new regime emerges after 1930 with the widespread dissemination of new medical technologies that drastically reduced the burden of infectious and parasitic diseases. Following a short-lived period of contraction, health disparities increased and then mutated to become characteristic of modern epidemiological regimes that followed massive reductions in mortality. It is indeed paradoxical that the most significant achievement in human history, the conquest of infectious diseases, became a new vehicle for the establishment of one of the most disfiguring features of modern societies, the gaping social class disparities in health status and length of life. Their receding importance, and the consequent improvements in survival to older ages, opened the gates to chronic illnesses, sensitive to an entirely new array of determinants that are also unequally distributed across social classes, including new exposures and new behaviors. Disparities associated with chronic illness would have grown because of larger fractions of the population reaching post-reproductive senescence (Medawar, 2019). Because fatality rates associated with post-reproductive diseases are tied to access to modern medical care, there is plenty of room for health and mortality disparities to persist and grow in any stratified societies with high life expectancy. We will argue below that modern health and mortality inequalities are powered by a unique set of pathways, the product of a peculiar combination of modern epidemiological regimes and social class systems. These pathways are by no means the only ones that account for modern disparities and, in some populations, may not even be the most important. Their uniqueness is rooted in that (i) they are only viable in populations characterized by modern stratification and epidemiological systems and (ii) can only be manifested over long periods of time, throughout the life cycle of individuals. Our goal 233

234  Handbook of health inequalities across the life course is to define a framework that includes these pathways as determinants of modern social class health and mortality disparities. The chapter is organized in six main sections. The first section describes the uniqueness of modern epidemiological and social stratification systems that combine to produce a modern regime of health and mortality disparities. The second section briefly reviews recent estimates of mortality differentials by education in the US and Europe and confirms that, over the past 20 years or so, health and mortality disparities have become a systemic feature of advanced societies. The third section defines a causal model that includes early direct and indirect conditions as explanatory factors, highlights the importance of early health and early acquisition of capabilities and skills, and points to the potential role for self-replication of disparities across successive generations. In the fourth section we quantify the contribution of direct and indirect pathways to mortality disparities associated with obesity (and metabolic diseases) and smoking. The fifth section briefly discusses the intergenerational transmission of health disparities and the last section summarizes and concludes.

EPIDEMIOLOGICAL REGIMES, SOCIAL CLASSES, AND HEALTH DISPARITIES The literature in epidemiology and population health has long since adhered to the idea that human populations have undergone multiple epidemiological transitions. In the standard rendition (Omran, 1971), there are three stages culminating in a shift from a regime of infectious and parasitic diseases to one dominated by chronic conditions. In more recent versions, a new epidemiological regime is added to represent populations in which illnesses affecting the very old become salient (emerging chronic illnesses, cognitive and neurological disorders) (Olshansky and Ault, 1986). Next, we focus on three dimensions of epidemiological regimes that are not explicit in the original epidemiological transition framework. Unequal Allocation of Early Conditions and of Opportunities of Class Accession In contrast to pre-modern regimes, adult health and mortality in modern epidemiological regimes can be associated with exposures whose manifestations may take a lifetime to be observed. The chain of events that leads from an initial risk of exposure to a final health outcome may take a very long time to unfold, could be heavily influenced by initial conditions experienced early in life, and may be modified to variable extents by subsequent events. Some infectious and parasitic diseases, such as rheumatic heart fever, HPV, respiratory TB, and some forms of malaria (plasmodium vivax), may also have delayed effects. But this is not typical of most infectious and parasitic diseases prevailing during the period 1750–1930. In modern epidemiological regimes, very low levels of infant and child mortality translate into large increases of the probability of surviving to critical ages after which adult delayed effects of early exposures may be observable. When child mortality levels are of the order of 200 per thousand, less than 45% of the population survives to age 60, in contrast to 95% in cohorts born after 1930. Because the quality of early exposures across a population is not random but strongly influenced by parental social class, they may have impacts on health and mortality disparities far exceeding those they had in preceding epidemiological regimes. This

Early childhood origins of modern social class health disparities  235 is reinforced in modern social stratification systems characterized by significant correlations between parental and offspring social class. Early exposures may alter processes of cell growth and differentiation in early embryonic life, fetal organogenesis, and epigenetic dysregulation, all related to maternal and parental exposures (including nutritional challenges and life stressors). In our framework, this source of disparities brings together two strands of empirical research that, until recently, grew disconnected from each other, namely, the analysis of life course determinants and the developmental origins of health and disease (DOHaD) (Gluckman, 2006; Gluckman, Buklijas, and Hanson, 2016; Gluckman and Hanson, 2004). Unequal Allocation of Early Resources that Minimize Health Risks and Optimize Social Class Accession Another peculiarity of modern epidemiological regimes is that they are shaped by conditions highly dependent on individual preferences, choices, and behaviors, some of which are rooted in early formative years and modified by socialization processes strongly influenced by family and class of origin. This, by itself, is enough to promote social class health inequalities. However, the mechanism is buttressed because modern stratification systems are founded on selective social class accession rules that favor individuals with skills and traits that happen to be associated with characteristics that minimize exposure to chronic illnesses. The net result is that those born in favorable social class positions are more likely to develop traits that simultaneously minimize health risks and favor their own social class standing. There will be a correlation between traits acquired early in life that influence both individuals’ opportunities of social class accession and their exposure to health risks. In an offspring generation, these traits are the outcome of genetic, epigenetic, and sociocultural inheritance from parents, family of origin, and school experiences. An important body of empirical research on some aspects of this source of disparities refers to it as health selection. Replication of Disparities: Heritability of Health-related Traits and Social Class In societies founded on a social class system, there are multiple mechanisms that generate a correlation between parental and offspring social class. Because some of these mechanisms are associated with early health conditions and trait formation and their interrelations, a regime of health and mortality disparities prevailing at one time will depend on its past, has a momentum of its own, and may persist for at least one generation, even after the conditions that originated it weaken or vanish. Thus, a modern epidemiological regime facilitates the intergenerational transmission of health disparities. There is a body of very rich research on the aforementioned sources of modern health disparities. But this research strand has grown separately from DOHaD and our contribution is to integrate them into a single framework that draws from past work in epidemiology and population health (Kuh and Ben-Shlomo, 2004), health and labor economics (Cunha and Heckman, 2007; Grossman, 1972, 2000), and the literature on sociocultural heredity (Bonduriansky and Day, 2018; Cavalli-Sforza and Feldman, 1981). Emphasis on these sources of disparities (and associated mechanisms) is neither meant to ignore direct, causal, impacts (see below) nor serve as a distraction from the importance of social class as a “fundamental” determinant (Link and Phelan, 1995). Health disparities are an inevitable product of stratified social systems, not

236  Handbook of health inequalities across the life course just the result of weakly related components that can be understood separately from each other. Our objective is to highlight mechanisms that are unique in modern stratified social systems and that happen to coexist with a modern epidemiological regime.

HEALTH AND MORTALITY INEQUALITIES: HOW LARGE ARE THEY? Analysis of health and mortality inequalities has been carried out on multiple health outcomes and using a broad array of metrics of individuals socioeconomic ranking. Indicators based on mortality rates have been the workhorse in this literature whereas social class and socioeconomic ranking are traditionally assessed with occupation categories, occupational prestige scores, income, wealth, educational attainment, self-assessments, and combinations thereof. There is agreement, however, that estimates of educational disparities are the most robust to multiple pitfalls, are moderately comparable across populations and over time, and are most abundantly documented in the literature. Magnitude of Recent Mortality Disparities by Education in the US and Europe In the online Appendix to this chapter,2 we compute harmonized estimates of life expectancy at age 45 and associated relative mortality risks. These calculations include data from the US and selected European countries and refer to birth cohorts and cross-sections whose mortality experiences reflect conditions experienced after 1950 approximately. These estimates lead to two main inferences. US adult mortality educational disparities reflected in life expectancy at age 45, E(45), are relatively large, they hover in the range 3.0 to 6.0 years and, on average, suggest that those with lowest education (less than high school) experience residual lifetimes 10–20% shorter than those with more than high school education. These disparities are ubiquitous across the life span, graduated by education categories and, finally, increasing over time and across birth cohorts. A review of recent estimates of mortality disparities by education in multiple countries in the EU point to inequalities in life expectancy and relative mortality risks at age 45 of strikingly similar magnitude to those found in the US, they also point to an increasing trend in the most recent periods and, finally, are ubiquitous although variable across countries. Causes of Death Responsible for Mortality Disparities A detailed accounting of disparities by causes of deaths is beyond the scope of the chapter. However, there is strong empirical evidence showing that the main causes of death in modern mortality regimes are also responsible for the bulk of education mortality differentials (Glei, Lee, and Weinstein, 2020). These include ischemic heart disease (CHD), hemorrhagic and ischemic stroke, cancers (breast, bronchial and lung, bladder, liver), Type 2 diabetes and, finally, accidents, homicides and suicides (AHS) (Mackenbach, 2006). With the exception of AHS, these are all chronic conditions whose causal roots are multifactorial and have been associated with environmental exposures of various kinds and individuals’ life styles. Although this remains to be investigated more thoroughly, there are two features of modern

Early childhood origins of modern social class health disparities  237 individual life styles that may ultimately account for a significant fraction of cause-specific mortality disparities: obesity and smoking persistence.3 Smoking is known to be a behavior that starts very early in life and is responsive to parental, kin, and peer influences. Obesity, on the other hand, is a condition influenced by a host of allelic variants, maternal obesity before and during pregnancy, infant and childhood diet, adolescent food preferences, and physical activity. Note that in the US at least, smoking accounts for most of socioeconomic disparities in mortality from smoking-related diseases. In addition, smoking alone contributes to about 38% of cardiovascular disease mortality and 34% of all-cause mortality. Thus, a low bound estimate of the contribution of smoking to overall mortality differentials is about 21% (Glei, Lee, and Weinstein, 2020). A similar, if not a higher, figure must apply to obesity and, as a consequence, it may well be that jointly these two conditions could account for over 40% of observed disparities. Because disparities in smoking and obesity prevalence have been widening over time, their contribution to overall disparities must be increasing also. In short, identifying the role that early conditions may have as triggers of smoking adherence and behaviors that induce obesity, may go a long way in explaining mortality disparities documented before. We end this section by noting that it is nothing short of remarkable that highly heterogeneous populations that differ in their social class stratification, central government organization, medical health care systems, and the reach of social programs enacted to level the playing field, all exhibit roughly similar levels and trends in mortality disparities and comparable contributions of causes of death. Below, we argue that this is the result of conditions prevailing in all stratified social systems that share a modern epidemiological regime.

A CAUSAL MODEL FOR HEALTH DISPARITIES We now introduce a model that identifies mechanisms responsible for modern social class disparities emphasizing direct and indirect pathways that originate in early exposures. We also discuss conditions required for these pathways to be efficient vehicles for the production and replication of social class disparities. The model represents chains of events over the life-course, highlights intergenerational links, and emphasizes the relevance of cognitive and non-cognitive skills (human capital). The variables in Figure 16.1 represent constructs measured at some point in the life course, while the arrows connecting them represent causal relationships.4 In high-income societies at least, the bulk of mortality disparities is attributable to chronic illnesses (Mackenbach, 2006). At least part of the observed association between social class and adult health and mortality associated with chronic conditions have roots in processes that start early in life and influence both social class accession and determinants of health. In Figure 16.1, we distinguish between parents’ social class and childhood environments (including pregnancy and maternal conditions). We use a rather broad definition of childhood environment or early life conditions and include factors that can jeopardize, impose constraints on, or enhance children’s early development. For example, the quantity and quality of maternal care, environments experienced at home (parental education, family income, welfare, stress, diet and physical activity, alcohol and tobacco exposure), and school experiences. We also include exogenous events that may alter children’s life trajectories, such as wars, famines, or social contexts such as neighborhood poverty, violence, and crime. For the sake of simplicity,

238  Handbook of health inequalities across the life course

Figure 16.1

Life-course causal model for health disparities

the model ignores the timing of events and duration of exposures (in-utero, throughout infancy or childhood, or during adolescence), even though some of their impact is highly sensitive to these dimensions. Figure 16.1 includes two different sets of pathways, direct and indirect. We describe each in turn. Direct Pathways that Originate in Early Conditions There is abundant empirical evidence showing that adverse environments during the fetal period, infancy, and early childhood increase the adult risk of chronic illnesses such as cancers, and metabolic and cardiovascular diseases (see pathway 1 in Figure 16.1) (Gluckman, Hanson, and Beedle, 2007). The mechanism depends on parental factors that influence preconception exposures, in-utero, and post-natal conditions associated with nutritional challenges and stress response. Some of these involve epigenetic modifications, others include alteration of cell specialization and growth. In all cases, they may have a long reach depending on the timing of exposures and individuals’ resilience. Parental circumstances and early life play a significant role, since adversity that occurs during these critical periods can permanently set health trajectories in ways that equivalent exposures experienced outside of those periods do not. Maternal conditions prior to gestation can affect offspring early health, and influence cognitive and non-cognitive skills independently from adversity experienced during gestational, infancy and childhood periods. Additionally, there is considerable evidence suggesting that parents’ genetic predisposition to deleterious health conditions influences offspring risks of adverse health and undermines cognitive and non-cognitive trajectories. Aside from parental circumstances or genetics, adversity encountered while in the womb and during early infancy will compound the effects of parental circumstances prior to gestation. Table 16.1 is a highly stylized summary of the nature of sources, mechanisms and health outcomes related to exposures experienced preconception, in-utero and during infancy and early childhood. The table is maximally simplified as, among other things, it only highlights a very selected set of direct pathways leading from early exposure to adult delayed effects.

Early childhood origins of modern social class health disparities  239 Table 16.1

Classes of direct pathways related to obesity and T2D

Source

Child outcome

Adult outcome

Mechanisms

References

Maternal prepregnancy

Child obesity

Obesity

Genetic

Oken et al., 2009

BMI, gestational

Metabolic dysregulation T2D

Epigenetic

Simeone et al., 2015

Physiological

weight gain, gestational diabetes Fetal caloric nutritional

Child obesity

Obesity

Homeostasis

Barker et al., 1989

restriction

Organosesis

T2D

Epigenetic

Gluckman, Hanson and

irregularities

CHD

Fetal programming

Beedle, 2007

Metabolic dysregulation HBP

PAR

Kidney diseases COPD Fetal nutrient imbalance Adipogenesis irregularities

Obesity

Epigenetic

T2D

Gernand et al., 2016 Gertler and Gracner, 2021

Metabolic dysfunction

Patti, 2013 Maternal, fetal, infant

HPA dysregulation

stress

Obesity

Epigenetic

Entringer, Buss, and

T2D

Physiological

Wadhwa, 2015

Cognitive and

Kuzawa and Quinn,

non-cognitive

2009

limitations

Thayer and Kuzawa, 2015

Maternal smoking

HPA dysregulation

Obesity

Epigenetic

Al-Amri et al., 2021

T2D

Physiological

Mattsson et al., 2013

Respiratory disease Cognitive and non-cognitive limitations

They share one property: they are all associated with adult conditions that account for an important fraction of chronic conditions associated with modern health and mortality differentials, namely, obesity, Type II diabetes (T2D), stroke and CHD. Indirect Pathways that Originate in Early Conditions Early conditions may influence the acquisition of early skills and capabilities (paths 2, 3). Maternal behaviors right before and during pregnancy, fetal environments, conditions in infancy and early childhood, can affect processes of cell growth and specialization and, through them, have an impact on connectivity of brain tissue, on which the development of cognitive and non-cognitive capabilities depends. As a result, these early experiences may have a long reach. First, in modern stratification systems, educational attainment strongly affects the allocation of individuals in social classes. It happens to be the case that early cognitive (and some non-cognitive) skills will partially determine adolescent cognitive skills (path 14), adult education and, through it, an individual’s social class (path 16). Second, cognitive and non-cognitive abilities are strongly and independently associated with a range of factors that influence health outcomes (path 15). Most importantly, among these, are those involved in the formation of individuals’ health behaviors on which the timing

240  Handbook of health inequalities across the life course and duration of exposure to risks of chronic illness depends. They also regulate preferences and choices that may interfere with individuals’ ability to access resources that enhance resistance and recovery from disease. As a result, early conditions and experiences may constrain the repertoire of skills an individual will command, derail education trajectories and, simultaneously, reduce the array of resources available to minimize exposures or improve resistance to and recovery from illness. Thus, the link between early health and cognitive and non-cognitive skills (path 13) and the effect these have on both adult health (paths 15, 17) and social class (paths 16, 18) is a pathway for health and mortality disparities. In past empirical research this has been referred to as indirect health-selection. There is a robust body of empirical evidence showing that this pathway could be partly responsible for the education (or any social class) health gradient (Manor, Matthews, and Power, 2003; Power and Jefferis, 2002; Stern, 1983; West, 1991).5 Other Causal Pathways Although our interest revolves around direct and indirect pathways rooted in early conditions, modern health disparities have other sources. Most importantly, membership in a social class by itself constrains access to resources that minimize exposures to risks and augment resistance and recovery (path 19). These effects do not operate through childhood environment and early health, genetic predispositions, or cognitive/non-cognitive skills. Their impact is causally direct in the sense that it stems from differential access to resources available to a social class. These include wealth, income, education, all of which facilitate purchase in open markets of medical care, favorable residential settings, health insurance, and health information (Cutler and Lleras-Muney, 2010). Some of these resources might influence the formation of adult attitudes, preferences, tastes, choices, or behaviors that minimize risk exposures. Research on educational mortality disparities attribute some of the effects to resources endowed to those who attain some level of accreditation by virtue of the accreditation itself (Halpern-Manners et al., 2020). Thus, educational attainment facilitates acquisition of personal traits and capabilities such as self-efficacy (Mirowsky and Ross, 2005) implicated in the adoption of beneficial health behaviors (paths 23, 24). This is different from the alternative explanation we favor here, in which the association between education and mortality is a result of complex pathways that produce both low mortality risks and high educational attainment. Although empirical evidence supports both approaches, the empirical evidence is partial, as it only applies to some subpopulations, and incomplete, as the models tested do not include all the relevant pathways. Manifestation of Pathways As it stands, the causal model we formulated above is incomplete and should be extended in two directions. First, health and mortality differentials are the result of mechanisms that minimize exposure to illness (E) and maximize individual capabilities of resistance (R) to and recovery (Re) from illness over the life-course. Any indicator of health status or stock we care to use is invariably the joint result of exposure, resistance, and recovery (ERRe). But since each dimension of the triplet ERRe may be the result of partially independent causal processes, using a shortcut to bundle them into a single outcome will necessarily obscure important mechanisms. For example, the impact of unequal access to modern medical care, an increasingly important factor accounting for modern disparities, is only relevant to the extent

Early childhood origins of modern social class health disparities  241 that it affects the R or Re dimensions. This, in turn, may be a function of pathways involving early health and cognitive and non-cognitive capabilities. Second, the explanatory power of a mechanism generating disparities is a function of the precise nature of the dominant epidemiological regime. As a result, health and mortality disparities of similar magnitudes in different populations or historical periods could be the result of different processes. Any general model of health disparities should include parameters that regulate changes in the nature of gradients as a function of shifts in epidemiological regimes. A formal rendition of the model in Figure 16.1 could be as follows: 





H i  t   hEit * Z1,i1,Ht k  * Z 2,2i,Ht k  * Z 3,i3,Ht k  (16.1)

Ei  t   eFit k  *W1,i1,tHk  *W2,i2,tHk  * Z 3,3i,Ht k  (16.2) 





where H and E are health status and education attainment, the Zs and Ws are vectors of early health conditions (1), cognitive (2) and non-cognitive (3) traits, all fixed at time t − k and sharing at least some common elements. F is a vector of family characteristics, h and e are additive effects of exogenous factors (including random errors) and, finally, Greek letters are elasticities which are implicit functions of characteristics of the prevailing epidemiological regimes. For example, a shift from a regime dominated by infectious diseases to one dominated by chronic illness is manifested in an increase first of σ1,H and then of σ2,H and σ3,H. Similarly, in modern stratification systems the magnitudes of the θ’s elasticities should be larger than in stratification systems that constrain social class mobility. Modern populations that are combinations of different epidemiological and stratification systems will be characterized by different values of σs and θs. A final but important feature is that the size of ε, the direct causal effect of E, is likely to be a function not only of epidemiological and stratification regimes, but also of some of the early inputs included in vector Zs and Ws. For example, the executive capacity with which highly educated people may be endowed is likely to be dependent of early non-cognitive capabilities, such as self-reliance and perseverance.

ILLUSTRATIONS OF DIRECT AND INDIRECT PATHWAYS: OBESITY, TYPE II DIABETES (T2D) AND SMOKING In this section we review empirical estimates of the contribution to education mortality disparities that could be associated with two of the multiple causal pathways described before. First, we show that a direct pathway from early conditions influences adult risks of obesity and metabolic diseases and can account for a significant fraction of the education mortality gradient. Second, we illustrate the impact of one (out of many) indirect pathways involving early non-cognitive traits related to smoking behavior and estimate the magnitude of mortality disparities associated with it. In both cases, we only focus on key results and leave details of statistical analyses to the online Appendix. Education disparities in the prevalence of obesity are quite large and vary widely across populations. In the US older adult population (50 and above), about 30% of those with less than HS education are obese versus 19% among those with more than HS (see online Appendix).

242  Handbook of health inequalities across the life course These gross disparities in obesity conceal heterogeneity associated with population ancestry. This is due to recent migrant flows from West Africa and North-Central America, in the case of US and Canada, and from West and East Africa as well as South Asia, in the case of some Western European populations. We emphasize this distinction because direct pathways that increase obesity risks differ by population ancestry. In fact, first generation migrants from Asia, Africa or Latin America to North America or Europe are the subpopulations that face the largest contrasts between ancestral and modern environment and are thus at highest risk of manifesting health outcomes predicted by the Predictive Adaptive Response (PAR) conjecture (Gluckman, 2006). These populations are concentrated in the lower social classes and are simultaneously exposed to a whole array of social mobility obstacles. Consequently, it is in these populations that the direct mechanisms described above will operate maximally to produce social class disparities. Instead, the local, native ancestry subpopulations are more likely to avoid PAR and experience instead risks associated with maternal health status, maternal nutrition, and early exposure to familial obesogenic environments. Below, we quantify the relevance of direct mechanisms associated with obesity only among local populations. Consequently, the resulting estimates must be lower bounds of the target quantities. We use the Health and Retirement Survey (HRS) and estimate the contribution of an indicator of adverse early conditions to observed education mortality differentials associated with adult obesity and T2D (paths 1, 12 and 1, 10, 22 in the model). We use the elderly adult population of the HRS and compute the difference in life expectancy between those with no high school degree (LHS) and those with some college (MHS). We then estimate the impact of an indicator of adverse early conditions on the risk of old adult obesity, separately by education category. In addition, we calculate the relative risk of T2D among the obese and non-obese and, finally, compute the relative risk of mortality associated with T2D. Under some assumptions spelled out in the online Appendix, these computations yield proper estimates of the fraction of differences in life expectancy at age 45 between those with no high school and those with more than high school education that is attributable to the effects of poor early conditions on adult obesity and T2D. Education mortality disparities in the HRS survey are stark. In the sample followed since 1992, those with LHS experience mortality risks twice as high (2.08) relative to those with MHS. In terms of life expectancy at age 45, E(45), this translates into a disparity of about 4.1 years of residual life, very close to the estimates from independent sources described before and for a similar period of time. The issue of interest to us is this: how much of this difference can be attributed to early conditions? Since this is too broad a question, we narrow it down further and ask: how much of the difference can be attributed to the effects that early exposures have on obesity and of a single chronic illness associated with obesity, namely, T2D? The sequence of required steps to arrive at suitable estimates is fully described in the online Appendix. Here we only review key results. First, we estimate that those with the poorest values on the indicator of early conditions who were alive in 2006, experience an excess probability of becoming obese before 2017 that is 1.72 times as high relative to those with better values of the same indicator. Because all variables contributing to the indicator of early conditions are set before completing education, it is legitimate to assess its association with educational attainment. It turns out that about 78% of those with LHS are classified in the worst category of the early conditions indicators versus 38% among those with MHS. Second, according to HRS data, being obese increases the risk of T2D almost twofold. If we conservatively assume that this relative risk does not vary with early conditions (or education)

Early childhood origins of modern social class health disparities  243 and, in addition, that the excess mortality risks associated with T2D do not vary with early conditions (or education) and obesity status, we can compute mortality risks by early conditions in each education group. We then convert these relative risks into a metric of life expectancy at age 45, E(45).6 Figure 16.2 displays values of life expectancy at age 45, E(45), under three alternative scenarios labeled MINobesity, HRSobesity, and MAXobesity. These scenarios correspond to the initial levels of obesity prevalence in HRS. The scenario HRSobesity utilizes observed values in 2006 whereas MINobesity and MAXobesity reduce (increase) the observed values by 15% respectively. The Y-axis is for values of E(45). The X-axis is for alternative values of the relative risks of obesity associated with adverse early conditions (lower values correspond to higher risks). Because estimates are sensitive to initial levels of adult prevalence in HRS, the figure includes three lines whose points were computed using different assumptions about the age pattern of obesity prevalence in the 2006 HRS. The highest line assumes levels of obesity prevalence 10% lower than observed in 2006 HRS. Those defining the middle line assume a prevalence identical to that observed in HRS. Finally, those in the lower line assume levels of prevalence 10% higher than observed. In addition, the figure includes two horizontal lines. The lower one is drawn at the value of E(45) observed among those with LHS. The highest horizontal line is drawn at the value of E(45) observed among those with MHS. The difference between these two horizontal lines is about 4.1 years, the original estimate of education adult mortality disparity. Points along the three non-horizontal lines represent values of E(45) that would be attained by the LHS group if everything except differential allocation of early conditions were the same across education groups. One way to read the figure is to focus on the difference between any point in any of the three lines and the top horizontal line: this

Figure 16.2

Predicted life expectancy at age 45, E(45), associated with T2D excess mortality risks (data from US HRS)

244  Handbook of health inequalities across the life course difference is equivalent to the education disparity attributable to effects of early conditions operating via obesity and T2D. In the case when we use the observed prevalence of obesity in HRS (diamond point), this difference is approximately years (56.9 − 55.5) or, equivalently, about 34% (1.4/4.1) of the observed education disparity (difference between the two horizontal lines). This is not a small number. Furthermore, because of the assumptions we made before to simplify computations, the actual contribution of early conditions to adult mortality disparities might be even more sizable. Indirect Pathways: Smoking This section describes the influence that indirect pathways originating in early conditions may have on sustaining social class disparities. We focus on paths 3, 20, 16 and 3, 20, 18. These are pathways which involve exposure to early conditions that result in preferences and behaviors that influence both educational attainment and health conditions. Educational contrasts in smoking behavior are even starker than those for obesity. In the US, the fraction of adults 25 and older who are current smokers is of the order of .35 among those with LHS and .16 among those with MHS (Cornelius et al., 2020) or odds of the order of 3. In the UK the figures are .22 and .10 respectively. A 2000 report on smoking in Europe shows that differences among adults older than 20 are of similar order of magnitude though they vary broadly by countries: from odds ratios as high as 2.5 in Norway to values below 1 in Portugal (Cavelaars, 2000). Past research on the relations between early upbringing, smoking, and mortality has unveiled abundant and robust evidence for two regularities: i. Smoking and adult mortality: The linkage between smoking and mortality is well established and, despite some remaining uncertainty regarding magnitudes, we know that smoking significantly increases the risk of CHD, stroke, and cancer to several sites (breast and lung and upper respiratory tract), the three most powerful causes of adult mortality in modern populations (CDC, 2021). ii. Smoking and early upbringing: It is known that smoking is a behavior adopted early, difficult to abandon, and that its effects are felt only after long latency periods. For the most part, tobacco addiction is acquired during early adolescence, is influenced by parental, sibling, and peer smoking, and its initiation and persistence are tightly associated with individuals’ future outlooks and time preferences, risk aversion, self-control, and discipline. Some of these factors also contribute to sort out who graduates from high school and who drops out as well as those who are successful in the labor market and those who are not (Simmons-Morton et al., 2001). iii. Education and early upbringing: Educational attainment is a prime example of a mortality determinant that is highly sensitive to cognitive and non-cognitive traits. The connection between cognitive traits and education is well-established and so is the importance of non-cognitive traits such as industriousness, discipline, and perseverance (Cunha and Heckman, 2007). At least some of these traits are involved in the formation of individual preferences, tastes, and choices that also influence health-related behaviors, and these, in turn, shape the individual exposure–resistance–recovery triplet. Protective health behaviors, such as avoidance of smoking, moderate or no alcohol consumption, healthy diets, and physical activity are more likely to be adopted by individuals who have strong future

Early childhood origins of modern social class health disparities  245 outlooks and low discount rates of the future. These same time preferences influence decisions about schooling continuation, training, and acquisition of skills. From the above, we conjecture that it is possible that at least part of the association between education and adult health and mortality is due to the fact that both educational attainment and adult health status are influenced by time preferences set early in life (paths 3, 20, 17).7 Individual time preferences are just one in an otherwise much larger bundle of health and education-related cognitive and non-cognitive traits that are partially or totally transferred from parents to offspring. They are all part of the sociocultural contribution to individual phenotypes (paths 2, 3) to which also contribute genes and epigenes (path 4) and non-family environments (Cavalli-Sforza and Feldman, 1981). Just how large can the contribution of indirect pathways be? To offer an approximate answer to this question, we use the 1958 British Cohort and estimate the contribution a single early behavioral trait to adult health and mortality disparities associated with smoking. We choose a single trait, T, in the offspring generation, namely, having a diagnosis of experiencing behavioral problems or “externalizing behaviors” at ages 6 and 11. Estimates from a model estimated with NCDS data shows that this trait influences both offspring smoking and educational attainment. In addition, T is unequally distributed by offspring social class of origin, being much more prevalent in lower social classes (see online Appendix for details of the model and statistical analysis). Life expectancy at age 50, E(50), in the UK is of the order of 31.70 and 28.90 years for high and low educated individuals, giving a total disparity of about 2.8 years. The question we pose is as follows: how much can this disparity be reduced if the impact of T on smoking is suppressed? To assign a numeric value to this quantity we use estimates of three parameters: effects of T on education, effects of T on adult smoking, and effects of education on smoking. We then perform simple simulations that combine these parameter values with alternative values of T’s prevalence. To increase realism, we also choose a range of values of parameters centered on the point estimated in NCDS data. Combining values of T and of parameter estimates, we perform a total of 50 simulations and choose those that yield values of E(50) for each education group that differ by less than one year relative to those observed in the population. We then use the simulations thus selected and compute counterfactual values of E(50) for the lowest education group that would be observed if either the prevalence of T was set equal in both education groups or if the effects of T on smoking were set to zero (see the online Appendix for details about the model used, parameter estimates and simulations). Figure 16.3 displays the fraction of reduction of education disparities in E(50) that would be obtained if either the effects of T on smoking were eliminated (triangles) or the prevalence of T were set equal in both education categories (“x”). These two quantities, (y-axis), are plotted against the absolute value (in years) of the difference between E(50) observed and predicted in each simulation that produced the quantities. Preferred simulations yielding errors below 5% are enclosed in a rectangle. These yield counterfactual reductions in the range .05–.15. This means that, under the estimated model and simulated data, one would expect that education disparities in E(50) would be reduced between 5% and 15%. As in the case of obesity studied before, these contributions are not small and suggest that observed education mortality disparities could feasibly be rooted in at least one indirect mechanism. As before in the illustration with obesity, we are probably erring on the conservative side and downplay the role of indirect mechanism. First, there are determinants of smoking other

246  Handbook of health inequalities across the life course

Figure 16.3

Education differences in life expectancy at age 50, E(50): total and associated with smoking (data from the UK NCDS-1958)

than early behavior problems and some of them may be sensitive to unequally distributed other early traits. Second, these estimates ignore the contribution of parental effects and heritability, including those associated with maternal smoking during pregnancy and with early family-shared environments. Third, the health impacts of smoking depend on a host of factors, including age of onset, dosage, regularity, and duration. It is unlikely that conditions that influence initiation are the same as those that affect desistance. Nor can smoking behavior be easily decoupled from other health-related behaviors, such as alcohol intake and diet, that have independent additive effects and are influenced by non-overlapping determinants.

INTERGENERATIONAL TRANSMISSION OF DISPARITIES: DIRECT AND INDIRECT PATHWAYS Both direct and indirect pathways rooted in early conditions may produce disparities in one generation and can also contribute to their self-replication in subsequent ones. Below we discuss different forms of heritability associated with obesity and metabolic disorders, on one hand, and smoking, on the other. Direct Pathways and Intergenerational Transmission There are three sources of heritability of disparities associated with direct pathways increasing risks of obesity and metabolic disorders. The first source is via sociocultural transmission. In fact, parental and offspring diet and physical activity preferences, the two most important behavioral drivers of obesity, are closely related. To the extent that the phenotype is unequally

Early childhood origins of modern social class health disparities  247 distributed by social classes in one generation, it will continue to be so in subsequent ones when paired with imperfect social class mobility.8 The second mechanism involves epigenetic changes. There are many animal studies that demonstrate that alteration of the macro and micronutrient composition of the maternal diet can exert powerful influences on the hypo (hyper) methylation and other types of chromatin changes. This, in turn, “programs” fetal body type and risks of metabolic disorders (Bateson and Gluckman, 2011). Although similar results have been obtained in human populations subjected to several types of nutritional constraints and stresses of the maternal diet or the balance of nutrients, the empirical evidence pertains to very selected populations and is difficult to generalize. The involvement of the epigenome is sensitive to deficiencies of some micronutrients that are themselves, or participate in the production of, methyl donors and imbalanced diet composition (e.g. carbohydrates excess). This suggests that inequities in allocation of resources that promote a balanced maternal diet will reinforce conditions to develop obesity and metabolic disorders. In addition, it is possible that the epigenetic marks that trigger obesity and metabolic disorder risk in the fetus, infant, and child can be reliably inherited across at least two generations. If this is the case, replication of disparities is enhanced even if the original source (unequal distribution of nutritional deficiencies) disappears (Patti, 2013; Pembrey, 2010; Pembrey et al., 2006). Indirect Pathways and Intergenerational Transmission The most direct route through which indirect pathways may reproduce inequalities across generations is by increasing the risk of smoking via shared maternal and parental environments. First, although the mechanisms are unclear, there is some evidence that maternal smoking during pregnancy increases risks of addiction in the newborn (Blood-Siegfried and Rende, 2010; Niaura et al., 2001). If this is the case, smokers’ offspring are exposed to a double risk: to become smokers as adults and to have diminished cognitive abilities due to in-utero induced damage. Second, there is a large body of evidence supporting the idea that an individual’s likelihood of smoking is higher among those whose parents or siblings smoke. This is a classic form of sociocultural inheritance of a health behavior. To the extent that smoking behavior is unequally distributed across classes, this route alone could induce a replication of disparities across generations in stratification regimes that have restricted social mobility. There is, however, another gateway which reinforces replication. This is that membership in social classes in which smoking is more prevalent may carry an additional risk associated with the transmission of non-cognitive factors, such as poor outlooks and heavy discount of the future, that reduce the opportunities for the acquisition of skills that promote social class mobility even in stratification systems with high mobility.9

CONCLUSION AND DISCUSSION We propose a framework that accounts for systemic and growing social class health and mortality disparities in modern populations. Within the proposed framework we elaborate a theoretical model that emphasizes direct and indirect pathways rooted in early exposures. The model has room to account for differential sensitivity of observed disparities to secular

248  Handbook of health inequalities across the life course shifts or moderate alterations of the epidemiological and stratification regimes in a population. Using the example of obesity (metabolic and cardiovascular illnesses) and smoking (cancers and cardiovascular illnesses), we showed that attention to the direct and indirect pathways as well as to their power of replication across generations, can augment our ability to account for current modern disparities and past trends. Although we do not show directly how the framework would strongly support actual empirical research, the examples of obesity and smoking leave enough clues about how to do so. The chapter has several shortcomings. First, as is done in most of this literature, we ignore the importance of separating the three components that contribute to health status, namely the triplet ERRe. Doing so may simplify the descriptive task in a chapter like this, but it is not best practice in a research agenda on the subject. Second, we did not discuss in much detail the role of direct, causal, impacts of social class. This is because these have been thoroughly examined in most of this literature and we wished to highlight pathways that have been less addressed despite the fact that their importance may have increased over time. Third, we did not consider the potential influences that policy interventions, either local or national, may have and under what conditions they may do so. If, for example, the bulk of health and mortality disparities by education were traceable to differential obesity, would a policy intervention targeting inequities in access to modern medical care to deal with T2D have durable effects? And how large could these be? Would a policy targeting instead maternal nutrition and BMI be more effective in the long run? The model must be adapted to ensure that, just as it makes explicit that the power of some pathways is a function of epidemiological and stratification regimes, the reach of some of these interventions may be equally dependent on the power of those pathways. Finally, illustrations of the efficiency of direct and indirect pathways are based on only two data sets from two quite singular populations. Because of this, strong generalization of our findings is inappropriate but we believe they do serve to prove the point that these mechanisms can, under some conditions, be influential.

NOTES 1. This project has received funding from the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation program (grant agreement No 788582). This publication reflects only the authors’ view and the Research Executive Agency and the Commission are not responsible for any use that may be made of the information it contains. In addition, Palloni also acknowledges research support from the National Institute on Aging (https://​www​.nia​.nih​.gov/​), the National Institute of Child Health and Development (https://​www​.nichd​.nih​.gov), and the Fogarty International Center Global Research Training in Population and Health (https://​www​.fic​.nih​.gov) via the following project grants R01-AG016209, R03-AG015673, R01-AG018016, R37-AG025216, RO1-AG056608, RO1-AG052030; D43-TW001586, R24 HD047873 P30-AG-017266. 2. https://​www​.e​-elgar​.com/​textbooks/​hoffmann 3. Alcohol intake plays a less important role but is also included as part of the array of behaviors that make up modern “life styles”. 4. This is a simple graphical causal model, a directed acyclic graph (DAG), and is meant to be a heuristic tool, not a detailed causal model. 5. Perhaps the most influential theoretical work in this area is by Grossman (1972, 2000). More recent contributions are the work by Conti et al. (2010a, b), Hoffmann et al. (2018, 2019), Mackenbach (2012), Palloni (2006) and Palloni et al. (2009). 6. It bears repeating that the assumptions just made about independence between conditional risks of T2D and adult mortality among diabetics, on the one hand, and early conditions and education, on

Early childhood origins of modern social class health disparities  249 the other, are violated in the HRS data. As a result of this violation, we will estimate lower bounds of the contribution of early conditions to disparities in adult mortality. 7. The mechanism is likely to be more complicated than this. Thus, some empirical evidence suggests that time preferences are themselves influenced by educational attainment (Becker and Mulligan, 1997; Oreopoulos and Salvanes, 2011; Perez-Arce, 2017). 8. It is important to note that food preferences at least have been shown to be influenced in-utero and involve changes in the fetal epigenome (Bateson and Gluckman, 2011; Bonduriansky and Day, 2018). This reinforces the sociocultural mechanism. 9. There is, of course, a final source for the replication of inequalities associated with indirect pathways and smoking, the genetic route. There is some empirical evidence of alleles that increase the risks of nicotine addiction. If these also have pleiotropic effects and influence levels of cognitive and non-cognitive capacities, a genetic source for the correlation between health status and social class would be established. We know of no direct empirical evidence suggesting the existence of such allelic variants.

REFERENCES Al-Amri, I. S. et al. (2021). “Effects of Prenatal Tobacco Smoke on Cerebellum Histomorphological Changes at Critical Developmental Stages in CD-1 Mice”. In: International Journal of Morphology 39.1, pp. 318–326. Antonovsky, A. (1967). “Social Class and Illness: A Reconsideration”. In: Sociological Inquiry 37.2, pp. 311–322. Barker, D. J. et al. (1989). “Weight in Infancy and Death from Ischaemic Heart Disease”. In: Lancet (London, England) 2.8663, pp. 577–580. Bateson, P. and P. Gluckman (2011). Plasticity, Robustness, Development and Evolution. Cambridge University Press. Becker, G. S. and C. B. Mulligan (1997). “The Endogenous Determination of Time Preference”. In: The Quarterly Journal of Economics 112.3, pp. 729–758. Bengtsson, T. and M. Dribe (2011). “The Late Emergence of Socioeconomic Mortality Differentials: A Micro-level Study of Adult Mortality in Southern Sweden 1815–1968”. In: Explorations in Economic History 48.3, pp. 389–400. Blood-Siegfried, J. and E. K. Rende (2010). “The Long-Term Effects of Prenatal Nicotine Exposure on Neurologic Development”. In: Journal of Midwifery & Women’s Health 55.2, pp. 143–152. Bonduriansky, R. and T. Day (2018). Extended Heredity: A New Understanding of Inheritance and Evolution. Princeton: Princeton University Press. Cavalli-Sforza, L. L and M. W. Feldman (1981). Cultural Transmission and Evolution. Princeton, N.J. Cavelaars, A. E. J. M. (2000). “Educational Differences in Smoking: International Comparison”. In: BMJ 320.7242, pp. 1102–1107. Centers for Disease Control (CDC) (2021): https://​www​.cdc​.gov/​tobacco/​data​_statistics/​fact​_sheets/​ health​_effects/​effects​_cig​_smoking/​index​.htm Conti, G., J. J Heckman, and S. Urzua (2010a). “The Education–Health Gradient”. In: American Economic Review 100.2, pp. 234–238. Conti, G., J. J. Heckman, and S. Urzua (2010b). “Early Endowments, Education, and Health”. Working Paper No 2011-001, Human Capital and Economic Opportunity: A Global Working Group. October 2011. Cornelius, M. E. et al. (2020). “Tobacco Product Use among Adults – United States, 2019”. In: MMWR Morb Mortal Wkly Rep 69, pp.  1736–1742. DOI:  http://​dx​.doi​.org/​10​.15585/​mmwr​.mm6946a4. Cunha, F. and J. Heckman (2007). “The Technology of Skill Formation”. In: American Economic Review 97.2, pp. 31–47. Cutler, D. M. and A. Lleras-Muney (2010). “Understanding Differences in Health Behaviors by Education”. In: Journal of Health Economics 29.1, pp. 1–28.

250  Handbook of health inequalities across the life course Entringer, S., C. Buss, and P. D. Wadhwa (2015). “Prenatal Stress, Development, Health and Disease Risk: A Psychobiological Perspective – 2015 Curt Richter Award Winner”. In: Psychoneuroendocrinology 62, pp. 366–375. Fogel, R. W. (2012). Explaining Long-Term Trends in Health and Longevity. Cambridge University Press. Gernand, A. D. et al. (2016). “Micronutrient Deficiencies in Pregnancy Worldwide: Health Effects and Prevention”. In: Nature Reviews Endocrinology 12.5 (5), pp. 274–289. Gertler, P. and T. Gracner (2021). “The Sweet Life: The Long-Term Effects of a Sugar-Rich Early Childhood”. NBER Working Paper 30799 – http://​www​.nber​.org/​papers/​w30799. Glei, D. A., C. Lee, and M. Weinstein (2020). “Socioeconomic Disparities in U.S. Mortality: The Role of Smoking and Alcohol/Drug Abuse”. In: SSM – Population Health 12, p. 100699. Gluckman, P. D. (2006). Developmental Origins of Health and Disease. 1st edition. Cambridge University Press. Gluckman, P. D., T. Buklijas, and M. A. Hanson (2016). “The Developmental Origins of Health and Disease (DOHaD) Concept”. In: The Epigenome and Developmental Origins of Health and Disease. Ed. by C. Rosenfeld (Ed.). Elsevier, pp. 1–15. Gluckman, P. D. and M. A. Hanson (2004). “Developmental Origins of Disease Paradigm: A Mechanistic and Evolutionary Perspective”. In: Pediatric Research 56.3 (3), pp. 311–317. Gluckman, P. D., M. A. Hanson, and A. S. Beedle (2007). “Early Life Events and their Consequences for Later Disease: A Life History and Evolutionary Perspective”. In: American Journal of Human Biology 19.1, pp. 1–19. Grossman, M. (1972). “On the Concept of Health Capital and the Demand for Health”. In: Journal of Political Economy 80.2, pp. 223–255. Grossman, M. (2000). “Chapter 7 – The Human Capital Model”. In: Handbook of Health Economics. Ed. by A. J. Culyer and J. P. Newhouse. Vol. 1, Part A. Elsevier, pp. 347–408. Halpern-Manners, A. et al. (2020). “The Effects of Education on Mortality: Evidence from Linked U.S. Census and Administrative Mortality Data”. In: Demography 57.4, pp. 1513–1541. Hoffmann, R., H. Kröger, and E. Pakpahan (2018). “Pathways between Socioeconomic Status and Health: Does Health Selection or Social Causation Dominate in Europe?” In: Advances in Life Course Research 36, pp. 23–36. Hoffmann R., H. Kröger, and S. Geyer (2019). “Social Causation versus Health Selection in the Life Course – Does their Relative Importance Differ by Dimension of SES?” In: Social Indicators Research 141(3), pp. 1341–1367. Hollingsworth, T. H. (1977). “Mortality in the British Peerage Families since 1600”. In: Population (French Edition) 32, pp. 323–352. Kuh, D. and Y. Ben-Shlomo, eds. (2004). A Life Course Approach to Chronic Diseases Epidemiology. 2nd edition. Oxford University Press. Kuzawa, C. W. and E. A. Quinn (2009). “Developmental Origins of Adult Function and Health: Evolutionary Hypotheses”. In: Annual Review of Anthropology 38.1, pp. 131–147. Link, B. G. and J. Phelan (1995). “Social Conditions as Fundamental Causes of Disease”. In: Journal of Health and Social Behavior, pp. 80–94. Mackenbach, J. P. (2006). Health Inequalities: Europe in Profile. Expert Report, European Commission. Mackenbach, J. P. (Aug. 2012). “The Persistence of Health Inequalities in Modern Welfare States: The Explanation of a Paradox”. In: Social Science and Medicine 75.4, pp. 761–769. Manor, O., S. Matthews, and C. Power (2003). “Health Selection: The Role of Inter- and Intra-generational Mobility on Social Inequalities in Health”. In: Social Science & Medicine 57.11, pp. 2217–2227. Mattsson, K. et al. (2013). “Maternal Smoking During Pregnancy and Daughters’ Risk of Gestational Diabetes and Obesity”. In: Diabetologia 56.8, pp. 1689–1695. Medawar, P. B. (2019). The Uniqueness of the Individual. Routledge. Mirowsky, J. and C. E. Ross (2005). “Education, Cumulative Advantage, and Health”. In: Ageing International 30.1, pp. 27–62. Niaura, R. et al. (2001). “Maternal Transmission of Nicotine Dependence: Psychiatric, Neurocognitive and Prenatal Factors”. In: The American Journal on Addictions 10.1, pp. 16– 29. Oken, E. et al. (2009). “Associations of Gestational Weight Gain with Short- and Longer-term Maternal and Child Health Outcomes”. In: American Journal of Epidemiology 170.2, pp. 173–180.

Early childhood origins of modern social class health disparities  251 Olshansky, S. J. and A. B. Ault (1986). “The Fourth Stage of the Epidemiologic Transition: The Age of Delayed Degenerative Diseases”. In: The Milbank Quarterly 64.3, pp. 355–391. Omran, A. R. (1971). “The Epidemiologic Transition: A Theory of the Epidemiology of Population Change”. In: The Milbank Memorial Fund Quarterly 49.4, pp. 509–538. Oreopoulos, P. and K. G. Salvanes (2011). “Priceless: The Nonpecuniary Benefits of Schooling”. In: Journal of Economic Perspectives 25.1, pp. 159–184. Palloni, A. (2006). “Reproducing Inequalities: Luck, Wallets, and the Enduring Effects of Childhood Health”. In: Demography 43.4, pp. 587–615. Palloni, A. et al. (May 2009). “Early childhood health, reproduction of economic inequalities and the persistence of health and mortality differentials”. In: Social Science & Medicine. Special Issue: Early life effects on socioeconomic performance and mortality in later life: A full life course approach using contemporary and historical sources 68.9, pp. 1574–1582. Patti, M.E (2013). “Intergenerational Programming of Metabolic Disease: Evidence from Human Populations and Experimental Animal Models”. In: Cellular and Molecular Life Sciences 70.9, pp. 1597–1608. Pembrey, M. E. (Dec. 2010). “Male-line Transgenerational Responses in Humans”. In: Human Fertility (Cambridge, England) 13.4, pp. 268–271. Pembrey, M. E. et al. (Feb. 2006). “Sex-specific, Male-line Transgenerational Responses in Humans”. In: European Journal of Human Genetics 14.2 (2), pp. 159–166. Perez-Arce, F. (2017). “The Effect of Education on Time Preferences”. In: Economics of Education Review 56, pp. 52–64. Power, C. and B. Jefferis (2002). “Fetal Environment and Subsequent Obesity: A Study of Maternal Smoking”. In: International Journal of Epidemiology 31.2, pp. 413–419. Simeone, R. M. et al. (2015). “Diabetes and Congenital Heart Defects: A Systematic Review, Meta-Analysis, and Modeling Project”. In: American Journal of Preventive Medicine 48.2, pp. 195–204. Simmons-Morton, B. et al. (2001). “Peer and Parent Influences on Smoking and Drinking among Early Adolescents”. In: Health Education and Behavior 28(1), pp. 95–107 Stern, J. (1983). “Social Mobility and the Interpretation of Social Class Mortality Differentials”. In: Journal of Social Policy 12.1, pp. 27–49. Thayer, Z. M. and C. W. Kuzawa (2015). “Ethnic Discrimination Predicts Poor Self-rated Health and Cortisol in Pregnancy: Insights from New Zealand”. In: Social Science & Medicine 128, pp. 36–42. West, P. (Jan. 1, 1991). “Rethinking the Health Selection Explanation for Health Inequalities”. In: Social Science & Medicine 32.4, pp. 373–384.

17. The long arm hypothesis: childhood poverty, epigenetic ageing, and late-life health in America, Britain, and Europe Gindo Tampubolon

INTRODUCTION Childhood conditions have been increasingly considered a risk factor for old age disability, dysfunction, disease and mortality. The term “long arm of childhood conditions” was coined by Hayward and Gorman (2004) who studied the influence of childhood poverty on mortality in American males in their 40s and 50s. Recently, nationally representative surveys or large cohort surveys have emerged to extend and provide empirical support to the long arm thesis (Berndt & Fors 2016; Lennartsson et al. 2018; Pakpahan et al. 2017; Tampubolon 2015; Vable et al. 2019). These authors unpacked the outcome of mortality into various types of ill health and dysfunction, while simultaneously extending the age range to the ninth decade. Studies from Britain and Sweden, as well as other European countries, examined whether childhood poverty relates to general health, physical health, mental health, and cognitive health of people in their 50s and older. Not only were the health outcomes diverse, the definition of childhood poverty also varied. It is therefore necessary to examine the thesis with a common definition of childhood condition and comparable health outcomes when broadening the scope of countries under study. Twenty-eight countries on both sides of the Atlantic have collected life histories using similar instruments and health outcomes, using clinically relevant measures such as probable sarcopenia; see European working group on sarcopenia (Cruz-Jentoft et al. 2019). With such comparative design emerging, findings are consolidated and knowledge is accumulated to advance efforts in delivering healthy ageing (UN 2021) to increasingly large numbers of people throughout the world. This chapter therefore aims to investigate the long arm thesis in many countries, which suggests that childhood conditions continue to mark old age health, and that even after youth or adult conditions are considered, the partial or direct association between childhood and old age remains. To achieve this aim, a family of longitudinal ageing studies from 28 countries is used in a cross-country fixed effect design adopted from Pakpahan et al. (2017: 5) which investigated similar questions in 13 of these countries. Then to explain the mechanism behind the continuing association, this chapter also aims to investigate an epigenetic ageing hypothesis using new nationwide representative data from America: childhood poverty is associated with changes in the epigenome of individuals, especially increasing rates of methylation which turns genes on and off. With changes in the epigenome or increases in the methylation rates which control normal organ functions, such individuals have older epigenetic age for the same calendar age. Childhood conditions associate with epigenetic age even after calendar age is considered.

252

The long arm hypothesis  253 A broad conceptual framework is used to bind recent results together and form an organising principle capable of generating policy input and scientific progress. One conceptual framework which puts important weight on the conditions in utero is the developmental origin of health and disease (Barker et al. 2005; Gillman 2005). Another framework is the social determinant of health, which is defined as “conditions in the environments in which people are born, live, learn, work, play, worship, and age that affect a wide range of health, functioning, and quality-of-life outcomes” (my emphasis, Office of Disease Prevention and Health Promotion 2016, quoted in Birn et al. 2017: 280). To these activities I insert “are born and raised”, giving emphasis to childhood conditions recorded in the study samples below. Although both can illuminate how childhood conditions can have long-run consequences, including ill health in older ages, I have chosen the social determinants framework because of its explicit recognition of ageing and the lack of information on birth conditions of the study samples. Put differently, the “social determinants of health” framework, while efficacious and influential, has yet to recognise the long arm of childhood conditions reaching into old age. The emerging empirical research strongly suggests that childhood lasts a lifetime. Emerging seams of research reinforce this reach. The English Longitudinal Study of Ageing (ELSA) furnished data which showed that the childhood poor have slower walking speed, lower cognitive function and worse depression in old age (Tampubolon 2015). Across the English Channel, the corresponding study in Europe, the Survey of Health, Ageing and Retirement (SHARE), was used to study childhood socioeconomic position and old age self-rated health. Though initial findings were similar, the association between childhood and old age disappeared when controlling for midlife factors (Pakpahan et al. 2017). Across the Atlantic, another longitudinal ageing study, the US Health and Retirement Study (HRS), was used to derive trajectories of socioeconomic condition from childhood (Vable et al. 2019). Some echoes were found: those tracing the poor trajectory have slower walking speed, weaker grip strength and worse lung function. In sum, how one was raised materially can have lasting effects in countries of America, Britain and Europe. Why the Need for So Many Countries? Using ELSA, Tampubolon (2015) was the first to study the long arm of childhood reaching into midlife and old age (≥ 50 years) on a nationwide scope affecting wide health outcomes (physical and mental health as well as cognition). There have been cohort studies, including one of 60–64-year-olds or the 1946 cohort (Stafford et al. (2015) on mental well-being), but such cohort studies are by construction limited in many ways, especially in the childhood exposure and timing of outcome. For instance no nationwide cohort studies were initiated before World War II (the exposure) to study the experience of the oldest old (≥ 85 years), an increasingly large section of the older age group (Suzman et al. (1996) coined the term, while Aronson coined a new term “elderhood” to rhyme with childhood and adulthood (Jain 2021)). Of course the study which posited and tested the thesis used a nationwide male sample of Americans aged ≤ 59 years, which again was silent on the experience of large sections of much older people (Hayward & Gorman 2004). Therefore there were limits to the information on childhood exposure in the earlier cohort surveys, unlike the national sample ELSA. As to the place-based cohort surveys, these are equally limited in similar ways. By design the childhood exposure was limited to one place, making tenuous any attempts to stretch

254  Handbook of health inequalities across the life course the findings nationwide. Lastly, rarely were wide outcomes of ageing examined, unlike the national sample ELSA which enabled studies of a wide range of disability, dysfunction and disease. Empirical investigations of the long arm thesis have been fruitful in uncovering new determinants of health inequalities both in levels of health and in its rates of change. Exposure to World War II matters. In a series of cross-country studies we showed that such an exposure shapes old age health trajectories in America and Britain (Tampubolon & Fajarini 2018; Tampubolon & Maharani 2017a, 2017b). We found that cohort makes a difference to the levels and trajectories of depression and allostatic load. Recently, we studied the long arm of childhood condition and how it relates to trajectories of grip strength using the European survey SHARE (Tampubolon & Fajarini 2018). We found that childhood conditions match the levels of muscle function in old age but not its rate of decline. The childhood poor have weaker grip strength in old age though they do not experience more rapid worsening of muscle function. Encouraged by the results of examining complex patterns of health inequalities (cohort effect, childhood conditions, levels versus rates), we expand on the cross-country study by bringing America, Britain and Europe into one study and focusing on the difference in the levels of health as shaped by childhood conditions. Cross-country studies constitute a specific test of the long arm thesis, helping us to fathom its geographic reach. By putting these countries together we shall also appreciate its historical import. In our studies on allostatic load and depression trajectories on both sides of the Atlantic, we revealed that World War II made a difference. The War cohort born in the American Midwest experience ageing differently from the War cohort in the British Midlands, because the theatre of War was on this side of the Atlantic. Growing up in the theatre of War matters for health in old age. By bringing more countries together we shall learn more about whether the thesis has a broad reach that crosses historical experience. Mainland America was never under attack, Britain was under attack, much of Europe was occupied to different degrees during the War. The long arm thesis can only be strengthened by subjecting it to such varying historical experience. In between the original submission of this work and its final form, Russia invaded Ukraine, and by early April 2022 the war had displaced 4 million Ukrainians to various countries included in this work, mostly mothers and children. This work speaks to the possible consequences for the refugee children which may reach far into their old age. Framework, Life Course Model and Theory When designing the tests it is important to recall that there are different concepts of childhood conditions and for each concept different indicators were used in the empirical literature. Childhood poverty, childhood socioeconomic position and adverse childhood experience were some of the concepts used. These were captured as a scale or latent constructs with various indicators including parental social class and education and employment, family financial hardship (bankruptcy and self-report of financial hardship, for example), being raised in a foster family, number of rooms in childhood home, number of people in childhood home, and number of books in childhood home, and its facilities. The last in turn includes running hot and cold water, fixed bath, central heating and indoor lavatory – all material facilities.

The long arm hypothesis  255

Figure 17.1

Social determinants framework for testing the long arm of childhood conditions (growing up) shaping old age health

This study does not use the adverse childhood experience scale for two reasons. First, the scale is complementary to the other concepts. Second, the scale has more psychological or subjective elements, which suggests more susceptibility to bias. I am not aware of any test of the magnitude of the bias, for instance by comparing an independent report during childhood and during old age, unlike childhood conditions which have undergone such a test; see Tampubolon (2010, 2015). In short there are variables in the concepts and indicators which require a clear framework and latent constructs to organise them. Following Tampubolon (2015), a conceptual framework and equations capable of being tested in many countries are applied here (see Figure 17.1). Equations to estimate the associations between childhood poverty and old age health when childhood conditions are recalled retrospectively follow Tampubolon (2015) equations 3 and 4. For simplicity the framework does not decompose the childhood poverty–old age health association into direct and youth-mediated associations, or show any additional path from childhood poverty to youth or adult health. Nevertheless the framework is capable of giving an indication of whether the mediation path absorbs all the direct association, by including youth health in the complete model and examining the net association of childhood poverty. This framework follows the previous designs of Case, Fertig and Paxson (2005), who studied Britons aged 42, and Tampubolon (2015), who studied Britons aged 50–90. This framework is an instance of life course models as opposed to pathway models (Case et al. 2005; Tampubolon 2010, 2015). Throughout, a set of confounders or social determinants is included as standard in the literature on life course ageing (dashed box). The standard practice of controlling for education and class in this way, that is simultaneously instead of using education as a mediator before class, is similar to this framework. For an example of how to unpack this box see Nazroo (2017). Theory: Epigenetics Under the Skin The long arm thesis and the wide outcomes together suggest that child poverty gets under the skin. Epigenetics has been proposed as the probe to uncover evidence of the operating mechanism underneath. Two children with identical genes can appear different to one another because chemical or molecular changes acted upon their genetics – the Greek prefix “epi” can be translated as “upon”. The changes acting upon genetics or epigenetics turn the same genes

256  Handbook of health inequalities across the life course on and off differently depending on chemical additions or deletions to specific sites in the vast DNA. All along, the genes stay the same, but epigenetic changes turn certain genes on allowing them to be expressed as organ function while other genes are turned off with consequent changes to their expression. Two kinds of molecular changes are thought largely responsible, namely methylation and histone modification (Carey 2011); the first is increasingly available in epidemiologic studies thanks to reduction in the cost of chips for epigenomic sequencing (Crimmins et al. 2021). The processes of transcription and translation from DNA to RNA then to protein are susceptible to environment and experience influence. Szyf (2013) found in macaques (primates like humans) that early experience of being groomed (“licking by older macaques”) modified phenotypic expressions. Two offspring of the same parent were separated early, and one of them was put to a family which groomed young macaques differently (“less licking”). The two grew to express a key gene for regulating the hypothalamic-pituitary-adrenal axis, namely the glucocorticoid receptor NR3C1. This receptor is differentially expressed according to the experience of social deprivation, by epigenetic modification, resulting in heightened stress response and demodulated immune response (Szyf 2013; Tampubolon & Fajarini 2018). Mechanisms that turn off DNA transcription into phenotypic expression are various, including histone modification, DNA and RNA methylation. In empirical research, methylation is mostly tissue-specific, involving specific proteins, for example along the hypothalamic– pituitary–adrenal axis, but recent works suggest that blood-based (general) methylation shows useful variations of modification in the transcription, thereby expressing how much experience and environment gets under the skin (Crimmins et al. 2021). Further opportunities for experience-dependent pervasive methylation which modifies messenger RNA have been revealed within the last decade, using a technique of methylated RNA immunoprecipitation sequencing (see Figure 17.2). A recent review of subsequent experiments (Widagdo & Anggono 2018) proposes an epigenetic model of specific RNA methylation responsible for psychiatric disorder, in which experience and environment (top left) interfere in the neuron that eventually modifies the expression of brain function of mice (top right). The focus is on N6 methyladenosine (m6A) methylation. The review suggests that m6A is capable of behavioural and sensory regulation in different brain regions, including the hippocampus, prefrontal cortex, and amygdala. Subsequently, molecular manipulation that changes the levels of m6A results in behavioural changes such as addiction and reward learning. This makes m6A a relevant methylation for physiological control of cognition, including memory. It is a long and winding line to draw from mouse neurons to episodic memory of older Americans, Britons, and Europeans. Moreover, there is no population sample of m6A methylation in these populations. It would be unwise, however, to dismiss entirely the possibility, as demonstrated in the review of experiments, that childhood experience influences the amount of RNA methylation under the skin, which then affects various tissues and organs and their function, including the brain. Two Hypotheses: The Long Arm of Childhood Conditions and Epigenetic Ageing I tie this discussion into two hypotheses. First, while youth health relates to old age health, childhood poverty retains a direct relation to old age health in older people: the long arm of childhood conditions hypothesis. If the hypothesis receives support in nearly 30 countries

The long arm hypothesis  257

Source: From Widagdo & Anggono (2018).

Figure 17.2

The model of N6 methyladenosine (m6A) methylation modifying neuron functions in the brain

with an ageing population, the social determinants of healthy ageing around the world can no longer ignore childhood poverty, especially in delivering the UN Decade of Healthy Ageing. To explain the mechanism behind this long arm hypothesis, I propose an epigenetic ageing hypothesis: childhood poverty is directly associated with epigenetic age, where the childhood poor show higher rates of methylation as summarised in their epigenetic age (Crimmins et al. 2020, 2021).

METHODS AND MATERIALS Latent Construct to Recover Child Poverty Status Obtained in Hindsight In this study measures of childhood conditions are based on retrospective reports of adults aged 50 and older, raising concerns about recall error. While these concerns are often acknowledged, little has been done to deal with them until recently. I believe older Britons are typical in the fallibility of their memory. When asked the number of rooms and people in their child-

258  Handbook of health inequalities across the life course hood home, only one in three Britons aged 50 got both right. Their mothers had been visited 40 years earlier so records were available to check. Retrospective information about childhood obtained in midlife is error-laced. Believing otherwise yields unsafe inference (Tampubolon 2010, 2015). The forms of unsafe inference include both measurement error and measurement bias, and ways to deal with these using latent constructs were proposed and applied fruitfully in Tampubolon (2010) using the European Survey of Income and Living Conditions, and in Tampubolon (2015) using the English Longitudinal Study of Ageing (ELSA), as well as in Tampubolon and Fajarini (2018) using the European Survey of Health, Ageing and Retirement (SHARE). Recent works on childhood poverty assume recalled childhood information to be error-laced, thus begin by deriving latent constructs of child poverty, either latent class or latent factor (Berndt & Fors 2016; Lennartsson et al. 2018; Pakpahan et al. 2017; Vable et al. 2019). As demonstrated in these emerging studies, these latent constructs afford flexibility, adapting to available indicators to capture childhood poverty status. The methodological literature has also settled on retaining, rather than excluding, any cases with incomplete indicators, by applying an expectation-maximisation algorithm, thus avoiding a highly selective sample (Pakpahan et al. 2017; Tampubolon 2015; Vable et al. 2019). This approach is followed here, using latent class of poverty in childhood as the key risk factor in old age health. Doing Cross-Country Comparison When Indicators are Not Exactly the Same The empirical literature on cross-country comparison offers another building block in enabling comparison despite less than exact sameness (Courtin et al. (2015) on depression in America and Europe; Tampubolon and Maharani (2018) on allostatic load in America and Britain). Courtin and colleagues, for instance, compare two measures of depression: CESD and EuroD. The authors found that although these measures differ in their tendency to assign caseness (CESD assigned more cases of depression), they are nevertheless comparable in risk factor associations. The focus can therefore be placed on the associations rather than abandoning comparisons altogether because of any difference in tendency. If the aim is not to estimate prevalence but to test hypothesised associations, then this sharpened focus should enable comparison to proceed despite different available indicators. Similarly, Tampubolon and Maharani (2018) also used different sets of indicators for allostatic load on both sides of the Atlantic, motivated by the same principle of focusing not on the indicators but on the risk factor associations. In fact, as argued in presenting the methods above (Tampubolon 2015), a latent construct to assess childhood conditions is precisely at home with different contexts on different sets of indicators. I continue to apply this principle and am not unduly constrained by having to have the same set of indicators of childhood conditions. Lastly, following Pakpahan et al. (2017: 5) which reported the first cross-country study of the long arm thesis in 13 European countries, all analyses control for differences between countries by including country fixed effects. Material – Childhood Information: ELSA, SHARE and HRS The analytic sample was constructed from a family of ageing studies: the US Health and Retirement Study (HRS), the English Longitudinal Ageing Study (ELSA), and the Survey of Health, Ageing and Retirement in Europe (SHARE www​.share​-project​.org). ELSA collected

The long arm hypothesis  259 Table 17.1

Retrospective childhood information in the English Longitudinal Study of Ageing (ELSA), the Survey of Health, Ageing and Retirement in Europe (SHARE) and the Health and Retirement Study (HRS)

Country

Britain

Europe

Variable

Number of rooms

Number of rooms

America Number of rooms

Number of people

Number of people

Number of people

Number of books

Number of books

Number of books

Physical facilities:

Physical facilities:

Ever experienced:

indoor toilet, hot and cold

indoor toilet, hot and cold running

being financially poor had to

running water,

water,

move or be helped because of

central heating, fixed bath

central heating, fixed bath

financial difficulty, had to live with

Tampubolon 2015

Tampubolon & Fajarini 2018

grandparents, father unemployed Source

HRS Life History Mail Survey

life history in wave 3, while SHARE collected childhood information in its life history waves in wave 3 and 7, for those who did not provide it the first time. The ELSA and SHARE indicators have been described elsewhere (Tampubolon 2015; Tampubolon & Fajarini 2018). HRS collected some childhood information in the core interview at each wave but, crucially, this is not closely comparable with indicators from the other side of the Atlantic, and instead opted to elaborate on financial situation. This information is augmented with the ad hoc Life History Mail Survey 2017. Although ageing studies in America, Britain and Europe have close family resemblances, there are sufficient differences worth listing. Table 17.1 lists variables collected in each study, noting the emphasis in HRS on financial information, used to derive latent class of childhood poverty. To test the long arm thesis, the latent class of childhood poverty was then matched with the latest health outcomes (2020/2021), that is depression (following Courtin et al. 2015), episodic memory (following Tampubolon 2015) and probable sarcopenia (following the European Working Group on Sarcopenia, Cruz-Jentoft et al. 2019). To test the epigenetic ageing hypothesis in HRS, the childhood poverty status was matched with the epigenetic ages (methylation rates) sequenced from the epigenomes collected in the HRS Venous Blood Sample (Crimmins et al. 2020, 2021). In the Venous Blood Sample Study 4021 HRS respondents consented to have their blood taken. Sequencing of the blood sample gives the rates of methylation across various sites in the human epigenome, giving 12 different variables of epigenetic ages. The higher the rates, the older the epigenetic ages, expressing more potential deleterious methylation to the epigenome. Following the nomenclature established by the authors, there are 12 DNA methylation rates or epigenetic ages used here: Bocklandt, Dunedin, Garagnani, Hannum, Horvath, Levine, Lin, Lu, VidalBralo, Weidner, Yang and Zhang (Crimmins et al. 2020, 2021). These numerous variations arise from two sources – the vast epigenome commensurate with the similarly vast genome it controls, and how information is read from the epigenome by calibrating or training it against outcomes. Early constructs of epigenetic ages were calibrated or trained against age and mortality; thus the units are typically age in year (horizontal axes in Figure 17.6) while new constructs were trained on health outcomes for example diseases and on pace of ageing; thus the units are typically dimensionless (absolute value around one reflecting one epigenetic age per one calendar age; horizontal axes in Figure 17.6). As Crimmins et al. (2021: 1118) state, epigenetic ages do not “relate to outcomes in the same

260  Handbook of health inequalities across the life course way … nor to independent variables … Understanding which risk factors relate to which [constructs] may help researchers synthesize existing research findings.” The study here is another exploration of available constructs to enhance this understanding. So when validating the epigenetic ages, the authors found some surprising and null associations. These were explained as arising from the novelty of the ideas and the experimental character of the epigenetic age measures. For instance, common associations such as with age were not statistically significant. Nonetheless the measures and report are seminal. More details are given in the original reports (Crimmins et al. 2020, 2021) though their smoothed distributions are shown here for the first time. To test the epigenetic ageing hypothesis, I fit a linear model explaining methylation rates in terms of the key exposure to childhood poverty. Because the Venous Blood Sample is experimental, little overlap is observed with the main analytic sample and its covariates are also limited (see below). All estimation applied corrections to the standard errors because the key exposure of childhood poverty is not an observed variable but constructed in an earlier stage of latent class analysis (Vermunt 2010). Moreover, Bonferroni correction was further applied in the testing of the epigenetic ageing hypothesis.

RESULTS Table 17.2 collects features of the analytic sample which has 57% female, one in five was childhood poor and mean age 66.3 years. The bottom part shows the numbers contributed to the analytic sample by each country. Table 17.3 shows the epigenetic sample’s features including 16% of the childhood poor (a higher percentage in females compared to males). Compared to the cross-country sample (Table 17.2 – 20% were childhood poor), this sample (16% were childhood poor) comes from more advantaged members, as history would also attest. In the 20th century America has been in the ascendancy and she has fewer poor children (16% vs 20%). Compared to the main HRS sample, the epigenetic ageing sample is more advantaged (Crimmins et al. 2020). Clearly this difference accords with the exploratory nature of the test of the epigenetic ageing hypothesis, namely more information comes from the more advantaged section of the population. The other limitation of this sample relative to the main one is the limited number of variables with non-missing responses. I would have preferred the main sample to be sequenced for their epigenomes but this epigenetic sample already holds very promising first results, as we shall see shortly. The coefficients explaining health outcomes of depression caseness using probit model, episodic memory (Poisson model) and probable sarcopenia (probit model) are collected in Table 17.4, with one column block for each health outcome. All the coefficients are partial or net associations after controlling for all covariates and country fixed effects. Because the key exposure of childhood poverty is a latent class (childhood poor and non-poor) which was constructed instead of observed decades before, adjustment to standard error is made accordingly (Vermunt 2010). Failure to recognise the constructed nature of the key exposure to childhood poverty renders inference unsafe. In all models country fixed effects were included to capture the uniqueness of each country such as its health system or history, following Pakpahan et al. (2017). The results show that the childhood poor are significantly more adversely affected in

The long arm hypothesis  261 Table 17.2

Summary features of main analytic sample with categorical variables in numbers and percentages and continuous variables in means and standard deviations Sex

All

Male

Female

N = 33275

N = 44814

Non-poor

26875 (80.8%)

35829 (80.0%)

62704 (80.3%)

Poor

6400 (19.2%)

8985 (20.0%)

15385 (19.7%)

Age

66.4 (9.7)

66.2 (10.4)

66.3 (10.1)

Times ill since 16

0.2 (0.6)

0.2 (0.7)

0.2 (0.6)

Lower degree

26487 (79.6%)

36996 (82.6%)

63483 (81.3%)

College or higher

6788 (20.4%)

7818 (17.4%)

14606 (18.7%)

N = 78089

Childhood poverty

College

Marital status Single Married/Union Separated/Widowed

2072 (6.2%)

2411 (5.4%)

4483 (5.7%)

20057 (60.3%)

21513 (48.0%)

41570 (53.2%)

11146 (33.5%)

20890 (46.6%)

32036 (41.0%)

Father’s job: elementary occupation No

26911 (80.9%)

36179 (80.7%)

63090 (80.8%)

Yes

6364 (19.1%)

8635 (19.3%)

14999 (19.2%)

27961 (84.0%)

35038 (78.2%)

62999 (80.7%)

5314 (16.0%)

9776 (21.8%)

15090 (19.3%)

Bottom wealth tertile No Yes Race Other White Caucasian

2874 (8.6%)

3976 (8.9%)

6850 (8.8%)

30401 (91.4%)

40838 (91.1%)

71239 (91.2%)

Country Britain

2789 (8.4%)

3558 (7.9%)

6347 (8.1%)

Austria

778 (2.3%)

1181 (2.6%)

1959 (2.5%)

Germany

1700 (5.1%)

1936 (4.3%)

3636 (4.7%)

Sweden

1291 (3.9%)

1513 (3.4%)

2804 (3.6%)

Netherlands

749 (2.3%)

977 (2.2%)

1726 (2.2%)

Switzerland

1046 (3.1%)

1296 (2.9%)

2342 (3.0%)

France

1420 (4.3%)

1991 (4.4%)

3411 (4.4%)

Denmark

1322 (4.0%)

1574 (3.5%)

2896 (3.7%)

Luxembourg

377 (1.1%)

467 (1.0%)

844 (1.1%)

Finland

519 (1.6%)

605 (1.4%)

1124 (1.4%)

Italy

1439 (4.3%)

1817 (4.1%)

3256 (4.2%)

Spain

1330 (4.0%)

1661 (3.7%)

2991 (3.8%) 4259 (5.5%)

Greece

1878 (5.6%)

2381 (5.3%)

Malta

330 (1.0%)

424 (0.9%)

754 (1.0%)

Cyprus

180 (0.5%)

309 (0.7%)

489 (0.6%)

Latvia

264 (0.8%)

474 (1.1%)

738 (0.9%)

Slovenia

935 (2.8%)

1362 (3.0%)

2297 (2.9%)

Poland

1307 (3.9%)

1663 (3.7%)

2970 (3.8%)

Czechia

1238 (3.7%)

1928 (4.3%)

3166 (4.1%)

Estonia

1034 (3.1%)

1840 (4.1%)

2874 (3.7%)

Croatia

492 (1.5%)

643 (1.4%)

1135 (1.5%)

262  Handbook of health inequalities across the life course Sex Male Lithuania

All Female

N = 33275

N = 44814

N = 78089

475 (1.4%)

872 (1.9%)

1347 (1.7%)

Hungary

262 (0.8%)

416 (0.9%)

678 (0.9%)

Israel

331 (1.0%)

485 (1.1%)

816 (1.0%) 1220 (1.6%)

Romania

509 (1.5%)

711 (1.6%)

Bulgaria

346 (1.0%)

535 (1.2%)

881 (1.1%)

Slovakia

430 (1.3%)

539 (1.2%)

969 (1.2%)

America

8504 (25.6%)

11656 (26.0%)

20160 (25.8%)

Sources: HRS, ELSA and SHARE.

Table 17.3

Summary features of epigenetic ageing sample with categorical variables in numbers and percentages and continuous variables in means and standard deviations Sex

All

Male

Female

N = 1672

N = 2349

N = 4021

Non-poor

1419 (84.9%)

1958 (83.4%)

3377 (84.0%)

Poor

253 (15.1%)

391 (16.6%)

644 (16.0%)

Variable Childhood poverty

Race Other

370 (22.1%)

635 (27.0%)

1005 (25.0%)

Non-Hispanic white

1302 (77.9%)

1714 (73.0%)

3016 (75.0%)

13.5

13.5

13.5

(0.1)

(0.1)

(0.1)

Education, years Household size

0.3

0.3

0.3

(0.7)

(0.6)

(0.7)

Source: HRS Venous Blood Sample (HRS 2020).

old age, reporting higher probabilities of depression, worse episodic memory and more probable sarcopenia. Beyond Britain, growing up in poverty goes with growing old in infirmity. Briefly, the other covariates suggest that females report more depression but better memory; however there is no difference in reporting probable sarcopenia. Education is associated with better health all round, and so is more wealth. In particular, those at the bottom wealth distribution report more depression, have lower episodic memory and higher probable sarcopenia. Importantly, to reflect the social determinant framework above, youth/adult illness is significant and in expected directions. This enables childhood poverty coefficients to be interpreted as direct or partial associations after considering all the other covariates. To bring this forward, for each of the 28 countries the predicted values of health outcomes are plotted as distinguished by childhood poverty over age 70 to 90, beginning with depression, then episodic memory and finally probable sarcopenia. This shows at a glance how childhood poverty shapes old age infirmity. This set of plots is evidence that childhood poverty not only shapes health long into old age, it does so across a wide range of health outcomes and, critically, across various health systems and history in rich countries.

The long arm hypothesis  263 Table 17.4

Coefficients explaining depression (probit model), episodic memory (Poisson model) and probable sarcopenia (probit model) in America, Britain and Europe with Vermunt correction for latent class covariate; body mass index was included in explaining probable sarcopenia, and depression (mood) was included in explaining episodic memory

Childhood poor Sex, Female Age College Marital status Single Married/union Father’s occupation Adult illness Bottom wealth White Caucasian Constant Country effects

Depression

Episodic memory

0.185

–0.079

Probable sarcopenia 0.090

(0.00)

(0.00)

(0.00)

0.349

0.090

0.001

(0.00)

(0.00)

(0.94)

0.008

–0.014

0.054

(0.00)

(0.00)

(0.00)

–0.148

0.122

–0.166

(0.00)

(0.00)

(0.00)

Ref: widow 0.084

–0.026

0.140

(0.00)

(0.00)

(0.00)

–0.050

0.005

–0.035

(0.00)

(0.10)

(0.10)

0.090

–0.0548

0.019

(0.00)

(0.00)

(0.40)

0.208

–0.018

0.131

(0.00)

(0.00)

(0.00)

0.300

–0.110

0.203

(0.00)

(0.00)

(0.00)

–0.206

0.145

–0.130

(0.00)

(0.00)

(0.01)

–1.591

3.135

–4.288

(0.00)

(0.00)

(0.00)

Included

Included

Included

Sources: HRS, ELSA and SHARE.

The depression trellis plot (Figure 17.3) reveals several key patterns. First, age is an important risk factor. Second, childhood poverty again puts people at a disadvantage consistent with the other two trellis plots. However, in this health outcome, childhood poverty exerts the strongest association: for any given age in any given country the distance between the probabilities of depression between the childhood poor and non-poor is greater than the difference in the probabilities of sarcopenia (Figure 17.5). Third, unlike in sarcopenia, here the regional patterns of Europe are not as pronounced (Northern, Southern and Eastern Europe). In each region variation is marked (France and Denmark in northern Europe, Spain and Cyprus in southern Europe and Lithuania and Slovenia in eastern Europe). These variations are comparable across regions, preventing one single summary pattern from representing all high-income countries. I have previously noted that depression in older ages in these countries is associated with childhood conditions, especially through the experience of war (Tampubolon & Maharani 2017a), and this may be consistent with what is apparent here, especially in northern European countries.

264  Handbook of health inequalities across the life course

Source: Analysis of HRS, ELSA and SHARE.

Figure 17.3

Probabilities of depression among the childhood poor (dash) and non-poor (solid) in older people aged 70 to 90 years in Britain, Europe and America based on models in Table 17.4 where all covariates are set at the sample averages

The episodic memory trellis plot shows several important patterns (Figure 17.4). First, age is the major factor, invariably showing lower episodic memory as people age. Second, even at such an advanced age, childhood poverty continues to make a difference with the childhood poor at a disadvantage. The regional pattern is not as pronounced. Even so, variations across countries are wide enough to prevent one pattern being sufficient to represent high-income countries. For example, comparing Britain and Spain (1st and 2nd rows) is instructive in this regard. Figure 17.5 shows a trellis plot of probable sarcopenia in these countries. First, age increases the risk of muscle dysfunction or probable sarcopenia, with a hint of acceleration in the oldest old. Second, childhood poverty additionally increases the probability of sarcopenia throughout old age: the dashed lines trace higher points throughout. Third, there are striking regional patterns, with northern Europeans showing lowest levels of probable sarcopenia throughout old age. Southern Europeans (Italy to Cyprus, except Greece) tend to have higher probabilities of sarcopenia. Eastern Europeans again tend to have higher probabilities than northern Europeans. The cases of Finland, Spain and Slovakia are instructive. Americans tend to be higher than northern Europeans and Britons higher still. Last, there are enough variations

The long arm hypothesis  265

Source: Analysis of HRS, ELSA and SHARE.

Figure 17.4

Episodic memory (number of words recalled) among the childhood poor (dash) and non-poor (solid) in older people aged 70 to 90 years in Britain, Europe and America based on models in Table 17.4 where all covariates are set at the sample averages

across these countries, even when ages are held constant from 70 to 90, to prevent generalisations across all high-income countries. For the same ages, probabilities of sarcopenia in Slovakia lie entirely above those of Germany (both in column 3). This is a novel finding. There are four summary points I want to raise with these three sets of plots. Across all health outcomes in old age, childhood poverty goes with worse health irrespective of health systems and history. Secondly, childhood poverty shows more effect on mental health relative to probable sarcopenia. Thirdly, there is no obvious correlation between these profiles with what is known about the wealth of these countries. Take depression in this case, and compare Britain and America (first and last plots in Figure 17.3). Though America spends more on health, and its history of childhood was not marked by the War to the same extent as the Europeans, depression probability in America is higher. Lastly, although there is uniform association with childhood poverty (first point), there is no uniformity across countries. Phrased differently, a country that is ahead in one health outcome may not hold that advantage in another. Take probable sarcopenia and, again, Britain and America. This time the advantage is reversed. A similar exercise comparing Britain and Austria will illustrate these points. I have argued elsewhere that part of the explanation can be found in the cohort compositions of older people

266  Handbook of health inequalities across the life course

Source: Analysis of HRS, ELSA and SHARE.

Figure 17.5

Probabilities of sarcopenia among the childhood poor (dash) and non-poor (solid) in older people aged 70 to 90 years in Britain, Europe and America based on models in Table 17.4 where all covariates are set at the sample averages

in these countries (Tampubolon & Fajarini 2018; Tampubolon & Maharani 2017a, 2017b). For now, given the complex pictures laid out here we need to go deeper. Epigenetic Ageing Hypothesis Result The analysis went into the blood stream: the epigenome from the Venous Blood Sample was sequenced to derive methylation rates or epigenetic ages and their smoothed distributions are plotted in Figure 17.6. Some variations in the features of the distribution are apparent, and all point to the need for further exploration instead of picking a select few; the range, skew and kurtosis vary notably. The Yang index, for instance, is right skewed leptokurtic, while Dunedin is almost symmetric and broader. These variations are evidence of the different calibration noted above, some against mortality, others against diseases while yet others against pace of ageing (Crimmins et al. 2021). Critical to note is that all measures of epigenetic age are unimodal, giving some support to using a linear model to explain epigenetic age or methylation rates in terms of childhood poverty and other covariates.

The long arm hypothesis  267

Figure 17.6

Smoothed distributions of epigenetic age in HRS Venous Blood Sample (HRS 2020)

Table 17.5

Linear model of epigenetic age (Yang) with Bonferroni correction for multiple testing and Vermunt correction for latent class covariate Coeff (p value)

Childhood poor

0.004 (4.2 E-9)

Sex, Female

0.0005 (0.39)

Age

0.0005 (1.0 E-58)

Education, year

0.0004

Household size

0.0007

Non-Hispanic

–0.0082

white vs other

(1.8 E-31)

(0.92) (0.11)

Constant

0.0346 (0.48)

N

Source: HRS Venous Blood Sample (Crimmins et al. 2020).

4021

268  Handbook of health inequalities across the life course In testing the epigenetic ageing hypothesis I found only one epigenetic age that is highly significant in association with childhood poverty, that is Yang, shown in Table 17.5. It says the childhood poor aged epigenetically faster. To ease interpretation of the association between childhood poverty and epigenetic age over a range of chronological age, I plot predicted epigenetic age (Yang) on chronological age while distinguishing those whose childhoods were poor from the non-poor (Figure 17.7). The plot shows that the childhood poor undergo a faster ageing from midlife onward. Using the coefficients and significance from Table 17.5 and marginal effects from Figure 17.6, the childhood poor experienced epigenetic ageing that is 3.7 times the calendar age for those aged 50 to 90. This is evidence of the childhood conditions in the epigenome.

Source: HRS Venous Blood Sample (HRS 2020).

Figure 17.7

Predicted epigenetic age/methylation (Yang) by age of older Americans aged 50–90 among the childhood poor (dash) and non-poor (solid)

DISCUSSION AND CONCLUSION Evidence is accumulating and compelling that the long arm of childhood conditions reaches into old age. Childhood conditions work under the skin shaping the health of older people, including the oldest old across wide outcomes of ageing around the world. Epigenetic changes objectively measured in DNA methylation rates across the vast epigenome are higher among the childhood poor. This is the first evidence in a nationwide representative population of how childhood poverty plays a role in epigenetic changes observed in the oldest old.

The long arm hypothesis  269 The result suggests two key claims. First, childhood continues to directly affect healthy ageing for wide outcomes of ageing, even after some control of youth health and other social determinants. Second, the mechanism underlying this direct link works through epigenetic changes or methylation of the epigenome. Material lack or family financial hardship when raising a child has long-term consequences for the child. Childhood conditions get into the blood stream to change the epigenome with consequences up to the ninth decade of life. This operates in diverse health systems: in a national health system (UK), a private health system (US) and everything in between (27 European countries). The UK NHS is more than 70 years old and has accompanied Britons in the study cohorts throughout their life courses. Using latent class analysis to recover child poverty status from different sets of error-laced indicators, the result shows important associations between childhood poverty and health outcomes that are consistently observed across 28 countries extending the long arm thesis in many directions. What Is the Mechanism? I have tried to offer an epigenetic explanation for the role of childhood conditions as a risk factor. Poverty in childhood is posited to induce epigenetic change through increasing methylation rates (Tampubolon & Fajarini 2018) which leaves a mark that lasts until old age. Children with similar genetic make-up may nevertheless differ in their phenotypic expression, hence organ function, because their childhood conditions differ in ways that change methylation rate in one child more than in another. The analysis uncovered the first evidence in support of the epigenetic theory of the long arm of childhood conditions in a nationally representative American sample. Out of 12 measures of DNA methylation rates I found one strongly significant association, showing more childhood poverty goes with more methylation. The childhood poor aged epigenetically faster. This new hypothesis, tested here on a nationally representative sample for the first time, adds to the importance of tackling childhood poverty. That childhood lasts a lifetime, we knew. Now we know it is partly through epigenetic changes. The methylation rates have been found not to associate with common risk factors in expected ways (Crimmins et al. 2021). I also found the other methylation measures to be neither significant nor inversely related to childhood poverty. Part of the missing associations may hide in the epigenome. The transcription from DNA to protein is fraught with methylation. Widagdo and Anggono (2018) showed that in addition to DNA methylation, RNA methylation was also shaped by experience and environment to modify the expression of neuron functioning. This first analysis of population-wide DNA methylation rates has captured some epigenetic changes but may have missed RNA methylation. Clearly, this is an obvious path to explore in further longitudinal ageing studies with venous blood sample, especially studies comparable to the HRS. Limitations and Strengths There are limits to what can be learned from this work. First, its simple framework is not a full structural framework such as that of Pakpahan and co-authors (2017), although this has precedents (Case et al. 2005; Tampubolon 2015). The structural framework alternative has encouraged the authors to suggest that were more covariates considered, the long arm of childhood would be entirely weakened. This suggestion is of course an eminently empirical

270  Handbook of health inequalities across the life course question. The imperative remains that a fuller framework should be considered when comparable information on youth and adult conditions in all these countries is available. Second, this work is also limited in its focus on material lack in childhood, lacking psychological variables such as parent-child bonding or nurturing relationships. Cultivating psychological development is of course core to the stage of raising children, a distinct stage in the social determinants framework above. Third, by including fixed effects for countries this work controls for unobserved differences between countries such as their histories and health systems. Despite a precedent (Pakpahan et al. 2017) in using country fixed effects, this may be considered insufficient and thus constitutes another limitation. Country by country analyses for these 28 countries are deferred to future monographs. Last, a key limitation arises from the design being observational rather than a randomised study (to childhood material intervention and to control), preventing causal interpretation of these results. The childhood poor may differ from the non-poor in ways not observed. Moreover, some of the childhood poor may not have survived to take part in these longitudinal ageing studies. Although this survivor effect may suggest a direction of bias, its magnitude is unknown. Also, a moment’s thought would suggest that randomisation may make it impossible to draw causal inference for the thesis of long arm of childhood conditions because of the necessary long time span separating treatment (childhood) and outcomes (elderhood). Half a century is typical and that is long enough for participants to find out about their assignments at randomisation and modify their behaviours accordingly. Observational studies such as this do not allow strong causal inference, although support for the long arm thesis can be strengthened in the two ways demonstrated here, namely by going across countries and going under the skin. The strengths were mentioned throughout but three need emphasis. First, this is the first analysis to study 28 countries in the field of Health and Retirement Study, with closely comparable key exposure and health outcomes that speak directly to the UN Decade of Healthy Ageing (UN 2021). Second, all the parallel surveys are ongoing, making this work a unique pad from which to launch future works on the consequences of child poverty in old age, on age trajectories of myriad health outcomes in ageing populations in industrial societies. Lastly, it shows what can be achieved with a clear framework, a new biological theory and a robust method applied to parallel studies. Reflection on Recent Literature These findings echo recent studies in different countries though not all (Berndt & Fors 2016; Lennartsson et al. 2018; Pakpahan et al. 2017; Tampubolon 2015; Tampubolon & Fajarini 2018; Vable et al. 2019). While consensus has not yet arrived, the evidence is largely supportive of a long arm of childhood conditions. The contrast with the new findings here can arise from the outcome studied or the special nature of the sample. Here nationally representative samples are used, comparable across 28 countries. Further, fixed effects capturing idiosyncratic country effects, such as unique health system and history, are included. Even further, wide health outcomes were examined and all agree in showing that the childhood poor have worse old age health. A few studies on the long arm of childhood conditions have reported different results, which may be due to variations in variable construction, outcomes explained, methods and broader structures of societies. Vable et al. (2019) in America suggest that once

The long arm hypothesis  271 socioeconomic mobility in adulthood is considered, no association remains between childhood economic status and old age health. On this side of the Atlantic, studies in Sweden and Europe have also shown varying associations between poverty or economic status in childhood with old age health (Berndt & Fors 2016; Lennartsson et al. 2018; Pakpahan et al. 2017). But the construction of childhood conditions varies. Lennartsson and colleagues depart from the usual practice of ignoring recall bias by using latent construct of childhood poverty, in fact a full structural model, to explain old age health indicated by pain, fatigue and breathing difficulty in Sweden; meanwhile Pakpahan and colleagues also used a structural model to explain self-rated health in old age in 13 European countries. In Sweden the authors suggest that childhood conditions are no longer binding on old age health once adult conditions are considered. This may be explained in two ways. First, the extensive welfare state of Sweden has been known to deliver exceptional health services to its older population. Our own work comparing 17 countries in Europe (SHARE), analysed using latent class analysis, shows that Sweden has managed to break the link between economic position or wealth and sensory impairment (Maharani et al. 2021). Lennartsson and colleagues may have found a manifestation of Sweden’s exceptionalism in ameliorating pain, fatigue and breathing difficulty. Second, this, however, does not contradict the results above. On cognition, muscle function and depression the long arm continues to bind. Sweden’s exceptionalism may not have succeeded in breaking all the links between childhood conditions and the spectrum of health in old age. If indeed the mechanism involves epigenetic changes, amelioration is possible, although complete elimination of the links is perhaps more difficult. Future Research and Policy including the UN Decade of Healthy Ageing The strong evidence laid out here coming from 28 countries concerning wide outcomes of ageing, including probable sarcopenia, episodic memory and mental health, should enhance confidence in the long arm thesis around the world. It also offers several key implications for research and global health policy. First, retrospective childhood information can be more fruitfully examined by recognising that this information is laced with error. Childhood poverty, a latent construct indicated by several items of recalled information, can be obtained as a latent factor or a latent class. But there is an advantage to interpretation offered by a latent class (see plots above). In fact, the latent factor (a continuous variable) is often discussed in terms of childhood poor and non-poor, a compliment to latent class. Second, cohort studies such as the British 1956 and 1970 cohorts are ripe for testing the epigenetic ageing hypothesis. The childhood poor, as indicated by lack of material resources, are undergoing more DNA methylation. In due course, venous blood samples from these cohorts can be used to test this epigenetic ageing hypothesis in efforts to advance knowledge on the epigenetics of ageing and to offer stronger input to healthy ageing policy. These cohorts are ideal because their childhood information was collected prospectively. In this connection, a note is important. While this study uses broad-based DNA methylation, future work can consider tissue-specific methylation rates as well as RNA methylation specific to explaining variation in brain or psychological functioning (Widagdo & Anggono 2018). Although this study has shown for the first time that childhood poverty affects healthy ageing in all 28 countries, the epigenetic basis has come from the experience of one country, America. While encouraging, this needs to be replicated. In fact, I have shown that America’s experience, specifically how her older population was raised, differed from how Britain’s

272  Handbook of health inequalities across the life course older population grew up. Some cohorts of older Americans were raised free from the adverse effects of World War II, whereas the same cohorts on the other side of the Atlantic grew up under wartime conditions. This difference mattered for their old age trajectories of health, specifically their trajectories of depression and of allostatic load (Tampubolon & Maharani 2017b, 2018). Although it is not yet possible to study European allostatic load trajectories, because no repeated blood biomarkers are available, trajectories of depression, sarcopenia and cognition can certainly be examined. Furthermore, in a rejoinder (Tampubolon & Maharani 2017a), I conjectured that the cohort difference between the two sides of the Atlantic bookends the health trajectories of older Europeans: some who grew up under heavy influence of war (the French and the Germans) will trace trajectories akin to those of older Britons, while some who grew up under lighter war influence (the Swedes, again) will trace trajectories akin to those of older Americans. This conjecture is eminently testable with the new results on epigenetic mechanisms demonstrated here. Third, the strong result from many countries speaks directly to global health policy such as the UN Decade of Healthy Ageing. The initiative has missed an opportunity to grab the long arm of childhood conditions reaching into old age around the world. The results here furnish evidence on probable sarcopenia (derived from grip strength) and episodic memory (derived partly from delayed word recall) as well as depression. The first two are also the essential indicators of the UN initiative and both are shaped by childhood conditions in these rich countries. If the rich countries with their advanced health systems have experienced this, the poor countries are also likely to carry the long reach of childhood conditions. Moreover, the 28 countries in this study are high-income countries; those who grew rich before growing older. Other countries are in a decidedly more difficult predicament of growing old before growing rich, for example India and China. The WHO observed that the countries with the difficult predicament are also growing old faster: what took France more than a century will take China only a few decades (Beard et al. 2016). These countries should be the focus of future research on the long arm of childhood conditions. With the higher prevalence of childhood poverty in low- and middle-income countries and its long arm reaching into old age, as evinced here, the need for research in low- and middle-income countries is now urgent. Note that many of these countries straddle the hot and humid equator, rendering one of the indicators in ELSA and SHARE wholly inappropriate, i.e. central heating. Replacement indicators that are more appropriate to geography and climate must be sought. The knowledge that childhood lasts a lifetime and the tools to construct latent childhood poverty from available indicators combine to motivate such an effort. In conclusion, the long arm of childhood conditions reaches into old age, risking wide variation in outcomes of ageing through its mark on the epigenome which compromises various organ systems. While more knowledge is needed, especially from low- and middle-income countries on the rates of epigenetic change, enough strong evidence is at hand to help secure a decade of healthy ageing by proving the necessity of eliminating child poverty.

REFERENCES Barker, DJP et al. 2005 ‘Trajectories of growth among children who have coronary events as adults’ New England Journal of Medicine 353/17: 1802–1809 Beard, JR et al. 2016 ‘The World report on ageing and health: a policy framework for healthy ageing’ The Lancet 387/10033: 2145–2154

The long arm hypothesis  273 Berndt, H & Fors, S 2016 ‘Childhood living conditions, education and health among the oldest old in Sweden’ Ageing & Society 36/3: 631–648 Birn, AE, Pillay, Y & Holtz, TH 2017 Textbook of Global Health Fourth Edition, Oxford, New York: Oxford University Press Carey, N 2011 The Epigenetics Revolution London: Icon Books Case, A, Fertig, A & Paxson, C 2005 ‘The lasting impact of childhood health and circumstance’ Journal of Health Economics 24/2: 365–389 Courtin, E et al. 2015 ‘Are different measures of depressive symptoms in old age comparable? An analysis of the CES‐D and Euro‐D scales in 13 countries’ International Journal of Methods in Psychiatric Research 24/4: 287–304 Crimmins, EM et al. 2020 HRS Epigenetic Clocks Ann Arbor, Michigan: Survey Research Center, University of Michigan Crimmins, EM et al. 2021 ‘Associations of age, sex, race/ethnicity, and education with 13 epigenetic clocks in a nationally representative U.S. sample: The Health and Retirement Study’ The Journals of Gerontology: Series A 76/6: 1117–1123 Cruz-Jentoft, AJ et al. 2019 ‘Sarcopenia: Revised European consensus on definition and diagnosis’ Age and Ageing 48/1: 16–31 Gillman, MW 2005 ‘Developmental origins of health and disease’ New England Journal of Medicine 353/17: 1848–1850 Hayward, MD & Gorman, BK 2004 ‘The long arm of childhood: The influence of early-life social conditions on men’s mortality’ Demography 41/1: 87–107 Jain, B 2021 ‘Social determinants of elderhood’ Nature Aging 1/5: 413–415 Lennartsson, C et al. 2018 ‘Social class and infirmity. The role of social class over the life-course’ SSM – Population Health 4: 169–177 Maharani, A et al. 2021 ‘Healthcare system performance and socioeconomic inequalities in hearing and visual impairments in 17 European countries’ European Journal of Public Health 31/1: 79–86 Nazroo, J 2017 ‘Class and health inequality in later life: Patterns, mechanisms and implications for policy’ International Journal of Environmental Research and Public Health 14/12: 1533 Pakpahan, E, Hoffmann, R & Kröger, H 2017 ‘The long arm of childhood circumstances on health in old age: Evidence from SHARELIFE’ Advances in Life Course Research 31: 1–10 Stafford, M et al. 2015 ‘Childhood environment and mental wellbeing at age 60–64 years: Prospective evidence from the MRC National Survey of Health and Development’ PLOS ONE 10/6: e0126683 Suzman, RM, Willis, DP & Manton, KG eds. 1996 The Oldest Old Oxford, New York: Oxford University Press Szyf, M 2013 ‘How do environments talk to genes?’ Nature Neuroscience 16/1: 2–4 Tampubolon, G 2010 ‘Recall error and recall bias in life course epidemiology’ Available at: https://​mpra​ .ub​.uni​-muenchen​.de/​23847/​ Accessed 2.2.2022 Tampubolon, G 2015 ‘Growing up in poverty, growing old in infirmity: The long arm of childhood conditions in Great Britain’ PLOS ONE 10/12: e0144722 Tampubolon, G & Fajarini, M 2018 ‘The strong grip of childhood conditions in older Europeans’ biorxiv: 267526 Tampubolon, G & Maharani, A 2017a ‘Depression trajectories of older Americans and Britons 2002–2012: A rejoinder and a hope’ American Journal of Geriatric Psychiatry 25/11: 1198 Tampubolon, G & Maharani, A 2017b ‘When did old age stop being depressing? Depression trajectories of older Americans and Britons 2002–2012’ American Journal of Geriatric Psychiatry 25/11: 1187–1195 Tampubolon, G & Maharani, A 2018 ‘Trajectories of allostatic load among older Americans and Britons: Longitudinal cohort studies’ BMC Geriatrics 18/1: 255 UN 2021 https://​www​.who​.int/​initiatives/​decade​-of​-healthy​-ageing. Accessed 3 February 2022. Vable, AM, Gilsanz, P & Kawachi, I 2019 ‘Is it possible to overcome the “long arm” of childhood socioeconomic disadvantage through upward socioeconomic mobility?’ Journal of Public Health 41/3: 566–574 Vermunt, JK 2010 ‘Latent class modeling with covariates: Two improved three-step approaches’ Political Analysis 18/4: 450–469

274  Handbook of health inequalities across the life course Widagdo, J & Anggono, V 2018 ‘The m6A-epitranscriptomic signature in neurobiology: From neurodevelopment to brain plasticity’ Journal of Neurochemistry 147/2: 137–152

18. Childhood conditions and health later in life: examples from Sweden Serhiy Dekhtyar and Stefan Fors

INTRODUCTION Childhood and childhood experiences hold a special place in life-course research. There are several reasons for this. First and foremost, childhood is a formative period during which crucial physical, psychological, and social faculties develop, and any disturbances to this development may leave life-long consequences (Kuh et al. 2003). Yet, childhood also holds a special place in life-course research on inequalities for other reasons. While some might argue that inequalities observed in adulthood are, to some extent, contingent on the efforts and behaviours of the individuals themselves, it is more complicated to hold individuals responsible for inequalities rooted in childhood conditions. The main purpose of this chapter is to leverage research from Sweden on long-term associations between childhood socioeconomic conditions and health later in life, to illustrate and discuss the evidence for long-term health effects of exposures during childhood. We have chosen to focus on two types of outcomes: cognitive function and mortality. We chose these outcomes for three reasons. First, they constitute major public health concerns. Second, a substantial amount of research has been conducted on these outcomes in relation to childhood conditions in Sweden. Third, these outcomes are different in nature. While they partially share similar causative factors (e.g., cardiovascular health), they are characterized by substantive differences, which have to be considered when setting up the analyses and interpreting the results. For one thing, cognitive function represents a continuum, whereas mortality is a binary, absorbing, state. In addition, cognitive function is highly heritable, while the heritability of longevity is limited. We believe that these similarities and differences are useful for illustrating and discussing life-course processes. While there are other approaches to studying and understanding health inequalities from a life course perspective (e.g. cumulative disadvantage theory; ecosocial theory), much of the research have been motivated by the heuristic models described by Kuh et al. (2003). In terms of childhood consequences, these models tend to distinguish between two fundamentally different types of mechanisms. The first type of mechanism, the critical or sensitive period, suggests that exposures in childhood that hamper optimal development may have life-long consequences for health that are difficult to compensate for later in life. Much of the research on the consequences of disturbances during sensitive periods have focused on cardiovascular health outcomes later in life. The potentially long-lasting effects of childhood exposures on cardiovascular health have a bearing on our examples in this chapter as well, since cardiovascular health affects both cognitive aging and mortality risk. The other type of heuristic model, the chains of risk model, suggests that rather than having a direct detrimental effect on health itself, childhood conditions may affect health later in life by serving as the starting point for subsequent chains of risk (Kuh et al. 2003). This type of 275

276  Handbook of health inequalities across the life course model is sometimes also referred to as an “unhealthy life career” model (Lundberg 1993). Here, the mechanisms linking childhood conditions to later health runs through a series of associations running across the life course. Typically, educational attainment is assumed to play a pivotal role in chains of risks that lead to inequalities in cognition and mortality later in life. It is well known that children who grow up in working-class homes are less likely to attend higher education than children who grow up with white-collar parents (Bukodi, Erikson, and Goldthorpe 2014). This selection into higher education works through two types of mechanisms: primary and secondary selection. Primary selection occurs as children from privileged social backgrounds, on average, outperform children from disadvantaged social backgrounds in school. As a consequence, children from privileged backgrounds are disproportionally often eligible for advanced education. Secondary selection occurs as children from privileged social backgrounds are more likely to attend academic educational tracks than children from disadvantaged social backgrounds – even when they perform at the same level academically. Taken together, these mechanisms generate systematic inequalities in educational attainment based on the class background of children (Erikson and Rudolphi 2010; Fors, Torssander, and Almquist 2018). Such differences in educational attainment are, in turn, likely to affect the likelihood of behaviours and exposures in ways that generate inequalities in cognition and mortality risk later in life. Life course epidemiology models offer a useful framework for assessing the long-term impacts of childhood social exposures, as we will clearly show in this chapter. Yet, they also occasionally fall short of accounting for the full range of complexity, particularly as it relates to outcomes occurring during aging. By using the examples of late-life cognition and mortality, we will expose some limitations of conventional life course models, particularly with respect to their handling of non-binary endpoints, the heterogeneity of change trajectories in the lead-up to functional impairment thresholds, and genetic confounding. In our conclusions, we will also highlight some conceptual and methodological extensions to life course epidemiology models in light of the two outcomes we chose to highlight. Our intention with this chapter is to put forward the argument that in order to be most effective, heuristic life-course models ought to be complemented with specific knowledge about the features of the outcomes under study.

COGNITION THROUGHOUT LIFE: DISTINGUISHING BETWEEN LEVELS AND CHANGES Cognitive decline is pervasive even in healthy older adults without dementia. It is associated with a reduction in the ability to perform everyday functions, make important decisions, and live independently. Within any given population, there is substantial heterogeneity in terms of cognitive development and subsequent decline throughout the life course. Different cognitive domains may follow different patterns of change. For example, fluid abilities, which reflect processing aspects of cognition such as speed and memory, begin to decline already during middle age (Rönnlund and Nilsson 2006). In contrast, crystallized abilities, which reflect declarative and procedural knowledge such as vocabulary or specialized domain skills, are much less adversely affected by old age. Overall, correlations across domains of cognitive ability levels are both positive and sizeable. This covariation is commonly operationalized as a factor

Childhood conditions and health later in life  277 variable indicating general intelligence (g). Such factor variables tend to account for about half of the variation in the individual cognitive ability domains. Similarly, longitudinal rates of change correlate to a significant extent – with about two-thirds of the variance in change being shared across different cognitive domains (Lövdén et al. 2020). And yet, level of cognitive function exhibits only inconsistent and small associations with subsequent aging-related cognitive changes. For example, Tucker-Drob et al. (2019), have shown that the meta-analytic correlation between levels of cognitive abilities and rates of longitudinal cognitive changes was just r = .047, suggesting that aging may not necessarily be kinder to those initially more able (Gow et al. 2012; Tucker-Drob, Brandmaier, and Lindenberger 2019). In fact, predicting between-person differences in the rates of cognitive change during aging has proven to be considerably more difficult than predicting the differences in the levels of cognitive performance in old age (Lövdén et al. 2020). For instance, Ritchie and colleagues have shown that a wide range of sociodemographic, clinical, functional, and genetic indicators explained about 80% of the variance in cognitive levels at 70 years but only about 16% of the variance in decline between ages 70 and 76 years (Ritchie et al. 2016). These observations bear important implications for interpreting the epidemiological literature on the link between childhood social experiences and cognitive outcomes subsequently discussed in this chapter.

DEMENTIA DIAGNOSIS THRESHOLD: IMPLICATIONS FOR COGNITIVE LEVEL AND CHANGES In addition to cognitive decline due to normative aging-related processes, cognitive impairment may also occur due to disease-related influences. Distinguishing between the two can be difficult, especially during the initial stages of decline and among very old individuals (Grande, Qiu, and Fratiglioni 2020). Dementia is primarily diagnosed on the basis of an individual’s cognitive ability declining below a functional threshold in more than one domain, with implications for their daily activities and independence. Whether or not a person performs below a diagnostic threshold for dementia may be influenced by their peak attained cognitive levels, and/or by the factors influencing the rate of decline (e.g., the severity of the pathological insult, the extent of reserve capacity that enables the maintenance of function in the face of neurologic damage, or other factors (Liu, Jones, and Glymour 2010; Stern 2012); see Figure 18.1 for an illustration). As we turn to examining the impact of early-life factors on late-life cognitive outcomes we will find ourselves needing to reconcile the associations between early-life factors and cognitive change (generally inconsistent) with the associations between early-life factors and dementia risk (generally much more consistent). Doing so will require accepting that due to the threshold nature of dementia diagnosis (i.e., the diagnosis of dementia is given when cognitive function dips below a minimum performance threshold), the association between early-life factors and peak cognitive levels (often strong and consistent) is alone sufficient to account for the early-life factor–dementia risk association (Fratiglioni, Marseglia, and Dekhtyar 2020; Lövdén et al. 2020). This implies that a factor only associated with cognitive levels may still contribute to a differential timing of dementia onset, even if it has no relevance for the rate of cognitive decline.

278  Handbook of health inequalities across the life course

Note: Hypothetical individual curves [thinner lines] and population average [thick line]. Curves discontinue at different ages representing individual variation in the age at death. The dotted line represents a functional level, below which cognitive function is too low for independence in everyday life. Individuals cross the functional impairment threshold at different ages, depending on the peak cognitive levels attained, the speed of decline, or the combination of the two processesple Source: Figure adapted from Tucker-Drob (2019).

Figure 18.1

A representation of life-course development of cognitive function

FRAMEWORK FOR UNDERSTANDING THE IMPACT OF EARLY-LIFE FACTORS ON COGNITION AND DEMENTIA Given the lifelong variability in cognitive development and the multifactorial nature of dementia risk throughout life (Whalley, Dick, and McNeill 2006), the framework offered by life-course epidemiology (Kuh et al. 2003) is particularly well-suited to assessing the contribution of childhood factors to both these outcomes. Life-course epidemiology focuses on stages, durations, and sequences of exposures to co-occurring and interacting genetic and environmental factors as they shape health and disease risk throughout life. In the next section, we provide an account of the possible connections between childhood factors, cognition, and dementia using the life-course epidemiology models of sensitive periods, and chains of risk. Sensitive period models emphasize exposures occurring during key phases, such as during rapid tissue growth (Liu et al. 2010), at periods of key neurologic structure development (Fox, Levitt, and Nelson III 2010), or in moments of heightened plasticity in epigenetic marks (Karlsson et al. 2021). The foundations of reserve, the capacity to withstand pathological insults and retain function in the face of damage due to structural properties (brain reserve) or enhanced efficiency and flexibility of cognitive function (cognitive reserve), are both laid out during the early years of life (Stern et al. 2019). For instance, the volumes of the hippocampus (a brain area often first impaired in Alzheimer’s disease) have been shown to be reduced in children from low socioeconomic backgrounds (Yu et al. 2018). Furthermore, in a cohort of

Childhood conditions and health later in life  279 1,099 individuals between three and 20 years from the US, parental education and family income have been linked with brain structural development in regions that are critical for language, executive function, and memory (Noble et al. 2015). Findings linking dementia with childhood personality (Chapman et al. 2020) and early-life cognitive ability (Dekhtyar et al. 2016) – two exposures strongly influenced by childhood social conditions – are also consistent with the predictions of the sensitive period model (particularly when these associations are preserved after accounting for the mediating influence of adult factors (Greenfield, Moorman, and Rieger 2021; Richards et al. 2019)). Chains of risk models envision that childhood exposures are associated with old-age cognitive outcomes not as a result of direct linkages, but by enabling adult exposures which then influence cognitive outcomes. For example, experience of adverse social conditions in childhood may impair educational and occupational attainment – both well-established correlates of dementia risk (Meng and D’arcy 2012; Then et al. 2014), but arguably less robust predictors of cognitive change (Berggren, Nilsson, and Lövdén 2018; Seblova, Berggren, and Lövdén 2020). Furthermore, through physiological, psychological, and health-behavior mechanisms adverse childhood conditions may increase the likelihood of poor health in adulthood (Cohen et al. 2010). In turn, the link between single diseases, particularly of the cardiovascular system, and cognitive impairment and dementia is well established (Qiu and Fratiglioni 2015), while recent evidence is also pointing towards complex clinical phenotypes such as multimorbidity and frailty as being implicated in dementia risk (Bai et al. 2021; Grande et al. 2021).

ROLE OF GENES IN CHILDHOOD EXPOSURES–LATE-LIFE COGNITION ASSOCIATIONS The heritability of general cognitive ability is both substantial and variable throughout life: it increases from about 40% in childhood, peaks at 66–80% in early adulthood, and settles at about 60% by late old age (Haworth et al. 2010; Johnson, McGue, and Deary 2014). In addition to heritability varying by age, it is believed to also vary in accordance with individuals’ social origins: it is thought to be greater in socioeconomically advantaged families than in the disadvantaged ones (Giangrande et al. 2019). Described as the Scarr–Rowe hypothesis, this observation is attributed to environmental disadvantages hindering the ability of individuals reared in lower SES households to realize their intellectual potential, while the socioeconomically advantaged families are thought to be better placed to provide environmental conditions matching their children’s genetic dispositions (Rowe, Jacobson, and Van den Oord 1999; Scarr-Salapatek 1971). The genes × SES interaction on intelligence and scholastic achievement has been shown to be more pronounced in contexts with higher social stratification in access to education and with more restricted social-welfare regimes, such as the US as opposed to Europe and the Nordics (Tucker-Drob and Bates 2016), although it has been observed in German samples too (Baier and Lang 2019). In addition to genetic impacts on cognitive abilities (which may be differentially expressed in accordance with childhood socioeconomic origins), genetic factors are also likely involved in the shaping of the childhood social exposures themselves. For example, genetic influences on educational attainment (Lee et al. 2018; Okbay et al. 2016), but also on childhood environments and early life socioeconomic status have been reported previously, with heritability estimates in the region of 20–40% (Trzaskowski et al. 2014; Vinkhuyzen et al. 2010). Therefore,

280  Handbook of health inequalities across the life course the life-course association between childhood social experiences and long-term cognitive outcomes may to some extent be attributable to common genetic influences. Indeed, a study from the Swedish Adoption/Twin Study of Aging comparing reared-apart monozygotic and dizygotic twins with twins reared together has found that in unrelated individual analysis, childhood social class was associated with mean-level cognitive performance at age 65, but not with the rate of cognitive change. Furthermore, in between–within twin-pair analyses, childhood social class associations with cognitive levels were attenuated, suggesting the influence of genetic confounding (Ericsson et al. 2017).

CHILDHOOD SOCIAL EXPOSURES AND LONG-TERM COGNITIVE OUTCOMES: EVIDENCE FROM SWEDEN Associations with cognitive levels and changes. In a nationally representative sample from the Longitudinal Study of Living Conditions of the Oldest Old, Fors and colleagues (2009) examined the impact of several early life conditions, including financial difficulty, family size, parental absence, father’s social class, and conflicts in the family on cognitive functioning after age 77 (measured with an abridged version of Mini-Mental State Examination). They found that the father’s social class and conflicts in the family were associated with late-life cognitive levels, although the impact of the former was considerably attenuated by adulthood SES (as would be predicted by the chain or risk model), while the impact of the latter was largely independent of adulthood measures (as would be predicted by the sensitive period model) (Fors, Lennartsson, and Lundberg 2009). Interestingly, a later study using the same material (Andel, Silverstein, and Kåreholt 2015) has shown that the sensitive-period influences of early-life family conflicts found in Fors et al. (2009) were attenuated upon accounting for individuals’ mild-life occupational complexity and leisure participation, suggesting the lifelong chain of psychosocial risk factors for late-life cognition. When it comes to examining cognitive change trajectories using Swedish data, the aforementioned work by Ericsson et al. (2017) is one of the few examples specifically focusing on childhood social conditions. Their finding of a limited contribution of early life social conditions is echoed in two studies from Sweden examining the role of education in cognitive aging (Berggren et al. 2018; Karlsson et al. 2015), which also didn’t yield a change effect. Associations with dementia risk. Two Swedish studies have related measures of childhood academic performance at age ten (based on grades obtained from school archives) with clinical- (Dekhtyar et al. 2016), as well as register-based diagnoses of dementia (Dekhtyar et al. 2015). Both findings revealed a weak-to-moderate sensitive-period effect of early academic performance on dementia risk that remained even after including midlife indicators; a finding echoing earlier reports of sensitive-period influences of early-life education in dementia, which did not appear to be mediated by adult SES or socioeconomic mobility (Karp et al. 2004). The importance of early-life periods in the formation of dementia risk is further highlighted by the study from the Swedish Twin Register that found birth characteristics, such as size, weight, and head circumference – although not parental SES or early-life-education – to be associated with cognitive impairment and dementia diagnosis before 89 years (Mosing et al. 2018). Authors attributed the lack of social circumstances’ effects to the likely underestimation of dementia cases in inpatient records, limited age span for dementia ascertainment, and differences in help-seeking behavior across SES, which could lead to lower sensitivity of diag-

Childhood conditions and health later in life  281 nosis in less advantaged groups, thus concealing the beneficial impacts of early social class or education. In addition to the sensitive-period influences of early-life factors, chain-of-risk patterns have been reported in Swedish samples too. A study by Wang et al. (2017) has found that an index incorporating early-life education, occupation (before 20 years), and family size was associated with dementia risk after 75 years in an urban population-based sample from Stockholm, although nearly 40% of this effect was mediated by mid-life measures of occupational attainment, and late-life measures of social network and leisure participation (Wang et al. 2017). This brief synopsis of studies linking childhood exposures with cognitive levels, changes, and dementia in late life suggests that in Sweden, just like elsewhere in the West, the contribution of childhood social exposures to late-life cognition is likely attributed to their role in shaping peak cognitive levels that are then preserved throughout life (i.e., preserved differentiation), rather than in differentially affecting the rates of cognitive change (i.e. differential preservation) (Lövdén et al. 2020; Salthouse et al. 1990). And given the threshold nature of dementia diagnosis, these preserved differences in cognitive levels can yield differences in the timing of the diagnosis according to background SES or childhood education. Importantly, these effects can still have major public health relevance, as delaying dementia by only a few years could substantially reduce its prevalence and related human and economic burden (Brookmeyer, Gray, and Kawas 1998)

CHILDHOOD SOCIOECONOMIC CONDITIONS AND MORTALITY LATER IN LIFE A substantial chunk of the literature on socioeconomic conditions in childhood and health later in life concerns differences in mortality. There is good reason for this. It has been argued that inequalities in length of life represent the most fundamental form of inequalities, as all other forms of inequalities are conditional on being alive (van Raalte, Sasson, and Martikainen 2018; however, see Davey Smith et al. 1992 for a notable exception). Yet, mortality can be, and has been, assessed in several different ways in the literature on socioeconomic conditions in childhood and health later in life. Perhaps the most common way of assessing mortality disparities is through all-cause mortality. That is, by assessing differences in all mortality, regardless of the cause. This presents an efficient way of analysing the sum of differences in all causes of death. Thus, the strength of this approach is that it provides a summary account of all mortality inequalities, regardless of the causal mechanisms at play. The main drawback of this approach is that it does not provide any information on the aetiological processes that underlie the observed inequalities. In order to get closer to understanding the causal nature of the associations between socioeconomic conditions in childhood and later mortality, we need measures that account for the cause of death. Different types of mortality can be classified in different ways. One common way of classifying mortality by cause of death that is commonly used in the literature on health inequalities is to distinguish between amenable causes of death and non-amenable causes of death (Phelan et al. 2004). This distinction has been used to design studies based on the logic behind fundamental cause theory (FCT) (cf. Mackenbach et al. 2015; Plug et al. 2012). In short, FCT posits that individuals will use their available resources to avoid morbidity and mortality or, as

282  Handbook of health inequalities across the life course Deaton (2013) puts it: “power and money seek out health improvements”. Yet, this can only be efficiently achieved for causes of poor health that are amenable. Thus, to the extent that FCT holds true, we would expect to find stronger health inequalities for amenable mortality than for non-amenable mortality. In order to get closer to specific etiological models for the associations between socioeconomic conditions in childhood and mortality later in life, we need to assess inequalities in cause-specific mortality. That is, mortality from specific causes of death, typically operationalized by the international classification of diseases (ICD) nomenclature. For example, much of the research on socioeconomic conditions in childhood and mortality later in life have focused on mortality from different types of cardiovascular diseases. There are empirical and theoretical reasons to believe that childhood disadvantage may lead to increased risks of cardiovascular mortality, either by disrupting the biological development leading to a life-long increased vulnerability to cardiovascular events, or by serving as a steppingstone for life-course chains of exposures and behaviours that eventually increase the risk of cardiovascular disease (Galobardes, Davey Smith, and Lynch 2006).

FRAMEWORK FOR UNDERSTANDING THE IMPACT OF SOCIOECONOMIC CONDITIONS IN CHILDHOOD ON MORTALITY LATER IN LIFE Childhood as a Sensitive Period Much of the literature on the associations between socioeconomic conditions in childhood and mortality later in life have concerned the extent to which early life and childhood serves as a sensitive period, where sub-optimal conditions lead to an increased vulnerability to morbidity and mortality later in life. This notion has a long history in public health research, reaching back at least to Francis Bacon’s writings from the 16th century (Christensen 2019). In recent times, two of the most prominent proponents of this perspective are the late David Barker and the Nobel-prize winning economist James Heckman. David Barker is known for the “Barker hypothesis”, suggesting that suboptimal environmental conditions during pregnancy and infancy leave the child at an increased risk of a range of health problems, including cardiovascular mortality, throughout the rest of the life course. There is evidence that there are indeed associations between growth during pregnancy and early infancy and later cardiovascular mortality, but the causal nature of these associations has still not been convincingly established (Barker 1995; Christensen 2019). James Heckman and colleagues have mainly focused on environmental conditions during the first years of life. By following-up and evaluating pre-school programmes geared towards disadvantaged children, they argue that such programmes have substantial beneficial consequences for a wide range of outcomes later in life, including risk of cancer, cardiovascular health, and all-cause mortality (García and Heckman 2021). Childhood as a Steppingstone (for Chains of Risk) Another type of causal mechanism, through which socioeconomic conditions in childhood may affect mortality in later life is chains of risks that start in childhood and cascade throughout

Childhood conditions and health later in life  283 the life course. Perhaps the most important lever in these types of mechanisms is educational attainment. As described in the introduction, differences in childhood socioeconomic conditions tend to lead to inequalities in educational attainment, where children from working-class families are less likely to get high grades and go on to further education. Such inequalities in educational attainment may, in turn, translate into differences in socioeconomic trajectories throughout the life course, where those who do not attend higher education are more likely to be exposed to a range of hazardous exposures that increase the risk of morbidity and mortality. These exposures include both behavioural risk factors, such as smoking and sedentary lifestyles, and environmental conditions, such as hazardous working conditions (Cutler and Lleras-Muney 2012). Confounding The vast majority of studies of the associations between socioeconomic conditions in childhood and mortality later in life are based on observational data. Thus, the basis for causal inference is hampered by the threat of residual confounding. That is, there is always a possibility that there are variables not accounted for in the analytical model that may explain the observed associations (Mackenbach 2019). As described for cognitive outcomes, one such potential source of residual confounding is genetics. It is possible that some characteristics of the childhood environment may be, partially or wholly, attributable to the parents’ genetic make-up. These characteristics may, in turn, be inherited by the child and affect her mortality risk later in life. For example, in some instances an adverse childhood environment may be caused by poor health or addiction among the parents. These are traits that are known to be partially heritable. Hence, the child then runs an increased risk of poor health or addiction herself – partially as a consequence of her genetic inheritance. In other words, one may reasonably expect part of the observed associations between socioeconomic conditions in childhood to be confounded by genetics. This is a problem, as most (albeit not all, see more below) observational studies lack genetically informative data (D’Onofrio et al. 2013). Yet, there is reason to believe that the problem of genetic confounding is likely to be limited when it comes to the association between childhood socioeconomic conditions and later mortality. In order for there to be a strong confounding effect of genetics on the association between socioeconomic conditions in childhood and mortality later in life – there would have to be a substantial effect of genetics on later mortality. One way of assessing the effect of genetics on mortality is to assess the heritability of longevity, that is, by using genetically informative data to analyse to what extent differences in longevity can be attributed to genetics. There are several examples of this in the literature, based on different data sources and study design. Historically, the estimates have ranged between 15% and 30%. That is, about 15–30% of the individual variation in length of life can be attributed to genetics (Ruby et al. 2018). Yet, these estimates have been challenged lately. In 2018, researchers from Calico Life Sciences LLC in collaboration with researchers from the Ancestry corporation presented a new, large-scale, study of the heritability of longevity. Drawing on pedigree data from Ancestries registries, including hundreds of millions of historical persons, they were able to leverage the relationships between individuals to assess the heritability of human longevity. They found that the “nominal heritability” estimates, based on the correlation between genetic

284  Handbook of health inequalities across the life course relatives, largely mimicked the estimates from the previous literature. However, when they accounted for assortative mating, the estimates were substantially reduced. In other words, the results suggest that previous estimates have been inflated as they have not accounted for the fact that individuals with traits beneficial to longevity tend to choose, and have children with, mates with similar traits. When accounting for this, the findings suggest that the true heritability of human longevity during the 1800s and early 1900s was less than 10%. That is, less than 10% of the individual variation in longevity could be attributed to genetics (Ruby et al. 2018). Thus, while some of the observed associations between socioeconomic conditions in childhood and mortality later in life could potentially be attributable to genetic confounding, this source of confounding is bounded by the limited heritability of longevity. Other potential sources of confounding, such as geographical and cultural differences, may also play a role. Yet, indicators for these are more likely to be available in the data, which makes it possible to adjust for them statistically.

CHILDHOOD AND MORTALITY LATER IN LIFE: EVIDENCE FROM SWEDEN In the Swedish context, several studies have analysed the potential associations between socioeconomic conditions during childhood and mortality during adulthood and old age, using different data sources and study designs. Most studies do find an association between childhood socioeconomic conditions and later mortality, where those born in families with less favourable socioeconomic profiles (most commonly measured by parents’ social class) have an increased risk of mortality later in life. These patterns seem to hold true for all-cause mortality as well as for a wide range of cause-specific mortality types (Donrovich, Drefahl, and Koupil 2014; Ericsson et al. 2019; Falkstedt, Lundberg, and Hemmingsson 2011; Fors et al. 2011, 2012; Hemmingsson and Lundberg 2005; Jackisch, Brännström, and Almquist 2019; Juárez, Goodman, and Koupil 2016; Kåreholt 2001; Lawlor et al. 2006). In addition, some of these studies provide further evidence on the nature of the associations. In several studies a substantial attenuation of the associations are observed when own socioeconomic position in adulthood is adjusted for (Donrovich, Drefahl, and Koupil 2014; Fors, Lennartsson, and Lundberg 2011; Kåreholt 2001; Lawlor et al. 2006). These findings are in line with the chains of risks hypothesis. In addition, in one study an interaction effect between childhood social class and own social class in adulthood was observed. The association between own social class and mortality was strongest for those who grew up with fathers who were non-manual workers. The authors suggest that this may be due to selective social mobility – where many of those who were downwardly socially mobile were so due to poor health or health-related problems (Fors et al. 2011). On the other hand, Hemmingsson and Lundberg (2005) found that the father’s social class, together with crowded housing in childhood, low stature, and low education could statistically explain substantial parts of the association between own social class and mortality in adulthood. These findings suggest that not all the association between childhood socioeconomic conditions and social disadvantage is mediated through adult social class, but rather that exposures in childhood in conjunction with selective social mobility contribute to mortality inequalities in adulthood.

Childhood conditions and health later in life  285 Jackisch et al. (2019) found that adjusting for indicators of childhood disadvantage (involvement with child welfare services and family dysfunction) fully attenuated the link between childhood social class and later mortality, suggesting that the link may be attributed to an increased risk of childhood disadvantages in the working-class families, rather than to socioeconomic variation within the normal range. Finally, Ericsson et al. (2019) used genetically informative data to analyse the associations between socioeconomic conditions at different stages of the life course and preventable and non-preventable mortality later in life, while accounting for potential genetic confounding. This was possible by using data from twins reared apart, who were discordant in terms of childhood social class. The results showed that those who grew up in lower social class homes had an increased risk for preventable mortality in adulthood. For non-preventable mortality, the results were ambiguous and characterized by lack of precision. Comparing twins reared apart, the results also suggests that the associations with preventable mortality may be largely attributable to familial and genetic confounding. In sum, most Swedish studies suggest that socioeconomic conditions in childhood are associated with mortality later in life. The effect sizes are substantial for several causes of death but tend to be substantially attenuated when adjusting for potential confounders and for own socioeconomic conditions later in life, suggesting that part of the effect may be as a steppingstone for chains of risk cascading forward over the life course. The only study that used genetically informative data suggests that the association between childhood class and later mortality may be largely driven by familial and genetic confounding. Yet, the authors acknowledge that the results should be interpreted cautiously, as the sample size of twins reared apart is small and the estimates are bounded by substantial uncertainty.

CONCLUSIONS In this chapter we discussed evidence (mostly from Sweden) linking childhood socioeconomic exposures with cognition and mortality later in life. In doing so, we aimed to highlight the complexities of relating separate long-term outcomes with seemingly the same early-life risk factors. Our chapter revealed that while traditional models of life course epidemiology (e.g., sensitive periods, chains of risk) offer a useful unified framework for mapping the long-term impact of early-life factors, special considerations unique to the outcomes under study also apply, as evidenced by our discussion of issues relating to the level and change influences on cognition, the threshold nature of dementia diagnosis, the dichotomy of preventable and non-preventable causes of mortality, and, last but not least, genetic confounding. Our discussion in many ways follows recent critiques and extensions of the life-course epidemiology framework, particularly as it relates to aging research (Ben-Shlomo, Cooper, and Kuh 2016). Original life-course epidemiology models were developed with a view on binary endpoints. For aging, this view may be reductionist, given the profound role that maintenance of cognitive and physical function plays in older adults’ well-being (Beard et al. 2016). Binary outcomes such as mortality, while informative, can ignore preceding functional trajectories, thus concealing important phenotypic heterogeneity that has been well described in older adults (Newman et al. 2016). As evidenced by our discussion of cognition and dementia, the same endpoint may be reached through suboptimal development, accelerated decline, or a combination of both processes. Identifying early life social correlates of these distinct

286  Handbook of health inequalities across the life course pathways can lead to more effective risk stratification, improved understanding of etiological pathways, and enhanced ability to study preclinical disease, thus widening the scope for potential intervention (Ben-Shlomo et al. 2016). Our chapter has also touched upon the concepts of reserve and resilience, which while only superficially discussed in the original papers on life course epidemiology, may represent a likely pathway connecting early life social exposures with late-life outcomes. Reserve has been widely employed in neuroepidemiology to explain why some individuals maintain cognitive function and remain dementia-free despite considerable brain pathology (Stern et al. 2020). A related concept, resilience, refers to the organism’s ability to withstand decline or recover function in the face of physical stressors (Whitson et al. 2016). For example, a recent study has shown that as many as 16% of older adults exhibit lower levels of frailty (a state of increased biological vulnerability), than would be predicted by their clinical diseases and background characteristics, suggesting some may be more resilient to the pathological declines associated with aging (Wu et al. 2020). Because both reserve and resilience are believed to be shaped by the individuals’ intrinsic biological resources, but also by the social and psychosocial environments they inhabit (Colón-Emeric et al. 2020), there is a real possibility that childhood social exposures affect late-life health by enriching or depleting the organism’s resilience against a plethora of life-course stressors. And while considerable questions remain about how reserve and resilience are formed, or how they should be measured (Stern et al. 2020; Whitson et al. 2018), future studies on childhood social experiences ought to consider this pathway, alongside the more conventional life-course epidemiology models. In addition to conceptual considerations, life-course-informed studies on the long-term consequences of childhood social exposures face substantial methodological challenges. There are obvious data availability issues when exposures and outcomes are separated in time by extended periods; and when the data are available, they are often based on retrospective self-reports, which may lead to recall bias. This issue is less severe in Sweden with access to objectively measured administrative data from registers, although information on highly relevant psychosocial exposures is typically not available in administrative sources. Then, there are analytical challenges of simultaneously modelling unconfounded, cumulative, and interacting effects of time-varying exposures and mediators (i.e., the ultimate goal of life-course epidemiology models), which is largely unachievable with standard regression models, as carefully shown by De Stavola and Daniel using the directed acyclic graph (DAG) framework (De Stavola and Daniel 2016). In addition to genetic confounding, extensively discussed in this chapter, carefully designed life-course studies need to account for time-dependent confounding when using repeated measures of exposures and mediators, to mitigate the consequences of participant attrition over extended follow-ups, and to limit measurement error. Inevitably, addressing all these challenges will prove to be difficult, if not impossible, and researchers seeking to evaluate the long-term effect of childhood social exposures will need to make assumptions about various biases. It is important that the sensitivity of these assumptions is tested as they relate to the different outcomes under study. Finally, as findings discussed here were obtained in Sweden, they may be less generalizable elsewhere. However, the landscape of risk and protective factors for major causes of death, including cognitive impairment and dementia, is likely to be similar in the West. In conclusion, the literature on the long-term health consequences of childhood social exposures represents a burgeoning field in Sweden. Future investigations in the field, while anchored in life-course epidemiology, ought to give special consideration to unique outcomes

Childhood conditions and health later in life  287 under study, thus necessitating a fine-tuning of conventional sensitive period or chain of risk models. Further developments to incorporate conceptual models of reserve and resilience, a shift from binary endpoints to change analyses, and continued methodological advancement are all warranted if this field is to push the aging research frontier.

REFERENCES Andel, R., M. Silverstein, and I. Kåreholt. 2015. ‘The Role of Midlife Occupational Complexity and Leisure Activity in Late-Life Cognition’. The Journals of Gerontology: Series B 70(2):314–21. doi: 10.1093/geronb/gbu110. Bai, G., Y. Wang, R. Kuja-Halkola, X. Li, Y. Tomata, I. K. Karlsson, N. L. Pedersen, S. Hägg, and J. Jylhävä. 2021. ‘Frailty and the Risk of Dementia: Is the Association Explained by Shared Environmental and Genetic Factors?’ BMC Medicine 19(1):248. doi: 10.1186/s12916-021-02104-3. Baier, T., and V. Lang. 2019. ‘The Social Stratification of Environmental and Genetic Influences on Education: New Evidence Using a Register-Based Twin Sample’. Sociological Science 6:143–71. Barker, D. J. P. 1995. ‘Fetal Origins of Coronary Heart Disease’. BMJ 311(6998):171–74. Beard, J. R., A. Officer, I. Araujo De Carvalho, R. Sadana, A. Margriet Pot, J.-P. Michel, P. Lloyd-Sherlock, J.E. Epping-Jordan, GMEE Geeske Peeters, and W. Retno Mahanani. 2016. ‘The World Report on Ageing and Health: A Policy Framework for Healthy Ageing’. The Lancet 387(10033):2145–54. Ben-Shlomo, Y., R. Cooper, and D. Kuh. 2016. ‘The Last Two Decades of Life Course Epidemiology, and its Relevance for Research on Ageing’. International Journal of Epidemiology 45(4):973–88. Berggren, R., J. Nilsson, and M. Lövdén. 2018. ‘Education Does Not Affect Cognitive Decline in Aging: A Bayesian Assessment of the Association between Education and Change in Cognitive Performance’. Frontiers in Psychology 9:1138. Brookmeyer, R., S. Gray, and C. Kawas. 1998. ‘Projections of Alzheimer’s Disease in the United States and the Public Health Impact of Delaying Disease Onset’. American Journal of Public Health 88(9):1337–42. doi: 10.2105/ajph.88.9.1337. Bukodi, E., R. Erikson, and J. H. Goldthorpe. 2014. ‘The Effects of Social Origins and Cognitive Ability on Educational Attainment: Evidence from Britain and Sweden’. Acta Sociologica 57(4):293–310. Chapman, B. P., A. Huang, K. Peters, E. Horner, J. Manly, D. A. Bennett, and S. Lapham. 2020. ‘Association Between High School Personality Phenotype and Dementia 54 Years Later in Results from a National US Sample’. JAMA Psychiatry 77(2):148–54. doi: 10.1001/jamapsychiatry.2019.3120. Christensen, K. 2019. ‘Early Life Events and Later Life Health: Twin and Famine Studies’. Pp. 311–17 in Old and New Perspectives on Mortality Forecasting. Springer, Cham. Cohen, S., D. Janicki-Deverts, E. Chen, and K. A. Matthews. 2010. ‘Childhood Socioeconomic Status and Adult Health’. Annals of the New York Academy of Sciences 1186(1):37–55. doi: 10.1111/j.1749 -6632.2009.05334.x. Colón-Emeric, C., C. F. Pieper, K. E. Schmader, R. Sloane, A. Bloom, M. McClain, J. Magaziner, K. M. Huffman, D. Orwig, and D. M. Crabtree. 2020. ‘Two Approaches to Classifying and Quantifying Physical Resilience in Longitudinal Data’. The Journals of Gerontology: Series A 75(4):731–38. Cutler, D. M., and A. Lleras-Muney. 2012. Education and Health: Insights from International Comparisons. National Bureau of Economic Research. Davey Smith, G., D. Carroll, S. Rankin, and D. Rowan. 1992. ‘Socioeconomic Differentials in Mortality: Evidence from Glasgow Graveyards.’ BMJ: British Medical Journal 305(6868):1554. Deaton, A. 2013. ‘What Does the Empirical Evidence Tell Us about the Injustice of Health Inequalities’. Inequalities in Health: Concepts Measures, and Ethics 263:281. Dekhtyar, S., H.-X. Wang, L. Fratiglioni, and A. Herlitz. 2016. ‘Childhood School Performance, Education and Occupational Complexity: A Life-Course Study of Dementia in the Kungsholmen Project’. International Journal of Epidemiology 45(4):1207–15. Dekhtyar, S., H.-X. Wang, K. Scott, A. Goodman, I. Koupil, and A. Herlitz. 2015. ‘A Life-Course Study of Cognitive Reserve in Dementia—from Childhood to Old Age’. The American Journal of Geriatric Psychiatry 23(9):885–96.

288  Handbook of health inequalities across the life course De Stavola, B. L., and R. M. Daniel. 2016. ‘Commentary: Incorporating Concepts and Methods from Causal Inference into Life Course Epidemiology’. International Journal of Epidemiology 45(4):1006–10. doi: 10.1093/ije/dyw103. D’Onofrio, B. M., B. B. Lahey, E. Turkheimer, and P. Lichtenstein. 2013. ‘Critical Need for Family-Based, Quasi-Experimental Designs in Integrating Genetic and Social Science Research’. American Journal of Public Health 103(S1):S46–55. Donrovich, R., S. Drefahl, and I. Koupil. 2014. ‘Early Life Conditions, Partnership Histories, and Mortality Risk for Swedish Men and Women Born 1915–1929’. Social Science & Medicine 108:60–67. Ericsson, M., C. Lundholm, S. Fors, A. K. Dahl Aslan, C. Zavala, C. A. Reynolds, and N. L. Pedersen. 2017. ‘Childhood Social Class and Cognitive Aging in the Swedish Adoption/Twin Study of Aging’. Proceedings of the National Academy of Sciences 114(27):7001. doi: 10.1073/pnas.1620603114. Ericsson, M., N. L. Pedersen, A. L. V. Johansson, S. Fors, and A. K. Dahl Aslan. 2019. ‘Life-Course Socioeconomic Differences and Social Mobility in Preventable and Non-Preventable Mortality: A Study of Swedish Twins’. International Journal of Epidemiology 48(5):1701–9. Erikson, R., and F. Rudolphi. 2010. ‘Change in Social Selection to Upper Secondary School—Primary and Secondary Effects in Sweden’. European Sociological Review 26(3):291–305. Falkstedt, D., I. Lundberg, and T. Hemmingsson. 2011. ‘Childhood Socio-Economic Position and Risk of Coronary Heart Disease in Middle Age: A Study of 49 321 Male Conscripts’. European Journal of Public Health 21(6):713–18. Fors, S., C. Lennartsson, and O. Lundberg. 2009. ‘Childhood Living Conditions, Socioeconomic Position in Adulthood, and Cognition in Later Life: Exploring the Associations’. The Journals of Gerontology: Series B 64B(6):750–57. doi: 10.1093/geronb/gbp029. Fors, S., C. Lennartsson, and O. Lundberg. 2011. ‘Live Long and Prosper? Childhood Living Conditions, Marital Status, Social Class in Adulthood and Mortality during Mid-Life: A Cohort Study’. Scandinavian Journal of Public Health 39(2):179–86. Fors, S., B. Modin, I. Koupil, and D. Vågerö. 2012. ‘Socioeconomic Inequalities in Circulatory and All-Cause Mortality after Retirement: The Impact of Mid-Life Income and Old-Age Pension. Evidence from the Uppsala Birth Cohort Study’. J Epidemiol Community Health 66(7):e16–e16. Fors, S., J. Torssander, and Y. B. Almquist. 2018. ‘Is Childhood Intelligence Associated with Coexisting Disadvantages in Adulthood? Evidence from a Swedish Cohort Study’. Advances in Life Course Research 38:12–21. Fox, S. E., P. Levitt, and C. A. Nelson III. 2010. ‘How the Timing and Quality of Early Experiences Influence the Development of Brain Architecture’. Child Development 81(1):28–40. doi: 10.1111/j.1 467-8624.2009.01380.x. Fratiglioni, L., A. Marseglia, and S. Dekhtyar. 2020. ‘Ageing without Dementia: Can Stimulating Psychosocial and Lifestyle Experiences Make a Difference?’ The Lancet Neurology 19(6):533–43. doi: 10.1016/S1474-4422(20)30039-9. Galobardes, B., G. Davey Smith, and J.W. Lynch. 2006. ‘Systematic Review of the Influence of Childhood Socioeconomic Circumstances on Risk for Cardiovascular Disease in Adulthood’. Annals of Epidemiology 16(2):91–104. García, J. L., and J. J. Heckman. 2021. ‘Early Childhood Education and Life‐Cycle Health’. Health Economics 30(Suppl 1):119–41. Giangrande, E. J., C. R. Beam, S. Carroll, L. J. Matthews, D. W. Davis, D. Finkel, and E. Turkheimer. 2019. ‘Multivariate Analysis of the Scarr–Rowe Interaction across Middle Childhood and Early Adolescence’. Intelligence 77:101400. doi: https://​doi​.org/​10​.1016/​j​.intell​.2019​.101400. Gow, A. J., W. Johnson, G. Mishra, M. Richards, D. Kuh, I. J. Deary, and HALCyon Study Team. 2012. ‘Is Age Kinder to the Initially More Able? Yes, and No’. Intelligence 40(1):49–59. Grande, G., A. Marengoni, D. L. Vetrano, A. Roso-Llorach, D. Rizzuto, A. Zucchelli, C. Qiu, L. Fratiglioni, and A. Calderón-Larrañaga. 2021. ‘Multimorbidity Burden and Dementia Risk in Older Adults: The Role of Inflammation and Genetics’. Alzheimer’s & Dementia 17(5):768–76. doi: 10.1002/alz.12237. Grande, G., C. Qiu, and L. Fratiglioni. 2020. ‘Prevention of Dementia in an Ageing World: Evidence and Biological Rationale’. Ageing Research Reviews 64:101045. doi: 10.1016/j.arr.2020.101045.

Childhood conditions and health later in life  289 Greenfield, E. A., S. Moorman, and A. Rieger. 2021. ‘Life Course Pathways from Childhood Socioeconomic Status to Later-Life Cognition: Evidence from the Wisconsin Longitudinal Study’. The Journals of Gerontology: Series B 76(6):1206–17. doi: 10.1093/geronb/gbaa062. Haworth, C. M. A., M. J. Wright, M. Luciano, N. G. Martin, E. J. C. de Geus, C. E. M. van Beijsterveldt, M. Bartels, D. Posthuma, D. I. Boomsma, O. S. P. Davis, Y. Kovas, R. P. Corley, J. C. DeFries, J. K. Hewitt, R. K. Olson, S. A. Rhea, S. J. Wadsworth, W. G. Iacono, M. McGue, L. A. Thompson, S. A. Hart, S. A. Petrill, D. Lubinski, and R. Plomin. 2010. ‘The Heritability of General Cognitive Ability Increases Linearly from Childhood to Young Adulthood’. Molecular Psychiatry 15(11):1112–20. doi: 10.1038/mp.2009.55. Hemmingsson, T., and I. Lundberg. 2005. ‘How Far Are Socioeconomic Differences in Coronary Heart Disease Hospitalization, All-Cause Mortality and Cardiovascular Mortality among Adult Swedish Males Attributable to Negative Childhood Circumstances and Behaviour in Adolescence?’ International Journal of Epidemiology 34(2):260–67. Jackisch, J., L. Brännström, and Y. B. Almquist. 2019. ‘Troubled Childhoods Cast Long Shadows: Childhood Adversity and Premature All-Cause Mortality in a Swedish Cohort’. SSM – Population Health 9:100506. Johnson, W., M. McGue, and I. J. Deary. 2014. ‘Normative Cognitive Aging’. Pp. 135–67 in Behavior Genetics of Cognition across the Lifespan. Springer. Juárez, S. P., A. Goodman, and I. Koupil. 2016. ‘From Cradle to Grave: Tracking Socioeconomic Inequalities in Mortality in a Cohort of 11 868 Men and Women Born in Uppsala, Sweden, 1915–1929’. J Epidemiol Community Health 70(6):569–75. Kåreholt, I. 2001. ‘The Long Shadow of Socioeconomic Conditions in Childhood: Do They Affect Class Inequalities in Mortality?’ Pp. 29–44 in Cradle to Grave. Sociology Press. Karlsson, I. K., M. Ericsson, Y. Wang, J. Jylhävä, S. Hägg, A. K. Dahl Aslan, C. A. Reynolds, and N. L. Pedersen. 2021. ‘Epigenome-Wide Association Study of Level and Change in Cognitive Abilities from Midlife through Late Life’. Clinical Epigenetics 13(1):85. doi: 10.1186/s13148-021-01075-9. Karlsson, P., V. Thorvaldsson, I. Skoog, P. Gudmundsson, and B. Johansson. 2015. ‘Birth Cohort Differences in Fluid Cognition in Old Age: Comparisons of Trends in Levels and Change Trajectories over 30 Years in Three Population-Based Samples.’ Psychology and Aging 30(1):83. Karp, A., I. Kåreholt, C. Qiu, T. Bellander, B. Winblad, and L. Fratiglioni. 2004. ‘Relation of Education and Occupation-Based Socioeconomic Status to Incident Alzheimer’s Disease’. American Journal of Epidemiology 159(2):175–83. doi: 10.1093/aje/kwh018. Kuh, D., Y. Ben-Shlomo, J. Lynch, J. Hallqvist, and C. Power. 2003. ‘Life Course Epidemiology’. Journal of Epidemiology and Community Health 57(10):778. Lawlor, D. A., J. A. C. Sterne, P. Tynelius, G. Davey Smith, and F. Rasmussen. 2006. ‘Association of Childhood Socioeconomic Position with Cause-Specific Mortality in a Prospective Record Linkage Study of 1,839,384 Individuals’. American Journal of Epidemiology 164(9):907–15. Lee, J. J., R. Wedow, … D. Cesarini, 23andMe Research Team, COGENT (Cognitive Genomics Consortium), and Social Science Genetic Association Consortium. 2018. ‘Gene Discovery and Polygenic Prediction from a Genome-Wide Association Study of Educational Attainment in 1.1 Million Individuals’. Nature Genetics 50(8):1112–21. doi: 10.1038/s41588-018-0147-3. Liu, S., R. N. Jones, and M. M. Glymour. 2010. ‘Implications of Lifecourse Epidemiology for Research on Determinants of Adult Disease’. Public Health Reviews 32(2):489–511. doi: 10.1007/BF03391613. Lövdén, M., L. Fratiglioni, M. M. Glymour, U. Lindenberger, and E. M. Tucker-Drob. 2020. ‘Education and Cognitive Functioning Across the Life Span’. Psychological Science in the Public Interest 21(1):6–41. doi: 10.1177/1529100620920576. Lundberg, O. 1993. ‘The Impact of Childhood Living Conditions on Illness and Mortality in Adulthood’. Social Science & Medicine 36(8):1047–52. Mackenbach, J. P. 2019. Health Inequalities: Persistence and Change in Modern Welfare States. Oxford University Press, USA. Mackenbach, J. P., I. Kulhánová, M. Bopp, P. Deboosere, T. A. Eikemo, R. Hoffmann, M. C. Kulik, M. Leinsalu, P. Martikainen, G. Menvielle, E. Regidor, B. Wojtyniak, O. Östergren, O. Lundberg, for the EURO-GBD-SE Consortium 2015. ‘Variations in the Relation between Education and Cause-specific Mortality in 19 European Populations: A Test of the “Fundamental Causes” Theory of Social Inequalities in Health’. Social Science & Medicine 127:51–62.

290  Handbook of health inequalities across the life course Meng, X., and C. D’arcy. 2012. ‘Education and Dementia in the Context of the Cognitive Reserve Hypothesis: A Systematic Review with Meta-Analyses and Qualitative Analyses’. PloS One 7(6):e38268. Mosing, M. A., C. Lundholm, S. Cnattingius, M. Gatz, and N. L. Pedersen. 2018. ‘Associations between Birth Characteristics and Age-Related Cognitive Impairment and Dementia: A Registry-Based Cohort Study’. PLOS Medicine 15(7):e1002609. doi: 10.1371/journal.pmed.1002609. Newman, A. B., J. L. Sanders, J. R. Kizer, R. M. Boudreau, M. C. Odden, A. Zeki Al Hazzouri, and A. M. Arnold. 2016. ‘Trajectories of Function and Biomarkers with Age: The CHS All Stars Study’. International Journal of Epidemiology 45(4):1135–45. Noble, K. G., …, and E. R. Sowell. 2015. ‘Family Income, Parental Education and Brain Structure in Children and Adolescents’. Nature Neuroscience 18(5):773–78. doi: 10.1038/nn.3983. Okbay, A., J. P. Beauchamp, … and D. J. Benjamin. 2016. ‘Genome-Wide Association Study Identifies 74 Loci Associated with Educational Attainment’. Nature 533(7604):539–42. doi: 10.1038/ nature17671. Phelan, J. C., B. G. Link, A. Diez-Roux, I. Kawachi, and B. Levin. 2004. ‘“Fundamental Causes” of Social Inequalities in Mortality: A Test of the Theory’. Journal of Health and Social Behavior 45(3):265–85. Plug, I., R. Hoffmann, B. Artnik … and J. P. Mackenbach. 2012 ‘Socioeconomic Inequalities in Mortality from Conditions Amenable to Medical Interventions: Do They Reflect Inequalities in Access or Quality of Health Care?’ BMC Public Health 12:346. Qiu, C., and L. Fratiglioni. 2015. ‘A Major Role for Cardiovascular Burden in Age-Related Cognitive Decline’. Nature Reviews: Cardiology 12(5):267–77. doi: 10.1038/nrcardio.2014.223. Richards, M., S-N. James, A. Sizer, N. Sharma, M. Rawle, D. H. J. Davis, and D. Kuh. 2019. ‘Identifying the Lifetime Cognitive and Socioeconomic Antecedents of Cognitive State: Seven Decades of Follow-up in a British Birth Cohort Study’. BMJ Open 9(4). doi: 10.1136/bmjopen-2018-024404. Ritchie, S. J., E. M. Tucker-Drob, S. R. Cox, J. Corley, D. Dykiert, P. Redmond, A. Pattie, A. M. Taylor, R. Sibbett, J. M. Starr, and I. J. Deary. 2016. ‘Predictors of Ageing-Related Decline across Multiple Cognitive Functions’. Intelligence 59:115–26. doi: 10.1016/j.intell.2016.08.007. Rönnlund, M., and L-G. Nilsson. 2006. ‘Adult Life-Span Patterns in WAIS-R Block Design Performance: Cross-Sectional versus Longitudinal Age Gradients and Relations to Demographic Factors’. Intelligence 34(1):63–78. doi: 10.1016/j.intell.2005.06.004. Rowe, D. C., K. C. Jacobson, and E. J. C. G. Van den Oord. 1999. ‘Genetic and Environmental Influences on Vocabulary IQ: Parental Education Level as Moderator’. Child Development 70(5):1151–62. Ruby, J. G., K. M. Wright, K. A. Rand … and I. Dorfman. 2018. ‘Estimates of the Heritability of Human Longevity Are Substantially Inflated Due to Assortative Mating’. Genetics 210(3):1109–24. Salthouse, T. A., R. L. Babcock, E. Skovronek, D. Mitchell, and R. Palmon. 1990. ‘Age and Experience Effects in Spatial Visualization.’ Developmental Psychology 26(1):128. Scarr-Salapatek, S. 1971. ‘Race, Social Class, and IQ’. Science 174(4016):1285–95. Seblova, D., R. Berggren, and M. Lövdén. 2020. ‘Education and Age-Related Decline in Cognitive Performance: Systematic Review and Meta-Analysis of Longitudinal Cohort Studies’. Ageing Research Reviews 58:101005. Stern, Y. 2012. ‘Cognitive Reserve in Ageing and Alzheimer’s Disease’. The Lancet Neurology 11(11):1006–12. Stern, Y., E. M. Arenaza‐Urquijo, D. Bartrés‐Faz, S. Belleville, M. Cantilon, G. Chetelat, M. Ewers, N. Franzmeier, G. Kempermann, and W. S. Kremen. 2020. ‘Whitepaper: Defining and Investigating Cognitive Reserve, Brain Reserve, and Brain Maintenance’. Alzheimer’s & Dementia 16(9):1305–11. Stern, Y., C. A. Barnes, C. Grady, R. N. Jones, and N. Raz. 2019. ‘Brain Reserve, Cognitive Reserve, Compensation, and Maintenance: Operationalization, Validity, and Mechanisms of Cognitive Resilience’. Neurobiology of Aging 83:124–29. Then, F. S., T. Luck, M. Luppa, M. Thinschmidt, S. Deckert, K. Nieuwenhuijsen, A. Seidler, and S. G. Riedel-Heller. 2014. ‘Systematic Review of the Effect of the Psychosocial Working Environment on Cognition and Dementia’. Occupational and Environmental Medicine 71(5):358–65. Trzaskowski, M., N. Harlaar, R. Arden, E. Krapohl, K. Rimfeld, A. McMillan, P. S. Dale, and R. Plomin. 2014. ‘Genetic Influence on Family Socioeconomic Status and Children’s Intelligence’. Intelligence 42:83–88. doi: 10.1016/j.intell.2013.11.002.

Childhood conditions and health later in life  291 Tucker-Drob, E. M. 2019. ‘Cognitive Aging and Dementia: A Life Span Perspective’. Annual Review of Developmental Psychology 1:177–96. doi: 10.1146/annurev-devpsych-121318-085204. Tucker-Drob, E. M., and T. C. Bates. 2016. ‘Large Cross-National Differences in Gene × Socioeconomic Status Interaction on Intelligence’. Psychological Science 27(2):138–49. doi: 10.1177/0956797615612727. Tucker-Drob, E. M., A. M. Brandmaier, and U. Lindenberger. 2019. ‘Coupled Cognitive Changes in Adulthood: A Meta-Analysis.’ Psychological Bulletin 145(3):273. van Raalte, A. A., I. Sasson, and P. Martikainen. 2018. ‘The Case for Monitoring Life-Span Inequality’. Science 362(6418):1002–4. Vinkhuyzen, A. A. E., S. Van Der Sluis, E. J. C. De Geus, D. I. Boomsma, and D. Posthuma. 2010. ‘Genetic Influences on “Environmental” Factors’. Genes, Brain and Behavior 9(3):276–87. doi: 10.1111/j.1601-183X.2009.00554.x. Wang, H.-X., S. W. S MacDonald, S. Dekhtyar, and L. Fratiglioni. 2017. ‘Association of Lifelong Exposure to Cognitive Reserve-Enhancing Factors with Dementia Risk: A Community-Based Cohort Study’. PLoS Medicine 14(3):e1002251. Whalley, L. J., F. D. Dick, and G. McNeill. 2006. ‘A Life-Course Approach to the Aetiology of Late-Onset Dementias’. The Lancet Neurology 5(1):87–96. doi: 10.1016/S1474-4422(05)70286-6. Whitson, H. E., H. J. Cohen, K. E. Schmader, M. C. Morey, G. Kuchel, and C. S. Colon-Emeric. 2018. ‘Physical Resilience: Not Simply the Opposite of Frailty’. Journal of the American Geriatrics Society 66(8):1459–61. doi: 10.1111/jgs.15233. Whitson, H, E., W. Duan-Porter, K. E. Schmader, M. C. Morey, H. J. Cohen, and C. S. Colón-Emeric. 2016. ‘Physical Resilience in Older Adults: Systematic Review and Development of an Emerging Construct’. Journals of Gerontology Series A: Biomedical Sciences and Medical Sciences 71(4):489–95. Wu, C., Y.-X. Li, M. M. Marron, M. C. Odden, A. B. Newman, and J. L. Sanders. 2020. ‘Quantifying and Classifying Physical Resilience among Older Adults: The Health, Aging, and Body Composition Study’. The Journals of Gerontology: Series A 75(10):1960–66. Yu, Q., A. M. Daugherty, D. M. Anderson, M. Nishimura, D. Brush, A. Hardwick, W. Lacey, S. Raz, and N. Ofen. 2018. ‘Socioeconomic Status and Hippocampal Volume in Children and Young Adults’. Developmental Science 21(3):e12561. doi: 10.1111/desc.12561.

19. The influence of early health on educational and socioeconomic outcomes Marco Cozzani and Juho Härkönen

INTRODUCTION There is a large literature on socioeconomic disparities in health, much of it discussed in this volume. Many explanations for health disparities assume that the material and working conditions as well as social and psychological stressors that come with different socioeconomic positions are responsible for these disparities. In other words, socioeconomic position affects health (Link and Phelan, 1995; Lynch et al., 2000; Mackenbach, 2012; Marmot and Wilkinson, 2001; Wilkinson et al., 2009). An alternative explanation suggests that the causality flows in the opposite direction, so that being healthy enables one to reach high socioeconomic positions, whereas poor health restricts these opportunities (e.g., Lynch and von Hippel 2016; West 1991). While some of the literature on health selection refers to contemporaneous associations between health and socioeconomic positions such as unemployment or earnings, a literature that grew in the past two decades has documented strong associations between early health – from the prenatal stage to adolescence – on socioeconomic attainment in later life. Early health has thus emerged as an important predictor of attainment in adulthood, and because early health itself is socially stratified, it has also been regarded as a potential pathway in the intergenerational reproduction of inequality (Case et al., 2001; Palloni, 2006). The purpose of this chapter is to pull together and discuss research on the effects of early health on socioeconomic attainment. Our focus is on research from developed countries, which has grown substantially in the last two decades. This choice omits research from developing countries from our review, as the latter differ from developed ones in disease determinants and socioeconomic environments. As socioeconomic outcomes, we consider educational, occupational and income attainment as well as related outcomes, such as cognitive performance, which are sometimes used to explain the effects of health on socioeconomic outcomes. We start by reviewing research on the effects of early health during the prenatal period and infancy as well as later childhood and adolescence, and continue to discuss a life-course framework of how early health can affect socioeconomic attainment, which builds on the well-known life-course models by Kuh and Ben-Shlomo (Ben-Shlomo and Kuh, 2002; Kuh et al., 2003) and complements these epidemiological models with social scientific scholarship. We conclude by discussing some underrepresented and emerging questions in the literature.

EARLY HEALTH AND SOCIOECONOMIC ATTAINMENT: AN OVERVIEW OF FINDINGS Just before the turn of the new millennium, James P. Smith (1999) pointed to “new theories” stressing the importance of health in childhood and “even intrauterine factors” for wealth in 292

The influence of early health on educational and socioeconomic outcomes  293 adulthood. Since then, research in economics, sociology, demography and epidemiology has produced a large body of evidence in support of the important role of early health in socioeconomic attainment. In this section, we provide an overview of this research, proceeding along the life course from research on perinatal health (which includes in utero (intrauterine) health and the very early childhood) to health during later childhood and adolescence. Perinatal and Childhood Health Three decades or so ago, consideration of intrauterine effects on socioeconomic attainment was novel and epidemiological research had recently been introduced to the fetal origin or programming hypothesis by David Barker, who argued that the nutritional conditions in the intrauterine environment program the fetus’s future metabolic characteristics, which in turn may lead to the onset of diseases in adulthood (Barker, 1990). Support for this hypothesis was found, among other sources, from the Dutch hunger winter of 1944, where survivors exposed to the hunger winter in utero had elevated obesity but also higher incidence of schizophrenia later in life (Hoek et al., 1998; Susser and Stein, 1994). More broadly, the fetal origins hypothesis postulates that many diseases experienced in adulthood have their origin in the womb or to put it in Barker’s own words, “the womb may be more important than home” (Barker, 1990, p. 1111). Although first formulated to identify precedents of common chronic diseases, mounting evidence has supported the idea that insults to fetal health have far-reaching consequences on many other domains of human life, so that negative consequences of the prenatal period are “surprisingly general phenomena” across human populations, affecting health, cognitive development as well as educational achievement and labor market outcomes (Almond et al., 2018). Generally, this line of research falls within two groups: studies using direct measures of perinatal health, such as birth weight, and those taking a design-based approach, which use natural experiments to identify the causal effect of adverse exposures during pregnancy. Much research has focused on the long-term effects of being born with low birth weight (LBW – being born smaller than 2,500 grams) (Boardman et al., 2002; Conley and Bennett, 2000, 2001; Conley et al., 2003; Currie and Hyson, 1999). This focus reflects Barker and associates’ attention to the effects of fetal growth, and LBW is used as a proxy for the intrauterine environment, and, in general, as a measure of developmental potential (Torche and Conley, 2016). Currie and Hyson (1999) were among the first to document persistent negative effects of LBW on a large set of socioeconomic outcomes such as school performance, the probability of having a full-time job and hourly wages. Subsequent research supported these findings and reported that LBW children were less likely to graduate from high school on time (Conley and Bennett, 2000), and more likely to earn less (Black et al., 2007) and achieve lower social class positions as adults (Palloni, 2006). Many of these studies used sibling and twin fixed-effects models to purge the estimates from unobserved environmental and in the case of twin studies, genetic, confounding by factors shared by the siblings (cf. Torche and Conley 2016). These findings were also integrated within the status attainment model to investigate whether health was a mediator of the relationship between parental socioeconomic background and children’s socioeconomic success (Conley and Bennett, 2000; Haas, 2006; Haas et al., 2011; Palloni, 2006; Palloni et al., 2009). Overall, this research has found a link between parental socioeconomic background and children’s achievement passing through infant health. To put it in Conley and Bennett’s own words there is an “intergenerational loop of social inequality and

294  Handbook of health inequalities across the life course low birth weight” (Conley & Bennett, 2000, p. 465) in which socioeconomic status influences health at birth, which in turn affects future socioeconomic attainment. In his early study, Palloni (2006) estimated that roughly 10% of the intergenerational persistence of social class could be attributed to class differences in low birth weight, a contribution comparable to the role of class differences in educational choices. Although LBW has been, by far, the most used marker of prenatal health, some studies have moved beyond the use of LBW as a compound measure of health in utero. First, the effects of birth weight may not be confined to a threshold effect at 2,500 grams (Black et al., 2007). Second, birth weight is a function of gestational age and intrauterine growth. These have related yet independent etiologies, with prematurity being more important than intrauterine growth in determining LBW at the population level in rich countries (Behrman and Butler, 2007). Yet most studies continue to refer to intrauterine growth when discussing the effects of LBW, although some studies have separately analyzed the effects of gestational age and of intrauterine growth (Härkönen et al., 2012; Oreopoulos et al., 2008). Third, some studies have broadened the scope of measures to other common threats to prenatal health, such as the mother’s cigarette smoking, which has been found to be associated with poorer educational and labor market outcomes (Härkönen et al., 2012; Jackson, 2010, 2015a). Another group of studies has tried to isolate random shocks to prenatal health. This scholarship has studied random variation by using twins models to isolate presumably random variation in intrauterine growth (Black et al., 2007; Figlio et al., 2014; Torche and Echevarria, 2011), as well as various exposure to health insults such as exceptional events that generate prenatal maternal stress (Currie and Rossin-Slater, 2013; Persson and Rossin-Slater, 2018; Torche, 2018), pandemic flu (Almond, 2006), pollution (Currie and Schwandt, 2016) and radioactive fallout (Almond et al., 2009; Kelly, 2011). All these studies point to a long-lasting negative effect of health insults during the prenatal period, reporting effects on impaired cognitive performance at different ages (Black et al., 2007; Kelly, 2011; Torche, 2018; Torche and Echevarria, 2011), weaker school performance (Almond et al., 2009; Figlio et al., 2014), and poorer labor market outcomes (Almond, 2006; Black et al., 2007). The effects of poor early health are not restricted to the prenatal period. There is a surprising shortage of studies of the effects of specific health problems in the very first years, which possibly has to do with a shortage of such measures in commonly used data. Yet studies have identified the importance of early life health environments on socioeconomic outcomes by studying different interventions. For example, access to free infant health care (Bütikofer et al., 2019), the US food stamp program (Hoynes et al., 2016), nutritional programs (Jackson, 2015b) or breastfeeding interventions (Kramer et al., 2008) have been all shown to boost cognitive and educational outcomes. Other studies have used height as a proxy of the early nutritional and disease environment. Case and Paxson (2010), for instance, concluded that height is a robust predictor of socioeconomic outcomes in adolescence and adulthood. These effects were partly, but not fully, attributable to prenatal health, thus proposing an additional role for early childhood years. Together, these studies point to the importance of early nutrition – and additionally, to diseases such as diarrhea (cf. Bütikofer et al., 2019) that directly impact on nutrition – in the first years as a predictor of educational attainment and adulthood income. Other studies have highlighted the role of environmental toxins. The adverse consequences of lead exposure on cognitive development and scholastic performance have received wide attention in the literature (Muller et al., 2018). Likewise, exposure to air pollutants (i.e. carbon monoxide, nitrogen dioxide) can have adverse effects on school attendance and learning (Currie et

The influence of early health on educational and socioeconomic outcomes  295 al., 2009). Pollution exposure leads to adverse children health outcomes (i.e. asthma) (Han et al., 2021), which may influence school attendance. Furthermore, the substances contained in pollution can hamper cognitive development as they act as neuro-developmental toxicants that affect the gene expression in the developing brain (Grandjean and Landrigan, 2014; Slama et al., 2008). Adolescent Health: “the Missing Middle” While much research has focused on the role of prenatal and early childhood health in shaping socioeconomic outcomes, less scholarship has paid attention to later childhood and the adolescent years, which Currie (2020) referred to as the “missing middle” in the literature. Yet late childhood and adolescence is in general a formative life course stage when many decisions of future educational paths are made, and which is also the time when several health conditions – such as mental health disorders – often develop (Bernard et al., 2022; Klocke and Stadtmüller, 2022). Furthermore, to the extent that social background disparities in health widen over the childhood and teenage years (Currie and Stabile, 2003), health problems beyond pregnancy and early childhood can be more important in the intergenerational reproduction of inequality (Mikkonen et al. 2020). Studies that have analyzed the effects of health conditions in later childhood and adolescence on socioeconomic attainment have implicated the importance of poor health of this life course stage on long-term attainment (Case et al., 2005; Case and Paxson, 2010; Currie and Stabile, 2009; Jackson, 2009, 2010, 2015b; Mikkonen et al., 2020; Smith, 2009). These studies have used different measures of health as well as studying different socioeconomic outcomes. One group of studies has used self-reported or compound measures of health. Adolescents who self-reported better health attain more education (Haas, 2006; Haas and Fosse, 2008; Lynch and von Hippel, 2016; Smith, 2009) and respondents who self-reported being in better health during childhood also had stronger labor market attachment, higher occupational status, earnings, household incomes and wealth, and stronger income growth trajectories in adulthood (Haas 2006; Smith 2009), often leading to larger income disparities by early health status over the career (Haas et al., 2011). With the exception of educational attainment, effects of childhood health persisted in a sibling comparison design (Smith 2009). Currie and Stabile (2009) reported that poor physical health in childhood has negative effects on grades and social benefit recipiency, primarily through poor health in young adulthood. Case, Fertig and Paxson (2005) showed, using British birth cohort data, that chronic health problems at age 7 and 16 were robust predictors of educational and occupational attainment over the adulthood years, but perinatal conditions did not have direct effects once the former had been adjusted for. Jackson (2010) reported that children who were in persistently poor health during childhood and adolescence had particularly disadvantaged occupational outcomes, whereas children with transitory health problems were likely to catch up. In another study (Jackson, 2015b), she found that the relationship between poor health and academic achievement emerged early and could compound over time as those with poor health at birth often experienced additional poor health when growing up. Next to global measures of health, scholars have analyzed the importance of specific health conditions or groups of conditions. Of somatic conditions, severe conditions such as metabolic diseases and childhood cancer have been identified as important for educational attainment (Champaloux and Young, 2015; Ghaderi et al., 2016; Persson et al., 2013). Frederiksen et al.

296  Handbook of health inequalities across the life course (2019) furthermore conclude that the negative effects of childhood cancer are mainly confined to survivors of central nervous system (CNS) tumors. Other results can be used to compare the effects of groups of health conditions, particularly separating between physical and mental health problems in childhood and adolescence (Case et al., 2005; Currie and Stabile, 2009; Mikkonen et al., 2018, 2020). Although both groups of health problems can have adverse socioeconomic effects, these studies highlight the importance of mental health on socioeconomic attainment (Fried et al., 2016; Goodman et al., 2011; Goulding et al., 2010; McLeod and Fettes, 2007), and poor mental health in childhood and adolescence typically has stronger effects on education and labor market outcomes than physical health conditions (cf. Currie, 2020). Goodman, Joyce and Smith (2011) reported that psychological problems by age 16 were associated with a 28% reduction in family incomes in the 50s, whereas low birth weight and physical health problems in particular had clearly weaker effects. Mikkonen et al. (2018) concluded that 11% of the failure to complete upper secondary school in Finland could be attributed to mental health problems at ages 10–16, compared to 5% that was attributable to somatic health conditions and 30% to all health conditions. Other studies have decomposed mental disorders into more specific conditions, typically differentiating between internalizing (depression and anxiety disorders) and externalizing (attention and conduct disorders) disorders. Focusing on the role of internalizing problems in educational attainment, McLeod and Fettes (2007) concluded that young people experiencing internalizing disorders were less likely to enroll in college, but no less likely to complete high school, whereas Fletcher (2010) found that adolescent depression had a negative effect on educational attainment, primarily through increasing the probability of early school drop-out, and this association was robust in a sibling comparison design. Yet internalizing problems often co-occur together with externalizing problems and other comorbidities. Studies on the effects of adolescent mental health on schooling outcomes have concluded that the negative effect of internalizing disorders disappears once one accounts for comorbidity and externalizing problems, in particular (Evensen et al., 2016; Miech et al., 1999). All in all, a large body of research has in the last two decades convincingly documented the potential of health at different stages of childhood and adolescence – starting from the prenatal period – for affecting socioeconomic attainment. In the following section, we continue discussing a life course framework to theorize these effects and their interrelations.

A LIFE-COURSE FRAMEWORK OF EARLY HEALTH AND SOCIOECONOMIC ATTAINMENT We discuss a life-course framework that starts from the life-course epidemiology approach promoted, in particular, by Yaov Ben-Shlomo and Diana Kuh and colleagues (Ben-Shlomo and Kuh, 2002; Ben-Shlomo et al., 2014, 2016; Kuh et al., 2003) to understand how health develops over the life course and how exposures at different life-course stages facilitate this development. We then discuss how insights from social science can be used to complement this approach in an interdisciplinary framework combining research from the medical and the social sciences.

The influence of early health on educational and socioeconomic outcomes  297 Life-Course Epidemiological Models for Understanding Early Health and Socioeconomic Attainment Central questions in life-course epidemiology concern three factors: (1) how duration of an exposure is related to a health outcome; (2) whether the timing when an exposure is experienced matters; (3) and how exposures are related to one another as part of causal pathways (Ben-Shlomo et al., 2014, p. 1523). These questions are closely related to early formulations of the life-course epidemiological approach, which distinguished between accumulation and chains of risk models on the one hand, and critical and sensitive periods models on the other (Ben-Shlomo and Kuh, 2002; Kuh et al., 2003), even if the authors themselves later questioned such a strict distinction (Ben-Shlomo et al., 2014, 2016). References to critical and sensitive periods models have been particularly popular in research on early health effects, and prenatal health effects in particular, and fit contemporaneous emphasis in the economics of human capital formation on the importance of early childhood conditions (Heckman, 2007). According to critical and sensitive periods models, the effects of exposures are moderated by their timing in the life course, with effects being particularly pronounced within some narrow time windows in the life course. Whether the time window should be called critical or sensitive depends on whether effects are limited only to exposures within the time window (critical periods) or whether effects are simply stronger within the time window than in other periods (sensitive periods). A good example of the use of the idea of critical periods in the literature on early health and socioeconomic attainment is the study on exposure to the radioactive fallout from the Chernobyl disaster, where – based on biological models of brain development – in utero exposure was identified during weeks 8–25 post-conception (Almond et al., 2009). In another example, children of mothers who smoked during the whole pregnancy attained less education than those of mothers who quit during the first trimester, in line with critical period models of nicotine exposure and brain development (Härkönen et al., 2012). The idea of sensitive periods has been used, for instance, in understanding the adverse effects of lead exposure on cognitive skills, test scores, health and criminality, where exposure in early childhood is considered more detrimental than in later childhood (Muller et al., 2018). Critical and sensitive periods models have been primarily used to conceptualize the effects of health insults through biological pathways, as in the above examples which related toxin exposure to critical stages in brain development. The biological pathways are typically theorized to operate as latent effects, where early health insults lead to permanent changes in the developing fetus, which affect distal outcomes such education and labor market success through more proximal mechanisms such as academic aptitude and personality traits. These models resemble the stunting variant of health selection models in social epidemiology, where poor health limits human capital accumulation with life-long consequences (Haas, 2006; West, 1991). Even though critical and sensitive periods models are typically used in reference to biological pathways, owing to the origin of these models in epidemiology, the main idea of effect moderation depending on the timing of the exposure is central in non-epidemiological life-course approaches as well (Elder, 1994) and has been applied to non-biological pathways. Researchers have analyzed whether events such as parental divorce happening close to important educational transition points have particularly strong effects (Sigle-Rushton et al., 2014). Similar approaches can be considered for health adversities, such as an onset of mental health problems close to such an educational transition point.

298  Handbook of health inequalities across the life course Risk accumulation models have been less popular theoretical frameworks for conceptualizing the effects of early health on socioeconomic outcomes, and this is also reflected in the statistical models that researchers have used. According to life course epidemiological accumulation models (Ben-Shlomo and Kuh, 2002; Ben-Shlomo et al., 2014, 2016; Kuh et al., 2003), illnesses, injuries, adverse environmental conditions and behaviors that gradually accumulate over the life course expose the body to wear and tear that produces ill health. Some studies have analyzed the cumulative effects of poor health. An example is the above-cited study by Jackson (2010), which found stronger effects of persistently poor childhood health than of transitory health problems. This example highlights a feature of accumulation models, where rather than the timing of poor health, its duration is considered to shape its effect in a dose-response fashion. The chain-of-risks model (Ben-Shlomo and Kuh, 2002; Kuh et al., 2003) theorizes that the effects rather develop through a sequence of disadvantages. For example, a child born exposed to smoke while in utero, is more likely to experience health issues at birth (D’Onofrio et al., 2010), which may lead to a frail health over the childhood (Persson and Rossin-Slater, 2018), poorer cognitive development (Torche, 2018), and thus a lower chance to achieve education (Conley and Bennett, 2001). A first disadvantage generated a chain of events which led to a health problem that later shaped the child’s socioeconomic opportunities. In line with the chain-of-risks model, several studies have considered the role of poor health in later stages of childhood and adolescence or in adulthood as mediating the effects of poor health at earlier life course stages (Case et al., 2005; Haas et al., 2011; Jackson, 2015b; Smith, 2009). The persistence of poor health into adulthood reduces labor market participation and, in line with the drift selection model of socioeconomic disparities in health, leads to lower occupational and income attainment and thus a “health disparity in socioeconomic attainment” (Lynch and von Hippel, 2016). Persistently poor health in adulthood is also one of the pathways that can lead to growing occupational and income disparities over the life course, as those in poorer health miss out on career development in relation to their healthier peers (Case and Paxson, 2011; Haas et al., 2011; Smith, 2009). Although typically discussed separately, especially in the early days of life course epidemiology, epidemiological models on critical and sensitive periods on the one hand, and risk accumulation and chain-of-risks on the other, are not mutually exclusive. Critical and sensitive periods can be seen as subsets of accumulation models, in which health insults add up with different intensities (Ben-Shlomo et al., 2016). And, as discussed above, the chain-of-risks model can be applied to understand how a health insult at a critical or sensitive period (such as smoking in pregnancy) of child development transforms into lower socioeconomic attainment through a pathway of poor health in later childhood, adolescence and adulthood. Social and Institutional Environments and Early Health Effects The literature on early health effects is necessarily interdisciplinary, and the incorporation of biological and epidemiological models into social science scholarship on the predictors of socioeconomic success and social inequality has been a welcome contribution of this literature. These models have been complemented by social scientific approaches that draw attention to other pathways through which early health may operate, as well as to the moderating role of social environments, thus emphasizing the interactions between biological and social mechanisms.

The influence of early health on educational and socioeconomic outcomes  299 A large share of the research on early health effects has focused on educational outcomes, measured as attained grades and test scores, or achieved educational levels. Following from the emphasis on biological pathways, many studies have highlighted the potential role of shocks to early development in cognitive development, learning skills and academic achievement. Following Boudon (1974), Mikkonen and colleagues (2021) called these primary effects of health, which affect educational attainment through educational performance, whether measured by the researcher as school grades or test scores. Implicit emphasis on primary effects is common in studies on the effects of prenatal health. Low birth weight, for instance, can affect brain development through reduced brain growth or brain damage (Abernethy et al., 2002; Bharadwaj et al., 2018), and the above examples of toxin exposure in the womb likewise emphasize effects on learning either directly or as an important pathway to reduced educational attainment. Next to the effects of prenatal insults, health status in later childhood and adolescence may likewise affect learning by influencing learning motivation, attention and school presence (Basch, 2011; Mikkonen et al., 2021). According to Boudon (1974), secondary effects refer to differences in educational choices that students and parents – and occasionally, teachers – from different social backgrounds make at similar levels of the students’ school performance. Students coming from higher socioeconomic backgrounds typically choose academically more ambitious tracks, whereas students from lower backgrounds are more likely to pursue vocational or other academically less ambitious tracks, where the probability of successful completion is higher (Breen and Goldthorpe, 1997). Health may have similar effects on educational choices, if students in poorer health perceive themselves or are perceived as frailer and less likely to succeed in academically demanding tracks (Mikkonen et al. 2021), or if they discount their future differently and focus on shorter-term human capital investments (Becker and Mulligan, 1997; Pampel et al., 2010). Mikkonen and colleagues (2021) found that secondary effects of adolescent health are confined to the effects of mental health, where students with mental health problems are less likely to pursue academic tracks leading to university. This result resonates with findings of the role of common mental health disorders and conduct problems as “non-cognitive skills” that affect educational attainment and can mediate the effects of other health conditions (Currie and Stabile, 2009). Effects of childhood and adolescent health can be modified by responses from the social environment. Economists have theorized that parents do not treat their children alike and can respond to their children’s strengths and weaknesses either by reinforcing them or compensating for weaknesses by directing additional attention and investments to the children with weaker starting points (Almond and Mazumder, 2013; Aquino et al., 2022; Becker and Tomes, 1976; Behrman et al., 1982; Conley, 2008). Social stratification in these behaviors can either amplify or reduce the inequality consequences of disparities in children’s health (cf. Bernardi, 2014). Some studies have analyzed whether the consequences of children’s poor health are socially stratified, with mixed results. On the one hand, Almond and colleagues (2009) found that the long-term consequences of poor perinatal health induced by the exposure to the Chernobyl fallout in Sweden were substantial only among children in less advantaged households; Torche (2018) found that the negative consequences of in utero exposure to the Tarapaca earthquake on cognition were concentrated among children from a low socioeconomic background; and Bütikofer, Løken and Salvanes (2019) found that the effects of extended access to free infant health provision on educational outcomes were stronger for children from low socioeconomic backgrounds. These results suggest that parents of higher socioeconomic status may be able

300  Handbook of health inequalities across the life course to compensate for these adverse starting points, possibly through active parenting strategies or by being able to provide a more stable and stimulating environment. More direct evidence on the role of parents’ active strategies was provided by Hsin (2012), who found that parents with low education allocated more time and more “quality time” to their children who were born heavier, whereas highly educated parents favored children with lower birth weights. Despite these results, other studies have not found evidence for socioeconomically heterogeneous effects, for example of poor perinatal health on cognition (Grätz and Torche, 2016; Black et al., 2007). It is not clear why these results differ, as they do not seem to be driven by either differences in the research design or by the context analyzed (Torche and Conley, 2016). Parents are not the only ones in the child’s close social environment who can shape the effects of their health on educational outcomes. Teachers may likewise either reinforce or compensate for a disadvantage stemming from the child’s poor health. As an example of the former, Evensen et al. (2016) reported that a part of the reason why young people with externalizing problems have lower school performance is that teachers penalize them when setting grades. Together, these results show how the social environment can moderate the consequences of health, and expanding on this line of research can be a fruitful future direction in the field. Furthermore, theories of compensation or reinforcement suggest that parents (and other social actors) are able to appropriately identify their children’s health problems or identify the (possibly latent) sources of their children’s cognitive or behavioral problems from other sources. A promising avenue can be to understand which health conditions trigger compensating or reinforcing behaviors from different actors in the social environment. As the above examples of low birth weight and externalizing problems suggest, the reactions to different health issues may differ. Beyond the immediate social environment, early health effects can be shaped by the broader macrostructural and institutional environment. To our knowledge, there is only limited previous research on the variation in the effects of early health across countries or other relevant units, and more research could be done on whether institutional differences and reforms moderate the effects of early health on later life outcomes (Hoffmann, 2011). Several institutions can be relevant. First, health care and social policy systems may vary in how well they can ameliorate the consequences of poor health at different childhood stages, either through medical treatments or non-medical measures. Second, educational systems can shape the effects of early health. Stratification scholars have paid much attention to the importance of educational tracking, and generally found more educational inequality in strongly tracked educational systems where tracking occurs at a younger age (Van de Werfhorst and Mijs, 2010). Such systems can reinforce the effects of some forms of early health, possibly through lowering the pupil’s, her parents’ and teachers’ evaluations of success in more demanding educational tracks. Alternatively, by anticipating the age when crucial educational decisions are made, they can reduce the effects of health conditions which are likely to emerge in adolescence.

DISCUSSION The effect of health on educational and other socioeconomic outcomes has been an active field of recent research. Whereas two decades ago, early health as a source of socioeconomic attainment and inequality was seen as a promising and exciting avenue for research (Palloni,

The influence of early health on educational and socioeconomic outcomes  301 2006; Smith, 1999), a considerable amount of evidence has accumulated to support the important role of early health in shaping children’s long-term outcomes, as this overview chapter has shown. As final words, we discuss what we consider some fruitful avenues for future research on early health effects on socioeconomic attainment. First, future research will benefit from continuing to analyze the linkages of health problems from the prenatal period through adolescence and beyond to understand how early life health impacts on later life socioeconomic outcomes through health in later childhood, adolescence, and adulthood – akin to the chain-of-risks or the accumulation models. Such research has been done, as discussed above, but future research is welcome that considers how important predictors, such as poor prenatal health and adolescent mental health problems, interact in restricting educational attainment and future socioeconomic success. These questions are not merely of theoretical interest but can be of policy relevance if they are able to pinpoint the most important health conditions that policies may effectively target. Such research can benefit from models for causal mediation analysis (VanderWeele, 2015). Similarly, although the critical and sensitive periods models have been widely referred to in the literature, few studies have empirically attempted to identify whether these can be found in the data, especially for common health problems that are not restricted to a specific stage in the early life course. Again, scholars can benefit from recently developed models for distinguishing critical and sensitive periods from accumulation models (Kröger et al., 2016; Mishra et al., 2009; Potente et al., 2021). Second, we consider research that combines theoretical insights from epidemiology, other medical sciences, child development and the social sciences as a promising approach to improve our understanding of how early health impacts on socioeconomic outcomes. Early health does not operate in a vacuum, and, as we have discussed above, its impacts can depend on the social environment in which the child grows up. Such research can consider the close social environment of the family at one end, or the macro-social structural and institutional environment at the other. Research on whether the latter moderates the effects of early health has been particularly absent, probably due to a lack of suitable (cross-national) data. Research on how early health has different impacts depending on one’s socioeconomic environment is also relevant for understanding how early health can affect the intergenerational transmission of inequality (Bernardi, 2014). Above, we already mentioned research that attempts to identify which health problems are likely to draw attention from parents, teachers and other actors as an interesting line of future research. Third, future research can pay more attention to the effects of specific common and clinically-relevant health conditions. The variables included in commonly used data have partly limited researchers’ attention to broad measures of health, such as self-rated health and low birth weight. These studies, as well as studies exploiting natural experiments that allow identifying causal effects of prenatal health, for example, have provided important information on the effects of early health on later outcomes. Moving forward, researchers can attempt to identify the effects of more specific health conditions, which may also point to conditions that require particular policy focus. Research that has distinguished between the effects of internalizing and externalizing mental health disorders on education is an example of such research (Evensen et al., 2016; Miech et al., 1999). Fourth, and relatedly, future studies can pay more attention to the population-level relevance of different health conditions at different early life course stages. Most research done to do date has reported on the individual-level effects of early health. Yet even strong individual-level

302  Handbook of health inequalities across the life course effects need not translate into population-level importance, if the health conditions or shocks considered are sufficiently rare. On the other hand, even relatively weak effects can have large population-level impacts – for example, on the fraction of young people who drop out of school – if they are experienced by a large enough share of the population. Thus far, only a handful of studies have provided such estimates. A rare example includes Mikkonen and colleagues (2018) who estimated that up to 30% of school drop-out at age 17 in Finland was attributable to adolescent health problems. In a related example, Palloni (2006), Härkönen and colleagues (2012), and Bütikofer, Løken and Salvanes (2019) estimated that roughly one-tenth of the intergenerational persistence in socioeconomic status could be explained by low birth weight, prenatal smoking, and access to early health care, respectively. Fifth, this chapter is being written two years into the COVID-19 pandemic. There is no doubt that we will continue to see important research on how the health consequences of the pandemic – both for those who were infected as well as affected (Settersten et al., 2020) – influence children and young people’s long-term outcomes.

ACKNOWLEDGEMENT The authors acknowledge funding from the Academy of Finland (decision number 324613).

REFERENCES Abernethy, L. J., Palaniappan, M. & Cooke, R. W. 2002. Quantitative magnetic resonance imaging of the brain in survivors of very low birth weight. Archives of Disease in Childhood, 87, 279–283. Almond, D. 2006. Is the 1918 influenza pandemic over? Long-term effects of in utero influenza exposure in the post-1940 US population. Journal of Political Economy, 114, 672–712. Almond, D., Currie, J. & Duque, V. 2018. Childhood circumstances and adult outcomes: Act II. Journal of Economic Literature, 56, 1360–1446. Almond, D., Edlund, L. & Palme, M. 2009. Chernobyl’s subclinical legacy: Prenatal exposure to radioactive fallout and school outcomes in Sweden. Quarterly Journal of Economics, 124, 1729–1772. Almond, D. & Mazumder, B. 2013. Fetal origins and parental responses. In: Arrow, K. J. & Bresnahan, T. F. (eds.) Annual Review of Economics, Vol. 5. Palo Alto: Annual Reviews. Aquino, T., Brand, J. E. & Torche, F. 2022. Unequal effects of disruptive events. Sociology Compass, 16, e12972. Barker, D. J. P. 1990. The fetal and infant origins of adult disease. BMJ: British Medical Journal, 301, 1111. Basch, C. E. 2011. Healthier students are better learners: A missing link in school reforms to close the achievement gap. Journal of School Health, 81, 593–598. Becker, G. S. & Mulligan, C. B. 1997. The endogenous determination of time preference. Quarterly Journal of Economics, 112, 729–758. Becker, G. S. & Tomes, N. 1976. Child endowments and the quantity and quality of children. Journal of Political Economy, 84, S143–S162. Behrman, J. R., Pollak, R. A. & Taubman, P. 1982. Parental preferences and provision for progeny. Journal of Political Economy, 90, 52–73. Behrman, R. E. & Butler, A. S. 2007. Preterm Birth: Causes, Consequences, and Prevention. National Academies Press. Ben-Shlomo, Y., Cooper, R. & Kuh, D. 2016. The last two decades of life course epidemiology, and its relevance for research on ageing. International Journal of Epidemiology, 45, 973–988.

The influence of early health on educational and socioeconomic outcomes  303 Ben-Shlomo, Y. & Kuh, D. 2002. A life course approach to chronic disease epidemiology: Conceptual models, empirical challenges and interdisciplinary perspectives. International Journal of Epidemiology, 31, 285–293. Ben-Shlomo, Y., Mishra, G. & Kuh, D. 2014. Life course epidemiology. In: Ahrens, W. & Pigeot, I. (eds.) Handbook of Epidemiology. Springer, New York. Bernard, M., Winter, K. & Moor, I. 2022. Health inequalities in adolescence from a life course perspective. In: Hoffmann, R. (ed.) Handbook of Health Inequalities across the Life Course. Edward Elgar Publishing, Cheltenham, UK and Northampton, MA, USA. Bernardi, F. 2014. Compensatory advantage as a mechanism of educational inequality: A regression discontinuity based on month of birth. Sociology of Education, 87, 74–88. Bharadwaj, P., Eberhard, J. P. & Neilson, C. A. 2018. Health at birth, parental investments, and academic outcomes. Journal of Labor Economics, 36, 349–394. Black, S. E., Devereux, P. J. & Salvanes, K. G. 2007. From the cradle to the labor market? The effect of birth weight on adult outcomes. Quarterly Journal of Economics, 122, 409–439. Boardman, J. D., Powers, D. A., Padilla, Y. C. & Hummer, R. A. 2002. Low birth weight, social factors, and developmental outcomes among children in the United States. Demography, 39, 353–368. Boudon, R. 1974. Education, Opportunity, and Social Inequality: Changing Prospects in Western Society. John Wiley & Sons Inc. Breen, R. & Goldthorpe, J. H. 1997. Explaining educational differentials towards a formal rational action theory. Rationality and Society, 9, 275–305. Bütikofer, A., Løken, K. V. & Salvanes, K. G. 2019. Infant health care and long-term outcomes. Review of Economics and Statistics, 101, 341–354. Case, A., Fertig, A. & Paxson, C. 2005. The lasting impact of childhood health and circumstance. Journal of Health Economics, 24, 365–389. Case, A., Lubotsky, D. & Paxson, C. 2001. Economic Status and Health in Childhood: The Origins of the Gradient. National Bureau of Economic Research. Case, A. & Paxson, C. 2010. Causes and consequences of early-life health. Demography, 47, S65–S85. Case, A. & Paxson, C. (2011). The long reach of childhood health and circumstance: Evidence from the Whitehall II study. Economic Journal, 121(554), F183–F204. Champaloux, S. W. & Young, D. R. 2015. Childhood chronic health conditions and educational attainment: A social ecological approach. J Adolesc Health, 56, 98–105. Conley, D. 2008. Bringing sibling differences in: Enlarging our understanding of the transmission of advantage in families. Social Class: How Does It Work, 179–200. Conley, D. & Bennett, N. G. 2000. Is biology destiny? Birth weight and life chances. American Sociological Review, 65, 458–467. Conley, D. & Bennett, N. G. 2001. Birth weight and income: Interactions across generations. Journal of Health and Social Behavior, 42(4), 450–465. Conley, D., Strully, K. W. & Bennett, N. G. 2003. The Starting Gate: Birth Weight and Life Chances, University of California Press. Currie, J. 2020. Child health as human capital. Health Economics, 29, 452–463. Currie, J., Hanushek, E. A., Kahn, E. M., Neidell, M. & Rivkin, S. G. 2009. Does pollution increase school absences? Review of Economics and Statistics, 91, 682–694. Currie, J. & Hyson, R. 1999. Is the impact of health shocks cushioned by socioeconomic status? The case of low birthweight. American Economic Review, 89, 245–250. Currie, J. & Rossin-Slater, M. 2013. Weathering the storm: Hurricanes and birth outcomes. Journal of Health Economics, 32, 487–503. Currie, J. & Schwandt, H. 2016. The 9/11 dust cloud and pregnancy outcomes: A reconsideration. Journal of Human Resources, 51, 805–831. Currie, J. & Stabile, M. 2003. Socioeconomic status and child health: Why is the relationship stronger for older children? American Economic Review, 93, 1813–1823. Currie, J. & Stabile, M. 2009. Mental health in childhood and human capital. The Problems of Disadvantaged Youth: An Economic Perspective, 115–148. D’Onofrio, B. M., Singh, A. L., Iliadou, A., Lambe, M., Hultman, C. M., Neiderhiser, J. M., Långström, N. & Lichtenstein, P. 2010. A quasi‐experimental study of maternal smoking during pregnancy and offspring academic achievement. Child Development, 81, 80–100.

304  Handbook of health inequalities across the life course Elder, G. H. 1994. Time, human agency, and social change: Perspectives on the life course. Social Psychology Quarterly, 57, 4–15. Evensen, M., Lyngstad, T. H., Melkevik, O. & Mykletun, A. 2016. The role of internalizing and externalizing problems in adolescence for adult educational attainment: Evidence from sibling comparisons using data from the Young HUNT Study. European Sociological Review, 32, 552–566. Figlio, D., Guryan, J., Karbownik, K. & Roth, J. 2014. The effects of poor neonatal health on children’s cognitive development. American Economic Review, 104, 3921–3955. Fletcher, J. 2010. Adolescent depression and educational attainment: results using sibling fixed effects. Health Economics, 19(7), 855–871. Frederiksen, L. E., Mader, L., Feychting, M., Mogensen, H., Madanat‐Harjuoja, L., Malila, N., Tolkkinen, A., Hasle, H., Winther, J. F. & Erdmann, F. 2019. Surviving childhood cancer: A systematic review of studies on risk and determinants of adverse socioeconomic outcomes. International Journal of Cancer, 144, 1796–1823. Fried, R., Petty, C., Faraone, S. V., Hyder, L. L., Day, H. & Biederman, J. 2016. Is ADHD a risk factor for high school dropout? A controlled study. Journal of Attention Disorders, 20, 383–389. Ghaderi, S., Engeland, A., Gunnes, M. W., Moster, D., Ruud, E., Syse, A., Wesenberg, F. & Bjørge, T. 2016. Educational attainment among long-term survivors of cancer in childhood and adolescence: A Norwegian population-based cohort study. J Cancer Surviv, 10, 87–95. Goodman, A., Joyce, R. & Smith, J. P. 2011. The long shadow cast by childhood physical and mental problems on adult life. Proceedings of the National Academy of Sciences, 108, 6032–6037. Goulding, S. M., Chien, V. H. & Compton, M. T. 2010. Prevalence and correlates of school drop-out prior to initial treatment of nonaffective psychosis: Further evidence suggesting a need for supported education. Schizophrenia Research, 116, 228–233. Grandjean, P. & Landrigan, P. J. 2014. Neurobehavioural effects of developmental toxicity. The Lancet Neurology, 13, 330–338. Grätz, M. & Torche, F. 2016. Compensation or reinforcement? The stratification of parental responses to children’s early ability. Demography, 53, 1883–1904. Haas, S. A. 2006. Health selection and the process of social stratification: The effect of childhood health on socioeconomic attainment. Journal of Health and Social Behavior, 47, 339–354. Haas, S. A. & Fosse, N. E. 2008. Health and the educational attainment of adolescents: Evidence from the NLSY97. Journal of Health and Social Behavior, 49, 178–192. Haas, S. A., Glymour, M. M. & Berkman, L. F. 2011. Childhood health and labor market inequality over the life course. Journal of Health and Social Behavior, 52, 298–313. Han, K., Ran, Z., Wang, X., Wu, Q., Zhan, N., Yi, Z. & Jin, T. 2021. Traffic-related organic and inorganic air pollution and risk of development of childhood asthma: A meta-analysis. Environmental Research, 194, 110493. Härkönen, J., Kaymakçalan, H., Mäki, P. & Taanila, A. 2012. Prenatal health, educational attainment, and intergenerational inequality: The Northern Finland Birth Cohort 1966 Study. Demography, 49, 525–552. Heckman, J. J. 2007. The economics, technology and neuroscience of human capability formation. Proceedings of the National Academy of Sciences, 104(33): 13250–13255. Hoek, H. W., Brown, A. S. & Susser, E. 1998. The Dutch famine and schizophrenia spectrum disorders. Social Psychiatry and Psychiatric Epidemiology, 33, 373–379. Hoffmann, R. 2011. Socioeconomic inequalities in old-age mortality: A comparison of Denmark and the USA. Social Science & Medicine, 72, 1986–1992. Hoynes, H., Schanzenbach, D. W. & Almond, D. 2016. Long-run impacts of childhood access to the safety net. American Economic Review, 106, 903–934. Hsin, A. 2012. Is biology destiny? Birth weight and differential parental treatment. Demography, 49, 1385–1405. Jackson, M. I. 2009. Understanding links between adolescent health and educational attainment. Demography, 46, 671–694. Jackson, M. I. 2010. A life course perspective on child health, cognition and occupational skill qualifications in adulthood: Evidence from a British cohort. Social Forces, 89, 89–116. Jackson, M. I. 2015a. Cumulative inequality in child health and academic achievement. Journal of Health and Social Behavior, 56, 262–280.

The influence of early health on educational and socioeconomic outcomes  305 Jackson, M. I. 2015b. Early childhood WIC participation, cognitive development and academic achievement. Social Science & Medicine, 126, 145–153. Kelly, E. 2011. The scourge of Asian flu in utero exposure to pandemic influenza and the development of a cohort of British children. Journal of Human Resources, 46, 669–694. Klocke, A. & Stadtmüller, S. 2022. Social inequalities, social capital, and health inequalities in the process of growing up. In: Hoffmann, R. (ed.) Handbook of Health Inequalities across the Life Course. Edward Elgar Publishing, Cheltenham, UK and Northampton, MA, USA. Kramer, M. S., Aboud, F., Mironova, E., Vanilovich, I., Platt, R. W., Matush, L., Igumnov, S., Fombonne, E., Bogdanovich, N., Ducruet, T., Collet, J.-P., Chalmers, B., Hodnett, E., Davidovsky, S., Skugarevsky, O., Trofimovich, O., Kozlova, L., Shapiro, S. & Group, P. O. B. I. T. S. 2008. Breastfeeding and child cognitive development: New evidence from a large randomized trial. Archives of General Psychiatry, 65, 578–584. Kröger, H., Fritzell, J. & Hoffmann, R. 2016. The association of levels of and decline in grip strength in old age with trajectories of life course occupational position. PLoS One, 11, e0155954. Kuh, D., Ben-Shlomo, Y., Lynch, J., Hallqvist, J. & Power, C. 2003. Life course epidemiology. Journal of Epidemiology and Community Health, 57, 778. Link, B. G. & Phelan, J. 1995. Social conditions as fundamental causes of disease. Journal of Health and Social Behavior, 35, 80–94. Lynch, J. L. & Von Hippel, P. T. 2016. An education gradient in health, a health gradient in education, or a confounded gradient in both? Social Science & Medicine, 154, 18–27. Lynch, J. W., Smith, G. D., Kaplan, G. A. & House, J. S. 2000. Income inequality and mortality: Importance to health of individual income, psychosocial environment, or material conditions. British Medical Journal, 320, 1200–1204. Mackenbach, J. P. 2012. The persistence of health inequalities in modern welfare states: The explanation of a paradox. Social Science & Medicine, 75, 761–769. Marmot, M. & Wilkinson, R. G. 2001. Psychosocial and material pathways in the relation between income and health: A response to Lynch et al. British Medical Journal, 322, 1233–1236. McLeod, J. D. & Fettes, D. L. 2007. Trajectories of failure: The educational careers of children with mental health problems. American Journal of Sociology, 113, 653–701. Miech, R. A., Caspi, A., Moffitt, T. E., Wright, B. R. E. & Silva, P. A. 1999. Low socioeconomic status and mental disorders: A longitudinal study of selection and causation during young adulthood. American Journal of Sociology, 104, 1096–1131. Mikkonen, J., Moustgaard, H., Remes, H. & Martikainen, P. 2018. The population impact of childhood health conditions on dropout from upper-secondary education. Journal of Pediatrics, 196, 283–290. Mikkonen, J., Remes, H., Moustgaard, H. & Martikainen, P. 2020. Evaluating the role of parental education and adolescent health problems in educational attainment. Demography, 57, 2245–2267. Mikkonen, J., Remes, H., Moustgaard, H. & Martikainen, P. 2021. Early adolescent health problems, school performance, and upper secondary educational pathways: A counterfactual-based mediation analysis. Social Forces, 99, 1146–1175. Mishra, G., Nitsch, D., Black, S., De Stavola, B., Kuh, D. & Hardy, R. 2009. A structured approach to modelling the effects of binary exposure variables over the life course. International Journal of Epidemiology, 38, 528–537. Muller, C., Sampson, R. J. & Winter, A. S. 2018. Environmental inequality: The social causes and consequences of lead exposure. Annual Review of Sociology, 44, 263–282. Oreopoulos, P., Stabile, M., Walld, R. & Roos, L. L. 2008. Short-, medium-, and long-term consequences of poor infant health: An analysis using siblings and twins. Journal of Human Resources, 43, 88–138. Palloni, A. 2006. Reproducing inequalities: Luck, wallets, and the enduring effects of childhood health. Demography, 43, 587–615. Palloni, A., Milesi, C., White, R. G. & Turner, A. 2009. Early childhood health, reproduction of economic inequalities and the persistence of health and mortality differentials. Social Science & Medicine, 68, 1574–1582. Pampel, F. C., Krueger, P. M. & Denney, J. T. 2010. Socioeconomic disparities in health behaviors. Annual Review of Sociology, 36, 349. Persson, P. & Rossin-Slater, M. 2018. Family ruptures, stress, and the mental health of the next generation. American Economic Review, 108, 1214–1252.

306  Handbook of health inequalities across the life course Persson, S., Dahlquist, G., Gerdtham, U. G. & Steen Carlsson, K. 2013. Impact of childhood-onset type 1 diabetes on schooling: A population-based register study. Diabetologia, 56, 1254–1262. Potente, C., Harris, K. M., Chumbley, J., Cole, S. W., Gaydosh, L., Xu, W., Levitt, B. & Shanahan, M. J. 2021. The early life course of body weight and gene expression signatures for disease. American Journal of Epidemiology, 190, 1533–1540. Settersten, R. A., Bernardi, L., Härkönen, J., Antonucci, T. C., Dykstra, P. A., Heckhausen, J., Kuh, D., Mayer, K. U., Moen, P. & Mortimer, J. T. 2020. Understanding the effects of Covid-19 through a life course lens. Advances in Life Course Research, 45, 100360. Sigle‐Rushton, W., Lyngstad, T. H., Andersen, P. L. & Kravdal, Ø. 2014. Proceed with caution? Parents’ union dissolution and children’s educational achievement. Journal of Marriage and Family, 76, 161–174. Slama, R., Darrow, L., Parker, J., Woodruff, T., Strickland, M., Nieuwenhuijsen, M., Glinianaia, S., Hoggatt, K. J., Kannan, S. & Hurley, F. 2008. Atmospheric pollution and human reproduction: Report of the Munich International Workshop. Environ Health Perspect, 116, 791–798. Smith, J. P. 1999. Healthy bodies and thick wallets: The dual relation between health and economic status. Journal of Economic Perspectives, 13, 145–166. Smith, J. P. 2009. The impact of childhood health on adult labor market outcomes. The Review of Economics and Statistics, 91, 478–489. Susser, M. & Stein, Z. 1994. Timing in prenatal nutrition: A reprise of the Dutch Famine Study. Nutrition Reviews, 52, 84–94. Torche, F. 2018. Prenatal exposure to an acute stressor and children’s cognitive outcomes. Demography, 55, 1611–1639. Torche, F. & Conley, D. 2016. A pound of flesh: The use of birthweight as a measure of human capital endowment in economics research. In: Komlos, J. & Kelly, I. (eds.) The Oxford Handbook of Economics and Human Biology. New York, NY: Oxford University Press. Torche, F. & Echevarria, G. 2011. The effect of birthweight on childhood cognitive development in a middle-income country. International Journal of Epidemiology, 40, 1008–1018. Van de Werfhorst, H. G. & Mijs, J. J. 2010. Achievement inequality and the institutional structure of educational systems: A comparative perspective. Annual Review of Sociology, 36, 407–428. VanderWeele, T. 2015. Explanation in Causal Inference: Methods for Mediation and Interaction. Oxford University Press. West, P. 1991. Rethinking the health selection explanation for health inequalities. Social Science & Medicine, 32, 373–384. Wilkinson, R., Pickett, K. & Cato, M. S. 2009. The Spirit Level: Why More Equal Societies Almost Always Do Better. Penguin, London.

20. Divergence and convergence: how health inequalities evolve as we age Johan Fritzell and Johan Rehnberg

INTRODUCTION The interdependencies between age and health inequalities are many. First, the socioeconomic conditions that people live under are to a large extent dependent on age-stratified institutions and processes such as the education system, social protection schemes, the labour market, and pension systems. Second, in high- and middle-income countries, rates of the most common illnesses and causes of death – heart disease, stroke, dementia, and many cancers – increase dramatically with age. Two theoretical perspectives are often used to discuss and understand age effects in health inequalities: the age-as-leveller hypothesis and theories of accumulation. The age-as-leveller hypothesis proposes that age somehow acts as an equalizer of inequalities between groups; with increased age, the relationship between socioeconomic position and health should attenuate. In contrast, theories of accumulation suggest that the accumulation of advantage and disadvantage leads to greater diversity within cohorts as time passes. In the first part of the chapter, we present a short overview of the commonly used theoretical perspectives on why and how age influences health inequalities. While inequalities can be studied in relation to a wide variety of social conditions and categories, we will in this chapter concentrate on socioeconomic inequality. Moreover, our focus of attention will be on the distinction between absolute and relative inequalities. This distinction is not merely, or even primarily, a methodological issue. Instead, as we will show in this chapter, the interpretation of how health inequalities evolve as we age differs substantially depending on the operationalisation of inequalities, a topic that is surprisingly often neglected. Absolute measures of inequality are commonly expressed on the outcome scale, for example differences in life expectancy measured in years between educational groups or the difference in prevalence rates between income groups. Relative measures, on the other hand, are expressed as a ratio, for example how many times higher mortality is in a group of low-educated persons compared to a group of high-educated persons. When comparing health inequalities across different groups or across time, a reduction of relative inequality may well coincide with an increase of absolute inequality. Moreover, suggested actions and policies for reducing health inequalities may be different if the focus is on relative or absolute inequalities. For example, evaluations of policy reforms or comparisons between regions and countries may come to different conclusions depending on what measure is used. In many countries, public health authorities and supranational organisations have plans for reducing health inequality. However, the type of inequality targeted, whether absolute or relative, is seldom specified. In the second part of the chapter, we present an overview of previous research that has examined the magnitude of health inequalities across the life course and that included older persons in their analyses. Next, we present an empirical illustration with the total population 307

308  Handbook of health inequalities across the life course register data from Sweden to show the (possibly) stark difference between absolute and relative measures of income inequality in mortality across the adult life course. We conclude with a discussion on how theoretical perspectives of health inequalities and ageing relate to empirical data, how they can be disentangled and how they relate to public health.

THEORIES ON AGE AND HEALTH INEQUALITIES It is evident that socioeconomic conditions, health, and mortality vary over the life course and that the causes of this variation are attributable to widely different and complex mechanisms and processes. The theories and hypotheses that have been used to interpret how socioeconomic inequalities in health vary across the life course can be divided into two categories: the first accounts for patterns of decreasing health inequalities with increased age, and the second focuses on patterns of increasing health inequalities with increased age. Decreasing Inequalities across the Life Course: Age-as-a-Leveller The age-as-a-leveller hypothesis is used in the literature as an umbrella term that includes several different processes which predict a convergence of socioeconomic health inequalities in older ages. The literature on this hypothesis can be traced back to the 1960s and 1970s, when it was used to understand why racial inequalities sometimes declined in old age (National Urban League, 1964; Dowd and Bengtson, 1978). This was followed by an increasing number of studies on ageing and health during the 1980s that included the age-as-leveller explanation for decreased health inequalities in older ages (George, Okun and Landerman, 1985; Ferraro, 1987). Since then, it has become an often cited hypothesis in social gerontology and studies examining health inequalities in old age (House et al., 1994; Ferraro and Farmer, 1996; Dupre, 2007; Kim and Durden, 2007; Hoffmann, 2011; Rehnberg, 2020). The hypothesis proposes that age may act as a leveller of inequalities between groups, meaning that age can modify the relationship between socioeconomic position and health. Four processes have been suggested for the levelling of health inequalities in older ages: biological processes, welfare state policies, changing impact of major social stratification processes, and selective mortality. The specific processes mentioned in the literature vary from study to study, as does the emphasis on which is the most important process. Biological processes Biological ageing can refer to a multitude of physiological processes related to decay or breakdown of cells or bodily functions over time. Persons with the same chronological age score differently on various measures of biological ageing, which in turn is related to future health decline and mortality risks (Li et al., 2020). Biological limitations of the human body result in declining health for everyone at some stage of the ageing process, and according to this hypothesis, levelling of health inequalities would occur when biological ageing-related processes become stronger predictors for future health than any previous social exposure (House, Kessler and Herzog, 1990; Liang et al., 2002). There is empirical support for this hypothesis as studies have shown that illness, rather than age, account for a substantial levelling of health inequalities among older persons (Hoffmann, 2011; Rehnberg, 2020). In sum, an inevitable

How health inequalities evolve as we age  309 health decline in old age implies a weaker association between socioeconomic position and health among older persons. Changing impact of major social stratification processes The second reason for a supposed convergence of socioeconomic inequalities in health and mortality is that the direct impact of social stratification systems in capitalist societies is lessened in old age. A person’s socioeconomic position is primarily determined by their position in the labour market, which in turn determines level of income, power, and working conditions. A combination of these processes drive health inequalities (Mackenbach, 2019), but change in character later in life. Some are no longer active when people retire, such as working conditions, while other determinants partly shift to become outcomes from previous labour market achievements, such as income in old age (Hoffmann, Kröger and Pakpahan, 2018; Rehnberg et al., 2021). As a result, life course social stratification processes continue to play a role for living conditions and health in old age, but many of the processes that generate inequalities in socioeconomic position are weakened after retirement from the labour market. Welfare state policies The third set of factors in the age-as-leveller hypothesis concern age-varying processes related to social protection schemes of the welfare state. Old-age benefits, both cash and care, are a fundamental part the welfare state. Most modern welfare states have advanced pension systems that redistribute resources both horizontally across the life course and vertically across social strata. For retirees, pension income is often determined by accumulated income during their working lives, with little or no room for change after labour market exit. Important to note, however, is that the policies that determine pensions vary between countries and over time, and that in general, income inequalities tend to be smaller in old age than in younger ages, but with variations across countries (OECD, 2019). Like pension incomes, health care and long-term care rely on a re-distribution of resources from younger to older generations due to their greater need of care and support (Birnbaum et al., 2017). If health care policies are more equally distributed among persons in old age compared to other age groups, it could potentially result in more equal health care utilisation and, by extension, smaller health inequalities in older ages. Both Dupre (2007) and Preston (1984) argued this to be the case with Medicare in the United States. In sum, the tendencies for redistribution of welfare resources to primarily target older persons can potentially lead to reduced socioeconomic and health inequalities among older persons. Selective mortality The fourth process, sometimes highlighted as the main explanation in the age-as-leveller hypothesis, is selective mortality or selective survival (Vaupel and Yashin, 1985; Ferraro and Farmer, 1996; Dupre, 2007). This selective process results from socially patterned mortality: as time passes, a higher mortality rate in lower socioeconomic groups leaves a surviving population that is more homogenous, healthier, and more resilient to biological ageing processes. This selection process has been described as a type of cohort inversion, where the removal of persons that were likely to have severe health problems from a cohort will give the appearance of better health among lower socioeconomic groups than in the initial cohort where everyone was alive (Ferraro and Shippee, 2009). This process also occurs in higher socioeconomic groups but is less pronounced due to lower mortality rates at younger ages. The change in

310  Handbook of health inequalities across the life course the composition of surviving persons may at the aggregate level thus lead to a convergence between socioeconomic groups in terms of future mortality risk. The potentially large impact of selective mortality has led researchers to sometimes try to adjust for the process in order to quantify its effects on inequalities (Beckett, 2000; Hoffmann, 2008; Benzeval, Green and Leyland, 2011; Celeste and Fritzell, 2018). Increasing Inequalities across the Life Course: Theories of Accumulation In social gerontology, cumulative (dis)advantage theory is often used to explain patterns of increased inequality in old age. The theory suggests that the accumulation of advantage and disadvantage leads to greater diversity within cohorts as time passes (Dannefer, 1987, 2003; Rowe and Kahn, 1987; O’Rand, 1996; Kim and Durden, 2007). More general theories on accumulation originate from Merton’s (1968) so-called ‘Matthew effect’, where he describes a process of disproportional credit given to prominent researchers compared to less well known researchers for comparable work. On the basis of this Matthew effect, the accumulation of advantage and disadvantage has been generalised to age-related social processes that accumulate to generate intra-cohort heterogeneity with increased age (Dannefer, 1987, 2003). When health inequalities are concerned, both socioeconomic advantage and health status vary simultaneously as well as separately over time. This type of accumulation has been referred to as an extrinsic type of process wherein accumulation of one phenomenon affects accumulation of another phenomenon (Ferraro and Morton, 2016).

RELATIVE AND ABSOLUTE INEQUALITIES Diagnostic processes employed by medical doctors often seek to classify patients in a binary state, ill or healthy, and help doctors take decisions on appropriate medical treatments. The focus on binary health states has popularized statistical methods that estimate binary outcomes, which is reflected in the frequent use of regression methods such as logistic regression and survival analysis in epidemiology and public health research. Statistical measures of binary health outcomes are typically variants of risk or odds ratios, and such ratio measures are often presented without considering how common the outcome of the study is. Ratio measures are also commonly estimated when comparing mortality rates between groups in a population. Another common approach is to operationalise health outcomes on continuous scales with no clear distinction between healthy and sick individuals. In such cases, commonly used statistical methods are variations of linear ordinary least square regressions that estimate coefficients on the outcome scale and are expressed on an absolute scale. Absolute differences are also common when comparing mortality rates or life expectancy measures. Both relative and absolute measures of inequality can be found in the literature. However, King et al. (2012) showed that out of studies on health inequalities published in a selection of public health journals in 2009, only 7 per cent reported both absolute and relative measures, while 88 per cent only reported a relative measure. The common use of relative measures of inequality is perhaps partly explained by the strong tradition of a binary operationalisation of illness and disease and the subsequent focus on ratio measures described above. With that said, a binary operationalisation of health is certainly a valid strategy in many cases, but to

How health inequalities evolve as we age  311 Table 20.1

Infant mortality per thousand births by class and absolute and relative differences in England and Wales in 1911 and 2001 Class I

Class V

Absolute difference

Relative difference

1911

76.4

152.4

76 deaths per thousand

2 (152.4/76.4)

2001

3.8

7.4

3.6 deaths per thousand

1.95 (7.4/3.8)

Source: Graham (2007, Table 2.2).

exclusively concentrate on relative inequalities may provide an incomplete picture of health inequalities, especially from a policy planning and evaluation perspective. Another central consideration when measuring health inequalities is that both relative and absolute measures tend to be affected by the prevalence of the outcome. When an outcome is rare, for example mortality among younger persons, the possible range of values that absolute differences can vary between will be small due to the low prevalence in the studied population. When an outcome is more prevalent, for example mortality in older ages, absolute differences have the potential to be larger. Conversely, relative differences tend to be large when the prevalence of the outcome is rare, as it requires small absolute changes for large relative differences to appear. If, for example, the prevalence of the outcome increases, relative differences will decline if absolute differences remain constant. These arithmetic properties are elaborated in detail by Mackenbach et al. (2016). The interpretation of inequality measures when the prevalence of the outcome is different between and within groups over time is highly relevant when studying health inequalities across the life course. Mortality rates and most common diseases – heart disease, stroke, Alzheimer’s, dementia, and many cancers – increase dramatically with age (Kaeberlein, 2013). Therefore, studies that compare health inequalities in different age groups often compare groups with widely different prevalence of the outcome. To exemplify the different conclusions that might emerge from these two concepts and the importance for evaluating health inequalities and policy reforms more generally, we will use an empirical example. Table 20.1 shows infant mortality rates for social class I (highest) and social class V (lowest) in England and Wales from the years 1911 and 2001 (Graham, 2007). Not surprisingly, the risk of dying during the first year of life decreased tremendously during these 90 years. Interestingly, if we evaluate inequalities on a relative scale, we will conclude that inequalities remained at similar levels, with relative mortality differences at around two times higher in the lower class. On the other hand, if we evaluate absolute inequalities, we will conclude that there have been dramatic reductions in inequality, from a difference of 76 deaths per thousand births to 3.6 deaths. And, we would argue, it is these spectacular improvements during the 20th century in both classes that should be emphasized in the first place. Still, from a scientific and policy perspective, it is of course an interesting question why relative risks have prevailed despite fundamental societal changes over the century. The prevailing relative difference may suggest that despite medical and societal advances that have led to lower infant mortality, some factors that generate inequalities in infant mortality remain. One explanation that is put forward in the fundamental cause theory by Link and Phelan (1995) is that specific processes that generate inequalities can be prevented, but if the underlying, or fundamental, differences in socioeconomic resources and power relations remain, other inequality shaping processes will replace those processes that were prevented.

312  Handbook of health inequalities across the life course It should be stressed that the strong focus on relative inequalities is not solely found in studies of health inequalities but is often found in sociological studies that examine other types of inequalities. One example is the seminal work by Erikson and Goldthorpe (1992) on intergenerational social mobility, where they argue for the importance of relative measures and conclude that the relative chances of ending up in a certain social class position have been more or less constant during most of the 20th century. Still, as in the example on infant mortality, overwhelming changes in the composition of the class structure have increased the absolute chance for a person from the working class to end up in a higher non-manual position. Finally, the way inequalities are quantified can potentially have a strong impact on the interpretation of empirical findings and the extent to which they are supportive of a theory. In the next section, we will give a brief overview of the literature examining health inequalities in different age groups and provide an empirical example of how absolute and relative differences in mortality by income vary across ages in Sweden.

PREVIOUS STUDIES – A MAPPING REVIEW Theory identifies mechanisms and processes by which health inequalities may vary across the life course. But when are health inequalities largest – in older ages or in younger ages? Are these patterns different across measures of inequality? And what does previous research tell us about the mechanisms that generate these patterns? The empirical literature on health inequalities is voluminous, many studies are descriptive, and no single study gives a definite answer to these questions. A surprisingly large majority of studies on health inequalities ignore age variations and study persons of working age as one homogenous group, without including older persons. Other studies focus on children, such as infant mortality, and often with a global perspective (Houweling and Kunst, 2010). The fact that older persons are often neglected in health inequality studies is paradoxical, since most health problems that people experience typically occur later in life. In addition, mortality in affluent countries is nowadays low at ages below 60. For example, during the year 2020 in Sweden, 92 per cent of all deaths occurred at or above age 60 (Human Mortality Database, 2022). Another complicating factor is the type of indicators that should be used to assess the broad concept of health. A recent review on topics and indicators for measuring and monitoring health inequalities found that over 600 different health indicators were used in government related reports and databases (Albert-Ballestar and García-Altés, 2021). As mentioned earlier, whether the indicator is binary or continuous will influence the interpretation, but perhaps most important for age-related patterns of health inequalities is how a specific health indicator is related to age. Some health indicators are heavily age patterned, such as mortality or Alzheimer’s disease, while others have a weaker, if any, connection to age, such as self-rated health (Idler, 1993; Zajacova and Woo, 2016). We performed a mapping review of studies measuring socioeconomic inequalities in health and mortality across age groups. We selected studies that included persons aged 70 or older and quantified health inequalities in at least two age groups. Table 20.2 presents 31 studies that fulfil these criteria. The studies were found mainly by searching Google Scholar and PubMed for keywords related to the topic (health inequalities, age). Moreover, studies were found through references in other studies, comparable to a snowball sampling method. The 31 studies presented here should be considered only as a sample of studies that examine health

Country

Occupation

Education

measure

Socioeconomic

(2011)

Hoffmann

Herd (2006)

(2009)

and Breeze

Tabassum,

Gjonca,

(1996)

Preston

Elo and

UK

1980–2002 Denmark

1992–2002 USA

2002

1979–1985 USA

Dupre (2007) 1982–1992 USA

al. (2007)

59+

51–71

50+

25–89

25–85

Mortality

Functional health

ADL/IADL

Mortality

attack

diabetes, stroke, heart

Hypertension,

health

Physical and mental

35–74

Chandola et

1985–2002 UK

musculoskeletal pain

(2018)

Income

Education

wealth

social class,

hazard model

Proportional

curve model

Ordinal growth

regression

Logistic

regression

Education,

Logistic

income

prevalence rates

estimated

hazard ratio,

regression,

Logistic

curve model

Linear growth

identity link

regression with

binomial

regression,

Poisson

measures model

repeated

hierarchical

regression Linear

Ordinal logistic

Method

Education, log

Education

Occupational

psychological distress, Poverty

binary)

SRH (continuous and

impairment

Functional

Health outcome

and Fritzell

15+

15–76

32–86

Age

Poor teeth,

1968–2011 Sweden

1987–2007 Scotland

1982–1992 USA

Period

Relative risk

Odds ratio

Odds ratio

Odds ratio

hazard ratio

Odds ratio,

scale

on outcome

Visual

Rate difference

differences

ratio

Coefficient

Prevalence

Prevalence

probabilities

in predicted

Difference

probabilities

Predicted

coefficient

coefficient Log odds

Absolute

Relative

↓

then ↓

↑ up to age 63,

↓

↓

↓

↓

inequalities

Relative

↓

↑, then ↓

↑

↑ up to 45–64, then ↓

↑ up to age 65, then mixed

then ↓

↑ up to age 60–75, and

Absolute inequalities

31 studies examining health inequalities in at least two age groups that included persons aged 70 or older

Celeste

(2011)

Leyland

Green, and

(2000) Benzeval,

Beckett

Reference

Table 20.2

How health inequalities evolve as we age  313

(2019)

Leopold

(2016)

Leopold

(2014)

Korda et al.

(2007)

Durden

Kim and

(2004)

Huisman

(2003)

Mackenbach

Kunst, and

Huisman,

(2020)

Hu et al.

(1990)

and Herzog

Kessler,

House,

Reference

Country

11

countries

European

11

Australia

USA

2006–2014 Germany

1991–2010 Sweden

2006

1989, 1994

1986,

countries

1990–1995 European

1994

1988–2007

1985, 1986 USA

Period

25–83

24–75

45+

25+

– 90+

30

60–80+

50–79

25+

Age measure

Socioeconomic

grip strength

SRH, physical health,

SRH

psychological distress

physical functioning,

and osteoarthritis,

disease, cancer

diabetes, Parkinson’s

Education

income

Education,

Income

income

depression Heart disease,

Education,

Physical impairment,

Education

income

Mortality

Education,

disability

Education

income

SHR, daily activities,

Days in hospital

health

Self–reported physical Education,

Health outcome

↓ or stable

Coefficient on outcome scale

linear regression

measures

among women, ↓ in other

men, ↑ for grip strength

Mixed, ↑ for SRH among

age, stable in older ages

↓

↑ from early to middle

↓

differences in depression

↑, except for income

↑

↑ for men, ↓ for women

probabilities

differences

↓

men, ↓ for women

Remained or ↓ for

↓ between 50–69, ↑ between 70–79

Predicted

Prevalence ratio

outcome scale

Coefficient on

Rate difference

Rate difference

Rate difference

and over

Absolute inequalities

Largest in mid age, ↓ 75

Prevalence

Rate ratio

RII

Odds ratio,

Rate ratio

inequalities

Relative

outcome scale

coefficient

coefficient Coefficient on

Absolute

Relative

Hierarchical

panel regression

average logistic

Population

Prevalence rates

models

Latent growth

Mortality rates

Prevalence rates

regression

binomial

Negative

regression

least square

Ordinary

Method

314  Handbook of health inequalities across the life course

Change in: ADL,

Health outcome

(2019)

Fritzell

Fors, and

Rehnberg,

(2020)

Rehnberg

(2014)

Rickertsen

and

Gustavsen,

Øvrum,

al. (2016)

Mortensen et

(2008)

and Breeze

Nazroo,

McMunn,

(2014)

Mather et al.

(2013)

1990–2009 Sweden

1990, 2005 Sweden

1997–2011 Norway

countries

1995, 2003 Nordic

Four

2002–2004 UK

2006–2008 Australia

2004–2007 Europe

31–99

65–75

and

50–60

25–79

25+

50+

45+

50–80

Mortality

Mortality

SRH, physical activity

Mortality

mortality

heart disease (IHD),

SRH, ADL, ischemic

SRH, heart disease

strength

self–rated health, grip

chronic diseases,

Age

IADL, mobility,

Country

Engelhartdt

Period

Leopold and

Reference

Income

death

Probability of

Cox regression

regression

income

Income

Logistic

Mortality rates

regression

Education,

Income

Wealth

Areas (SEIFA) Logistic

regression

Risk ratio

ratios

Hazard

Odds ratio

difference

Probability

probabilities

Predicted

Rate difference

differences

Prevalence

outcome scale

regression

Poisson

Coefficient on

Economic Indexes for

coefficient

coefficient

RII

Absolute

Relative

Linear

Method

Socio–

education,

Income,

Education

measure

Socioeconomic

inclusion

↓

↑ (50–60 > 65–75)

outcome and mortality ADL

↑ up to age 80–90, then ↓

physical activity

↓ for income and SRH,

Mixed

↑ up to age 75, then ↓

Mixed, depending on IHD, remained for

were stable

diseases and SRH that

↑ except for chronic

Absolute inequalities

↓ for SRH and

↓

inequalities

Relative

How health inequalities evolve as we age  315

Country

States

United

Netherlands

The

SRH

Linear regression

Education, income

Education

regression

Logistic

curve model

Education

Linear growth

income, wealth

(LOWESS)

 

Odds ratio

 

outcome scale

 

↓ in later old age

↑

Stable or ↑

then ↓

mortality: ↓

Coefficient on

↑ to late middle age, Only estimated for

outcome scale

mortality

↑

wealth

↑ for education, ↓ for

Absolute inequalities

Coefficient on

↓ for disability and

hazard ratio

outcome scale

Coefficient on

wealth

education, ↓ for

Remained for

inequalities

Relative

Odds ratio,

RII

Rate difference

coefficient

coefficient Rate ratio,

Absolute

Relative

Prevalence rates Relative rate

Education,

income

occupation,

Education,

cox regression

regression and

Logistic

regression

wealth

Wealth

Poisson

Method

Education,

measure

Socioeconomic

regression

Herd (2014)

40–74

functioning,

and cognitive

Physical, emotional,

SRH

mortality

SRH, disability,

inactivity

obesity, physical

Disability, mortality,

SRH, physical health

Mortality

Health outcome

Nonparametric

1997–2010 USA

65+

26–92

18–80

51+

18+

65+

Age

Montez, and

1998–2011 USA

1984–2001

1983–2000

1998–2008 USA

1979, 1990 USA

2001–2008 Spain

Period

Zajacova,

(2015)

Xu et al.

Elder (2007)

Shuey, and

Willson,

et al. (2010)

Kippersluis

van

(2014)

Shaw et al.

(1996)

Ross and Wu

(2015)

Reques et al.

Reference

316  Handbook of health inequalities across the life course

How health inequalities evolve as we age  317 inequalities in different age groups, since there are most certainly other studies that fulfil these criteria but were not found in our search. Nonetheless, the included studies represent a wide range of disciplines. We found 15 studies with no upper age limit, eight studies with an age limit above 80 and eight studies with an age limit between 70 and 80. The measurement of health in these studies varied, but the most common outcome was self-rated health, which was used in 11 studies. The second most common health outcome was mortality, used in nine studies. Other common outcome measures were physical and cognitive functioning, and disease specific outcomes such as heart disease. Similarly, the socioeconomic measures that were used varied between the studies, with the most common measures being education or income, followed by occupational class and wealth. The method used for estimating the magnitude of health inequalities determines on what scale the measure is reported. In 11 of the studies, both absolute and relative measures of inequality were used. In 13 studies, only absolute measures were reported, and in seven studies only relative measures were reported. Our primary aim in this part of the chapter is to illustrate the differences in age patterns that arise depending on which measure of inequality that is used. In none of the 18 studies that used relative measures of inequality did the authors observe increased inequalities in older ages. Several studies, however, reported stable relative inequalities across age groups. Huisman et al. (2003) observed declining relative inequalities with increased age for several health outcomes, using income as the stratifying variable. Using education, however, they observed varying patterns, and in certain health outcomes the magnitude of relative inequalities remained also in older ages. Similarly, Reques et al. (2015) observed remaining relative health inequalities among older persons when using education, but declining patterns of inequality when using indicators of wealth. Lastly, Hu et al. (2020) found declining relative health inequalities with increased age in some age groups, while in other age groups they found stable inequalities with increased age, using number of hospital days as the health outcome. In contrast, absolute inequalities in health showed more mixed patterns than relative inequalities. In 17 of the 24 studies that included absolute measures, the patterns showed increasing inequalities with older age. In several studies, absolute inequalities increased up to ages between 65 and 85, after which they declined. In five studies, the patterns were less clear and showed mixed findings depending on the outcome and age group that was included. In two studies, absolute health inequalities declined with age. The results from the 31 studies described above indicate that when health inequalities are quantified with relative measures, the patterns tend to be that health inequalities decline in older ages. However, when health inequalities are quantified on an absolute scale, the patterns tend to be more mixed, with either decreasing, stable or increasing inequalities. In sum, we can conclude that with increased age, poor or declining health becomes increasingly prevalent, and, as a result, relative inequalities tend to decline, but absolute inequalities tend to show more varied patterns.

318  Handbook of health inequalities across the life course

ABSOLUTE AND RELATIVE INEQUALITIES IN MORTALITY: AN EMPIRICAL ILLUSTRATION In this section, we make an empirical illustration of a typical pattern of mortality inequalities across the life course. Using Swedish population register data, we calculated three-year all-cause mortality rates per 1000 observed person years in two income groups, those with the lowest 30 per cent and those with the highest 30 per cent incomes. Income was measured as household income adjusted for number of household members (household income divided by the square root of number of household members), and income percentiles were calculated within each one-year age group. The dataset included all persons aged 30 or older living in Sweden between the years 2010 and 2017, giving a total of 4,157,612 persons that were included from the bottom and top 30 per cent of the income distribution. To reduce the impact of yearly variations in income, household income was measured as the average household income between 2010 and 2013, and three-year mortality was measured during 2015, 2016, and 2017. The data was retrieved from official records on taxable income and from the cause of death register. Figure 20.1a shows the mortality rate by income from age 30 to age 90 and over. In both income groups, the mortality rate is at low levels from ages 30 to 55. At age 55 there is a discernible increase among those with the lowest incomes, and we observe diverging absolute patterns. After the age of around 70 the mortality rates in both income groups start to increase rapidly with each increasing year. Relative inequalities in mortality between these two income groups, measured by the risk ratio, are shown in Figure 20.1b. Despite the low levels reported in Figure 20.1a, the relative risk ratio is high, between 2 and 4.5, up to age 60. After age 60 a continuous decrease of relative inequality mortality is seen. This decline coincides with the start of the rapidly increasing mortality rates after age 60 seen in Figure 20.1a. Above age 85, the risk ratio is around 1, indicating no remaining inequalities in mortality between the two income groups. Using the same data, we report absolute inequalities in mortality in Figure 20.1c. Contrary to the patterns observed for the relative inequalities (Figure 20.1b), absolute inequalities are low between ages 30 to 50, when the morality rates also are low. After age 50, absolute inequalities increase rapidly and peak around age 75 to 80, after which they start to decline. This example has illustrated a common age pattern in health inequalities, namely that relative inequalities are low at younger ages and decline in older age groups, while absolute inequalities show an opposite pattern. Several factors influence these results. As previously discussed, the choice of health outcome is important. In the same data, other patterns would likely be found if we had used another health measure (e.g., Celeste and Fritzell, 2018). Mortality is perhaps the most extreme example of an outcome that everyone experiences, and that is very much a function of age, as compared to more subjective feelings of health status such as self-rated health. Because of this, the arithmetic of how inequalities in mortality are measured becomes central to whether we conclude that inequalities increase or decrease with age. Under conditions when the prevalence of the outcome, in this case mortality, increases rapidly from low starting levels, we are unlikely to observe decreasing relative differences (Mackenbach et al., 2016).

How health inequalities evolve as we age  319

Note: Lines in the lower two figures estimated with locally estimated scatterplot smoothing (LOESS).

Figure 20.1

Three-year mortality rate per 1000 person years in Sweden by income group, and relative and absolute inequalities in mortality by income group, year 2015–2017

What we have shown here is a central part to answering the question whether health inequalities increase or decrease with age. Specifically, we showed that an outcome with large age variations, such as mortality, requires careful attention to the scale on which inequalities are evaluated. To rely solely on relative measures of inequality will in many cases only tell half the story. Further implications of these insights are elaborated on in the next section.

320  Handbook of health inequalities across the life course

CONCLUDING DISCUSSION In this chapter we have highlighted how socioeconomic health inequalities tend to evolve over the life course, with a special focus on older ages. There are at least three reasons for this focus. First, it is surprisingly common for studies on health inequalities to neglect older people. Research on health and living conditions among older person has come a long way since Suzman and Riley’s (1985) seminal work where they introduced the term the oldest old. Yet, our knowledge of socioeconomic health inequalities at older ages is relatively sparse, and our knowledge is even more limited about the fastest growing group of older persons: the oldest-old. Second, deaths and most illnesses are rare occurrences in younger ages, and to exclude older persons in studies of health inequalities implies the exclusion of most cases of death and illnesses. Third, the inclusion of and focus on older persons is also motivated by the fundamental demographic change that our societies have and continue to undergo, captured by the term “ageing societies”, with repercussions on both the macro- and the micro-level. Given these facts, old-age health inequalities and life course determinants of inequalities are likely to receive more attention in future research. We identified two contrasting hypotheses on how socioeconomic health inequalities evolve at different life stages, which can be summarized by convergence or divergence of health inequalities with increased age. The umbrella theory of age-as-a-leveller, the convergence view, highlights both social and biological processes, as well as statistical reasons for why a convergence of inequalities may occur. Theories of divergence focus more on social constructs of the life course and rely on the presumption that life stages are interlinked and that resources and conditions at earlier time points can accumulate, and in turn lead to growing inequalities. The standpoint we have promoted in this chapter is that the “or” in the question of convergence or divergence over the life course should be replaced by “and”. We have highlighted that the question at hand depends on whether we measure absolute or relative socioeconomic inequalities. In the case of mortality, it seems to be an empirical regularity that relative socioeconomic inequalities converge before the age of 60, and that they are substantially smaller among older persons than among younger persons. In contrast, relying on absolute differences, it seems to be almost an empirical regularity that inequalities in mortality diverge up to advanced old age. In our own empirical illustration using Swedish register data, we find that the peak of inequalities measured on an absolute scale occurred at around 75 to 80 years of age, after which inequalities start to rapidly converge. What, then, explains the empirical patterns that we have shown? One intriguing viewpoint that can be found in the literature is the need to separate processes that occur at the individual level and the population level. Dupre (2007) suggests that selective mortality and the impact it has on the composition of the surviving population is a mechanism that occurs at the population level. At the same time, on the individual level, mechanisms related to cumulative advantage can be active. This implies that there are not necessarily two opposing theoretical views between either convergence or divergence; rather, they are complementary for understanding health inequalities across the life course. Selective mortality is certainly a convincing theoretical model for the weakened association between socioeconomic conditions and health among older persons, and there is some empirical support for this hypothesis (Willson, Shuey and Elder, 2007). In contrast, and perhaps surprisingly, several studies have found a small or, at most, a modest impact of selective mortality on health inequalities (Beckett, 2000; Herd, 2006; McMunn, Nazroo and Breeze, 2008; Celeste

How health inequalities evolve as we age  321 and Fritzell, 2018; Rehnberg, Fors and Fritzell, 2019). In these studies, converging patterns of health inequalities remained when employing different techniques that adjust for mortality, suggesting that selective mortality is, in fact, not the main driver for convergence. The answer might instead lie in the process of gradual decline in health as people age – a process that is seldom covered under the age-as-leveller hypotheses and that, like selective mortality, affects the composition of the population. With increasing age, a continuously increasing share of the population will experience the transition from good health to poor health. The implications for converging health inequalities are that people with higher socioeconomic position can perhaps delay the onset of poor health, but when a large enough share of the population has transitioned into poor health the remaining health inequalities will be diminished. This has been shown empirically: there is a strong socioeconomic gradient in the transition from good to poor health, but the gradient is weaker in the transition from poor health to death (Buckley et al., 2004; Hoffmann, 2011; Robitaille et al., 2018; Rehnberg, 2020). Consistent with this hypothesis, researchers have observed socioeconomic gradients in cancer incidence and in survival from cancers with high survival rates (Birch-Johansen et al., 2008; Dalton, Schüz, et al., 2008). However, in people diagnosed with cancers with lower survival rates, socioeconomic gradients in mortality are less pronounced or even non-existent (Dalton, Steding-Jessen, et al., 2008). Both the selective mortality and the transition to poor health hypotheses are related to compositional changes of the population and do not necessarily imply that the causal mechanisms that generate health inequalities are modified in old age. An important caveat for these two hypotheses is that compositional processes are less valid for health indicators that have a weaker relation to age, such as self-rated health, or indicators in which the onset of a disease starts early, which has been reported in studies on oral health (Celeste and Fritzell, 2018). Compositional changes related to ageing are doubtless central for understanding converging patterns of health inequalities, but what about the more traditional social determinants of health? As described earlier in the chapter, several of the processes that connect social stratification to health originate from working conditions and income, conditions which either disappear or change in character after retirement. In old age, living conditions and income shift to become outcomes from earlier stratification processes, supplemented by a stronger reliance on welfare state programmes that redistribute resources horizontally over the life course and from younger to older generations (Birnbaum et al., 2017). There is, however, a notable lack of research that has directly investigated these dynamics. Lastly, we take a similar position to that suggested by Dupre (2007), in that it is imperative to take into consideration the level of analysis. Processes that occur at the population level between groups, such as selective mortality, tend to generate converging population-level health inequalities. On the other hand, processes that occur at the individual level, such as cumulative advantage, tend to generate diverging individual-level health inequalities. Public Health and Socioeconomic Inequality Both absolute and relative differences are valuable in order to understand inequalities. We have, however, shown that the interpretation of how inequalities evolve across the life course often differs between absolute and relative measures, in part depending on how common the outcome measure is. From a public health perspective, it can be misleading to interpret a relative risk without considering the absolute change in mortality, as illustrated in the example of changing inequal-

322  Handbook of health inequalities across the life course ities in infant mortality in England and Wales between 1911 and 2001. Interestingly, the type of measure used is seldom discussed in the literature on age patterned health inequalities, and rather seems to be a consequence of the method that best suited the data. We suggest that when evaluating whether health inequalities diverge or converge with age, both absolute and relative measures of inequality need to be assessed. An example of the important distinction between absolute and relative inequalities comes from the first, out of many, important European cross-national studies on socioeconomic inequalities in health by Mackenbach and colleagues (1997). They came to the surprising conclusion that, based on analyses of relative inequalities, the Nordic countries (in this case Norway and Sweden) had higher levels of inequality than most other European countries. In a comment Vågerö and Erikson (1997), however, noted that despite higher relative inequalities between manual and non-manual workers, manual workers in Sweden had the lowest risk of dying among the included countries and the absolute difference in mortality between manual and non-manual workers was among the smallest. This example highlights the problem of focusing on one measure of inequality: absolute and relative measures showed opposite patterns, and whether the Nordic model was a success or a failure required a more nuanced answer. This controversy was one of many insights that led to a change in how inequalities were reported in research on cross-national comparison of health inequalities, and now it is common to report both absolute and relative measures, as well as level of mortality.1 The same principle, we argue, should be strived for in studies on how health inequalities develop over the life course. The interpretation of age patterns in health inequalities differs between absolute and relative measures, and any public health evaluation of the magnitude of these inequalities must take this distinction into account. The absolute perspective is important to stress also from a strict scientific perspective. A common approach in epidemiology has been to compare relative differences when assessing causality. However, the theories we have discussed in this chapter are mainly concerned with the actual health status or death risk of people, rather than their health status or death risk relative to other persons. With this in mind, it is not self-evident that relative measures of difference are preferable when assessing causality. Poole (2010) extends this discussion and shows that the arguments in epidemiology that were originally favouring the risk ratio for assessing causality were overstated, and that under several conditions a rate difference measure is equally valid. Poole argues that the traditional use of relative measures has been “handed down from one generation to the next, without citation or critical reflection, as though [its] truth were self-evident” (Poole, 2010, p. 3). Hence, when investing in processes with the aim of assessing causality, absolute difference measures can sometimes be of equal use or even preferable over relative measures, and we believe that an informed choice of measure should be a priority.

ACKNOWLEDGEMENTS We would like to thank the reviewers and our colleagues at the Social Gerontology unit of the Aging Research Center, Karolinska Instiutet & Stockholm University for valuable comments on an earlier version of this chapter. We acknowledge financial support from Marianne and Marcus Wallenberg Foundation (MMW 2016.0017) and Forte (2016-07206 and 2020-00071).

How health inequalities evolve as we age  323

NOTE 1.

The example here should by no means be read as if the Nordic countries do not face problems with mortality inequality. On the contrary, in a recent cross-national study on trends in socioeconomic inequality in 15 countries, the Nordic countries were among the worst performers (Mackenbach et al., 2019).

REFERENCES Albert-Ballestar, S. and García-Altés, A. (2021). ‘Measuring health inequalities: a systematic review of widely used indicators and topics’. International Journal for Equity in Health, 20 (1). doi: 10.1186/ s12939-021-01397-3. Beckett, M. (2000). ‘Converging health inequalities in later life – an artifact of mortality selection?’ Journal of Health and Social Behavior, 41 (1), pp. 106–119. doi: 10.2307/2676363. Benzeval, M., Green, M. J. and Leyland, A. H. (2011). ‘Do social inequalities in health widen or converge with age? Longitudinal evidence from three cohorts in the West of Scotland’. BMC Public Health, 11 (1), p. 947. doi: 10.1186/1471-2458-11-947. Birch-Johansen, F., Hvilsom, G., Kjær, T. and Storm, H. (2008). ‘Social inequality and incidence of and survival from malignant melanoma in a population-based study in Denmark, 1994–2003’. European Journal of Cancer, 44 (14), pp. 2043–2049. doi: 10.1016/j.ejca.2008.06.016. Birnbaum, S., Ferrarini, T., Nelson, K. and Palme, J. (2017). The generational welfare contract: justice, institutions and outcomes. Cheltenham, UK and Northampton, MA, USA: Edward Elgar Publishing. Buckley, N. J., Denton, F. T., Leslie Robb, A. and Spencer, B. G. (2004). ‘The transition from good to poor health: an econometric study of the older population’. Journal of Health Economics, 23 (5), pp. 1013–1034. doi: 10.1016/j.jhealeco.2004.03.001. Celeste, R. K. and Fritzell, J. (2018). ‘Do socioeconomic inequalities in pain, psychological distress and oral health increase or decrease over the life course? Evidence from Sweden over 43 years of follow-up’. J Epidemiol Community Health, 72 (2), pp. 160–167. doi: 10.1136/jech-2017-209123. Chandola, T., Ferrie, J., Sacker, A. and Marmot, M. (2007). ‘Social inequalities in self reported health in early old age: follow-up of prospective cohort study’. BMJ, 334 (7601), p. 990. doi: 10.1136/ bmj.39167.439792.55. Dalton, S. O., Schüz, J., Engholm, G., Johansen, C., Kjær, S. K., Steding-Jessen, M., Storm, H. H. and Olsen, J. H. (2008). ‘Social inequality in incidence of and survival from cancer in a population-based study in Denmark, 1994–2003: summary of findings’. European Journal of Cancer, 44 (14), pp. 2074–2085. doi: 10.1016/j.ejca.2008.06.018. Dalton, S. O., Steding-Jessen, M., Engholm, G., Schüz, J. and Olsen, J. H. (2008). ‘Social inequality and incidence of and survival from lung cancer in a population-based study in Denmark, 1994–2003’. European Journal of Cancer, 44 (14), pp. 1989–1995. doi: 10.1016/j.ejca.2008.06.023. Dannefer, D. (1987). ‘Aging as intracohort differentiation: accentuation, the Matthew effect, and the life course’. Sociological Forum, 2 (2), pp. 211–236. doi: 10.1007/BF01124164. Dannefer, D. (2003). ‘Cumulative advantage/disadvantage and the life course: cross-fertilizing age and social science theory’. The Journals of Gerontology Series B: Psychological Sciences and Social Sciences, 58 (6), pp. S327–S337. Dowd, J. and Bengtson, V. (1978). ‘Aging in minority populations: an examination of the double jeopardy hypothesis’. Journal of Gerontology, 33 (3), pp. 427–436. doi: 10.1093/geronj/33.3.427. Dupre, M. E. (2007). ‘Educational differences in age-related patterns of disease: reconsidering the cumulative disadvantage and age-as-leveler hypotheses’. Journal of Health and Social Behavior, 48 (1), pp. 1–15. doi: 10.1177/002214650704800101. Elo, I. T. and Preston, S. H. (1996). ‘Educational differentials in mortality: United States, 1979–1985’. Social Science & Medicine, 42 (1), pp. 47–57. doi: 10.1016/0277-9536(95)00062-3. Erikson, R. and Goldthorpe, J. H. (1992). The constant flux: a study of class mobility in industrial societies. Oxford: Clarendon Press.

324  Handbook of health inequalities across the life course Ferraro, K. F. (1987). ‘Double jeopardy to health for black older adults?’ Journal of Gerontology, 42 (5), pp. 528–533. doi: 10.1093/geronj/42.5.528. Ferraro, K. F. and Farmer, M. M. (1996). ‘Double jeopardy, aging as leveler, or persistent health inequality? A longitudinal analysis of white and black Americans’. The Journals of Gerontology: Series B, 51B (6), pp. S319–S328. doi: 10.1093/geronb/51B.6.S319. Ferraro, K. F. and Morton, P. M. (2016). ‘What do we mean by accumulation? Advancing conceptual precision for a core idea in gerontology’. The Journals of Gerontology Series B: Psychological Sciences and Social Sciences, 73 (2), pp. 269–278. doi: 10.1093/geronb/gbv094. Ferraro, K. F. and Shippee, T. P. (2009). ‘Aging and cumulative inequality: how does inequality get under the skin?’ The Gerontologist, 49 (3), pp. 333–343. doi: 10.1093/geront/gnp034. George, L. K., Okun, M. A. and Landerman, R. (1985). ‘Age as a moderator of the determinants of life satisfaction’. Research on Aging, 7 (2), pp. 209–233. doi: 10.1177/0164027585007002004. Gjonca, E., Tabassum, F. and Breeze, E. (2009). ‘Socioeconomic differences in physical disability at older age’. Journal of Epidemiology & Community Health, 63 (11), pp. 928–935. doi: 10.1136/ jech.2008.082776. Graham, H. (2007). Unequal lives: health and socioeconomic inequalities. McGraw-Hill Education (UK). Herd, P. (2006). ‘Do functional health inequalities decrease in old age?’ Research on Aging, 28 (3), pp. 375–392. doi: 10.1177/0164027505285845. Hoffmann, R. (2008). Socioeconomic differences in old age mortality. Berlin: Springer (Springer series on demographic methods and population analysis, 25). Hoffmann, R. (2011). ‘Illness, not age, is the leveler of social mortality differences in old age’. The Journals of Gerontology Series B: Psychological Sciences and Social Sciences, 66 (3), pp. 374–379. doi: 10.1093/geronb/gbr014. Hoffmann, R., Kröger, H. and Pakpahan, E. (2018). ‘Pathways between socioeconomic status and health: Does health selection or social causation dominate in Europe?’ Advances in Life Course Research, 36, pp. 23–36. doi: 10.1016/j.alcr.2018.02.002. House, J. S., Kessler, R. C. and Herzog, A. R. (1990). ‘Age, socioeconomic status, and health’. The Milbank Quarterly, 68 (3), pp. 383–411. doi: 10.2307/3350111. House, J. S., Lepkowski, J. M., Kinney, A. M., Mero, R. P., Kessler, R. C. and Herzog, A. R. (1994). ‘The social stratification of aging and health’. Journal of Health and Social Behavior, 35 (3), pp. 213–234. doi: 10.2307/2137277. Houweling, T. A. and Kunst, A. E. (2010). ‘Socio-economic inequalities in childhood mortality in low-and middle-income countries: a review of the international evidence’. British Medical Bulletin, 93 (1), pp. 7–26. doi: 10.1093/bmb/ldp048. Hu, Y., Leinonen, T., Myrskylä, M. and Martikainen, P. (2020). ‘Changes in socioeconomic differences in hospital days with age: cumulative disadvantage, age-as-leveler, or both?’ The Journals of Gerontology: Series B, 75 (6), pp. 1336–1347. doi: doi​.org/​10​.1093/​geronb/​gbx161. Huisman, M. (2004). ‘Socioeconomic inequalities in mortality among elderly people in 11 European populations’. Journal of Epidemiology & Community Health, 58 (6), pp. 468–475. doi: 10.1136/ jech.2003.010496. Huisman, M., Kunst, A. E. and Mackenbach, J. P. (2003). ‘Socioeconomic inequalities in morbidity among the elderly; a European overview’. Social Science & Medicine, 57 (5), pp. 861–873. doi: 10.1016/S0277-9536(02)00454-9. Human Mortality Database. (2022). University of California, Berkeley and Max Planck Institute for Demographic Research. Available at: http://​www​.mortality​.org/​(Accessed: 23 February 2022). Idler, E. L. (1993). ‘Age differences in self-assessments of health: age changes, cohort differences, or survivorship?’ Journal of Gerontology, 48 (6), pp. S289–S300. doi: 10.1093/geronj/48.6.S289. Kaeberlein, M. (2013). ‘Longevity and aging’. F1000Prime Reports, 5. doi: 10.12703/P5-5. Kim, J. and Durden, E. (2007). ‘Socioeconomic status and age trajectories of health’. Social Science & Medicine, 65 (12), pp. 2489–2502. doi: 10.1016/j.socscimed.2007.07.022. King, N. B., Harper, S. and Young, M. E. (2012). ‘Use of relative and absolute effect measures in reporting health inequalities: structured review’. BMJ, 345, p. e5774. doi: 10.1136/bmj.e5774. Korda, R. J., Paige, E., Yiengprugsawan, V., Latz, I. and Friel, S. (2014). ‘Income-related inequalities in chronic conditions, physical functioning and psychological distress among older people in

How health inequalities evolve as we age  325 Australia: cross-sectional findings from the 45 and up study’. BMC Public Health, 14 (1), p. 741. doi: 10.1186/1471-2458-14-741. Leopold, L. (2016). ‘Cumulative advantage in an egalitarian country? Socioeconomic health disparities over the life course in Sweden’. Journal of Health and Social Behavior, 57 (2), pp. 257–273. doi: 10.1177/0022146516645926. Leopold, L. (2019). ‘Health measurement and health inequality over the life course: a comparison of self-rated health, SF-12, and grip strength’. Demography, 56 (2), pp. 763–784. doi: 10.1007/ s13524-019-00761-x. Leopold, L. and Engelhartdt, H. (2013). ‘Education and physical health trajectories in old age. Evidence from the Survey of Health, Ageing and Retirement in Europe (SHARE)’. International Journal of Public Health, 58 (1), pp. 23–31. doi: 10.1007/s00038-012-0399-0. Li, X., Ploner, A., Wang, Y., Magnusson, P. K., Reynolds, C., Finkel, D., Pedersen, N. L., Jylhävä, J. and Hägg, S. (2020). ‘Longitudinal trajectories, correlations and mortality associations of nine biological ages across 20-years follow-up’. Elife. eLife Sciences Publications Limited, 9, p. e51507. doi: 10.7554/eLife.51507. Liang, J., Bennett, J., Krause, N., Kobayashi, E., Kim, H., Brown, J. W., Akiyama, H., Sugisawa, H. and Jain, A. (2002). ‘Old age mortality in Japan: does the socioeconomic gradient interact with gender and age?’ The Journals of Gerontology: Series B, 57 (5), pp. S294–S307. doi: 10.1093/geronb/57.5.S294. Link, B. G. and Phelan, J. (1995). ‘Social conditions as fundamental causes of disease’. Journal of Health and Social Behavior. pp.  80–94. doi: doi​.org/​10​.2307/​2626958. Mackenbach, J. P. (2019). Health inequalities: persistence and change in modern welfare states. New York: Oxford University Press. Mackenbach, J. P., Kunst, A. E., Cavelaars, A. E., Groenhof, F. and Geurts, J. J. (1997). ‘Socioeconomic inequalities in morbidity and mortality in western Europe’. The Lancet, 349 (9066), pp. 1655–1659. doi: 10.1016/S0140-6736(96)07226-1. Mackenbach, J. P., Martikainen, P., Menvielle, G. and de Gelder, R. (2016). ‘The arithmetic of reducing relative and absolute inequalities in health: a theoretical analysis illustrated with European mortality data’. Journal of Epidemiology and Community Health, 70 (7), pp. 730–736. doi: 10.1136/ jech-2015-207018. Mackenbach, J. P., Rubio Valverde, J., Bopp, M., Brønnum-Hansen, H., Costa, G., Deboosere, P., Kalediene, R., Kovács, K., Leinsalu, M., Martikainen, P., Menvielle, G., Rodriguez-Sanz, M. and Nusselder, W. J. (2019). ‘Progress against inequalities in mortality: register-based study of 15 European countries between 1990 and 2015’. European Journal of Epidemiology, 34 (12), pp. 1131–1142. doi: 10.1007/s10654-019-00580-9. Mather, T., Banks, E., Joshy, G., Bauman, A., Phongsavan, P. and Korda, R. J. (2014). ‘Variation in health inequalities according to measures of socioeconomic status and age’. Australian and New Zealand Journal of Public Health, 38 (5), pp. 436–440. doi: 10.1111/1753-6405.12239. McMunn, A., Nazroo, J. and Breeze, E. (2008). ‘Inequalities in health at older ages: a longitudinal investigation of the onset of illness and survival effects in England’. Age and Ageing, 38 (2), pp. 181–187. doi: 10.1093/ageing/afn236. Merton, R. K. (1968). ‘The Matthew effect in science: the reward and communication systems of science are considered’. Science, 159 (3810), pp. 56–63. doi: 10.1126/science.159.3810.56. Mortensen, L. H., Rehnberg, J., Dahl, E., Diderichsen, F., Elstad, J. I., Martikainen, P., Rehkopf, D., Tarkiainen, L. and Fritzell, J. (2016). ‘Shape of the association between income and mortality: a cohort study of Denmark, Finland, Norway and Sweden in 1995 and 2003’. BMJ Open, 6 (12), p. e010974. doi: 10.1136/bmjopen-2015-010974. National Urban League. (1964). Double jeopardy: the older negro in America today. National Urban League. Available at: https://​books​.google​.se/​books​?id​=​o2SPHwAACAAJ. OECD. (2019). ‘Old-age income inequality’. In Pensions at a Glance 2019. OECD, pp. 190–191. doi: 10.1787/b00d8f0e-en. O’Rand, A. M. (1996). ‘The precious and the precocious: understanding cumulative disadvantage and cumulative advantage over the life course’. The Gerontologist, 36 (2), pp. 230–238. doi: 10.1093/ geront/36.2.230.

326  Handbook of health inequalities across the life course Øvrum, A., Gustavsen, G. W. and Rickertsen, K. (2014). ‘Age and socioeconomic inequalities in health: examining the role of lifestyle choices’. Advances in Life Course Research. Elsevier, 19, pp. 1–13. doi: 10.1016/j.alcr.2013.10.002. Poole, C. (2010). ‘On the origin of risk relativism’. Epidemiology, 21 (1), pp. 3–9. Doi: 10.1097/ EDE.0b013e3181c30eba. Preston, S. H. (1984). ‘Children and the elderly: divergent paths for America’s dependents’. Demography, 21 (4), pp. 435–457. Doi: 10.2307/2060909. Rehnberg, J. (2020). ‘What levels the association between income and mortality in later life: age or health decline?’ The Journals of Gerontology: Series B, 75 (2), pp. 426–435. Doi: 10.1093/geronb/gbz082. Rehnberg, J., Fors, S. and Fritzell, J. (2019). ‘Divergence and convergence: how do income inequalities in mortality change over the life course?’ Gerontology, 65 (3), pp. 313–322. Doi: 10.1159/000494082. Rehnberg, J., Östergren, O., Esser, I. and Lundberg, O. (2021). ‘Interdependent pathways between socioeconomic position and health: a Swedish longitudinal register-based study’. Social Science & Medicine, 280, p. 114038. Doi: 10.1016/j.socscimed.2021.114038. Reques, L., Santos, J. M., Belza, M. J., Martínez, D. and Regidor, E. (2015). ‘Inequalities in mortality at older ages decline with indicators of material wealth but persist with educational level’. European Journal of Public Health, 25 (6), pp. 990–995. Doi: 10.1093/eurpub/ckv110. Robitaille, A., Hout, A., Machado, R. J. M., Bennett, D. A., Čukić, I., Deary, I. J., Hofer, S. M., Hoogendijk, E. O., Huisman, M., Johansson, B., Koval, A. V., Noordt, M., Piccinin, A. M., Rijnhart, J. J. M., Singh‐Manoux, A., Skoog, J., Skoog, I., Starr, J., Vermunt, L., Clouston, S. and Muniz Terrera, G. (2018). ‘Transitions across cognitive states and death among older adults in relation to education: A multistate survival model using data from six longitudinal studies’. Alzheimer’s & Dementia, 14 (4), pp. 462–472. Doi: 10.1016/j.jalz.2017.10.003. Ross, C. E. and Wu, C.-L. (1996). ‘Education, Age, and the Cumulative Advantage in Health’. Journal of Health and Social Behavior, 37 (1), pp. 104–120. Doi: 10.2307/2137234. Rowe, J. and Kahn, R. (1987). ‘Human aging: usual and successful’. Science, 237 (4811), pp. 143–149. Doi: 10.1126/science.3299702. Shaw, B. A., McGeever, K., Vasquez, E., Agahi, N. and Fors, S. (2014). ‘Socioeconomic inequalities in health after age 50: are health risk behaviors to blame?’ Social Science & Medicine, 101, pp. 52–60. Doi: 10.1016/j.socscimed.2013.10.040. Suzman, R. and Riley, M. W. (1985). ‘Introducing the “oldest old”’. The Milbank Memorial Fund Quarterly. Health and Society, 63 (2), pp. 177–186. Vågerö, D. and Erikson, R. (1997). ‘Correspondence: Socioeconomic inequalities in morbidity and mortality in western Europe’. The Lancet, 350 (9076), p. 516. Doi: 10.1016/S0140-6736(97)26033-2. Van Kippersluis, H., O’Donnell, O., van Doorslaer, E. And Van Ourti, T. (2010). ‘Socioeconomic differences in health over the life cycle in an egalitarian country’. Social Science & Medicine, 70 (3), pp. 428–438. Doi: 10.1016/j.socscimed.2009.10.020. Vaupel, J. W. and Yashin, A. I. (1985). ‘Heterogeneity’s ruses: some surprising effects of selection on population dynamics’. The American Statistician, 39 (3), pp. 176–185. Doi: 10.1080/00031305.1985.10479424. Willson, A. E., Shuey, K. M. and Elder, Jr., Glen H. (2007). ‘Cumulative advantage processes as mechanisms of inequality in life course health’. American Journal of Sociology, 112 (6), pp. 1886–1924. Doi: 10.1086/512712. Xu, X., Liang, J., Bennett, J. M., Botoseneanu, A. and Allore, H. G. (2015). ‘Socioeconomic stratification and multidimensional health trajectories: evidence of convergence in later old age’. The Journals of Gerontology Series B: Psychological Sciences and Social Sciences, 70 (4), pp. 661–671. Doi: 10.1093/geronb/gbu095. Zajacova, A., Montez, J. K. and Herd, P. (2014). ‘Socioeconomic disparities in health among older adults and the implications for the retirement age debate: a brief report’. The Journals of Gerontology Series B: Psychological Sciences and Social Sciences, 69 (6), pp. 973–978. Zajacova, A. and Woo, H. (2016). ‘Examination of Age variations in the predictive validity of self-rated health’. The Journals of Gerontology Series B: Psychological Sciences and Social Sciences, 71 (3), pp. 551–557. doi: 10.1093/geronb/gbv050.

21. Environmental inequality and health outcomes over the life course Christian König and Jan Paul Heisig

INTRODUCTION Environmental inequalities – persistent social inequalities in exposure to environmental bads such as air pollution or noise and to environmental goods such as green spaces – are increasingly recognized as potentially important sources of social inequalities in health and other major life outcomes. The main goals of the present chapter are to review conceptual and empirical work on the health impacts of environmental quality and on the social processes that lead to environmental inequality as defined above. We begin by embedding our review in a life-course framework (Ben-Shlomo and Kuh, 2002; Kuh et al., 2003) that allows for interdependent and dynamic linkages between environmental quality, health, and other life domains (e.g., educational attainment). Life-course approaches suggest that the effects of early-life insults, including exposure in utero, may not be confined to their often severe short-run consequences (Barker and Osmond, 1986). Early-life experiences might additionally trigger processes that lead to manifest disease (much) later in life, after long periods of latency – possibly reinforced by processes of cumulative disadvantage (DiPrete and Eirich, 2006). The main part of the chapter reviews empirical evidence for a (causal) effect of environmental quality on health. We begin with the health impacts of air pollution, the most widely studied aspect of environmental quality. We discuss some of the most obvious threats to causal identification and review a by now large body of literature that has exploited quasi-experimental variation to provide compelling evidence for substantial causal effects of air pollution on health. Our extensive discussion of research on the effects of air pollution is complemented by shorter sections on the health impacts of extreme heat, green space presence, and noise. While our review covers a sizable literature, we do have to be somewhat selective. We encourage readers to consult excellent earlier reviews on the health effects of air pollution (Pope and Dockery, 2006; Currie, 2013; Graff Zivin and Neidell, 2013; Currie et al., 2014; Kelly and Fussell, 2020) and green spaces (Kondo et al., 2018) that we build upon and extend. We also note that we focus on evidence from high-income countries where environmental regulation is rather strict and where levels of pollution tend to be low by international standards. Our review demonstrates that the health impacts of air pollution and environmental influences more broadly can be substantial, even under these comparably favorable conditions. The final sections of the chapter turn towards the extent and sources of social inequalities in residential environmental quality. Research on the health effects of environmental quality treats residential sorting as a nuisance that complicates causal identification: it represents a form of selection into treatment. From a social stratification and health equity perspective, sorting constitutes a major social problem and a phenomenon of substantive interest. Why are some social groups more likely to move into or continue living in unhealthy environments than 327

328  Handbook of health inequalities across the life course others? We briefly review the most prominent answers to this question, including financial constraints, housing and credit market discrimination, as well as “legacy effects” due to the historical clustering of (minority) population groups in certain areas. We also present German data that highlight the importance of fine-grained spatial data for capturing the full extent of environmental inequality. The chapter concludes with a summary and some open questions.

A LIFE-COURSE PERSPECTIVE ON ENVIRONMENTAL AND HEALTH INEQUALITIES Life-course perspectives (Ben-Shlomo and Kuh, 2002; Kuh et al., 2003) highlight a number of general explanations and explanatory mechanisms for health inequalities, including critical developmental periods and long-term effects of early-life experiences, the dynamic interplay of biological and social factors, as well as processes of cumulation and cumulative disadvantage (see Figure 21.1 for a stylized graphical representation). An influential example of an explanation emphasizing long-term effects is the fetal origins hypothesis (Barker, 1995), the idea that insults to health experienced in utero can lead to a fundamental “reprogramming” that puts the organism on a path towards later-life disease. The related literature on adverse childhood experiences (Hughes et al., 2017) emphasizes the lasting impact of childhood adversity (e.g., poverty, abuse, or violence) on health in adulthood. While the strong version of the fetal origins hypothesis is distinct from explanations emphasizing childhood experiences in that it implies a discontinuity at the time of birth (Almond and Currie, 2011), both rest on a critical period model assuming that early insults to health can have lasting – sometimes irreversible – effects on health and other life outcomes through a variety of both biological (e.g., physiological or epigenetic) and social pathways (Ben-Shlomo and Kuh, 2002). The characterization of early (prenatal and postnatal) life as “critical” further suggests that adverse exposures during these periods tend to have stronger effects on later outcomes than comparable experiences during later stages of the life course (Ben-Shlomo and Kuh, 2002). A similar idea is that developmental capacities may be confined to certain “developmental windows” and that delayed and/or impaired growth during these windows is difficult to compensate for or reverse after a certain point (Andersen, 2003). A related concept is that of latency. While some health effects of early-life insults may become apparent more or less immediately, others may remain latent for long periods, as with cardiovascular disease and other health conditions that typically do not manifest themselves until middle age (Almond and Currie, 2011). Long latency does not imply that early-life insults have no effects in the short term. It could even be argued that some changes (e.g., on a physiological or epigenetic level) have to occur immediately in order to pave the way for measurable impacts on health later in life. The notion of latency – represented by the dashed arrows in Figure 21.1 – rather emphasizes the idea that these alterations do not result in manifest disease right away but only after a certain (possibly long) period of time. Whether initial insults eventually result in pathology may depend on the presence of further stressors over the subsequent life course, that is, on whether life courses follow a pattern of cumulative disadvantage (DiPrete and Eirich, 2006). A specific form that cumulative disadvantage may take is that of spillover and feedback effects between different life domains. Environmental insults early in life may cause direct short- and long-term harm to health, but they may also do so indirectly by affecting cognitive development, education, employment,

Environmental inequality and health outcomes over the life course  329 or income, which subsequently influence health (Zhang, Chen and Zhang, 2018).1 Cumulative disadvantage in this sense can be conceptualized as a twofold process consisting of latent health but also human capital effects that further affect later-life health. Crucially, both mechanisms can be assumed to cluster in socially patterned ways (Ben-Shlomo and Kuh, 2002), that is, to vary according to socio-economic status (SES).2

Figure 21.1

Environmental inequalities and health over the life course

Moving from these general considerations to possible impacts of environmental quality, there are many reasons why children might be particularly susceptible to (positive and negative) environmental influences from air pollution, green space presence, or heat. First, children’s bodily systems are not yet fully developed making them particularly sensitive to environmental conditions. For example, the respiratory system grows and develops rapidly during early childhood, which may render it more vulnerable to assaults, with limited ability to fully repair after disruption of the morphogenesis (Bateson and Schwartz, 2007). Children’s immune systems are immature, which does play an important role in the development of diseases such as asthma (Schwartz, 2004; Bateson and Schwartz, 2007). Similarly, children’s thermoregulatory systems have not yet fully developed, resulting in lower capacity to adapt to extreme temperatures (Xu et al., 2014). Second, children physiologically differ from adults in other ways relevant to susceptibility. Their baseline ventilation rates are higher, thus exposing their lungs to more air pollution even under similar conditions. They also have a greater body surface area-to-mass ratio compared to adults, allowing greater heat and cold transfer between the environment and the body (Xu et al., 2014). Finally, children may be disproportionately affected by outdoor conditions as they spend more time outside and are physically more active than adults.

330  Handbook of health inequalities across the life course

ADVERSE HEALTH EFFECTS OF AIR POLLUTION We now review empirical work on the health effects of air pollution. After discussing associational epidemiological work and selected toxicological studies, we turn to key problems of causal identification and to research that tries to overcome these challenges with quasi-experimental designs. We show that empirical evidence is broadly consistent with critical period models in that the most severe short-term effects of pollution exposure are found for young children (including those exposed in utero) and for older people with pre-existing conditions. Evidence for long-term effects of early pre- and/or postnatal exposure seems to be growing as well, although compelling quasi-experimental studies remain scarce, partly due to demanding data requirements. Epidemiological and Toxicological Evidence Much of our understanding about the relationship between exposure to air pollution and health stems from epidemiological and toxicological studies. The list of air pollutants is long and includes compounds such as carbon monoxide (CO), ground-level ozone (O3), lead (Pb), and nitrogen (NOx) or sulphur oxides (SOx). Perhaps the most widely studied aspect of air quality is particulate matter (PM) concentration, usually measured in microgram per cubic meter (μg/ m³) and typically classified into fine particulate matter with particle diameters below 2.5 μm (PM2.5) and coarse particulate matter with diameters below 10 μm (PM10). Epidemiological studies have established strong and robust associations between ambient air pollution and various health outcomes. Particularly strong links have been found for respiratory and cardiovascular morbidity, including reduced lung function and severity of symptoms in individuals with asthma and chronic obstructive pulmonary disease (COPD) as well as ischemic heart disease (Brook et al., 2010; Kelly and Fussell, 2011; WHO, 2013). Temporary spikes in air pollution have been linked to increased hospital admissions. Using data from Chicago area hospitals between 1988 and 1993, Schwartz (2001) found an increase in PM10 by 10 μg/m³ to be associated with 2.00 percent, 1.45 percent, and 1.27 percent increases in admissions for pneumonia, chronic obstructive pulmonary disease, and heart disease, respectively (see also Zanobetti, Schwartz and Dockery, 2000). With regard to mortality, Ostro, Hurley, and Lipsett (1999) found that an increase in PM10 concentration by 10 μg/m³ in Coachella Valley, California, was associated with a 1 percent increase in daily all-cause mortality. An analysis of ten large US cities similarly indicated that heightened PM10 concentration (10 μg/m³) was associated with a 0.67 percent increase in total mortality and a 0.89 percent increase in out-of-hospital mortality (Schwartz, 2000). For the European context, Katsouyanni et al. (2001) demonstrated that PM10 and black smoke predicted daily mortality in 29 countries. In addition to these short-term associations, prospective cohort studies with long-term follow-up periods have documented robust links between levels of air pollution and lung cancer, cardiopulmonary, and all-cause mortality. One of the most comprehensive early studies – the American Cancer Society study – linked individual health risks of residents from approximately 150 US cities with data on ambient air pollution and found a risk ratio of 1.17 for all-cause mortality between the areas most and least affected by fine particulates (Pope et al., 1995; see also Dockery et al., 1993). Results of a subsequent follow-up study, observing death or survival of participants over more than 16 years (1982–1998), indicated that the risk

Environmental inequality and health outcomes over the life course  331 of dying from lung cancer and cardiopulmonary diseases increased by 8 percent and 6 percent, respectively, for each 10 μg/m³ increase in PM2.5 (Pope, 2002). Several groups have been found to be at higher risk for air-pollution-induced cardiovascular morbidity and mortality, including individuals with pre-existing cardiovascular conditions, people with diabetes, and elderly individuals (Zanobetti and Schwartz, 2002; Devlin et al., 2003). As many relevant conditions are more prevalent among low-SES populations – due in part to long-term exposure to pollution itself – these differential vulnerabilities tend to amplify social inequalities in health and can be interpreted as examples of cumulative disadvantage (McNamara et al., 2017). Epidemiological evidence is bolstered by research on biological and physiological responses to air pollution exposure in humans and non-human species (Salvi et al., 1999). Bové et al. (2019) show that black carbon particles stemming from combustion-derived PM pass the human placental barrier, elucidating a potential mechanism for fetal origins of later-life disease. They further find the black carbon load from placentae to be positively associated with residential black carbon exposure of mothers throughout pregnancy (see Luyten et al., 2018 for a review of placental markers associated with prenatal air pollution exposure). A number of studies on adults have examined immediate effects of pollution exposure on humans, for example, by exposing human volunteers to diesel exhaust for 1–2h (Salvi et al., 1999, 2000; Stenfors et al., 2004; Behndig et al., 2006). These studies provide compelling evidence of systemic and pulmonary inflammatory responses following even short-term exposure to toxic pollutants. Toxicological examinations have shown that exposure to air pollution leads to changes in many cardiovascular indicators, with some (e.g., in heart rate, heart rate variability, blood pressure, vascular tone, and blood coagulability) occurring more or less immediately, and others, such as accelerated progression of arthrosclerosis, after more prolonged exposure (Simkhovich, Kleinman and Kloner, 2008). While such toxicological approaches provide valuable evidence on biological and physical pathways, they also have clear limitations. Studies involving long-term or high levels of exposure of humans are ethically indefensible, as are studies of high-risk populations. Such concerns do not loom quite as large in studies of nonhuman animal species (e.g., Sun et al., 2005), but their results cannot necessarily be extrapolated to humans. Analysis of observational data on real-world exposures, as in the epidemiological literature discussed above, is therefore indispensable for gauging the full impact of air pollution on human health and health inequalities. Such analysis is complicated by well-known threats to causal identification in observational studies, and this is where the social sciences have made significant contributions through the application of quasi-experimental approaches (Currie et al., 2014). Challenges to Identification As in practically any observational study, a general concern in estimating the health impacts of air pollution is selection on unobservables. Greater exposure to air pollution might go hand in hand with other individual and environmental risk factors, including socio-economic deprivation, unhealthy behaviors, work-related health hazards, or lack of green spaces. Failure to adequately control for such confounding factors will lead to biased estimates of pollution exposure effects.

332  Handbook of health inequalities across the life course In the context of environmental inequality, selection on unobservables is often linked to residential sorting. A growing body of literature indicates that sorting based on environmental quality is a key factor shaping residential segregation. For example, Banzhaf and Walsh (2008) show that high-income families tend to move away from polluted areas (see also Best and Rüttenauer, 2018). In addition, research highlights the tendency of potentially health-relevant neighborhood characteristics such as crime, school quality, and public infrastructure more broadly to cluster in space (Alba and Logan, 1993), suggesting possible confounding at the local contextual level. From the viewpoint of identifying the health effects of air pollution and other dimensions of environmental inequality, residential sorting is a nuisance that complicates causal identification. In the remainder of this chapter, we will initially adopt this perspective and focus on studies that seek to address the resulting selectivities through clever (quasi-experimental) designs. Yet residential sorting clearly is a social process that is of utmost substantive interest, a point that we emphasize when turning towards the mechanisms underlying environmental inequality in the later parts of the chapter. Avoidance behavior is another potential source of bias. People may try to protect themselves and their children– temporarily or persistently – against poor environmental conditions, for example, by spending more time indoors (Neidell, 2009; Currie et al., 2014). In the presence of avoidance behavior, measures of ambient air pollution will overstate actual differences in exposure. Avoidance behavior thus results in the underestimation of health effects, in the sense that the latter would be even larger under unmitigated exposure to measured differentials in outdoor pollution (Neidell, 2009). Another interpretation is that a narrow focus on health outcomes misses some of the welfare costs of air pollution by ignoring the individual and social costs of avoidance behavior (Moretti and Neidell, 2009; Currie et al., 2014). Quasi-Experimental Approaches Given the ethical and practical unviability of randomized experiments, identification of plausibly causal effects of air pollution on health mostly comes from various types of quasi-experimental designs (see Morgan and Winship, 2007 for a thorough discussion; see also Chapters 7 and 9 by Gebel and by Hoffmann and Doblhammer in this volume). Plausibly exogenous variation in pollution concentration often stems from so-called natural experiments that lead to sudden, unexpected, and sometimes unobserved changes in environmental conditions. Important types of events include policy changes, changes in weather conditions (e.g., changes in wind directions or temperature inversions), and unexpected changes in industrial production or traffic intensity. In the absence of anticipation effects, such “shocks” can be argued to affect the same population just before and after their implementation, making distortions due to changes in population composition unlikely, at least in the short run. Whenever an event and its implications for pollution exposure are difficult to observe, avoidance and other changes in health-relevant behaviors do not pose major threats to identification either. Fixed-effects designs that exploit (short-term) variation in measured pollution within geo-spatial and administrative units follow a similar basic logic. The primary goal of such designs again is to rule out confounding from unobserved factors that might bias estimates based on between-unit comparisons (e.g., population composition or healthcare infrastructure). While this assumption is often plausible over short periods of time, it becomes more difficult to defend when the observation period spans several months or even years. Another limitation

Environmental inequality and health outcomes over the life course  333 of designs that take measured pollution as given, even when focusing on short-term variability, is that the sources of this variation are unclear and can include factors (e.g. fireworks or traffic congestion due to major sport events) that may be associated with other (health-relevant) behavioral changes. This is a potentially crucial difference to natural-experiment approaches that can render such changes unlikely by focusing on specific (weather) events. Another kind of fixed-effects design focuses on within-family variation, usually comparing the health outcomes of siblings who have experienced different environmental conditions in utero and in childhood, induced by families moving to another area or by changes in environmental quality over time. The strength of this approach is to control for time-invariant unobserved characteristics shared by children from the same family, including parental health behaviors or genetic predispositions. At the same time, there are obvious threats to a causal interpretation of within-family associations. For example, residential moves might go hand in hand with changes in numerous other health-relevant domains, not all of which may be adequately controlled in a given application (for a useful critical discussion of within-family designs, see Engzell and Hällsten, 2022). The designs discussed in this section address key limitations of simple associational and conditioning-on-observables approaches, yet they certainly are no panacea. Generally, the more time elapses after the onset of a (plausibly exogenous) event, the greater the scope for residential sorting to complicate the picture (Currie et al., 2014). When natural experiments are based on unexpected shocks or policy changes, these events can also directly affect health through other mechanisms. Shocks to industrial productivity due to recessions, for example, have been used to investigate the effects of decreased industrial emissions of air pollutants (Sanders, 2012). However, economic recessions may affect health through other channels (e.g. loss of employment and/or health insurance) that, if not accounted for, can result in biased estimates. Finally, approaches exploiting temporary weather events or short-term spatiotemporal variability more generally are not well-suited for identifying health effects that have long latency, nor for studying the impact of long-term (accumulated) exposure. With these caveats in mind, we now turn to a review of quasi-experimental evidence on the health impacts of air pollution. Short-Term Variability in Exposure and Immediate Health Consequences Deryugina et al. (2019) exploit variation in local wind direction at the US county level to investigate adverse health effects of air pollution on US Medicare beneficiaries. They find that a 1 μg/m³ increase in PM2.5 leads to 2.7 additional Emergency Room (ER) visits and 0.69 additional deaths per million beneficiaries over the three-day window that spans the day of the increase and the following two days. Jans, Johansson and Nilsson (2018) examine whether poor air quality affects Swedish children’s respiratory health, exploiting variation in air quality induced by night-time temperature inversions.3 They find that temperature inversions temporarily raise PM10 and NO2 concentrations by 25 and 16 percent, respectively, and that they lead to increases in health care visits due to respiratory illness of around 5.5 percent. Low-income children are more strongly affected, which seems to be primarily due to differences in baseline health status and preexisting conditions. These effects are unlikely to be confounded by avoidance behavior, as (night-time) temperature inversions usually cannot be observed with the naked eye and are rarely reported in the media.

334  Handbook of health inequalities across the life course Dominici et al. (2006) use a design with local area fixed effects to investigate short-term health effects of air pollution using a database comprising daily information on hospital admissions of Medicare enrollees (aged > 65 years), ambient PM2.5 levels, and weather controls for 204 US urban counties. They find short-term increases in hospital admission for all health outcomes under study except for injuries. Effects are largest for heart failure, respiratory tract infections, and COPD with statistically significant estimated increases of 1.28, 0.92, and 0.91 percent per 10 μg/m³ increase in PM2.5. Ground-level ozone (O3) concentration is another aspect of air quality that has received quite a bit of attention. Neidell (2009) shows that accounting for avoidance behavior drastically increases estimated health effects of ozone on vulnerable groups – by roughly 160 percent for children and 40 percent for the elderly – while having no effect on impact estimates for other adults. Moretti and Neidell (2009) use boat traffic at the port of Los Angeles – a major source of nitrogen dioxide (NO2), which forms O3 when reacting with sunlight – as an instrumental variable (IV) in estimating effects of O3 on respiratory-related hospitalizations. Simple OLS yields a statistically significant but relatively small estimate of $11.1 million ozone-related hospital costs per year. The IV estimate is roughly four times as large ($44.5 million per year). Medium- to Long-Term Health Consequences of Air Pollution Policy changes and other events that lead to sudden and persistent changes in environmental conditions can be exploited to investigate longer-term consequences of air pollution exposure. However, this goes hand in hand with challenges to causal identification such as changes in the composition of the affected populations. Currie and Walker (2011) use the introduction of an electronic toll collection scheme (called E-ZPass) across the northeastern United States around the year 2000. E-ZPass reduced delays and traffic congestion at toll plazas, thereby leading to local declines in pollution caused by decelerating, idling, and accelerating. Currie and Walker (2011) compare health outcomes of infants born to mothers living near toll plazas (0–2 km) to those born to mothers living farther away (2–10 km) but close to major highways (0–3 km), starting three to six years prior to the implementation of the E-ZPass until one to two years thereafter. Difference-in-difference models suggest that the incidence of low birth weight fell by 8.5–11.3 percent among infants born to mothers within 2 km of a toll plaza, while prematurity fell by 6.7–9.2 percent (Currie and Walker, 2011). Housing prices and mothers’ observable characteristics remained largely unchanged around the adoption of the E-ZPass, assuaging concerns about residential sorting. Following a very similar logic, several studies have investigated health effects of newly established low emission zones4 (LEZs) in urban areas. Gehrsitz (2017) studies the impact of the staggered introduction of LEZs in multiple German cities and finds modest average reductions in air pollution of approximately 2.5 percent for PM10, with larger effects in cities that implemented stricter policies. Gehrsitz reports a moderate negative and statistically significant effect on stillbirths but none for average birthweight or the prevalence of low-weight births. Another German study by Margaryan (2021) finds similar reductions of PM10 levels and a 2–3 percent reduction in the number of patients with cardiovascular disease based on outpatient care data, with stronger effects for cerebrovascular disease (e.g., stroke) and for older people aged 65 and above. Studies looking at the introduction of an LEZ in London, United Kingdom, in February 2008 have found that the initial introduction had little to no effect on air pollution and health outcomes (Wood et al., 2015; Mudway et al., 2019).

Environmental inequality and health outcomes over the life course  335 Another widely studied policy change is the 1970 Clean Air Act Amendments (CAAA) which required so-called “nonattainment counties” with total suspended particulate (TSP) levels above threshold values to reduce air pollution, while “attainment counties” below the threshold were not required to do so. Chay and Greenstone (2003) use the nonattainment status of counties to instrument for changes in TSP between 1971 and 1972. They find that nonattainment status was indeed associated with subsequent reductions in air pollution and that a 1 percent reduction in TSP results in a 0.5 percent decline in overall infant mortality rate at the county level. Using a similar empirical approach, Chay, Dobkin and Greenstone (2003) find no systematic effect of reduced pollution on adult and elderly mortality rates, however. In a more recent paper focusing on labor market outcomes, Isen, Rossin-Slater, and Walker (2017) study the long-term consequences of pollution exposure in utero by comparing individuals born just before and just after the mandated improvements in air quality in nonattainment counties, using individuals born in attainment counties as a control group (difference-in-differences approach). They find that regulation-induced improvements in air quality are associated with a 0.7 percent increase in the number of quarters worked and a 1 percent increase in mean annual earnings at age 30. Their study is a so far rare case of a quasi-experimental study that provides evidence for long-term effects of in utero exposure. Applying a sibling design, Currie, Neidell and Schmieder (2009) investigate health effects of O3, CO, and PM10 on birth outcomes and infant mortality in New Jersey, USA, in the 1990s. Their models include pollution monitor and maternal fixed effects to account for time-invariant characteristic of neighborhoods and mothers. Results indicate that a one-unit rise in average CO parts per million (ca. 1.27 standard deviations) during the last trimester of gestation increases the risk of low birth weight by 8 percent. Additionally, a one-unit increase in mean CO during the first two weeks after birth increases the risk of infant mortality by 2.5 percent, conditional on health at birth. In a recent study, Alexander and Schwandt (2022) exploit spatial variation in the share of emissions-cheating Volkswagen “clean” diesel cars over the 2008–2015 period as a natural experiment. They estimate that an additional cheating diesel car per 1,000 cars increases PM2.5 concentrations by 2 percent, and low birth weight and infant mortality rates by 1.9 and 1.7 percent, respectively. The research reviewed in this section provides compelling evidence that air pollution has substantial adverse effects on health even in high-income countries where levels of pollution tend to be low by international standards. Short-term effects of pollution exposure are strongest among the old, partly as a result of preexisting conditions, for young children and in utero (as indicated by effects on birth weight and stillbirths). Extant research also points towards long-term effects of early-life exposure, although compelling quasi-experimental evidence remains rare, partly due to formidable data requirements (e.g., information on parental and childhood residential location). Nevertheless, empirical support for the fetal origins hypothesis seems to be growing, also in terms of mechanistic biological evidence. In the remaining sections, we extend our discussion in two main ways: The next section discusses selected dimensions of environmental quality – access to green space and exposure to noise or extreme heat – that have so far received less attention than air pollution. The subsequent section then focuses on environmental inequality as an explanation for health inequalities and on underlying processes of residential sorting and segregation.

336  Handbook of health inequalities across the life course

GOING BEYOND AIR POLLUTION: OTHER DIMENSIONS OF ENVIRONMENTAL QUALITY Green Space Proximity and access to green spaces, particularly in urban areas, is another aspect of environmental quality whose potential health-promoting effects have received increasing attention in recent years. While proximity to green spaces, an environmental “good”, is likely to correlate negatively with exposure to environmental “bads” such as pollution and noise, there are also plausible channels through which green spaces might independently affect both physical and mental health. For example, Markevych et al. (2017) argue that green space can affect health by restoring (e.g. stress recovery) and building capacities (e.g. encouraging physical exercise). In addition, negative correlations with environmental bads could partly be viewed as mediation (rather than confounding), for example, if the creation of a new park at the cost of road space leads to a reduction in traffic. So far, evidence on the health effects of green spaces mostly comes from conditioning-on-observables approaches that try to account for potential confounders with multiple regression and related approaches. This may also be because natural experiments involving clear temporal and/or spatial discontinuities in green space “exposure” are more difficult to find than in the case of ambient pollution. Extant studies report associations between exposure to natural environments and a wide range of physical and mental health outcomes, such as preterm birth (Abelt and McLafferty, 2017), childhood asthma (Donovan et al., 2018), student performance (Sivarajah, Smith and Thomas, 2018), anxiety and depression (Cohen-Cline, Turkheimer and Duncan, 2015), more aggregated indices of psychiatric disorders and mental health (Alcock et al., 2014), and respiratory as well as cancer mortality (James et al., 2016). In an exemplary and widely cited recent study, Engemann et al. (2019) combine satellite images with Danish population register data to investigate the association between green space presence in childhood (before age ten) and mental health in adolescence and young adulthood, controlling for multiple alternative explanations. High levels of green space presence in childhood are negatively associated with several psychiatric disorders. With regard to the underlying mechanisms, several studies have investigated the role of natural environments in recovering from psychological stress, relying on salivary and hair cortisol concentration as short- and long-term biomarkers (Ward Thompson et al., 2012; Roe et al., 2013; Gidlow et al., 2016). Results suggest that exposure to natural environments is indeed negatively correlated with short-term cortisol levels (Roe et al., 2013), although associations are not always robust to controls for confounding factors such as neighborhood deprivation (Gidlow et al., 2016). In an analysis of a small-scale intervention in the UK, Chalmin-Pui et al. (2021) find that adding plants to formerly bare front gardens led to lower perceived stress levels and improvements in diurnal cortisol patterns. Results concerning physical activity as another potential mediator between residential green space and health are mixed (Triguero-Mas et al., 2015; James et al., 2016). As noted above, clean natural experiments are more difficult to find for green space than for pollution exposure. At the same time, concerns about residential sorting and other threats to causal interpretation loom just as large as in the pollution case. Several studies have therefore tried to address concerns about unobserved confounding, mostly using (longitudinal) designs

Environmental inequality and health outcomes over the life course  337 with individual fixed effects. Using such an approach, both Alcock et al. (2014) and White et al. (2013) find positive effects of green space presence on mental health in the UK, while Noordzij et al. (2020) find no evidence for such a relationship using Dutch data. Even these longitudinal results have to be interpreted cautiously, however, as other neighborhood characteristics that co-vary with green space presence may not be fully accounted for. Cohen-Cline, Turkheimer and Duncan (2015) investigate the link between access to green space and self-reported depression, stress, and anxiety in a sample of same-sex monozygotic twin pairs. Pooled models show negative relationships between green space presence and all three mental health outcomes, but only the association with depression remains statistically significant in within-pair models, suggesting partial genetic confounding. Extreme Heat Extreme heat is another environmental factor with potentially severe consequences. Studies of high-income countries have so far mostly focused on countries and regions with a warm climate such as California, USA, or Australia. Climate change, however, will likely result in heat waves occurring more often, including in regions where they have been rare in the past (Koppe et al., 2004). We will not be able to review the sizable literature on the health effects of extreme heat in detail, but robust short-term association between heat waves and health have been documented for a variety of specific outcomes – including sunburn, heat stress, heat exhaustion, kidney failure, and heart attacks (Koppe et al., 2004; Guirguis et al., 2014) – as well as for broader measures of morbidity and mortality (Nitschke et al., 2011). Urban areas are generally more likely to be affected due to “heat islands” resulting from high levels of soil sealing, high-rise buildings, and the absence of green and open spaces. Importantly, both the risks of exposure and vulnerability to extreme heat are likely to be shaped by social inequalities, an aspect that has received relatively little attention so far (for initial evidence, see Vandentorren et al., 2006; Gronlund et al., 2015). Disadvantaged populations are likely to be at greater risk of experiencing extreme heat (e.g., due to lower levels of green space presence, lower building quality, or inability to afford air conditioning and other protective measures) and tend to be more vulnerable because of pre-existing conditions. Extreme heat thus seems likely to become an increasingly important source of health inequalities in the coming years. Noise Associational evidence suggests a link between environmental noise exposure and a range of health and well-being outcomes such as noise annoyance, sleep quality, cognitive impairment, birth outcomes, cardiovascular disease, and metabolic disorders. Aircraft and road traffic are the most important sources of noise, suggesting that it is highly correlated with other measures of environmental quality, most importantly air pollution, but also green space presence (or the absence thereof). Establishing causality therefore requires adjustment for confounders not only on the individual (e.g. age, SES, BMI) but also on the residential neighborhood level. While this also applies to other aspects of environmental quality, the link between noise and air pollution is particularly obvious.5 Munzel et al. (2014) review the cardiovascular effects of noise exposure and highlight that night-time noise exposure in particular can cause sleep disturbance and increased blood pres-

338  Handbook of health inequalities across the life course sure, heart rate, and stress hormone levels, which in turn may result in arterial hypertension. Kälsch et al. (2014) jointly investigate the cardiovascular effects of PM and noise and find that both air pollution and night-time noise exposure are associated with subclinical atherosclerosis after mutual adjustment. In a study by Liu et al. (2014), the association between blood pressure in children and air pollution even becomes insignificant once controlling for noise but not vice versa. Gehring et al. (2014) show that noise exposure during pregnancy is negatively associated with birth weight even after controlling for air pollution. Exploiting plausibly exogenous variation in night-time noise introduced by a night flight ban at Frankfurt Airport, Germany, Müller et al. (2016) find that the number of awakenings per night declined from 2.0 to 0.8 after the implementation of the ban. While the list of environmental factors considered in this chapter inevitably remains incomplete, we have discussed four of the most important and widely studied dimensions of environmental quality: air pollution, green space presence, heat, and noise. We have reviewed multidisciplinary evidence that collectively provides compelling evidence for both short- and longer-term effects of environmental quality on health and mortality. Environmental quality is not distributed equally. Disadvantaged and vulnerable populations live in less healthy environments, that is, there is environmental inequality. From a causal identification standpoint, such inequalities are a nuisance, a source of confounding that may bias estimates of the health effects of environmental quality. But of course, environmental inequality and its contribution to health inequalities and inequities more broadly are important matters in their own right. In the following sections, we therefore turn to the processes shaping environmental inequality and to some aspects of its measurement. Mechanisms Underlying Environmental Inequality Motivated by the profound effects of environmental quality on health and life chances reviewed in the previous section, research on environmental inequality seeks to understand, first, which population subgroups face low levels of residential environmental quality and, second, which mechanisms explain the socially unequal distributions of environmental goods and bads. A key finding of the international empirical literature on social inequalities in pollution exposure is that individuals with low income (Ash and Fetter, 2004), and even more so racial or ethnic minorities (Pais, Crowder and Downey, 2013; Ash and Boyce, 2018), carry a disproportionate burden of exposure. With regard to urban green space, studies similarly show clear associations between neighborhood- or community-level greenspace presence and socio-economic (Chen et al., 2020) as well as ethnic composition (Byrne, Wolch and Zhang, 2009; Matthew McConnachie and Shackleton, 2010; Kabisch and Haase, 2014). The processes that underlie these empirical regularities are not yet well understood. Theoretically, environmental inequality has often been attributed to two broad classes of causal mechanisms. The selective siting hypothesis states that environmental hazards such as polluting facilities or airports are disproportionately sited in or close to areas characterized by economic deprivation and/or high minority shares. Regarding green space, this would suggest that investments in public green spaces or urban renewal would primarily target residential areas with affluent and/or majority (white) residents. The selective migration hypothesis, in contrast, assumes that environmental inequality is the result of (post-siting) residential sorting,

Environmental inequality and health outcomes over the life course  339 with advantaged groups leaving areas with low environmental quality and moving into environmentally attractive areas at higher rates. Residential sorting along environmental lines could occur for several reasons that are not mutually exclusive (for a detailed overview, see Banzhaf, Ma and Timmins, 2019). Housing prices, which have been shown to be sensitive to air pollution (Chay and Greenstone, 2005) as well as green space presence (Franco and Macdonald, 2018; Panduro et al., 2018), likely play an important role (Mohai and Saha, 2015): disadvantaged groups may be priced out of environmentally attractive neighborhoods. The racial income inequality hypothesis (Crowder and Downey, 2010) accordingly states that ethnic and racial minorities move to and live in lower-quality neighborhoods because they have, on average, fewer economic resources. This seems to be only part of the explanation for ethno-racial disparities in environmental quality, however, as many studies show that ethnic minorities face lower environmental quality even after controlling for income and other resources (Downey et al., 2008; Rüttenauer, 2018). Housing and credit market discrimination are further structural explanations, in terms of both institutional rules and regulations (e.g. restrictive zoning, steering, credit scoring) and individual discrimination by landlords and other actors (Morello-Frosch and Jesdale, 2006; Small and Pager, 2020). Finally, historical “legacy effects” could explain the persistence of environmental inequality. If (for whatever reasons) racial and ethnic minorities settled in areas with low environmental quality in the past, current members of these groups may continue to be drawn towards these areas. Research based on observational (Mossaad et al., 2020) and experimental data (Ibraimovic and Masiero, 2014) suggests that living with co-ethnics plays an important role in the internal location and migration decisions of immigrants and ethnic minorities. Studies of residential mobility show that the presence of close relatives and friends reduces the probability of moving (Clark, Duque-Calvache and Palomares-Linares, 2017; Ermisch and Mulder, 2019; Hünteler and Mulder, 2020). Such “local social capital” is assumed to deter residential mobility because the resources stemming from it are location-specific and will be less valuable if a household moves (Kan, 2007). Qualitative research from the Moving to Opportunity (MTO) experiment corroborates the importance of residential proximity to close ties, showing that many households that initially moved out of high-poverty neighborhoods either keep spending much of their time at their former communities or even move back (Popkin et al., 2002). The bottom line is that, in the presence of legacy effects, ethno-racial minorities may face a trade-off between living with co-ethnics and enjoying better environmental conditions that does not exist for members of majority groups.

ENVIRONMENTAL INEQUALITY AT DIFFERENT LEVELS OF SPATIAL RESOLUTION Investigating patterns of environmental inequality and testing underlying causal mechanisms with observational data are complex tasks that require spatially fine-grained and longitudinal data on different dimensions of environmental quality as well as individual- or neighborhood-level SES. Longitudinal data is required to shed light on the mechanisms of selective siting and sorting. While large industrial plants may affect the health of residents that live even some kilometers away, other dimensions of environmental quality such as air pollution and noise from road traffic primarily affect those who live close by. Similarly, some

340  Handbook of health inequalities across the life course of the health-promoting effects of green space (e.g. stress restoration) have been argued to be concentrated among those who live in sight. High spatial granularity can also be important for exploiting quasi-experimental (spatial) variation. Conversely, measuring environmental quality and/or ethnic and socio-economic composition at overly coarse levels of spatial aggregation – e.g. counties, districts, or municipalities – can conceal the true extent of environmental inequality. We illustrate the importance of fine-grained spatial data with a concrete example in Figure 21.2, which shows the relationship between neighborhood ethnic composition and green space presence at different levels of spatial resolution for German municipalities with at least 100,000 inhabitants. Combining data on the proportion of non-nationals at the block level6 with high-resolution satellite data,7 we calculated the surface share of natural environments (urban green space, sport and leisure facilities, and forests) for three levels of spatial aggregation (both the share of non-nationals and the green space surface share are expressed in percentage terms). The first two levels of aggregation are the administrative units of city/municipality and of districts within cities and municipalities. The third and most fine-grained level of aggregation, the neighborhood level, was constructed by calculating the natural environment share in a 250 m circular buffer around each block. We then ran bivariate OLS models regressing the share of non-nationals, that is, residents without German citizenship, on our measure of

Note: The figure shows the association between the population share of non-nationals and urban green space presence (measured as the surface share covered by natural environments) at different levels of spatial aggregation, with measures expressed in percentage terms. Depicted are the coefficients from three separate bivariate linear regressions for the three different levels of aggregation (OLS estimates, all standard errors clustered at the municipality level). For example, the neighborhood-level coefficient estimate of –0.12 implies that the (neighborhood-level) population share of non-nationals declines by 0.12 percentage points for each percentage point increase in green space surface share.

Figure 21.2

Association between urban green space availability and share of non-nationals at different levels of spatial aggregation

Environmental inequality and health outcomes over the life course  341 green space presence. Figure 21.2 shows the resulting regression coefficient estimates and 95 percent confidence intervals for the three different levels. At the city level, we do not observe a statistically significant association between green space availability and demographic composition. The city-level point estimate is positive, suggesting that – if anything – higher green space presence goes hand in hand with higher population share of non-nationals. The picture changes completely once we look at the association at the level of city districts: districts with higher shares of non-national residents, on average, tend to have lower green space presence. Importantly, this association becomes even stronger when we turn to the neighborhood level. The neighborhood-level coefficient estimate implies that the population share of non-nationals declines by 0.12 percentage points per percentage-point increase in the green space surface share. The fully standardized coefficient estimate is –.13, that is, a standard deviation increase in the surrounding green space surface share is associated with a decline in the block-level share of non-nationals of approximately .13 standard deviations on average.8 These results are a simple, yet powerful illustration of how spatial aggregation can understate or even conceal the magnitude of environmental inequalities.

CONCLUSION We have reviewed multidisciplinary evidence on the health effects of various dimensions of residential environmental quality. In so doing, we have paid particular attention to issues of causal identification and to the by now sizable body of quasi-experimental research indicating substantial causal effects of environmental quality on health, particularly for air pollution. As for the life course patterning of these effects, research supports the notion that in utero and childhood exposures are “critical”. This is certainly true in the sense that children, in addition to the elderly and adults with preexisting health problems, are often found to suffer the most from adverse environmental influences in the short run, in terms of both mortality (stillbirths) and morbidity (e.g., asthma) as well as with respect to indicators of overall development (e.g., birthweight). There is some empirical support for long-term cumulative effects of environmental exposures in utero and in childhood, although evidence – and especially quasi-experimental evidence – remains much more limited on this front, partly due to formidable data requirements and the potentially long latency of effects. In the later parts of this chapter, we have discussed the social mechanisms that generate and reproduce patterns of environmental inequality. We have emphasized and illustrated the importance of granular geo-spatial data for mapping the full extent of environmental inequality and for implementing compelling quasi-experimental designs. On a theoretical level, we have discussed several explanations for environmental inequality. In addition to resource constraints and housing or credit market discrimination, we have pointed to “legacy effects” as one potentially important reason why ethno-racial minorities may face a disproportionate environmental burden. If co-ethnics tend to cluster in areas with low environmental quality, minority individuals will face a trade-off between environmental quality and living close to co-ethnic friends and infrastructures – even if the reasons for residential clustering are purely historical. This trade-off may simply not exist for non-minority individuals, at least not to the same extent. While we have mostly focused on the health effects of environmental quality, a growing body of literature investigates effects on human capital development (Nilsson et al., 2009;

342  Handbook of health inequalities across the life course Sanders, 2012; Persico, 2020) that are likely to mediate a portion of the total long-term effect of (early-life) environmental conditions on health (cf. Figure 21.1 above). The literature reviewed in this chapter is, thus, also linked to the broader literature on trends in morbidity and mortality. For example, Case and Deaton (2017) show that college education protects against a range of undesirable outcomes, including (involuntary) singlehood, social isolation, and detachment from the labor market, all of which are linked to rising mortality from alcohol and drug abuse as well as suicide (“deaths of despair”). We have reviewed a substantial literature from a fairly wide range of disciplines, but many open questions remain. We highlight two of them here. First, as noted at the beginning of the chapter, the evidence discussed above stems from high-income countries. This is partly due to space limitations but also a matter of data availability: estimating (long-term) causal health effects of environmental quality requires spatially fine-grained and longitudinal data that are more readily available for high-income countries (e.g. Sweden, Denmark, USA). Such countries tend to be characterized by rather moderate levels of, for example, air pollution. Adverse health effects of environmental pollution are likely to be more severe in the high-pollution contexts found in many low- and middle-income countries. Second, we have focused on outdoor residential environmental quality, ignoring environmental influences experienced in non-residential settings and environments, for example, in and around schools or workplaces. Approximating an individual’s level of exposure solely by measures of outdoor environmental quality does not account for aspects of indoor environments that may be both conducive (e.g., protection from pollution and heat through insulation or air conditioning) or detrimental to health (e.g., indoor smoking). Restricting measures of environmental quality to the place of residence may introduce considerable measurement error, as individuals spend considerable amounts of time elsewhere. Moreover, different social groups may do so to different extents and face very different exposures in non-residential environments, as illustrated by recent work on social inequalities in infection risks during the first wave of the Covid-19 pandemic (Chang et al., 2021). Longitudinal and spatio-temporally fine-grained data at the contextual-environmental, household, and individual levels is needed to answer these and other important questions about the environment-health nexus – and to foster a richer evidence base that helps build healthy environments for all members of society.

NOTES 1. Of course, it is equally possible that health partly or largely mediates the effects of environmental insults on these outcomes, e.g., when children miss school because of health issues. 2. Ben-Shlomo and Kuh (2002) provide a more fine-grained conceptualization of four different pathways: social, biological, socio-biological, and bio-social. The fact that infants born to households with low socio-economic status, on average, show less favorable birth outcomes (e.g. due to differential in utero environmental quality) qualifies as a socio-biological pathway, whereas early life illnesses affecting educational participation and performance as well as subsequent labor market outcomes constitute a bio-social pathway. 3. Under normal conditions, temperature decreases with altitude, leading air pollutants emitted close to the ground to rise and disperse. During inversion periods, a warmer layer of air traps air pollutants close to the ground, thereby slowing down or preventing vertical exchange of air and increasing pollution concentration close to the surface. 4. Low Emission Zones are (mostly inner-city) areas where polluting vehicles are regulated. This usually implies that vehicles have to meet certain emission standards to be allowed to enter the area.

Environmental inequality and health outcomes over the life course  343 5. The same problem, intuitively, also applies for many of the studies investigating adverse health effects of traffic-induced air pollution. Natural experiments exploiting variation in traffic-related pollution do not necessarily isolate the health effects of air pollution from the health effects of noise. Policy changes affecting traffic, such as the introduction of LEZs, could also affect noise. For studies based on weather events, such as inversion periods, this should be less of a problem. 6. Socio-demographic data at the neighborhood level were obtained from infas 360 GmbH (https://​ www​.infas360​.de/​). There are 186,423 neighborhood blocks containing around 13.92 million households across all German municipalities with at least 100,000 inhabitants, so the average block comprises approximately 75 households. 7. High-resolution land use maps for urban areas based on satellite imagery were obtained from the EU Copernicus Urban Atlas (https://​land​.copernicus​.eu/​local/​urban​-atlas). 8. The population-weighted average share of non-nationals is equal to 17.36 for all three levels of spatial disaggregation. The city/district/block-level standard deviations are 5.3, 9.3, and 10.7, respectively. The population-weighted average city/district/block-level green space surface shares are 27.0/22.6/12.6, with standard deviations of 9.7/15.2/11.7.

REFERENCES Abelt, K. and McLafferty, S. (2017) ‘Green Streets: Urban Green and Birth Outcomes’, International Journal of Environmental Research and Public Health, 14(7), p. 771. doi:10.3390/ijerph14070771. Alba, R. D. and Logan, J. R. (1993) ‘Minority Proximity to Whites in Suburbs: An Individual-Level Analysis of Segregation’, American Journal of Sociology, 98(6), pp. 1388–1427. doi:10.1086/230193. Alcock, I., White, M. P., Wheeler, B. W., Fleming, L. E. and Depledge, M. H. (2014) ‘Longitudinal Effects on Mental Health of Moving to Greener and Less Green Urban Areas’, Environmental Science & Technology, 48(2), pp. 1247–1255. doi:10.1021/es403688w. Alexander, D. and Schwandt, H. (2022) ‘The Impact of Car Pollution on Infant and Child Health: Evidence from Emissions Cheating’, Review of Economic Studies, 89(6), pp. 2872–2910. doi:10.1093/ restud/rdac007. Almond, D. and Currie, J. (2011) ‘Killing Me Softly: The Fetal Origins Hypothesis’, Journal of Economic Perspectives, 25(3), pp. 153–172. doi:10.1257/jep.25.3.153. Andersen, S. L. (2003) ‘Trajectories of Brain Development: Point of Vulnerability or Window of Opportunity?’, Neuroscience & Biobehavioral Reviews, 27(1), pp. 3–18. doi:10.1016/ S0149-7634(03)00005-8. Ash, M. and Boyce, J. K. (2018) ‘Racial Disparities in Pollution Exposure and Employment at US Industrial Facilities’, Proceedings of the National Academy of Sciences, 115(42), pp. 10636–10641. doi:10.1073/pnas.1721640115. Ash, M. and Fetter, T. R. (2004) ‘Who Lives on the Wrong Side of the Environmental Tracks? Evidence from the EPA’s Risk-Screening Environmental Indicators Model’, Social Science Quarterly, 85(2), pp. 441–462. Banzhaf, H. S., Ma, L. and Timmins, C. (2019) ‘Environmental Justice: Establishing Causal Relationships’, Annual Review of Resource Economics, 11(1), pp. 377–398. doi:10.1146/ annurev-resource-100518-094131. Banzhaf, H. S. and Walsh, R. P. (2008) ‘Do People Vote with Their Feet? An Empirical Test of Tiebout’s Mechanism’, American Economic Review, 98(3), pp. 843–863. doi:10.1257/aer.98.3.843. Barker, D. J. P. (1995) ‘Fetal Origins of Coronary Heart Disease’, BMJ, 311(6998), pp. 171–174. doi:10.1136/bmj.311.6998.171. Barker, D. J. P. and Osmond, C. (1986) ‘Infant Mortality, Childhood Nutrition, and Ischaemic Heart Disease in England and Wales’, The Lancet, 327(8489), pp. 1077–1081. doi:10.1016/ S0140-6736(86)91340-1. Bateson, T. F. and Schwartz, J. (2007) ‘Children’s Response to Air Pollutants’, Journal of Toxicology and Environmental Health, Part A, 71(3), pp. 238–243. doi:10.1080/15287390701598234.

344  Handbook of health inequalities across the life course Behndig, A. F. et al. (2006) ‘Airway Antioxidant and Inflammatory Responses to Diesel Exhaust Exposure in Healthy Humans’, European Respiratory Journal, 27(2), pp. 359–365. doi:10.1183/090 31936.06.00136904. Ben-Shlomo, Y. and Kuh, D. (2002) ‘A Life Course Approach to Chronic Disease Epidemiology: Conceptual Models, Empirical Challenges and Interdisciplinary Perspectives’, International Journal of Epidemiology, 31(2), pp. 285–293. doi:10.1093/ije/31.2.285. Best, H. and Rüttenauer, T. (2018) ‘How Selective Migration Shapes Environmental Inequality in Germany: Evidence from Micro-level Panel Data’, European Sociological Review, 34(1), pp. 52–63. Bové, H. et al. (2019) ‘Ambient Black Carbon Particles Reach the Fetal Side of Human Placenta’, Nature Communications, 10(1), p. 3866. doi:10.1038/s41467-019-11654-3. Brook, R. D. et al. (2010) ‘Particulate Matter Air Pollution and Cardiovascular Disease: An Update to the Scientific Statement from the American Heart Association’, Circulation, 121(21), pp. 2331–2378. doi:10.1161/CIR.0b013e3181dbece1. Byrne, J., Wolch, J. and Zhang, J. (2009) ‘Planning for Environmental Justice in an Urban National Park’, Journal of Environmental Planning and Management, 52(3), pp. 365–392. doi:10.1080/09640560802703256. Case, A. and Deaton, A. (2017) ‘Mortality and Morbidity in the 21st Century’, Brookings Papers on Economic Activity, pp. 397–476. Chalmin-Pui, L. S. et al. (2021) ‘“It Made Me Feel Brighter in Myself” – the Health and Well-Being Impacts of a Residential Front Garden Horticultural Intervention’, Landscape and Urban Planning, 205, p. 103958. doi:10.1016/j.landurbplan.2020.103958. Chang, S. et al. (2021) ‘Mobility Network Models of COVID-19 Explain Inequities and Inform Reopening’, Nature, 589(7840), pp. 82–87. doi:10.1038/s41586-020-2923-3. Chay, K., Dobkin, C. and Greenstone, M. (2003) ‘The Clean Air Act of 1970 and Adult Mortality’, Journal of Risk and Uncertainty, 27(3), pp. 279–300. doi:10.1023/A:1025897327639. Chay, K. Y. and Greenstone, M. (2003) Air Quality, Infant Mortality, and the Clean Air Act of 1970. Working Paper 10053. National Bureau of Economic Research. doi:10.3386/w10053. Chay, K. Y. and Greenstone, M. (2005) ‘Does Air Quality Matter? Evidence from the Housing Market’, Journal of Political Economy, 113(2), pp. 376–424. doi:10.1086/427462. Chen, X. et al. (2020) ‘The Effects of Prenatal Exposure to Temperature Extremes on Birth Outcomes: The Case of China’, Journal of Population Economics, 33(4), pp. 1263–1302. doi:10.1007/ s00148-020-00768-4. Clark, W. A. V., Duque-Calvache, R. and Palomares-Linares, I. (2017) ‘Place Attachment and the Decision to Stay in the Neighbourhood’, Population, Space and Place, 23(2), p. e2001. doi:10.1002/ psp.2001. Cohen-Cline, H., Turkheimer, E. and Duncan, G. E. (2015) ‘Access to Green Space, Physical Activity and Mental Health: A Twin Study’, J Epidemiol Community Health, 69(6), pp. 523–529. doi:10.1136/ jech-2014-204667. Crowder, K. and Downey, L. (2010) ‘Interneighborhood Migration, Race, and Environmental Hazards: Modeling Microlevel Processes of Environmental Inequality’, American Journal of Sociology, 115(4), pp. 1110–1149. doi:10.1086/649576. Currie, J. (2013) ‘Pollution and Infant Health’, Child Development Perspectives, 7(4), pp. 237–242. doi:10.1111/cdep.12047. Currie, J., Neidell, M. and Schmieder, J. F. (2009) ‘Air Pollution and Infant Health: Lessons from New Jersey’, Journal of Health Economics, 28(3), pp. 688–703. Currie, J. and Walker, R. (2011) ‘Traffic Congestion and Infant Health: Evidence from E-ZPass’, American Economic Journal: Applied Economics, 3(1), pp. 65–90. doi:10.1257/app.3.1.65. Currie, J. et al. (2014) ‘What Do We Know About Short- and Long-Term Effects of Early-Life Exposure to Pollution?’, Annual Review of Resource Economics, 6(1), pp. 217–247. doi:10.1146/ annurev-resource-100913-012610. Deryugina, T. et al. (2019) ‘The Mortality and Medical Costs of Air Pollution: Evidence from Changes in Wind Direction’, American Economic Review, 109(12), pp. 4178–4219. Devlin, R. B. et al. (2003) ‘Elderly Humans Exposed to Concentrated Air Pollution Particles Have Decreased Heart Rate Variability’, European Respiratory Journal, 21(40 suppl), pp. 76s–80s. doi:10 .1183/09031936.03.00402403.

Environmental inequality and health outcomes over the life course  345 DiPrete, T. A. and Eirich, G. M. (2006) ‘Cumulative Advantage as a Mechanism for Inequality: A Review of Theoretical and Empirical Developments’, Annual Review of Sociology, 32(1), pp. 271–297. doi:10.1146/annurev.soc.32.061604.123127. Dockery, D. W. et al. (1993) ‘An Association between Air Pollution and Mortality in Six U.S. Cities’, The New England Journal of Medicine, 329(24), pp. 1753–1759. doi:10.1056/NEJM199312093292401. Dominici, F. et al. (2006) ‘Fine Particulate Air Pollution and Hospital Admission for Cardiovascular and Respiratory Diseases’, Journal of the American Medical Association, 295(10), pp. 1127–1134. Donovan, G. H. et al. (2018) ‘Vegetation Diversity Protects against Childhood Asthma: Results from a Large New Zealand Birth Cohort’, Nature Plants, 4(6), pp. 358–364. doi:10.1038/s41477-018-0151-8. Downey, L. et al. (2008) ‘Environmental Inequality in Metropolitan America’, Organization & Environment, 21(3), pp. 270–294. doi:10.1177/1086026608321327. Engemann, K. et al. (2019) ‘Residential Green Space in Childhood is Associated with Lower Risk of Psychiatric Disorders from Adolescence into Adulthood’, Proceedings of the National Academy of Sciences, 116(11), pp. 5188–5193. doi:10.1073/pnas.1807504116. Engzell, P. and Hällsten, M. (2022) ‘A Caution on the Discordant Parenting Design’. SocArXiv. doi:10.31235/osf​.io/​rx4z6. Ermisch, J. and Mulder, C. H. (2019) ‘Migration Versus Immobility, and Ties to Parents’, European Journal of Population, 35(3), pp. 587–608. doi:10.1007/s10680-018-9494-0. Franco, S. F. and Macdonald, J. L. (2018) ‘Measurement and Valuation of Urban Greenness: Remote Sensing and Hedonic Applications to Lisbon, Portugal’, Regional Science and Urban Economics, 72, pp. 156–180. doi:10.1016/j.regsciurbeco.2017.03.002. Gehring, U. et al. (2014) ‘Impact of Noise and Air Pollution on Pregnancy Outcomes’, Epidemiology, 25(3), pp. 351–358. Gehrsitz, M. (2017) ‘The effect of low emission zones on air pollution and infant health’, Journal of Environmental Economics and Management, 83, pp. 121–144. doi:10.1016/j.jeem.2017.02.003. Gidlow, C. J. et al. (2016) ‘Natural Environments and Chronic Stress Measured by Hair Cortisol’, Landscape and Urban Planning, 148, pp. 61–67. doi:10.1016/j.landurbplan.2015.12.009. Graff Zivin, J. and Neidell, M. (2013) ‘Environment, Health, and Human Capital’, Journal of Economic Literature, 51(3), pp. 689–730. doi:10.1257/jel.51.3.689. Gronlund, C. J. et al. (2015) ‘Vulnerability to Extreme Heat by Socio-Demographic Characteristics and Area Green Space among the Elderly in Michigan, 1990–2007’, Environmental Research, 136, pp. 449–461. doi:10.1016/j.envres.2014.08.042. Guirguis, K. et al. (2014) ‘The Impact of Recent Heat Waves on Human Health in California’, Journal of Applied Meteorology and Climatology, 53(1), pp. 3–19. doi:10.1175/JAMC-D-13-0130.1. Hughes, K. et al. (2017) ‘The Effect of Multiple Adverse Childhood Experiences on Health: A Systematic Review and Meta-Analysis’, The Lancet Public Health, 2(8), pp. e356–e366. doi:10.1016/S2468-2667(17)30118-4. Hünteler, B. and Mulder, C. H. (2020) ‘Geographic Proximity to Parents, Intergenerational Support Exchange, and Migration Within Germany’, European Journal of Population = Revue Européenne de Démographie, 36(5), pp. 895–918. doi:10.1007/s10680-020-09558-w. Ibraimovic, T. and Masiero, L. (2014) ‘Do Birds of a Feather Flock Together? The Impact of Ethnic Segregation Preferences on Neighbourhood Choice’, Urban Studies, 51(4), pp. 693–711. doi:10.1177/0042098013493026. Isen, A., Rossin-Slater, M. and Walker, W. R. (2017) ‘Every Breath You Take – Every Dollar You’ll Make: The Long-Term Consequences of the Clean Air Act of 1970’, Journal of Political Economy, p. 55. James, P. et al. (2016) ‘Exposure to Greenness and Mortality in a Nationwide Prospective Cohort Study of Women’, Environmental Health Perspectives, 124(9), pp. 1344–1352. doi:10.1289/ehp.1510363. Jans, J., Johansson, P. and Nilsson, J. P. (2018) ‘Economic Status, Air Quality, and Child Health: Evidence from Inversion Episodes’, Journal of Health Economics, 61, pp. 220–232. doi:10.1016/j. jhealeco.2018.08.002. Kabisch, N. and Haase, D. (2014) ‘Green Justice or Just Green? Provision of Urban Green Spaces in Berlin, Germany’, Landscape and Urban Planning, 122, pp. 129–139.

346  Handbook of health inequalities across the life course Kälsch, H. et al. (2014) ‘Are Air Pollution and Traffic Noise Independently Associated with Atherosclerosis: The Heinz Nixdorf Recall Study’, European Heart Journal, 35(13), pp. 853–860. doi:10.1093/eurheartj/eht426. Kan, K. (2007) ‘Residential mobility and social capital’, Journal of Urban Economics, 61(3), pp. 436–457. doi:10.1016/j.jue.2006.07.005. Katsouyanni, K. et al. (2001) ‘Confounding and Effect Modification in the Short-Term Effects of Ambient Particles on Total Mortality: Results from 29 European Cities within the APHEA2 Project’, Epidemiology, 12(5), pp. 521–531. Kelly, F. J. and Fussell, J. C. (2011) ‘Air Pollution and Airway Disease’, Clinical & Experimental Allergy, 41(8), pp. 1059–1071. doi:10.1111/j.1365-2222.2011.03776.x. Kelly, F. J. and Fussell, J. C. (2020) ‘Toxicity of Airborne Particles – Established Evidence, Knowledge Gaps and Emerging Areas of Importance’, Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences, 378(2183), p. 20190322. doi:10.1098/ rsta.2019.0322. Kondo, M. et al. (2018) ‘Urban Green Space and Its Impact on Human Health’, International Journal of Environmental Research and Public Health, 15(3), p. 445. doi:10.3390/ijerph15030445. Koppe, C. et al. (2004) Heat Waves: Risks and Responses. Copenhagen: Regional Office for Europe, World Health Organization. Kuh, D. et al. (2003) ‘Life Course Epidemiology’, Journal of Epidemiology and Community Health, 57(10), pp. 778–783. doi:10.1136/jech.57.10.778. Liu, C. et al. (2014) ‘The Associations between Traffic-Related Air Pollution and Noise with Blood Pressure in Children: Results from the GINIplus and LISAplus studies’, International Journal of Hygiene and Environmental Health, 217(4–5), pp. 499–505. doi:10.1016/j.ijheh.2013.09.008. Luyten, L. J. et al. (2018) ‘Air Pollution and the Fetal Origin of Disease: A Systematic Review of the Molecular Signatures of Air Pollution Exposure in Human Placenta’, Environmental Research, 166, pp. 310–323. doi:10.1016/j.envres.2018.03.025. Margaryan, S. (2021) ‘Low Emission Zones and Population Health’, Journal of Health Economics, 76, p. 102402. doi:10.1016/j.jhealeco.2020.102402. Markevych, I. et al. (2017) ‘Exploring Pathways Linking Greenspace to Health: Theoretical and Methodological Guidance’, Environmental Research, 158, pp. 301–317. doi:10.1016/j. envres.2017.06.028. Matthew McConnachie, M. and Shackleton, C.M. (2010) ‘Public Green Space Inequality in Small Towns in South Africa’, Habitat International, 34(2), pp. 244–248. doi:10.1016/j.habitatint.2009.09.009. McNamara, C.L. et al. (2017) ‘The Socioeconomic Distribution of Non-Communicable Diseases in Europe: Findings from the European Social Survey (2014) Special Module on the Social Determinants of Health’, European Journal of Public Health, 27(suppl_1), pp. 22–26. doi:10.1093/eurpub/ckw222. Mohai, P. and Saha, R. (2015) ‘Which Came First, People or Pollution? A Review of Theory and Evidence from Longitudinal Environmental Justice Studies’, Environmental Research Letters, 10(12), p. 125011. Morello-Frosch, R. and Jesdale, B. M. (2006) ‘Separate and Unequal: Residential Segregation and Estimated Cancer Risks Associated with Ambient Air Toxics in U.S. Metropolitan Areas’, Environmental Health Perspectives, 114(3), pp. 386–393. doi:10.1289/ehp.8500. Moretti, E. and Neidell, M. (2009) ‘Pollution, Health, and Avoidance Behavior: Evidence from the Ports of Los Angeles’, Journal of Human Resources, 46(1), pp. 154–175. Morgan, S. L. and Winship, C. (2007) Counterfactuals and Causal Inference: Methods and Principles for Social Research. New York: Cambridge University Press (Analytical Methods for Social Research). Mossaad, N. et al. (2020) ‘In search of opportunity and community: Internal migration of refugees in the United States’, Science Advances, 6(32), p. eabb0295. doi:10.1126/sciadv.abb0295. Mudway, I. S. et al. (2019) ‘Impact of London’s Low Emission Zone on Air Quality and Children’s Respiratory Health: A Sequential Annual Cross-Sectional Study’, The Lancet. Public Health, 4(1), pp. e28–e40. doi:10.1016/S2468-2667(18)30202-0. Müller, U. et al. (2016) ‘The NORAH-Sleep Study: Effects of the Night Flight Ban at Frankfurt Airport’, INTER-NOISE and NOISE-CON Congress and Conference Proceedings, 253(7), pp. 1260–1264. Munzel, T. et al. (2014) ‘Cardiovascular Effects of Environmental Noise Exposure’, European Heart Journal, 35(13), pp. 829–836. doi:10.1093/eurheartj/ehu030.

Environmental inequality and health outcomes over the life course  347 Neidell, M. (2009) ‘Information, Avoidance Behavior, and Health: The Effect of Ozone on Asthma Hospitalizations’, Journal of Human Resources, 44(2), pp. 450–478. Nilsson, J. P. et al. (2009) ‘The Long-term Effects of Early Childhood Lead Exposure: Evidence from Sharp Changes in Local Air Lead Levels Induced by the Phase-Out of Leaded Gasoline’, in Childhood Obesity: Introducing the Issue, The Future of Children, pp. 345–352. Nitschke, M. et al. (2011) ‘Impact of Two Recent Extreme Heat Episodes on Morbidity and Mortality in Adelaide, South Australia: A Case-Series Analysis’, Environmental Health, 10(1), p. 42. doi:10.1186/1476-069X-10-42. Noordzij, J. M. et al. (2020) ‘Effect of Changes in Green Spaces on Mental Health in Older Adults: A Fixed Effects Analysis’, Journal of Epidemiology and Community Health, 74(1), pp. 48–56. doi:10.1136/jech-2019-212704. Ostro, B. D., Hurley, S. and Lipsett, M. J. (1999) ‘Air Pollution and Daily Mortality in the Coachella Valley, California: A Study of PM10 Dominated by Coarse Particles’, Environmental Research, 81(3), pp. 231–238. doi:10.1006/enrs.1999.3978. Pais, J., Crowder, K. and Downey, L. (2013) ‘Unequal Trajectories: Racial and Class Differences in Residential Exposure to Industrial Hazard’, Social Forces, 92(3), pp. 1189–1215. doi:10.1093/sf/ sot099. Panduro, T. E. et al. (2018) ‘Eliciting preferences for urban parks’, Regional Science and Urban Economics, 73, pp. 127–142. doi:10.1016/j.regsciurbeco.2018.09.001. Persico, C. (2020) ‘Can Pollution Cause Poverty? The Effects of Pollution on Educational, Health and Economic Outcomes’, SSRN Electronic Journal [Preprint]. doi:10.2139/ssrn.3539513. Pope, C. A. (2002) ‘Lung Cancer, Cardiopulmonary Mortality, and Long-term Exposure to Fine Particulate Air Pollution’, JAMA, 287(9), p. 1132. doi:10.1001/jama.287.9.1132. Pope, C. A. and Dockery, D. W. (2006) ‘Health Effects of Fine Particulate Air Pollution: Lines that Connect’, Journal of the Air & Waste Management Association, 56(6), pp. 709–742. doi:10.1080/10 473289.2006.10464485. Pope, C. A. et al. (1995) ‘Particulate Air Pollution as a Predictor of Mortality in a Prospective Study of U.S. Adults’, American Journal of Respiratory and Critical Care Medicine, 151(3 Pt 1), pp. 669–674. doi:10.1164/ajrccm/151.3_Pt_1.669. Popkin, S. J. et al. (2002) Families in Transition: A Qualitative Analysis of the MTO Experience, U.S. Department of Housing and Urban Development, Office of Policy Development and Research. Roe, J. J. et al. (2013) ‘Green Space and Stress: Evidence from Cortisol Measures in Deprived Urban Communities’, International Journal of Environmental Research and Public Health, 10(9), pp. 4086–4103. doi:10.3390/ijerph10094086. Rüttenauer, T. (2018) ‘Neighbours Matter: A Nation-Wide Small-Area Assessment of Environmental Inequality in Germany’, Social Science Research, 70, pp. 198–211. Salvi, S. et al. (1999) ‘Acute Inflammatory Responses in the Airways and Peripheral Blood After Short-Term Exposure to Diesel Exhaust in Healthy Human Volunteers’, American Journal of Respiratory and Critical Care Medicine, 159(3), pp. 702–709. doi:10.1164/ajrccm.159.3.9709083. Salvi, S. S. et al. (2000) ‘Acute Exposure to Diesel Exhaust Increases IL-8 and GRO-α Production in Healthy Human Airways’, American Journal of Respiratory and Critical Care Medicine, 161(2), pp. 550–557. doi:10.1164/ajrccm.161.2.9905052. Sanders, N. J. (2012) ‘What Doesn’t Kill You Makes You Weaker: Prenatal Pollution Exposure and Educational Outcomes’, Journal of Human Resources, 47(3), pp. 826–850. doi:10.1353/ jhr.2012.0018. Schwartz, J. (2000) ‘Assessing Confounding, Effect Modification, and Thresholds in the Association between Ambient Particles and Daily Deaths’, Environmental Health Perspectives, 108(6), pp. 563–568. doi:10.1289/ehp.00108563. Schwartz, J. (2001) ‘Is There Harvesting in the Association of Airborne Particles with Daily Deaths and Hospital Admissions?’, Epidemiology, 12(1), pp. 55–61. Schwartz, J. (2004) ‘Air Pollution and Children’s Health’, Pediatrics, 113(Supplement 3), pp. 1037–1043. doi:10.1542/peds.113.S3.1037. Simkhovich, B. Z., Kleinman, M. T. and Kloner, R. A. (2008) ‘Air Pollution and Cardiovascular Injury’, Journal of the American College of Cardiology, 52(9), pp. 719–726. doi:10.1016/j.jacc.2008.05.029.

348  Handbook of health inequalities across the life course Sivarajah, S., Smith, S. M. and Thomas, S. C. (2018) ‘Tree Cover and Species Composition Effects on Academic Performance of Primary School Students’, PLoS ONE, 13(2), p. e0193254. doi:10.1371/ journal.pone.0193254. Small, M. L. and Pager, D. (2020) ‘Sociological Perspectives on Racial Discrimination’, Journal of Economic Perspectives, 34(2), pp. 49–67. doi:10.1257/jep.34.2.49. Stenfors, N. et al. (2004) ‘Different Airway Inflammatory Responses in Asthmatic and Healthy Humans Exposed to Diesel’, European Respiratory Journal, 23(1), pp. 82–86. doi:10.1183/09031936.03.000 04603. Sun, Q. et al. (2005) ‘Long-term Air Pollution Exposure and Acceleration of Atherosclerosis and Vascular Inflammation in an Animal Model’, JAMA, 294(23), pp. 3003–3010. doi:10.1001/jama.294.23.3003. Triguero-Mas, M. et al. (2015) ‘Natural Outdoor Environments and Mental and Physical Health: Relationships and Mechanisms’, Environment International, 77, pp. 35–41. doi:10.1016/j. envint.2015.01.012. Vandentorren, S. et al. (2006) ‘August 2003 Heat Wave in France: Risk Factors for Death of Elderly People Living at Home’, European Journal of Public Health, 16(6), pp. 583–591. doi:10.1093/ eurpub/ckl063. Ward Thompson, C. et al. (2012) ‘More Green Space is Linked to Less Stress in Deprived Communities: Evidence from Salivary Cortisol Patterns’, Landscape and Urban Planning, 105(3), pp. 221–229. doi:10.1016/j.landurbplan.2011.12.015. White, M. P. et al. (2013) ‘Would You Be Happier Living in a Greener Urban Area? A Fixed-Effects Analysis of Panel Data’, Psychological Science, 24(6), pp. 920–928. doi:10.1177/0956797612464659. WHO (2013) Review of Evidence on Health Aspects of Air Pollution – REVIHAAP Project: Final Technical Report. Available at: https://​www​.euro​.who​.int/​en/​health​-topics/​environment​-and​-health/​ air​-quality/​publications/​2013/​review​-of​-evidence​-on​-health​-aspects​-of​-air​-pollution​-revihaap​-project​ -final​-technical​-report (accessed: 22 January 2022). Wood, H. E. et al. (2015) ‘Effects of Air Pollution and the Introduction of the London Low Emission Zone on the Prevalence of Respiratory and Allergic Symptoms in Schoolchildren in East London: A Sequential Cross-Sectional Study’, PLOS ONE, 10(8), p. e0109121. doi:10.1371/journal. pone.0109121. Xu, Z. et al. (2014) ‘The Impact of Heat Waves on Children’s Health: A Systematic Review’, International Journal of Biometeorology, 58(2), pp. 239–247. doi:10.1007/s00484-013-0655-x. Zanobetti, A. and Schwartz, J. (2002) ‘Cardiovascular Damage by Airborne Particles: Are Diabetics More Susceptible?’, Epidemiology, 13(5), pp. 588–592. Zanobetti, A., Schwartz, J. and Dockery, D. W. (2000) ‘Airborne Particles are a Risk Factor for Hospital Admissions for Heart and Lung Disease’, Environmental Health Perspectives, 108(11), pp. 1071–1077. doi:10.1289/ehp.001081071. Zhang, X., Chen, X. and Zhang, X. (2018) ‘The Impact of Exposure to Air Pollution on Cognitive Performance’, Proceedings of the National Academy of Sciences, 115(37), pp. 9193–9197. doi:10.1073/pnas.1809474115.

22. Infectious diseases across the life course: an inequalities perspective Nico Dragano

INTRODUCTION Infections with viruses, bacteria or parasites are an unavoidable part of life for most living beings on this planet – they are, so to speak, biological normality. Humans are hosts to hundreds of such pathogens, some of which are entirely harmless, while others can have serious or even fatal consequences. Diseases caused by infections continue to shape global population health despite the epidemiological transformation of the last 150 years, which has been marked by a sharp increase in the importance of chronic degenerative diseases. Accordingly, among the ten leading causes of death currently identified by the World Health Organization (WHO), three are communicable diseases (neonatal conditions, diarrhoeal disease, lower respiratory diseases) (World Health Organization, 2022). The burden of disease from non-fatal infectious diseases is also enormous, although difficult to quantify at a global level. Therefore, to take just one – arbitrary – example from the country where I work: in 2019, 52,979,966 sick days were reported for acute upper respiratory infections (ICD-10 J00-J06) for employees in the statutory health insurance system alone (Bundesgesundheitsministerium, 2021). In addition, infections also play a role in the aetiology of non-communicable diseases, so that their importance tends to be underestimated when looking at main diagnoses alone. Not unlike chronic degenerative diseases, social determinants play a central role in the spread of infectious diseases. This is most obvious for agents that are transmitted from person to person, as they necessarily require social interaction to transmit. Social practices such as dietary or hygienic behaviour, as well as processes and structures at the societal level, such as the regulation of food production, are also essential for pathogens that are ingested through food or water (van Seventer & Hochberg, 2017). And like all diseases in which the social environment has a strong influence on their development, social inequalities can be observed for infectious diseases. To take up an already mentioned example: if one compares the main causes of death between high- and low-income countries, the burden of disease due to infections is exorbitantly higher in low-income countries. In these countries, six of the ten main causes of death are communicable diseases (e.g. lower respiratory infections, diarrhoeal disease, malaria, HIV), while in high-income countries there is only one (lower respiratory infections) (World Health Organization, 2022). Poverty and deprivation have always been a breeding ground for infectious diseases and therefore a social perspective is necessary when dealing with these diseases (Mamelund & Dimka, 2021; Quinn & Kumar, 2014). This applies to both the cross-sectional and the longitudinal view in a life-course perspective. Nevertheless, in the research on health inequality in general as well as in research oriented towards the ideas of life-course epidemiology in particular, infectious diseases have so far played a minor role, as the focus of previous life-course epidemiological research was primarily on non-communicable diseases. With the global crisis 349

350  Handbook of health inequalities across the life course of the COVID-19 pandemic this could change, and in the two years of the pandemic so far, numerous new research papers on social inequality in COVID-19 have been published. They allow new insights into the extent and causes of health inequalities not only for the current pandemic, but also in relation to infectious diseases in general. Nevertheless, research has not yet come to a point where a coherent life-course picture of the origins of inequalities in infectious diseases can be drawn. Many questions remain open and therefore this chapter is somewhat preliminary. It begins with a description of basic characteristics of infectious diseases and the introduction of the key terminology. It then summarises the research on social inequalities in infectious diseases and presents a preliminary aetiological model. This model is initially cross-sectional. Based on general concepts of life-course epidemiology, the concluding sections then discuss which longitudinal influences act on infectious diseases and whether these could contribute to the emergence of social inequalities in disease.

GENERAL CHARACTERISTICS OF INFECTIOUS DISEASE CAUSATION Infectious diseases have peculiarities that must be taken into account when analysing their aetiology. The essential difference compared to non-communicable diseases is the necessary presence of a specific infectious agent (or pathogen). These agents have biologically different natures, but can be roughly divided into viruses, bacteria, fungi, parasites and other forms of microbes. Agents can colonise a host – in the case of this chapter, a human being – and possibly infect it, that is infiltrate host tissue, and trigger a pathogenic reaction or disease (= infectious disease) (Horsburgh & Mahon, 2008). The third variable is the environment in which both pathogen and host are present and which has a significant influence on whether transmission of the pathogen to the host occurs and what the consequences are. This so-called epidemiological triad of agent, host, environment is an old concept in epidemiology to describe and analyse infectious disease causation (Snieszko, 1974). Following a concept paper by van Seventer and Hochberg (2017), the three variables of the epidemiological triad can be assigned properties that determine the probability of infection and the severity of the course of a disease. First, the biological properties of the agent must be considered. They can be described along the temporal sequence of exposure, colonisation, infection, disease and outcome. Crucial at the beginning is infectivity, which describes the potential of the agent to actually infect a host after exposure. If an infection occurs, the pathogenicity of the agent determines whether a symptomatic disease occurs at all. Closely related to this is virulence, that is the extent of pathogenic damage or the resulting severity of the symptomatic disease as it progresses (Barreto et al., 2006; Krämer et al., 2010). These properties are determined to a certain extent by the biology of the pathogen. However, this does not mean that they are static. Rather, they can change through mutations as an evolutionary adaptation process – rapidly, as in the case of the SARS-CoV-2 virus, or rather slowly, as in the case of longer-lived parasites, depending on the life cycle of the pathogen. The specific mode of transmission, that is the way in which a pathogen colonises the host, can also be assigned to the agents’ properties. Here, a rough distinction can be made between direct (i.e. from person-to-person) and indirect transmission (Krämer et al., 2010). Indirect transmission means that the agent uses another vehicle on its way to colonise the host. This vehicle can be an organism (vector-born) or inanimate matter (e.g. water-borne, airborne, food-borne, object-borne) (Giesecke, 2017).

Infectious diseases across the life course: an inequalities perspective  351 A well-known example of a vector-borne disease is malaria, in which the plasmodia parasite is transmitted to the human host via the vector of the “malaria mosquito” (Anopheles gambiae). It should be noted, however, that very few pathogens are fixed to one pathway and both direct and indirect transmission are often possible at the same time (Giesecke, 2017). These properties of the agent interact with those of the host. First of all, the host can, through its behaviour, (a) avoid exposure or (b) prevent an infection if exposure has occurred, for example by eliminating superficially colonising pathogens through hand hygiene. However, the decisive factor for the overall course is the susceptibility of the host, which denotes the probability that a host gets infected if exposed to the agent (Abubakar et al., 2016). On a biological level, this is determined by the immune system and the general constitution or resilience of the organism. Both have a genetic component, but are also variable. The immune system, for example, is capable of learning and can either develop immunity through previous infections or be trained through external immunisation (i.e. vaccination) in order to be able to show a defence reaction in subsequent exposures to the same pathogen (Abubakar et al., 2016). If an infection nevertheless occurs, immunity and constitution remain important, as they also influence the course of the disease (pathogenicity and virulence). For example, older people with chronic non-infectious diseases react more frequently with more severe symptoms in case of infection than younger, healthier people do. In the example of COVID-19, certain biomedical risk factors or pre-existing conditions such as obesity or diabetes significantly increase the likelihood of a severe course of disease in the case of infection (Gao et al., 2021). Susceptibility to infection and vulnerability to severe courses of disease thus depend on biological and behavioural characteristics of the host (age, immunisation status, stress, previous illnesses, hygiene behaviour, etc.). The third variable is the environment, which influences both agent and host and also moderates the probability of whether and under what conditions agent and host meet. Environment is to be understood here in the broadest sense, that is in addition to the natural environment, it also includes the social environment with all its levels (Krickenberg & Klemperer, 2010). Properties of the natural environment are decisive for the evolutionary development and spread of pathogens. For example, numerous bacterial, viral or parasitic pathogens are bound to certain climatic zones, weather conditions and/or seasons and need certain biotopes to reproduce effectively and, if necessary, to find suitable vehicles (van Seventer & Hochberg, 2017). Equally important is the social environment, which, in the sense of an ecosocial model of disease causation, moderates the interaction of exposure and susceptibility/resilience (Krieger, 2008). The first level is the man-made physical environment which is closely interwoven with the natural environment. This refers to aspects such as the built environment, degree of urbanisation, supply infrastructure or land use, which directly influence pathogens, hosts and transmission routes (van Seventer & Hochberg, 2017). Man-made climate change, with its fundamental consequences for the global ecosystem, can also be included here (Krämer & Khan, 2010). Other social environmental determinants are technological and/or organisational systems that are related to transmission. These are, for example, production, storage and transport systems for food, water treatment systems or the handling of waste. The probability of transmission of infections is also determined by the type, duration and organisation of personal human contacts. Next, everyday living and working conditions are important. The extent of occupational and private mobility, work-related contacts, private contacts, the occupancy density of homes and neighbourhoods, childcare, education and many other areas determine both the extent and form of social contacts (and thus possible exposures) and the exposures to

352  Handbook of health inequalities across the life course vehicles (e.g. working in indoor air) and thus form the framework for possible transmissions (Arthur et al., 2017). The organisation and effectiveness of health care and the public health system are another aspect. Targeted prevention and care of infectious diseases influence all stages of the spread and development of disease and are therefore a key area of the social environment. Naturally, social norms related to the contact and/or hygiene behaviour of individuals are also highly relevant. (Arthur et al., 2017). Examples are sexual norms, conventions for distance or closeness in everyday encounters, hygiene norms, for example regarding hand washing or illness behaviour when having a symptomatic disease. This description of the social determinants of infectious diseases is not exhaustive, but it should make clear the elementary importance of social structures and practices for the spread of infectious diseases. With this knowledge, a bridge can be built to the unequal distribution of disease risks, since social determinants tend to be unequally distributed and affect certain groups in society more or less frequently (Krieger, 2008).

SOCIOECONOMIC INEQUALITIES AND INFECTIOUS DISEASE: GENERAL CONSIDERATIONS Before adopting a longitudinal perspective, it is important to clarify whether patterns of social inequality in the incidence and severity of infectious diseases can be identified. The empirical findings to date clearly point in this direction: numerous infectious diseases affect socially disadvantaged people more often, both in terms of incidence and severity (Mamelund & Dimka, 2021). Such inequalities are already historically documented, for example in Friedrich Engels’ reports on the health situation of the working class in England in 1845 or Rudolf Virchow’s report on an outbreak of typhus in Upper Silesia in 1848. Retrospective data analyses of major historic pandemics such as the Spanish flu, also show socially differential distributions, such as higher mortality rates in poorer neighbourhoods of Chicago in the autumn of 1918 (Grantz et al., 2016). This pattern has not changed significantly to date and there is much evidence of inequalities both between social classes within societies and between rich and poor countries globally. Higher incidences and severe courses in socioeconomically disadvantaged groups compared to more advantaged groups are, for example, found for seasonal influenza (Chandrasekhar et al., 2017; Lowcock et al., 2012; Manabe et al., 2012), SARS (Bucchianeri, 2010), gastrointestinal infections in children (Adams et al., 2018), severe respiratory disease in children (Rod et al., 2021), sexually transmitted diseases such as HIV or gonorrhoea (Ekholuenetale et al., 2021; Harling et al., 2013; Nakagawa et al., 2014), neglected tropical disease (Houweling et al., 2016) or hepatitis (Tosun et al., 2018). The same applies to the country comparison. The populations of low-income countries not only have higher incidences of hepatitis (Cao et al., 2021; Jacobsen & Koopman, 2005), HIV (Challacombe, 2020) or malaria (Carrasco-Escobar et al., 2021), but also severe courses of common infections such as respiratory or diarrhoeal diseases (Ugboko et al., 2020). Another recent example is the COVID-19 pandemic. Worldwide, marked socioeconomic inequalities (measured in terms of socioeconomic position, e.g. income, education, low occupational status and/or sociodemographic characteristics such as gender, race/ethnicity, migration background) in the risk of infection, hospitalisation and mortality have been documented for all age groups (Batty et al., 2020; Bergman et al., 2021;

Infectious diseases across the life course: an inequalities perspective  353 Chadeau-Hyam et al., 2020; Chang et al., 2020; Ginsburgh et al., 2021; Hoebel, Michalski, et al., 2021; Muhsen et al., 2021; Patel et al., 2020). How such inequalities emerged must be answered in detail on an agent-specific basis, since the interactions between agent–host–environment take on specific forms and the connection to societal inequality structures can be correspondingly diverse. However, general principles can be derived, and the most elaborate model of health inequalities in infectious disease to date is a model proposed by Quinn and Kumar (Quinn & Kumar, 2014). Building on an older concept by Blumenshine et al. (2008) and informed by empirical findings on epidemic and pandemic influenza waves, the two researchers propose to organise the social determinants along the central variables of the temporal course of infectious diseases. First is the probability of exposure, which depends on numerous social factors such as living conditions, occupational activities or hygiene behaviour. This is followed by influences on susceptibility, which modify the probability of infection (e.g. immune status, nutritional status), and the severity of the disease (e.g. previous illnesses). The third mediating pathway in this model is access to and quality of medical care in the case of illness, which also influence the severity and duration of the disease. Figure 22.1 shows a modified version of Quinn and Kumar’s model specifically for SARS-CoV-2 and COVID-19, respectively, prepared for this chapter. The figure displays three broader domains of mediating factors, all related to the process of infection and disease (i.e. exposure, susceptibility and vulnerability, health care; see above). It also lists single risk factors in each domain that are known from previous research to be socially unequally distributed. For example, the likelihood of exposure increases when people live together in crowded living conditions (Ahmad et al., 2020). Since the size of the home and the number of its occupants correlate with income in many regions of the world, this is a pathway that can lead from social disadvantage to infection. Factors that influence susceptibility are also found more often in socially disadvantaged people. For example, the prevalence of risk factors such as stress, smoking or obesity is generally higher in socially disadvantaged populations than in better-off population groups. The issue of immunisation is also important. Recent data on global immunisation campaigns suggest that people with low incomes, low education or in discriminated social situations are less likely to have access to vaccines and, when access is available, are less likely to take it up (Bergmann et al., 2021; Butter et al., 2021; Caspi et al., 2021; Momplaisir et al., 2021; Muhsen et al., 2021). Most factors that increase susceptibility also increase the risk of a severe course of COVID-19 (prediction). In addition to the factors already mentioned, pre-existing conditions or comorbidities are of central importance. Numerous contributions in this book clearly show that chronic degenerative disease such as hypertension, cardiovascular disease, chronic obstructive pulmonary disease or diabetes are more prevalent among people in disadvantaged social positions. At the same time, all these diseases increase the risk of a severe course of COVID-19, including case fatality, so that a plausible mediating path can be assumed here as well. Furthermore, the care system plays a role in several respects. Good care for chronic diseases, for example, could attenuate the relationship described above. COVID-19-specific care can also explain social inequalities, for example when people with low incomes do not have access to tests or medical care in case of illness (Hoebel, Grabka, et al., 2021). The aetiological model, simplistic as it is, nevertheless illustrates how and at what point in the disease inequalities might arise. The question of this chapter is whether this cross-sectional

354  Handbook of health inequalities across the life course

Source: Disease-specific adaptation from Quinn and Kumar (2014).

Figure 22.1

Model of associations between social position, mediating factors and COVID-19

model can and should be extended to include the dimension of time (in the sense of development over the life course).

BASIC PRINCIPLES OF LIFE-COURSE EPIDEMIOLOGY The examination of infectious diseases with concepts and methods of life-course epidemiology has so far remained an exception, and this is even more true with regard to the development of health inequalities over the life course (Ben-Shlomo et al., 2016; Hall et al., 2002). Yet assumptions of life-course epidemiology are transferable – even if they were primarily developed for researching the development of chronic diseases over the course of people’s lives. Before discussing this transferability, general principles of life-course epidemiology should first be brought to mind. They can then serve as a framework for later consideration of infectious diseases over the life course. For the purposes of this chapter, it will be sufficient to stick to simple principles and not to exhaust the entire wealth of current epidemiological knowledge on the life course. The core idea of this research paradigm is that time plays an elementary role in the development of diseases and must accordingly be explicitly taken into account in causal research (Ben-Shlomo et al., 2016). From this, life course epidemiology derives various principles. A classical hypothesis is that the effect of an exposure on health depends on the timing of

Infectious diseases across the life course: an inequalities perspective  355 its impact (Ben-Shlomo & Kuh, 2002; Burton-Jeangros et al., 2015; Kuh et al., 2003). This timing determines in particular the degree of possible pathogenic damage. The assumption is that there are certain sensitive periods in the life course in which people’s health is particularly susceptible to external and internal disturbances. If exposures act in such periods, they can cause more lasting damage than in other periods. The focus is usually on early phases of life, such as pregnancy or infancy, but sensitive periods can also occur later in life. In this context, a distinction is often made between critical and sensitive periods (Kuh et al., 2003). Critical periods are time windows with special significance for the further development of health, for example the maturation steps of the foetus during pregnancy. If disturbances occur during this time, their consequences may be no longer reversible (i.e. biological programming). Sensitive periods are also times of increased vulnerability, but the effects of disruptions are less absolute and may disappear or at least be alleviated over time. Another central idea is that exposures or first subclinical damage in earlier phases of the life course influence later health trajectories as a whole, as well as increasing the probability of the occurrence of individual diseases in later phases of life (lag between exposure and disease). The respective time periods can vary depending on the underlying pathogenic processes, but in certain cases can even span many decades. For example, negative influences on the development of the foetus may set the course for chronic degenerative diseases in older adulthood (Barker, 1990). It is also possible that exposures and their consequences skip generations, for example when traumas of grandparents still shape the mental health of grandchildren (Bowers & Yehuda, 2016). However, the correlations are not necessarily linear or deterministic. Rather, it is postulated that additional factors that occur later in life can have an influence by either mediating the effect of early exposures or modifying their effect (Ben-Shlomo & Kuh, 2002). This idea of mediation and modification leads to another principle, namely accumulation. Especially in the case of chronic degenerative diseases, the aetiology is usually complex and it is not a single exposure that triggers the disease, but the interaction of several exposures over time. Such risks can accumulate and only when sufficient component causes come together is the critical threshold for pathogenic damage crossed (Blane et al., 2007). A variant of this concept is the pathway model. Here, certain trajectories in the life course lead to later exposures, for example when a person acquires a low educational qualification and then has to perform physically demanding work in adult life because other, less harmful activities are not available to him or her. This example is not chosen at random. Accumulation and pathways are closely linked to social structure and social dynamics in the concepts of life-course epidemiology. Social characteristics may thus predetermine the accumulation of risks on the one hand, and, at a long run, social transitions and trajectories are component parts of aetiological pathways that lead to health-relevant exposures, mediators and modifiers. Life-course epidemiology has therefore dealt with the aspect of social inequality early in its development. This can be interpreted as a necessity, since the diseases that have been researched with the new methods and approaches are all socially unequally distributed (e.g. ischaemic heart disease, diabetes, obesity). The insight that health inequalities in adulthood may at least partly be attributed to disadvantages in childhood and adolescence was an important milestone in health inequality research that opened up new explanatory possibilities. Health inequalities arise over the life course when people in low social positions experience health-related risks more frequently during critical or sensitive periods or systematically accumulate more risks over time. Various models exist on the emergence of social inequalities in health during the life course, describing how socioeconomic position, risk factors and diseases

356  Handbook of health inequalities across the life course interact over time. With regard to the accumulation of risks over the life course, it is assumed that this is strongly socially patterned (Bartley, 2017; Ferraro & Shippee, 2009; Larson et al., 2018). Social disadvantage in childhood and adolescence is associated with increased health burdens in this period, which marks the beginning of accumulation (Kelly-Irving & Delpierre, 2021). This then continues, as social disadvantage in early life makes disadvantage in later life more likely, which in turn is accompanied by higher health burdens, and so on. In addition, health burdens in the life course have an influence on social mobility and can perpetuate or reinforce social disadvantage (for example, if a chronic illness prevents the acquisition of a higher educational qualification). Thus, complex chains of risk can emerge, in which social and health trajectories combine and ultimately lead to socially unequal patterns of disease across the life course (Arcaya et al., 2015; Ferraro & Shippee, 2009; Larson et al., 2018). Empirically, such cumulative inequalities or chains of risk have been demonstrated in many cases, ranging from correlations between social disadvantage in childhood and premature mortality in older adulthood to inequities in complex health trajectories of individual biological variables (e.g. cardiovascular risk factors) over longer periods of time (Bartley, 2017; Karlamangla et al., 2005; Pavalko & Caputo, 2013). However, these general concepts need to be specified for the study of individual diseases. In a recent article, Kelly-Irving and Delpierre call for a close look at the “social-to-biological processes” and for the investigation of which socially patterned exposures have which biological consequences and how these consequences reverberate over time (Kelly-Irving & Delpierre, 2021). How this idea could be applied for infectious disease is the subject of the next section.

APPLYING BASIC PRINCIPLES OF LIFE-COURSE EPIDEMIOLOGY TO INFECTIOUS DISEASE AETIOLOGY AND INEQUALITIES As an introduction to life-course epidemiology of infectious diseases, a look at the age-specific development of individual infectious diseases should demonstrate the significance of a longitudinal view on the emergence of infectious disease. Although there are numerous pathogens that can infect and make people ill at any age, a majority of them show certain age peaks in the probability of infection. Well-known examples are childhood diseases such as rubella, sexually transmitted diseases with a high incidence in adulthood or the high susceptibility to certain bacterial diseases in older age. The variance in the life course is clearly visible in the descriptive representation and there is hardly an infectious disease that would not show a specific age course. Figure 22.2 lists four examples of this. The data come from a register of notifiable diseases in 2020 in Germany. The cumulative age-specific incidences are shown, each of which reveals very characteristic age curves. Systematic age differences also exist in the severity of the disease (Glynn & Moss, 2020), although these do not necessarily have to show the same pattern as the incidences. For example, people may be more likely to be infected at a younger age but not develop symptoms, while older people may be less likely to be infected but become more severely ill when infected. A recent example of this is the SARS-CoV-2 variant Omicron (B.1.1.529), which showed the same pattern in the winter of 2021/2022.

Infectious diseases across the life course: an inequalities perspective  357

Source: RKI, 2021 (Robert Koch Institut, 2021).

Figure 22.2

Cumulative incidences per 100,000 for COVID-19, norovirus, chickenpox and hepatitis B by age groups in Germany in 2020

Critical and Sensitive Periods Whether and why there is increased exposure, infectivity, susceptibility or vulnerability at certain stages in the life course must be answered on a pathogen-specific basis. Jia and col-

358  Handbook of health inequalities across the life course leagues suggest, for example, that the epidemiological triad for a particular pathogen should be considered separately for all developmental stages of human ageing because, on the one hand, all components of the triad could have an influence on whether and why an age-specific sensitive or critical period occurs and, on the other hand, different components of the triad could have different effects at different age stages (Jia et al., 2020). It is not possible to examine this in detail here. However, two variables seem to be particularly relevant: the probability of exposure and the stage of development of the immune system. Although it is secondary to the risk of exposure in the temporal sequence, it makes sense to start with the immune system. It is of particular importance because it influences the probability of infection independently of the pathogen and at the same time represents the key biological mechanism for critical and sensitive periods. In terms of a life-course perspective, it should be noted that the immune system is in a constant state of change in the course of biological ageing (Hall et al., 2002). Particularly sensitive periods are the beginning of life and old age. In the first years of life, both the innate and the adaptive immune system of humans are not yet fully developed, so that an immune response to defend against pathogens is absent or at least underdeveloped (Simon et al., 2015; Zhang et al., 2017). This makes perfect biological sense, because the adaptive immune system must first be trained and antibodies developed after birth, which requires a certain susceptibility. On the other hand, it also makes young children particularly susceptible to certain pathogens, for example to gastro interstitial pathogens such as norovirus (Figure 22.2) or respiratory pathogens such as respiratory syncytial virus (RSV) (Glynn & Moss, 2020; Ruckwardt et al., 2016). After the immune system reaches its peak performance in adolescence and young adulthood, its functionality decreases again significantly in old age (Hall et al., 2002; Simon et al., 2015). Thus, the older as well as the very young age can be interpreted as a sensitive period. Certain periods could even be considered critical periods, for example, there is evidence that intrauterine developmental disorders can impair the immune system throughout life (Hall et al., 2002). If we now look at the factors that have a negative influence on the immune system during these periods, it quickly becomes clear that many of them are socially unequally distributed. An important example is malnutrition and undernourishment, which severely impair the development of the immune system in children, having lifelong consequence for the functionality of the immune system (Rytter et al., 2014). Malnutrition in childhood is clearly a consequence of poverty (in low- as well as in high-income countries) and can thus represent a mediating mechanism. This also applies to other factors with a negative influence on the immune system, such as maternal stress during pregnancy (Marques et al., 2015) or chronic psychosocial stress due to disadvantaged living conditions, both of which are often socially unequally distributed. Individual longitudinal studies have then also been able to demonstrate associations between social disadvantage in childhood and impaired immune function in older age (Austin et al., 2018; West et al., 2012). Such associations can also occur in the sensitive period of old age, since factors such as malnutrition or stress do not only have a negative effect on the immune system in childhood (Wick & Grubeck-Loebenstein, 1997). Another aspect is indirect immunisation through vaccinations. Effective vaccinations have been developed for diseases that pose a particular threat to people in young and old age, which can prevent infections in vulnerable phases of life and – in the case of many childhood vaccinations – can in some instances establish lifelong immunity. However, despite global efforts, coverage of these vaccines remains socially unequal, with coverage mostly lower in poorer countries than in richer ones and lower in socioeconomically disadvantaged populations than in better-off

Infectious diseases across the life course: an inequalities perspective  359 populations. This inequality exists both for childhood vaccinations (Bobo et al., 2022) and for vaccines that protect older people in particular (e.g. influenza or COVID-19) (Ayers et al., 2021; Muhsen et al., 2021). The second variable is the probability of exposure. Whether exposure occurs depends on environmental conditions that change over the life course. This is especially true for social environments. Infants, for example, have significantly fewer social contacts than a working person and are therefore less likely to be exposed to certain pathogens. Behaviour also changes over the life stages, which has an influence on the probability of pathogen contacts. An example is sexually transmitted diseases, whose transmission is largely limited to the sexually active phase of life. Links to social inequality can now be drawn here. They arise from the logic of transmission, which usually takes place in concrete social interactions or social practices (e.g. eating, drinking). Many of these interaction contexts and practices are in turn socio-culturally shaped, including class-specific differences. This applies both to behaviour and to the respective living and working conditions that constitute the environment for transmission. Poverty is a factor that is particularly associated with infectious environments. For example, children in poor families are more likely to grow up in cramped housing with critical hygiene conditions, which can significantly increase the risk of exposure (Evans, 2004). This issue can be exacerbated by clustering, as living with people who have an increased individual susceptibility to infection due to poverty and/or health conditions also increases one’s own risk of infection (both direct and indirect transmission). Low material resources can also lead to a lack of access to clean water or non-contaminated food, resulting in the need to resort to potentially contaminated food (Evans, 2004). Furthermore, personal behaviour plays a role. Socially disadvantaged people sometimes show riskier hygiene behaviour and have poorer access to information on how to avoid infections, as well as lower overall health literacy, than more privileged groups of people (Crichton et al., 2014; Eaton et al., 2003; Stormacq et al., 2019). Lag between Exposure and Disease The principle of the time lag between exposure and disease can also play a role in infectious diseases, but it is less central than in chronic diseases. The reason is that the incubation period, i.e. the time between exposure and the onset of the first symptoms, is only long enough for a few pathogens to justify a general consideration in the life course (Horsburgh & Mahon, 2008). Nevertheless, there are individual pathogens that only become apparent and trigger symptoms after years in the host. These are often bacterial (e.g. mycobacterium tuberculosis) pathogens, but individual viruses, such as varicella-zoster virus, can also remain unapparent for long periods after initial infection (Behr et al., 2018; Laing et al., 2018). However, the extent to which incubation periods for such diseases show social inequality has hardly been researched. Yet, an influence of social disadvantage cannot be ruled out in view of the connections between social factors and the immune system already described. Accumulation Moving on to the principle of accumulation and pathways, a fundamental difference from the analysis of chronic diseases must first be highlighted. While the aetiology of most chronic diseases is multifactorial in most cases, an infectious disease is always monocausal. This is true at

360  Handbook of health inequalities across the life course least with regard to the exposure, which must always be the specific pathogen. In this respect, a pure accumulation of exposures is not possible, even if sometimes a repeated infection with the same pathogen can increase the probability of disease. Nevertheless, accumulation is also relevant here, because at least the susceptibility and vulnerability are influenced by multiple factors independent of the pathogen. Of particular importance is again the immune system, which, as has already been described, changes over the course of life. For example, immunity against certain pathogens can be built up actively or passively. As a positive form of accumulation, immunization at a young age can, for example, protect against diseases in the following phase of life. Accumulation in the negative sense, on the other hand, occurs when factors such as malnutrition, stress or substance use negatively influence the immune system, thus increasing susceptibility over the life course and thus susceptibility to infection in later exposure. More broadly, behaviours could also be understood as factors that accumulate over time. Many of the behaviours that play a role in infections, such as hygiene, personal preventive action (e.g. attitudes towards vaccination) or sexual behaviour, are shaped in childhood, adolescence and young adulthood. If risk behaviour is formed there, this can increase the risk of exposure in later phases of life. The fact that such immune characteristics and behaviours are often socially unequally distributed has already been mentioned. However, it is also important to consider effect modifiers that influence the impact of disadvantage over the life course. For example, there are studies that show that factors such as a close parent–child bond are able to modify the increased susceptibility to infection as a result of socioeconomic disadvantage in adulthood (Cohen et al., 2020). Other factors that influence susceptibility and virulence also develop over the life course and can thus be classified in a model of accumulation. An important point in this context is that infectious diseases interact with other, non-infectious diseases in the life course. This relationship is two-sided. First, non-infectious diseases can influence the likelihood and severity of infections. People who develop diseases such as diabetes or chronic obstructive pulmonary disease over the life course are both more susceptible to becoming infected when exposed to a pathogen and more at risk of suffering a severe course when infected. Currently, this can be observed in connection with the SARS-CoV-2 virus, which makes its host more likely to become ill and more severely if pre-existing conditions such as diabetes, obesity or hypertension are present (Gao et al., 2021). Since most of these risk factors and pre-existing disease are socially unequally distributed, they are likely to play a fundamental role in the accumulation of infectious disease risks over the life course. How inequality arises over the life course of these non-infectious risk factors can be seen from numerous contributions in this book. Secondly, infections in the life course can also be part of the aetiology of chronic degenerative diseases. For example, frequent infections in childhood and adolescence are suspected to promote the development of atherosclerosis and thus cardiovascular diseases (Burgner et al., 2015). Another example is an infection with cancer-causing human papillomavirus (HPV) types, which is often acquired at a young age, remains inconspicuous for a long time, but then significantly increases the risk of cervical cancer (Burd, 2003). Longitudinal studies suggest that such infections affect children from socially disadvantaged families more frequently and that this circumstance could be part of the explanation for the socially unequal distribution of diseases in adulthood (Liu et al., 2016). In this respect, a life-course approach to infectious diseases is also important for the study of social inequalities in non-communicable diseases.

Infectious diseases across the life course: an inequalities perspective  361

CONCLUSION Infections and infectious diseases accompany humans throughout their lives and, historically, they have been the defining cause of death in all age groups for the longest period of evolutionary development. Accordingly, the fight against infectious diseases was and is a central public health issue. It is important to consider in this respect that infectious diseases affect people in disadvantaged socio-economic positions more often. A recent example is pronounced inequalities in COVID-19 incidence and severity. This chapter describes how infectious diseases develop over the life course and how they can be explained with common concepts of life-course epidemiology. Behind this background, life-course-related reasons for the unequal distribution of infectious diseases are discussed. It becomes clear that socially disadvantaged people experience more exposures to pathogens during sensitive periods and have a higher susceptibility (mainly as a result of reduced immune responses and pre-existing conditions) to infection and severe disease. Accumulation of risk over the life course is also socially shaped. Against this background, increased longitudinal research on health inequalities in infectious diseases seems promising, for example to identify “infectious environments” in childhood or the influence that social disadvantage has on the immune system at different stages of the life course. Yet, it is also important to keep the global dimension of this topic in mind. Infectious diseases affect populations in low-income countries to a much greater extent and therefore researchers from these countries should be enabled to investigate the background of social inequalities on site with complex longitudinal study designs. However, and this applies to all countries, it also became clear that a life-course approach does not necessarily make sense per se. Especially for diseases with short latency periods and those that can occur at all stages of life, cross-sectional studies are still necessary.

REFERENCES Abubakar, I., Cohen, T., Stagg, H. R., & Rodrigues, L. C. (Eds.). (2016). Oxford specialist handbooks in infectious diseases: Infectious disease epidemiology (First edition). Oxford University Press. Adams, N. L., Rose, T. C., Hawker, J., Violato, M., O’Brien, S. J., Barr, B., Howard, V. J. K., Whitehead, M., Harris, R., & Taylor-Robinson, D. C. (2018). Relationship between socioeconomic status and gastrointestinal infections in developed countries: A systematic review and meta-analysis. PLOS ONE, 13(1), e0191633. https://​doi​.org/​10​.1371/​journal​.pone​.0191633 Ahmad, K., Erqou, S., Shah, N., Nazir, U., Morrison, A. R., Choudhary, G., & Wu, W-C. (2020). Association of poor housing conditions with COVID-19 incidence and mortality across US counties. PLOS ONE, 15(11), e0241327. https://​doi​.org/​10​.1371/​journal​.pone​.0241327 Arcaya, M. C., Arcaya, A. L., & Subramanian, S. V. (2015). Inequalities in health: Definitions, concepts, and theories. Global Health Action, 8, 27106. https://​doi​.org/​10​.3402/​gha​.v8​.27106 Arthur, R. F., Gurley, E. S., Salje, H., Bloomfield, L. S. P., & Jones, J. H. (2017). Contact structure, mobility, environmental impact and behaviour: The importance of social forces to infectious disease dynamics and disease ecology. Philosophical Transactions of the Royal Society of London. Series B, Biological Sciences, 372(1719). https://​doi​.org/​10​.1098/​rstb​.2016​.0454 Austin, M. K., Chen, E., Ross, K. M., McEwen, L. M., MacIsaac, J. L., Kobor, M. S., & Miller, G. E. (2018). Early-life socioeconomic disadvantage, not current, predicts accelerated epigenetic aging of monocytes. Psychoneuroendocrinology, 97. Ayers, C. K., Kondo, K. K., Williams, B. E., Kansagara, D., Advani, S. M., Smith, M., Young, S., & Saha, S. (2021). Disparities in H1N1 vaccination rates: A systematic review and evidence synthesis to inform COVID-19 vaccination efforts. Journal of General Internal Medicine, 36(6), 1734–1745. https://​doi​.org/​10​.1007/​s11606​-021​-06715​-7

362  Handbook of health inequalities across the life course Barker, D. J. (1990). The fetal and infant origins of adult disease. BMJ: British Medical Journal, 301(6761), 1111. https://​doi​.org/​10​.1136/​bmj​.301​.6761​.1111 Barreto, M. L., Teixeira, M. G., & Carmo, E. H. (2006). Infectious diseases epidemiology. Journal of Epidemiology and Community Health, 60(3), 192–195. https://​doi​.org/​10​.1136/​jech​.2003​.011593 Bartley, M. (2017). Health inequality: An introduction to concepts, theories and methods (Second edition). Polity. Batty, G. D., Deary, I. J., Luciano, M., Altschul, D. M., Kivimäki, M., & Gale, C. R. (2020). Psychosocial factors and hospitalisations for COVID-19: Prospective cohort study based on a community sample. Brain, Behavior, and Immunity, 89, 569–578. https://​doi​.org/​10​.1016/​j​.bbi​.2020​.06​.021 Behr, M. A., Edelstein, P. H., & Ramakrishnan, L. (2018). Revisiting the timetable of tuberculosis. BMJ, 362, k2738. https://​doi​.org/​10​.1136/​bmj​.k2738 Ben-Shlomo, Y., Cooper, R., & Kuh, D. (2016). The last two decades of life course epidemiology, and its relevance for research on ageing. International Journal of Epidemiology, 45(4), 973–988. https://​ doi​.org/​10​.1093/​ije/​dyw096 Ben-Shlomo, Y., & Kuh, D. (2002). A life course approach to chronic disease epidemiology: Conceptual models, empirical challenges and interdisciplinary perspectives. International Journal of Epidemiology, 31(2), 285–293. Bergman, J., Ballin, M., Nordström, A., & Nordström, P. (2021). Risk factors for COVID-19 diagnosis, hospitalization, and subsequent all-cause mortality in Sweden: A nationwide study. European Journal of Epidemiology, 36(3), 287–298. https://​doi​.org/​10​.1007/​s10654​-021​-00732​-w Bergmann, M., Hannemann, T.-V., Bethmann, A., & Schumacher, A. (2021). Determinants of SARS-CoV-2 vaccinations in the 50+ population. SHARE Working Paper, 72. Blane, D., Netuveli, G., & Stone, J. (2007). The development of life course epidemiology. Revue d’Épidémiologie et de Santé Publique, 55(1), 31–38. https://​doi​.org/​10​.1016/​j​.respe​.2006​.12​.004 Blumenshine, P., Reingold, A., Egerter, S., Mockenhaupt, R., Braveman, P., & Marks, J. (2008). Pandemic influenza planning in the United States from a health disparities perspective. Emerging Infectious Diseases, 14(5), 709–715. https://​doi​.org/​10​.3201/​eid1405​.071301 Bobo, F. T., Asante, A., Woldie, M., Dawson, A., & Hayen, A. (2022). Child vaccination in sub-Saharan Africa: Increasing coverage addresses inequalities. Vaccine, 40(1), 141–150. https://​doi​.org/​10​.1016/​ j​.vaccine​.2021​.11​.005 Bowers, M. E., & Yehuda, R. (2016). Intergenerational transmission of stress in humans. Neuropsychopharmacology: Official Publication of the American College of Neuropsychopharmacology, 41(1), 232–244. https://​doi​.org/​10​.1038/​npp​.2015​.247; Bucchianeri, G. W. (2010). Is SARS a poor man’s disease? Socioeconomic status and risk factors for SARS transmission. Forum for Health Economics & Policy, 13(2), 115. https://​doi​.org/​10​.2202/​1558​ -9544​.1209 Bundesgesundheitsministerium. (2021). Arbeitsunfähigkeit: Fälle und Tage nach Diagnosen 2019. Ergebnisse der Krankheitsartenstatistik der gesetzlichen Krankenversicherung. https://​www​ .bundesge​sundheitsm​inisterium​.de/​fileadmin/​Dateien/​3​_Downloads/​Statistiken/​GKV/​G​eschaeftse​ rgebnisse/​Diagnosen​_und​_Alter​_2019​.pdf Burd, E. M. (2003). Human papillomavirus and cervical cancer. Clinical Microbiology Reviews, 16(1), 1–17. https://​doi​.org/​10​.1128/​CMR​.16​.1​.1​-17​.2003 Burgner, D. P., Sabin, M. A., Magnussen, C. G., Cheung, M., Sun, C., Kähönen, M., Hutri-Kähönen, N., Lehtimäki, T., Jokinen, E., Laitinen, T., Viikari, J. S. A., Raitakari, O. T., & Juonala, M. (2015). Early childhood hospitalisation with infection and subclinical atherosclerosis in adulthood: The Cardiovascular Risk in Young Finns Study. Atherosclerosis, 239(2), 496–502. https://​doi​.org/​10​ .1016/​j​.atherosclerosis​.2015​.02​.024 Burton-Jeangros, C., Cullati, S., Sacker, A., & Blane, D. N. (Eds.). (2015). A life course perspective on health trajectories and transitions (First edition). Springer. Butter, S., McGlinchey, E., Berry, E., & Armour, C. (2021). Psychological, social, and situational factors associated with COVID-19 vaccination intentions: A study of UK key workers and non-key workers. British Journal of Health Psychology. Advance online publication. https://​doi​.org/​10​.1111/​ bjhp​.12530

Infectious diseases across the life course: an inequalities perspective  363 Cao, G., Jing, W., Liu, J., & Liu, M. (2021). The global trends and regional differences in incidence and mortality of hepatitis A from 1990 to 2019 and implications for its prevention. Hepatology International, 15(5), 1068–1082. https://​doi​.org/​10​.1007/​s12072​-021​-10232​-4 Carrasco-Escobar, G., Fornace, K., & Benmarhnia, T. (2021). Mapping socioeconomic inequalities in malaria in Sub-Sahara African countries. Scientific Reports, 11(1), 15121. https://​doi​.org/​10​.1038/​ s41598​-021​-94601​-x Caspi, G., Dayan, A., Eshal, Y., Liverant-Taub, S., Twig, G., Shalit, U., Lewis, Y., Shina, A., & Caspi, O. (2021). Socioeconomic disparities and COVID-19 vaccination acceptance: A nationwide ecologic study. Clinical Microbiology and Infection: The Official Publication of the European Society of Clinical Microbiology and Infectious Diseases, 27(10), 1502–1506. https://​doi​.org/​10​.1016/​j​.cmi​ .2021​.05​.030 Chadeau-Hyam, M., Bodinier, B., Elliott, J., Whitaker, M. D., Tzoulaki, I., Vermeulen, R., Kelly-Irving, M., Delpierre, C., & Elliott, P. (2020). Risk factors for positive and negative COVID-19 tests: A cautious and in-depth analysis of UK biobank data. International Journal of Epidemiology, 49(5), 1454–1467. https://​doi​.org/​10​.1093/​ije/​dyaa134 Challacombe, S. J. (2020). Global inequalities in HIV infection. Oral Diseases, 26(Suppl 1), 16–21. https://​doi​.org/​10​.1111/​odi​.13386 Chandrasekhar, R., Sloan, C., Mitchel, E., Ndi, D., Alden, N., Thomas, A., Bennett, N. M., Kirley, P. D., Hill, M., Anderson, E. J., Lynfield, R., Yousey-Hindes, K., Bargsten, M., Zansky, S. M., Lung, K., Schroeder, M., Monroe, M., Eckel, S., Markus, T. M., … Lindegren, M. L. (2017). Social determinants of influenza hospitalization in the United States. Influenza and Other Respiratory Viruses, 11(6), 479–488. https://​doi​.org/​10​.1111/​irv​.12483 Chang, S., Pierson, E., Koh, P. W., Gerardin, J., Redbird, B., Grusky, D., & Leskovec, J. (2020). Mobility network models of COVID-19 explain inequities and inform reopening. Nature. Advance online publication. https://​doi​.org/​10​.1038/​s41586​-020​-2923​-3 Cohen, S., Chiang, J. J., Janicki-Deverts, D., & Miller, G. E. (2020). Good relationships with parents during childhood as buffers of the association between childhood disadvantage and adult susceptibility to the common cold. Psychosomatic Medicine, 82(6), 538–547. https://​doi​.org/​10​.1097/​PSY​ .0000000000000818 Crichton, J., Hickman, M., Campbell, R., Heron, J., Horner, P., & Macleod, J. (2014). Prevalence of chlamydia in young adulthood and association with life course socioeconomic position: Birth cohort study. PLOS ONE, 9(8), e104943. https://​doi​.org/​10​.1371/​journal​.pone​.0104943 Eaton, L., Flisher, A. J., & Aarø, L. E. (2003). Unsafe sexual behaviour in South African youth. Social Science & Medicine, 56(1), 149–165. https://​doi​.org/​10​.1016/​s0277​-9536(02)00017​-5 Ekholuenetale, M., Onuoha, H., Ekholuenetale, C. E., Barrow, A., & Nzoputam, C. I. (2021). Socioeconomic inequalities in Human Immunodeficiency Virus (HIV) sero-prevalence among women in Namibia: Further analysis of population-based data. International Journal of Environmental Research and Public Health, 18(17). https://​doi​.org/​10​.3390/​ijerph18179397 Evans, G. W. (2004). The environment of childhood poverty. The American Psychologist, 59(2), 77–92. https://​doi​.org/​10​.1037/​0003​-066X​.59​.2​.77 Ferraro, K. F., & Shippee, T. P. (2009). Aging and cumulative inequality: How does inequality get under the skin? The Gerontologist, 49(3), 333–343. https://​doi​.org/​10​.1093/​geront/​gnp034 Gao, Y.‑D., Ding, M., Dong, X., Zhang, J.‑J., Kursat Azkur, A., Azkur, D., Gan, H., Sun, Y.‑L., Fu, W., Li, W., Liang, H.‑L., Cao, Y.‑Y., Yan, Q., Cao, C., Gao, H.‑Y., Brüggen, M.‑C., van de Veen, W., Sokolowska, M., Akdis, M., & Akdis, C. A. (2021). Risk factors for severe and critically ill COVID-19 patients: A review. Allergy, 76(2), 428–455. https://​doi​.org/​10​.1111/​all​.14657 Giesecke, J. (2017). Modern infectious disease epidemiology (Third edition). CRC Press. Ginsburgh, V., Magerman, G., & Natali, I. (2021). Covid-19 and the role of inequality in French regional departments. European Journal of Health Economics: HEPAC: Health Economics in Prevention and Care. Advance online publication. https://​doi​.org/​10​.1007/​s10198​-020​-01254​-0 Glynn, J. R., & Moss, P. A. H. (2020). Systematic analysis of infectious disease outcomes by age shows lowest severity in school-age children. Scientific Data, 7(1), 329. https://​doi​.org/​10​.1038/​s41597​-020​ -00668​-y Grantz, K. H., Rane, M. S., Salje, H., Glass, G. E., Schachterle, S. E., & Cummings, D. A. T. (2016). Disparities in influenza mortality and transmission related to sociodemographic factors within

364  Handbook of health inequalities across the life course Chicago in the pandemic of 1918. Proceedings of the National Academy of Sciences of the United States of America, 113(48), 13839–13844. https://​doi​.org/​10​.1073/​pnas​.1612838113 Hall, A. J., Yee, L. J., & Thomas, S. L. (2002). Life course epidemiology and infectious diseases. International Journal of Epidemiology, 31(2), 300–301. Harling, G., Subramanian, S., Bärnighausen, T., & Kawachi, I. (2013). Socioeconomic disparities in sexually transmitted infections among young adults in the United States: Examining the interaction between income and race/ethnicity. Sexually Transmitted Diseases, 40(7), 575–581. https://​doi​.org/​10​ .1097/​OLQ​.0b013e31829529cf Hoebel, J., Grabka, M. M., Schröder, C., Haller, S., Neuhauser, H., Wachtler, B., Schaade, L., Liebig, S., Hövener, C., & Zinn, S. (2021). Socioeconomic position and SARS-CoV-2 infections: seroepidemiological findings from a German nationwide dynamic cohort. https://​doi​.org/​10​.1101/​2021​.06​.21​ .21259001 Hoebel, J., Michalski, N., Diercke, M., Hamouda, O., Wahrendorf, M., Dragano, N., & Nowossadeck, E. (2021). Emerging socio-economic disparities in COVID-19-related deaths during the second pandemic wave in Germany. International Journal of Infectious Diseases: IJID: Official Publication of the International Society for Infectious Diseases, 113, 344–346. https://​doi​.org/​10​.1016/​j​.ijid​.2021​ .10​.037 Horsburgh, C. R., & Mahon, B. E. (2008). Infectious disease epidemiology. In K. J. Rothman, S. Greenland, & T. L. Lash (Eds.), Modern epidemiology (Third edition) (pp. 549–563). Wolters Kluwer Lippincott Williams & Wilkins. Houweling, T. A. J., Karim-Kos, H. E., Kulik, M. C., Stolk, W. A., Haagsma, J. A., Lenk, E. J., Richardus, J. H., & de Vlas, S. J. (2016). Socioeconomic inequalities in neglected tropical diseases: A systematic review. PLoS Neglected Tropical Diseases, 10(5), e0004546. https://​doi​.org/​10​.1371/​ journal​.pntd​.0004546 Jacobsen, K. H., & Koopman, J. S. (2005). The effects of socioeconomic development on worldwide hepatitis A virus seroprevalence patterns. International Journal of Epidemiology, 34(3), 600–609. https://​doi​.org/​10​.1093/​ije/​dyi062 Jia, P., Dong, W., Yang, S., Zhan, Z., La Tu, & Lai, S. (2020). Spatial lifecourse epidemiology and infectious disease research. Trends in Parasitology, 36(3), 235–238. https://​doi​.org/​10​.1016/​j​.pt​.2019​ .12​.012 Karlamangla, A. S., Singer, B. H., Williams, D. R., Schwartz, J. E., Matthews, K. A., Kiefe, C. I., & Seeman, T. E. (2005). Impact of socioeconomic status on longitudinal accumulation of cardiovascular risk in young adults: The CARDIA Study (USA). Social Science & Medicine, 60(5), 999–1015. https://​doi​.org/​10​.1016/​j​.socscimed​.2004​.06​.056 Kelly-Irving, M., & Delpierre, C. (2021). Framework for understanding health inequalities over the life course: The embodiment dynamic and biological mechanisms of exogenous and endogenous origin. Journal of Epidemiology and Community Health, 75(12), 1181–1186. https://​doi​.org/​10​.1136/​jech​ -2021​-216430 Krämer, A., Akmatov, M., & Kretzschmar, M. (2010). Principles of infectious disease epidemiology. In A. Krämer, M. Kretzschmar, & K. Krickeberg (Eds.), Modern infectious disease epidemiology: Concepts, methods, mathematical models, and public health (pp. 85–99). Springer New York. Krämer, A., & Khan, M. H. (2010). Global challenges of infectious disease epidemiology. In A. Krämer, M. Kretzschmar, & K. Krickeberg (Eds.), Modern infectious disease epidemiology: Concepts, methods, mathematical models, and public health (pp. 23–38). Springer New York. Krickenberg, K., & Klemperer, D. (2010). Social risk factors. In A. Krämer, M. Kretzschmar, & K. Krickeberg (Eds.), Modern infectious disease epidemiology: Concepts, methods, mathematical models, and public health (pp. 100–115). Springer New York. Krieger, N. (2008). Proximal, distal, and the politics of causation: What’s level got to do with it? American Journal of Public Health, 98(2), 221–230. https://​doi​.org/​10​.2105/​AJPH​.2007​.111278 Kuh, D., Ben-Shlomo, Y., Lynch, J., Hallqvist, J., & Power, C. (2003). Life course epidemiology. Journal of Epidemiology and Community Health, 57(10), 778–783. https://​doi​.org/​10​.1136/​jech​.57​ .10​.778 Laing, K. J., Ouwendijk, W. J. D., Koelle, D. M., & Verjans, G. M. G. M. (2018). Immunobiology of Varicella-Zoster virus infection. Journal of Infectious Diseases, 218(Suppl 2), S68–S74. https://​doi​ .org/​10​.1093/​infdis/​jiy403

Infectious diseases across the life course: an inequalities perspective  365 Larson, K., Russ, S. A., Kahn, R. S., Flores, G., Goodman, E., Cheng, T. L., & Halfon, N. (2018). Health disparities: A life course health development perspective and future research directions. In N. Halfon, C. B. Forrest, E. M. Faustman, & R. M. Lerner (Eds.), Handbook of life course health development (pp.  499–520). Springer. https://​doi​.org/​10​.1007/​978​-3​-319​-47143​-3​_20 Liu, R. S., Burgner, D. P., Sabin, M. A., Magnussen, C. G., Cheung, M., Hutri-Kähönen, N., Kähönen, M., Lehtimäki, T., Jokinen, E., Laitinen, T., Taittonen, L., Dwyer, T., Viikari, J. S. A., Kivimäki, M., Raitakari, O. T., & Juonala, M. (2016). Childhood infections, socioeconomic status, and adult cardiometabolic risk. Pediatrics, 137(6). https://​doi​.org/​10​.1542/​peds​.2016​-0236 Lowcock, E. C., Rosella, L. C., Foisy, J., McGeer, A., & Crowcroft, N. (2012). The social determinants of health and pandemic H1N1 2009 influenza severity. American Journal of Public Health, 102(8), e51–58. https://​doi​.org/​10​.2105/​AJPH​.2012​.300814 Mamelund, S.‑E., & Dimka, J. (2021). Social inequalities in infectious diseases. Scandinavian Journal of Public Health, 49(7), 675–680. https://​doi​.org/​10​.1177/​1403494821997228 Manabe, T., Higuera Iglesias, A. L., Vazquez Manriquez, M. E., Martinez Valadez, E. L., Ramos, L. A., Izumi, S., Takasaki, J., & Kudo, K. (2012). Socioeconomic factors influencing hospitalized patients with pneumonia due to influenza A(H1N1)pdm09 in Mexico. PLOS ONE, 7(7), e40529. https://​doi​ .org/​10​.1371/​journal​.pone​.0040529 Marques, A. H., Bjørke-Monsen, A‑L., Teixeira, A. L., & Silverman, M. N. (2015). Maternal stress, nutrition and physical activity: Impact on immune function, CNS development and psychopathology. Brain Research, 1617, 28–46. https://​doi​.org/​10​.1016/​j​.brainres​.2014​.10​.051 Momplaisir, F. M., Kuter, B. J., Ghadimi, F., Browne, S., Nkwihoreze, H., Feemster, K. A., Frank, I., Faig, W., Shen, A. K., Offit, P. A., & Green-McKenzie, J. (2021). Racial/ethnic differences in COVID-19 vaccine hesitancy among health care workers in 2 large academic hospitals. JAMA Network Open, 4(8), e2121931. https://​doi​.org/​10​.1001/​jamanetworkopen​.2021​.21931 Muhsen, K., Na’aminh, W., Lapidot, Y., Goren, S., Amir, Y., Perlman, S., Green, M. S., Chodick, G., & Cohen, D. (2021). A nationwide analysis of population group differences in the COVID-19 epidemic in Israel, February 2020–February 2021. The Lancet Regional Health. Europe, 7, 100130. https://​doi​ .org/​10​.1016/​j​.lanepe​.2021​.100130 Nakagawa, F., Phillips, A. N., & Lundgren, J. D. (2014). Update on HIV in Western Europe. Current HIV/AIDS Reports, 11(2), 177–185. https://​doi​.org/​10​.1007/​s11904​-014​-0198​-8 Patel, A. P., Paranjpe, M. D., Kathiresan, N. P., Rivas, M. A., & Khera, A. V. (2020). Race, socioeconomic deprivation, and hospitalization for COVID-19 in English participants of a national biobank. International Journal for Equity in Health, 19(1), 114. https://​doi​.org/​10​.1186/​s12939​-020​-01227​-y Pavalko, E. K., & Caputo, J. (2013). Social inequality and health across the life course. American Behavioral Scientist, 57(8), 1040–1056. https://​doi​.org/​10​.1177/​0002764213487344 Quinn, S. C., & Kumar, S. (2014). Health inequalities and infectious disease epidemics: A challenge for global health security. Biosecurity and Bioterrorism: Biodefense Strategy, Practice, and Science, 12(5), 263–273. https://​doi​.org/​10​.1089/​bsp​.2014​.0032 Robert Koch Institut. (2021). SurvStat@​RKI 2.0. https://​survstat​.rki​.de Rod, N. H., Bengtsson, J., Elsenburg, L. K., Taylor-Robinson, D., & Rieckmann, A. (2021). Hospitalisation patterns among children exposed to childhood adversity: A population-based cohort study of half a million children. The Lancet Public Health, 6(11), e826–e835. https://​doi​.org/​10​.1016/​ S2468​-2667(21)00158​-4 Ruckwardt, T. J., Morabito, K. M., & Graham, B. S. (2016). Determinants of early life immune responses to RSV infection. Current Opinion in Virology, 16, 151–157. https://​doi​.org/​10​.1016/​j​.coviro​.2016​ .01​.003 Rytter, M. J. H., Kolte, L., Briend, A., Friis, H., & Christensen, V. B. (2014). The immune system in children with malnutrition – a systematic review. PLOS ONE, 9(8), e105017. https://​doi​.org/​10​.1371/​ journal​.pone​.0105017 Simon, A. K., Hollander, G. A., & McMichael, A. (2015). Evolution of the immune system in humans from infancy to old age. Proceedings: Biological Sciences, 282(1821), 20143085. https://​doi​.org/​10​ .1098/​rspb​.2014​.3085 Snieszko, S. F. (1974). The effects of environmental stress on outbreaks of infectious diseases of fishes Journal of Fish Biology, 6(2), 197–208. https://​doi​.org/​10​.1111/​j​.1095​-8649​.1974​.tb04537​.x

366  Handbook of health inequalities across the life course Stormacq, C., van den Broucke, S., & Wosinski, J. (2019). Does health literacy mediate the relationship between socioeconomic status and health disparities? Integrative review. Health Promotion International, 34(5), e1–e17. https://​doi​.org/​10​.1093/​heapro/​day062 Tosun, S., Aygün, O., Özdemir, H. Ö., Korkmaz, E., & Özdemir, D. (2018). The impact of economic and social factors on the prevalence of hepatitis B in Turkey. BMC Public Health, 18(1), 649. https://​doi​ .org/​10​.1186/​s12889​-018​-5575​-6 Ugboko, H. U., Nwinyi, O. C., Oranusi, S. U., & Oyewale, J. O. (2020). Childhood diarrhoeal diseases in developing countries. Heliyon, 6(4), e03690. https://​doi​.org/​10​.1016/​j​.heliyon​.2020​.e03690 van Seventer, J. M., & Hochberg, N. S. (2017). Principles of infectious diseases: Transmission, diagnosis, prevention, and control. In International Encyclopedia of Public Health (pp. 22–39). Elsevier. https://​doi​.org/​10​.1016/​B978​-0​-12​-803678​-5​.00516​-6 West, D. A., Leung, G. M., Jiang, C. Q., Elwell-Sutton, T. M., Zhang, W. S., Lam, T. H., Cheng, K. K., & Schooling, C. M. (2012). Life-course origins of social inequalities in adult immune cell markers of inflammation in a developing southern Chinese population: The Guangzhou Biobank Cohort Study. BMC Public Health, 12(1), 269. https://​doi​.org/​10​.1186/​1471​-2458​-12​-269 Wick, G., & Grubeck-Loebenstein, B. (1997). The aging immune system: Primary and secondary alterations of immune reactivity in the elderly. Experimental Gerontology, 32(4–5), 401–413. https://​doi​ .org/​10​.1016/​s0531​-5565(96)00152​-0 World Health Organization. (2022). The top 10 causes of death. https://​www​.who​.int/​news​-room/​fact​ -sheets/​detail/​the​-top​-10​-causes​-of​-death Zhang, X., Zhivaki, D., & Lo-Man, R. (2017). Unique aspects of the perinatal immune system. Nature Reviews: Immunology, 17(8), 495–507. https://​doi​.org/​10​.1038/​nri​.2017​.54

PART V POLICY PERSPECTIVES AND EMPIRICAL EVALUATIONS OF INTERVENTIONS AGAINST HEALTH INEQUALITIES

23. Policy, inequity, and the life course in the US Sarah Petry

INTRODUCTION The United States’ policy space is fragmented. This fragmentation is, in part, a consequence of the federalist system in the US. Federalism, in the US case, maintains that the federal government has significant oversight in drafting and implementing health and social policy. This is then diminished to some extent by the rights designated to state governments. State governments – and often county and city governing bodies – make their own decisions about how to implement policies devised by the federal government. It is this distribution of power, in part, that produces a fragmented social safety net in the US. Fragmentation produced via federalism means that people in different states, counties, and cities may have radically different experiences with the social safety net. This is significant for several reasons. First, fragmentation in this way produces predictable health inequities at the state level. Second, because states have distinct racial, ethnic, and socioeconomic profiles, fragmented policies may play a role in creating, maintaining, exacerbating, or alleviating health inequities between marginalized groups. Finally, and perhaps most importantly, because policy power is distributed between federal, state, and local governments, attempts at any level to address predictable health inequities will not necessarily be effective or broadly accessible. This creates the potential for the most vulnerable, the sickest, the poorest, to remain the worst off. The second feature of the US policy space that produces health inequities across the life course is its elective nature. To access the social safety net, individuals must apply, demonstrate their own eligibility for a given program, then, if they are deemed eligible, they must continue to demonstrate their need over time. This creates barriers to access for people who could benefit from the various health and social programs that make up the US’s fragmented social safety net. In this chapter I outline how fragmentation and electiveness create barriers to access for people and populations across the life course. I describe the current US policy landscape, though not in its entirety as each state has a distinct collection of programs, detailing how policies contribute to or alleviate health inequities. I note significant policies targeting health at different periods of the life course, and review what is known about the short- and long-term health consequences of those policies. I will identify methodological techniques, innovations, and challenges in this area of research, as well as identify emergent themes and unresolved issues. I conclude with directions for future research in this critical area of study.

THE US WELFARE STATE The current US welfare state can be traced to policies and programs from the early 20th century. In particular, the New Deal programs of the 1930s laid the foundation for the various 368

Policy, inequity, and the life course in the US  369 protections offered to US citizens today. A welfare state is a “collection of programs designed to assure economic security to all citizens by guaranteeing the fundamental necessities of life: food, shelter, medical care, protection in childhood, support in old age” (Katz, 2001, p.9). However, because the US welfare state is fragmented and elective, it is not the case that all citizens are guaranteed these necessities. Developments of the Welfare State Prior to the 1930s, those social programs that existed in the US were localized (Katz, 2001). The federal government did not, at this time, provide standardized support to any population, let alone the country overall (Katz, 2001). In 1933 President Franklin Delano Roosevelt, FDR, took office and established the early welfare state in the US. The key piece of New Deal legislation that became the foundation of the US welfare state was the Social Security Act of 1935. The Social Security Act of 1935 established two programs: Social Security and Aid to Dependent Children (Katz, 2001). The first was a federal program focused on providing, primarily, economic support in old age (Katz, 2001). Social Security provides public pensions to retirees (Aldy, 2016), and was intended to provide all retirees with economic security in their late life. However, this pension program aims to complement two other forms of post-retirement income that many retirees do not have: private pensions and private savings (Poterba, 2014). It is intended to replace only a percentage of pre-retirement earnings. Aid to Dependent Children, which required federal–state partnership, focused instead on providing support in early life. This program provided states with funds to support “needy” children (Falk, 2021; Gordon and Batlan, 2011). States had discretion to determine need, establish benefit and eligibility levels, and administer this program (Falk, 2021; Gordon and Batlan, 2011). Unlike Social Security, this program was not designed to support all children. Instead, only children who lived in single-parent households with a relatively low income (as determined by a given state) were eligible (Gordon and Batlan, 2011; Katz, 2001). Both Social Security and Aid to Dependent Children provided cash payments to eligible beneficiaries to support citizens during childhood and old age. Each program has been amended since 1935, but both remain critical components of the modern US welfare state. Indeed, these foundational New Deal programs established a precedent for US welfare programs to target economically vulnerable groups primarily during these two periods of life. Federalism remains a key element at the core of modern US welfare programs, producing variable access, eligibility, and benefits across states (Falk, 2021; Gordon and Batlan, 2011). The Childhood Welfare State The earliest policy supporting childhood well-being was Aid to Dependent Children, which later became Aid to Families with Dependent Children (AFDC) and then Temporary Assistance to Needy Families (TANF). While the former programs were intended to provide economic support to (in general) single mothers and children, TANF was intended to move families off welfare (Hildebrandt and Stevens, 2009). This is, again, a unique feature of the US welfare state: over time, the US has increasingly structured programs to foster economic self-sufficiency and encourage families to cease reliance on the welfare state. A welfare state can go beyond ensuring a basic level of income (Katz, 2001). The Social Security Act of 1965 established the Medicaid program (Holohan, 1975). Medicaid was

370  Handbook of health inequalities across the life course targeted, not universal, but provided a basic level of health insurance to families covered by AFDC (Davis and Schoen, 1978). This link between Medicaid and AFDC meant that only children whose families were enrolled in AFDC had access to this insurance. Because AFDC was administered by the states, there were significant variations in who was able to access AFDC and, by extension, Medicaid in the period following implementation (Davis and Schoen, 1978). In the 1990s, AFDC was replaced by the Temporary Assistance for Needy Families (TANF) program. AFDC was politically unpopular given the construction of beneficiaries as Black welfare queens, who did not work, abused the social safety net, and were undeserving of aid (Triece, 2013). TANF, which evolved to counter the political unpopularity of AFDC, is a much more restrictive program, with greater work requirements, and a lifetime enrollment limit of 60 months (Falk, 2021). It remains a cash assistance program, administered by states to support low-income families and children. While TANF became more restrictive, Medicaid coverage has expanded for children. The Children’s Health Insurance Program (CHIP) was implemented in 1997 to encourage states to provide public health insurance to children whose family income is up to 200 percent of the federal poverty line (FPL). The Patient Protection and Affordable Care Act (ACA, otherwise known as “Obamacare”) similarly expanded the populations of children eligible for Medicaid coverage (Barr, 2011). Medicaid, though, is still administered by states, who have discretion to establish eligibility requirements and benefits. One other program targeted to support children is the Supplemental Nutrition Program for Women, Infants, and Children (WIC). WIC provides nutrition counseling, vouchers for nutritious foods, and other supportive services to pregnant women and children up to age 5 (Gray et al., 2019). To be eligible, a family must have an income less that 185 percent of the FPL (Gray et al., 2019; Kline et al., 2020). WIC is administered by the federal government, so these eligibility guidelines are uniform across the US. WIC, then, is not fragmented in the same way as Medicaid and TANF. However, the benefits of WIC enrollment may vary across states. WIC participants must attend in-person appointments quarterly. At these appointments, staff assess nutritional risk and dietary preferences, and collect children’s anthropometric data, such as height and weight (Barnes and Petry, 2021). During these appointments, staff may also refer participants to other social service programs, such as Medicaid or TANF (Gray et al., 2019). Given that these programs are not uniform across the country, though, WIC participants may have distinct experiences with the welfare state based on their geographic location. As the US welfare state expanded during the latter half of the 1900s, programs aimed at poverty alleviation and improving childhood well-being have proved durable (Howell and Kenney, 2012; Starr, 1982). These programs, though, are deeply fragmented because they are administered within a federalist system: there are at least 50 different childhood welfare states. Children across the US have inequitable access to remarkably different welfare systems, and where they do have access, the benefits received can vary widely. The Old Age Welfare State The exception to the federalist rule has been in the US’s approach to providing support for aged Americans. Since 1935, Social Security has been amended several times (Aldy, 2016). One such amendment, Medicare, was added in 1965, providing health insurance to all US

Policy, inequity, and the life course in the US  371 citizens over 65 (Sullivan-Marx, 2015). The US welfare state targeting older adults is stable: these two programs together provide some economic support to all aged Americans. Social Security and Medicare are targeted to and universal for all citizens over 65. Both are funded through payroll taxes, which aligns with US values of self-help and self-sufficiency: if a person worked long enough and paid into Social Security and Medicare then they are seen as deserving of benefits once they retire (Cox, 2015; Yoshida, 2018). It is in part due to this alignment with an intrinsic US value, self-sufficiency, that both programs have remained politically popular and relatively stable since their respective implementations (Yoshida, 2018). In 1965, Medicare was designed to do two things: provide hospital insurance (Part A) and cover out-patient services (Part B) (Barr, 2011). All adults over 65 are covered by Medicare Part A. Individuals do not pay for this coverage, but instead are responsible for a deductible when they do access hospital care (Barr, 2011). Part B, though, is voluntary. Adults over 65 may elect to enroll in Part B, which requires that a premium be deducted from their monthly Social Security payment (Barr, 2011). Eligible adults are automatically enrolled in both and must then opt-out of Part B if they do not want to pay the premium. Medicare Parts A and B are known as “traditional Medicare.” In 2006, traditional Medicare was expanded to include Part D, which is elective like Part B, and provides prescription drug coverage (Engelhardt and Gruber, 2011). Both Social Security and Medicare are federally administered. Therefore, federalism has not produced fragmentation in either. However, Medicare has become fragmented for other reasons. The first is the voluntary nature of Parts B and D. Older adults who do not elect to enroll in these programs face potentially greater economic insecurity than those who do. The second reason is that older adults now may choose between original Medicare, where they are automatically enrolled in Part A and B and then choose a Part D plan, or they may elect to choose a Medicare Advantage Plan. Medicare Advantage Plans are bundled plans offered by private companies. As choices for health care increase for older adults, the level of support offered by any one of these programs can vary greatly from others. The Welfare State across the Life Course Although the US welfare state provides the most support to citizens in childhood and old age, several welfare programs do support people across the life course. I offer a brief overview of how the US provides for basic needs of certain populations at certain times. As noted, the US welfare state is unique in that it does not provide for all citizens, instead supporting only some. Social Security, Medicare, Medicaid, and AFDC or TANF can provide support to US citizens during young adulthood and midlife. For example, disabled adults can receive Social Security, Medicare, and Medicaid if they are eligible (Barr, 2011). Also, it is parents, not children, who receive the cash payments linked to TANF. These large public programs, then, can facilitate access to shelter, medical care, and food for individuals at any age. The US social safety net for working age adults is more fragmented than that for either children or elderly populations. As with welfare for the young and old, many programs are administered by states. The simple accident of geography, then, impacts what resources are available and whether similarly situated individuals are eligible for state support (Campbell, 2014; Michener, 2018). It is also the case that eligibility for a given program does not indicate enrollment. All welfare programs for working-age adults are elective and means-tested. Individuals must

372  Handbook of health inequalities across the life course apply, demonstrate their own need, recertify at various increments, and, often, fulfill other program requirements to retain their benefits. Unlike Medicare and Social Security, which are universal programs and have essentially 100% coverage of eligible populations, means-tested programs have many requirements that might deter participation (Burden et al., 2012; Moynihan, Herd and Harvey, 2015). Another concerning feature of the US welfare state programs is the political nature of their implementation, maintenance, and longevity. For example, when it became clear that AFDC welfare recipients were primary single Black mothers, it became a racialized and gendered policy, providing support to an “underserving” population. This construction made it a politically unpopular program, resulting in decreased benefits and the eventual dissolution of the program (Triece, 2013). Social Security, on the other hand, supports older adults who – in theory, at least – paid into this program and deserve those benefits. Social Security is a politically popular program, supporting a “deserving” group. The historical, racial, gendered, economic, and political composition of the US creates challenges for the welfare state. The welfare state automatically supports deserving populations: the elderly, children, and, generally, white people. The welfare state excludes or requires voluntary enrollment for undeserving populations: individuals who could work, and, generally, racially marginalized populations. Working-age adults face a complicated policy landscape and significant barriers to accessing welfare programs. The very design of these policies produces inequities in access and in potential benefits. These inequities can worsen or maintain existing health disparities across the life course.

WELFARE PROGRAMS AND HEALTH INEQUITIES The United States has several public policies aimed at improving well-being for various populations. Given the public cost of such programs, it is important to determine whether those programs improve population health. Policymakers, researchers, and individuals across the US have a vested interest in the extent to which public programs improve well-being and health and alleviate disparities across the life course. Childhood Interventions and Outcomes Prior research has defined “critical periods” of development that can be formative for long-term health trajectories (Barker, 1990; Hoynes, Schanzenbach and Almond, 2016). For example, the fetal origins hypothesis suggests that a lack of investment in fetal health (especially nutrition) can prevent proper development and establish trajectories of poor health (Barker, 1990; Boudreaux, Golberstein and McAlpine, 2016). In addition, the early childhood period, often defined as age 0–5, is commonly cited in the literature as a meaningful period for child development when access to resources or deprivation can have lasting consequences for health (Boudreaux, Golberstein and McAlpine, 2016; Currie and Rossin-Slater, 2015; Hoynes, Schanzenbach and Almond, 2016; Schafer, Ferraro, and Mustillo, 2011). Therefore, the childhood welfare state may provide short- and long-term health benefits.

Policy, inequity, and the life course in the US  373 Cash Assistance: Economic Security is not Health Aid to Dependent Children was the first attempt to provide a social safety net to perhaps the most vulnerable population in the US: children living in poverty. Over time, this program was renamed Aid to Families with Dependent Children (AFDC). AFDC provided cash assistance to families experiencing economic hardship. In many states, AFDC was limited to single-mother households (Davis and Schoen, 1978). Because of the different intents behind each program, the effects are often measured as the extent to which a program’s consequences aligned with that intent. So, because TANF is intended to move families off welfare, most studies examine the economic consequences for participants. There is robust evidence that socioeconomic status is associated with health across the life course, yet it is beyond the scope of this chapter to assess that relationship. TANF recipients report more physical and mental health conditions than non-TANF recipients (Hildebrandt and Kebler, 2005; Hildebrandt and Stevens, 2009; Wise at al., 2002). This is reasonable given that participation is tied to poverty. However, children whose mothers successfully enter the workforce following TANF participation are as unhealthy as those whose mothers do not (Hildebrandt and Stevens, 2009). This is concerning because, although the intent is to move families off welfare, there appear to be no health benefits to this. Conversely, participation in AFDC, which had no time limits for participation, is associated with greater life-years gained (in relation to life-expectancy) than TANF, which has a lifetime limit for participation of five years (Muennig et al., 2015). Although the intent differs between these two programs, one has the potential to reduce lifetime health inequities while the other does not. Health Insurance: Early and Often Public health insurance programs are not intended to have direct economic consequences. Instead, these programs are designed to improve health by facilitating access to health care. Therefore, researchers have examined the short- and long-term health effects of accessing Medicaid or the Children’s Health Insurance Program (CHIP). Ideally, providing access to public health insurance will improve population health and reduce otherwise predictable health disparities. There is robust evidence that programs aimed at increasing access to health care in early life have profound, positive impacts on childhood health and well-being (Currie and Rossin-Slater, 2015; De La Mata, 2012; Goodman-Bacon, 2018; Howell and Kenney, 2012). For example, early Medicaid expansions in the 1980s reduced infant and child mortality (Currie and Gruber, 1996; Goodman-Bacon, 2018) and were associated with greater childhood health care utilization (Currie and Gruber, 1996; De La Mata, 2012). Among marginalized groups, these policies targeting childhood health and nutrition can reduce the immediate, short-term experiences and consequences of disadvantage. Emerging evidence suggests that these programs also have long-term effects, impacting health well into adulthood (Boudreaux, Golberstein and McAlpine, 2016; Brown, Kowalski and Lurie, 2019; Campbell et al., 2014; Duncan, Ziol‐Guest and Kalil, 2010; Hoynes, Schanzenbach and Almond, 2016; Wherry et al., 2018). Exposure to Medicaid in childhood is associated with lower mortality (Brown, Kowalski and Lurie, 2019) and fewer hospitalizations in young adulthood (Wherry et al., 2018), and with fewer risk factors for cardiovascular and metabolic conditions in early- to mid-adulthood (Boudreaux, Golberstein and McAlpine,

374  Handbook of health inequalities across the life course 2016; Campbell et al., 2014). Together, this suggests that policies aimed at improving access to health care in early life can have profound impacts not only on short-term health outcomes, but also, critically, on later life health outcomes. Nutrition: Promising Pathways The Supplemental Nutrition Program for Women, Infants, and Children (WIC) has a clearer relationship to health than cash assistance, yet perhaps not as clear as health insurance. This component of the childhood welfare state is designed to promote good nutrition and eating behaviors (Barnes and Petry, 2021; Bitler and Currie, 2005). Indeed, WIC does improve nutrition and eating behaviors for children who are enrolled, which can have lasting benefits for health well into adulthood (Siega-Riz, et al. 2004). WIC targets a critical period of development, pre-natal until age five. Investments in this period, particularly by promoting healthy eating and good health behaviors, can have lasting effects on health and well-being. Several studies have addressed the impacts of WIC participation on childhood health. For example, WIC participation during pregnancy reduces the probability of low birth weight (Bitler and Currie, 2005). Low birth weight is associated with poor health outcomes during childhood (Choi & Martinson, 2018), so individuals whose mothers participated in WIC may avoid those negative health outcomes. In addition, WIC participation during this critical period of childhood fosters positive brain and cognitive development (Jackson, 2015). WIC demonstrably improves developmental outcomes in childhood. Finally, WIC may improve health through another pathway: facilitating access to health care. As noted, the WIC program design is intended to provide families with referrals to other welfare programs and services that may benefit their health (Bersak and Sonchak, 2018; Buescher et al., 2003). For example, because WIC, Medicaid, and CHIP are all means-tested, WIC recipients are often also eligible for Medicaid or CHIP. They must elect to apply and enroll, but, if they are enrolled in WIC, the barriers for enrollment in public health insurance may be lower. WIC participation is, generally, associated with greater preventative care use and better childhood health (Buescher et al., 2003). WIC participants are more likely to attend the recommended number of well-child pediatric visits during a child’s first year of life (Bersak and Sonchak, 2018). In addition, WIC participation is associated with a greater likelihood of vaccination (Bersak and Sonchak, 2018). Finally, children who are enrolled in both WIC and Medicaid are more likely than children enrolled only in Medicaid to receive preventative and curative medical treatments (Buescher et al., 2003). The benefits of WIC enrollment are wide-ranging and significant. WIC, like Medicaid and TANF, is voluntary, so women must apply, enroll, and engage in recertification processes to maintain benefits. There are barriers to continued enrollment, but for families that maintain WIC access while children are under five there are many positive health benefits. Those health benefits may last into adulthood, potentially establishing better health trajectories than would otherwise be expected based on recipients’ economic status and other disadvantages. Children who are supported by more than one welfare program may experience the greatest benefits.

Policy, inequity, and the life course in the US  375 Old Age Interventions and Outcomes The US welfare state addresses well-being in early life and in old age. Although investments in health and well-being in childhood are significant for establishing long-term health (Barker, 1990; Hoynes, Schanzenbach and Almond, 2016), many US welfare programs instead address poor health much later in life. The US spends much more per capita on welfare for the elderly than they do on children (Hoynes and Schanzenbach, 2018). These programs are therefore aimed at alleviating rather than preventing disparities. Social Security: Economic Insecurity Social Security was designed to provide older adults with a minimum level of economic security once they leave the workforce (Yoshida, 2018). Social Security, though, is not the only source of income that retirees may draw on. In the US, retirement income is a “three-legged stool,” including Social Security, private pensions, and private savings (Poterba, 2014). Social Security alone does not provide economic security, as a welfare state should, for most Americans. This means that many older adults live in poverty. In 2017, at least 9.2% of older adults had incomes below the Federal Poverty Line (FPL) and more than 30% of older adults had incomes below 200% of the FPL (Cubanski et al., 2018). Poverty is widespread among older adults who cannot draw on private savings or employer-based pensions (Poterba, 2014). For one third of older Blacks and Hispanics, Social Security represents 90% of retirement income, while another quarter rely on Social Security as their only source of income (Cox, 2015). Economic inequity, tied to racial inequity and health inequities, is already deeply entrenched by the time individuals in the US retire. Medicare: Too Little, Too Late Medicare was intended to protect older adults from economic crises due to health care costs (Barr, 2011). And much like Medicaid and CHIP, for children, Medicare is intended to facilitate access to health care. Medicare, as the policy stands, ensures that all adults over 65 have access to in- and out-patient care at no or low cost. Much like Social Security, though, Medicare is available once health trajectories have already been established and health disparities have been cemented. Although health disparities are well-established by age 65, gaining access to Medicare at this point in the life course does have some positive health benefits overall. Previously uninsured individuals who enroll in Medicare at age 65 report improved self-rated health and better cardiovascular function (McWilliams et al., 2007). Enrolling in Medicare Part D, which covers prescription drugs and is elective, is associated with improved self-rated health and lower incidence of high-blood pressure (Diebold, 2018). These studies, however, do not address disparities. Several studies demonstrate that racial and socioeconomic health disparities persist after Medicare enrollment. Among Medicare beneficiaries aged 65 and 66, the mortality rate for Blacks is twice that of Whites (Yang, Huang and Phillips, 2014). Similarly, in this age group, Black beneficiaries are 40% more likely to be hospitalized than White beneficiaries (Yang, Huang and Phillips, 2014). Nearly half of all Non-Hispanic Black Medicare beneficiaries between 74 and 85 have four or more comorbid conditions (Lochner and Cox, 2013).

376  Handbook of health inequalities across the life course Similarly, non-Hispanic Black women and Hispanic women have higher rates of comorbid conditions than White women (Lochner and Cox, 2013). The cohorts currently enrolled in Medicare have different lived experiences and they have been growing more racially and ethnically diverse over time (Ortman, Velkoff and Hogan, 2014). These changes make racial health disparities in late life more apparent. Thus, Medicare, does not appear to address health disparities. Instead, this late-life policy intervention simply highlights racial, gendered, socioeconomic, and other health disparities. Rather than address early life inequalities that establish divergent trajectories in access to resources, including health care, and health, generally, US health policy focuses on late-life vulnerability. That focus, though, fails to alleviate health disparities that are already entrenched due to inequities experienced across the life course. Life-Course Interventions and Outcomes There are some programs that focus on periods outside childhood and old age. Some of those already discussed fit into this category. WIC, for example, supports women if they are pregnant, post-partum, or breast-feeding (Gray et al., 2019), while TANF similarly supports parents during the period of eligibility (Gordon and Batlan, 2011). The Supplemental Nutrition Assistance Program (SNAP) is more widely available, both to parents and childless adults (Gregory and Deb, 2015). Finally, Medicaid has expanded to include broader populations, though this is haphazard across the US states (Michener, 2018). WIC enrollment improves health care utilization among pregnant women as well as birth outcomes for beneficiaries (Bitler and Currie, 2005). TANF, which is administered by states and is designed to promote economic security, has variable impacts on health. Mothers report worse mental health in states with more restrictive enrollment criteria than in states with fewer TANF conditions (Davis, 2019). SNAP, which was intended to reduce hunger, not necessarily to improve health, does seem to have a positive association with self-rated health (Bleich, Rimm, and Brownell, 2017; Gregory and Deb, 2015). It is unclear whether the effects of these programs persist over the long term. Medicaid, which, like Medicare, facilitates access to health care, may more directly reduce health disparities. Among working-age adults, Medicaid expansions in three large states were associated with significant declines in absolute mortality, especially for non-white populations and adults aged 35–64 (Sommers, Baicker, and Epstein, 2012). In addition, low-income mothers have better access to care with Medicaid than when uninsured (Long, Coughlin, and King, 2005). This is probably because women with Medicaid coverage are more likely to have a usual source of care and to access preventative care services than women who are uninsured (Long, Coughlin, and King, 2005). Although Medicaid can facilitate access to health care, the underlying federalist nature of this policy creates distinct disparities in access and outcomes. Medicaid eligibility criteria are stricter and take-up rates are lower in states where the proportion of the population that is Black is high (Andrews, 2014). Indeed, Andrews (2014) notes: “Studies across the fields of political science, public health, and social welfare have consistently found a strong, inverse relationship between the size of a state’s African American population and the generosity of cash assistance and health and social welfare benefits” (p. 132). Taken together, this suggests that Medicaid’s potential to reduce health disparities is limited and that the current collection

Policy, inequity, and the life course in the US  377 of Medicaid policies across the country likely widen health disparities at the national level and at every age. The US welfare state for individuals between ages 6 and 64 is fragmented and fraught. Access to any of the programs discussed here is voluntary, requiring applications, re-certifications, and other demonstrations of eligibility over time. Take-up rates for these programs vary greatly, such that the benefits are concentrated among those populations who ultimately enroll in these programs. States, too, play a significant role in administering these policies, such that geography drives inequitable access to programs and benefits. The evidence is mixed: the US welfare state may reduce health inequities for some groups, at some points in the life course, but the US welfare state often instead exacerbates existing health disparities.

DISCUSSION Scholars and policymakers are contributing to the mounting body of work demonstrating how the diverse policies that compose the US welfare state impact health across the life course. Although some earlier literature documents the role of the state in structuring the life course (Leisering, 2003; Mayer and Schoepflin, 1989), this critical area of research, integrating health and the life course, is relatively new. One reason for this is that the modern US welfare state only emerged during the latter half of the 20th century. In addition, because the social safety net is fragmented at various levels, researchers must be selective in the programs they study and the geographical area of interest. Finally, the US, a country with a particular racist past, is increasingly racially and ethnically diverse. Studying life-course processes and health in this context requires careful consideration of how to operationalize race, gender, and other dimensions of marginalization. I document some limitations of the current literature and offer suggestions for future research. Limitations The US welfare state is comprised of many programs, some of which are specifically intended to improve individual and population health, while others are not. This social safety net is strongest for children and older adults but is nonetheless fragmented for all US citizens. In addition, because most programs are means-tested, many Americans will not interact with the welfare state until they enroll in Medicare in older adulthood. Given these features of the US welfare state, it is reasonable that scholars tend to focus their research on only one policy or program. Research on how access to policies impacts health is broadly concentrated on Medicaid, WIC, cash assistance, and Medicare. For example, scholars assess whether access to Medicaid during the critical period between birth and age five can improve health (e.g., Boudreaux, Golberstein and McAlpine, 2016). However, because Medicaid is means-tested, individuals and families who are eligible for Medicaid are likely eligible for WIC, SNAP, and other social welfare programs. Some researchers do account for this, such as Buescher et al. (2003), who examine whether children who receive WIC benefits and Medicaid experience better health outcomes than those only enrolled in one. Given the inherent ties between these programs, it is critical to determine their effects separately and in combination.

378  Handbook of health inequalities across the life course Because the US has at least 50 different welfare states, researchers must incorporate states into their analyses. Although WIC and SNAP are federal programs, Medicaid is not. Therefore, analyses of the effects of Medicaid coverage on health must operationalize state and policy context in some way. For example, Davis (2019) constructs a typology of states based on the stringency of their social safety nets. Other studies select only one state to study, such as Campbell et al. (2014). Overall, research in this area is complicated by the nature of the US federalist system. The strongest analyses must demonstrate how and why they incorporate states into their study, detailing how their choice impacts what we can infer from their findings. Perhaps the most significant limitation of this body of research, as it stands, is that measures of health are included less frequently than economic outcomes. Cash assistance programs, such as AFDC and TANF, are not intended to improve health. The way these policies are written, the key outcome that policymakers intend is economic self-sufficiency. The legislation itself lends itself to determining the economic consequences for participants, rather than any health outcomes. Examinations of economic outcomes associated with welfare involvement directly align with the unique goal of the US welfare state: to help people over the short term and, ideally, push people out of the social safety net. When health outcomes are included in policy analyses, they are not standardized. For example, over the short term, scholars study birth outcomes (e.g., Bitler and Currie, 2005), self-rated health (e.g., McWilliams et al., 2007), high blood pressure (e.g., McWilliams et al., 2007), and infant and child mortality (e.g., Currie and Gruber, 1996; Goodman-Bacon, 2018). Over the medium and long term, scholars study cardiovascular health (e.g., Campbell et al., 2014), high blood pressure (e.g., Boudreaux, Golberstein and McAlpine, 2016), hospitalizations (e.g., Wherry et al., 2018), and mortality (e.g., Brown, Kowalski and Lurie, 2019). This makes synthesis of these studies complex and difficult to convey to policymakers. Finally, and most relevant to this book and this chapter, in the US context very few studies directly assess the impact of policies on health disparities. There is a significant body of evidence documenting racial (e.g., Haas & Rohlfsen, 2010; Phelan & Link, 2015), gender (e.g., Bird and Rieker, 2008), socioeconomic (e.g., Dupre, 2007; Hayward, Hummer, & Sasson, 2015), and intersectional health disparities (e.g., Brown et al., 2016; Richardson & Brown, 2016). There is also a growing body of evidence documenting the relationship between access to US welfare programs and health, generally (e.g., Bersak and Sonchak, 2018; Currie and Rossin-Slater, 2015; De La Mata, 2012). These two literatures, though, are relatively siloed. In order to understand to what extent US policies exacerbate, maintain, or reduce health disparities, researchers must explicitly link these two areas of research. It is this last limitation that fosters the most exciting potential for researchers. Does access to Medicaid in childhood reduce health disparities across the life course? Does participation in WIC reduce health disparities across the life course? These, and many other, empirical questions remain. Future Directions Despite the challenges to studying policies and health in the US, this is an area ripe for further research. There are opportunities to link and generate new data sources, employ novel methodological techniques, and be creative about how to answer critical questions.

Policy, inequity, and the life course in the US  379 First, because many of the policies comprising the modern US welfare state have now existed for nearly 60 years it is possible to study the long-term consequences of accessing these programs. Medicare and Medicaid, for example, have proved durable and popular. Researchers must assess whether access to these programs at different stages of life can reduce health disparities. Scholars should pay particular attention to the period and cohort effects, as well as the timing of access to these policies. Second, to complement and deepen the existing literature in this area scholars must examine the mechanisms through which policies improve health. For example, access to Medicaid in childhood is associated with reduced mortality in young adulthood (Brown, Kowalski and Lurie, 2019). Is this because Medicaid fosters increased preventative care use, as De La Mata (2012) suggests? Or is there another pathway, such as establishing positive health behaviors in early life? In addition, scholars must attend to the racialized Medicaid landscape. Does access to these policies improve health, therby reducing inequities, for Black and White Americans equally? Do they do so through the same mechanism? Finally, this area of research can benefit from interdisciplinary collaboration, methodological rigor, and qualitative analyses. As noted, research on policy implementation and outcomes is relatively siloed from that on health disparities. Connecting these bodies can offer new insights about the relationship between the US welfare state and racial, gendered, and socioeconomic health disparities. Scholars can also deploy novel quantitative techniques, such as triple-difference models (i.e., Boudreaux, Golberstein and McAlpine, 2016), to assess the mutliple dimensions of complexity inherent in the US welfare state. Where quantitative work falls short, though, is in determining mechanisms where there are not good priors for hypotheses. Qualitative studies offer the potential to explore relationships and describe how individuals experience the welfare state over their lives. There are many empirical questions remaining, and many scholars eager to tackle them. This chapter has only touched on the many exciting avenues to expand this area of research. In addition, policymakers at every level are wrestling with questions about how to provide for the welfare of citizens across the life course. This is an exciting time for researchers to participate in and contribute to the further development of US welfare and population health.

CONCLUSION The US welfare state, as it stands, is fragmented and fails to support population health and reduce health inequities. Most welfare programs and policies in the US focus on early childhood or old age, and most are elective. In addition, states, not the federal government, administer most of the social and health policies in the US. Policies vary across states in terms of accessibility, eligibility stringency, benefit generosity, and more. From birth until death, the US welfare state provides for citizens selectively and insufficiently. The childhood welfare state is perhaps the most robust in the US. The collection of programs targetted to children under five offer the opportunity to establish good individual and population health trajectories from an early age. The Supplemental Nutrition Program for Women, Infants, and Children (WIC) is the most promising policy in this childhood welfare state. Access to WIC during early childhood does improve health outcomes over the short and medium term. Because WIC is targetted at economically disadvantaged children and families, this has the potential to prevent health disparities across the life course.

380  Handbook of health inequalities across the life course The old age welfare state provides a minimal economic and health safety net, but does not fully support citizens late in life. At this stage of the life course, health inequities are entrenched and embodied. Research on this population demostrates that race, gender, and socio-economic status across the life course contribute to predictable health inequities in late life. Policy interventions late in life, then, offer too little support, too late. Further research is needed to determine whether access to the fragmented social safety net during young- and mid-adulthood can prevent these inequities. As more cohorts in the US reach older adulthood, facing greater medical and social needs, it is essential to consider how their health today has been shaped by processes that began in their early childhood. Even more pressing, though, is to determine whether, when, and how access to health and social programs cement, alter, or radically disrupt expected inequitable trajectories of disadvatange as people age.

REFERENCES Aldy, J. E. (2016). Mobilizing political action on behalf of future generations. The Future of Children, 26(1), 157. Andrews, C. (2014). Unintended consequences: Medicaid expansion and racial inequality in access to health insurance. Health & Social Work, 39(3), 131–133. Barker, D. J. (1990). The fetal and infant origins of adult disease. British Medical Journal, 301(6761), 1111–1111. doi:10.1136/bmj.301.6761.1111 Barnes, C., & Petry, S. (2021). “It was actually pretty easy”: COVID-19 compliance cost reductions in the WIC program. Public Administration Review, 81(6), 1147–1156. doi:https://​doi​.org/​10​.1111/​puar​ .13423 Barr, D. H. (2011). Introduction to U.S. health policy. Baltimore, MD: The Johns Hopkins University Press. Bersak, T., & Sonchak, L. (2018). The impact of WIC on infant immunizations and health care utilization. Health Services Research, 53 (Suppl Suppl 1), 2952–2969. doi:10.1111/1475-6773.12810 Bird, C. E., & Rieker, P. P. (2008). Gender and health: the effects of constrained choices and social policies. Cambridge, United Kingdom: Cambridge University Press. Bitler, M. P., & Currie, J. (2005). Does WIC work? The effects of WIC on pregnancy and birth outcomes. Journal of Policy Analysis and Management, 24(1), 73–91. doi:10.1002/pam.20070 Bleich, S. N., Rimm, E. B., & Brownell, K. D. (2017). U.S. nutrition assistance, 2018 – modifying SNAP to promote population health. New England Journal of Medicine, 376(13), 1205–1207. doi:10.1056/ NEJMp1613222 Boudreaux, M. H., Golberstein, E., & McAlpine, D. D. (2016). The long-term impacts of Medicaid exposure in early childhood: evidence from the program’s origin. Journal of Health Economics, 45, 161–175. doi:https://​doi​.org/​10​.1016/​j​.jhealeco​.2015​.11​.001 Brown, D. W., Kowalski, A. E., & Lurie, I. Z. (2019). Long-term impacts of childhood Medicaid expansions on outcomes in adulthood. Review of Economic Studies, 87(2), 792–821. doi:10.1093/restud/ rdz039 Brown, T. H., Richardson, L. J., Hargrove, T. W., & Thomas, C. S. (2016). Using multiple-hierarchy stratification and life course approaches to understand health inequalities: the intersecting consequences of race, gender, ses, and age. Journal of Health and Social Behavior, 57(2), 200–222. doi:10.1177/0022146516645165 Buescher, P. A., Horton, S. J., Devaney, B. L., Roholt, S. J., Lenihan, A. J., Whitmire, J. T., & Kotch, J. B. (2003). Child participation in WIC: Medicaid costs and use of health care services. American Journal of Public Health, 93(1), 145–150. doi:10.2105/AJPH.93.1.145 Burden, B. C., Canon, D. T., Mayer, K. R., & Moynihan, D. P. (2012). The effect of administrative burden on bureaucratic perception of policies: evidence from election administration. Public Administration Review, 72(5), 741–751. doi:10.1111/j.1540-6210.2012.02600.x

Policy, inequity, and the life course in the US  381 Campbell, A. L. (2014). Trapped in America’s safety net: one family’s struggle. Chicago, IL: University of Chicago Press. Campbell, F., Conti, G., Heckman, J. J., Moon, S. H., Pinto, R., Pungello, E., & Pan, Y. (2014). Early childhood investments substantially boost adult health. Science, 343(6178), 1478–1485. Choi, K. H., & Martinson, M. L. (2018). The relationship between low birthweight and childhood health: disparities by race, ethnicity, and national origin. Annals of Epidemiology, 28(10), 704–709. doi:https://​doi​.org/​10​.1016/​j​.annepidem​.2018​.08​.001 Cox, C. B. (2015). Social policy for an aging society: a human rights perspective. New York, NY: Springer Publishing Company. Cubanski, J., Koma, W., Damico, A., & Neuman, T. (2018). How many seniors live in poverty? Retrieved from kff.org: https://​www​.kff​.org/​medicare/​issue​-brief/​how​-many​-seniors​-live​-in​-poverty/​ Currie, J., & Gruber, J. (1996). Health insurance eligibility, utilization of medical care, and child health. Quarterly Journal of Economics, 111(2), 431–466. Currie, J., & Rossin-Slater, M. (2015). Early-life origins of life-cycle well-being: research and policy implications. Journal of Policy Analysis and Management, 34(1), 208–242. doi:10.1002/pam.21805 Davis, K., and Kathy Schoen. (1978). Health and the war on poverty: a ten-year appraisal. Washington, DC: Brookings Institution. Davis, O. (2019). What is the relationship between benefit conditionality and mental health? Evidence from the United States on TANF policies. Journal of Social Policy, 48(2), 249–269. doi:10.1017/ S0047279418000363 De La Mata, D. (2012). The effect of Medicaid eligibility on coverage, utilization, and children’s health. Health Economics, 21(9), 1061–1079. doi:10.1002/hec.2857 Diebold, J. (2018). The effects of Medicare Part D on health outcomes of newly covered Medicare beneficiaries. The Journals of Gerontology: Series B, 73(5), 890–900. doi:10.1093/geronb/gbw030 Duncan, G., Ziol-Guest, K. & Kalil, A. (2010). Early-childhood poverty and adult attainment, behavior, and health. Child Development, 81, 306–325. Dupre, M. E. (2007). Educational differences in age-related patterns of disease: reconsidering the cumulative disadvantage and age-as-leveler hypotheses. Journal of Health and Social Behavior, 48(1), 1–15. doi:10.1177/002214650704800101 Engelhardt, G. V., & Gruber, J. (2011). Medicare Part D and the financial protection of the elderly. American Economic Journal: Economic Policy, 3(4), 77–102. doi:10.1257/pol.3.4.77 Falk, G. (2021). The temporary assistance for needy families (TANF) block grant: a legislative history. Retrieved from Congressional Research Service: https://​sgp​.fas​.org/​crs/​misc/​R44668​.pdf Goodman-Bacon, A. (2018). Public insurance and mortality: evidence from Medicaid implementation. Journal of Political Economy, 126(1), 216–262. Gordon, L., & Batlan, F. (2011). The legal history of the aid to dependent children program. Retrieved from the Social Welfare History Project: http://​socialwelfare​.library​.vcu​.edu/​public​-welfare/​aid​-to​ -dependent​-children​-the​-legal​-history/​ Gray, K., Trippe, C., Tadler, C., Perry, C., Johnson, P., & Betson, D. (2019). National- and state-level estimates of WIC eligibility and WIC program reach in 2017. Retrieved from https://​ fns​ -prod​ .azureedge​.net/​sites/​default/​files/​resource​-files/​WICEligibles2017​-Volume1​.pdf Gregory, C. A., & Deb, P. (2015). Does SNAP improve your health? Food Policy, 50, 11–19. doi:https://​ doi​.org/​10​.1016/​j​.foodpol​.2014​.09​.010 Haas, S., & Rohlfsen, L. (2010). Life course determinants of racial and ethnic disparities in functional health trajectories. Social Science & Medicine, 70(2), 240–250. doi:https://​doi​.org/​10​.1016/​j​ .socscimed​.2009​.10​.003 Hayward, M. D., Hummer, R. A., & Sasson, I. (2015). Trends and group differences in the association between educational attainment and U.S. adult mortality: implications for understanding education’s causal influence. Social Science & Medicine, 127, 8–18. doi:https://​doi​.org/​10​.1016/​j​.socscimed​.2014​ .11​.024 Hildebrandt, E., & Kelber, S. T. (2005). Perceptions of health and well-being among women in a work-based welfare program. Public Health Nurs, 22(6), 506–514. doi:10.1111/j.0737-1209.2005.220606.x Hildebrandt, E., & Stevens, P. (2009). Impoverished women with children and no welfare benefits: the urgency of researching failures of the Temporary Assistance for Needy Families program. American Journal of Public Health, 99(5), 793–801. doi:10.2105/AJPH.2006.106211

382  Handbook of health inequalities across the life course Holohan, J. (1975). Financing health care for the poor: the Medicaid experience. Lexington, MA: Lexington Books. Howell, E. M., & Kenney, G. M. (2012). The impact of the Medicaid/CHIP expansions on children: a synthesis of the evidence. Med Care Res Rev, 69(4), 372–396. doi:10.1177/1077558712437245 Hoynes, H. W., & Schanzenbach, D. W. (2018). Safety net investments in children. Brookings Papers on Economic Activity, 89+. Retrieved from https://​link​-gale​-com​.proxy​.lib​.duke​.edu/​apps/​doc/​ A560688966/​EAIM​?u​=​duke​_perkins​&​sid​=​summon​&​xid​=​cb6c9f3c Hoynes, H. W., Schanzenbach, D. W., & Almond, D. (2016). Long-run impacts of childhood access to the safety net. American Economic Review, 106(4), 903–934. doi:http://​dx​.doi​.org/​10​.1257/​aer​ .20130375 Jackson, M. I. (2015). Early childhood WIC participation, cognitive development and academic achievement. Social Science & Medicine, 126, 145–153. doi:10.1016/j.socscimed.2014.12.018 Katz, M. B. (2001). The price of citizenship: redefining the American welfare state. New York, NY: Metropolitan Books. Kline, N., Thorn, B., Bellows, D., Wroblewska, K., & Wilcox-Cook, E. (2020). WIC participant and program characteristics 2018. Retrieved from Alexandria, VA: https://​fns​-prod​.azureedge​.net/​sites/​ default/​files/​resource​-files/​WICPC2018​.pdf Leisering, L. (2003). “Government and the life course.” In J. T. Mortimer and M. J. Shanahan (eds) Handbook of the Life Course. New York: Springer, pp. 205–228. Lochner, K. A., & Cox, C. S. (2013). Prevalence of multiple chronic conditions among Medicare beneficiaries, United States, 2010. Preventing Chronic Disease, 10, E61. doi:10.5888/pcd10.120137 Long, S. K., Coughlin, T., & King, J. (2005). How well does Medicaid work in improving access to care? Health Serv Res, 40(1), 39–58. doi:10.1111/j.1475-6773.2005.00341.x Mayer, K. U., & Schoepflin, U. (1989). The state and the life course. Annual Review of Sociology, 15, 187–209. http://​www​.jstor​.org/​stable/​2083224 McWilliams, J. M., Meara, E., Zaslavsky, A. M., & Ayanian, J. Z. (2007). Health of previously uninsured adults after acquiring Medicare coverage. JAMA, 298(24), 2886–2894. doi:10.1001/jama.298.24.2886 Michener, J. (2018). Fragmented democracy: Medicaid, federalism, and unequal politics. Cambridge, United Kingdom: Cambridge University Press. Moynihan, D., Herd, P., & Harvey, H. (2015). Administrative burden: learning, psychological, and compliance costs in citizen-state interactions. Journal of Public Administration Research and Theory, 25(1), 43–69. Muennig, P., Caleyachetty, R., Rosen, Z., & Korotzer, A. (2015). More money, fewer lives: the cost effectiveness of welfare reform in the United States. American Journal of Public Health, 105(2), 324–328. doi:10.2105/AJPH.2014.302235 Ortman, J. M., Velkoff, V. A., & Hogan, H. (2014). An aging nation: the older population in the United States, Report P25-1140. Washington, DC: US Census Bureau. Phelan, J. C., & Link, B. G. (2015). Is racism a fundamental cause of inequalities in health? Annual Review of Sociology, 41(1), 311–330. doi:10.1146/annurev-soc-073014-112305 Poterba, J. M. (2014). Retirement security in an aging population. American Economic Review, 104(5), 1–30. doi:10.1257/aer.104.5.1 Richardson, L. J., & Brown, T. H. (2016). (En)gendering racial disparities in health trajectories: a life course and intersectional analysis. SSM Popul Health, 2, 425–435. doi:10.1016/j.ssmph.2016.04.011 Schafer, M. H., Ferraro, K. F., & Mustillo, S. A. (2011). Children of misfortune: early adversity and cumulative inequality in perceived life trajectories. American Journal of Sociology, 116(4), 1053–1091. doi:10.1086/655760 Siega-Riz, A. M., Kranz, S., Blanchette, D., Haines, P. S., Guilkey, D. K., & Popkin, B. M. (2004). The effect of participation in the WIC program on preschoolers’ diets. Journal of Pediatrics, 144(2), 229–234. doi:10.1016/j.jpeds.2003.10.052 Sommers, B. D., Baicker, K., & Epstein, A. M. (2012). Mortality and access to care among adults after state Medicaid expansions. N Engl J Med, 367(11), 1025–1034. doi:10.1056/NEJMsa1202099 Starr, P. (1982). The social transformation of American medicine. New York, NY: Basic Books. Sullivan-Marx, E. M. (2015). Looking ahead after 50 years of Medicare. Journal of Gerontological Nursing, 41(9), 15–18. doi:http://​dx​.doi​.org/​10​.3928/​00989134​-20150814​-01

Policy, inequity, and the life course in the US  383 Triece, M. E. (2013). Tell it like it is: women in the national welfare rights movement. Columbia, SC: University of South Carolina Press. Wherry, L. R., Miller, S., Kaestner, R., & Meyer, B. D. (2018). Childhood Medicaid coverage and later-life health care utilization. Review of Economics and Statistics, 100(2), 287–302. doi:10.1162/ REST_a_00677 Wise, P. H., Wampler, N. S., Chavkin, W., & Romero, D. (2002). Chronic illness among poor children enrolled in the temporary assistance for needy families program. Am J Public Health, 92(9), 1458–1461. doi:10.2105/ajph.92.9.1458 Yang, Z., Huang, C., & Phillips, V. (2014). Medicare eligibility age, health disparities, and Medicare reform. Journal of Health Care for the Poor and Underserved, 25(3), 1379–1383. doi:10.1353/ hpu.2014.0119 Yoshida, K. (2018). The stability of Social Security in the United States: the need for a durable institutional design. Journal of Social Policy, 47(2), 397–415. doi:10.1017/S0047279417000599

24. The role of Social Protection Policies in reducing health inequalities Amanda Aronsson, Hande Tugrul, Clare Bambra and Terje Andreas Eikemo

INTRODUCTION Addressing the fundamental causes of health inequalities from a life-course perspective requires policies and programs that are sensitive to all phases of life. Since risk exposure in childhood is particularly impactful and can start an accumulation of negative health consequences, policies that consider a child’s position would be particularly useful in addressing and potentially reducing health inequalities. A distinction can be made between public health policies, health care policies and social policies. In this chapter, the focus is on the latter and especially on Social Protection Policies (SPPs) because they tend to address inequalities early in the chain of causes as they are designed to reduce risks by protecting people against the negative consequences of social disadvantage (UNICEF, 2019). Thus, SPPs should, in theory, level out the unequal distribution of the exposures and vulnerabilities that contributes to health inequalities. This chapter discusses research evidence on the role of Social Protection Policies in reducing child health inequalities and does this from a global perspective. The importance of SPPs in reducing health inequalities was recognized in the highly influential report provided by the WHO Commission on the Social Determinants of Health (Marmot et al., 2008). However, global evidence on the effectiveness of SPPs to reduce health inequalities is scattered, not least because of the large variety of policies, but also because the effectiveness of SPPs is sensitive to context (Moore et al., 2021). Important aspects of context are time and space, cultural, economic, social, and political circumstances and local practices. All of these features may interact with the intervention and thereby potentially produce variation in outcomes. This makes it challenging to synthesize and generalize relevant evidence of SPPs’ effectiveness in reducing child health inequalities. Most evidence for SPPs’ effectiveness is found in high-income countries (HICs) because they have longer-standing SPPs systems and higher investment in evaluation; but because low- and middle-income countries (LMICs) have a greater urgency concerning the identification of effective policies, we will also focus on available evidence in LMICs. We acknowledge the wide variety of SPPs and welfare systems across countries and that comparing policy effectiveness between countries directly is therefore complicated. The aim of this chapter is not to identify the SPP design that works best, but rather offer a broad overview of why and how SPPs can improve health from a life-course perspective. We draw on existing studies and focus especially on parental leave as a “case study” to illustrate our arguments. The chapter starts by clarifying SPPs in terms of their definition and key components, and by highlighting the theoretical pathways as to how they can be understood as a tool to reduce health inequalities. The second part discusses SPPs and child health from a life-course perspec384

The role of Social Protection Policies in reducing health inequalities  385 tive by describing how childhood conditions may be linked to child and adult health and how child-sensitive SPPs can moderate potentially harmful links between childhood conditions and health. The final section focuses on parental leave as a specific example to demonstrate how our theoretical arguments can be applied in terms of its potential inequality-reducing benefits.

SOCIAL PROTECTION POLICIES (SPPs) Definition and Purpose of SPPs Social protection has been conceptualized in multiple ways by different actors and scholars. There are, however, commonalities in the use of the concept that allow for a broad definition of social protection as a set of policies and programs (hence, SPPs) that reduces poverty and deprivation by providing resources that help people manage social risks, labor market risks, and financial risks related to unemployment, poor working conditions, disease, financial crises, and social exclusion (Esping-Andersen, 1990). Examples of policies and programs that can mitigate such risks are unemployment benefits, pensions, health insurance, fee waivers and cash transfers. SPPs are commonly understood as publicly funded policies provided by governments. In HICs, these policies are most commonly institutionalized in welfare state systems. Here, the idea of “decommodification” is central, meaning that a person’s well-being and living standards should not be attached to his or her market performance and that the state should provide key resources such as education and health care to prevent future adverse effects (Schrecker and Bambra, 2015; Bambra et al., 2018). In LMICs, the provision of SPPs more frequently involves international organizations and non-governmental organizations (Barrientos and Hulme, 2009) as well as private actors and community networks. Here, a general rationale behind implementing SPPs is often linked to the perceived need to provide people with measures to cope and mitigate the effects of external shocks and to increase resilience, especially among the poorest who are the most exposed to adverse health outcomes (Gentilini and Omamo, 2011; Jawad, 2019). Moreover, SPPs can be developed with the objectives to (1) prevent the adverse effects of risk by enforcing measures before the risk has occurred, (2) mitigate the impact of a risk by applying measures before the risk occurs (e.g., health insurance), (3) cope with the impact of a risk when it occurs (e.g., in-kind transfers such as food transfers during crisis), and (4) transform the impact of the risk by addressing the underlying sociopolitical causes that drive poverty and vulnerability in the first place (Holzmann and Jørgensen, 2001; Banks et al., 2017). Components of SPPs The design of an SPP determines the criteria for eligibility and access to the program, which in turn are crucial mechanisms in any relationship between interventions and health inequalities. The design rests on the actors, the scope of the policy, and the financing structure. In terms of the actors, we may differentiate between public, private, organizations, and aid donors. Regarding the scope of the policy, it is important to determine whether there are certain conditions attached to accessing the policy, and to decide whether the target group is universal (reaching everybody in the population) or targeted (aimed at a population or geographical

386  Handbook of health inequalities across the life course subgroup). Finally, the financing structure may involve general taxes, employment-based contributions or other financial structures. We may distinguish between four main types of SPPs: social transfers, social insurances, labor market policies (which may also fall under social insurance policies) and social services (which may also fall under social insurance schemes). First, social transfers are often called social safety nets or social assistance. These are non-contributory schemes and are usually provided to the most vulnerable populations (i.e., targeted) to alleviate poverty or to increase access to education, nutrition, and health services. Social transfers can encompass cash transfers, school feeding programs, fee waivers (e.g. removal of school enrollment fees) and in-kind transfers (i.e. non-cash transfers such as food) (Honorati et al., 2015). Second, social insurance usually refers to publicly provided insurances to mitigate loss of income due to sickness, unemployment, old age, parental leave, or disability and encompass health insurance, unemployment benefits, pensions, or allowances. Insurance schemes can be redistributed to reach all people (universal) and are either financed through taxation or contributions linked to employment. Labor market policies include sickness benefits, parental leave, and programs to increase employment opportunities (e.g. skills training or job search assistance) and to reduce working risks (e.g. controlled working hours or minimum wage). Finally, social services, see for example UNICEF (2019), refer to the provision of basic services such as health care, water and sanitation, education, childcare services, and family support services (Bachelet, 2011; Gentilini and Omamo, 2011). Universal and governmentally provided SPPs (as compared to targeted SPPs or SPPs provided by non-governmental actors) may be more effective in reducing poverty and income inequality and improving health inequality (Diderichsen et al., 2012), as exemplified by the successful health outcomes of the Nordic social democratic welfare states up until the 1970s (Bambra, 2021). This argument is based on the assumption that universal policies are more politically sustainable (Diderichsen et al., 2012), less exclusionary and reduce the chances of inequitable, targeted distributions (UNICEF, 2019). Governmentally provided policies, on the other hand, are more likely to be institutionalized and long-lasting compared to the volatility of aid-supported SPPs or SPPs provided by NGOs (Devereux and McGregor, 2014). Although this line of reasoning argument might be valid, some countries also face challenges in the implementation of sufficiently universal and government-supported SPPs. LMICs often face financial constraints and have taxation systems that cannot support a universal scheme, partly as a result of the large informal economy in LMICs that does not contribute to taxes (Marmot et al., 2008). These factors give some explanation as to why universal SPPs are more rare in LMICs and why targeted social assistance programs, often supported by international organizations, donors, NGOs or civil society actors, are more common (Gentilini and Omamo, 2011). Linking SPPs and Health Inequalities People characterized by high SES have more access to material resources, knowledge, power and social networks than people characterized by a lower SES (Bambra, 2016). “Health inequality” refers to the systematic differences in health which exist by socio-economic status (SES) (usually measured in terms of income, education, occupation or area-level deprivation). Inequalities in health are “systematic differences in health between different socio-economic groups within a society. As they are socially produced, they are potentially avoidable and

The role of Social Protection Policies in reducing health inequalities  387 widely considered unacceptable in a civilised society” (Whitehead, 2007). Inequalities in health by SES are not restricted to differences between the most privileged groups and the most disadvantaged; health inequalities exist across the entire social gradient (Marmot, 2006). The social gradient in health runs from the top to the bottom of society and “even comfortably off people somewhere in the middle tend to have poorer health than those above them” (Marmot, 2006). People with higher occupational status (e.g. professionals such as teachers or lawyers) have better health outcomes than those with lower occupational status (e.g. manual workers) (Eikemo et al., 2017). Similarly, people with a higher income or tertiary-level education have better health outcomes than those with a low income or no educational qualifications (Bambra, 2016). SPPs have the potential to reduce or mitigate the effect of poverty and deprivation on health by providing people with more of these resources. For example, SPPs can help people manage social risks, labor market challenges, and financial risks by the provision of more monetary resources, but SPPs can also provide better working conditions and stronger labor rights, and greater access to social services, education and food. SPPs aim to make these resources accessible to either all social groups (universal policies), or to those who are defined as the most vulnerable (targeted policies). While targeted policies aim to reduce health gaps between disadvantaged and less disadvantaged social groups, universal policies aim to reduce the social gradient in health along the whole social ladder. Common for both targeted and universal policies is that they can affect health inequalities through material, behavioral, and psychosocial pathways. In this section we will explain and exemplify how this can be achieved. The materialist approach explains SES inequalities in health by focusing on income and on what income enables such as access to goods and services and the limitation of exposures to physical and psychosocial risk factors (Bartley, 2017). The main social determinants of health are widely considered to be: access to essential goods and services (specifically water and sanitation, and food); housing and the living environment; access to health care; unemployment and social security; working conditions; and transport (Dahlgren and Whitehead, 1991). By way of illustration, a decent income enables access to health care, transport, an adequate diet, quality housing and opportunities for social participation; all of which are health promoting. Material wealth also enables people to limit their exposures to known risk factors for disease such as physical hazards at work or adverse environmental exposures. Materialist approaches give primacy to structure in their explanation of health and health inequalities, looking beyond individual-level factors (agency) in favour of the role of public policy and services such as schools, transport and welfare in the social patterning of inequality. The behavioural approach asserts that the link between SES and health is a result of differences by SES in terms of health-related behaviour. The “pure” behavioural approach asserts that risky health behaviours are more concentrated amongst lower SES groups due to the concentration of individuals with less self-control, lower responsibility, poorer coping abilities, lower health knowledge, and a more short-term outlook on life – an agency-focused explanation (Mackenbach, 2011). A more structural version of the behavioural model – the cultural-behavioural approach – takes into consideration the role of culture and how different cultural norms can pattern the distribution of unhealthy behaviours (Bartley, 2017). It argues that unhealthy behaviours are more common in lower SES groups where these behaviours represent the cultural norm and are more acceptable. Psychosocial explanations focus on how social inequality makes people feel and the effects of the biological consequences of these feelings on health (Bartley, 2017). Feelings of subor-

388  Handbook of health inequalities across the life course dination or inferiority stimulate stress responses which can have long-term consequences for physical and mental health, especially when they are prolonged and chronic. Psychosocial risk factors include low levels of control at work or in the community, plus the stigma as well as the “stress” that results from the lived experience of poverty. It is not straightforward exposures to stressors that matter but the stress response that these stressors produce. In this way the model combines both structure and agency. For example, it may not simply be income level or an adequate working environment alone that leads to good health, but rather how good income and good quality work can make people feel, especially in relation to others (Bartley, 2017). The material, behavioral, and psychosocial explanations of health inequalities are core pathways from which SPPs can be understood to influence SDH and health. We also argue that since SPPs provide resources linked to all these three factors, they could be potential moderating policies in the pathways to obtain equal health. For instance, a policy that can mediate the pathway between occupation and health would be labor legislation that regulates working hours or sets minimum standards for occupational health and safety practices in the workplace. Indeed, the importance of SPPs to health inequalities has led to a further theory of health inequalities – the political economy approach. The political economy approach combines aspects of the materialist, behavioral, and psychosocial explanations with the recognition that the social determinants of health (SDH) are themselves shaped by macro-level structural determinants: politics, the economy, the state, the organization of work, and the labor market (Schrecker and Bambra, 2015). This is referred to collectively as the political economy of health (Doyal and Pennell, 1979). Health inequalities are thus considered as politically determined by institutional (in)action (Beckfield et al., 2015). A wide range of research has demonstrated that even within the constraints of unequal societies, the behavioral, material, and psychosocial determinants of health inequalities are themselves amenable to public policy interventions. Not all countries have the same levels of health inequality, and the political economy approach argues that political choices and the resulting public policies are responsible for these differences (Beckfield and Bambra, 2016). In this way, SPPs can be understood to influence the SDH and health inequalities through the material, behavioral, and psychosocial pathways. From a material explanation in the context of SPPs, resources such as income can enable or prevent access to other goods, services and other material risk factors that have an impact on health (e.g. housing). Unemployment has been linked to poor health (Jin et al., 1995; Bartley et al., 2006; Bambra and Eikemo, 2009) and a material explanation for this association is that unemployment leads to a loss of income which in turn reduces access to goods that are essential for good health, such as health services, a high standard of housing, and nutritious foods. The generosity of replaced wages by means of unemployment insurance can reduce the likelihood of poverty (O’Campo et al., 2015). The material explanation for how cash transfers can improve health equality is similar to that of unemployment benefits: disadvantaged people are provided with material resources (income) to afford access to goods and services that are essential for health. Cash transfers can be efficient in helping households to afford food, school fees and health care through the added income (Adato and Bassett, 2009). Health insurance also exemplifies a material pathway to increasing health equality. Behavioral explanations to health assume that health-related behaviors, such as smoking or dietary choices, are associated with SES. From this perspective SPPs can reduce health inequalities through behavioral pathways whereby the program in some way nudges people towards healthier behaviors. The behavioral assumption is incorporated

The role of Social Protection Policies in reducing health inequalities  389 into the design of conditional cash transfers (CCTs) as well. CCTs are programs in which cash is given if participants comply with certain conditions, assuming that behavioral change is expected from the monetary incentives. Some studies show that CCTs influence behavior but have moderate or no effect on changes in health (Lagarde et al., 2007; Owusu-Addo et al., 2018), and other studies conclude that CCTs are efficient in reducing health inequalities. For instance, Rasella et al. (2013) found that CCTs can reduce childhood mortality by increasing health care access and vaccination coverage. However, the debate about the effectiveness of CCTs in reducing health inequalities through conditions of behavioral change is ongoing, with evidence of no relative effectiveness of CCT compared to unconditional cash transfers (UCTs) in changing behavior. A systematic review by Baird et al. (2014) found that the odds of increased school attendance was equally seen in CCTs and UCTs. Finally, SPPs can also reduce health inequality through psychosocial pathways. This approach illustrates how disadvantage and vulnerability have a psychological impact on people, which in turn can influence health. Poverty and social disadvantage are associated with increased stress, social exclusion, or exposure to trauma or violence, and can cause poor health (Lund et al., 2011). Cash transfers can also improve health by reducing the psychosocial burden on people, as they can increase self-acceptance, hopefulness and autonomy (Attah et al., 2016), including increased social cohesion and civic participation (Owusu-Addo et al., 2018). However, it should be noted that although cash transfers are a widely studied topic in LMICs, the results are mixed. For example, Hjelm et al. (2017) found that cash transfers did not reduce stress and there are continuous concerns that cash transfers might increase stigma and social exclusion resulting from perceptions of unfair eligibility criteria or feelings of shame in receiving cash transfers (Devereux et al., 2011). Psychosocial pathways to health through SPPs are also evident in labor policies. In the case of poor health outcomes from employment conditions stressors, feelings of low control over work, high work demands, insecure employment conditions and lack of support can lead to increased stress, anxiety, and other risk factors for adverse health (Michie and Williams, 2003; Smith et al., 2008; Bambra, 2016). Labor policies such as regulated working hours and skills training can be implemented to reduce work-related psychosocial risk factors and to increase workers’ health. While SPPs can influence the material, psychosocial, and behavioral pathways towards greater equality in health, it must still be acknowledged that SPPs do not emerge or exist in a vacuum. Instead, the international and national political economy sets the agenda for what SPPs can and cannot do in a country as a result of political and economic priorities and opportunities (Schrecker and Bambra, 2015). Schrecker and Bambra (2015) exemplify the political economy perspective in HICs by describing how national priorities have shaped different welfare state regimes (Liberal, Bismarckian, and Social Democratic) and illustrates how these regimes differ in aspects of provision, access, generosity, eligibility and scope of education, health care, housing, and not least SPPs. These aspects differ in terms of their provision between the three welfare state regimes. Generally, in the Liberal regime the state provides little welfare, which instead is subsidized to private schemes, criteria to access SPPs are strict and the support given is minimal. The Bismarckian regime is largely contributory-based with low redistribution. The Social Democratic regime is more universal, with higher redistribution and more generous support (Schrecker and Bambra, 2015). The aspects of access, generosity, eligibility, and scope of the SPPs provided have been seen to affect population health. Health measures such as infant mortality rate have been shown to differ depending on the welfare regime, with lowest mortality rates in the Social Democratic regimes and the highest in the

390  Handbook of health inequalities across the life course Liberal regimes (Schrecker and Bambra, 2015). Overarching political regimes influence the design of SPPs and the design determines access and availability, which in turn can contribute to increased or reduced health inequalities. The political economy perspective reveals that it is not SPP design in itself that determines SPP effectiveness; instead effectiveness is highly dependent on the larger political context that sets the agenda for what SPPs can and cannot do, as illustrated by the health effects of different welfare regimes. Departing from the same argument but turning to LMICs, the political economy perspective can explain both the development of an SPP agenda in LMICs and the opportunities and constraints of SPPs. The SPP agenda in LMICs partly emerged from the Asian financial crisis in the late 1990s, followed by an acceleration in the 2000s as a response to the observed negative consequences of structural adjustment programs and global crises such as the financial crisis of 2008 (Holzmann and Jørgensen, 2001; Gentilini and Omamo, 2011; Devereux and McGregor, 2014; de Haan, 2014). The SPP agenda was created as a tool for development and to protect people from external shocks, and has since developed into a variety of different kinds of policies operating in different domains, protecting people not only from financial shocks, but also from climate change, food insecurity, and health insecurity. Aspects of access, generosity, eligibility, and scope of SPPs in LMICs, as shaped by the political economy, are complex and must be understood both in terms of the global and national political economy structures. For example, globalization has incentivized countries to liberalize and relax their taxation systems to attract foreign investment, resulting in less national funds to be directed to social welfare and SPPs. Furthermore, SPPs in LMICs are generally less institutionalized and involve more non-state actors compared to the welfare state systems in HICs. This creates not only a large variety of SPP types, but also generates a plethora of SPP types that vary in their effectiveness due to the design differences in terms of access, eligibility, and generosity.

SPPs IN A LIFE-COURSE PERSPECTIVE As argued above, SPPs can potentially reduce health inequalities by acting on the SDH through material, behavioral, and psychosocial pathways. We also argued that these links are strongly context dependent. In this section, we bring in the life-course perspective. This is essential for two reasons. First, adverse SDH and conditions in childhood are risk factors for poor health in adulthood. Secondly, parental disadvantage is inherited by their children through intergenerational spillover effects. At the end of this section, it should be evident that child-sensitive SPPs targeted at parents can reduce child health inequalities and be beneficial for health later in life. The life-course approach illustrates the importance of child-sensitive SPPs, taking into account the assumptions in life-course theory that risks accumulate throughout life, that childhood entails critical and sensitive periods, and that intergenerational spillover effects contribute to the health status of children in adult life. Links between Adverse Childhood Exposures and Health Risk accumulates throughout life, where one risk often leads to another, which in turn adds to or exacerbates the chances of experiencing adverse health outcomes later in life. The same is true for protective factors, which too can accumulate and instead lower the chances of expe-

The role of Social Protection Policies in reducing health inequalities  391 riencing poor health outcomes (Braveman and Barclay, 2009). Life-course studies have confirmed the link between childhood conditions and adult health outcomes. It has, for example, been shown that socio-economic conditions in childhood are associated with all-cause and cause-specific mortality, cardiovascular disease, obesity, alcohol consumption, smoking and depression (Braveman and Barclay, 2009) and that childhood maltreatment is linked to inflammation (Danese et al., 2007), depression (Li et al., 2016), diabetes, lung disease, malnutrition, and vision problems (Widom et al., 2012). The exposure to different types of risks (e.g., environmental, socio-economic or behavioral) accumulates during a lifetime, so children born into a high-risk environment have greater chances of being exposed to a chain of risk factors. For example, children growing up in low SES conditions compared to higher SES conditions tend to have lower educational attainment and are more likely to be unemployed, to live in inadequate housing conditions, and to eat less healthily (Law, 2009); conditions which are known to increase the likelihood of poor health. Secondly, so-called critical and sensitive periods occur throughout childhood, and even during pregnancy (Hertzman and Power, 2003), which is when exposures to risks are particularly impactful for health later in life because exposure during critical periods are often irreversible (Ben-Shlomo and Kuh, 2002; Braveman and Barclay, 2009). Therefore, exposure to the SDH in childhood will have either a protective or harmful impact on health during childhood and in adult life. Studies that have focused on critical and sensitive periods have, for example, shown links between low birthweight, coronary heart disease, hypertension, stroke and type 2 diabetes in adulthood (Godfrey and Barker, 2001), and it has been shown that breastfeeding is associated with improved neurocognitive development and a reduced risk of obesity and diabetes in adulthood (Chai et al., 2018). Thirdly, intergenerational spillover effects happen most often when parental risks and parental conditions influence the risk exposure and health of children (Ben-Shlomo and Kuh, 2002). Fetal development is particularly influenced by maternal conditions, especially during pregnancy. A mother who does not have access to nutritious foods has an increased chance of giving birth to a child with low birthweight, and low birthweight in turn is a risk factor for several poor health outcomes, such as type 2 diabetes. Both low birthweight and diabetes could thus have been prevented if the mother had had access to better conditions. Similarly, maternal stress can lead to preterm birth (Shapiro et al., 2013) and maternal SES and poverty have been associated with poor infant health (Astone et al., 2007). While neither of these studies specifically considered health in adulthood, this kind of evidence on the link between maternal and parental conditions and child health have led to an increased understanding of the dependency of child health now and in adult life on parental conditions (Simpson et al., 2021). Evidence from studies on the issues above make it clear that because childhood conditions are a risk factor for poor health later in life, and because parental conditions have a strong influence on those childhood conditions, paying attention to the conditions of parents would be of benefit for the children. It can therefore be theorized that SPPs directed at adults with children during a critical or sensitive period, could interrupt an otherwise potential adverse pathway of risk transferal between parents and children, thereby lowering the accumulation of risks and increasing the chances of a healthier life.

392  Handbook of health inequalities across the life course Child-Sensitive SPPs As discussed in the previous section, accumulated exposures to physical, psychosocial, and material factors early in life influence not only child health but also health later in life. Many of the adverse childhood experiences are preventable, some through SPPs, which is why it is crucial to develop Child-Sensitive Social Protection (CSSP) to reduce childhood vulnerabilities. UNICEF has proposed seven principles for CSSP (UNICEF, 2014): 1. Avoid adverse impacts on children and reduce or mitigate social and economic risks that directly affect children’s lives. 2. Intervene as early as possible where children are at risk, in order to prevent irreversible impairment or harm. 3. Consider the age- and gender-specific risks and vulnerabilities of children. 4. Mitigate the effects of shocks, exclusion, and poverty on families, recognizing that families raising children need support to ensure equal opportunity. 5. Make special provision to reach children who are particularly vulnerable and excluded, including children without parental care, and those who are marginalized within their families or communities due to their gender, disability, ethnicity, HIV and AIDS or other factors. 6. Consider the mechanisms and intra-household dynamics that may affect how children are reached, with particular attention paid to the balance of power between men and women within the household and broader community. 7. Include the voices and opinions of children, their caregivers and youth in the understanding and design of social protection systems and programs. The principles above do not necessitate SPPs to be aimed at children directly, and they might more effectively be targeted at the household or parents in order to improve the living situation of the child (UNICEF, 2019). For example, the insight that a malnourished mother increases the chances of a having a child of low birthweight (Godfrey and Barker, 2001) implies that SPPs with the purpose of improving the financial status of mothers can lead her to afford more nutritious foods and thus reduce the risk of a child of low birthweight. SPPs targeted at mothers can also improve maternal SES and poverty and reduce maternal stress (Shapiro et al., 2013), thereby reducing the likelihood of adverse childhood experiences and increasing the likelihood of a healthier child.

PAID PARENTAL LEAVE To summarize, SPPs can potentially reduce health inequalities by acting on the SDH. SPPs can reduce gaps and/or the whole social gradient and their effects can work through different pathways (material, behavioral, and psychosocial), but it will also depend both on where (context) and when (life-course) the policy is implemented. In particular, we have argued that SPPs targeted at parents can be beneficial for child health, both now and later in life. To illustrate this, this section focuses on a particular type of SPPs, namely paid parental leave, and the evidence on how it can influence child health inequalities.

The role of Social Protection Policies in reducing health inequalities  393 Definition and Current Policy Landscape Paid parental leave (PPL) allows parents to take time away from work following childbirth or adoption while maintaining their jobs and at least a certain share of their income. Though the term may be used for distinctively defining the leave available equally to mothers and fathers, we use it as an umbrella term comprising maternity and paternity leave. PPL policies are typically designed for reconciling competing work and family responsibilities. They aim to promote career continuity and improve family members’ physical and mental health. Seemingly more adult-centric, PPL policies also constitute an essential early intervention to the children’s life. Their benefits on children are channeled through enhancing parents’ time investments, physical and psychological health, and financial status in the short and long term. Regarding children’s health, such factors are associated with improved physical health as well as enhanced cognitive and non-cognitive development. Hence, besides the immediate health outcomes, PPL policies are likely to improve a child’s later life health through their influence on social determinants such as educational attainment and labor market outcomes. Against this backdrop, PPL policies carry important implications for decreasing health inequalities. If designed as universally accessible, they may generate an equalizing effect between families of different income levels and socio-economic backgrounds. Over the past two decades, there has been meaningful progress regarding PPL provision, yet further action is still needed to reach universal and equal access. Today, all OECD countries except for the United States entitle employees to some form of national PPL with partial or full compensation (Raub et al., 2018). Globally, all but eight of the 195 countries provide paid leave to mothers, and about half provide leave to fathers (OECD, 2021). Besides this prevalence, the design of PPL policies still varies considerably across countries.1 The duration of leave, availability of job protection, wage replacement level, and eligibility requirements constitute essential aspects of PPL policies. They determine the availability and generosity of the leaves, thus influencing whether and for how long parents can afford to take leave as well as the direction and magnitude of returns. Whereas the leave duration is critical for the efficient operation of beneficial mechanisms on children’s outcomes, the level of wage replacements, job protection, and eligibility requirements determine the leave’s accessibility. There is sufficient evidence to argue that longer paid leaves reduce the likelihood of adverse health outcomes. Nevertheless, it should be noted that these findings concentrate on the absolute effect of the policy among parents taking up the leave. What about the ones who do not meet eligibility requirements or are eligible but just cannot afford to take the leave due to low wage replacement or lack of job protection? From a social standpoint, high eligibility requirements and low affordability cause disproportionate leave availability to the groups which are already socially advantaged and own the resources to take time off work. Hence, such leaves are prone to exacerbate already existing disparities across socio-economic groups. For instance, the Family Leave and Medical Act in the US provides 12 weeks of unpaid leave following childbirth. Research showed that the take up of leave is lower among the mothers who are younger, lower educated, Black and Hispanic because they cannot afford to be out of work (Hawkins, 2020). Conservative eligibility requirements can also critically decrease the accessibility of PPL for lower socio-economic groups. An employee’s eligibility for PPL can be dependent on conditions such as the length of time at an employer, the formality of employment, or the company’s size. The leave policies are often tied to formal employment where workers contribute to

394  Handbook of health inequalities across the life course the policy through taxation and payments to social security. For instance, a common design in a HIC is the following: the main actor providing maternity leave is the government; the policy is targeted at parents currently in registered employment; and the policy is financed through tax-based contributions and employment-based social insurance contributions. This design is often institutionalized in well functioning welfare states where unemployed parents are also usually entitled to other benefits. Since the policy builds on employment and accounts for the unemployed to some extent, parents in the informal economy classified in neither of these groups face specific challenges. Evidently, the informal sector is significantly smaller in HICs compared to LMICs. Thus, the severity of the issue increases for LMICs where a significant portion of the population participates in the informal economy and consequently does not have access to paid parental leave. How Might Paid Parental Leave Influence Children’s Health Outcomes? PPL’s impact on children’s well-being flows through its influence on parental involvement and physical environment during the early ages. Hence, previous studies often identified two main channels of impact: time investments and family income. The primary and the most direct channel is the increased time invested in the child due to delay in the return to work following birth or adoption (Ruhm, 2000; Dustmann and Schönberg, 2012). Health economic theories consider parental time as one of the direct inputs of child health capital (Leibowitz, 2005). Accordingly, extra time spent with children may decrease mortality and improve physical health by easing crucial childcare practices such as breastfeeding and vaccination. Moreover, psychologists agree that child–mother interactions enhance cognitive and non-cognitive development, essential for future socialization and academic and job performance. Attachment theory posits that a secure and caring mother–child relationship built during the first year is critical for developing a better sense of self-efficacy and trust (Bowlby, 1969). Though children should be exposed to other children and adults in further years, the nurturing of the primary caregiver is particularly advantageous during the first year. Parental leave is likely to increase the quantity of these early interactions. Nevertheless, the quality of nurturing still depends on various factors such as the household’s socio-economic status, education level, degree of parental stress, and the relative quality of alternative daycare arrangements. Another channel that PPL may impact children’s well-being through is family income. The family’s financial resources are essential for buying goods (e.g. dietary supplements, books, games) beneficial for the child’s physical and cognitive development. Furthermore, household income may influence parental stress and anxiety, and thus the quality of parent–child interaction. These suggest that the leave’s impact may differ according to the level of financial benefits offered and the presence of job protection. For instance, during unpaid leave, the family is expected to experience a sudden and temporary reduction in income, particularly if they lack enough means (e.g. accumulated wealth, inheritance) to compensate for the lost wage. Such a reduction in income may extend to the long term if the leave does not include job protection, which may cause parents to experience career disruptions or exit the labor market. Paid Parental Leave’s Short-Term Impact on Children’s Health Outcomes There is ample evidence that more generous PPLs significantly reduce perinatal, neonatal, and child mortality rates during the first year of life. Though these studies mainly focus on

The role of Social Protection Policies in reducing health inequalities  395 European and OECD countries (Winegarden and Bracy, 1995; Ruhm, 2000; Tanaka, 2005; Nandi et al., 2018), their conclusions are also corroborated by the evidence drawn from maternity leaves in LMICs (Nandi et al., 2016). Studying 20 LMICs, Nandi et al. (2016) found that each additional month of paid maternity leave is associated with 7.9 fewer infant deaths per 1,000 live births. Also, these reductions were concentrated mainly on the increase in duration during the post-neonatal period. In contrast, unpaid leaves fail to generate such improved mortality rates. Studies focusing on the extension of unpaid and non-job-protected leaves found no impact on the reductions in infant mortality (Ruhm, 2000; Tanaka, 2005; Staehelin, Bertea and Stutz, 2007; Shim, 2016). Furthermore, focusing on the US, Rossin (2011) found that federally mandated unpaid maternity leave provides minor improvements in infant mortality and premature birth rates only among socio-economically advanced groups. As mothers are reluctant to take the leave without financial compensation, the policy disproportionally benefited those capable of affording the leave, aggravating inequality. Identifying the exact pathways leading to children’s improved health is relatively complex. Nevertheless, studies demonstrated that generous PPLs increase the prevalence of certain childcare practices associated with improved health, such as breastfeeding and vaccination. Breastfeeding is often associated with protection against sudden death syndrome, leukemia, diarrhea, asthma, and diabetes (León-Cava, Pan American Health Organization and LINKAGES Project, 2002; Ip et al., 2007), better child growth (Giugliani et al., 2015) and enhanced cognitive development (León-Cava, Pan American Health Organization and LINKAGES Project, 2002). For optimal support for an infant’s health, the WHO recommends exclusive breastfeeding of six months and continued breastfeeding with complementary foods (World Health Organization, 2001).2 Nevertheless, previous studies have shown that full-time working mothers have a shorter breastfeeding duration than part-time employed or unemployed mothers (Fein and Roe, 1998; Hawkins et al., 2007). Besides, breastfeeding often stops when the mother returns to work (Chen, Wu and Chie, 2006) and women planning on returning to work are less likely to initiate breastfeeding in the first place (Mirkovic et al., 2014; Rollins et al., 2016). PPL policies constitute essential opportunities for achieving the breastfeeding goals indicated by public health agencies. For example, California was the first state in the US to introduce paid family leave in 2004. Huang and Yang (2015) showed that this leave led to increased exclusive and overall breastfeeding through the first three, six, and nine months of infancy. The extension of parental leave is also associated with higher breastfeeding intentions, initiation rates, and duration (Baker and Milligan, 2008; Andres et al., 2016; Heymann et al., 2017). While maternity leave policies are relatively less common in LMICs, the existing evidence aligns with the evidence drawn from HICs. In a longitudinal study assessing the effect of paid maternity leave in 38 LMICs, Chai et al. (2018) found that each additional month is associated with an improved prevalence of early breastfeeding initiation, exclusive breastfeeding, and an increase of 2.2 months in breastfeeding duration. Research has demonstrated that the importance of preventative care is heightened specifically during the prenatal phase and the first 12 months of the child’s life. In a wide range of countries, it is found to lower the rates of diseases causing child mortality such as influenza, measles, and gastroenteritis (Theodoratou et al., 2010; Simons et al., 2012; Enane et al., 2016). Unfortunately, even when vaccinations are free and low cost, thus accessible by low SES groups, the on-time uptake is not universal. Given that one of the barriers is parents’ working schedules, PPLs may enhance adherence to vaccination schedules. Nevertheless, the conclusions drawn from HICs remain mixed. Numerous studies associated longer paid leaves

396  Handbook of health inequalities across the life course in HICs with higher on-time immunization uptake (Berger, Hill and Waldfogel, 2005; Daku, Raub and Heymann, 2012; Ueda et al., 2014). Yet, in a longitudinal study of OECD countries, Tanaka (2005) found no significant link between PPL extension (or any other leaves, including unpaid and non-job protected) and immunization coverage for DTP and measles. The author reasoned this result in the light of already high and stable DTP vaccination rates across these countries. Though the number of studies focusing on PPL’s impact on immunization in LMICs is low, existing evidence specifies a significant positive relationship (Daku, Raub and Heymann, 2012; Hajizadeh et al., 2015). Studying 20 LMICs Hajızadeh et al. (2015) demonstrated that the extension of PPL duration enhanced rates of DTP vaccination which is administered several weeks after birth when women might be expected to return to work. Paid Parental Leave’s Long-Term Impact on Social Determinants of Health The SDH are built on the premise that beyond the availability of health care, health and quality of life are determined by the conditions in which individuals are born, grow, live, work, and age. Such conditions shape the quantity and quality of resources the individual possesses to satisfy needs and stay healthy (Raphael, 2009). Education and income are the most often cited SDH. Channeling through the individual’s socio-economic status, they both predict health during the life course. A well-paid and steady job secures food, housing, and access to quality medical care. Also, studies on education’s lifelong impact on health showed that higher education is strongly associated with increased life expectancy, reduced morbidity, and healthy behaviors (Raphael, 2009; The Lancet Public Health, 2020). It may increase the capacity for better decision-making regarding one’s health and provide better employment opportunities, thus a chance for upward mobility and better financial conditions (Shankar et al., 2013). From a holistic perspective, social determinants are responsible for most of the entrenched health inequalities between and within countries (Marmot et al., 2008). Therefore, any interventions generating an equalizing impact on the social determinants are expected to reduce health inequalities. Studies on PPL’s long-term influence on children found that the introduction of PPL or the generous extensions of modest mandates are associated with enhanced educational and labor market outcomes (Dustmann and Schönberg, 2012; Carneiro, Løken and Salvanes, 2015). Carneiro et al. (2015) built a natural experiment exploiting Norway’s introduction of PPL in 1977 through a policy reform that replaced the three months of unpaid leave with four months of paid plus 12 months of unpaid leave. They found evidence for reduced high school dropout, increased college attendance, and increased earnings. Moreover, evaluating three successive PPL reforms in Germany, Dustmann and Schönberg (2012) showed that the first two reforms, which extended two months of PPL first to six then to ten months, had a slightly positive impact on educational attainment and earnings. The authors based this on the enhanced time invested in the child and slightly increased income available to mothers. On the contrary, the extension of unpaid leave failed to achieve such positive and significant impacts. Dustmann and Schönberg (2012) saw that the third reform in Norway, which extended unpaid leave from 18 to 36 months, actually negatively impacted educational attainment. The main mechanism was argued to be the loss in household income since they also observed that this reform lowered maternal employment in the medium term. Besides average impacts on the overall population, do PPL’s long-term influences hold implications for lowering inequality across socio-economic groups? The answer to this

The role of Social Protection Policies in reducing health inequalities  397 remains highly dependent on the context and institutional background. The introduction of PPL may make leave affordable to parents from low socio-economic backgrounds for the first time, hence, benefit their children’s long-term outcomes more than the rest. This equalizing impact may be more evident if the context lacks a subsidized high-quality daycare system. In such cases, unable to afford parental leave and high-quality private care, low SES parents would have to rely on informal care or low-quality private care. For instance, Carneiro et al. (2015) observed that the long-term positive impacts of the introduction of PPL are greater in magnitude for children of less-educated mothers. Whereas the fall in dropout rates was only 1.8% for the children of mothers with ten years of education or more, it was 3.6 for the children of mothers with less than ten years of education. Also, regarding college attendance rates, they observed a more substantial impact on children born to mothers with lower education. The authors attributed this equalizing effect to the significance of parental nurturing in the first months and its relative advantage compared to daycare alternatives available in Norway during the late 1970s, namely informal care or low-quality private sector care. However, the extension of already long PPL durations may aggravate inequality, especially if an affordable and good-quality formal daycare option is available (Liu and Skans, 2010; Danzer and Lavy, 2018; Danzer et al., 2022). Focusing on Sweden’s 1988 reform which extended PPL from 12 to 15 months, Liu and Skans (2010) observed no impact on overall scholastic performance but a positive effect for children of well-educated mothers. Given that the main alternative daycare arrangement was formal subsidized daycare, they interpret their results as the effects of shifting time from subsidized formal childcare to subsidized parental leave. They concluded that in contexts where subsidized high-quality formal daycare is available, longer PPLs reinforce the relationship between maternal education and school outcomes. This mechanism is likely to transmit inequalities to further generations. Likewise, Danzer et al. (2022) showed that the extension of PPL generated positive effects only in communities without nurseries; no significant effect was observed for children living in communities where nurseries are available. The studies of PPL’s long-term impacts are mainly natural experiments exploiting the policy landscape in a single country. Most of them focus on HICs with social-democratic or conservative welfare systems. Therefore, it is not possible to draw conclusions for countries of different income levels. However, their causal inferences may carry implications to a certain extent for LMICs if they share similar institutional backgrounds and PPL policy design.

CONCLUSION Health inequalities continue to be a key public health problem throughout the world. This is not only a matter of health differences between the most vulnerable and the wealthiest; health inequalities extend along the whole societal hierarchy, with better health enjoyed in the highest social strata. Thus, they are not “natural” or “inevitable”; health inequalities are socially distributed and socially determined. This is also why they are modifiable. In this chapter, we have argued that health inequalities can be reduced by means of SPPs, such as social transfers, social insurances, labor market policies, and social services that target the unequal distribution of social determinants between and within countries. However, identifying the most effective policies is challenging, partly because of the large variety of social policies, health care policies, and public health policies being implemented at the same time, but also because the

398  Handbook of health inequalities across the life course effectiveness of SPPs is sensitive to cultural, economic, social, and political circumstances as well as local practices. What we do know is that SPPs have the potential to reduce or mitigate the effect of poverty and deprivation via material, behavioral, and psychosocial pathways. This can be obtained through universal or targeted policies, in which people are provided with more resources, such as stronger labor rights or greater access to social services, education and food. As we have shown in more detail, child-sensitive SPs directed at adults, such as paid parental leave, can greatly improve childhood conditions, which is an important prerequisite for health later in life. If SPPs are implemented as intended, we will come closer to reducing unnecessary health gaps between more and less disadvantaged social groups and decreasing the social gradient in health along the whole social ladder across the world.

ACKNOWLEDGEMENT This study was supported by a grant awarded by the Research Council of Norway (project number 288638) to the Centre for Global Health Inequalities Research at the Norwegian University for Science and Technology.

NOTES 1. During the past two decades, the majority of the countries experienced a significant upward trend in the duration of PPL. Today more than half of all countries (54%), meet the ILO standard of at least 14 weeks of paid maternity leave. However, the gap between high- and low-income countries has grown considerably. In 1995 the percentage of countries guaranteeing at least 14 weeks of leave was 56% among high-income and 28% among low-income countries. By 2015 the contrast reached 77% vs. 44% (UNICEF, 2019). As per wage replacements, ILO recommends at least two-thirds of regular wages to keep families out of poverty. Most of the HIC countries meet this recommendation, whereas the rate falls relatively in LMICs. Presence of job protection is fairly prevalent among PPLs worldwide. Seventy-eight percent of all countries provide job protection guarantees during the entire leave. However, countries of different income levels again show a certain divergence in this aspect. Whereas 87% of high-income countries guarantee job protection, this figure falls to 67% among low-income countries (UNICEF, 2019). 2. However, the existence of causality between breastfeeding and the child’s health, as well as the persistence of its short-term benefits in the future, remains inconclusive.

REFERENCES Adato, M. & Bassett, L. (2009) ‘Social protection to support vulnerable children and families: the potential of cash transfers to protect education, health and nutrition’, AIDS Care, 21(1), 60–75. Andres, E. et al. (2016) ‘Maternity leave access and health: A systematic narrative review and conceptual framework development’, Maternal and Child Health Journal, 20(6), pp. 1178–1192. doi:10.1007/ s10995-015-1905-9. Astone, N. M., Misra, D. & Lynch, C. (2007) ‘The effect of maternal socio-economic status throughout the lifespan on infant birthweight’, Paediatr Perinat Epidemiol, 21(4), 310–318. Attah, R., et al. (2016) ‘Can social protection affect psychosocial wellbeing and why does this matter? Lessons from cash transfers in sub-Saharan Africa’, Journal of Development Studies, 52(8), 1115–1131.

The role of Social Protection Policies in reducing health inequalities  399 Bachelet, M. C. (2011) Social protection floor for a fair and inclusive globalization: Report of the Social Protection Advisory Group). Geneva: International Labour Office. Baird, S., et al. (2014) ‘Conditional, unconditional and everything in between: a systematic review of the effects of cash transfer programmes on schooling outcomes’, Journal of Development Effectiveness, 6(1), 1–43. Baker, M. and Milligan, K. (2008) ‘Maternal employment, breastfeeding, and health: Evidence from maternity leave mandates’, Journal of Health Economics, 27(4), pp. 871–887. doi:10.1016/j. jhealeco.2008.02.006. Bambra, C. (2016) Health divides: Where you live can kill you. Bristol: Bristol University Press. Bambra, C. (2021) ‘Levelling up: Global examples of reducing health inequalities’, Scandinavian Journal of Public Health, 140349482110224-14034948211022428. Bambra, C. & Eikemo, T. A. (2009) ‘Welfare state regimes, unemployment and health: a comparative study of the relationship between unemployment and self-reported health in 23 European countries’, J Epidemiol Community Health, 63(2), 92–98. Bambra, C., Reibling, N. & Mcnamara, C. L. (2018) ‘States of health: Welfare regimes, health and healthcare’. In: Greve, B. (ed.) Routledge Handbook of the Welfare State. Second edition. London: Routledge. Banks, L. M., et al. (2017) ‘Disability and social protection programmes in low- and middle-income countries: A systematic review’, Oxford Development Studies, 45(3), 223–239. Barrientos, A. & Hulme, D. (2009). ‘Social protection for the poor and poorest in developing countries: Reflections on a quiet revolution’, Oxford Development Studies, 37(4), 439–456. https://​doi​.org/​10​ .1080/​13600810903305257 Bartley, M. (2017) Health inequality: An introduction to concepts, theories and methods. Cambridge: Polity Press. Bartley, M., Ferrie, J. & Montgomery, S. M. (2006) ‘Health and labour market disadvantage: Unemployment, non-employment, and job insecurity’. In: Marmot, M. & Wilkinson, R. G. (eds.) Social determinants of health. Oxford, UK: Oxford University Press. Beckfield, J. & Bambra, C. (2016) ‘Shorter lives in stingier states: Social policy shortcomings help explain the US mortality disadvantage’, Social Science & Medicine, 171, 30–38. Beckfield, J., et al. (2015) ‘An institutional theory of welfare state effects on the distribution of population health’, Social Theory & Health, 13(3–4), 227–244. Ben-Shlomo, Y. & Kuh, D. (2002) ‘A life course approach to chronic disease epidemiology: Conceptual models, empirical challenges and interdisciplinary perspectives’, Int. J. Epidemiol, 31(2), 285–293. Berger, L. M., Hill, J. and Waldfogel, J. (2005) ‘Maternity leave, early maternal employment and child health and development in the US’, Economic Journal, 115(501), pp. F29–F47. doi:10.1111/j.0013-0133.2005.00971.x. Bowlby, J. (1969) Attachment and loss. New York: Basic Books. Braveman, P. & Barclay, C. (2009) ‘Health disparities beginning in childhood: A life-course perspective’, Pediatrics, 124, pp. S163–S175. Carneiro, P., Løken, K. V. and Salvanes, K. G. (2015) ‘A flying start? Maternity leave benefits and long-run outcomes of children’, Journal of Political Economy, 123(2), pp. 365–412. doi:10.1086/679627. Chai, Y., Nandi, A. & Heymann, J. (2018) ‘Does extending the duration of legislated paid maternity leave improve breastfeeding practices? Evidence from 38 low-income and middle-income countries’, BMJ Glob Health, 3(5), e001032–e001032. Chen, Y. C., Wu, Y.-C. and Chie, W.-C. (2006) ‘Effects of work-related factors on the breastfeeding behavior of working mothers in a Taiwanese semiconductor manufacturer: A cross-sectional survey’, BMC Public Health, 6(1), p. 160. doi:10.1186/1471-2458-6-160. Dahlgren, G. & Whitehead, M. (1991) Policies and strategies to promote social equity in health. Stockholm: Institute for Future Studies. Daku, M., Raub, A. and Heymann, J. (2012) ‘Maternal leave policies and vaccination coverage: A global analysis’, Social Science & Medicine, 74(2), pp. 120–124. doi:10.1016/j.socscimed.2011.10.013. Danese, A., et al. (2007) ‘Childhood maltreatment predicts adult inflammation in a life-course study’, Proceedings of the National Academy of Sciences, 104(4), 1319–1324.

400  Handbook of health inequalities across the life course Danzer, N. and Lavy, V. (2018) ‘Paid parental leave and children’s schooling outcomes’, Economic Journal, 128(608), pp. 81–117. doi:10.1111/ecoj.12493. Danzer, N. et al. (2022) ‘Parental leave, (in)formal childcare and long-term child outcomes’, Journal of Human Resources, 57(6), pp. 1826–1884. de Haan, A. (2014) ‘The rise of social protection in development: Progress, pitfalls and politics’, European Journal of Development Research, 26(3), 311–321. Devereux, S. & Mcgregor, J. A. (2014) ‘Transforming social protection: Human wellbeing and social justice’, European Journal of Development Research, 26(3), 296–310. Devereux, S., Mcgregor, J. A. & Sabates-Wheeler, R. (2011) ‘Introduction: Social protection for social justice’, IDS Bulletin, 42(6), 1–9. Diderichsen, F., et al. (2012) ‘Health inequality – determinants and policies’, Scand J Public Health, 40(8_suppl), 12–105. Doyal, L. & Pennell, I. (1979) The political economy of health. London: Pluto Press. Dustmann, C. and Schönberg, U. (2012) ‘Expansions in maternity leave coverage and children’s long-term outcomes’, American Economic Journal: Applied Economics, 4(3), pp. 190–224. doi:10.1257/app.4.3.190. Eikemo, T. A., et al. (2017) ‘The First Pan-European Sociological Health Inequalities Survey of the General Population: The European Social Survey Rotating Module on the Social Determinants of Health’, European Sociological Review, 33(1), 137–153. Enane, L. A., et al. (2016) ‘Impact of rotavirus vaccination on hospitalizations and deaths from childhood gastroenteritis in Botswana’, Clinical Infectious Diseases, 62(suppl_2), pp. S168–S174. doi:10.1093/ cid/civ1210. Esping-Andersen, G. (1990) The three worlds of welfare capitalism. Cambridge: Polity Press. Fein, S. B. and Roe, B. (1998) ‘The effect of work status on initiation and duration of breast-feeding’, American Journal of Public Health, 88(7), pp. 1042–1046. Available at: https://​www​.ncbi​.nlm​.nih​ .gov/​pmc/​articles/​PMC1508266/​ (accessed: 9 December 2021). Gentilini, U. & Omamo, S. W. (2011) ‘Social protection 2.0: Exploring issues, evidence and debates in a globalizing world’, Food Policy, 36(3), 329–340. Giugliani, E. R. J., et al. (2015) ‘Effect of breastfeeding promotion interventions on child growth: a systematic review and meta-analysis’, Acta Paediatrica, 104(467), pp. 20–29. doi:10.1111/apa.13160. Godfrey, K. M. & Barker, D. J. (2001) ‘Fetal programming and adult health’, Public Health Nutrition, 4(2b), 611–624. Hajizadeh, M., et al. (2015) ‘Paid maternity leave and childhood vaccination uptake: Longitudinal evidence from 20 low-and-middle-income countries’, Social Science & Medicine, 140, pp. 104–117. doi:10.1016/j.socscimed.2015.07.008. Hawkins, D. (2020) ‘Disparities in the usage of maternity leave according to occupation, race/ethnicity, and education’, American Journal of Industrial Medicine, 63(12), pp. 1134–1144. doi:10.1002/ ajim.23188. Hawkins, S. S., et al. (2007) ‘The impact of maternal employment on breast-feeding duration in the UK Millennium Cohort Study’, Public Health Nutrition, 10(9), pp. 891–896. doi:10.1017/ S1368980007226096. Hertzman, C. & Power, C. (2003) ‘Health and human development: Understandings from life-course research’, Developmental Neuropsychology, 24(2–3), 719–744. Heymann, J., et al. (2017) ‘Paid parental leave and family wellbeing in the sustainable development era’, Public Health Reviews, 38(1), p. 21. doi:10.1186/s40985-017-0067-2. Hjelm, L., et al. (2017) ‘Poverty and perceived stress: Evidence from two unconditional cash transfer programs in Zambia’, Soc Sci Med, 177, 110–117. Holzmann, R. & Jørgensen, S. (2001) ‘Social risk management: A new conceptual framework for social protection, and beyond’, International Tax and Public Finance, 8(4), 529–556. Honorati, M., Gentilini, U. & Ruslan, Y. (2015) The state of social safety nets 2015. Washington, DC: World Bank Group. Huang, R. and Yang, M. (2015) ‘Paid maternity leave and breastfeeding practice before and after California’s implementation of the nation’s first paid family leave program’, Economics & Human Biology, 16, pp. 45–59. doi:10.1016/j.ehb.2013.12.009.

The role of Social Protection Policies in reducing health inequalities  401 Ip, S., et al. (2007) ‘Breastfeeding and maternal and infant health outcomes in developed countries’, Evidence Report/Technology Assessment, (153), pp. 1–186. Jawad, R. (2019) ‘A new era for social protection analysis in LMICs? A critical social policy perspective from the Middle East and North Africa region (MENA)’, World Development, 123, 104606. Jin, R. L., Shab, C. P. & Svoboda, T. J. (1995) ‘The impact of unemployment on health: A review of the evidence’, CMAJ, 153(5), 529–540. Lagarde, M., Haines, A. & Palmer, N. (2007) ‘Conditional cash transfers for improving uptake of health interventions in low- and middle-income countries: A systematic review’, JAMA, 298(16), 1900–1910. Law, C. (2009) Life-course influences on children’s futures. In: Graham, H. (ed.) Understanding Health Inequalities, 2nd ed. McGraw Hill/Open University Press. Leibowitz, A. A. (2005) ‘An economic perspective on work, family, and well-being’, Work, Family, Health, and Well-Being, pp. 187–200. León-Cava, N., Pan American Health Organization and LINKAGES Project (eds) (2002) Quantifying the benefits of breastfeeding: A summary of the evidence. Washington DC: Food and Nutrition Program, Pan American Health Organization : LINKAGES Project, Academy for Educational Development. Li, M., D’arcy, C. & Meng, X. (2016) ‘Maltreatment in childhood substantially increases the risk of adult depression and anxiety in prospective cohort studies: Systematic review, meta-analysis, and proportional attributable fractions’, Psychological Medicine, 46(4), 717–730. Liu, Q. and Skans, O.N. (2010) ‘The duration of paid parental leave and children’s scholastic performance’, B.E. Journal of Economic Analysis & Policy, 10(1). doi:10.2202/1935-1682.2329. Lund, C. D., et al. (2011) ‘Poverty and mental disorders: Breaking the cycle in low-income and middle-income countries’, Lancet, 378(9801), 1502–1514. Mackenbach, J. P. (2011) ‘Can we reduce health inequalities? An analysis of the English strategy (1997–2010)’, Journal of Epidemiology & Community Health, 65(7), 568–575. Marmot, M. (2006) Introduction. In: Marmot, M. & Wilkinson, R. G. (eds.) Social determinants of health. Oxford: Oxford University Press. Marmot, M. P., et al. (2008) ‘Closing the gap in a generation: Health equity through action on the social determinants of health’, Lancet, 372(9650), 1661–1669. Michie, S. & Williams, S. (2003) ‘Reducing work related psychological ill health and sickness absence: A systematic literature review’, Occup Environ Med, 60(1), 3–9. Mirkovic, K. R., et al. (2014) ‘Maternity leave duration and full-time/part-time work status are associated with US mothers’ ability to meet breastfeeding intentions’, Journal of Human Lactation, 30(4), pp. 416–419. doi:10.1177/0890334414543522. Moore, G., et al. (2021) ‘Adapting interventions to new contexts—the ADAPT guidance’, BMJ (Online), 374, n1679–n1679. Nandi, A., et al. (2016) ‘Increased duration of paid maternity leave lowers infant mortality in low- and middle-income countries: A quasi-experimental study’, PLOS Medicine. Edited by D. Osrin, 13(3), p. e1001985. doi:10.1371/journal.pmed.1001985. Nandi, A., et al. (2018) ‘The impact of parental and medical leave policies on socioeconomic and health outcomes in OECD countries: A systematic review of the empirical literature: Parental and medical leave policies in OECD countries’, Milbank Quarterly, 96(3), pp. 434–471. doi:10.1111/1468-0009.12340. O’Campo, P., et al. (2015) ‘Social welfare matters: A realist review of when, how, and why unemployment insurance impacts poverty and health’, Social Science & Medicine, 132, 88–94. OECD (2021) OECD Family Database. OECD. Available at: https://​www​.oecd​.org/​els/​family/​database​ .htm (accessed: 29 November 2021). Owusu-Addo, E., Renzaho, A. M. N. & Smith, B. J. (2018) ‘The impact of cash transfers on social determinants of health and health inequalities in sub-Saharan Africa: A systematic review’, Health Policy Plan, 33(5), 675–696. Raphael, D. (2009) Social determinants of health: Canadian perspectives. Canadian Scholars’ Press. Rasella, D. P., et al. (2013) ‘Effect of a conditional cash transfer programme on childhood mortality: A nationwide analysis of Brazilian municipalities’, Lancet, 382(9886), 57–64. Raub, A., et al. (2018) ‘Paid parental leave: A detailed look at approaches across OECD countries’, WORLD Policy Analysis Center [Preprint].

402  Handbook of health inequalities across the life course Rollins, N. C., et al. (2016) ‘Why invest, and what it will take to improve breastfeeding practices?’, The Lancet, 387(10017), pp. 491–504. doi:10.1016/S0140-6736(15)01044-2. Rossin, M. (2011) ‘The effects of maternity leave on children’s birth and infant health outcomes in the United States’, Journal of Health Economics, 30(2), pp. 221–239. doi:10.1016/j.jhealeco.2011.01.005. Ruhm, C. J. (2000) ‘Parental leave and child health’, Journal of Health Economics, 19(6), pp. 931–960. doi:10.1016/S0167-6296(00)00047-3. Schrecker, T. & Bambra, C. (2015) How politics makes us sick: Neoliberal epidemics. London: Palgrave Macmillan. Shankar, J., et al. (2013) ‘Education as a social determinant of health: Issues facing indigenous and visible minority students in postsecondary education in western Canada’, International Journal of Environmental Research and Public Health, 10(9), pp. 3908–3929. doi:10.3390/ijerph10093908. Shapiro, G. D., et al. (2013) ‘Psychosocial stress in pregnancy and preterm birth: Associations and mechanisms’, J Perinat Med, 41(6), 631–645. Shim, J. (2016) ‘Family leave policy and child mortality: Evidence from 19 OECD countries from 1969 to 2010’, International Journal of Social Welfare, 25(3), pp. 215–221. doi:10.1111/ijsw.12186. Simons, E., et al. (2012) ‘Assessment of the 2010 global measles mortality reduction goal: Results from a model of surveillance data’, The Lancet, 379(9832), pp. 2173–2178. doi:10.1016/ S0140-6736(12)60522-4. Simpson, J., et al. (2021) ‘Effects of social security policy reforms on mental health and inequalities: A systematic review of observational studies in high-income countries’, Social Science & Medicine, 272, 113717. Smith, P. M., et al. (2008) ‘Examining the relationships between job control and health status: A path analysis approach’, J Epidemiol Community Health, 62(1), 54–61. Staehelin, K., Bertea, P. C. and Stutz, E. Z. (2007) ‘Length of maternity leave and health of mother and child – a review’, International Journal of Public Health, 52(4), pp. 202–209. doi:10.1007/ s00038-007-5122-1. Tanaka, S. (2005) ‘Parental leave and child health across OECD countries’, Economic Journal, 115(501), pp. F7–F28. doi:10.1111/j.0013-0133.2005.00970.x. The Lancet Public Health (2020) ‘Education: A neglected social determinant of health’, The Lancet Public Health, 5(7), p. e361. doi:10.1016/S2468-2667(20)30144-4. Theodoratou, E., et al. (2010) ‘The effect of Haemophilus influenzae type b and pneumococcal conjugate vaccines on childhood pneumonia incidence, severe morbidity and mortality’, International Journal of Epidemiology, 39(Suppl 1), pp. i172–i185. doi:10.1093/ije/dyq033. Ueda, M., et al. (2014) ‘Maternal work conditions, socioeconomic and educational status, and vaccination of children: A community-based household survey in Japan’, Preventive Medicine, 66, pp. 17–21. doi:10.1016/j.ypmed.2014.05.018. UNICEF. (2014) Assessing child-sensitivity in social protection. A toolkit for social transfers. United Nations Children’s Fund Regional Office for South Asia. UNICEF. (2019) UNICEF’s Global social protection programme framework. New York: UNICEF. Whitehead, M. (2007) ‘A typology of actions to tackle social inequalities in health’, Journal of Epidemiology & Community Health, 61(6), 473–478. Widom, C. S., et al. (2012) ‘A prospective investigation of physical health outcomes in abused and neglected children: New findings from a 30-year follow-up’, Am J Public Health, 102(6), 1135–1144. Winegarden, C. R. and Bracy, P. M. (1995) ‘Demographic consequences of maternal-leave programs in industrial countries: Evidence from fixed-effects models’, Southern Economic Journal, 61(4), p. 1,020-035. World Health Organization (2001) Report of the expert consultation of the optimal duration of exclusive breastfeeding, Geneva, Switzerland, 28–30 March 2001. World Health Organization.

Index

absolute inequalities 307, 310–312, 318–19, 321–2 accumulation of adversity in form of stressors 68 as driving impulse of capitalism 36 human capital 205, 297 process of positive feedback 23 in relation to disease 359–60 as relatively simple model 119 of risks 2, 298, 300, 355–6, 361, 391 as specific case of cumulative dis/advantage 36, 41 theories of 307, 310 active labour market policies (ALMP) 182–3 adolescence cognitive skills 239 cohesion with grandparents 195 health during affecting learning 299 effect of anticipating age when educational decisions are made 300 future research avenues 301 immune system reaching peak performance 358 infectious disease 360 mental, and green space presence in childhood 336 poor 298 school drop-out 302 and social disadvantage 153, 356 and socioeconomic attainment 152, 295–6 urgent need for early life intervention 154 health inequalities long-term impact of socioeconomic, on later life 150–153 nature of, and causes of 147–50 need for early life intervention 154 health trajectories 151–2, 153 height as predictor of socioeconomic outcomes 294 importance of development stage 146–7 as important milestone in health inequality research 355 life course perspective on 17, 24, 146, 147, 151–2, 160–161

nutrition transition 35 obesity 152, 237 smoking 148, 149, 151, 152, 244 and social capital concept of 162 findings 165–9 hypotheses, data and measures 162–5 index 165 and social inequality 161, 162 and work in later life 173, 175 age-as-a-leveller hypothesis 51, 218, 307, 308–10, 320 age gradients 24, 25 age-period-cohort (APC) models 84 age roles 15, 17, 24 age, theories on 308–10 ageing biological 49, 65, 308–9, 358 bringing increased risk of chronic diseases, disability and mortality 215 and children 221, 223 chronological age and functional age 24 and cognitive abilities 205, 277, 280, 285 compositional changes 321 effect of negative stereotypes on grandparents’ well-being 192 family relationships becoming increasingly important 196 in health capital theory 49 healthy, UN decade of 271–2 negative stereotypes 192 and resilience 65, 286 social determinants 255 theory of selective optimization with compensation 20 viewed as gradual succession of assimilation and accommodation 20–21 see also epigenetic ageing; health in later life; later life air pollution adverse health effects earlier reviews on 327 epidemiological and toxicological evidence 330–331 identification challenges 331–2 medium- to long-term health consequences 334–5 quasi-experimental approaches 332–3

403

404  Handbook of health inequalities across the life course short-term variability in exposure and immediate health consequences 333–4 child susceptibility to 329 housing prices sensitive to 339 link with noise 337–8, 343 avoidance behaviour 332, 334 before–after estimator 105–6 bias applied research guidance 108–9 avoidance behavior as source of 332 baseline 95–6, 100 causal effect heterogeneity 95–6, 100, 104 collider 96–7, 98, 108 confounding 96–7, 98, 99, 103, 106, 130, 138, 180, 331, 332, 338 endogenous selection 96–7 and inverse probability weighting 121 and IV method 128, 130, 138 latent class analysis 90 overcontrol 96–7, 98, 99, 108 parametric regression 102 present 47, 53 recall 271, 286 regression discontinuity design 104 simultaneity 128 “survivorship” 219, 270 treatment effect 128 versus variance 112–14, 116, 118 bioecological model/enhancement 66 biological ageing 49, 65, 308–9, 358 biological processes 218, 308–9, 320, 356 biology 5–6 black box models 113, 114–15, 120, 121 cash assistance 373, 376, 377 causal graphs 96–8 causal hypotheses 93, 94, 108 causal identification 97–8, 99, 105, 106, 108–9, 327, 331, 332, 334, 338, 341 causal inference applied research guidance 108–9 causal graphs, hypotheses and counterfactuals 94–8 cross-sectional methods selection on observables 98–102 selection on unobservables 103–4 experiments randomized 93–4 seen as gold standard for 93, 128 explanation 93 limitations to using observable data 128–9, 270, 283 longitudinal methods 105–8

machine learning for 124 problem of mixing aims 93 causal model for health inequalities benefits and limitations of study 247–8 constructs and causal relationships 237–8 direct pathways originating in early conditions 238–9 indirect pathways originating in early conditions 239–40 manifestation of pathways 240–241 other causal pathways 240 causal relationships health inequalities 237–8 between origin/education/destination 22–3 with respect to health effects of retirement 204, 205, 210–211, 212 causality and absolute perspective 322 adjusting for confounders 337 counterfactual model of 94–6, 180 health inequalities across life course 4 IVs used to address issues of 128, 133–4, 140 mixed methods to address 133–4, 140–141 reverse 205, 210–211, 222, 225 social, and selection processes 217–18 chains of risk 25, 133, 275–6, 279, 282–3, 284, 285, 297, 356 child health child sensitive SPPs 392, 398 links between adverse childhood exposures and health 390–391 paid parental leave influencing outcomes 394–7 role of SPPs components 385–6 context 384–5 definition and purpose 385 in life course perspective 390–392 linking with health inequalities 386–90 in social inequalities and social capital theoretical framework 161–2 and socioeconomic attainment 293–5 child-sensitive SPPs 392 childhood genetic confounding 283–4 interventions and outcomes 372–4 as sensitive period 282 as steppingstone for chains of risk 282–3 childhood exposures adverse, links with health 390–391 and chains of risk model 279 as “critical” 341 long-lasting effects on cardiovascular health 275

Index  405 role of genes in 279–80 social consequences of long-term 286–7 genetic influences 279–80 and long-term cognitive outcomes 280–281 childhood poverty context 252–3 framework and life course model 254–5 future research avenues and policy 271–2 hypotheses 256–7 methods and materials childhood information 258–60 doing cross-country comparison when indicators are not exactly the same 258 latent construct to recover child poverty status obtained in hindsight 257–8 number of countries studied 253–4 results 260–267 study discussion and conclusion 267–9 limitations and strengths 270–271 theory 255–6 childhood socioeconomic conditions and health in later life cognition levels and changes 276–8 context 275–6 dementia diagnosis threshold 277–8 evidence from Sweden 280–281 framework for understanding impact of early life factors on cognition and dementia 278–9 link with adult health outcomes 391, 398 role of genes in childhood exposures– late-life cognition associations 279–80 seen as risk factor for old age disability, dysfunction, disease and mortality 252 social exposures and long-term cognitive outcomes 280–281 and mortality in later life 281–2 evidence from Sweden 284–5 framework for understanding impact 282–4 study conclusions 285–7 see also long arm of childhood conditions childhood welfare state 369–70, 372, 379 chronic disease see disease: chronic classification and regression trees (CART) 116 cognition framework for understanding impact of early-life factors on 278–9

late-life associations, role of genes 279–80 levels and changes and childhood social exposures 280 and dementia 277–8, 285–6 distinguishing between 276–7 long-term cognitive outcomes and childhood social exposures 280–281, 285 representation of life course development of cognitive function 278 research recommendations 285–6 cognitive ability and ageing 205, 277, 280, 285 effects of retirement on 26, 205–6, 211 gene-environment interaction 66, 68, 279 of grandparents 194 heritability of 379 influencing health outcomes 239–40 of parents with many children 221 preventing morbidity and premature mortality 61 of smokers’ offspring 247 comparative cohort analysis 33–6 compensation 66 complexity 3 compulsory schooling 52, 129, 140 conditional independence assumption (CIA) 98–9, 101, 102 counterfactual model 94–6, 180 see also potential outcome (PO) approach Covid-19 as biological threat 5–6 biomedical risk factors or pre-existing conditions 351, 353 cumulative incidences per 100,000 357 emphasising SEP-based differences for health risks 153 genetic susceptibility 61 model of association between social position, mediating factors and 353–4 pronounced inequality in incidence and severity 361 research interest in social inequality in 350, 352–3 risk of aggravating inequalities 46, 173, 342, 352–3 and social capital of young people 169 vaccination inequality 358–9 critical periods and adverse childhood experiences 328, 355, 391 and immune system 358 parental circumstances and early life playing significant role 238 pollution exposure 330

406  Handbook of health inequalities across the life course relating primarily to prenatal and early infancy periods 24, 173, 355 and sensitive periods distinguishing between 355 infectious disease 357–9, 361 instrumental variable approaches, early in life 132–3 models 297, 298, 301 occurrence throughout childhood 391 as relatively simple models 119 US welfare interventions and outcomes 372 Medicare 377 WIC participation during 374 use in explaining chronic diseases 26 cross-sectional data 17, 83–4, 218–19, 223 cross-sectional methods selection on observables 98–9 inverse probability weighting 102 matching 101–2 regression and regression adjustment 99–100 selection on unobservables instrumental variable estimator 103–4 regression discontinuity design 104 cumulative age-specific incidences of infectious disease 356–7 cumulative dis/advantage definition 36–7 in emerging adulthood 151–2, 153 on global scale 36–8 in life course perspective on environmental and health inequalities 328–9 longitudinal perspective, and stressors 67–8 operating at level of nation states and in historical time 41 and pollution exposure 331 social relationships as central to 220 tending to generate diverging individual-level heath inequalities 321 theory of 151, 310 dementia and childhood social exposures 280–281 diagnosis threshold 277–8 framework for understanding impact of early-life factors on 278–9 rates increasing with age 307, 311 research recommendations 285–6 depression and access to green space 337 during adolescence 147, 153, 296 in adolescent/adult grandchildren 195 and childhood poverty 253, 254, 258, 259, 260, 262–4, 265, 271, 272

and childlessness 221 effects of retirement 206–7 family relationships and later life health 221 gender differences 222–3 and grandparents 191, 194 and SES-related health inequalities 217 socioeconomic conditions in childhood associated with 391 and volunteering 204 and work environment 178 description 93 diabetes and air pollution risks 331 breastfeeding associated with protection against 395 childhood maltreatment linked to 391 and early-life nutritional scarcity 35 global rates of 32 and health inequalities 241–4 interaction with infectious diseases 360 life table methods 83 in socially disadvantaged populations 353 in sub-Saharan Africa 33 see also type II diabetes (T2D) difference-in-differences (DID) estimator 106–8, 129, 132, 211, 335 direct effect hypothesis 220 directed acyclic graphs (DAGs) 96–100, 103–4, 109, 286 disease and air pollution 330–331, 334 binary operationalisation of 310–311 burden of 52, 132, 233, 349 childhood conditions as risk factor for later life 252, 327 chronic ageing process and increased risk of 215 consequences of good care 353 dominant and ascending views of 26 formula-fed infants having higher risk of 39 mapping review 315 mostly multifactorial aetiology 359 principle of time lag between exposure and 359 as socially acceptable condition in later life 24 turning points 25 and educational attainment 295–6 and fetal origins hypothesis 293, 328 and genetics 59–65 growing evidence on developmental origins of 47 and mediation analysis 179 non-communicable 32, 152

Index  407 occupational 176–8, 181–2 prevalence, wide differences between countries 26 and retirement 207–8 sensitive periods and chains of risk models 278–9 and SEP 179, 180, 181 smoking-related 237 social capital as resource for recovery from 67 see also infectious disease disparities see inequalities DNA methylation see methylation early childhood origins of social class health inequalities benefits and limitations of study 247–8 causal model for health inequalities 237–41 context 233–4 epidemiological regimes, social classes, and health inequalities 234–6 examples of direct and indirect pathways 241–6 intergenerational transmission of inequalities 246–7 magnitude of health and mortality inequalities 236–7 unequal allocation of 234–5 early health conditions correlation with higher rates of ill health in later life 24 direct pathways originating in 238–9 indirect pathways originating in 239–40 manifestation of pathways 240–241 other causal pathways 240 effects of 298–300, 301 and socioeconomic attainment adolescent health 295–6 future research avenues 301–2 introduction 292–3 life course framework of 296–300 perinatal and childhood health 293–5 strong associations between 292 early-life factors causality question 4 and cognitive change 277 framework for understanding impact on cognition and dementia 278–9 economic development in agricultural economies 39–40 bringing nutritional diversity to country 35 and cohorts 34, 35 and heterogeneity within SSA countries 41

nutrition transition and cumulative dis/ advantage on global scale 36–8 obesity and metabolic patterns 38 economic independence, undermining 39–40, 42 economic theories of health economic approach basic pillars 46 challenges to assumptions 46–7 shift from traditional approaches 47 shift towards natural experiments 47–8 health capital theory 48–9 conceptual insights 49–52 further extensions 52–4 on parental time 394 recommendations 54 typically stated as mathematical optimization problem 53 education of adult children, affecting their parents’ health in later life 224 and allocation of individuals in social classes 239–40 and chains of risk 283 in counterfactual model of causality 94–6 and dementia risk 280–281 and early health effects 299–300, 301 economic approach 46–8, 50, 52, 53–4 educational decisions and transitions, rational choice theory 21 and exposure to treatment 135 and genetics 5, 61–2, 65, 279 and grandparenthood 192–3 and health measures 219 and health outcomes 387 influence of early health adolescence and childhood 295–6 perinatal and childhood health 293–5 magnitude of recent mortality by, in US and Europe 236–7 at meso-level 149–50 and obesity 241–4 origin/education/destination triangle 22–3 paid parental leave’s long-term impact 396–7 and quality of work 172–3, 182–3 regression adjustment 99–101 and selection according to ability and performance 25 primary and secondary 276 selection processes and social causation 217–18 SES as moderator of effect of genes on 5 and smoking 244–6, 297, 298 and social networks 223

408  Handbook of health inequalities across the life course as socioeconomic measure, studies examining health inequalities 313–16, 317 effect heterogeneity 95–6, 100, 102, 104, 180, 205–6, 207, 212 employment see work English Longitudinal Study of Ageing (ELSA) 1, 133, 177, 208, 253–4, 258–9, 262–6, 272 environmental inequality adverse health effects of air pollution 330–335, 341, 343 definition and context 327–8 at different levels of spatial resolution 339–40 legacy effects 339, 341 life course perspective on 328–9 mechanisms underlying 338–9, 341 open questions remaining 342 other dimensions of environmental quality 336–9 residential sorting 332, 333, 334, 335, 336–7, 338–9, 341–2 study conclusions 341–2 epidemiological regimes 234–6, 241, 248 see also life course epidemiology epigenetic ageing context 252–3 future research avenues and policy 271–2 hypotheses 256–7 linear model of 268 methods and materials 257–60 number of countries studied 253–4 predicted 269 results 260–267 smoothed distributions of 267 study discussion and conclusion 267–9 limitations and strengths 270–271 summary features of sample 262 theory 255–6 epigenetic ageing hypothesis explanation 252, 257 results 260, 262, 266–7 testing 259–60, 271–2 epigenetic clock 65, 66, 68, 69 epigenetics direct pathways obesity and metabolic disorders risk in fetus 247 originating in early conditions 238–9 enrichment of life course studies on health by 63–6 rapid development of 59 research on, and future research avenues 68–9

studies highlighting role of strains and resources across life course 66 and traumatic experiences 67 episodic memory 256, 259, 260, 262–3, 264, 265, 272 events and accumulative advantage/disadvantage 67–8 distinction with states 67 life course as series of 17, 25 methods for analyzing occurrence of 75–83 sequencing and timing 120 states as always implying 18 stressful 67–8, 151, 162–3, 192, 203, 225, 294 extreme gradient boosting (XGBoost) 116, 118–19, 121–3 extreme heat 337 factor analysis 116–17 family and adolescence 146–7, 148, 153, 160–169 within-family variation 333 financial hardship of 254, 268 family affluence scale (FAS) 163–9 family income 237, 279, 296, 370, 394 family relationships and health inequalities 188–96 as key social determinant of health 220 and later life cognition 280 and later life health 221–2 gender differences 222–3, 225 socioeconomic group differences 223–4 summary 225 linked lives 26 and social capital 162–3 family SEP in early adolescence, and depressive symptoms in adulthood 153 and social capital 160, 162–9 FAS see family affluence scale (FAS) fetal origins hypothesis 293, 328, 335, 372 fixed effects (FE) panel estimator 106, 108, 133 fixed effects models 84–6, 87 fundamental cause theory (FCT) 281–2 gender in counterfactual model of causality 100 grandparenthood 192, 193 inclusion in UNICEF’s principles 392 and later life health 222–3, 225 and measurement of SES at older ages 218 and US welfare state 372, 376, 378, 379, 380 gene–environment interaction 62, 66–7 genetic confounding 283–4, 285, 293, 337

Index  409 genetic variation and heritability 62–3 influence on health 60–61 research on 68 genetics in baseline bias example 96 causal effect heterogeneity bias example 96, 100 childhood and mortality later in life 283–4, 285 as focus of life course epidemiology 278 genes relevance of 59–60 role in childhood exposures–late-life cognition associations 279–80 types of gene–environment interaction 62, 66–7 genetic predisposition 62 and health enrichment of life course studies by epigenetic information 63–6 future research avenues 68–9 gene-health studies through lens of life course 66–8 heritability estimates 62–3 heritability of health-related traits and social class 235–6 pathways of influence 60–62 relevance 59–60 research on, and future research avenues 68–9 infectious disease 351 and instrumental variables 131, 133 pathways originating in early conditions 238–9 see also epigenetics global metabolic inequalities 38–41 Gossen’s Law 20 grandchildren future research avenues 196 health inequalities from perspective of 195–6 grandparents future research avenues 196 grandparenthood 191–3 grandparenting 193–5 health inequalities from perspective of 190–195 seen as increasingly central to families 188 green spaces 327, 329, 336–7, 338–9, 340–341 growing up heterogeneous opportunities and conditions of 146 increasing effect of social capital on health in process of 167 in low SES conditions 391

and poor health at birth 295 in poverty 262 social determinants framework for testing 255 and social inequality 161 see also adolescence growth models 88–9 hazard models 76–80 health across life course development of research on 1, 2 environmental inequalities and 329 machine learning algorithms to predict self-rated 120 overview 224–5 social determinants predicting 396 social gradient in 218–19 social networks and family 219–24 and socioeconomic resources 217–19 socioeconomic status associated with 4, 373 US welfare state policies impacting 377 consequences of poor quality work and employment 176–8 as input into utility, leading to trade-offs 49–50 and intergenerational relationships 189–96 links with adverse childhood exposures 390–391 pathways to 219–20 requirements of examining determinants of 132 scoping review of IV studies on 137–8, 139, 140 and social stratification 23, 321 in utero 293–4 value over life cycle 51 Health and Retirement Study (HRS) 1, 83, 133, 134, 177, 209, 242–4, 248–9, 253, 258–9, 260, 262–9 health capital theory assumption versus evidence 47 conceptual insights 49–52 explanation 48–9 further extensions 52–4 and parental time 394 health cost 50 health effects of air pollution 330–335, 343 early future research avenues 301 and social and institutional environments 298–300 of environmental quality 327, 338, 341, 342

410  Handbook of health inequalities across the life course of grandparent role 191–5 of human capital development 341–2 of intergenerational relations 189–90 of obesity 194 of retirement on cognitive abilities 205–6 introduction to 203–4 on mental health 206–7 natural experiments and identification issues 210–212 overview 212 on physical health and mortality 207–10 health index 163, 164, 165–8 health inequality/inequalities across life course biology 5–6, 308–9 causality 4 complexity 3 decreasing 308 economic theories 46–54 increasing 310 perspective approaches 1 policy 6 processes contributing to development of research on 1–2 thematic structure and summary 6–11 applicability of machine learning 112, 119–20, 121 and environmental inequalities 328–9 evolution over age context 307–8 inequalities in mortality 318–19 public health and socioeconomic inequality 321–2 relative and absolute inequalities 307, 310–312, 318–19, 321–2 review of previous studies 312–17 study conclusions 320–322 theories 308–10 from grandchildren’s perspective 195–6 from grandparents’ perspective 190–195 instrumental variables estimation 134–5 future potential for sociological analyses 140 journals publishing studies on 139 scoping review of studies on 137–8, 139, 140 time trend of publications 138 treatment effect and policy effect 139 intergenerational transmission of 235, 246–7, 301 and life course inequalities 23–7 in parent–child relationship 189–90

requirements of examining determinants of 132 and social inequality conceptualizations 215–16 in process of growing up 162 and SPPs links between 386–8 role in reducing 384–98 value over life cycle 51 health insurance 26, 240, 349, 370, 373–4, 385, 386 health measures in childhood and adolescence 295–6 differing depending on welfare regime 389–90 general 207–8, 301 objective, widening educational inequalities in 219 health policy see policy health selection hypothesis 217 health trajectories see trajectories: health heritability associated with obesity and metabolic disorders 246–7 of general cognitive ability 279 of health outcomes 62–3 of health-related traits and social class 235–6 of longevity 283–4 and smoking 247 varying by age 279 human capital application to life course 21 and critical and sensitive periods models 297 and cumulative disadvantage 329 effects of retirement 205 and environmental quality 341–2 focus on individuals’ abilities 162 identification causal 97–8, 99, 105–9, 211, 327, 331–2, 334, 338, 341 environmental quality 338 of genetic influences 59–60 of health effects of retirement 204, 210–212 health impacts of air pollution 331–2, 334, 341 in utero developmental origin of health and disease 253 direct pathway related to obesity and T2D 238–9 epigenetic changes 64 exposures as critical 341 environmental conditions 330, 333, 335

Index  411 fallout from Chernobyl disaster 297 Tarapaca earthquake 299 health 293–4 insults to health experiences 328 maternal diets 35 smoking effects 247, 298 independence and dementia 277, 278 life course 26–7 of older adults 221 of young adults 151 independence assumption 84–5, 95, 98 “independent risk” approach 67 infant health and air pollution 334, 342 breastfeeding for 395 exposure to pathogens 359 formula-fed infants 39 free infant health care 294, 299 in instrumental variable example 131 in intergenerational loop 293–4 obesity and T2D 237, 239, 247 poor, maternal SES and poverty associated with 391 US nutrition programme for 370, 374, 379 infant mortality and air pollution 335 early Medicaid expansions reducing 373 in instrumental variable example 132–3 and paid maternity leave 395 per thousand births by class and absolute and relative differences 311, 312, 321–2 rates differing depending on welfare regime 389–90 unequal allocation of early conditions 234–5 infections 334, 349, 351, 352, 358, 359, 360–361 infectious disease accumulation 359–60, 361 conquest of, becoming new vehicle for social class inequalities 233 context 349–50 critical and sensitive periods 357–9 epidemiological triad of agent, host and environment 350–352 health inequalities models of 353, 355–6 need for increased longitudinal research on 361 socioeconomic health inequalities 352–4 and instrumental variables 131 lag between exposure and 359 life course epidemiology application of principles to aetiology and inequalities 356–8 basic principles 352–6

longitudinal versus cross-sectional studies 361 and socioeconomic inequalities 352–4 unequal allocation of 234 institutional contexts 26, 150, 153 institutional environment, and early health effects 298–300, 301 instrumental variables (IV) estimator 103–4 instrumental variables (IVs) age-based eligibility thresholds 211–12 approaches of critical and sensitive periods early in life 132–3 used at adult ages 133–4 assumptions of approach 129–31, 140–141 context 128–9 estimating effects of O3 on respiratory-related hospitalizations 334 estimation 134–5 literature review of studies on health and health inequalities 137–8, 139, 140 strengths and weaknesses 138–40 types and examples 131–2 used to estimate complier-specific treatment effects 135–7 interdisciplinarity avenues for 68, 141, 379 in life course framework of early health and socioeconomic attainment 296–300 in relation to causality 140, 141 intergenerational relationships adult parent–child 189–90 buffering stressors and minimizing health risks 224 context 188–9 future research avenues 196 grandparent–grandchild 190–196 sandwich generation 197 intergenerational spillover effects 390–391 intergenerational transmission of health inequalities 235, 246–7, 301 inverse probability weighting (IPW) as cross-sectional method dealing with selection of observables 102 for survey selective drop-out 121–3 IVs see instrumental variables (IVs) LASSO 115–16, 119, 120, 122–3 latency 244, 327, 328, 333, 341, 361 latent class analysis (LCA) 88, 90, 116 latent class models 88–90 latent transmission see chains of risk later life health

412  Handbook of health inequalities across the life course and childhood socioeconomic conditions 275–81 and cumulative disadvantage 329 and differences in socioeconomic status 223–4 and family relationships 221–2 and gender 222–3, 225 human capital effects 329 importance of social networks 216, 219–20, 222, 225 paid parental leave (PPL) policies for 393 possibility of sensitive periods occurring in 355 health inequalities age-as-leveller hypothesis 308 concepts 215–16 and health care policies 309 overview 224–5 previous studies 312–17 reasons for focus on 320 social networks, family and health across life course 219–24 socioeconomic, in adolescence, long-term impact 150–153 socioeconomic resources 217–19 interventions and outcomes 375–6 mortality and childhood, evidence from Sweden 284–5 and childhood socioeconomic conditions 281–2 framework for understanding impact of socioeconomic conditions in childhood on 282–4 study conclusions 285–7 welfare programs 321, 375–6 welfare state 370–371, 375, 380 see also grandparents; health in later life; retirement life course and adolescence 17, 24, 146, 147, 151–2, 160–161 applicability of machine learning 112, 113, 115, 116–18, 119–20, 121, 124 concepts as sequence of states 18 as series of events and transitions 17 as series of life phases or age roles 17 as trajectories 18 development of cognitive function, representation of 278 as expression and mechanism of social inequalities 22–3

framework of early health and socioeconomic attainment 296–300 genes, genetics, epigenetics childhood poverty and epigenetic ageing 254–5 future research avenues 68–9 gene-health studies through lens of 66–8 and genetic predispositions 62 interplay of genetic variation with social environment over 61 lessons from epigenetic mechanisms 64–5 and social origin 63 studies on health, enrichment by epigenetic information 63–6 heterogeneity in cognitive development and decline throughout 276–7 history of sociology of 15–16 infectious diseases across basic principles of epidemiology and their application 354–60 causation 350–352 context 349–50 longitudinal versus cross-sectional studies 361 and socioeconomic inequalities 352–4 interest in accumulation of experiences over 59 manifestation of health trajectories through 151–2 paradigms and theories 19–21 perspective centrality of work in 172 on environmental and health inequalities 328–9 immune system 358 intragenerational mobility 65 on social inequality and health in adolescence 161 on SPPs 384–5, 390–392 three assumptions of 19 as well-integrated in sociology 3 research challenge of complexity 3 childhood holding special place in 275 contributions to development of 1–2 instrumental variables 138, 140 and nutrition transition 33 policy perspective 6 possibilities and limitations for future of see machine learning (ML) transitions to adulthood 22 as yet unanswered causal question 4 SEP seen as predictor for developing comorbidities through 153

Index  413 social gradients in health across 218–19 and socialisation phases 147 stressors over 67–8 as unfolding “exposure to risk” 23 in United States elective nature of policy space 368 future research avenues 378, 379 gaining access to Medicare 375 interventions and outcomes 376–7, 380 welfare state across 371–2 see also health: across life course; health inequalities: across life course life course cube 19–20, 26 life course epidemiology basic principles 354–6 application to infectious disease aetiology and inequalities 356–60 and causality 322 epidemiological triad of agent, host and environment 350–352, 358 framework for understanding impact of childhood socioeconomic conditions on mortality later in life 282–4, 285 life phases with salience for unequal health outcomes 24 models benefits and limitations 276, 285–6 examining childhood factors, cognition and dementia 278–9 for understanding early health and socioeconomic attainment 297–8 social epidemiology as 3 life course etiology 24 life course inequalities and health inequalities 23–7 methods used in studies of for analyzing occurrence of events 75–83 for analyzing repeated measures 83–90 future research avenues 91 life course observatory 26 “life course paradigm” 19 life cycle theory of reproduction and longevity 20 of savings and consumption 21 life phases and duration of states 67 and health inequality 24 history of concept 16 and investments in human capital 21 life course as series of 17 life table methods 80–83 lifelong learning 182–3 lifelong observations 25 lifetime as metric 24

linear regression 76, 102, 166, 182, 314–15 linked lives 19, 26, 224 local average treatment effects (LATE) 55, 104, 133, 136–7 logistic regression 79, 83, 120, 121–3, 310, 313, 315–16 long arm of childhood conditions conditions hypothesis 256–7 future research and policy 271–2 limitations and strengths 270 mechanism 268–9 methods and materials 257–60 origins and context 252–4 reflection on recent literature 270–271 regaining momentum in emerging adulthood 151 results 260–268 social determinants framework for testing 254–5 theory 255–6 longevity constraint of 49 education and cognition influencing expectations of 54 heritability of 283–4 impact of retirement 209, 211, 213 importance of ability to influence 51–2 life cycle theory of reproduction and 20 longitudinal data era of popularity 17 growing availability of 1 often used in social sciences 77 scarcely available data on health from birth to old age 132 used to study environmental quality 339–40, 342 used to study grandparenthood 191, 192–3 longitudinal methods before–after estimator 105–6 difference-in-differences estimator 106–8 longitudinal design 105 “lumping” approach 67 machine learning (ML) bias versus variance 112–14, 116, 118 black box models 114–15 for causal inference 124 circumstances under which strengths are shown 118–19 cross-validation 114, 117–18 examples of use of drop-out from longitudinal studies 121–3 from research literature 120–121 future research avenues 124

414  Handbook of health inequalities across the life course prediction 112 black box models 114 supervised 115–16, 123 validity of 114 for research on health inequalities and life course 119–20 supervision need for 115 supervised methods 115–16 unsupervised methods 116–17 versus traditional methods 119 training and test data 114 translation of terms to statistics 113 matching 101–2 measurement in biomedical and socioeconomic studies 25 of SES 218, 219 of “states” 18 mediation analysis 179–80, 181, 301 Medicaid 213, 369–70, 371, 373, 374, 375, 376–8, 379 Medicare 309, 333–4, 370–372, 375–6, 377, 379 mental health of adolescents 147–8, 150, 173, 295–6, 299, 301, 336 child 195, 295–6 childhood poverty 265, 271 depression and family SEP 153 effects of retirement on 206–7 feelings of subordination or inferiority 387–8 female caregivers 222 and grandparental childcare 193–4 green spaces 336–7 in health index 163, 164 heritability 62 no single gene accounting for 60 of older parents 189–90 and paid parental leave policies 393 parental, adult children’s higher education affecting 224 policy changes 52 poor working conditions 174 and roles invested with importance 188 SES-related health inequalities 217 social isolation associated with increased risk of poor 220 studies examining 313 of TANF recipients 373, 376 and volunteering 204 methylation associations with childhood poverty and epigenetic age 252, 257, 259, 260, 266–9, 271–2 DNA 256, 259, 268–9, 271–2 and maternal diet 247

N6 methyladenosine (m6A) 256, 257 predicted 268 RNA 256, 269, 272 ML see machine learning (ML) modelling for analyzing occurrence of events 75–80, 82, 83 for analyzing repeated measures 84–90 counterfactual 94–6, 180 economic 46–54 machine learning 112–24 regression adjustment 99–100 moderation analysis 180–181 mortality and ageing 24, 215, 259–60, 266–7 and air pollution 330–331, 335, 341 all-cause 237, 281, 282, 284, 318, 330–331, 391 bias 109 bulk of inequalities attributable to chronic illnesses 237 causal pathways 240 examples of direct and indirect 241–6 cause-specific 237, 282, 284, 391 causes of death responsible for inequalities 236–7 childhood health conditions as risk factor 252 childhood socioeconomic conditions context 275–6 evidence from Sweden 284–5 framework for understanding impact 282–4 and mortality in later life 281–2 study conclusions 285–7 chronic conditions associated with differentials 239 classifications 281–2 cognitive ability preventing premature 61 differences between countries 26 effects of retirement early retirement versus normal retirement age 203–4 heterogeneity of results 212 increasing effects 210 measuring 207, 208–9 reducing effects 209 zero effects 209–10, 213 elevated risks for grandmothers 192–3 and exposure to Medicaid as adults 376 in childhood 373, 375, 379 and family relationships 221 and health inequalities

Index  415 absolute and relative 307, 310–312, 318–19, 321–2 context 233–4 unequal allocation of early conditions and of opportunities of class accession 234–5 higher prevalence among lower socioeconomic groups 46, 217, 224 impact of paid parental leave 394–5 and infectious diseases 352–3, 356 instrumental variables 132 and investments in health 52 magnitude of recent inequalities by education in US and Europe 236 manifestation of pathways 240–241 mapping review 312–17 modelling 77–8, 80–83, 86 other environmental conditions 337, 342 selective 309–10, 320–321 social isolation associated with 220 socioeconomic status and gender differences 218 study on effect of prolongation of working life on 101 and welfare regimes 389–90 work risk factors 176, 179–80 see also infant mortality natural experiments benefits to empirical research 48 effect of income on health 133–4 and emissions-cheating diesel cars 335 examples of 129 and green spaces 336 identifying causal effect of adverse exposures during pregnancy 133–4 identifying health effects of retirement 210–212 and paid parental leave 396–7 and shocks 332–3 theory assisting 48 validity of 133 where source of variation in exposure is clear 47–8 see also quasi-experiments neoliberalism 36, 38–41 noise environmental 327, 337–8, 339, 343 random 113, 114, 118, 119 working conditions 174, 176 nutrition support program 370, 374, 376, 377–8, 379 nutrition transition comparative cohort analysis and social change 33–6

conceptual figure, and timing of impact on sequential birth cohorts 34 economic development and cumulative dis/ advantage on global scale 36–8 global perspective 32–3 neoliberalism and increasing inequality in global metabolic conditions 38–41 summary 41–2 obesity in adolescents 152 and breastfeeding 39, 391 childhood 32, 38, 195 direct pathways 239, 241–4 and early-life nutritional scarcity 35 global rates of 32 and health inequalities 237, 241–4, 246–7 in high-income countries 33 intergenerational transmission 246–7 in middle-income countries 38, 39 in poor and middle-income members of society 40 in socially disadvantaged populations 353 in sub-Saharan Africa 33, 40–41 in urban areas 39, 41 obesogenic foods cheap availability of 40–41 unregulated consumption of processed foods 33 occupational careers theory 21 old age/older age see health in later life; later life paid parental leave (PPL) definition and current policy landscape 393–4 influence on children’s health outcomes 394 long-term impact 396–7 short-term impact 394–6 as type of SPP 392 parent–child relationship 188, 189–90, 192, 196, 270, 294, 360 parental leave see paid parental leave (PPL) pathways behavioural 149, 216, 388–9, 390, 392, 398 biological 297, 299, 328, 331, 342 direct and intergenerational transmission 246–7 obesity and T2D 241–4 originating in early conditions 238–9 early retirement 203, 206 to health 219–20 health and mortality inequalities benefits and limitations of study 247–8 in causal model 237

416  Handbook of health inequalities across the life course indirect, originating in early conditions 239–40 intergenerational transmission via direct and indirect 246–7 manifestation of 240–241 obesity, T2D and smoking 241–6 originating in early conditions 238–40 other causal 240 powered by unique set of 233–4 indirect and intergenerational transmission 247 originating in early conditions 239–40 smoking 244–6 materialist 149, 216, 388, 390, 392, 398 nutrition 374 pathway model 255, 355 psychosocial 149, 216, 389, 390, 392, 398 by which genes influence health 60–62 perinatal health 293–4, 299–300 physical health of adolescents 151, 295–6 in childhood 295–6 effects of retirement on 133, 207–8 extra time spent with children possibly improving 394 feelings of subordination or inferiority 387–8 and grandparental childcare 193–4 green spaces 336 in health index 163, 164 nutrient-poor diets 32 and obesity 32, 152 and paid parental leave policies 393 parental, adult children’s higher education affecting 224 poor working conditions 174 SES-related health inequalities 217 social isolation associated with increased risk of poor 220 studies examining 313, 314, 315 of TANF recipients 373 policy health inequalities across life course 6 implications of social inequalities in relation to work 181–3 paid parental leave 393–7 Social Protection 385–92 in United States elective nature of 368, 371, 375, 379 as fragmented 368 future research avenues 378–9 limitations 377–8, 380 welfare programs 372–7, 380 welfare state 368–72, 379 policy change and air pollution 335

for identifying longevity effects of retirement 211 interpreting 52 IV approach 129 and natural experiments 332, 333 population health infectious disease shaping global 349 literature in 234 in low- to middle-income countries 32 past research 235 sociology’s acknowledgement of 2 SPPs affecting 389–90 United States welfare programs intended to improve 372–7, 379 welfare state failing to support 379 potential outcome (PO) approach 94–6, 98, 100, 102, 104, 105, 109, 135 see also counterfactual model prediction in relation to causal inference 93, 96, 99, 108 in relation to machine learning 112 black box models 114 supervised 115–16, 123 validity of 114 principal component analysis (PCA) 116–17 propensity score matching (PSM) 101, 129, 133 public health cognitive function and mortality as concerns of 275 diseases with high level of relevance to 178 fight against infectious diseases as issue of 361 health inequalities as problem of 397 obesity as one of most severe issues of 152 and socioeconomic inequality 321–2 public health insurance programs 370, 373–4 quality of work influences of 172 poor, health consequences of 176–8 social inequalities in 172–5 policy implications 181–3 quasi-experiments 46–7, 129, 332–3, 335, 340, 341 see also natural experiments random effects models 86–8 random forests 116, 118–19, 120, 123 regression 99–100 see also linear regression; logistic regression regression adjustment 47, 99–100, 102, 129, 212 regression discontinuity design (RDD) 104, 109, 118, 129, 210, 211–12 relative inequalities 307, 310–312, 318–19, 321–2

Index  417 repeated measures methods for analyzing 83–90 need to account for time-dependent confounding 286 reserve 277, 278–9, 286, 287 residential sorting 332, 333, 334, 335, 336–7, 338–9, 341–2 resilience 23, 286, 287 retirement context 203–4 effects on cognitive abilities 205–6 on mental health 206–7 on physical health and mortality 207–10 natural experiments and identification issues 210–212 risk see chains of risk role strain theory 193–4 sarcopenia (probable) 259, 260, 262–3, 264–6, 272 Scarr–Rowe hypothesis 66, 279 selection health class and social status as possible outcomes of 25 hypothesis of 217–18 importance in relation to health inequalities 9 indirect 240 models 297–8 and unequal allocation of early resources 235 health impairments resulting in lower social positions 160 mortality 9, 51, 309–10 on observables 98–102 primary and secondary 276 and selective optimization with compensation 20 self-selection 68 stratification as process of 25 on unobservables 103–5, 331–2 selection processes 59, 217–18, 221–2 selective mortality 309–10, 320–321 selective optimization with compensation (SOC) theory 20 sensitive periods changes in epigenome likely emerging in 63 childhood as 282 future research avenues 287 impact of chronic stress during 173 impact of early-life factors on cognition and dementia 278–9, 280–281, 285 possibility of occurring in later life 355

research focus on cardiovascular health 275 see also critical periods: and sensitive periods SEP see socioeconomic position/status (SEP/ SES) sequence analysis 18, 26, 117 sequences examples of longitudinal 16 identifying heterogeneity in life course 120 life course as 18 SES see socioeconomic position/status (SEP/ SES) smoking adolescence carried over into adulthood 151 cumulative approach 152 link to parental SEP 152 risk of 148 behavioural pathway 149, 216, 388 genetic route 249 and health inequalities 237, 244–5, 247, 249 and higher education 283 indirect pathways 244–6 intergenerational transmission 247 and IV approach assumptions 129–30 maternal 239, 246, 247, 294, 298, 302 and rationality 50 and social control 66–7 in socially disadvantaged populations 353 as trigger of epigenetic changes 64 social capital concept 162, 219 effects on health index 166 findings 165–9 hypotheses, data and measures 162–5 importance in health development of young people 169 “local” 339 as major resource for healthy lifestyles and recovery from diseases 67 as mediator 160, 163 predicted values for health index of interactions between FAS and 167 of students with different ages and varying 168 of students with low versus high values of FAS and 166 social capital index 165 theoretical frameworks 160–162 social causation 4, 217–18 social causation hypothesis 217 social change 33–6 social classes in causal model for health inequalities 237–41

418  Handbook of health inequalities across the life course childhood and mortality later in life 284–5 and cognitive performance 280–281 definition 22 gaping inequalities in health status and length of life 233 and heritability of health-related traits 235–6 infant mortality rates 311, 312 obesity and metabolic disorders 242, 246–7 perinatal and childhood health 293–4 and smoking 244–5, 247 unequal opportunities of class accession 234–5 social control 66–7, 151, 220 social determinants of health family relationships as 220 framework for testing long arm of childhood conditions shaping later life health 253, 255, 257, 262, 268, 270 longitudinal perspectives important for 59 macro-level structural determinants shaping 388 PPL policies influencing 393, 396–7 primary 387 in relation to infectious diseases 349, 350–352, 356–60 and social protection policies 384 social environment and early health effects 298–300, 301 individuals’ self-selection into, based on IQ 68 and natural environment 351–2 as of particular importance to adolescents 146, 153 and probability of exposure 359 and socio-environmental approach 150 types of gene–environment interaction 66–7 social gradients in certain diseases 176, 179, 181 in health 181, 215, 217, 218–19, 225, 387, 398 in occupational attainment 173 of poor working conditions 174–5 SPPs reducing 392, 398 social inequality/inequalities in access to work and employment 172–5 policy implications 181–3 and environmental inequalities 327, 331, 337, 338, 342 epigenetic changes closely linked to 63–5, 68 and health inequality conceptualizations 215–16 in process of growing up 162 and infectious diseases 349, 352, 353, 355, 359, 361

life course as expression and mechanism of 22–3 link with low birth rate 293–4 in relation to health as ambiguous concept 23 causation and selection 160 important dimension 2 role of work in explaining 179–81 and social capital 162 temporal dynamics 22 social mobility health burdens in life course having influence on 356 and health selection hypothesis 217–18 intergenerational 22, 312 in relation to machine learning 119–20 selective 284 in stratification regimes 247 theory of 151–2 social networks definition 219–20 family 188, 220 higher educated older adults having more diverse 223 importance of, in later life 216, 219–20, 222, 225 linked lives 19 methodological issues 221–2 as part of life course 19 and pathways to health 219–20 and young people 169 social origin 8, 22–3, 63, 68, 99–100, 161, 279 social policy see policy Social Protection Policies (SPPs) components 385–6 context 384–5 definition and purpose 385 in life course perspective child-sensitive SPPs 392 links between adverse childhood exposures and health 390–391 linking to health inequalities 386–8 paid parental leave as type of 392–7 summary and recommendation 397–8 social security systems 64, 204, 211, 212, 218, 219, 387, 393–4 Social Security (US) 370–371, 375 social stratification changing impact of processes of 309 dimensions of 24 field of sociology of 27 and health 23, 321 and stratification orders 22 socialisation theory 146–7, 153 socioeconomic attainment

Index  419 and adolescent health 152, 295–6 and early health effects, and social and institutional environments 298–300 future research avenues 301–2 life course epidemiological models for understanding 297–8 life course framework 296–300 overview of findings 292–6 strong associations between 292 perinatal and childhood health 293–5 socioeconomic conditions 147, 307, 308, 320–321 see also childhood socioeconomic conditions socioeconomic differentials adverse working conditions 180–181 and air pollution 331 approaches to analysis 134–5 effects of retirement 203, 206–7, 208, 212 historic pandemics 352 life time as metric 24 in older age 223–4 as quite recent phenomenon 233 socioeconomic gradient of obesity as much steeper in urban areas 39 in southern Africa 41 as steepest in countries with moderate GDPs 38 in transition from good to poor health, and from poor health to death 321 socioeconomic health inequalities in adolescence, long-term impact on later life 150–153 and infectious disease 352–4 in older ages 308, 320 and public health 321–2 socioeconomic outcomes see socioeconomic attainment socioeconomic position/status (SEP/SES) of adolescents 148–53 and age-as-leveller hypothesis 218, 307, 308–9, 321 association with health across life course 4, 218–19, 373 air-pollution-induced conditions 331 at birth 294 causal effects 48, 53 children 280–281, 391 cost of 50 differences between socioeconomic groups in later life 223–4 former affecting latter 160, 292 immunization take-up 395–6 inequalities in, and SPPs 386–90, 392

infants 299–300, 391 inverse relationship 223 longevity 51–2 role of work 179–81, 183 selection processes and social causation 217–18 smoking 50 value placed on 51 childcare costs 397 childhood, and old age self-rated health 253 as closely linked to different levels of epigenetic imprint 64 cognitive levels and changes, associations with 280 components complemented by genetic origin 68 and cumulative disadvantage 328–9 dementia risk, associations with 280–281 example measures 86, 352 exposure to stress and capacity to draw upon social relationships as coping resource shaped by 222 family effect on adolescent health 160, 161, 162–4, 165–6, 169 effect on children’s development 173 genetic influences 5, 53, 279 importance of measurement 219 intergenerational persistence in 302 neighbourhood-level, and environmental quality 339 and overweight/obesity in Sub-Saharan African (SSA) countries 41 and rational decision-making 50, 54 understanding of multiple dimensions of 2 socioeconomic resources selection processes and social causation 217–18 social gradient in health across life course 218–19 sociology of life course history 15–16 influential program for 19 prominent topics in 24–7 synthesis with other sociologies 27 recent interest in health research agenda 2 in relation to causality 4 “splitting” approach 67 SPPs see Social Protection Policies (SPPs) stability 20, 22, 26–7 states in counterfactual model of causality 94–6 distinction with events 67 life course as sequence of 18, 117

420  Handbook of health inequalities across the life course methods for analyzing 26, 82–3 in multistate life tables 82–3 statistics, traditional 113, 119–20, 124 stress-buffering hypothesis 220 stress-diathesis 66 stressors during adolescence 162–3 familial and social ties buffering 220, 224 over life course 67–8 resilience against 286 stress responses to 388 at work 193–4, 389 supervised machine learning (SML) and life course research 113, 117 methods 115–16, 120–121 need for 115 Supplemental Nutrition Program for Women, Infants, and Children (WIC) 370, 374, 376, 377–8, 379 Survey of Health, Ageing and Retirement in Europe (SHARE) 1, 25, 133, 175, 191, 206, 253, 254, 258–9, 262–6, 271, 272 T2D see type II diabetes (T2D) Temporary Assistance to Needy Families (TANF) 369, 370, 371, 373, 374, 376, 378 theory of cumulative advantage 151, 152 theory of primary and secondary control 20–21 toxicological evidence 330–331 trajectories cognitive change 280 early life inequalities establishing divergent 376 education 240 of epigenetic aging 65 health adolescent 151–2, 153 adversity during early years affecting 238 critical periods of development as formative for long-term 372 early developmental experiences and exposures affecting 63 epigenetic changes providing explanations for 64 exposure to war 254, 272 and infectious diseases 355–6 knowledge of support as crucial to 219 and traumatic experiences 67 and US welfare state 374, 375, 379 life course gender differences in 218 pathway model 355 sequences analysis for 117 as trajectories 18

“resilience” and “skidding” 23 and social inequalities 22 of socioeconomic conditions from childhood 253 transitions to adulthood adolescent phase 146–7, 151–2 research in 22 educational 21, 297 from good to poor health 321 to grandparenthood 190–193 life course as series of 17 and timing 19 in life tables 82–3 to old age, and social causation 217 to retirement 203 social-biological 5 see also nutrition transition traumatic experiences 61, 64, 67 treatment effects doubts about external validity of local 47–8 IVs for estimating complier-specific 135–7 and matching 102 nesting marginal estimation into event study design 205–6 of raising minimum school-leaving age 52 see also local average treatment effects (LATE) triggering 66 turning points 17, 23, 25–6 type II diabetes (T2D) critical and sensitive periods 391 direct pathways 239, 242–4 and education mortality differentials 236 intergenerational spillover effects 391 and work environment 178 United States health inequalities elective nature of policy space 368, 379 fragmentation produced via federalism 368 policies contributing to or alleviating 372–7, 380 welfare programs childhood interventions and outcomes 372–4 later life interventions and outcomes 375–6 life course interventions and outcomes 376–7 welfare state across life course 371–2 childhood 369–70, 379

Index  421 comparison with Sweden 219 developments 369 as fragmented 379 later life 370–371, 380 limitations 372, 377–8 see also policy: United States unsupervised machine learning (UML) 113, 115, 116–17, 120 “use-it-or-lose-it” hypothesis 203, 205, 206, 212 variance versus bias 112–14, 116, 118 in cognitive abilities 277 in “hierarchical model” 87 meaning of term used in life course research 113 in multi-level regression model 165, 166 standard analysis in life course research having low 117 wealth of countries, double standard example 37 and difference between absolute and marginal utility 50–51 enabling people to limit exposure to disease risk factors 387 and health examples of other factors influencing 53 impact studies 133–4, 262–3, 313–17 health costs increasing with 50 life cycle theory of 21 obesity as symbol of 40 welfare programs (US) childhood interventions and outcomes 372–4 later life interventions and outcomes 375–6 life course interventions and outcomes 376–7 welfare state definition 369 welfare state policies 309 welfare state (US) across life course 371–2 childhood 369–70, 379

comparison with Sweden 219 developments 369 as fragmented and elective 368–9, 379 future research avenues 378–9 later life 370–371, 375, 380 limitations 372, 377–8 well-being adolescent 147, 148, 150 of children adult, and parents 189–90 paid parental leave influencing 394 role of grandparents 195–6 and “decommodification” 385 economic approach 46 entering grandparenthood negative effects on 192 positive effects on 191–2 and environmental noise exposure 337 impact of grandchild care provision on 193–4 in later life crucial factors for 219 maintenance of cognitive and physical function for 285 support from children affecting 221 social capital improving 162 social integration as conducive to 188 US addressing 369, 370, 372, 373, 374, 375 WIC see Supplemental Nutrition Program for Women, Infants, and Children (WIC) work centrality in life course perspective 172 health consequences of poor quality of work and employment 176–8 as playing crucial role in adult life 172 role in explaining social inequalities in health 179–81 social inequalities in access to, and quality of work and employment 172–5 policy implications 181–3 work stress 175, 181, 182–3