Pandemic and the City 3031219821, 9783031219825

Citation preview

Footprints of Regional Science The Voice of Regional Science

Mehmet Güney Celbiş Karima Kourtit Peter Nijkamp Editors

Pandemic and the City

Footprints of Regional Science

The Voice of Regional Science Series Editors Bruce Newbold, School of Geography and Earth Sciences, McMaster University, Hamilton, ON, Canada Vicente Royuela , University of Barcelona, Barcelona, Spain Mia Wahlström, Tyréns, Gustavsberg, Sweden

The field of regional science continues to advance. The book series The Voice of Regional Science presents new scholarly thinking on important emerging issues and new scientific advances in the broad and multidisciplinary domain of regional science research. The contributions to the series invoke the aim of the Regional Science Academy to rethink “the spatial dynamics of people and socio-economic activities in the connected and complex spatial systems of our earth.” The volumes in this series will serve the global regional science community by providing new insights on, or by calling attention to, recent developments that are rarely covered by conventional conferences, or are indicative of novel research directions. The Voice of Regional Science welcomes both edited volumes and monographs, with works originating in the meetings of the Academy or proposed by external authors and volume editors. It features scholarly works that are retrospective only to the extent necessary to lay the groundwork for visions of future research in the area; all volumes have a forward-looking focus. Further, the series’ emphasis on thematic publications is intended to promote a more future-oriented strategic focus for regional science research. Applied studies in regional science are also welcome, provided they reflect this forward-looking focus. The two sub-series The Voice of Regional Science and Great Minds in Regional Science jointly form a series entitled Footprints in Regional Science. All three book series present work stemming from and related to the activities of The Regional Science Academy (www.regionalscienceacademy.org).

Mehmet Güney Celbi¸s · Karima Kourtit · Peter Nijkamp Editors

Pandemic and the City

Editors Mehmet Güney Celbi¸s Department of Economics Yeditepe University Istanbul, Turkey

Karima Kourtit Programme Smart Cities and Data Analytics Open University Heerlen, The Netherlands

Peter Nijkamp Programme Smart Cities and Data Analytics Open University Heerlen, The Netherlands

ISSN 2662-9623 ISSN 2662-9631 (electronic) Footprints of Regional Science ISSN 2662-9704 ISSN 2662-9712 (electronic) The Voice of Regional Science ISBN 978-3-031-21982-5 ISBN 978-3-031-21983-2 (eBook) https://doi.org/10.1007/978-3-031-21983-2 © The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 This work is subject to copyright. All rights are solely and exclusively licensed by the Publisher, whether the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed. The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use. The publisher, the authors, and the editors are safe to assume that the advice and information in this book are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or the editors give a warranty, expressed or implied, with respect to the material contained herein or for any errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. This Springer imprint is published by the registered company Springer Nature Switzerland AG The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland

Contents

The Pandemic and the City . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Mehmet Güney Celbi¸s, Karima Kourtit, and Peter Nijkamp

1

Cities, Cooperation, and Resilience in the Face of COVID-19 The Resilience of Cities to COVID-19: A Literature Review and Application to Dutch Cities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Jeroen van Haaren and Frank van Oort

11

A Subjective Geographer’s Experience of Pandemic and Confidence in Systems of Cities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Denise Pumain

33

City and Regional Demand for Vaccines Whose Supply Arises from Competition in a Bertrand Duopoly . . . . . . . . . . . . . . . . . . . . . . . . . . . . Amitrajeet A. Batabyal and Hamid Beladi

45

Post Pandemic Cities—Competing for Size or Cooperating for Interaction an Analysis of the Evolution of Portuguese Municipalities Based on an Organic and Rational Spatial Interaction Growth Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Tomaz Ponce Dentinho

61

Comparative Approaches on the Patterns and Effects of City and Location-Specific Policies and Socioeconomic Structures During COVID-19 The Social Digital Twin for Liveable Cities: A COVID-19 Case Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Corentin Kuster, Sanne Hettinga, Tim van Vliet, Henk Scholten, and Paul Padding

77

v

vi

Contents

The Impact of Differing COVID-19 Mitigation Policies: Three Natural Experiments Using Difference-in-Difference Modelling . . . . . . . . Kingsley E. Haynes, Rajendra Kulkarni, Abu Siddique, and Meng-Hao Li

97

On the Association Between Income Inequality and COVID Spread: A View into Spanish Functional Urban Areas . . . . . . . . . . . . . . . . 127 David Castells-Quintana and Vicente Royuela Urbanization Impact Arising from the Behavioral Shift of Citizens and Consumers in a Post-pandemic World . . . . . . . . . . . . . . . . . . . . . . . . . . . 139 Tannistha Maiti, Anwita Maiti, Biswajit Maiti, and Tarry Singh The Socioeconomic and Labor Market Effects of Pandemics on Cities and Local Economies What Happened After SARS in 2003? The Economic Impacts of a Pandemic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163 Ilan Noy, Yasuyuki Sawada, Nguyen Doan, and Canh Phuc Nguyen Industrial Composition, Remote Working and Mobility Changes in Canada and the US During the COVID-19 Pandemic: A SHAP Value Analysis of XGBoost Predictions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189 Mehmet Güney Celbi¸s, Cem Özgüzel, Karima Kourtit, and Peter Nijkamp The Need for New Types of Data and Applications, and Existing Challenges in Analysing the Effects of COVID-19 on Cities Problems with Recording the Spread of COVID-19 in Developing Countries: Evidence from a Phone Survey in Indonesia . . . . . . . . . . . . . . . 211 Budy P. Resosudarmo, Rus’an Nasrudin, Pyan A. Muchtar, Usep Nugraha, and Anna Falentina Pandemic Regional Recovery Index: An Adaptable Tool for Decision-Making on Regions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237 J. Irving, K. Waters, T. Clower, and W. Rifkin The Geography of Daily Urban Spatial Mobility During COVID: The Example of Stockholm in 2020 and 2021 . . . . . . . . . . . . . . . . . . . . . . . . . 261 Ian Shuttleworth, Marina Toger, Umut Türk, and John Östh Social Justice, Digitalization, and Health and Well-Being in the Pandemic City . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 279 Laurie A. Schintler and Connie L. McNeely

The Pandemic and the City Mehmet Güney Celbi¸s, Karima Kourtit, and Peter Nijkamp

1 The Pandemic Experience of the City The recent COVID-19 experience has left deep traces in the urban fabric and has prompted a re-orientation regarding the future of cities. Is an urban overhaul on the way? And will digital technology erode conventional urban agglomeration advantages? The socioeconomic impacts of the pandemic will definitely leave their mark on cities. For example, the digital divide has already become more apparent regarding equitable access to education or to virtual libraries. Furthermore, climate change will have a stronger influence on cities than in the past, given that avoiding close indoor activities or reducing dense concentrations of people may become commonplace. For example, operating restaurants with outside dining areas or organizing open-air activities which allow safe social distancing will face new challenges in cities, in particular in places where weather conditions may become less favorable. The rise of telecommuting (i.e. remote working)—as a consequence of digital technology—is a heavily debated new topic. It is argued that if the individuals who work from home move away from urban areas, a lot of office space might be converted into residences. However, the departure of the relatively rich who can work from home M. G. Celbi¸s (B) Department of Economics, Yeditepe University, Istanbul 34755, Turkey e-mail: [email protected] UNU-MERIT, 6211 Maastricht, Netherlands K. Kourtit · P. Nijkamp Faculty of Management, Open University, 6419 AT Heerlen, Netherlands e-mail: [email protected] P. Nijkamp e-mail: [email protected] K. Kourtit Centre for European Studies, Alexandru Ioan Cuza University, 700506 Ia¸si, Romania © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. G. Celbi¸s et al. (eds.), Pandemic and the City, Footprints of Regional Science, https://doi.org/10.1007/978-3-031-21983-2_1

1

2

M. G. Celbi¸s et al.

in suburban or non-urban areas will rebound as a considerable loss of tax income for city governments. Their exodus, on the other hand, may lead to novel opportunities for businesses to disperse to new clusters. This being said, the COVID-19 pandemic has presented new opportunities for cities to become more inclusive and sustainable and to invest in essential health infrastructure and public services that may allow them to emerge ultimately stronger from the pandemic. On the other hand, reduced dynamism in urban areas may render cities less appealing to tourists or visitors. The empirical evidence on possible new urban trajectories is still fragmented and at times confusing. Human health has become an increasingly important concern in modern cities, as is evidenced by the current corona pandemic. COVID-19 has shaken some of the foundations of cities, such as density and proximity. Cities are challenged to maintain a high level of well-being or to develop into healthy places, without health risks for the inhabitants. The present volume contains a collection of novel and original contributions to the study of urban sustainability, mainly from a human health perspective. Questions addressed are inter alia: Will cities as we know them survive the pandemic or are we heading towards “doughnut cities”? How will the habits and expectations of the post-Covid generation shape the new city? How will the changes to the urban landscape affect the resilience of cities to other crises (e.g. natural disasters)? Will social distancing and isolation triumph over spatial sharing and togetherness? The future of cities is clearly at stake. This book aims to encompass new analytical contributions, to provide informed discussions, to present empirical findings, and to offer plausible predictions so as to assist policymaking alongside with opening new windows for research regarding the pandemics and the city. It contains and presents a rich set of original and challenging contributions on the implications of and responses to the corona pandemic for modern cities. The scholarly work in this book is separated into four interconnected research challenges and themes: I. Cities, cooperation, and resilience in the face of COVID-19 II. Comparative approaches on patterns and effects of city and location-specific policies and socioeconomic structures during COVID-19 III. The socioeconomic and labor market effects of pandemics on cities and local economies IV. The need for new types of data and applications in addressing challenges in analyzing the effects of COVID-19 on cities The contributions in this volume of Springer’s The Voice of Regional Science series are 14 in total, and are outlined based on the above categorization in the next synthesis section.

The Pandemic and the City

3

2 Synthesis of the Book 2.1 Cities, Cooperation, and Resilience in the Face of COVID-19 In the first chapter in this volume titled “The Resilience of Cities to COVID-19: A Literature Review and Application to Dutch Cities”, Jeroen van Haaren and Frank van Oort, discuss how cities may be affected by the COVID-19 pandemic in the medium to long run from the perspective of urban economics, focusing on the transmission of knowledge. This literature review on city resilience to the COVID-19 pandemic paves the way for the remaining part of this chapter by providing an overview of recent thoughts on the impact of the pandemic on urban economies and adding a case study of the four largest cities in the Netherlands. The scene is set in a further conceptual formulation in the following thought-provoking chapter by Denise Pumain in the chapter titled “A Subjective Geographer’s Experience of Pandemic and Confidence in Systems of Cities.” The chapter discusses the possibility of drastic changes in the territorial organization of the world in light on prior knowledge of the dynamics of cities within systems of cities, providing a framework for assessing the plausibility of predictions. The next chapter in Part I is written by Amitrajeet A. Batabyal and Hamid Beladi titled “City and Regional Demand for Vaccines Whose Supply Arises from Competition in a Bertrand Duopoly.” This chapter presents a one-period model of an aggregate economy composed of cities and regions that demand vaccines designed to fight a pandemic such as COVID-19. The equilibrium pricing strategies of two firms are modeled in a Bertrand Duopoly context, and the incentives to conduct R&D faced by two firms are analyzed under different competition scenarios. Part I is concluded with the chapter titled “Post-Pandemic Cities—Competing for Size or Cooperating for Interaction: An Analysis of the Evolution of Portuguese Municipalities Based on an Organic and Rational Spatial Interaction Growth Models”, by Tomaz Ponce Dentinho. The chapter focuses on the spatial interaction of Portuguese municipalities from 1960 until 2021, with regards to an implicit model of spatial interaction between municipalities, and the post-pandemic scenarios of spatial interaction between municipalities. Two scenarios of post-pandemic evolution of Portuguese City Regions are simulated; it is observed that the rational interaction expression leads to more concentration of population around the Lisbon and Porto metropolitan areas.

4

M. G. Celbi¸s et al.

2.2 Comparative Approaches on Patterns and Effects of City and Location-Specific Policies and Socioeconomic Structures During COVID-19 Part II of the volume delves further into an empirical analysis of the pandemic on places by juxtaposing location-specific socioeconomic characteristics and policies pertaining to slowing down the spread of the virus. In the chapter titled “The Social Digital Twin for Liveable Cities: A COVID-19 Case Study”, Corentin Kuster, Sanne Hettinga, Tim van Vliet, Henk Scholten and Paul Padding tackle the issue of privacyrelated difficulties when analyzing the behavior of urban population. The authors illustrate how synthetic populations can be used to shed light on the impact of policy measures (such as lockdown restrictions) in a case study context. The relationship between COVID-19 infections and Non-Pharmaceutical Interventions (NPI) such as Lockdown/Stay-at-Home policies is studied in a comparative context by Kingsley Haynes, Rajendra Kulkarni, Abu Siddique, and Meng-Hao Li in the next chapter, titled “The Impact of Differing Covid-19 Mitigation Policies: Three Natural Experiments Using Difference-in-Difference Modelling”. The authors conduct a comparative analysis of this relationship on a sample consisting of counties on the Iowa and Illinois border, the Dakotas (North and South) and Minnesota border and the Arkansas and Mississippi border. The study finds that state policies appear to make a significant difference in some of these specific border regions. Part II then continues with the chapter titled “On the Association between Income Inequality and COVID Spread: A View into Spanish Functional Urban Areas”, by David Castells-Quintana and Vicente Royuela, where the authors analyze the relationship between the size, income and income inequality of cities, during the spread of the COVID-19 in Spain. The authors identify several socioeconomic and demographic attributes of Functional Urban Areas that are related to increased virus transmission. This part is concluded with the chapter titled “Urbanization impact arising from the behavioral shift of citizens and consumers in a post-pandemic world” by Tannistha Maiti, Anwita Maiti, Biswajit Maiti, and Tarry Singh. The authors explore the urbanization impact from behavioral shifts of citizens and consumers due to Covid-19 pandemic and the way it impacted urban design and, in some cases, forced an urban redesign. In addition to the discussion of several simulated models, the chapter also reviews the economic impact of control strategies that allow social interaction between low-risk individuals in the post-pandemic world where both low-risk and high-risk populations are gradually being reunited towards a ‘new normal’.

2.3 Socioeconomic and Labor Market Effects of Pandemics on Cities and Local Economies The relationships of place-based economic characteristics with pandemic-related effects are addressed in two chapters in Part III, where the first provides an illustration

The Pandemic and the City

5

of the Synthetic Control method, and the latter uses machine learning techniques. The first chapter titled “What Happened after SARS in 2003? The Economic Impacts of a Pandemic” by Ilan Noy, Yasuyuki Sawada, Nguyen Doan, and Canh Phuc Nguyen, quantifies the economic impacts of SARS on the four affected Asian economies and the most affected Chinese provinces/cities using the Synthetic Control Method with macroeconomic and remote-sensing nightlight data. Among other findings, the authors observe that, while national economies have indeed bounced back quickly and fully, at a more local level, economies have taken longer to recover. Part III continues next with the chapter titled “Industrial Composition, Remote Working, and Mobility Changes in Canada and the US during the COVID-19 Pandemic: A SHAP value analysis of XGBoost predictions” by M. Güney Celbi¸s, Cem Özgüzel, Karima Kourtit, and Peter Nijkamp. Through the use of interpretable machine learning methods, the authors find, among other results, that in Canadian and US regions, the locations where mobility was reduced more were those with a higher share of persons in jobs amenable to remote working and persons working in the information and communication and science and technology-oriented sectors.

2.4 The Need for New Types of Data and Applications, and Existing Challenges in Analyzing the Effects of COVID-19 on Cities The final part of this volume highlights the crucial issue of challenges and needs for new types of data, and how they can be used in analyzing contemporary pandemic related-research questions. Various challenges are identified in the chapter titled “Problems with Recording the Spread of COVID-19 in Developing Countries: Evidence from a Phone Survey in Indonesia” by Budy P. Resosudarmo, Rus’an Nasrudin, Pyan A. Muchtar, Usep Nugraha, and Anna Falentina. This chapter focuses on evaluating the reliability and usefulness of a phone survey approach of directly asking respondents whether they are infected by COVID-19 and how the sickness has impacted their livelihood through a novel illustration of the method. Among other findings, the main results suggest that by developing a careful algorithm, a phone survey might be able to provide a better estimate than those announced by the government in developing countries where capabilities of conducting COVID-19 test are limited. In the second chapter of this part, Jacob Irving, Keith Waters, Terry Clower, and William Rifkin provide new data for researchers in the field by developing an index tool to identify regions that are likely to face the greatest difficulty in economic recovery from the COVID-19 pandemic in their chapter titled “Pandemic Regional Recovery Index: An Adaptable Tool for Decision-making on Regions.” After developing the index for six countries, the authors highlight that the tool can provide policy makers with important insights on which regions may require additional resources to fully recover from a pandemic or other unusual global impact. The following chapter titled “The Geography of Daily Urban Spatial Mobility during

6

M. G. Celbi¸s et al.

Covid: The Example of Stockholm in 2020 and 2021” by Ian Shuttleworth, Marina Toger, Umut Türk, and John Östh uses mobile phone data to assess daily spatial mobility in the Stockholm urban area during the Covid pandemic in 2020 and 2021. The chapter outlines various findings in mobility patterns throughout Stockholm during this period that present significant potential opportunities for targeted policymaking. The final chapter in this part, titled “Social Justice, Digitalization, and Health and Wellbeing in the Pandemic City” by Laurie A. Schintler and Connie L. McNeely provides an integrated and in-depth exploration of pandemics and social justice, as a contribution to a larger cross-cutting literature. The chapter devotes particular attention to artificial intelligence (AI) and algorithmic performance. The authors also specify imperatives for metrics development for assessing social justice implications of AI for urban pandemic management.

3 Retrospect and Prospect The above outlined fourteen chapters on topics pertaining to various place-based dynamics within the context of the pandemic constitute self-standing analytical studies. These different viewpoints on the topic constitute both applied and theoretical perspectives in addition to highlighting novel ways of using new types of data while underlining the various challenges. New research avenues can be opened through the diversity in scope and scientific approach of the chapters presented in this volume. From the studies in this volume, three main ways of how urban life and urban research is being transformed—depending on new human needs—can be highlighted. Firstly, as the effects of the pandemic have been the strongest particularly in locations where there is a high degree of face-to-face interaction among individuals, urban health is now becoming a more prominent topic in socioeconomic analysis of cities. The spatial dimension of the social, economic, and health effects of the pandemic on individuals, communities, and industries has now a much higher relevance than in the pre-corona age, both from a policy-making and a scientific perspective. Secondly, the dwindling effects of the COVID-19 pandemic are slowly but surely leaving their marks on cities. While dramatic changes are not expected, cities all over the world are, and will continue to, adjust themselves to new health challenges. The pandemic experience has granted urban administrations and urban dwellers various types of flexibilities that can be useful not only for pandemics, but also in situations involving other threats such as natural disasters. Ability to swiftly adjusting towards online education and work, stay-at-home practices, and setting up field hospitals can be seen as valuable new collective skills and experiences that are now ingrained in urban communities. Finally, as a closely related phenomenon to the first two points, data monitoring, decision support systems and health dashboards have been coming forward as important analytical vehicles to support the resilience of cities in the face of the pandemic (and other catastrophes). Fast and accurate spatial information on outbreaks, hospitalizations, alongside with mobility and population density has been

The Pandemic and the City

7

instrumental for the timely implementation of policy responses and for allocating the public and private resources of the city efficiently. For the safety of urban citizens, the protection of industry agglomerations, and the cultivation of lively and resilient communities, smarter and resilient cities are likely to characterize the post-pandemic world.

Cities, Cooperation, and Resilience in the Face of COVID-19

The Resilience of Cities to COVID-19: A Literature Review and Application to Dutch Cities Jeroen van Haaren and Frank van Oort

1 Introduction1 Highly dense cities are an ideal environment for the spread of viruses. COVID-19 is not the first pathogen to spread quickly in cities. The plague, cholera, and Spanish flu are examples of pandemics that have had a major impact urban populations (Glaeser & Cutler, 2021; Griffin & Denholm, 2020). In some cases, these pandemics have also led to new and innovative urban planning (Florida et al., 2021). For example, sewage systems can be directly traced back to efforts to control pathogens (Johnson, 2006). According to some, the advent of sewage has enhanced the benefits of urban agglomeration, as the advantages of living in high densities have been decoupled from hygiene and health drawbacks (Rosenthal & Strange, 2004; Rosewell, 2016). At the same time, the continued growth of cities that sanitation has enabled has resulted in other disadvantages such as congestion, crime, and poor air quality. COVID-19 may arguably be a driver for a new systemic leap in the organization of cities (Eltarabily & Elgheznawy, 2020). At the same time, the COVID-19 pandemic is affecting productivity, creating disruptions and driving the destruction of economic activity in cities (Kamal, 2020). Glaeser and Cutler (2021) called the pandemic, and the weak underlying institutional structure of society, “demons of density,” and argued that human interaction, which is 1 Parts of this chapter have been discussed (in Dutch) in: Economische Verkenning Rotterdam 2022, www.evr010.nl.

This work benefited from funding through ZonMw project 10430-03201-0006 (“The Resilient Region; Regional-Economic Impact Mitigation of Corona-related (De)escalation Policies”). J. van Haaren (B) Erasmus Centre for Urban Port and Transport Economics, Rotterdam, The Netherlands e-mail: [email protected] F. van Oort Erasmus School of Economics, Rotterdam, The Netherlands © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. G. Celbi¸s et al. (eds.), Pandemic and the City, Footprints of Regional Science, https://doi.org/10.1007/978-3-031-21983-2_2

11

12

J. van Haaren and F. van Oort

responsible for most of the prosperity of cities in past decades (Glaeser, 2011), is now associated with increased risk of contagion. To reduce the rate of contagion, many governments introduced social distancing measures, ranging from working from home, limiting transport capacity, closing schools so that children must be educated at home, to closing shops and personal services (Thissen et al., 2022). These measures were accompanied by a sharp downturn in economic activity. Interestingly, Sheridan et al. (2020) showed that the primary cause of the downturn in economic activity was the virus itself rather than social distancing measures. Following a longer than expected pandemic and its resultant measures, a sense of urgency emerged for developing digital technology (Piros, c˘a et al., 2021), resulting in the rapid development of remote working solutions, as evidenced by shifting patent applications (Bloom et al., 2021) as well as strong price increases in stocks for remote working solutions. Solutions range from advanced videoconferencing to remote learning options to cloud computing. Services which were previously provided in-person were done virtually, such as telehealth (Golinelli et al., 2020; Hollander & Carr, 2020). Many business processes were converted to digital equivalents, demonstrating the ability of firms and employees to adapt rapidly. As such, the pandemic has led to an accelerated “emergency” digitization of work in firms (de Lucas Ancillo et al., 2021; Kudyba, 2020; Soto-Acosta, 2020), public services (Agostino et al., 2021; Gabryelczyk, 2020), education (Cone et al., 2021; Crawford et al., 2020; Williamson & Hogan, 2020), and consumption patterns. Others argue that the pandemic has laid bare the limits of digitization with respect to organizing business processes (Faraj et al., 2021). Nevertheless, the degree and speed of digitization technology developed and adopted during the pandemic is unprecedented. This chapter presents a literature review on the spatial-economic implications of COVID-19 measures during and after the pandemic. This review focuses on the potential revival of the death of distance debate, where in line with the above narrative, it is argued that working from home can fundamentally change the need for commuting, the demand for office space, and in-person interactions with workers. Arguments in favor of and against this process are discussed. We tested these arguments with an expert focus group comprising real estate agents, local and regional policy-makers, and specialized spatial-economic consultants and reported the findings of this discussion. The chapter then presents a crude analysis that links three concepts that are central to the discussion. These concepts link working from home opportunities to sectoral urban and regional structure in the context of COVID-19 prevalence to evaluate whether more spiky spatial economies are resilient during and after the pandemic. The chapter concludes with the presentation of a research agenda.

2 Reviving the Death of Distance Discussion? Does this revive the long-lasting “death of distance” discussion prevalent in the popular literature? With San Francisco (Schram, 2022) and New York (Boyle &

The Resilience of Cities to COVID-19: A Literature Review …

13

Rockeman, 2022) not seeing the return of commuters to half of pre-pandemic capacity, this time, it seems different. Furthermore, the results of Mondragon and Wieland (2022) suggest a degree of permanency in this pattern of behavioral change, as they attributed half of the house price growth between December 2019 and November 2021 to the effect of remote working (see also Barrero et al., 2021). There have been many forecasts that the importance of distance has decreased (Cairncross, 1997; Friedman, 2005), starting with Future Shock (1970) and The Third Wave (1980) by futurist Alvin Toffler, who had little confidence that cities would survive and instead saw a big future for working from so-called electronic cottages built for telecommuting. However, such forecasts were followed by periods of increased importance of agglomeration economies and a spikier economic geography (Florida, 2005). According to Glaeser and Cutler (2021), this is because the working population grew more than the (limited) percentage of employed persons working from home, who must still overcome loneliness, lack of interaction, and coordination. Over the last decades, large technological leaps have coevolved with increased urbanization (Leamer & Storper, 2001). An important question is whether this coevolution is causal. Most evidence suggests that technological progress creates additional jobs, as previously unexplored ideas become feasible due to new enabling technologies and the automation of more mundane (i.e., routine) tasks (Florida et al., 2021). Additionally, technology may complement labor as technologies augment the productivity of workers (Autor et al., 2022). The lockdown during the recent pandemic and the resulting forced exploration, adoption, and innovation in tools for reducing physical proximity have led to rapid technological progress. These new tools are designed to substitute physical proximity, again potentially reducing its importance. New in the discussion on geographic proximity and the mediating role of information technology is the technology-readiness level of new collaboration tools, especially videoconferencing. Current technologies allow for dialog, more frequent interactions, and conveying (limited) emotion and body language, more readily emulating personal interaction, and its “capacity for interruption, repair, feedback, and learning” (Nohria & Eccles, 1992, p. 292 in: Bernela et al., 2019). This is important as communication via these media satisfies more conditions for conveying tacit knowledge. Most evidence on the transference of tacit knowledge dates prior to 2010, with lower technology levels and more one-dimensional communication, such as e-mail, chat or voice messaging. The work of Castaneda and Toulson (2021) suggests that a critical threshold has been passed whereby a higher degree of tacit knowledge can now be transferred through digital means. At the same time, Bernela et al. (2019) argued that face-to-face interaction, and by extension, colocation, are most important for the transference of highly tacit knowledge, as innovation requires frequent contact. Information technology now provides more opportunities for frequent contact while still maintaining a degree of the benefits of face-to-face interaction. However, this seems to be not a disruption or reversal but an acceleration of an existing trend. The question remains how agglomeration benefits are affected. With new forms of infrastructure and digitization, academia and practitioners wonder

14

J. van Haaren and F. van Oort

whether the importance of physical proximity decreases. At the same time, experience shows that the importance of physical proximity increases with technological breakthroughs. This can be explained by the scarcity of agglomeration advantage: if the accessibility of all places is increased, it is those locations that can still distinguish themselves by offering a unique advantage that prosper (Breznitz, 2021). In other words, the number of places that offer a significantly greater number of information and knowledge dissemination options than those offered by the recent technological advance, is decreasing. By extension, this makes places with an extraordinarily high rate of knowledge dissemination scarcer and at the same time more desirable from a firm perspective. In particular, highly dense cities are more conducive to more complex interactions (Balland et al., 2020). As a result, the most central, metropolitan living environments are arguably the most attractive locations for innovative activities, as they offer a scarce resource: the sharing of more tacit knowledge at lower transaction costs than other locales or using substituting technologies.

3 The Benefits and Costs of Location-Independent Work The COVID-19 pandemic also gave rise to a separate but highly related question. Given an unprecedented boost to digitization location independent work (in most cases working from home) has gained ground; see the developments in New York and San Francisco. How does this affect learning within and between organizations? The insight that digitization can boost productivity has been voiced before (Chan et al., 2018), but barriers to widespread adoption have prevented firms from capitalizing on these advantages. These barriers fall into four categories: . . . .

Culture: Work is at the office. Trust: Will employees be productive? Risk: Will it work? Will our clients accept it? Collective action: It only works if everyone does it.

The conditions of the pandemic created a sense of urgency that forced firms to rapidly adapt or perish. This created a unique set of conditions that caused a paradigm shift in the way that work is organized. The initial results suggest positive experiences and persistent behavior. At the macro level, the numbers suggest that most employees choose to perpetuate some but not all of the modified behavior: employees return to the office, although not with the same frequency as before. The pandemic has led to a forced experiment with digitization and working from home in many organizations. Early experiences suggest that working from home results in various cost savings: (1) monetary cost savings (Beno, 2021), (2) transaction cost savings, and (3) productivity gains (OECD, 2020). Conversely, there are indications that working from home also increases (1) social costs, (2) health costs, and (3) innovation costs. Working from home results in significant monetary cost savings, both for employers and employees. Accountancy firm PwC (2020) estimated that employers would save e1.7 billion per year if all Dutch employees work from home for one

The Resilience of Cities to COVID-19: A Literature Review …

15

day. The popular press suggests that employers are also more positive about the role of working from home in their business processes compared to the situation before the pandemic (CNN, 2022; Financial Times, 2021). Office space is a potentially significant source of cost savings for employers, but savings also occur through minimized catering, energy, and travel expenses (in the Dutch case, the variable costs of company cars are an important source of cost savings). Research by DJV Insights (2021) showed that even after the pandemic, many employees desire to continue to carry out part of their work at home. Employees gain back the non-reimbursable financial costs of commuting and may save on the number of hours of childcare needed (due to travel time savings outside work hours) (Lupu, 2017; in: Beno, 2021). It seems that the increase in working from home, at least in part, is permanent (Barrero et al., 2020). The Netherlands Bureau for Economic Policy Analysis (CPB) (Jongen et al., 2021) predicted a doubling of working from home compared to the pre-COVID-19 situation. This is in line with the results of Hamersma et al. (2020), who found that many employees reported positive experiences of working from home as well as an increased preference to work from home in the future. Beno and Hvorecky (2021) showed that working from home may be accompanied by a fall in productivity. Conversely, productivity gains are also plausible: the technology that we now use to work from home can be used post-COVID-19 to work at any desired location (Barrero et al., 2021). The time savings this creates can be an important reason to rearrange work patterns. Travel time also decreases during work, and appointments are handled digitally so that more conversations can be held in each time span. Even if digital conversations are less productive, overall productivity can still increase if the time savings outweigh the missed benefits of an in-person meeting. It is plausible that there is a decreasing marginal return on in-person meetings, which would imply an optimum that is partly digital and partly in person. However, this does require specific infrastructure and adaptive capacity on the part of the office worker. De Lucas Ancillo et al. (2021) argued that the digitalization response to the COVID-19 pandemic goes beyond simple digital substitution and requires “breaking with the past.” While companies initially used technology to emulate existing business processes remotely, they subsequently redesigned business processes, the workplace, governance, and corporate culture to new conditions and opportunities arising from digitization. It is not yet clear whether the average company adapts to digital change or applies a digital transformation (Soto-Acosta, 2020), the latter being transformational in the business process. However, there are also critical voices about this new way of working. The long-term effect of the pandemic is still unclear. Employees miss face-to-face interactions and social contact with colleagues and the substantive and social depth in the work (Glaeser & Cutler, 2021). This is at the expense of work quality, happiness at work, and, in some cases, even the well-being of the employee (Josten & Merens, 2021). Roper and Turner (2020) pointed out that innovation in firms is procyclical and that small and medium-sized enterprises (SMEs) face a varying degree of cash constraints to innovation because of the pandemic, resulting in uneven opportunities to invest in innovation and thus adapt to post-COVID-19

16

J. van Haaren and F. van Oort

conditions. This view is further supported by Bartik et al. (2020), who showed that the median SME (in the U.S.) had about a two-week worth of cash reserves on hand. Furthermore, it is conceivable that frequent working from home has negative consequences for learning and innovation not only within but especially between organizations (Glaeser & Cutler, 2021). Working from home does not fit well with the ideal image of innovation ecosystems, with nonlinear process planning, many personal interactions, and implicit knowledge transfer. Conversely, if we view the technology leap in digitization not as solely enabling working from home but as an enabler for location-independent work, interaction between people and organizations can become more frequent. Incidentally, the impact of COVID-19 on different organizations and groups of employees is certainly not uniform. There are major differences between large companies and small- and medium-sized enterprises and between younger or new employees and experienced employees. In SMEs, working in person “in the office” is more so the norm. Recently hired or young employees seem to experience the strongest benefits from interaction by learning how to work in the business (Glaeser & Cutler, 2021; Kahn, 2022). More experienced employees, who may have more familial obligations, experience more benefits from working from home, although this is strongly dependent on the family situation (e.g., having children who attend school), home characteristics, and personal preferences. However, the heterogeneous impact is of importance for the knowledge transfer between these groups, as differing preferences and benefits limit interactions (Grabner & Tsvetkova, 2022; Thissen et al., 2022) and therefore limit a source for learning within organizations. Furthermore, the workplace is not just a place to attain productivity, learn about business processes, and share knowledge. It is also a place where “learning the ropes” and adopting a broader skillset in the form of competency development takes place. The workplace can be viewed as a meaningful place for development that aids in providing purpose and belonging and helps shape identity (Michaelson et al., 2014). It is a place for personal development, as employees are also inspired by colleagues through mentoring and strategic leadership. The results of Younas and Bari (2020) even suggest that talent retention depends, in part, on these activities rather than knowledge sharing.

4 Consequences for the Urban and Regional Economy However, it seems that location-independent work finally has carved out a more permanent place in the knowledge economy, which has consequences for the urban economy. In the office market, this mainly leads to a change in the use of space. The value of the office will be assessed in a different way. The employee comes to the office for the possibility that the building offers to engage in valuable interactions with others people such as colleagues, customers, suppliers, and partners. This leads to a different set of requirements regarding the office and therefore also has consequences for the design, whereby common areas for consultation and informal spaces such as

The Resilience of Cities to COVID-19: A Literature Review …

17

coffee bars and sports facilities are more central (Florida et al., 2021). At the same time, separate units are important for digital meetings and concentration rooms. In consulting a group of Dutch experts,2 respondents doubted whether more locationindependent work leads to an increased or decreased demand for office space. The experts state that the firms they advise have not yet made firm decisions on future leases. For the time being, companies are closely monitoring new developments and try to optimize within existing leases. It appears that parties vested in keeping offices have limited short run options for adjustment. Relative radical behavioral changes offer room to accelerate developments that were initiated before COVID-19 (Van Oort & Thissen, 2021). For example, the pandemic has led to an increased focus on health in the city and a healthy life balance. Attention to greening the city and space for slow traffic (e.g., cycling and walking) is also increasing. Here, too, the pandemic has an accelerating effect. Resilience is increasingly associated with innovation: Challenges such as the COVID-19 pandemic create opportunities for improvement. However, the pandemic also accelerates inequality in the city. Knowledge workers (Autor et al., 2003; Drucker, 1999) in business services appear to be less affected by the pandemic than employees in personal services and production (Glaeser & Cutler, 2021). For people with freedom of choice, the palette of possibilities increases. On the other hand, the options for people in a profession in which working elsewhere is not possible, or for people with a more limited budget, are decreasing, and as such a digital divide may be occurring (Beno, 2021; Miladinovic, 2020). In their book Survival of the City, Glaeser and Cutler (2021) showed the other side of economies of scale and high densities in cities. They argue that the pandemic had a profound impact on cities because of their underlying problems. These problems center around the provision of universal access to the city and how “insiders have rigged the game.” Their argument revolves around the universal but unequally distributed benefits of cities and the lack of openness toward talent and new business. They argue that cities are at risk of demise because institutions hinder entrepreneurship and the entry of new residents, as well as perpetuate pockets of poverty by restricting new development. Of particular interest to this chapter is their argument that the services industry provides mass employment for those without a college degree as well as for high-paying talent. Glaeser and Cutler reason that in an age of digitization, humans retain an edge position in personal interaction and creativity. Their anecdotes show that many personal services in cities are more about experience rather than function and luxuries rather than necessities. We conducted a focus group interview with policy-makers and industry experts in December 2021, the results of which provide additional insight into the role of the city as a nexus for human interaction (reference: EVR, see also footnote 2). The experts agreed that the hypothesis of i.e. Glaeser and Cutler (2021) that the pandemic had a very disruptive effect on existing behavior that does not seem correct; rather, there is limited adjustment and especially even acceleration. This is related to the ability of people to resist and bounce back; as soon as an obstacle has been removed, we 2

The authors thank: Gilbert Bal, Klaas-Bart van den Berg, Marco Clarijs, Gabor Everraert, Walter Hulsker, Ruud Kruip, Jan-Daan Maasland and André Ouwehand for sharing their expertise.

18

J. van Haaren and F. van Oort

often return to established behavioral patterns. Our exploration shows that locations are increasingly evaluated on multiple criteria, which increases the complexity of the city. Before COVID-19, a preference seemed to arise for metropolitan, mixed-use interaction environments, with rich facilities and preferably sufficient green space. The pandemic has accelerated the reaffirmation of this preference. Another trend that has been reaffirmed is the rise of the services industry, particularly regarding the integration of digital and physical services, such as grocery shopping and online food delivery—(Gavilan et al., 2021). The widespread adoption of these technologies at an unprecedented rate has supplanted traditional services. However, this adoption process was in part forced and does not serve the human desire for human interaction that is so inherently ingrained in the service economy. Many services we consume are not necessities but luxuries, which we can do without if they are deemed too risky (Glaeser & Cutler, 2021). During the pandemic, digitization has created complements for many necessary services, but not the luxuries that are delivered face-to-face and whose primary value is human interaction and experience (Farag et al., 2006; Weltevreden et al., 2008). It is that part of the service economy which we forego during lockdown, and which we quickly readopt when restrictions are relaxed. This point of view is further strengthened by the insights of Correia et al. (2020), who showed that it is the health-related effects of pandemics that depress the economy rather than government intervention. At the same time, this also means that locations that were perceived as less attractive before the pandemic will face tougher challenges after the pandemic. Monofunctional and decentralized locations and outdated real estate will especially face challenges in competing for business. As the urgency to commute to work has been reduced for many office workers, the requirements that work locations need to meet to still attract workers have increased.

5 The Pandemic and Dutch Regions: Indications for Spikier Economies? In this section, we preliminarily evaluate several issues raised in the previous sections. Preliminary, as up-to-date data on economic behavior are scarce, the pandemic is not fully over. We work with hypotheses posted by Glaeser and Cutler (2021), linking density to the association between sector structure (with respect to knowledge intensive business services and hospitality), COVID-19 prevalence, and the ability to work from home just prior to the COVID-19 pandemic. We assess bivariate relations between these three indicators in the 40 NUTS-33 urban regions of the Netherlands and relate each of these to density to evaluate whether more spiky economies fare better or worse in and after the pandemic.

3

For more information on the Nomenclature of Territorial Units for Statistics (NUTS), see https:// ec.europa.eu/eurostat/web/nuts/background.

The Resilience of Cities to COVID-19: A Literature Review …

19

The ability to work independently of location differs by occupation and sector. It is mainly the service and business professions in the knowledge economy that offer opportunities for location-independent work. This poses an important dilemma. It is precisely in this knowledge-based economy that proximity plays a crucial role in knowledge exchange and therefore in fostering innovations from which firms, cities, and regions can derive competitive advantages. Therefore, we may expect that working from home is more prevalent in regions with higher shares of knowledgeintensive business services and in regions where face-to-face meetings may lead to more spiky economies. Figure 1 suggests a clear correlation between the ability to work from home and the share of knowledge-intensive business services in urban regions in the Netherlands. Although this does not prove causality, it is plausible that knowledge-based economies allow for more location independent work (in line with Glaeser & Cutler, 2021, Chap. 7) as Fig. 1 further confirms that density is positively related to both the share of knowledge intensive business services and the ability to work from home. In knowledge-based economies, hospitality services (e.g., hotels, restaurants, meeting places) are often identified as key facilitators of interaction between employees of different firms (Glaeser & Cutler, 2021; Thissen et al., 2022). Therefore, although the hospitality industry itself cannot work well from home, we would expect a correlation between high shares of the hospitality industry and the ability to work from home in regions if hospitality industries function as facilitators of knowledge-based economies. However, Fig. 2 shows that there is little to no correlation between hospitality industry shares and the ability to work from home prior to the pandemic. Although this result is surprising, it does not exclude the possibility 50% [CELLRANGE] [CELLRANGE]

Prevalence of working from home

[CELLRANGE]

45% [CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE]

40%

35%

[CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE]

30% [CELLRANGE]

[CELLRANGE] [CELLRANGE] [CELLRANGE]

25% [CELLRANGE]

20% 0.0%

5.0%

10.0%

15.0%

20.0%

25.0%

30.0%

Percentage of Knowledge Intensive Business Services

Fig. 1 Working from home and knowledge intensive office workers—a more knowledge intensive workforce is more adaptable. Source Centraal Bureau voor de Statistiek (Statistics Netherlands) 2018

20

J. van Haaren and F. van Oort

Prevalence of working from home

50% [CELLRANGE]

45%

[CELLRANGE] [CELLRANGE] [CELLRANGE]

[CELLRANGE]

[CELLRANGE]

[CELLRANGE]

[CELLRANGE]

35%

[CELLRANGE] [CELLRANGE]

[CELLRANGE] [CELLRANGE]

40%

[CELLRANGE]

[CELLRANGE] [CELLRANGE][CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE]

[CELLRANGE]

[CELLRANGE] [CELLRANGE]

30%

[CELLRANGE]

[CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE]

[CELLRANGE]

[CELLRANGE] [CELLRANGE] [CELLRANGE]

[CELLRANGE]

25% [CELLRANGE]

20% 0.0%

1.0%

2.0%

3.0%

4.0%

5.0%

6.0%

7.0%

8.0%

9.0%

10.0%

Percentage of hospitality services

Fig. 2 Working from home and hospitality services—no correlation. Source Centraal Bureau voor de Statistiek (Statistics Netherlands) 2018

of more complex conditional correlations resulting from counterbalancing forces.4 On the one hand, the hospitality industry may facilitate services that allow for more working from home, while on the other hand, working from home is rare in most occupations within the hospitality sector and other sectors that are strong drivers of the hospitality industry such as tourism, culture, the arts, and the events industry. However, for now, we conclude that regional economies with large shares of hospitality industries are unrelated to economic structures that enable working from home (contrary to the suggestion in Glaeser & Cutler, 2021, Chap. 6). Furthermore, Fig. 2 provides another surprising result: the share of hospitality services is not strongly correlated with density in Dutch regions, which is contrary to what was expected. As Glaeser and Cutler (2021) argued, the artistry, care, and effort of a barista making a latte for the enjoyment of the customer is hard to digitize; at the same time, this interaction is seen as a risk of contagion during a pandemic. We can derive two insights from this. First, in a service-based knowledge economy, direct interaction may be required to experience the benefits of personal attention, as is also argued in the case of knowledge exchange. Therefore, if the transmission of knowledge and personal experience follows the same mechanism of viral transmission, namely, frequent personal interaction, then we would expect the two to coincide. Second, this would result in higher COVID-19 prevalence in knowledge-based economies. Figure 3 suggests that COVID-19 incidence and knowledge-intensive business services coincide in urban regions in the Netherlands. Regions with larger shares of knowledge-intensive business services also carry a higher risk of COVID-19 infection (Glaeser & Cutler, 2021, Chap. 5). We do not claim that this is a causal connection but simply point out that the conditions for knowledge transfer are likely the same 4

One may argue that this argument only holds if the hospitality industry serves the knowledge economy. To address this concern, we also tested the conditional correlation in the setting of a parsimonious OLS model. This estimate yields similar (insignificant) results.

The Resilience of Cities to COVID-19: A Literature Review …

21

0.35 0.33 0.31

[CELLRANGE]

[CELLRANGE]

[CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE]

0.27 0.25

[CELLRANGE] [CELLRANGE] [CELLRANGE]

[CELLRANGE] [CELLRANGE]

0.29

Covid incidence

[CELLRANGE]

[CELLRANGE]

[CELLRANGE][CELLRANGE] [CELLRANGE]

[CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE]

0.23

[CELLRANGE] [CELLRANGE]

[CELLRANGE] [CELLRANGE]

0.21 [CELLRANGE]

0.19

[CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE]

0.17 0.15 0.0%

5.0%

10.0%

15.0%

20.0%

25.0%

30.0%

Percentage of Knowledge Intensive Business Services

Fig. 3 Risk of contagion and knowledge intensive office workers—places with more knowledge workers have a higher risk of contagion. Source Centraal Bureau voor de Statistiek (Statistics Netherlands) 2018, RIVM 2022

conditions relevant for mass contagion. The results with respect to density in this figure are also compelling, as all dense regions have high COVID-19 incidence, but not all low-density regions have low incidence. This suggests that while density may nearly guarantee contagion, low density does not necessarily protect against contagion. Additionally, and by extension of the earlier argument, we also assessed the association with hospitality services. Hospitality services, such as consumer amenities, are often argued to be important facilitators of the knowledge economy (Glaeser et al., 2001). They offer workspaces to gig workers and facilitate meetings and knowledge exchange. As such, they could be enablers of contagion, as well as knowledge exchange. Figure 4 suggests an absence of correlation between COVID-19 incidence and hospitality services. The figure can easily be interpreted as suggestive of a negative correlation in specific subgroups. The effects of national lockdowns on these results should be acknowledged, as hospitality services were closed or highly regulated for a long time. To relate whether regions in which working from home was easier prior to the pandemic and were also better protected from COVID-19 contagion, we examined Fig. 5. This figure suggests a weak positive association, indicating that the ability to work from home at higher frequencies did not protect these regions from contagion and that other factors related to human interaction may better explain COVID-19 incidence. We see a positive association between density and both the COVID-19 incidence and the ability to work from home. To sum up the results and assess whether Glaeser and Cutler’s “demons of density” are also at work in the Netherlands, we found that denser urban regions have higher COVID-19 incidences and have a sector structure favoring knowledge-intensive business services that simultaneously offer more opportunities to work from home.

22

J. van Haaren and F. van Oort 0.35 [CELLRANGE]

[CELLRANGE]

0.33

[CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE]

0.31

Covid incidence

0.29

[CELLRANGE]

[CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE]

[CELLRANGE]

0.27

[CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE]

0.25 0.23

[CELLRANGE]

[CELLRANGE]

[CELLRANGE]

[CELLRANGE] [CELLRANGE]

0.21

[CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE]

0.19

[CELLRANGE]

0.17 0.15 0.0%

[CELLRANGE] [CELLRANGE] [CELLRANGE]

1.0%

2.0%

3.0%

4.0%

5.0%

6.0%

7.0%

8.0%

9.0%

10.0%

Percentage of hospitality services

Fig. 4 Risk of contagion and hospitality services—no clear correlation. Source Centraal Bureau voor de Statistiek (Statistics Netherlands) 2018, RIVM 2022 0.35

0.31

Covid incidence

0.29

[CELLRANGE]

[CELLRANGE]

0.33

[CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE]

0.27

[CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE]

0.25

[CELLRANGE] [CELLRANGE]

0.23

[CELLRANGE]

[CELLRANGE]

[CELLRANGE] [CELLRANGE]

0.21 0.19

[CELLRANGE]

[CELLRANGE] [CELLRANGE] [CELLRANGE] [CELLRANGE]

0.17 0.15 20%

25%

30%

35%

40%

45%

50%

Prevalence of working from home

Fig. 5 Working from home and contagion—prevalence of working from home and COVID-19 incidence show no correlation. Source Centraal Bureau voor de Statistiek (Statistics Netherlands) 2018, RIVM 2022

However, the ability to work from home did not protect these regions from contagion. We found that in the Netherlands, density and COVID-19 incidence did not strongly correlate with the share of hospitality services in the region but note that the hospitality services industry was among the most tightly regulated sectors during the pandemic. Furthermore, our analysis shows that while density coincides with COVID-19 incidence, low density does not insure against contagion. This all points toward more vulnerable and spiky economic development during and after COVID19. To shed additional light on the dynamics between sector structure, the ability

The Resilience of Cities to COVID-19: A Literature Review …

23

to work from home, and COVID-19 incidence, we focused on the four largest and densest cities in the Netherlands.

6 The Pandemic in the Four Largest Cities in the Netherlands In this section, we briefly explore the impact of the pandemic in the four largest Dutch metropolitan regions: Amsterdam, Rotterdam, Utrecht, and The Hague. Based on secondary data on these regions (at the NUTS3 level), we assessed their economic structure in more detail, decomposed the ability to work from home prior to the pandemic, and monitored COVID-19 incidence and employment dynamics. A priori, we would expect economies with a large share of business services to have better potential to switch to remote work and subsequently have a lower COVID-19 incidence. However, the results in the previous section suggest that the reverse is true: COVID-19 incidence and working from home go hand-in-hand. Based on insights from the literature review, we explored whether we can observe a stronger impact of the pandemic in cities with a stronger specialization in the service industry compared to cities with other specializations based again on the premise that the mechanisms for the spread of knowledge are similar to the mechanisms for the spread of contagion. We calculated relative specializations for the four metropolitan regions based on firm establishment data (see Table 1). The data used are publicly available from Statistics Netherlands and include all firm establishments excluding the public sector (i.e., government, health care, and education). The four cities differ in their relative specializations. In all four cases, services are more prevalent in the metropolitan regional economy compared to the national economy. The metropolitan region of The Hague has a modest specialization in services (1.09) and, as expected, is less specialized in agriculture (0.82), industry (0.83), and trade and logistics (0.86) compared to the national economic structure. However, compared to the other three cities, these relatively space-intensive activities are most strongly represented. Specifically, for the region of The Hague, the location of public sector establishments affects the results, as this city houses the seat of parliament as well as different branches of the national government. The region of Amsterdam is strongly specialized in commercial services, both knowledge intensive (KIBS—1.18) and other commercial services (1.16), while agriculture (0.15) and industry (0.18) are underrepresented both compared to the national level as well as the other cities. The Rotterdam region specializes in trade and logistics (1.13), as the region houses the largest port in Europe. The region of Utrecht has the strongest specialization in knowledge intensive business services (0.13). As such, based on these data (as well as city characteristics), we can derive a distinct profile for each of the four cities: The Hague as a center for public services, Amsterdam as a services economy, Rotterdam as a major trade hub and Utrecht as a knowledge economy.

24

J. van Haaren and F. van Oort

Table 1 Relative specialization by firm establishments in major Dutch cities as measured by a location quotient Codes A BCDE GH ILNRS JKM

Activity Agriculture Industry Trade and logistics Commercial services (excl. KIBS) Knowledge intensive business services

NUTS-3 region The Hague 0,822 0,828 0,859 1,089

Amsterdam 0,148 0,178 0,865 1,160

Rotterdam 0,475 0,518 1,133 1,029

Utrecht 0,497 0,505 0,847 0,973

1,060

1,179

1,038

1,229

Source: Statistics Netherlands 2020-2021, author’s elaboration

Figure 6 provides insight into the impact of sector structure on the degree to which working from home was common before the pandemic in these city regions. We estimated the proportion of employment in which working from home is a common versus uncommon occurrence. We combined two secondary data sources at the national level from Statistics Netherlands on the propensities to work from home in different occupations with data on the frequency of occupation within sectors. The results were multiplied by sector structures in the respective regions to yield the results in the figure. Two key assumptions underlie the validity of these results: (1) sorting of occupations over sectors in regions mimics national patterns of sorting, and (2) propensities to work from home in specific regions are similar for specific occupations. The results of this exercise confirm that in the regions with the strongest services specializations, the ability to work from home is highest. Even before the pandemic, working from home was more prevalent in the occupations most dominantly represented in these regions. Interestingly, the region with the strongest specialization in knowledge intensive business services (Utrecht) is not the region in which working from home is most common. Rather, the region of Amsterdam, which specializes 80% 70% 60% 50% 40% 30% 20% 10% 0% Amsterdam

Utrecht Uncommon

Den Haag Common

Rotterdam

Unknown

Fig. 6 Distribution of employment in which working from home is common or uncommon by region. Source LISA 2017, Statistics Netherlands 2019, author’s elaboration

The Resilience of Cities to COVID-19: A Literature Review …

25

in other commercial services, had the highest incidence of working from home. Although tentative, this would suggest that working from home may more strongly be related to the services aspect of these jobs rather than their knowledge intensity. In Fig. 7, we plotted the cumulative number of COVID-19 infections over the total population for the four largest Dutch cities (at the municipal level), ordered from the largest to the smallest city by population. In the first one-and-a-half year of the pandemic, Rotterdam had the highest COVID-19 incidence. Under the social distancing measures and lockdown restrictions imposed, the sector structure of Rotterdam did not contribute to the ability to contain contagion (Thissen et al., 2022). Simply put, the city of Rotterdam housed an outsized number of jobs in which working from home was impossible or difficult. In the last three months of observations, we found the opposite result. Aside from the observation that COVID-19 incidence surges in the most recent three-month period under the relaxed restrictions as well as the emergence of the omicron variant, we see another interesting result. Under these relaxed restrictions, the surge in cases is strongest in Utrecht. While the city had the lowest cumulative incidence of the four cities up until December first, 2021, it now has the highest incidence. Although anecdotical, this reversal coincides with the end of government intervention. It would be interesting to study whether this re-enabled the high levels of interaction in a knowledge economy and assess whether these interactions are conductive to contagion. As a last step, the changes in gross regional product are presented in Table 2 for different stages of the pandemic. In the early pandemic, the city-region of Amsterdam was the most affected. The industry mix in the capital favors both financial institutions and tourism, sectors which were strongly affected by the pandemic. The city of Hague was relatively protected from adverse effects by its relatively large public sector. At the same time, sector structure explains the pattern of recovery. The strongest recovery occurred in Amsterdam, but to a level prior to the pandemic. The city of Rotterdam experienced the strongest growth between 2019 (prior to the 450 400 350 300 250 200 150 100 50 0 3/1/2021

6/1/2021 Amsterdam

9/1/2021 Rotterdam

's-Gravenhage

12/1/2021

3/1/2022

Utrecht

Fig. 7 Cumulative COVID-19 incidence per 1000 inhabitants in the four largest Dutch cities. Source RIVM 2022; Statistics Netherlands 2020–2021, aggregated by the authors

26

J. van Haaren and F. van Oort

Table 2 Changes in GRP for the four cities in various stages of the pandemic Region

Early pandemic 2019–2020 (%)

Control and recovery 2020–2021 (%)

Net effect 2019–2021 (%)

Rotterdam

−2.40

6.00

4.00

Amsterdam

−6.70

7.00

0.00

The Hague

−2.00

4.00

3.00

Utrecht

−2.60

5.00

3.00

Source Statistics Netherlands, 2021

pandemic) and 2021. Within the context of the four largest cities, the pandemic seems to have favored the development of cities with lower services-oriented agglomeration and more knowledge-intensive economies (e.g., Utrecht), allowing these cities to catch up with Amsterdam. These results suggest a slowdown of economic activity in services that require in-person interactions in Amsterdam, while the sector structures in Rotterdam and The Hague ensured continued activity and Utrecht’s knowledge economy enabled an easier than expected switch to remote work compared to that prior to the pandemic.

7 The Netherlands as a Special Case? Our results are preliminary in the sense that the full effects of the COVID-19 pandemic can only be analyzed in hindsight. However, our results provide a first indication for the direction of such a development in the Netherlands. Our results suggest that density does not have a strong effect on contagion, but low density in the context of the Netherlands is distinct from low density in other larger, less urbanized countries. Therefore, it is valuable to replicate the analysis with data from other countries with stronger variation between NUTS-3 areas. A similar argument can be made with respect to our case study on the four largest cities in the Netherlands. These cities are almost connected into one single urban agglomeration, locally called the “Randstad,” a polycentric urban region with a green area at its center surrounded by four large and many smaller agglomerations (Van Oort et al., 2010). Therefore, our results should be considered within this context. We preliminarily tested whether results from U.S. oriented literature hold in the distinct context of the Netherlands. The similarities and differences we identify may be instructive for identifying the underlying mechanisms of the processes discussed in the literature review to our chapter.

The Resilience of Cities to COVID-19: A Literature Review …

27

8 Conclusion and Research Agenda This chapter lists what is known about (post) pandemic impacts in urban regions. The chapter investigates how urban agglomeration and accessibility advantages affect the spread of COVID-19 and how similar these conditions are to the rapid spread of information. We discussed how cities may be affected by the COVID-19 pandemic in the medium to long run from the perspective of urban economics, focusing on the transmission of knowledge. An overview of recent thoughts on the impact of the pandemic on urban economies is provided. We then tentatively assessed whether a knowledgebased sector structure, COVID-19 incidence, and the ability to work remotely coincide at regional levels in the Netherlands, as argued by Glaeser and Cutler (2021). A special focus is on the four largest cities in the Netherlands (Amsterdam, Rotterdam, The Hague, and Utrecht), in which we studied how the sector composition in these cities affects the ability of the workforce to switch to working from home (moving away from the highly prized interaction environments), whether this enables people to avoid infection, and how regional domestic products developed during the pandemic. Our results suggest that although density, a sector structure favoring knowledgeintensive business services, and the ability to work from home go hand in hand in urban regions in the Netherlands, this does not ensure against contagion in a pandemic. While we found no link to hospitality services in our analysis, we recognize that these were highly regulated over the past two years. We did find that low density is no guarantee for low contagion in the context of the Netherlands. Our indicative analysis shows that the conditions for a spiky economic geography (e.g., density, knowledge intensity, and interaction) are not necessarily at odds with a post-COVID urban economy. We suggest that a likely outcome of the pandemic is an increase in the spikiness of the economic geography of innovation and that the impulse given to digitization further increases the spatially selective transmission of information. To a tacit degree, knowledge can be transmitted, as new levels of digital communications that foster more personal forms of communication, including dialog (Castaneda & Toulson, 2021), are now accessible, thereby reducing transmission costs. However, despite early signs in San Francisco and New York, this does not lead to a structural death of distance, but quite the opposite: reduced transmission costs suggest, in line with what we know of previous waves of digitization, that “truly” tacit knowledge that offers competitive advantage becomes scarcer and that urban locales that still offer efficiency advantages in the transmission of such knowledge are becoming scarcer as well. A key question for the future is whether the conditions that define where these spikes occur have remained the same or were altered in subtle ways. In addition, the question is put forth that as “truly tacit” knowledge becomes scarcer, it may also become relevant for a smaller subset of sectors in the economy. This requires further research. In terms of urban and regional resilience, the pandemic seems to demonstrate that an overspecialization in the services industry poses significant risks if there are forces at work that limit human interaction and mobility. Cities with more diverse economic structures, including manufacturing, logistics, health care, and government, were

28

J. van Haaren and F. van Oort

apparently more resilient during the pandemic. Perhaps analogous to the risks faced by highly specialized industrial cities of the twentieth century, a narrow specialization in specific sectors of the services economy may prove economically risky in the twenty-first century (Frenken et al., 2007). Conceptualization about COVID-19 incidence, working from home, knowledge intensity, and regional productivity is not readily available and is difficult to assess in practice. Therefore, a research agenda is needed to better study economic resilience in the future. Our literature review and our empirical exercise show that sector and occupational structure dictate the ability to work remotely. As such, we expect the largest changes to occur in those cities that have the largest shares of knowledge intensive business services. We suggest that further study is needed to assess how remote work affects knowledge-intensive business services, particularly in the form of the transmission of various degrees of tacit knowledge. Lockdown conditions may have disrupted both the transmission of disease and knowledge, but human interaction is not limited to the workplace or to the sharing of tacit knowledge. The identification of the resilience of workers and firms requires detailed study into the question of whether workers in knowledge-intensive business services are at a higher risk of contagion now that restrictions have been relaxed. Conversely, our overview also suggests that study is warranted into the question of whether the transmission of tacit knowledge in firms that have adopted more intense remote work regimes is lower compared to those firms that have worked more on premise. This is a long-term undertaking, as it requires assessing microlevel innovation and productivity outcomes of the next years in counterfactual regimes of firms. Complementarily, the reliance of local economies on global value chain relations also needs to be accounted for when determining regional economic resilience (Thissen et al., 2022).

References Agostino, D., Arnaboldi, M., & Diaz Lema, M. (2021). New development: COVID-19 as an accelerator of digital transformation in public service delivery. Public Money & Management, 41(1), 69–72. https://doi.org/10.1080/09540962.2020.1764206 Autor, D., Levy, F., & Murnane, R. (2003). The skill content of recent technological change: An empirical exploration. The Quarterly Journal of Economics, 118(4), 1279–1333. Autor, D., Mindell, D. A., & Reynolds, E. B. (2022). The work of the future: Building better jobs in an age of intelligent machines. The MIT Press. Balland, P. A., Jara-Figueroa, C., Petralia, S. G., Steijn, M. P. A., Rigby, D. L., & Hidalgo, C. A. (2020). Complex economic activities concentrate in large cities. Nature Human Behavior, 4, 248–254. https://doi.org/10.1038/s41562-019-0803-3 Barrero, J. M., Bloom, N., & Davis, S. J. (2020). COVID-19 is also a reallocation shock, Special edition of Brookings Papers on Economic Activity (Vol. 50). Barrero, J. M., Bloom, N., & Davis, S. J. (2021) Why working from home will stick, NBER Working Paper 28731. https://doi.org/10.3386/w28731

The Resilience of Cities to COVID-19: A Literature Review …

29

Bartik, A. W., Bertrand, M., & Cullen, Z. (2020). The impact of COVID-19 on small business outcomes and expectations. PNAS, 117(30), 17656–17666. https://doi.org/10.1073/pnas.200699 1117 Beno, M. (2021). Analysis of three potential savings in E-working expenditure. Frontiers in Sociology, 6, 675530. https://doi.org/10.3389/fsoc.2021.675530 Beno, M., & Hvorecky, J. (2021). Data on an austrian company’s productivity in the pre-covid-19 era, during the lockdown and after Its easing: To work remotely or not? Frontiers in communication. https://doi.org/10.3389/fcomm.2021.641199 Bernela, B., Ferru, M., & Rallet, A. (2019). The impact of digital technologies on perceptions of proximity, ffhal-02053306. https://hal.archives-ouvertes.fr/hal-02053306 Bloom, N., Davis, S. J., & Zhestkova, Y. (2021). COVID-19 shifted patent applications toward technologies that support working from home. AEA Papers and Proceedings, 111, 263–266. https://doi.org/10.1257/pandp.20211057 Boyle, M., & Rockeman, O. (2022). New Yorkers plan to cut time spent in the office by half, survey finds. Bloomberg City Lab. Breznitz, D. (2021). Innovation in real places. Strategies for prosperity in an unforgiving world. University Press. Cairncross, F. (1997). The death of distance: How the communications revolution is changing our lives. Harvard Business School Press. Castaneda, D. I., & Toulson, P. (2021). Is it possible to share tacit knowledge using information and communication technology tools? Global Knowledge, Memory and Communication, 70(8/9), 673–683. https://doi.org/10.1108/GKMC-07-2020-0102 Chan, C. M. L., Teoh, S. Y., Yeow, A., & Pan, G. (2018). Agility in responding to disruptive digital innovation: Case study of an SME. Information Systems Journal, 29(2), 436–455. https://doi.org/ 10.1111/isj.12215 CNN. (2022). These companies decided to go fully remote—Permanently. Retrieved January 27, 2022, from https://edition.cnn.com/2022/01/27/success/full-time-remote-decision-pandemic/ index.html Cone, L., Brøgger, K., Berghmans, M., Decuypere, M., Förschler, A., Grimaldi, E., Hartong, S., Hillman, T., Ideland, M., Landri, P., van de Oudeweetering, K., Player-Koro, C., Bergviken Rensfeldt, A., Rönnberg, L., Taglietti, D., & Vanermen, L. (2021). Pandemic acceleration: Covid19 and the emergency digitalization of European education. European Educational Research Journal (online First). https://doi.org/10.1177/14749041211041793 Correia, S., Luck, S., & Verner, E. (2020). Pandemics depress the economy, public health interventions do not: Evidence from the 1918 Flu. SSRN: 3561560. https://doi.org/10.2139/ssrn.356 1560 Crawford, J., Butler-Henderson, K. , Rudolph, J., Malkawi, B., Glowatz, M., Burton, R., Magni, P., & Lam, S. (2020). COVID-19: 20 countries’ higher education intra-period digital pedagogy responses. Journal of Applied Learning & Teaching, 3(1), 1–20. https://doi.org/10.37074/jalt. 2020.3.1.7 De Lucas Ancillo, A., del Val Núñez, M. T., & Gavrila Gavrila, S. (2021). Workplace change within the COVID-19 context: A grounded theory approach. Economic Research-Ekonomska Istraživanja, 34(1), 2297–2316. https://doi.org/10.1080/1331677X.2020.1862689 Drucker, P. (1999). Knowledge-worker productivity: The biggest challenge. California Management Review, 41(2), 79–92. https://doi.org/10.2307/41165987 DVJ Insights. (2021). Campagne-effectonderzoek: Thuiswerken. https://www.rijksoverheid.nl/bin aries/rijksoverheid/documenten/rapporten/2021/07/30/rapportage-campagne-effectonderzoekthuiswerken/Rapportage-campagne-effectonderzoek-Thuiswerken.pdf Eltarabily, S., & Elgheznawy, D. (2020). Post-pandemic cities—The impact of COVID-19 on cities and urban design. Architecture Research, 10(3), 75–84. https://doi.org/10.5923/j.arch.202010 03.02

30

J. van Haaren and F. van Oort

Farag, S., Weltevreden, J., van Rietbergen, T., Dijst, M., & van Oort, F. G. (2006). E-shopping in the Netherlands: Does geography matter? Environment and Planning B, 33(1), 59–74. https:// doi.org/10.1068/b31083 Faraj, S., Renno, W., & Bhardwaj, A. (2021). Unto the breach: What the COVID-19 pandemic exposes about digitalization. Information and Organization, 31(1), 100337. https://doi.org/10. 1016/j.infoandorg.2021.100337 Financial Times. (2021). PwC tells 40,000 US staff they can work remotely permanently. Retrieved October 1, 2021, from https://www.ft.com/content/70ce4afc-395b-48ea-88cc-7d0036b41b26 Florida, R. (2005). The world in numbers: The world is spiky. The Atlantic Monthly, 48–51. Florida, R., Rodríguez-Pose, A., & Storper, M. (2021). Cities in a post-COVID world. Urban Studies (online First). https://doi.org/10.1177/00420980211018072 Frenken, K., van Oort, F. G., & Verburg, T. (2007). Related variety, unrelated variety and regional economic growth. Regional Studies, 41, 685–697. https://doi.org/10.1080/00343400601120296 Friedman, T. L. (2005). The world is flat: A brief history of the twenty-first century. Farrar, Straus and Giroux. Gabryelczyk, R. (2020). Has COVID-19 accelerated digital transformation? Initial lessons learned for public administrations. Information Systems Management, 37(4), 303–309. https://doi.org/ 10.1080/10580530.2020.1820633 Gavilan, D., Balderas-Cejudo, A., Fernández-Lores, S., & Martinez-Navarro, G. (2021). Innovation in online food delivery: Learnings from COVID-19. International Journal of Gastronomy and Food Science, 24, 100330. https://doi.org/10.1016/j.ijgfs.2021.100330 Glaeser, E. L. (2011). Triumph of the city: How our greatest invention makes us richer, smarter, greener, healthier and happier. The Penguin Press. Glaeser, E. L., & Cutler, D. (2021). Survival of the city: Living and thriving in an age of isolation. Penguin Press. Glaeser, E. L., Kolko, J., & Saiz, A. (2001). Consumer city. Journal of Economic Geography, 1(1), 27–50. https://doi.org/10.1093/jeg/1.1.27 Golinelli, D., Boetto, E., Carullo, G., Nuzzolese, A. G., Landini, M. P., & Fantini, M. P. (2020). Adoption of digital technologies in health care during the COVID-19 pandemic: Systematic review of early scientific literature. Journal of Medical Internet Research, 22(11), e22280. https:// doi.org/10.2196/22280 Grabner, S. M., & Tsvetkova, A. (2022). Urban labour market resilience during the Covid-19 pandemic: What is the promise of teleworking? Regional Studies (online First). https://doi.org/ 10.1080/00343404.2022.2042470 Griffin, D., & Denholm, J. (2020). This isn’t the first global pandemic, and it won’t be the last. Here’s what we’ve learned from 4 others throughout history. https://theconversation.com/thisisnt-the-first-global-pandemic-and-it-wont-be-the-last-heres-what-weve-learned-from-4-othersthroughout-history-136231 Hamersma, M., De Haas, M., & Faber, R. (2020) Thuiswerken na de coronacrisis, Kennisinstituut voor Mobiliteitsbeleid. https://www.kimnet.nl/binaries/kimnet/documenten/rapporten/2020/ 08/31/thuiswerken-en-de-coronacrisis/KiM+rapport+Thuiswerken+en+de+coronacrisis.pdf Hollander, J. E., & Carr, B. G. (2020). Virtually perfect? Telemedicine for Covid-19. New England Journal of Medicine, 382, 1679–1681. https://doi.org/10.1056/NEJMp2003539 Johnson, S. (2006). The ghost map: The story of London’s most terrifying epidemic—And how it changed science, cities and the modern world. Riverhead Books. Jongen, E., Verstraten, P., & Zimpelmann, C. (2021). Thuiswerken vóór, tijdens en ná de coronacrisis, CPB. https://www.cpb.nl/sites/default/files/omnidownload/CPB-Achtergronddocu ment-Thuiswerken-voor-tijdens-en-na-de-coronacrisis_1.pdf Josten, E., & Merens, A. (2021). Thuis of terug naar kantoor: Plus- en minpunten van thuiswerken voor het welbevinden van werknemers. Sociaal en Cultureel Planbureau. https://www.scp.nl/ binaries/scp/documenten/publicaties/2021/07/14/thuis-of-terug-naar-kantoor/Thuis+of+terug+ naar+kantoor.pdf

The Resilience of Cities to COVID-19: A Literature Review …

31

Kahn, M. (2022). Going remote: How the flexible work economy can improve our lives and our cities. University of California Press. Kamal, M. M. (2020). The triple-edged sword of COVID-19: Understanding the use of digital technologies and the impact of productive, disruptive, and destructive nature of the pandemic. Information Systems Management, 37(4), 310–317. https://doi.org/10.1080/10580530.2020.182 0634 Kudyba, S. (2020). COVID-19 and the acceleration of digital transformation and the future of work. Information Systems Management, 37(4), 284–287. https://doi.org/10.1080/10580530.2020.181 8903 Leamer, E. E., & Storper, M. (2001). The economic geography of the internet age. Journal of International Business Studies, 32(4), 641–665. https://www.jstor.org/stable/3069470 Lupu, V. L. (2017). Teleworking and its benefits on work-life balance. International Multidisciplinary Scientific Conference Social Sciences and Arts SGEM, 1(2), 693–700. https://doi.org/10. 5593/sgemsocial2017/12/S02.087 Michaelson, C., Pratt, M. G., Grant, A. M., & Dunn, C. P. (2014). Meaningful work: Connecting business ethics and organization studies. Journal of Business Ethics, 121(1), 77–90. https://doi. org/10.1007/s10551-013-1675-5 Miladinovic, S. (2020). Is COVID-19 accelerating digitalisation or exposing the digital divide? New Europe, June 25th 2020. https://www.neweurope.eu/article/is-covid-19-accelerating-digita lisation-or-exposing-the-digital-divide/ Mondragon, J. A., & Wieland, J. (2022). Housing demand and remote work, NBER working paper 30041. http://www.nber.org/papers/w30041 Nohria, N., & Eccles, R. (1992). Networks and organizations: Structure, form, and action. Harvard Business School Press. OECD. (2020). Productivity gains from teleworking in the post COVID-19 era: How can public policies make it happen? OECD, September 7th 2020. https://www.oecd.org/coronavirus/policyresponses/productivity-gains-from-teleworking-in-the-post-covid-19-era-a5d52e99/ Piros, c˘a, G. I., S, erban-Oprescu, G. L., Badea, L., Stanef-Puic˘a, M. R., & Valdebenito, C. R. (2021). Digitalization and labor market—A perspective within the framework of pandemic crisis. Journal of Theoretical and Applied Electronic Commerce Research, 16(7), 2843–2857. https://doi.org/ 10.3390/jtaer16070156 PWC. (2020). The costs and benefits of working from home. https://www.pwc.nl/nl/actueel-public aties/assets/pdfs/pwc-the-costs-and-benefits-of-working-from-home.pdf Roper, S., & Turner, J. (2020). R&D and innovation after COVID-19: What can we expect? A review of prior research and data trends after the great financial crisis. International Small Business Journal: Researching Entrepreneurship, 38(6), 504–514. https://doi.org/10.1177/026624262094 7946 Rosenthal, S., & Strange, W. (2004). Evidence on the nature and sources of agglomeration economies. In V. Henderson & J. F. Thisse (Eds.), Handbook of regional and urban economics cities and geography. Elsevier. https://doi.org/10.1016/S1574-0080(04)80006-3 Rosewell, B. (2016). Economic impact of infrastructure. Welsh Economic Review, 24, 29–32. https:// doi.org/10.18573/j.2016.10054 Schram, E. (2022). After the pandemic San Francisco looks like the new Gotham. Het Financiële Dagblad, 18–19. Sheridan, A., Andersen, A. L., Hansen, E. T., & Johannesen, N. (2020). Social distancing laws cause only small losses of economic activity during the COVID-19 pandemic in Scandinavia. PNAS, 117(34), 20468–20473. https://doi.org/10.1073/pnas.2010068117 Soto-Acosta, P. (2020). COVID-19 pandemic: Shifting digital transformation to a high-speed gear. Information Systems Management, 37(4), 260–266. https://doi.org/10.1080/10580530.2020.181 4461 Thissen, M., van Oort, F. G., Weterings, A., Ivanova, O., & Bastiaansen, J. (2022). The wickedness of COVID19 policies: The importance of a regional approach. PBL Netherlands Environmental Assessment Agency.

32

J. van Haaren and F. van Oort

Toffler, A. (1970). Future shock. Random House. Toffler. (1980). The third wave. William Morrow and Company. Van Oort, F. G., Burger, M., & Raspe, O. (2010). On the economic foundation of the urban network paradigm. Spatial integration, functional integration and economic complementarities within the Dutch Randstad. Urban Studies, 47, 725–748. Van Oort, F. G., & Thissen, M. (2021). Networked shocks and regional resilience: Implications from Brexit and the Corona pandemic, Scienze Regionali, 20, 3–24. https://doi.org/10.14650/99722 Weltevreden, J., Atzema, O., Frenken, K., de Kruif, K., & van Oort, F. G. (2008). The geography of internet adoption by independent retailers in the Netherlands. Environment and Planning B, 35, 443–460. https://doi.org/10.1068/b33032 Williamson, B., & Hogan, A. (2020). Commercialisation and privatisation in/of education in the context of Covid-19. Education International. Younas, M., & Bari, M. W. (2020). The relationship between talent management practices and retention of generation ‘Y’ employees: Mediating role of competency development. Economic Research-Ekonomska Istraživanja, 33(1), 1330–1353. https://doi.org/10.1080/1331677X.2020. 1748510

A Subjective Geographer’s Experience of Pandemic and Confidence in Systems of Cities Denise Pumain

In this brief chapter, I dare for the first time to convey a few personal practices and feelings as a geographer regarding the pandemic and its probable effects on further changes in urban systems.

1 Which Data Make Sense for a Geographer Facing Pandemics? From a geographer’s view, the Covid-19 pandemic is striking because it bears little resemblance to the great historical precedents (the plague of the fourteenth century in Europe or the so-called “Spanish flu” of 1918). At first sight it emerged in China, very quickly reached Italy, and then spread in large jumps in just a few weeks to all parts of the world. It triggered several waves of stress on hospital intensive care units with shifts in time and varying amplitudes over months. It flared up and lasted for a long time, despite sometimes drastic restrictions on freedom of movement and technological prowess that led to the discovery and production of several vaccines in less than a year. Like many, I scrutinized the numbers by mentally locating them on the world map. At first, from March to mid-April 2020, I visited almost daily the website dedicated to Covid-19 by John Hopkins University, which offers maps, but the numbers given are raw, the death counts for the US are not aggregated but broken down for each of the 51 states. Getting an overview was difficult. Like other websites, this one provides several daily evolution indicators of the pandemic: The number of newly detected cases, newly hospitalized patients, tests, deaths, and recoveries. Only the number of deaths, which I relate to the population of the countries, seems to me to allow D. Pumain (B) Emeritus Professor, University Paris 1 Panthéon-Sorbonne, Paris, France e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. G. Celbi¸s et al. (eds.), Pandemic and the City, Footprints of Regional Science, https://doi.org/10.1007/978-3-031-21983-2_3

33

34

D. Pumain

international comparisons in terms of measuring the seriousness of the pandemic. Of course, the numbers are uncertain, approximate, and sometimes systematically biased. The United Kingdom and, for a time, France did not add the deaths in care homes to those in hospitals. Belgium, on the contrary, attributed to Covid-19 all the deaths sometimes produced by associated morbidities. After one year, demographers proposed reasoned estimates based on the difference with the mortality of a “normal” year, and also evaluate the number of “avoided” deaths that would have been due to influenza or road accidents, for example. In France, the registration of causes of death is not yet immediately digitized; it is done on paper and is sent from the reporting physicians to the centralized registry office sometimes with a delay of months. For many countries, it will be a long time before the estimation method used by demographers allows for the correction of the orders of magnitude displayed, and many deaths will probably never be attributed to their actual cause. I know these uncertainties, and the under-recording in poor countries and remote areas. I have no particular fascination with the deaths that normalize our awareness of the gravity of disasters, but there are still many unknowns throughout 2020 about the other questions that everyone is asking, about the biology of the virus, the effectiveness of treatments, the policies to be held to contain the epidemic, the arbitrariness of choices tending to preserve human lives by reducing freedoms, or to favor the economy that also contributes to sustaining life, the unequal qualities of life, confined or not… The litany of numbers however allows to start trying to understand, while also trying to foresee, according to the relations between what they display and what our knowledge allows us to expect. We need the world scale for the comparisons to start making sense. So I draw up my tables of the top 30 or 50 affected countries to try to know at least how the epidemic travels. From mid-April 2020 onwards, I turned to the Worldometer (2021) website,1 which has two advantages: First, it provides the number of deaths per capita (the case-fatality rate) and not just a raw total—as I write this, at the end of 2021, journalists still sometimes draw up lists of the countries that have “given” the most to the pandemic, without taking into account their inequalities in size. At the end of April 2021, I still hear several times that India had just beaten a world record with more than 350,000 contaminations in 24 h—without remembering its almost 1.4 billion inhabitants. “Do not compare the incomparable” was one of the leitmotifs of the geographer Philippe Pinchemel’s teachings that I have not forgotten. The second advantage of Worldometer is that it displays graphs of the evolution of the epidemic over time, which it proposes to see in two versions: Arithmetic and semi-logarithmic. I spent many hours to praise the visual interest of this semi-log graph to students! It allows you to compare curves, because if you put the time on the arithmetic scale and the variable on the logarithmic scale, a constant rate change, increasing or decreasing, becomes a straight line, and the slope of any line segment on this graph represents an identical rate of 1

It remains difficult to identify the origin of this website, which currently depends on a company registered in the state of Delaware in the United States, named Dadax LLC, and which would be animated by an international collective of researchers and others.

A Subjective Geographer’s Experience of Pandemic and Confidence …

35

growth, whether the curve shows high or low values.2 One can then compare the evolution of the disease in Luxembourg and India, in South Africa and China. And if we scale the axes properly, with the real values of the variable and not their transcription in logarithms (which unfortunately our mathematical or physics colleagues do too often), we know both the orders of magnitude and their relative variation. In the social sciences, demography or geography, these relative variations are called “rates”. Reasoning about rates is inevitable, and elementary, because we know that most of the quantities constructed to describe the social are likely to be, at least to a first approximation, proportional to the population.

2 Towards a Geographical Understanding of the Pandemic The geographer that I am walks her hypotheses on the map of the world, guessing very quickly that it is multi-factorial situations that explain the spread. In the hierarchy of factors, the candidate for the very first place is certainly the opening to circulations, especially international, or should we say “global” so much the situations of the hot spots of contamination join some of the hubs of globalization. This is not a scoop: As early as the 1950s, the geographer Torsten Hägerstrand (1952) had put at the center of his simulations of the geographic diffusion of innovations an “average information field”, which Peter Gould (1993) had illustrated with the hollow example of the Amish county of Lancaster, which hindered the spread of AIDS in the 1980s in North America. The networks that add their connectivity to the contiguity of geographical regions complicate the maps of a spread that no longer operates only in the vicinity but also in distant leaps guided by the hierarchy of the places of contact. But there is no fatality, all world’s cities with very large airports (these “nonplaces” according to the anthropologist Marc Augé) are not affected at the same time or with the same intensity; it is rather the chance that seems to regulate the irruption of contaminations. Gatherings of people from different residential locations in the same place also bring their share of “clusters” to reinforce the hypothesis of propagation through international exchanges. The hypothesis of high densities, on average higher in large cities, is also put forward and deserves attention, but does not play with such strong determinism. Conversely, the effectiveness of “barrier measures” was gradually confirmed throughout the year 2020, despite the denials of Matamorous (i.e. braggart) politicians who were soon caught up and denied by the outbreaks. The acceptability of tracing, isolation and containment measures is becoming a strong concept in the social sciences, as we try to understand the capacities of China and Korea to reduce and control the ravages of the pandemic compared to what the populations of the Western world would be able to endure. But we can also observe that the experience and the ability to adapt and take precautions play an important role in controlling the severity of the epidemic, since the previous pandemics (SARS 2

Olivier Finance (2021) has made a nice pedagogical exercise of reading the curves of evolution of the number of deaths, gross and cumulated.

36

D. Pumain

in 2003, in particular). It is probably also due to the preventive habits adopted by the populations that Africa has not been, until now, the continent most affected by the pandemic, contrary to what I had anticipated. I don’t have the means to test these hypotheses, from where I am now. But geographer colleagues have been able to put them together in simulation models, and to propose to the general public tools that are simple to manipulate, with explanations (Collective, 2020). Modelling in this way avoids focusing too much on the illusory search for the patient zero, which for a while held the attention of many commentators. We hope that they are now more convinced of this false security of genealogical research. The unfolding of a sequence of events, the recounting of a succession of facts, is essential to scientific construction, but is by no means the only path toward explanation. Simulation makes it possible to understand that the same set of factors and parameters is likely to generate a wider set of possible evolutions than the one that, by accident, has occurred, even indurated by path dependence effects (which I sometimes call “historical chaining”) in the dynamics of complex systems. Other epidemiologists have produced analyses of the uneven spread of the pandemic, across UK regions (Goriely & Grindrod, 2021), or have mapped population movements between cities and regions in India during the containment period based on telephone records (Denis et al., 2020). The city-countryside and interregional solidarities revealed by these maps are spectacular but hardly surprising to geographers familiar with India. Another original work explores the geographical configurations of the research sites from which numerous publications on Covid-19 have emerged (Maisonobe et al., 2020), showing the abundant interweaving of the networks that make up the research, between states or between cities of the world. It will be a long time before we have enough data and time to attempt syntheses, measure the relative importance of factors and weigh the effects of policy measures in different regional contexts around the world (Gilbert et al., 2020; Colizza et al., 2021). In its broad outlines, the pandemic corresponds well to the already known diffusion processes; it is the details of its localizations that we must try to explain. A better understanding of these processes will allow us to adjust prevention measures.

3 Can a “Break with the World Before” Be Predicted? At the middle of July 2022, Worldometer or John Hopkins Coronavirus resource center report about 540 million contaminations worldwide and 6.4 million deaths. Underestimates are likely: over time, regions that seemed to be spared, such as Central and Eastern Europe, appear to be among the most severely affected, and others will probably emerge: Thirteen countries among the CEEC (Central and Eastern European Countries) top with death rates close to 3000 per million inhabitants in the list of the 20 to date more impacted countries (among which Peru above 6000, Brazil 3132, Argentina 2806 and USA around 3030). The scale and brutality of the tragedy can be measured first of all at the individual and family level. But what lessons will

A Subjective Geographer’s Experience of Pandemic and Confidence …

37

be learned collectively, socially, from this unprecedented episode of a truly global pandemic? Can we believe that “nothing will ever be the same again”? I have already expressed myself in two editorials on these often supposed changes of perspective, which are only intuitions that the future will perhaps contradict (Pumain, 2020a, b). A first intuition is that of the greater diffusion within the society of an increased familiarity with representations on a global scale. Not always well enough mobilized in public communications, knowledge of how the same problem is approached elsewhere always brings a salutary perspective to policy discussions. Another intuition is that a mobilization against an adversary, no longer a rival or scapegoat, but one without a precisely known origin and common to all humanity, can undoubtedly encourage imaginations and practices of solidarity and compassion. That certainly does not break with the increased inequalities or previous conflicts, but opens up perspectives towards actions that were previously desired but remained unlikely. Thus, the pandemic could be credited with measures such as the one recommended by the WHO for the distribution of the COVAX vaccine to all countries. The disproportionate cost of this same pandemic for public finances is partly at the origin of the announcement by Europe and then the United States of the resumption of international negotiations to harmonize the taxation of companies, in particular those that have made enormous profits from the forced immersion of populations in the world of remote communications in digital form, for activities that are not only recreational. Finally, the sudden slowing down of the mobility that was exponentially growing at the beginning of the twenty-first century, induced by confinement measures, closing of borders and quasi halt of air communications, has made us reflect on the strong contribution to climate change of this exacerbated mobility. Indeed, the impact of mobility on climate change is not limited to the greenhouse gases that it directly generates. More broadly, high mobility results from the spatial organization of value chains that have been built without taking into account environmental costs, with the associated effects of deforestation, excessive specialization in unsustainable monocultures that penalize local populations, over-exploitation of resources and creation of all kinds of nuisances. Trying to bring order to this geographical complexity by optimizing the design of these value chains according to the criteria and technologies of the moment would be illusory and undoubtedly counterproductive. But correcting some of the most glaring anomalies would already be a good step towards what is known as the ecological transition. The withdrawal on oneself and on the home has undoubtedly widened the attention paid to the environment and the quality of living conditions, when they have not made them absolutely unbearable for the poorest. The optimistic side of the argument was that “we can hope for a greater attention in the future to the urgency of the ecological transition”. The intensified awareness of solidarity of all humans, in front of the ineluctable displacement of the centers of the pandemic and the similarities of the expansion of its deadly waves in all regions of the world, could go in the same direction. Faced with the rise of social inequalities and the triumph of neoliberalism of the previous period, the legitimacy of high salaries and even high profits is being questioned, while the “invisible” and underpaid professions, in personal care, commerce, maintenance, logistics and security, as well as teaching and many other activities, are revealing their indispensable usefulness.

38

D. Pumain

But I immediately added: “However, what we know about the processes of change in geography (Lane et al., 2009; Pumain, 2019) makes us highly doubtful about the possibility of drastic reorganizations” (Pumain, 2020b). Not all scientific certainties percolate through the population with the same ease. When adverse reactions to a vaccine were reported, it became clear how little many people listened to the notion of probability in order to assess the risk. Distrust of urban housing, especially in large cities, has been more successful. Whether it is a question of promoting “the city of the quarter of an hour” (Moreno et al., 2021) or of deploring “the messy space of low modernity”, a fashionable discourse, which has been used for many years, has had more success. Another discourse, which tends to impregnate common sense, announces or advocates the abandonment of cities, or at least the largest ones, in order to find a habitat (or a “living”, as initiated geographers say, sliding the function into the form) closer to rurality in its interactions with ecosystems. With the lockdown, many people have taken a dislike to cities! However, scientists such as Saskia Sassen and Karima Kourtit (2021) are skeptical. Neither do I do believe in a return to small and medium-sized cities. For three centuries, cities have only experienced phases of generalized decline during wars. In France, 1914– 1918 and 1940–1945 are the only two periods of urban exodus that can be observed since census statistics became available. If we were to look at the numbers without criticizing them, we would find another, more anecdotal, reason: Between 1836 and 1841, the population of French cities grew, but the larger they were, the less they grew, and this trend was rather the opposite of all the other intervals observed between censuses thereafter. In fact, there had been a rumor that the 1841 census might have been used for tax purposes, and the rumor seems to have spread more quickly in the larger cities, which probably had more “landlords,” a well-informed social category of the time. Five years later, at the next census, this socio-statistical artifact was forgotten, the underestimation was completely corrected and the population caught up. This anecdote also reveals the power of the circulation of information in and through cities whose actors are networked, the effects of which we have repeatedly demonstrated in terms of capturing innovations and strengthening the inequalities in the systems of cities they form (Pumain, 2021a). The first cities to experience a systematic downward trend were, on average, small towns, often specialized in production that was in recession, or located on the territorial margins away from major traffic routes. Spatial planning policies have sought to counteract this trend, and have contributed to redistributing certain activities, equipping the peripheries, raising income levels and caring for the countryside, but they have not reversed the quantitative inequalities and qualitative differences that networked adaptation processes renew and continue to deepen.

A Subjective Geographer’s Experience of Pandemic and Confidence …

39

Fig. 1 Human development index in cities and the country where they are located Source UN Habitat, 2015

4 Knowledge Based Models for Predicting the Future of Urbanization Is it really necessary to move away from big cities? We have known since Fernand Braudel (1979) that cities invented capitalism, and with Giovanni Botero (1588) that conquering politics, economic rivalry and what is now called urban marketing are processes that are part of the ordinary dynamics of systems of cities, whatever the regions of the world and the ways in which societies are politically and economically organized and controlled. For more than two centuries in developed countries, and for more than half a century in less urbanized countries, these city dynamics have been supported by demographic and economic growth, and have been accompanied by a tendency to concentrate in large urban areas.3 Will this trend continue after the end of the demographic transition, with aging or even shrinking populations, and a possible slowdown in economic growth? Although many argue that digital technologies will lead to decentralization, many other considerations continue to favor concentration, as evidenced by the statistics produced by the United Nations Habitat Division UN Habitat (2015), which show that not only income per inhabitant but also human development indices are significantly higher in large cities than in the countries where they are located (Fig. 1). Are the objectives of sustainable development compatible with the degree of urban concentration that will be reached? Research on city metabolism is progressing, but it is not yet clear whether the pressure on the environment increases with city size, or whether cities achieve economies of scale in this area that reduce the per capita burden. Measures of the exponent of the “scaling laws” that summarize these effects are still very incomplete and uncertain. As for development objectives, it is known with near certainty that the diversity of cities, in size and specialization, is an important condition for maintaining and renewing what makes them live together in mutually interdependent systems of cities (Pumain & Raimbault, 2020).

3

Provided that cities are defined and delineated according to a geographically sound ontology, see Bretagnolle et al., 2008; Rozenblat 2020; Denis, 2020).

40

D. Pumain

While concluding a chapter about geographical concepts that could be shared with demographers, I insisted on this disciplinary view: “For geographers, studying the spatiality of societies implies considering the planet’s organization with human requirements and assumes a clear understanding of settlement’s interactions with their environment. Figure 2 represents a schematic of the evolution of these relationships since the emergence of cities. By definition, cities base their development on networking more distant resources than the villages of the sedentary Neolithic period. However, during the four or five millennia following the emergence of cities in agrarian economies, their development remained strongly constrained by the availability of local resources and was influenced by the climatic and seismic fluctuations of their immediate environment. Besides, conflicts between neighboring cities or invasions of more distant origin often led to destruction, sometimes permanent, at least until the fourteenth century. The full realization of the amplifying role of the networks developed by the cities only emerged with the demographic and economic boom that prepared the first industrial revolution, spreading urbanization and its processes -including capitalism- all over the world in just two centuries. The power of urban networks seemed limitless until we recognized the depletion of specific resources, particularly in energy, or their excessive consumption or degradation, increasing at a higher rate than their renewal. But the established urban networks are now much more stable and universal in their coverage of territories. The interactions that support them and that they ensure should make it possible, in a system powered by science, technology, and territorial intelligence, to facilitate the necessary information exchanges to achieve the “ecological transition” and urban sustainability that are part of the United Nations’ programs (UN Habitat, 2020; Pumain, 2021b, page 9). Considering urban systems as complex systems leads to a different paradigmatic conception than in former classical geography and paves the way for a more intensive use of modelling: “With our models, we invert a sedimentary perspective that conceives geography as the last layer of history. We no longer seek to explain a location, a geographical entity, by taking into account its entire history, by reconstructing its unique genesis. We no longer construct history as a path that we follow backward to find an explanation in the biography of a place. Initially, we have been able to hypothesize that in some features, geographic objects represent specific achievements, among a range of possible achievements, of general dynamic processes with models that concede the existence of multiple solutions to the same dynamic process (Allen, 1997; White et al., 2015). Subsequently, we could postulate theoretical hypotheses describing the interaction patterns between cities that were tested through multiagent simulation models (Pumain, 2012; Pumain, 2021b, page 9–10). Since our first Simpop model (Bura et al., 1996) we experimented with several software. It is only when using powerful computational methods, including genetic algorithms and intensive distributed computing that we arrived at robust results regarding our confidence in the conclusions that were derived from the model. In a paper whose title emphasizes the very heavy computational cost of the experiment –half a billion simulation- (Schmitt et al., 2015) we considered that at last it is possible to bring a scientific “proof” in social sciences, solving the indetermination problem of many

A Subjective Geographer’s Experience of Pandemic and Confidence …

41

Fig. 2 Urban networking according to an evolutionary theory of urban systems

models attempting at reconstructing urban trajectories, through having demonstrated that all the included parameters were not only “sufficient” but as well “necessary”. Besides, this rather satisfactory result only validates the rules that were included in the frame of our model, but not that frame itself. Thus, the knowledge developed in GeoDiverCity seems to me to be valid for exploring trajectories of cities in different parts of the world, over durations of a few decades, based on information about their initial size and their main socio-economic functions resulting from their past adaptations to major innovation cycles. The construction of models that are constantly improved by adding empirical data and methodological innovations within a program whose name deliberately highlights urban geodiversity4 also contributes to identifying geographical singularities in the “classical” sense at different spatial and temporal scales. The inter-urban interactions are difficult to observe directly in the medium and long term but the synthetic reconstruction that was generated in silico with multi-agent models (Cottineau et al., 2015a, b; Schmitt et al., 2015) is consistent with available empirical observations of changes in urban systems of different types (Pumain et al., 2015). This knowledge did not allow me to anticipate a priori many of the major impacts caused by the shock of the pandemic. The greater availability of data at fine scales of time and space offers the hope of progress in the medium term. For example, 4

GeoDiverCity, ERC Advanced Grant 2011–2016, PI Denise Pumain. Empirical data bases were built that cover long periods of time for several large regions of the world (USA, Europe and all BRICS countries).

42

D. Pumain

thanks to the pooling of data on a global scale, some modelers have been able to analyse the spatio-temporal spread of the Covid-19 epidemic by simulating it using fractal models of spatial networks of individual and inter-city interactions (Abbasi et al., 2020). Other scientists propose “digital twins” of the UK territory for helping more precise locational analysis during the propagation of epidemics in a microsimulation model (Spooner et al., 2021). These authors were able to estimate the effect of the date chosen for the lockdown on the peak in COVID-19 cases. They claim that such a model is “crucial in gaining a greater understanding of the effects of policy interventions in different areas and within different populations”. I would better say that there is still a long way to go before we can provide planners with robust and reliable models, capable of experimenting with scenarios that enable them to make informed choices to anticipate the future of cities. One may here echo the two claims made by Helen Couclelis (2021) “How can the new science of information lead to the new science of cities? and how can big data lead to actionable wisdom under conditions of pervasive uncertainty and complexity?”. The question of the future of cities in our current context entails answering the following question: Will the inter-urban dynamics observed so far continue in a context of slowed or even negative demographic growth and more restricted economic growth? To what extent can it be affected by external shocks such as a pandemic or unexpected large geopolitical conflicts? I keep repeating that “Our research on the role and power of interactions in the morphogenesis of systems of cities tends to predict continuity of their hierarchical organization and a progression of their inequalities that only war periods have slowed down until now (Cura et al., 2017). Only massive and costly political interventions in small and medium-sized cities could lead to a complete reversal of the current dynamics. The evolutionary theory considers urban systems as complex, multi-scalar, and open social adapters. While always requiring a time and space context, the dynamics of these urban systems include regularities that allow them to be partly comparable and predictable across systems and time scales. Micro-geographic level interactions, resulting from the multiple interventions of many actors, produce the “behaviors” of cities and systems of cities at meso and macro-geographic levels due to the complex and reflexive feedback introduced by agents’ activities. Nevertheless, these individuals and institutions must be informed of this urban collective dynamic’s knowledge to benefit from the collective territorial intelligence they represent and to succeed in the essential adaptations required by our time’s ecological and social tensions” (Pumain, 2021b, page 10). The web is woven around the world by cities of all sizes, and functional and cultural specialization materializes the work accumulated on the earth for centuries to exploit and develop its resources, historically in confrontations between cities and then between states of the world than large transnational firms. Today, it is an instrument that can be used to share innovations and cooperate in emulation. But this is a real research issue: Intervening in a complex system to transform it requires a good level of science!

A Subjective Geographer’s Experience of Pandemic and Confidence …

43

References Abbasi, M., Bollini, A. L., Castillo, J. L. B., Deppman, A., Guidio, J. P., Matuoka, P. T., & Varona, A. R. P. (2020). Fractal Signatures of the COVID-19 spread. Chaos, Solitons & Fractals, 110119. Allen P. (1997) Cities and regions as self-organized systems. Models of complexity. Gordon and Breach, Amsterdam. Berque A. (2021), Mésologie urbaine. Editions Terre Urbaine, l’Esprit des villes, Paris. Botero G. (1588). Delle cause della grandezza et magnificenza delle città. Tre libri. New edition in 1598 inside the book “La Ragion di Stato” (7th ed.) in Venice, and a cura di Luigi Firpo, Unione Tipografico Editrice Torinese in 1948, pp. 341–409. Braudel F. (1979). Civilisation matérielle, économie et capitalisme (XVe – XVIIIe siècle), Armand Colin, Paris. Bretagnolle A., Giraud T., Mathian H. (2008) Measuring urbanization in United States, from the first trading posts to the Metropolitan Areas (1790–2000), Cybergeo : European Journal of Geography, 427, online. http://journals.openedition.org/cybergeo/19683 Bretagnolle, A., & Pumain, D. (2010). Simulating urban networks through multiscalar space-time dynamics (Europe and United States, 17th—20th centuries). Urban Studies, 47(13), 2819–2839. Bura, S., Guérin-Pace, F., Mathian, H., Pumain, D., & Sanders, L. (1996). Multi-agent systems and the dynamics of a settlement system. Geographical Analysis, 28(2), 161–178. Colizza, V., Grill, E., Mikolajczyk, R., Cattuto, C., Kucharski, A., Riley, S., & Fraser, C. (2021). Time to evaluate COVID-19 contact-tracing apps. Nature Medicine, 27(3), 361–362. Collective (2020). https://covprehension.org/en/about/ (consulted December 17th 2021) Cottineau, C., Chapron, P., and Reuillon, R. (2015a). Growing models from the bottom up. An evaluation-based incremental modelling method applied to the simulation of systems of cities. Journal of Artificial Societies and Social Simulation, 18(4), 1–9. Cottineau, C., Reuillon, R., Chapron, P., Rey-Coyrehourcq, S., & Pumain, D. (2015b). A modular modelling framework for hypotheses testing in the simulation of urbanisation. Systems, 3(4), 348–377. Couclelis, H. (2021). Conceptualizing the city of the information age. In Urban Informatics. Springer, Singapore, (pp. 133–145). Cura, R., Cottineau, C., Swerts, E., Ignazzi, C. A., Bretagnolle, A., Vacchiani-Marcuzzo, C., & Pumain, D. (2017). The old and the new: Qualifying city systems in the world with old models and new data. Geographical Analysis., 49(4), 363–386. https://doi.org/10.1111/gean.12129 Denis E. (2020), Population, Land, Wealth and the Global Urban Sprawl. Drivers of urban built-up expansion across the world from 1990 to 2015, in Theories and models of urbanization. Springer, 235–258. Denis E., Telle O., Benkimoun S. (2020) Mapping the lockdown effect in India: how geographers can contribute to tackle Covid-19 diffusion. https://cybergeo.hypotheses.org/category.englishconversations Finance O. (2021) Covid-19: L’empire des courbes. https://cybergeo.hypotheses.org/ 3 avril 2021. Gilbert, M., Pullano, G., Pinotti, F., Valdano, E., Poletto, C., Boëlle, P. Y., & Colizza, V. (2020). Preparedness and vulnerability of African countries against importations of COVID-19: A modelling study. The Lancet, 395(10227), 871–877. Gould, P. (1993). The slow plague: a geography of the AIDS pandemic. Blackwell Publishers. Goriely, A., & Grindrod, P. (2021). Lessons to be learned from the Covid-19 Experience in the UK. Rsearchgate. Hägerstrand, T. (1952). The propagation of innovation waves. Lund Studies in Geography: Series B, Human Geography, 4. John Hopkins (2020) Covid-19 map. Coronavirus resource center. https://coronavirus.jhu.edu/map. html (consulted December 17th 2021) Lane D., Pumain D., van der Leeuw S., West G. (eds.), (2009). Complexity perspectives on innovation and social change, ISCOM, Springer, Methodos Series 7.

44

D. Pumain

Maisonobe M. Jegou L. Eckert D. Grossetti M. Milard B. Cabanc G. (2020) Were do Covid19 research come from? https://geoscimo.univ-tlse2.fr/blog/2020/06/15/where-do-covid-19-resear ches-come-from/ (consulted December 17th 2021). Moreno, C., Allam, Z., Chabaud, D., Gall, C., & Pratlong, F. (2021). Introducing the “15-Minute City”: Sustainability, Resilience and Place Identity in Future Post-Pandemic Cities. Smart Cities, 4(1), 93–111. Neal, Z., & Rozenblat, C. (Eds.). (2021). Handbook of cities and networks. Edward Elgar. Pumain D. (2012), Multi-agents system modelling for urban systems : the series of SIMPOP models, in Heppenstall A. J., Crooks A.T., See L.M., Batty M. (eds), Agent-based Models of Geographical Systems. Springer, Population Studies, 721–738. Pumain, D. (2019), Geographical Modeling: Cities and Territories. ISTE Ltd, London, and Wiley, New York. Pumain D. (2020a) Geographical confinement or the virtues of an experiment, Cybergeo, European Journal of Geography, online, http://journals.openedition.org/cybergeo/34659. Pumain D. (2020b) Learning to adapt, Cybergeo, European Journal of Geography, online http:// journals.openedition.org/cybergeo/35362 Pumain D. (2021a), Co-evolution as the secret of urban complexity in J. Portugali (ed) Handbook on Complexity and Cities. Edward Elgar Publishing, pp 136–153. Pumain D. (2021b) Spatial analysis: a fertile ground for demography, in Degioanni A., Herrscher E., Naji S. (dir.), Journey of a committed paleodemographer. Farewell to Jean-Pierre BocquetAppel, Presses Universitaires de Provence, coll. Préhistoire de la méditerranée, Aix-en-Provence, pp. 7–14. Pumain D. Raimbault J. (2020), Conclusion: Perspectives on urban theories, in Pumain D. (ed.) (2020) Theories and Models of Urbanisation. Springer, Lecture Notes in Morphogenesis, 303-330. Pumain D., Swerts E., Cottineau C. Vacchiani-Marcuzzo C., Ignazzi A., Bretagnolle A., Delisle F., Cura R., Lizzi L, Baffi S. (2015) Multi-level comparison of large urban systems. Cybergeo, European Journal of Geography 706, http://cybergeo.revues.org/26730, https://doi.org/10.4000/ cybergeo.26730. Raimbault J. (2021). Modeling the co-evolution of cities and networks, in Neal Z. & Rozenblat C. (eds) Handbook of Cities and Networks, Edward Elgar. Raimbault J. Pumain D. (2021), Spatial dynamics of complex urban systems within an evolutionary theory frame, in A. Reggiani, L.A. Schintler, D. Czamanski, R. Patuelli (eds) Handbook on Entropy, Complexity and Spatial Dynamics. The Rebirth of Theory? Edward Elgar Publishing Ltd, Cheltenham, UK. Rozenblat C. (2020) Extending the concept of city for delineating large urban regions (LUR) for the cities of the world, Cybergeo: European Journal of Geography, 954, http://journals.opened ition.org/cybergeo/35411; Sassen, S., & Kourtit, K. (2021). A Post-Corona perspective for smart cities:‘Should I Stay or Should I Go?’ Sustainability, 13(17), 9988. Schmitt C., Rey-Coyrehourcq S., Reuillon R., Pumain D. (2015) Half a billion simulations, Evolutionary algorithms and distributed computing for calibrating the SimpopLocal geographical model, Environment and Planning B, 42, 2,300-315. Spooner, F., Abrams, J. F., Morrissey, K., Shaddick, G., Batty, M., Milton, R., Dennett, A., Lomax, N., Malleson, N., Nelissen, N., et al. (2021). A dynamic microsimulation model for epidemics. Social Science & Medicine, p. 114461. UN Habitat (2015) Annual Progress report. United nations for a better urban future. https://unhabi tat.org/annual-progress-report-2015 (consulted December 17th 2021) UN Habitat (2020), World cities report, the value of sustainable urbanization. United Nations Human Settlements Programme. https://unhabitat.org/sites/default/files/2020/10/wcr_2020_report.pdf (consulted December 17th 2021). White R. Engelen G and Uljee I. (2015) Modeling cities and regions as complex systems. From theory to planning applications. Cambridge (Mass.) MIT Press. Worldometer (2022) https://www.worldometers.info/coronavirus/ (last consulted July 18th 2022)

City and Regional Demand for Vaccines Whose Supply Arises from Competition in a Bertrand Duopoly Amitrajeet A. Batabyal and Hamid Beladi

1 Introduction We now know—see Chaplin (2020) and Batabyal and Beladi (2022)—that the cause of the severe acute respiratory syndrome or SARS-like illness that subsequently became known as Covid-19 was a novel coronavirus, in particular, the SARS-CoV2. On 30 January 2020, Covid-19 was declared by the WHO to be a Public Health Emergency of International Concern (PHEIC). The first case of Covid-19 arising from local person-to-person spread in the United States (U.S.) was confirmed in mid-February 2020. On 11 March 2020, the WHO declared COVID-19 a pandemic.1 Since that time, the pandemic—which is still not under complete control—has spread throughout the world. A recent study2 points out that the Covid-19 pandemic has led to 6.9 million deaths worldwide which is more than twice the number that has been officially reported. Focusing just on the U.S., this same study estimates that 905,000 people have died of Covid-19 since the start of the pandemic. As pointed out by Branswell (2021), it is worth emphasizing two points about this 905,000 number. First, it is about 61 percent higher than the death estimate of 561,594 provided in 1

We thank the Editors of this volume and three anonymous reviewers for their helpful comments on a previous version of this chapter. In addition, Batabyal acknowledges financial support from the Gosnell endowment at RIT. The usual disclaimer applies. 2 Go to http://www.healthdata.org/special-analysis/estimation-excess-mortality-due-covid-19-andscalars-reported-covid-19-deaths for more details. Accessed on 5 July 2022. A. A. Batabyal (B) Department of Economics, Rochester Institute of Technology, 92 Lomb Memorial Drive, Rochester, NY 14623-5604, USA e-mail: [email protected] H. Beladi Department of Economics, University of Texas at San Antonio, One UTSA Circle, San Antonio, TX 78249-0631, USA e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. G. Celbi¸s et al. (eds.), Pandemic and the City, Footprints of Regional Science, https://doi.org/10.1007/978-3-031-21983-2_4

45

46

A. A. Batabyal and H. Beladi

early May 2021 by the U.S. Centers for Disease Control.3 Second, this number also exceeds the estimated number of U.S. deaths in the so-called Spanish flu pandemic in 1918, which was estimated to have killed approximately 675,000 Americans. Given the significant health, social, and economic impacts of Covid-19, it is perhaps unsurprising to learn that a lot of faith has been placed on fighting this pandemic with a vaccine. As noted by Torreele (2020) and Stiglitz (2021), governments in the U.S., the United Kingdom, China, and Russia have poured large amounts of money into vaccine development. As a result, Katella (2021) rightly notes that the world now has several efficacious vaccines such as the two-shots messenger RNA or mRNA vaccines developed by Pfizer-BioNTech and Moderna and the one-shot vaccine developed by Johnson and Johnson. There is some research on how market competition between firms affects the development of vaccines.4 This notwithstanding, to the best of our knowledge, there is virtually no theoretical research on how alternate economic conditions such as changes in cost uncertainty and changes in the nature of the strategic competition between firms influence the incentives to conduct research and development (R&D) in the context of vaccine development. Given this lacuna in the extant literature, our objective in this chapter is to study a one-period model of an aggregate economy made up of cities and regions that demand vaccines designed to fight a pandemic such as Covid-19. The supply of the vaccines is the outcome of Bertrand’s competition between two firms A and B. The marginal cost of producing the vaccine for both firms is probabilistic and drawn from a uniform distribution. The model is based in part on the discussion of Schumpeterian economic growth in Acemoglu (2009, pp. 458–496). In this setting, we first show how the risk associated with the conduct of R&D affects whether both firms or only one firm ends up conducting R&D in equilibrium. Second, we demonstrate how increased competition influences the incentives to conduct R&D faced by the two firms in our model. The remainder of this chapter is organized as follows. Section 2 delineates our theoretical framework. Section 3 describes the equilibrium pricing strategies of the two firms and then computes their expected ex ante profits. Section 4 permits both firms to undertake risky R&D and then it ascertains the conditions under which only one firm engages in R&D and conditions under which both do. Section 5 introduces a way of mimicking the effect of increased competition and then analyzes the effect of this increased competition on the incentives to conduct R&D faced by the two firms. Finally, Sect. 6 concludes and then discusses three ways in which the research described in this chapter might be extended.

3

Go to https://www.cdc.gov/nchs/covid19/mortality-overview.htm to learn more about this point. Accessed on 5 July 2022. 4 See Sloan and Hsieh (2007), Fu et al. (2012), Gilchrist and Nanni (2013), Kremer et al. (2020), and Martonosi et al. (2021) for more on this literature.

City and Regional Demand for Vaccines Whose Supply Arises …

47

2 The Theoretical Framework Consider economic activities that take place in a single time period in an aggregate economy that consists of a finite number of cities and regions. A region, in the context of this chapter, is a sub-national geographic entity and hence a city is a particular kind of region. There are two vaccine producing firms denoted by A and B in our aggregate economy and both these firms produce a vaccine that we shall think of as being a homogeneous good. For concreteness, the reader may want to think of the two firms as the Pfizer and BioNTech alliance and Moderna and the good they produce as the mRNA vaccine. Since the mRNA vaccines produced by these two firms work very similarly and have comparable effectiveness levels,5 they can, for the purpose of our analysis, be thought of as a homogeneous good.6 The two vaccine producing firms A and B are the two Bertrand duopolists in our model.7 At the beginning of the time period under study, firm A' s marginal cost of production is given by a draw from the uniform distribution [0, c A ] and firm B ' s marginal is given by an independent draw from the uniform distribution [cost of] production 0, c B .8 Both duopolists first observe their costs and then they set their prices. Total demand for the vaccines produced by the two duopolists is generated by the governments of the various cities and regions that comprise our aggregate economy. This total demand or Q is given by the linear demand function.9 Ʌ

Ʌ

5

Go to https://share.upmc.com/2021/01/vaccines-moderna-pfizer/ for more details on this point. Accessed on 5 July 2022. 6 Go to https://www.fraserhealth.ca/health-topics-a-to-z/coronavirus/covid-19-vaccine/mrna#. YsT-pnbMKUc for additional details on this point. Accessed on 5 July 2022. 7 In the remainder of this chapter, we shall use the terms “firm” and “duopolist” interchangeably. 8 The uniform distribution is easy to work with and it is commonly used within many fields in economics to model the stochastic aspects of a variable. This is why we use the uniform distribution in our analysis. See Wanke (2008) for additional details on this point. In addition, we do not use the Weibull distribution because it is analytically more difficult to work with and because the most common applications of the Weibull distribution typically involve the analysis of survival data. See Mudholkar et al. (1996) for more details on this point. 9 An analysis of the competition between two firms to produce a vaccine for use against Covid-19 would be meaningless unless we also model the demand for these vaccines explicitly. That is why we have the total demand function that we do in Eq. (1). In this chapter, we are primarily interested in studying the nature and the properties of the competition between two firms to produce vaccines. That is why we have not imposed additional structure on the total demand function in Eq. (1). That said, two points are now worth emphasizing. First, clearly governments can and do perform a variety of roles in the context of vaccine development. One such role is to purchase the produced vaccines that they then make available to their citizens. This is the role that we model in the present chapter. Other possible roles include granting patents and making advance purchase agreements with the vaccine producing firms. These roles are discussed briefly in Sects. 4.1 and 6 below. Second, one could easily give an explicit spatial structure to this total demand function. Here are two examples to show how one could do this. Suppose the aggregate economy of interest is that of New York state in the United States. New York state is divided into 62 counties. Suppose the demand for vaccines from the ith county, i = 1, . . . , 62, at time t is denoted by qi . Then, aggregating over space, the total demand for vaccines at time t is the sum of the demand in these 62 counties. In this ∑ case, we 62 could write the inverse (and spatial) total demand function as P = H − Q where Q = i=1 qi .

48

A. A. Batabyal and H. Beladi

Q=H−P

(1)

( ) where P is the price and it is understood that the condition H > 2max c A , c B holds. With this description of the theoretical framework out of the way, we now proceed to first determine the equilibrium pricing strategies of the two firms and then we compute their expected ex ante profits.10 Ʌ

Ʌ

3 Pricing and Profits 3.1 Pricing Let us begin our study of the pricing decisions of the two firms by focusing on the firm with the lower cost realization. There is no loss of generality in this approach because, in equilibrium, this will be the only firm that produces the vaccine. That said, what we have to next ascertain is whether this lower cost firm will be able to behave like a monopolist and hence charge the monopoly price or whether this firm will be forced to use limit pricing.11 To determine the monopoly price, our lower cost firm uses the demand function given in Eq. (1) and solves the profit maximization problem max(H − P)(P − c)

(2)

{P}

where c > 0 is the realized marginal cost. Calculus shows that the monopoly price (P M ) is given by the ratio P M (c) =

H +c . 2

(3)

Note that it makes sense to write the monopoly price P M as a function of the marginal cost c because this cost realization is observed by our lower cost firm before it sets its price. ( ) our understanding that H > 2max c A , c B . Also, we know that c ∈ [ Recall ] 0, ci , i = A, B. Using these two pieces of information, we infer that Ʌ

Ʌ

Ʌ

P M (c) ≥ P M (0) =

) ( H > max c A , c B . 2 Ʌ

Ʌ

(4)

Continuing with the first example, suppose we wanted to give a probabilistic flavor to the analysis. ∑62 qi )/62. In this last case, we could write the Then, the average demand per county or q = ( i=1 inverse (and spatial) total demand function as P = H − Q where Q = 62q. 10 In what follows, the model solution techniques we employ are similar to those employed by Peters and Simsek (2009, pp. 287–291). 11 See Acemoglu (2009, p. 419) for a textbook exposition of limit pricing.

City and Regional Demand for Vaccines Whose Supply Arises …

49

Equation (4) tells us that the monopoly price will always be (weakly) higher than the competing duopolist’s marginal cost. This means that charging the monopoly price cannot constitute equilibrium behavior by our lower cost firm because if this firm attempted to charge the monopoly price then its rival could undercut this monopoly price by charging P M (c)− ε for some small ε > 0, and still make positive profits. This line of reasoning eliminates the possibility of the lower cost duopolist charging the monopoly price and therefore this finding tells us that in equilibrium, the lower cost duopolist must limit price. To determine the limit pricing equilibrium, let us begin by assuming, with no loss of generality, that c A < c B . Now observe that in order to ensure that it does not make a loss by producing the vaccine, duopolist B must charge c B in equilibrium. If it charged more than c B then, because c B < P M (c A ), duopolist A will also charge a price higher than c B . This pricing behavior cannot constitute an equilibrium because duopolist B will now want to undercut duopolist A' s price. In this regard, it is worth emphasizing that in our model, the lower cost firm— temporarily assumed to be duopolist A—must capture the total demand for vaccines in the aggregate economy. To see this, suppose that the above point is not the case and hence duopolist A captures only a fraction γ ∈ (0, 1) of the total demand for vaccines. Then, by charging PA (c A , c B ) = c B − ε for some small ε > 0, duopolist A would capture the entire market for vaccines in our aggregate economy. However, since c B − ε < P M (c A ), the revenue function is a decreasing function of ε. This line of reasoning tells us that there is no equilibrium in which PA (c A , c B ) = c B − ε < c B . In turn, this last finding leads to the conclusion that there is no equilibrium in which duopolist A captures only the fraction γ ∈ (0, 1) of the total demand for vaccines when both firms in our model charge c B . Note that when duopolist A captures the entire market for vaccines by charging c B , we have an equilibrium because duopolist B makes zero profit independent of its market share. Finally, the governments of the various cities and regions that are the consumers of the vaccines are indifferent about which firm to buy vaccines from because the produced vaccines are homogeneous.12 We can now conclude this discussion of pricing by pointing out that the equilibrium limit price or P L (c A , c B ) is given by P L (c A , c B ) = max(c A , c B ),

(5)

and that the lower cost firm captures the entire market for vaccines.13

12

If we think of the Pfizer and BioNTech alliance as firm A and Moderna as firm B, then we know that in the real world, it has not been the case that either only the Pfizer and BioNTech alliance or only Moderna has captured the total demand for mRNA vaccines. This discrepancy is most likely explained by the twin facts that we have not modeled advance purchase guarantees–-see Sect. 6 below–-by governments and that our model is static and hence unable to analyze repeated interactions between the Pfizer and BioNTech alliance, Moderna, and the pertinent governments. 13 We can think of the knife-edge case in which, notionally, the two firms A and B charge the same price for vaccines as one in which the two firms share the market for vaccines equally.

50

A. A. Batabyal and H. Beladi

3.2 Profits We now ascertain the expected profits E[πi ], i = A, B obtained by the two firms. In this regard, notice that because we are looking at the ex ante expected profits, the expectations we are considering are unconditional and they are taken over the joint distribution of the two costs c A and c B . Using the law of iterated expectations—see Ross (1996, p. 21)—we deduce that the ex ante profit of duopolist A is given by [ ] E[π A ] = E c B E c A [π A /c B ] .

(6)

To compute the above expectation, we need to consider the two relevant cases in which either c A ≥ c B or c A < c B . Let us focus on the c A ≥ c B case first. Using the properties of the uniform distribution—see Ross (2007, pp. 35–36)—and that of the conditional expectation— see Ross (2007, pp. 105–117)—we reason that Ʌ

Ʌ

Ʌ

Ʌ

Ʌ

Ʌ

∫ c B ⎧∫ Ʌ

[ ( ] ) E π A c A , c B /c A ≥ c B = Ʌ

Ʌ

Ʌ

Ʌ

0

cB

{(c B − c A )(H − c B )}

0

⎫ 1 1 dc B . (7) dc A cB cA Ʌ

Ʌ

After several steps of algebra, the right-hand-side (RHS) of Eq. (7) can be simplified. This simplification gives 3 2 [ ( ) ] c H cB − B . E π A c A , c B /c A ≥ c B = 8c A 6c A Ʌ

Ʌ

Ʌ

Ʌ

Ʌ

Ʌ

Ʌ

(8)

Ʌ

Ʌ

Ʌ

Next, focusing on the c A < c B case, we get ∫ c A ⎧∫ c B Ʌ

( ) ] E π A c A , c B /c A < c B = [

Ʌ

Ʌ

Ʌ

Ʌ

Ʌ

Ʌ

0 Ʌ

⎫ 1 1 {(c B − c A )(H − c B ) dc B dc A . (9) c c A B cA Ʌ

Ʌ

As in the case of the c A ≥ c B case, once again, we can simplify the RHS of Eq. (9), Doing this, we get 2 3 2 [ ( ) ] c c H (c B − c A ) c A c B HcA − A − B . (10) + + E π A c A , c B /c A < c B = 3 24c B 4 2 6c B Ʌ

Ʌ

Ʌ

Ʌ

Ʌ

Ʌ

Ʌ

Ʌ

Ʌ

Ʌ

Ʌ

Ʌ

Ʌ

Equations (8) and (10) together give us the ex ante profit that duopolist A can expect to earn by producing vaccines. Since duopolist B ' s problem is symmetric to that of duopolist A, we can easily write the analogous ex ante profit that duopolist B can expect to earn.14 14

In the real world, from the standpoint of 2021, the Pfizer and BioNTech alliance and Moderna were both expected to make significant profits in 2022 from the manufacture and sale of their

City and Regional Demand for Vaccines Whose Supply Arises …

51

Our next task is to allow both firms to undertake risky R&D and to then determine the conditions under which only one firm engages in R&D and conditions under which both do.

4 Risky R&D 4.1 Firm Payoffs Suppose that the two firms A and B in our aggregate economy [ ] begin their competition in the market for vaccines with the cost distribution 0, c . In addition, suppose that both firms can conduct R&D with fixed cost ζ > 0.15 If they do conduct R&D then with probability θ > 0, their cost distribution changes to [0, c − β] and we assume that c > β. We shall interpret the situation in which the cost distribution changes as one in which a firm is successful in coming up with an innovation that reduces the cost of producing a vaccine. We do not discuss patents explicitly in this chapter but consistent with an observation of Tirole (1988, p. 394), the sort of R&D competition that we are analyzing in this chapter can be thought of as a race for a patent. So, if our focus were to be on patents then we would want to recognize that either firm A or B might want to accelerate its R&D at the cost of incurring additional expenditures. Put differently, if an appropriate regulator in our aggregate economy were to give rise to a rent and here the rent would arise from the monopoly situation created by a patent, then there would be competition for this rent and hence it would be partly dissipated by the additional costs that would be incurred to appropriate it.16 Now, the decision to conduct R&D has to be made before the two firms realize what the actual cost of producing vaccines is going to be. Therefore, a firm will choose to conduct R&D if and only if this decision leads to higher ex ante profit. What complicates this decision for either firm is that as shown in Eqs. (8) and (10), firm A' s ex ante profit depends on whether firm B chooses to conduct R&D and vice versa. To study the equilibrium that arises when either firm makes a decision to conduct R&D, let us first define a firm’s expected profit as a function of the outcome of the decision to conduct R&D. To this end, let Π 00 denote the expected profit to a firm when the decision to conduct R&D leads to failure, i.e., results in no[ cost] reducing innovation, and hence the cost distribution for both firms remains 0, c . Let Π 10 Ʌ

Ʌ

Ʌ

Ʌ

respective mRNA vaccines. This standpoint is similar to the notion of ex ante expected profits that we have been discussing in this section. See Dunleavy (2021) for additional details on this point. 15 See Danzon and Pereira (2011) for a discussion of the importance of fixed costs in the context of the development of vaccines. 16 See Lee (2022) for a discussion of the practical pros and cons of granting patents to the two leading producers of mRNA vaccines and Gaviria and Kilic (2021) for a more general discussion of mRNA vaccine patents.

52

A. A. Batabyal and H. Beladi

denote the expected profit to a firm when its decision to conduct R&D leads ] to a [ cost reducing innovation, this firm’s cost distribution changes to 0, c − β , and hence this firm displaces the other firm and becomes a cost leader. Let Π 01 be the expected profit to our firm if its competitor’s decision to conduct R&D[ leads to ]a cost reducing innovation, the competitor’s cost distribution changes to 0, c − β , and hence this competitor displaces the first firm and becomes a cost leader. Finally, let Π 11 represent the case where the decision to conduct R&D by both firms leads innovations and therefore both face the changed cost distribution ] [to cost reducing 0, c − β . Some thought tells us that for, say, the ith firm, i = A, B, we get Ʌ

Ʌ

Ʌ

2 ( ] Hc c Π 00 = E[πi c, c/c A ≥ c B = − , 6 8 ( )2 ) ( ( ] c c−β H c−β Π 10 = E[πi c − β, c/c A < c B = + 6c 4 ( )3 2 c−β Hβ c − + − , 2 24c 3 )2 ( )3 ( ( ] H c−β c−β Π 01 = E[πi c, c − β/c A ≥ c B = − , 6c 8c Ʌ

Ʌ

Ʌ

Ʌ

Ʌ

Ʌ

Ʌ

Ʌ

Ʌ

Ʌ

Ʌ

(11)

Ʌ

Ʌ

Ʌ

Ʌ

Ʌ

Ʌ

Ʌ

Ʌ

Ʌ

Ʌ

(12)

Ʌ

Ʌ

Ʌ

Ʌ

(13)

and ) ( )2 ( ( ] c−β H c−β − . = E[πi c − β, c − β/c A ≥ c B = 6 8 Ʌ

Π 11

Ʌ

Ʌ

Ʌ

Ʌ

Ʌ

(14)

The decision by our vaccine producing firms to conduct R&D can be conceptualized as a static game in which each firm has two actions denoted by ai , i = A, B. These actions are “Conduct R&D” denoted by R and “Don’t conduct R&D” denoted by D. In symbols, we have ai ∈ {R, D}. The payoffs ( to the) two firms as a function of the two available actions can be expressed as Wi ai , a j , i /= j. In addition, the four specific payoffs written out in full detail are Wi (R, R) = (1 − θ )2 Π 00 + θ (1 − θ )Π 10 + θ (1 − θ )Π 01 + θ 2 Π 11 − ζ,

(15)

Wi (R, D) = (1 − θ )Π 00 + θ Π 10 − ζ,

(16)

Wi (D, R) = (1 − θ )Π 00 + θ Π 01 ,

(17)

Wi (D, D) = Π 00 .

(18)

and

City and Regional Demand for Vaccines Whose Supply Arises …

53

To interpret the above four payoffs, consider, for instance, the payoff Wi (R, R) given by Eq. (15). The situation in which both firms fail to generate a cost reducing innovation from their decision to conduct R&D occurs with probability (1 − θ )2 and the associated expected profit term is Π 00 . This explains the first term on the RHS of Eq. (15). The situation in which only one of the two firms generates a cost reducing innovation with its decision to conduct R&D occurs with probability θ (1 − θ ). The associated expected profit terms are either Π 10 or Π 01 . This explains the second and the third terms on the RHS of Eq. (15). The case where both firms generate a cost reducing innovation occurs with probability θ 2 and the related expected profit term is Π 11 . This explains the fourth term on the RHS of Eq. (15). Finally, observe that a firm has to pay the fixed cost of ζ whenever it decides to conduct R&D and this explains the fifth and last term on the RHS of Eq. (15). Similar interpretations can be given to the three remaining payoffs given in Eqs. (16) through (18).

4.2 Nash Equilibria Our task now is to determine the Nash equilibria of the static game that we have been describing thus far. To this end, let us first focus on the two possible symmetric equilibria. In these equilibria, both firms take similar actions as far as the decision to conduct or not conduct R&D is concerned. The first symmetric Nash equilibrium is where both firms conduct R&D. This happens when Wi (R, R) ≥ Wi (D, R) ⇐⇒ θ {θ (Π 11 − Π 01 ) + (1 − θ )(Π 10 − Π 00 )} ≥ ζ. (19) Let us interpret what the condition in (19) is telling us in three steps. First, by conducting R&D, a firm generates a cost reducing innovation with probability θ. This is the θ that appears outside the expression in the curly brackets in (19). Second, conditional on generating a cost reducing innovation, the marginal benefit to the firm is (Π 11 − Π 01 ) if its rival also generates a cost reducing innovation, which happens with probability θ, and the marginal benefit is (Π 10 −Π 00 ) if its rival fails to generate a cost reducing innovation, which happens with probability (1 − θ ). This explains the presence of the expression inside the curly brackets in (19). Finally, putting the preceding two points together, as long as the expected marginal benefit from conducting R&D exceeds the fixed cost ζ, it is a Nash equilibrium for each firm to conduct R&D. The second symmetric Nash equilibrium occurs when both firms decide not to conduct R&D. In this case, we have Wi (D, D) ≥ Wi (R, D) ⇐⇒ θ (Π 10 − Π 00 ) ≤ ζ.

(20)

54

A. A. Batabyal and H. Beladi

The condition in (20) tells us that the expected benefit to a firm from conducting R&D or θ (Π 10 − Π 00 ) does not exceed the fixed cost ζ incurred when conducting this R&D. When this condition holds we have another possible Nash equilibrium. In addition to the above two symmetric Nash equilibria, in principle, it is possible for there to exist an asymmetric Nash equilibrium in which only one firm conducts R&D. In this instance, we need two conditions to hold simultaneously. These two conditions are Wi (R, D) ≥ Wi (D, D) and W i (D, R) ≥ Wi (R, R).

(21)

In words, the two conditions in (21) tell us that we need one firm to want to conduct R&D when the other firm does not and we also need the other firm to not want to conduct R&D when the first firm wants to conduct R&D. As in the case of the two possible symmetric Nash equilibria discussed above, it is possible to use the various expected profit expressions in Eqs. (11) through (14) and rewrite the two conditions in (21). By doing this, we get θ (Π 10 − Π 00 ) ≥ ζ ≥ θ {θ (Π 11 − Π 01 ) + (1 − θ )(Π 10 − Π 00 )}.

(22)

To summarize the discussion in this section, we have seen that there is a symmetric Nash equilibrium—see (19)—in which both firms conduct R&D and, in addition, there is also an asymmetric Nash equilibrium—see (22)—in which only one of the two vaccine producing firms in our aggregate economy conducts R&D. Our final task in this chapter is to present a way of mimicking the effect of increased competition and to then analyze the impact of this increased competition on the incentives faced by the two firms to conduct R&D and generate potentially cost reducing innovations.

5 Increased Competition Ʌ

5.1 A Decrease in c Recall from [Sect.] 4 that the cost distribution faced by the duopolists in our model is given by 0, c . Now suppose that the upper endpoint of this distribution or c declines. We shall interpret this decline in c as being equivalent to an increase in the competition between the duopolists under study. To examine the impact of this increased competition on the innovation incentives faced by the duopolists, let us differentiate Eq. (11) with respect to c. This gives us Ʌ

Ʌ

Ʌ

Ʌ

Ʌ

c ∂Π 00 H − > 0. = ∂c 6 4 Ʌ

(23)

City and Regional Demand for Vaccines Whose Supply Arises …

55

The sign of the derivative in) Eq. (23) follows from our understanding—see ( Sect. 2—that H > 2max c A , c B = 2c. Given this sign result, we emphasize that a decline in c can be interpreted as an increase in competition because this decline reduces a firm’s pre-cost reducing innovation profits.17 Ʌ

Ʌ

Ʌ

Ʌ

5.2 Firm Incentives Now, to analyze the impact of this increase in competition on the incentives to conduct R&D and to generate possibly cost reducing innovations faced by the duopolists, let us define the two functions ψ1 (c, β, H ) and ψ2 (c, β, H ) where Ʌ

Ʌ

( ) β 2 ψ1 c, β, H = Π 10 − Π 00 = {β 2 + 4Hβ + c(4H − 3β) − 3c } 24c Ʌ

Ʌ

Ʌ

Ʌ

(24)

and ( ) ) ( )2 } β { ( ψ2 c, β, H = Π 11 − Π 01 = 4H c − β − 3 c − β . 24c Ʌ

Ʌ

Ʌ

Ʌ

(25)

In words, the two functions given by Eqs. (24) and (25) capture the benefit from generating a cost reducing innovation. Specifically, ψ1 (c, β, H ) describes the increase in expected profit obtained by a firm when it generates a cost reducing innovation and its rival does not conduct R&D. Similarly, ψ2 (c, β, H ) describes the increase in expected profit obtained by a firm when it generates a cost reducing innovation and its rival does conduct R&D. Differentiating the functions ψ1 (·) and ψ2 (·) with respect to c, it is straightforward to verify that Ʌ

Ʌ

Ʌ

⎧ ⎫ ( ) 2 ∂ψ1 c, β, H β β 2 + 4β H + 3c =− 0. = < ∂ cˆ 24 cˆ2

(27)

Even though the sign of the partial derivative in Eq. (27) is, in general, indeterminate, if β is large enough then we can reasonably expect the sign of this derivative to

17

Our focus on studying the effects of increased competition in the market for vaccines is consistent with the viewpoint of the United Nations Conference on Trade and Development (UNCTAD). Go to https://unctad.org/news/defending-competition-markets-during-covid-19 for more details on this point. Accessed on 6 July 2022.

56

A. A. Batabyal and H. Beladi

be positive. To keep the subsequent mathematical analysis tractable, in what follows, we shall assume that this is, in fact, the case. To comprehend why the signs of the derivatives in Eqs. (26) and (27) are as they are, note the following line of reasoning. First, with regard to Eq. (26), if competition between the two vaccine producing firms is intense (this happens when c is small) then the benefit to a firm from generating a cost reducing innovation is high {ψ1 (·) is high} when this firm’s rival does not conduct R&D. This explains the negative sign of the derivative in Eq. (26). Second and in contrast, the incentive to a firm to conduct R&D and generate a cost reducing innovation is low {ψ2 (·) is low} if this firm’s rival is also conducting R&D. This last finding arises because more intense competition between the two firms reduces the value of, so to speak, “getting ahead” when one’s rival is “already ahead.” Using the descriptions of the two functions ψ1 (·) and ψ2 (·), we can rewrite the three inequality conditions given in (19), (20), and (22). This gives us Ʌ

( ) )} ( { θ θ ψ2 c, β, H + (1 − θ )ψ1 c, β, H ≥ ζ,

(28)

) ( θ ψ1 c, β, H ≤ ζ,

(29)

Ʌ

Ʌ

Ʌ

and ( ( ( ) { ) )} θ ψ1 c, β, H ≥ ζ ≥ θ θ ψ2 c, β, H + (1 − θ )ψ1 c, β, H . Ʌ

Ʌ

Ʌ

(30)

Recall that (28) refers to the symmetric Nash equilibrium in which both vaccine producing firms conduct R&D. Similarly, (29) concerns the symmetric Nash equilibrium in which neither of the two firms conducts R&D. Finally, (30) refers to the asymmetric Nash equilibrium in which only one firm conducts R&D. Inspecting (28) through (30) carefully, it is clear that when there is increased competition in the market for vaccines, i.e., when c declines, the value of the function ψ1 (c, β, H ) rises and hence the condition given in (29) is less likely to be satisfied. Therefore, in an environment with increased competition, the benefit to a firm from conducting R&D and stochastically generating a cost reducing innovation is high. This means that a symmetric Nash equilibrium in which neither firm conducts R&D is unlikely to occur. If the probability of generating a cost reducing innovation or θ is high enough then, differentiating the condition given in (28) with respect to c, we get Ʌ

Ʌ

Ʌ

( ) ∂{θ ψ2 (c, β, H + (1 − θ )ψ1 c, β, H } > 0. ∂c Ʌ

Ʌ

Ʌ

(31)

The inequality in (31) tells us that with increased competition or with a decrease in c, the symmetric Nash equilibrium in which both firms conduct R&D is also unlikely. To see why this is the case a little differently, observe that as the probability θ of generating a cost reducing innovation approaches one, it becomes increasingly more Ʌ

City and Regional Demand for Vaccines Whose Supply Arises …

57

likely that a decision to conduct R&D by the two firms will lead to a cost reducing innovation for both of them. That said, if competition between the two firms in the market for vaccines is intense then even the profit after having generated a cost reducing innovation is likely to be low compared to the fixed cost ζ of conducting R&D. This line of reasoning tells us that the condition given in (28) is unlikely to be satisfied and therefore a symmetric Nash equilibrium in which both vaccine producing firms conduct R&D is also unlikely to exist. Having eliminated the conditions specified in (28) and (29), it follows that the only remaining case is the condition given in (30). We claim that this condition is most likely to be satisfied in an environment with increased competition. To see why, note the following three-part line of reasoning. First, in the absence of R&D, the rents obtained by the two firms are low because rising competition lowers expected profits for both firms. Second, to reduce competition, one firm will want to generate a cost reducing innovation but only if the other firm decides to not pursue the same strategy—of wanting to generate a cost reducing innovation—simultaneously. Finally, the preceding two points together lead to the conclusion that when competition between the two firms in the market for vaccines is intense, the asymmetric Nash equilibrium in which only one firm conducts R&D and hence potentially generates a cost reducing innovation is the most likely scenario. We have pointed out in our analysis thus far in this section that certain kinds of equilibria are less likely to occur in the static game between the Bertrand duopolists under study. That said, it should be noted that the effect of increased competition on aggregate expenditures on R&D in general is ambiguous. To see this, consider the following two cases. In the first case, suppose that in the status quo, neither firm conducts R&D. In this case, when competition policy reduces c over time, the condition in (29) for the symmetric Nash equilibrium in which neither firm conducts R&D will be violated. When this happens, at least one firm will conduct R&D. Clearly, in this case, increased competition between the duopolists will lead to an increase in aggregate expenditure on R&D. In the second case, suppose that in the status quo, both firms conduct R&D. In this case, increased competition between the two firms may move our aggregate economy to an asymmetric equilibrium in which only one firm conducts R&D. Obviously, in this second case, increased competition between the duopolists leads to a diminution in aggregate expenditures on R&D. This completes our discussion of the demand for vaccines by cities and regions when their supply is the outcome of competition in a Bertrand duopoly. Ʌ

6 Conclusions In this chapter, we analyzed a one-period model of an aggregate economy composed of cities and regions that demanded vaccines designed to fight a pandemic such as Covid-19. The supply of the vaccines was the outcome of Bertrand’s competition between two firms A and B. The marginal cost of producing the vaccine for both firms

58

A. A. Batabyal and H. Beladi

was stochastic and drawn from a uniform distribution. In this setting, we performed three tasks. First, we delineated the equilibrium pricing strategies of the two firms and then we computed their expected ex ante profits. Second, we allowed both firms to conduct risky R&D and then ascertained the conditions under which only one firm conducted R&D and conditions under which both did. Finally, we proposed a way of mimicking the effect of increased competition and then studied the impact of this increased competition on the incentives to conduct R&D faced by the two firms. The analysis in this chapter can be extended in a number of different directions. Here are two potential extensions. First, it would be interesting to model the Bertrand competition between the duopolists in the market for vaccines when one or more city and regional governments are able to commit to an advance agreement to purchase a certain quantity of the produced vaccines. Second, it would also be informative to study a scenario in which one or more city and regional governments attempt to increase competition in the market for vaccines by offering incentives such as subsidies to firms willing to enter this market. Such actions will involve an analysis not of a duopoly but, more generally, an oligopoly with n > 2 potential firms. Finally, one could also study scenarios in which the cost of producing a vaccine is quadratic, firms compete to produce a vaccine in order to obtain a patent and thereby secure monopoly rights over the produced vaccine, and the pertinent consumers are cities that are—like in the Hotelling (1929) model—uniformly distributed along a straight line with length one. Studies that analyze these aspects of the underlying problem will provide additional insights into the nature of competition policy, the behavior of firms with market power, and the ultimate development of vaccines.

References Acemoglu, D. (2009). Introduction to modern economic growth. Princeton University Press. Batabyal, A. A., & Beladi, H. (2022). Health interventions in a poor region and resilience in the presence of a pandemic. Forthcoming. Applied Spatial Analysis and Policy. Branswell, H. (2021). New analysis finds global Covid death toll is double official estimates. STAT, May 6. Retrieved July 5, 2022, from https://www.statnews.com/2021/05/06/new-analysis-findsglobal-covid-death-toll-is-double-official-estimates Chaplin, S. (2020). COVID-19: A brief history and treatments in development, Prescriber, May, 23–28. Danzon, P. M., & Pereira, N. S. (2011). Vaccine supply: Effects of regulation and competition. International Journal of the Economics of Business, 18, 239–271. Dunleavy, K. (2021). Pfizer, Moderna will rake in a combined $93 billion next year on COVID-19 vaccine sales: Report. Retrieved July 5, 2022, from. https://www.fiercepharma.com/pharma/pfi zer-moderna-will-rake-a-combined-93-billion-next-year-covid-19-sales-says-analytics-group Fu, Q., Lu, J., & Lu, Y. (2012). Incentivizing R&D: Prize or subsidies? International Journal of Industrial Organization, 30, 67–79. Gaviria, M., & Kilic, B. (2021). A network analysis of COVID-19 mRNA vaccine patents. Nature Biotechnology, 39, 546–548. Gilchrist, S. A. N., & Nanni, A. (2013). Lessons learned in shaping vaccine markets in low-income countries: A review of vaccine market segment supported by the GAVI alliance. Health Policy and Planning, 28, 838–846.

City and Regional Demand for Vaccines Whose Supply Arises …

59

Hotelling, H. (1929). Stability in competition. Economic Journal, 39, 41–57. Katella, K. (2021). Comparing the COVID-19 vaccines: How are they different? Yale Medicine, August 13. Retrieved July 5, 2022, from https://www.yalemedicine.org/news/covid-19-vaccinecomparison Kremer, M., Levin, J. D., & Snyder, C. M. (2020). Designing advance market commitments for new vaccines. National Bureau of Economic Research Working Paper 28168, Cambridge, MA. Lee, N. (2022). Experts seriously doubt whether patent waivers on Covid-19 vaccines will ever come to be. Retrieved July 5, 2022, from https://www.cnbc.com/2022/01/22/why-moderna-pfi zer-and-the-nih-debate-who-owns-the-covid-vaccine.html Martonosi, S. E., Behzad, B., & Cummings, K. (2021). Pricing the COVID-19 vaccine: A mathematical approach. Omega, 103, 102451. Mudholkar, G. S., Srivastava, D. K., & Kollia, G. D. (1996). A generalization of the Weibull distribution with application to the analysis of survival data. Journal of the American Statistical Association, 91, 1575–1583. Peters, M., & Simsek, A. (2009). Solutions manual for introduction to modern economic growth. Princeton University Press. Ross, S. M. (1996). Stochastic processes (2nd ed.). Wiley. Ross, S. M. (2007). Introduction to probability models (9th ed.). Academic Press. Sloan, F. A., & Hsieh, C. (Eds.). (2007). Pharmaceutical innovation. Cambridge University Press. Stiglitz, J. E. (2021). Globalization in the aftermath of the pandemic and Trump. Journal of Policy Modeling, 43, 794–804. Tirole, J. (1988). The theory of industrial organization. MIT Press. Torreele, E. (2020). Business-as-usual will not deliver the COVID-19 vaccines we need. Development, 63, 191–199. Wanke, P. (2008). The uniform distribution as a first practical approach to new product inventory management. International Journal of Production Economics, 114, 811–819.

Post Pandemic Cities—Competing for Size or Cooperating for Interaction an Analysis of the Evolution of Portuguese Municipalities Based on an Organic and Rational Spatial Interaction Growth Models Tomaz Ponce Dentinho

1 Introduction The spatial structure reveals the intrinsic potential of places and expresses in the resilience of their rank within the places of a country (Zipf, 1949). Notwithstanding this, external impacts have a role in the profile of the system of places namely the spatial redistribution of public money (Dentinho, 2017), the collapse of natural resources (Diamond, 2005); or technological innovations in transport and communication (Reggiani & Nijkamp, 1998). When interaction between city-regions is free the acts of each city-region generate impacts that modifies the context for other city-regions and interactions become complex leading to chaotic behaviours (Goldberger, 1996), new equilibriums (Gribbin, 2004) or self-organized dynamic equilibriums (Waldrop, 1992). Two types of interactions can be conceptualised: Organic and rational. Organic interaction exists between organisms that tend to maximize their stock and adapt through evolution (Pearce & Merletti, 2006) into spatially defined ecosystems (Müller-Schloer et al., 2011) eventually explainable by network models (Riley et al., 2015). On the other hand, Rational Interaction appears between people and places when their aim is to maximize interaction between them, as perceived by the seminal works of (Lösch, 1954; Alonso, 1969), further developed by Paelinck and Nijkamp (1976) and (Sen & Smith, 1995); and operationalized by the researchers inspired by the works of Wilson (1970) and Echenique et al. (2013). Places maximize flows as assumed in regional economic models or, alternatively, they maximize their stock or size promoting unsustainable paths (Dentinho, 2011). The maximization of size happens when the average cost of the city equals the average T. P. Dentinho (B) University of Azores, Faculty of Environmental and Agricultural Sciences, Angra Do Heroísmo, Portugal e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. G. Celbi¸s et al. (eds.), Pandemic and the City, Footprints of Regional Science, https://doi.org/10.1007/978-3-031-21983-2_5

61

62

T. P. Dentinho

benefits; whereas the maximization of interaction flows occurs when the marginal cost of the city equals the marginal benefit (Simões Lopes & Pontes, 2010). The methodological approach proposed by (Dentinho & Rodrigues, 2021) on models of complex spatial interaction, the authors differentiate rational spatial interaction, that maximize interaction flows between places, from organic spatial interaction, where places compete with each other to maximize their size. This chapter tries to perceive if, along three periods of twenty years from 1960 to 2021, the implicit interaction between Portuguese municipalities was more organic or more rational. Then to simulate post pandemic scenarios for each type of interaction behaviour: Organic or rational. Point 2 presents the growth models of spatial interaction with organic and rational collaboration. Point 3 estimates both models using the evolution of the population of Portuguese 308 municipalities and the distance matrix between them. Point 4, presents the simulations for future scenarios and discusses the results regarding the sustainability of Portuguese Municipalities. Point 5 presents conclusions and policy recommendations.

2 Organic and Human Spatial Interaction Models Table 1 synthesises the two models to be tested, using the same spatial interaction matrix {W i j =Pi P j ex p (−βd i j ) }, that represents the permanent spatial structure of the territory, where Pi = In averageweightofthepopulationofaplace(i)nthecountry and (ex p (−βd i j ) ) is an impedance function of the distance between (i) and (J). The second and third columns of Table 1 represent the Organic Spatial Interaction Model that assumes that the population evolves according to the entropic interaction between ∑ municipalities {− Nj [W i j X jt (ln(1 − δ)X jt −1)]} influenced by the existing population (X i , ,) and the Permanent Spatial Interaction spatial interaction matrix (W i j ), the economic ∑ base of the previous period (E it−1 ) and to the correction of external inputs (- Nj (W i j (1 − δ)X jt ). In the third and fourth columns, the Rational Spatial Interaction Growth Model assumes that demographic evolution depends on the structural spatial interaction (W i j ), on the economic base of the previous period (E it−1 ) and on two correction ( ) ∑ ∑ ∑ factors of the economic base { Nj Wi2j − Nj Wi j ln Wi j }; Nj Wi j that express the spatial spread effects of changes in the economic base. Notice that whereas the Organic Spatial Interaction Growth Model has cumulative effects represented by the existing population (X i ), the Rational Spatial Interaction Growth Model depends only on the expression of the permanent spatial structure of the territory {W i j ). The approach is first, to estimate the organic and rational spatial interaction factors, or organic and rational entropy. Second, based on the evolution of the entropy, economic base and correction factors for the organic and rational models we estimate the organic and rational regressions. Then we are able to not only see what is the

Extra Correction of external inputs—Spatial Interaction

εio

ε'io

External inputs

Correction of external inputs

ko

αo

Entropy

j (W i j (1 − δ)X jt

j [W i j X jt (ln(1 − δ)X jt − 1)]}

∑ N

(E it ) ∑ N

−

{−

εir (εir + νir )

εir

εir m

kr ( ) ) ( αr εir + νir + 1 exp −εir − νir

Coefficient

Intercept

Rational spatial interaction model

Coefficient

Variable

Organic spatial interaction model

Table 1 Coefficients and variables of organic and rational spatial interaction growth models

)) 2 ( ∑ N ( j [ −Wi j + ln Wi j Wi j exp(−Wi j )]} E it ) ( ∑ ∑ N 2 { N j Wi j − j Wi j ln Wi j } ∑ N j Wi j

{−

Variable

Post Pandemic Cities—Competing for Size or Cooperating … 63

64

T. P. Dentinho

better interaction growth model—organic or rational—but also to simulate future scenarios based on each one of the estimates.

3 Data on Spatial Interaction Factors and on the Evolution of the Population 3.1 Data on Population

Ln(Muncipal Population/ Country Population)

Figure 1 shows the evolution of the Zipf’s Curves (Zipf, 1949) for the Portuguese Municipalities from 1960 to 2021. Looking at the slopes of the Zipf’s Curves it is clear that, along these sixty years, there has been an increasing concentration of the Portuguese population in the bigger municipalities and in surrounding areas and a steady decrease in the population in the more depopulated areas. Looking at Fig. 2 it is evident that the winning regions are in the coastal strip between Braga in the North and Setubal, mostly in the expansion of the Metropolitan Areas of Lisbon, Porto and Braga; the cities of Viseu, Coimbra and Leiria; plus the touristic destination of Funchal in Madeira. The municipalities around Serra da Estrela lost their relative importance along these sixty years. Evolution of the Zipf's Coefficients for Portuguese Municipalities -2 0

1

2

3

4

5

-3 -4 -5 -6 Y(1960) = -0,75*X - 2,52 -7 Y(1981) = -0,98*X- 1,67 -8 Y(2001)= -1,07*X - 1,34 -9 y(2021) = -1,14x - 1,078 -10

Ln (rank) 1960

1981

2001

2021

Linear (1960)

Linear (1981)

Linear (2001)

Linear (2021)

Fig. 1 Evolution of the Zipf’s coefficients for portuguese municipalities (1960 to 2021)

6

Post Pandemic Cities—Competing for Size or Cooperating … 1960

1981

2001

2021

65

Fig. 2 Percentiles of the distribution of the portuguese population from 1961 to 2021

3.2 Data on the Spatial Interaction Permanent Potentials ∑ Figure 3 maps the Permanent Potential of Spatial Interaction {W i = nj=1 [W i j ]} to represent the estimates of the permanent spatial interaction factors {W i j = Pi P j ex p (−βd i j ) }. The impedance parameter calibrated to secure that∑ the average n i=1 {[W ii ] interaction } = N ∑ n within zones is 50% of the average total interaction { 0, 5{ j=1 [W i j ]/N }. Figure 3 reveals the strong Potential of Lisbon, Porto and Braga Metropolitan Areas and for the cities of Braga, Funchal, Coimbra and Leiria. Data on Population was collected from PORDATA (2021) and INE (2021). Data on distances were km per road and, to cross the sea, we use the formula of (Keeble et al, 1982).

4 Estimates of the Organic and Rational Spatial Interaction Growth Models 4.1 Estimation Process The estimation process involves different phases. First, we test three impedance parameters (β) not only for the estimated to secure que equilibrium interactions within and between municipalities but also for parameter values less and plus 50% of the impedance parameter; the results for the estimated impedance parameter show up much better than the results for higher and lower values.

66

T. P. Dentinho

Fig. 3 Permanent potential of spatial interaction

Second, we estimate a spatial weight matrix based on Euclidean distances with a limited threshold to cut interactions between the Portugal Main Land, Azores and Madeira but to allow interaction within each one of these areas and use these weights in regressions but the results were worse. Third, we eliminated the correction factor of the organic model and the first correction factor of the rational model because their coefficients were not significant. Fourth, because the variables of the rational model are the same for the three periods and we wanted to compare the two models, we divided the organic and rational entropies—this one with the second correction factor—into two sets of variables (for 1960–1981 and for1981-2001) considering the reference period of (2001–2021). To still consider the time variable we introduced two time dummies taking the reference of (1960–1981): One for (1981–2001), and another for (2001–2021). With two different reference periods, we could estimate the models avoiding intrinsic multicollinearity. Because we use panel data, we introduce spatial dummies and choose Lisbon as the reference municipality.

4.2 Organic Spatial Interaction Growth Model Table 2 presents the results of the Organic Spatial Interaction Growth Model that explains growth indices as a function of organic entropy, a proxy (*) of the economic base of the previous period assumed to be 20% of the population in the previous year, plus temporal and spatial dummy variables. The model with fixed effects presents an R2 = 0,760 and high overall significant results.

Post Pandemic Cities—Competing for Size or Cooperating …

67

Table 2 Estimates of the organic spatial interaction growth model R2 = 0,760 Sig = 0,000

B

Standard deviation

Standardized beta

(Constant)

5,195

0,288

OEntropy1960-1981

14,735,233

5585,484

0,757

OEntropy1981-2001

−20,891,159

8044,864

−0,745 −1,246

Economic Base*

−287,215

18,431

DD2001

0,020

0,013

0,036

DD2021

−0,038

0,013

−0,067

t

Sig

18,043

0,000

2,638

0,009

−2,597

0,010

−15,583

0,000

1,566

0,118

−2,932

0,003

According to the organic growth model, the constant growth increased in the period from 1981 to 2001, and decreased from 2001 to 2021. Furthermore, the municipalities with bigger economic bases grew less, and there are relevant cumulative effects in growth and decline explained by the organic entropy, that shows that population concentrates cumulatively in more central places in the period from 1960 to 1981 and spread cumulatively around those central places from 1981 to 2001, stabilizing afterwards. The coefficient of organic entropy is higher for the period 1960 to 1981, much lower for the other period (1981–2001) and in the middle taken as a reference implicit value for (2001–2021). The spatial dummies estimated for the Organic Spatial Interaction Growth Model (Fig. 5) reveal that the Metropolitan Areas of Lisbon, Porto and Braga could have had higher Permanent Potential of Spatial Interaction and the same could have happened with the Urban Areas around Coimbra, Leiria and Funchal. On the other hand, the Permanent Potentials of Spatial Interaction could be lower in the interior of the country. Using GEODA (2021) software Fig. 6 presents the Local Geary Cluster Map of the Spatial Dummies for Municipalities of Fig. 5. The analysis of Fig. 6 reinforces the robustness of the results presented in Table 2 and in Fig. 5. Only five municipalities around Lisbon show positive spatial correlation with the neighbours indicating positive complementarities, whereas almost half of the municipalities of the country (163) show negative spatial interaction suggesting the existence of a process of destructive competition in the interior of the country. Notwithstanding this, the overall spatial correlation of the dummy variables coefficients in very low and significant correlation indicators appear more in the interior of the country (small map in green).

4.3 Rational Spatial Interaction Growth Model Table 3 presents the results of the Rational Spatial Interaction Growth Model that tries to explain demographic growth as a function of the Permanent Potential of Spatial Interaction, not taking into account the cumulative positive and negative effects of the

68

T. P. Dentinho

Fig. 5 Coefficients of spatial dummies of the organic spatial interaction growth model

Fig. 6 Local geary cluster map and spatial correlation graphs

population in past periods. The proxy of the Economic Base of the previous period plus time and spatial dummy variables were also included. The rational model with fixed effects presents an R2 = 0,825 and high overall significant results, considerably better than the estimates of the organic model of Table 2. According to the estimates of the rational model, the constant growth increases for all the periods. Furthermore, municipalities with higher proxy values of the economic base grew less, and the effect of the Permanent Potential of Spatial Interaction, present in the Rational Entropy factors and Correction Factors, reduces over time.

Post Pandemic Cities—Competing for Size or Cooperating …

69

Table 3 Estimates of the rational spatial interaction growth model R2 = 0,825 Sig = 0,000 (Constant)

B

Standard deviation

Standardized beta

T

Sig

5,683

0,515

11,025

0,000

REntropy1981

3194,871

534,086

2,578

5,982

0,000

REntropy2001

2527,550

648,177

2,041

3,899

0,000

ECorrection1981

−0,730

0,115

−2,752

−6,352

0,000

ECorrection2001

−0,624

0,136

−2,353

−4,582

0,000

Economic Base*

−263,120

27,655

−1,141

−9,514

0,000

DD2001

0,060

0,012

0,107

4,991

0,000

DD2021

0,009

0,012

0,016

0,765

0,445

The spatial dummies estimated for the Rational Spatial Interaction Growth Model (Fig. 7) reveal a different pattern from the one indicated by the spatial dummies of the Organic Spatial Interaction Model (Fig. 5). As expected, whereas the organic formula reinforces the cumulative pressures of concentration and sprawl around major cities, the estimates of the rational formula indicate a concentration around the Permanent Potential of Spatial Interaction of Fig. 3 not yet fulfilled by the pressures of the real population evolution of Fig. 2. Clearly, Lisbon, Oeiras, Santiago de Cacém and the areas around Porto and Lisbon grew more than the estimated intuition of the Permanent Potential Indicated. The issue, not considered in this chapter, is to find and justify a Permanent Potential of each municipality consistent with the sustainable interaction of the urban network but this is not the aim of this chapter.

Fig. 7 Coefficients of spatial dummies of the rational spatial interaction growth model

70

T. P. Dentinho

Fig. 8 Local geary cluster map and spatial correlation graphs

Once more using GEODA (2021) software Fig. 8 presents the Local Geary Cluster Map of the Spatial Dummies of Fig. 7. It is similar to Fig. 6 but with a more clear indication of the remote areas of the Azores Island and the Northeast where decreasing regions interact negatively with other decreasing regions. It is also interesting to testify the negative correlation in Sines and Albufeira which can be the removal by industry and tourism. Finally, the negative interactions move more into the coastal areas of the centre indicating the spatial protection of the far North and far South of the country.

5 Projections and Discussion Using the estimates of the spatial dummies for both organic and rational models, Fig. 9 presents the projections of the Portuguese Population per Municipality for 2041, based in the rational and organic estimated models and the ratio between Rational and Organic Projections. Both projections reinforce the increase in the Metropolitan Areas of Lisbon, Porto and Braga and the municipalities of Leiria, Coimbra and Viseu; plus Aveiro for the Rational Projection. Notwithstanding this, there are revealing differences between the spatial dummies of the two models, rational and organic. Projections based on the Rational Spatial Interaction Growth Model are higher for the Metropolitan Areas of Lisbon, Porto and Braga and lower for the municipal areas close to the border with Spain. A question can arise: What is the best projection, organic or rational, after the pandemic? Since the Organic Model includes cumulative effects and the Rational Model takes on the spatial potential of places my arguments is that, since the memory

Post Pandemic Cities—Competing for Size or Cooperating …

Rational Projections for 2041

71

Organic Projections for 2041

Ratio between Rational and Organic Projections for 2041

Fig. 9 Projections of the population per municipality for 2041

of cumulative effects was disturbed by the pandemic, eventually the rational urban network adjustment will be stronger as it has been along the periods of major changes in the countries, from the sixties onwards. Moving beyond the spatial errors used extensively in spatial econometric literature after the seminal paper of (Anselin et. al., 1997), it is clear that spatial interaction and no errors can be, on the one hand, justified theoretically (Dentinho & Rodrigues, 2021) and, on the other hand, proved empirically as shown in this chapter. Another outcome of this essay is that there is not only Rational Interaction highlighted by the works of (Lösch, 1954; Alonso, 1969; Paelinck and Nijkamp, 1976); Sen & Smith, 1995); used by Wilson (1970), Echenique et al. (2013) and followers based in gravity impedance functions. There is also the possibility to operationalize organic (or irrational) spatial interaction to test the rules of the game of drastic Krugman´s (1991) competition between cities or interaction optimization between places.

6 Conclusion This chapter tried to see if spatial interaction between Portuguese municipalities followed an organic rule involving the competition between them or a rational process

72

T. P. Dentinho

that maximizes interaction flows between them. Using population data from 1960 to 2021 by municipality and the matrix of distances between these places to estimate an Organic Spatial Interaction Growth Model and a Rational Spatial Interaction Growth Model it is shown that the rational model generates better results. The chapter also simulates post pandemic scenarios of spatial interaction between municipalities and concludes that the rational rule although leading to more concentration in the coastal areas also allow different coastal cities to reveal their role (the case of Aveiro). The recommendation is a question for regional policy. If the rational interaction favours the development of poles in coastal areas, why should policy insist in the development of the interior? The implicit research question is what are the factors that influence the Permanent Potential of Spatial Interaction in Fig. 3? Do they change? Should we change them? Highlights. Spatial Interaction Models, Urban Structure, Pandemic, Portugal Acknowledgements to DRCTD—Direção Regional da Ciência e Transição Digital, that financed the workshop on Pandemic, Location and Mobility (project number M3.3.B/ORG.R.C./027/2021) and to the Portuguese Foundation for Science and Technology that financed the research project Operational Models of Complex Spatial Interaction (EXPL/GES-OUT/1325/2021).

References Alonso, W. (1969). Location and Land Use: Towards a General Theory of Land Rent (1969) ISBN 978-0674729568. Anselin, L., A. Varga, A., & Acs, Z. (1997). Local geographic spillovers between university research and high technology innovations, Journal of Urban Economics 42: 422–448. Diamond, J. (2005). Collapse. How Societies Choose to Fail or Succed. Penguin Books Ltd, ISBN 0-670-03337-5. Dentinho, T. (2017). Urban Concentration and Spatial Allocation of Rents from natural resources. A Zipf’s Curve Approach. Region 4(3), 77–86. https://doi.org/10.18335/region.v4i3.169 Dentinho, T. P., Rodrigues, A. F. (2021). Conceptual and operational models of complex spatial interaction. In: Reggiani, A., Schintler, L. A., Czamanski, D., Patuelli, R. (eds) Handbook on entropy, complexity and spatial dynamics. The rebirth of theory?. Edward elgar publishing Ltd, Cheltenham, UK. ISBN: 978 1 83910 058 1. Dentinho, T. P., & Serbanica C. (2020). Space justice, demographic resilience and sustainability. Revelations of the evolution of the population hierarchy of the regions of Romania from. (1948). to 2011. Eastern Journal of European Studies, 11(1), 27–44. Echenique, M. H., Grinevich, V., Hargreaves, A. J., & Zachariadis, V. (2013). LUISA: A LandUse interaction with social accounting model. Presentation and Enhanced Calibration Method, Environment and Planning B: Planning and Design., 40(6), 1003–1026. https://doi.org/10.1068/ b38202 GEODA (2021). An Introduction to Spatial Data Science (https://geodacenter.github.io/) Gribbin, J. (2004). Deep simplicity: Chaos, complexity and the emergence of life. Allen Lane. INE (2021). Instituto nacional de estatística de Portugal https://www.ine.pt/xportal/xmain?xpid= INE&xpgid=ine_main Keeble, D., Owens, P. L., & Thompson, C. (1982), Regional accessibility and economic potential in the European Community. Regional studies, (PDF) Market Potential and the curse of distance

Post Pandemic Cities—Competing for Size or Cooperating …

73

in European regions.16(6), 419–432. Available from: https://www.researchgate.net/publication/ 282943685_Market_Potential_and_the_curse_of_distance_in_European_regions Krugman, P. (1991). Increasing returns and economic geography, Journal of Political Economy, 1991, 99(3) 483–489. Lösch, A.(1954). The economics of location. New Haven: Yale University Press. ISBN 978-0-30000727-5. OCLC 876506870. Müller-Schloer, C., Schmeck, H., & Ungere T. (2011). Organic computing—a paradigm shift for complex systems. Springer, ISBN: 978–3–0348–0129–4. Paelinck, J. H. & Nijkamp, P. (1976). Operational theory and method in regional economics. Farnborough: Saxon House. Pearce, N., & Merletti, F. (2006). Complexity, simplicity, and epidemiology. International Journal of Epidemiology, 35, 515–519. PORTADA (2020). Base de Dados de Portugal Contemporâneo. Fundação Manuel dos Santos. https://www.pordata.pt/ PORDATA (2021).Censos 2021: Resultados Em Portugal E Por Concelho . https://www.pordata.pt/ Reggiani, A., & Nijkamp P. (1998). The Economics of Complex Systems. Elsevier Science CV. Riley, S., Eames, K., Isham, V., Mollison, D., & Trapman, P. (2015). Five challenges for spatial epidemic models. Epidemics, 10, 68–71. Sen, A., & Smith, T. E. (1995). Gravity models of spatial interaction behaviour. Springer. Simões Lopes, A, & Pontes, J. P. (2010). Introdução à Economia Urbana. Fundação Calouste Gulbenkian. Lisboa. ISBN: 9789723112863. Waldrop, M. M. (1992). (1992), Complexity: The emerging science at the edge of order and Chaos. Penguin. Wilson, A. G. (1970). Entropy in urban and regional modelling. Pion. Zipf, G. K. (1949). Human behaviour and the principle of least effort. Addison-Wesley.

Comparative Approaches on the Patterns and Effects of City and Location-Specific Policies and Socioeconomic Structures During COVID-19

The Social Digital Twin for Liveable Cities: A COVID-19 Case Study Corentin Kuster, Sanne Hettinga, Tim van Vliet, Henk Scholten, and Paul Padding

1 Introduction With increasing urbanization and inter-regional connectivity, people are mingling within several social groups, as well as spatial areas (Santana et al., 2020). According to the World Bank, already today 55% of people are living in urban areas, and this number is predicted to increase up to 70% by 2050 (Urban Development Overview, 2020). Cities are often a focal point for these connections and are therefore encountering higher densities of people moving within the city and between cities. Due to this development, local policymakers are faced with increased requirements regarding infrastructures, scarce spaces which require new strategies for crowd management during (large) events or crisis, and disaster management (Boukerche & Coutinho, 2019; Niu & Qin, 2021; Santana et al., 2020). These policymakers require insight into the flow of people moving in their system, using their infrastructure and how these flows can be managed, while ensuring privacy of individuals as much as possible. There are several technical smart city techniques for monitoring and managing crowds. Some examples are: Wi-Fi-tracking on smartphones, tracking cell phone locations, video cameras and radio frequency tracking (Santana et al., 2020; Sindagi & Patel, 2018). All of these techniques have associated advantages or disadvantages. For instance, cell phone tracking can be done everywhere in the world, C. Kuster (B) · T. van Vliet · H. Scholten Geodan, Amsterdam, Netherlands e-mail: [email protected] H. Scholten e-mail: [email protected] C. Kuster · S. Hettinga · H. Scholten Vrije Universiteit Amsterdam, Amsterdam, Netherlands S. Hettinga · P. Padding Geonovum, Amersfoort, Netherlands e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. G. Celbi¸s et al. (eds.), Pandemic and the City, Footprints of Regional Science, https://doi.org/10.1007/978-3-031-21983-2_6

77

78

C. Kuster et al.

without the need for installing local devices to measure pedestrians flow within individual street segments. However, the Dutch authority for personal data (Autoriteit Persoonsgegevens) has declared that privacy of cell phone data cannot be guaranteed and therefore cannot be used freely for crowd management in the COVID-19 crisis (Gebruik telecomdata tegen corona kan alléén met wet | Autoriteit Persoonsgegevens, 2021). Yet, local policymakers indicated how crucial it was for them to manage crowds, especially in the light of the COVID-19 crisis. To provide local policymakers with better insights, crowd modelling is a suitable approach, as it can be based on privacy sensitive data, but due to the modelling aspect, is easier to anonymize. Microsimulation crowd models are part of Agent-based Modelling and Simulation (ABMS) category (Niu & Qin, 2021). They offer the possibility to compute human behaviours at individual level and analyse urban systems in a bottom-up fashion. It enhances coarse-scale demographics to individual households or districts (Boukerche & Coutinho, 2019). This perspective allows researchers to integrate temporarily related processes for an improved, dynamic approach. Coupled with geospatial information, the models enable to predict movement patterns and study fine-granular spatial distribution (Sindagi & Patel, 2018). Microsimulation was first introduced in social-science by Orcutt et al. (Wager et al., 1962). Thereafter, in the mid-1960, demographers embraced the techniques to develop studies over fertility and population projections. Among them, some noticeable progress has been done by Hyrenius (Hyrenius & Adolfsson, 1964) with the development of a reproduction simulation model for age cohort. Ridley and Sheps developed the REPSIM to study demographic and biological influence on natality (Ridley & Sheps, 1966). Similar studies have been done by Jacquard (1967) and Barrett (1967). As computing power increased, the use of microsimulation models became widespread (Hyrenius, 1965), especially in the domain of transportation and land-use modelling (Miller, 2018). Certain models have proven efficiency such as ILUTE (Integrated Land Use Transport Environment) (Miller et al., 2004; Salvini & Miller, 2005) that disaggregate information to synthesize population from census data, travel survey data, activity data, and randomly generated proxy data; ILUMASS (Integrated Land Use Modelling and Transportation System Simulation) (Strauch et al., 2014), a similar system with a new approach for daily activity patterns; MUSSA (Martínez & Donoso, 2010), a model designed to forecast the expected location of agents, residents and firms, UrbanSIM (Waddell, 2011), MATSim (Multi-Agent Transport Simulation) software (Axhausen, 2016) and more. The models above made it possible for users to implement scenarios with the aim to forecast potential behavioural effects. It offers policymakers the opportunity to set strategies and policies on various topics and compare outcomes. Microsimulation models have been implemented for instance for demographic studies (Rephann & Holm, 2004), housing market (Rosenfield et al., 2013), taxation (Berry, 2019), mobility (Azevedo et al., 2016) or public health (Edwards & Clarke, 2009). However, to enable policymakers to actually work with the microsimulation and crowd modelling, they require a digital environment that enables decision-making and management of crowds in their city, with abilities to enact scenarios. In that perspective, digital twins have demonstrated utility. Indeed, Al Sehrawy et al. (2021)

The Social Digital Twin for Liveable Cities: A COVID-19 Case Study

79

explain how digital twins are used more and more for urban planning, despite it being a concept that is still under development. Same goes for Niu et Quin (2021) who describe the use of digital twins for integrating people- and space-related data for decision-making purposes. Because they hold a significant amount of data about the physical environment, digital twin not only serve visualization but also is leveraged to perform crowd simulation and microsimulation. The aim of this paper is using the abundance of spatial data with microsimulation data to allow policymakers to make decisions while respecting the privacy of all affected by this system. To evaluate whether the methodology designed in this paper has applications in practice, a case study is presented about the COVID-19 crisis. It was already indicated why the microsimulation data are relevant for this case study, and that in the Netherlands the privacy laws have restricted the policymakers in their decision making. This makes this a highly relevant case study. The goal of this paper is to design a system using spatial data and microsimulations using the input from different policymakers that is specifically aimed at managing the COVID-19 crisis in the Netherlands at the micro-scale (street and block level). In Sect. 2, the methodology designed for using digital twins to support microsimulations will allow crowd management without affecting the privacy and to design the system with the aid of an expert panel. In Sect. 3, the results were discussed, presenting the digital twin. Finally, in Sect. 4 the constraints and limitations of this study are highlighted, conclusions are drawn and the outlook for future research is described.

2 Methodology Modelling human spatial behaviour in an accurate manner requires the use of finegrained data that might fall under privacy requirements. Indeed, required datasets, for instance mobile phone tracking, are hardly accessible and/or violate regulations. The current study aims at leveraging a pool of proven solutions and integrate them in order to run satisficing models while complying with the upmost privacy regulations. In this section, the different components of such a model will be described. The following diagram (Fig. 1) gives an overview of the different modules from raw micro-data up to the end-user interface. Each of these modules is further explained in the remainder of this section. CBS microdata about the population, including attributes such as age, gender, household composition, status and more, are processed by an AI algorithm for the generation of a synthetic population (a replica of the original dataset but anonymized). This is a dataset that can be created at any scale, but for this case study had to be implemented on the national scale, as people move from city to city and province to province. This synthetic population is then matched against archetypal time-use diaries and the local environment GIS information in order to feed (1) the probabilistic activity model and (2) the digital twin canvas (which is still generated at the national scale). The probabilistic modelling is set following a number of rules defined within

80

C. Kuster et al.

Fig. 1 Human spatial behaviour modelling chart

scenarios e.g., closing shops, curfew, restricting area to pedestrian etc. The outcome of both processes are detailed diaries for each individual in our synthetic population and a 3D digital twin of our location of interest. Both outcomes form the basis for crowd simulation at the local level (specific set of streets for instance), as this requires immense computation power and is impossible to visualize or interpret on a higher scale level. The crowd simulation is performed using Unity Game engine where pedestrian movement are logged for each time step. This allow a live visualization within the game engine, which can be used as a foundation for the digital twin as well, but also within alternative technologies such as MapBox online viewer. In the present case, the purpose of a digital twin is twofold: it helped visualization and facilitated communication and understanding by user, and it is the canvas for modelling exercises.

2.1 Synthetic Population Microsimulation by nature requires the use of fine-grained datasets, mostly at the individual level when it comes to behavioural modelling. In practice, modellers perform as a first step a so-called population synthesis. The term synthetic population is used to describe the methods to generate spatial micro-data on which the spatial microsimulation approach depends. A synthetic population is an implementation of the lager data synthesis methodology. Data synthesis can be defined as follow: Synthetic datasets mimic real datasets by preserving their statistical properties and the relationships between variables (Quintana, 2020).

The Social Digital Twin for Liveable Cities: A COVID-19 Case Study

81

In other words, the synthetic dataset, although different from the original data, remains statistically relevant for any modelling and analysis purposes. The approach presents great advantages concerning privacy and security as no records can be traced back to any individuals. Several methods exist for the creation of a synthetic population. The most widespread method has been developed by Beckham and al. (Beckman et al., 1996). In this study, a 1990 census of the United States is used for the generation of synthetic populations of households. An iterative proportional fitting (IPF) was used to estimate the proportion of households in a certain area with an expected combination of demographic aspects. Applied to several variables, one can disaggregate coarse data into finer, more specific, dataset. In recent years, another more sophisticated approach has emerged using deep learning algorithms. Indeed, with an improved capacity of governmental statistical agencies to collect micro-data of their population, one could imagine using such pool of data as a base for his/her microsimulation. The advantage of the method lays in its direct use of micro-data and its reproducibility. In the project, data synthetic will be used to create a synthetic population from the original CBS microdata about individuals. Attributes such as age, address, household characteristic, means of transportation, incomes, social status etc. will be synthesized. In this way, we will obtain a usable and fine-grained dataset about the population that does not violate privacy and ethical considerations. For such a task, two deep learning models have been favoured in the literature, Variational Autoencoder (VAEs) and Generative Adversarial Networks (GANs). VAEs are unsupervised generative models. They are trained to encode the input information into a lower-dimensional space, also called latent space or code, and to decode it back to its original dimensional space while trying to minimize the reconstruction error. Subsequently, the latent space and decoder can serve synthetic data generation while minimizing divergence between probability distributions (Borysov et al., 2018; Salim, 2018). The outcome is a reconstructed dataset that conserves core statistical distributions without including any original information. Some tools have been developed for synthetic data generation using VAE such as in the Synthetic Data Vault, a Synthetic Data Generation ecosystem of libraries in Python (GitHub sdv-dev, SDV: Synthetic Data Generation for tabular, relational & time series data, 2021). Generative Adversarial Networks or GAN is another unsupervised generative model that has gained in popularity, thanks to its performance upon imagery, music, speech, text generation. GANs oppose two neural networks, the generator model and the discriminator model. The first generates new data from the input while the second aims at classifying the generated outcomes as either real or fake. The two models are trained in parallel until the discriminator is no longer able to classify the generated data, meaning that the generator is generating plausible examples. The two algorithms will be implemented and tested against test data. Ethical considerations are at the centre of the synthetic population technology as its main purpose is to leverage microdata without violating safety and privacy.

82

C. Kuster et al.

Beyond the technological “safeguard”, a methodological approach has also been favoured for its implementation. Indeed, a fit-for-purpose approach will be followed. The number of individuals as well as the attributes to be synthesized will be tailored to the case study and carefully selected on the basis of their relevance in the study. Such stance ensures (1) better performance of the algorithm because restricted, (2) a tailored made synthetic population that does not transgress ethical boundaries. In the present case, datasets required for microsimulations and population synthesis have been explicated by Miller et al. (1999), namely: . . . . . . . .

Urban land use and buildings stock, Households and organizations location, Demographic evolution of the population, Evolution of business establishments and jobs, Residence-Work/School linkages, Household ownership, Individual daily multi-modal travel, Good and services multi-modal movement

In addition, individual health-related, and COVID-19 specific data will be added when available. The synthesis will also be limited to specific locations of the case studies.

2.2 Probabilistic Activity modelling Activity-Based modelling is a part of transportation or mobility modelling where one aims at predicting individuals’ movement based on a set of activities (e.g. work, leisure, shopping, …) (Ettema, 1996). In this perspective, the travel’s demand is proxy for activities demand and agents are placeholders for whom we model behaviour within a network of activities. Various models have already been implemented such as TRANSIMS (TRansportation ANalysis Simulation System) (TRANSIMS, 2009) and ALBATROSS (A Learning-Based, Transportation-Oriented Simulation System) (Arentze & Timmermans, 2004). In practice, an activity-based approach for transportation that seeks satisficing heuristics relies on time-use survey (TUS) (Lawton, 1997). TUS collects data regarding all activities (in and out) followed by individuals over the course of a certain period (likely a day or multiple days). In the EU, the Harmonised European Time Use Surveys (HETUS) has been launched with a 10-min-based collections of various activities, including paid and unpaid work, household and family care, personal care, social life, travel and leisure (European Union, 2018). Thanks to the population synthesis, each and every individual of the synthetic population is assigned a diary for a specific day of the week. Imputation is performed upon common set of demographic and socio-economical attributes such as the age group, gender, household composition, labour status and more.

The Social Digital Twin for Liveable Cities: A COVID-19 Case Study

83

Open-source framework such as MATsim have already been used to complete multimodal agent-based transport simulations from activity-based modelling with great accuracy at aggregated levels (Bekhor et al., 2011; Delhoum et al., 2020). The framework is set using the synthetic population as well as GIS land use data and public transportation network and schedules. The outcome of such simulation is presented in the form of detailed itineraries for each individual. Those itineraries are subsequently used to set the boundaries conditions of the micro-level crowd simulation.

2.3 Crowd Simulation Crowd simulation is the process of simulating large numbers of agents interacting with their environment and with each other. It is an agent-based modelling where the agents are assigned a specific set of goals and behaviours. Crowd simulation emerged in the late 80 with behavioural models that aimed at modelling the flocking behaviour of birds (Reynolds, 1987). Following this early developments, the domain has improved greatly helped by great efforts in the fields of visual effects (Musse, 2001), games and safety policies (Moussaïd et al., 2016). Because crowd simulation is omnipresent in the field of computer games, game engines are proven technologies for such tasks. Moreover, game engine has capabilities to load and render 3D digital twins for a complete description of the simulation environment. As such, they are good candidates for digital twin and behavioural modelling, principal components of a crowd simulation.

2.3.1

Digital Twin

A digital twin is “a digital replica of a living or non-living physical entity. By bridging the physical and the virtual world, data is transmitted seamlessly allowing the virtual entity to exist simultaneously with the physical entity” (Joint Research Centre, 2021). Such a model can embed detailed information about infrastructures, terrains, subsurface and more, as well as semantic information across domains and scales. The objective is the creation of a holistic system where all relevant or desired attributes could be represented and linked. This system forms the base for a city information model that serves as a geodata hub for domain-specific spatial and temporal information (Lehner & Dorffner, 2020). In addition, a digital twin offers capabilities for modelling and simulation and can serve as a canvas to perform such tasks. For instance, White and al. have leverage a digital twin of the city of Dublin in order to perform flooding simulation, taking advantage of the 3D terrain, rainfall and river levels data (White et al., 2021); the DUET project has worked on the Digital Twin of the city of Antwerp and the capability to integrate noise model (Raes et al., 2021);

84

C. Kuster et al.

Fig. 2 BGT abstraction (Kadaster, 2021)

the SPHERE project has developed a BIM Digital Twin Platform in view to optimize the building lifecycle, reduce costs and improve energy efficiency in residential buildings (Alonso et al., 2019). The current Digital Twin (DT) implementation relies upon open geodatasets delivered by the Dutch cadastre, namely: . the Basisregistratie adressen en gebouwen (BAG) which contains registration information about addresses and buildings in the Netherlands, such as year of construction, surface area, purpose of use and location . Basisregistratie Grootschalige Topografie (BGT), a 20 cm-accurate land-use and topographic dataset. Figure 2 gives an overview of the land use classification with a distinction between erf (property), voetpad (footpath), rijbaan (roadway) and many more. Land classification is important for pathfinding of pedestrian who favoured footpath and walkable areas. . the Actueel Hoogtebestand Nederland (AHN), a point-cloud elevation map of the Netherlands. In addition, Open Street Map (OSM) data are used to complete any missing data, especially point of interests that might be missing or out-of-date. There are numerous technologies that can support the creation of digital twins and one must choose the most appropriate ones against his/her case study. In the present project, the DT should hold capabilities for dynamic crowd simulation. Games engines such as Unity have proven being able to render 3D Digital Twins as well as performing crowd simulation (White et al., 2021). In addition, the technology offers the possibility to interact with the simulation, as opposed to more common parameterized simulation, which enable end-user interaction and scenarios modelling. Figure 3 presents the overall architecture for the Digital twin generation using open GIS data. Spatial information taken from the Dutch cadastre and OSM are stored within a geodatabase (Postgresql database with PostGIS). The data have been processed to produced additional 3D datasets, namely 3D terrain and buildings, in a similar fashion to 3dfier reconstruction algorithm (Ledoux et al., 2021) for the terrain

The Social Digital Twin for Liveable Cities: A COVID-19 Case Study

85

Fig. 3 Digital twin generation stack

and Geoflow for the buildings (Ledoux et al., 2021). This value-added dataset is then fetch given a certain bounding box from a service hosted online. In the background, some data process is done using Trimesh (Borysov et al., 2018), a pure Python library for loading and using triangular meshes, in order to produce a COLLADA file (Barnes et al., 2008), an open standard digital asset schema for interactive 3D applications. The service will return a file which is composed of semantically distinct layers corresponding to each topographic layers and the buildings. The format is compatible with unity Game engine allowing the system to load all layers’ meshes. An additional service is hosted online which enable an end-user to download the urban furniture positions such as trees, light post, litter bins and more in a geojson format. The 2 files are then loaded within Unity with merging, colouring and shading the 3D objects taken care of. Note that if there are no restrictions in the extent a DT can be generated, it is yet limited by the hardware and software. A 2 × 2 km bounding box will produce a DT of around 50 MB (depending on the location). As such, there will be some memory issues when querying too large DT. Moreover, Unity game engine is efficient yet limited in performance. Loading a large DT will affect the crowd simulation performance significantly.

2.3.2

Behavioural Modelling

Behavioural models of pedestrian are at the core of crowd simulation. Each individual or agent is assigned a set of behaviours in interaction with its surrounding. Those

86

C. Kuster et al.

behaviours range across scales from microscopic such as collision avoidance and velocity to mesoscopic like group behaviour and formation to macroscopic with path finding and planning. Through the years, several behavioural models have been designed. To name a few: – Reciprocal Velocity Obstacle (RVO), Hybrid Reciprocal Velocity Obstacle HRVO (Snape et al., 2009) or optimal reciprocal collision avoidance ORCA (Berg et al., 2011) at microscopic level; – Social groups such as Moussaid and al. model (Moussaïd et al., 2010) or the Social Groups and Navigation (SGN) model (Jaklin et al., 2015) at mesoscopic level; – Cellular automata (Blue & Adier, 2000) or navigation meshes (Toll et al., 2012) at macroscopic level. In Unity, crowd simulation is natively composed of 2 principal behaviours: path finding and collision avoidance. Figure 4 taken from Unity documentation, describes the process. Two processes are run in parallel, a global one that set individual path finding from a position to another and a local one that covers steering, obstacle avoidance and movement. Path finding is performed, thanks to navigation meshes. A navigation mesh is a set of interconnected nodes and edges that cover the navigable surface. The A* Search algorithm uses the mesh to compute the shortest path between any 2 points. In addition, the edges in the mesh can hold weights. This will result in setting relative importance to portions for the overall navigation. For instance, roads can hold a higher weight than pedestrian areas so that pedestrians prioritize the latter. With a sequence of origins and destinations points for each agent, the shortest path that one must take to complete his/her journey can be computed. Of course, the shortest path is not always the chosen path in reality. Therefore, in order to obtain a better behavioural representation, those “baseline” paths can be pseudo-randomized in order to bring more variety. Obstacle avoidance is performed at a local level in order to steer our agents away from obstacles such as walls, street furniture or other agents. The steering logic

Fig. 4 Unity’s inner workings of the navigation system (Unity, 2021)

The Social Digital Twin for Liveable Cities: A COVID-19 Case Study

87

is based on RVO where an agent holds a desired velocity vector toward his/her destination. From this vector, a new vector is computed that will balance between moving in the desired direction and preventing future collisions with other agents and obstacles. By using the itineraries computed in the activity-base modelling, one can parameterize the unity crowd simulation for a specific location and period of time.

2.4 Case Study The system designed in Sect. 3.1–3.3 was used to create a digital twin around a specific theme: COVID-19 crisis management, even though other cases, such as mobility or event management could also be considered. When implementing the system, several steps needed to be defined. First, the intended users needed to be defined. The second step was to define upon policymaking scenario’s that the methodology would be applied. This step consists of two considerations: what scenarios can be tackled with this methodology and what questions the policymakers need answers to. The final step is to generate an applied workflow that allows the users of the methodology. This needs to take into account their interactions with the different elements, within the constraints of the modelling capabilities. We have achieved this by assembling an expert panel, consisting of several different policymakers with different responsibilities at the local scale in managing the COVID-19 crisis in the Netherlands. The local scale was selected, as this is the lowest scale of modelling (in the microsimulations), and the scale of visualization in the digital twin.

2.4.1

Expert Panel

At the start of the COVID-19 crisis in the Netherlands, there were several stakeholders that were responsible for designing effective measures to control the crisis. The national government set the rules and boundary conditions that were applied to the entire country. However, at the regional and local scale, the safety regions and municipalities were responsible for applying the national rules and make local policies that allowed for the implementations of these rules. Furthermore, the government institute for public health and the environment (RIVM) is responsible for the gathering knowledge on the COVID-19 crisis and insight in the public behaviors in the COVID-19 crisis. Based on this context, an expert group was assembled of people within the safety regions, local government and RIVM to help designing the system described in this study. Within these three categories, there are still different stakeholders that were assembled in this focus group. There were people from the local government and safety regions with expertise in event management (crowd control), direct advisors of the mayor, public health and IT services. For the local government it was ensured that both large and small municipalities were included, as different needs might arise

88

C. Kuster et al.

between these municipalities. Furthermore, from the RIVM people were included who study behavior of people specifically related to the COVID-19 crisis, as well as those who have expertise in risk factors of COVID-19 hospitalization. The first step was to interview all the experts in the expert group. The interview consisted of 9 open questions, to understand the responsibility of the person interviewed, the bottlenecks that they encounter during the management of the COVID-19 crisis and the needs they have for analyzing and communicating scenarios to different stakeholders in their daily work. The interview protocol can be found in Appendix 1. After the interviews, the findings of the expert panel were used to make a first version of the system described in this paper. This first version was presented to the expert panel, which was now gathered together to share their insights and give feedback on the design choices made. These group sessions occurred three times, in which feedback was given and (if possible) implemented in the next version. Some of the requests of the policymakers could not be implemented due to lack of data or modelling capabilities (e.g., the chance that a citizen would apply for additional benefits from their municipality due to COVID-19). This led to the final system with several scenarios and tools that were implemented in the digital twin, that allowed the experts from our expert panel to use the system to understand and communicate solutions in the COVID-19 crisis. The final system is described in Sect. 3.

3 Results 3.1 Tailored Digital twin design One of the initial questions that was asked by the focus group, was where and how the COVID-19 virus would spread. However, in discussion with the RIVM, it was decided that the knowledge of the spreading of COVID-19 virus and the detail of the methodology developed in this study would not allow for reliably modelling these outcomes. Therefore, together with the focus group it was decided to allow policymakers to watch the current scenarios in a digital twin that shows a lot of detail about obstacles and possible destinations. For the digital twin, the datasets open street map, BAG (building and address administration), AHN2 (pointcloud with height data) and BGT (detailed topological data, that also includes, for instance, park benches), as described in Sect. 3.3.1. When the environment was set up, functionality to move through the digital twin, such as panning and zooming and a sliding bar to move through the time, was implemented. Then, the COVID-19 specific data and scenarios were evaluated. For instantaneous feedback, the digital twin contains measures to block the street, make a street one-way and the forming of cues when a certain number of people are in a building is implemented (Fig. 5). However, when in the 3D view, it was difficult to evaluate the impact of a policy measure. Therefore, a split screen feature was implemented that allowed the users to see the impact in a heatmap (Fig. 6). The

The Social Digital Twin for Liveable Cities: A COVID-19 Case Study

89

Fig. 5 The dashboard showing one of the crowd management scenarios that policymakers could implement: blocking a road

scenarios that require larger computational time can be integrated into the digital twin, which can then be tested with the same integrated features.

3.2 COVID-19 Synthetic Population When applying the methodology designed in this study to the case study of COVID19 management in the Netherlands, one has to find data and assumptions that are relevant to the Dutch circumstances, and fit the questions from the expert group, as discussed in Sect. 2.4.1. The data used to construct the synthetic population, was taken from non-public microdata from Statistics Netherlands (CBS) and the decisions for including parameters that influence the risk of hospitalization from COVID-19 from the RIVM. These consist of smoking, obesity, age, gender and underlying suffering (e.g., COPD, heart failure, etc.). The risk of hospitalization increases when a person’s description consists of more risk factors. Apart from the COVID-19 specific properties, the synthetic population required data that allowed the diaries to be attached to the synthetic population. This consists of data about the employment status of the person, whether they have children to take to school, the assigned dwelling, etc. Obtaining the required information on employment for the synthetic people is an example of a model that combines several datasets from the CBS. Initially, a distinction is made between those too young or old to work. Next, data from CBS is used to distinguish between those on health or social benefits and those that are (self) employed. Next, data on sectorial employment is used in combination with the educational level to generate a more detailed form. Finally, data on location of employment from the CBS is used to not only know in

90

C. Kuster et al.

Fig. 6 Heat map that can be used to gain insight into the effect of the policy measures implemented in the dashboard

what sector someone works and if he or she can work from home, but also how they travel to work with the activity model. A result of the synthetic population that could be visualized in Fig. 7. Scenarios that were implemented were the blocking of streets, which ensured that people could no longer enter a crowded street, or a very narrow street. The microsimulation showed how people would most likely disperse. Furthermore, a single road implementation was made, where people could only move through a street in one direction. Again, the microsimulation showed the most likely new routes that the people would take. And finally, a function was implemented that counted the number of people in a specific building, in combination with a queuing function that showed policymakers the effects of maximum people in space, as queues also disrupted flow and influence crowdedness.

The Social Digital Twin for Liveable Cities: A COVID-19 Case Study

91

Fig. 7 A representation of a synthetic person. Some of the properties that are relevant to the focus group are available when clicking on an individual person

4 Discussion and Future Work 4.1 Constraint and limitation This study has shown that this methodology is well-designed to answer questions of crowd modelling that fully respect the privacy and other ethical boundaries of the public that is modelled. The application of the methodology described in this study shows that it can be used for policymaking for the case study COVID-19. However, as described in the scenario definition Sect. 0, not all scenarios that policymakers would like to make to come to prudent decisions on policy can currently be explored instantaneously. This is due to the long computation time (even with high computation power), where the modelling methodology has to span a full population of a full nation. This is a limitation but does not render the methodology useless. The users can define experimental scenarios in one policymaking session that will be tested before the next session, and the modelling results can then be implemented in the digital twin for perusal. However, when quick and efficient decision-making for complex policy scenarios is required, this methodology takes too long to calculate all the implications.

92

C. Kuster et al.

Another disadvantage of working with the synthetic population methodology is that it is still based on highly privacy sensitive census data. If not managed aptly, this can constitute an even higher privacy risk to the users. This study provides many considerations that will safeguard the protection of the privacy. But working with such privacy sensitive data constitutes another risk: the environment that is needed to access the data. Census data agencies have to protect the data that they manage, which makes generating a synthetic population in their protected environment complex and additionally time consuming. Researchers using this approach for other case studies have to prepare for both the fees and hurdles associated with handling the privacy sensitive data. This extends to using the diaries that are used to build the activity models. There is only a limited number of diaries available that require specific characteristics of the synthetic population, which limit the accuracy of the final methodology. Another complicating factor of this methodology is that it combines several modelling techniques, that can be individually validated (e.g., the representativity of the synthetic population), but when combined become more difficult to validate. In the COVID-19 case study for instance, one can validate the number of people present in a specific street with flow sensors, or the distance people keep to one another with camera images. However, validating the age, weight or destination of the people passing through that street is complicated, as well as ethically complex. Therefore, this methodology has to be applied to the correct case studies, to avoid answering questions that require a higher level of validation.

4.2 Conclusion In conclusion, this paper shows that it is possible to create a system that aids policymakers to make decisions at the microscale in the Dutch policy context. The use of geospatial data and microsimulations aid the policymakers to gain insight into the behavioral patterns of their citizens and implications of their proposed local policies, without invading their privacy in any way. This system is also made suitable to test policies at a larger scale, due to the activity model that is shown in Fig. 1, but from the interviews with the expert group, these policymakers shown no direct interest in these modelling capabilities. This level of modelling was crucial to determine the local flow of people. Also, having the activity modelling might be relevant to another group of policymakers that operate at a higher governmental scale, but this is a topic for further research.

4.3 Future work The present methodological framework sets solid ground for an all-round spatial behavioural modelling, from macro to microscopic scale. Leveraging open data as

The Social Digital Twin for Liveable Cities: A COVID-19 Case Study

93

well as microdata while preserving privacy, the framework offers flexibility in its uses as no tedious measurements such as cell phone gps data or flow sensors are required. In future works, the versatility of the approach should be demonstrated through implementations across various domains such as health and safety, energy transition and/or GHG emissions. Indeed, the methodology or some of its components can apply to any spatial behavioural studies that by proxy influence those themes. On a more technical perspective, calibration and validation could be performed within field labs where flow sensors are deployed, an important step in the completion of a satisficing solution. Finally, following the case study and domain above mentioned, new scenarios and behavioural models must be implemented. Therefore, additional efforts should be toward the rule engine at both macro and micro scale and the ability to plugin those new rules seamlessly. Testing these with new stakeholders that also include stakeholders at higher levels of government with different policy interests would validate this approach even further.

5 Appendix 1 VoorstelleN Introduceer jezelf en de high impact COVID case. Denk hierbij aan punten als: – – – – –

Synthetische populatie om privacy problemen te voorkomen Gedragsmodelering 3D visualisaties (digital twin) en interacties Beleid modeleren en visualiseren Maar ook de beperkingen: gebrek aan inzicht in hoe COVID zich gedraagt, en het is een model Introduceer ook het Franse voorbeeld, met als drie resultaten:

1 Waarschijnlijk aantal sociale contacten op basis van economische en sociale factoren. Bijvoorbeeld, iemand met schoolgaande kinderen bij de school. 2 Per economische activiteit hoeveel interacties zij hebben, op basis van verplaatsingen per auto of OV, thuiswerken, etc. 3 Evalueren van de ruimtelijke spreiding van de risico’s op basis van sociaaleconomische factoren in de regio en transportpatronen. Deze presenteren we om wat verwachtingen te managen. Vraag 1: Kan je jouw functie in de COVID crisis introduceren? Vraag 2: Welke verantwoordelijkheden horen daarbij? Vraag 3: Welke afhankelijkheden van andere partijen heb je daarbij? En welke interacties/synergiën?

94

C. Kuster et al.

Knelpunten Vraag 4: Welke knelpunten kom je tegen? Waarop loopt bepaalde besluitvorming vast? Vraag 5: Welke informatie/data zou kunnen helpen om deze knelpunten te verminderen/verwijderen? Vraag 6: Kan je 1 of 2 voorbeelden geven waarbij je deze knelpunten recent bent tegengekomen? Kijk of je hierbij kan sturen op voorbeelden waarbij wij streven naar een oplossing, maar niet te hard, anders kom je misschien niet uit bij de echte vragen die zij hebben. Communicatie Vraag 7: In de voorbeelden, naar wie moeten ze die informatie/oplossingen communiceren? Vraag 8: Op welke manier vindt deze communicatie nu plaats? Vraag 9: Zou een digital twin/visuele demo’s hierbij kunnen helpen?

References Alonso, R., Borras, M., Koppelaar, R. H. E. M., Lodigiani, A., Loscos, E., & Yöntem, E. (2019). SPHERE: BIM Digital Twin Platform. Proceedings, 20(1), 9. https://doi.org/10.3390/proceedin gs2019020009 Al-Sehrawy, R., Kumar, B., & Watson, R. (2021). A digital twin uses classification system for urban planning & city infrastructure management. Journal of Information Technology in Construction, 26, 832–862. https://doi.org/10.36680/J.ITCON.2021.045 Arentze, T. A., & Timmermans, H. J. P. (2004). A learning-based transportation oriented simulation system. Transp. Res. Part B Methodol., 38(7), 613–633. https://doi.org/10.1016/j.trb.2002.10.001 Axhausen, K. W. (2016). The multi-agent transport simulation MATSim. Azevedo, C. L., et al. (2016). Microsimulation of demand and supply of autonomous mobility on demand. Transportation Research Record, 2564(1), 21–30. https://doi.org/10.3141/2564-03 Barnes, M., Finch, E. L., & S. C. E. Inc. (2008). COLLADA—digital asset schema release 1.4.1 specification (2nd edn). Elements. Barrett, J. C. (1967). A Monte Carlo study of reproduction. Human Biology Symposium, 81–97. Beckman, R. J., Baggerly, K. A., & McKay, M. D. (1996). Creating synthetic baseline populations. Transportation Research Part A Policy and Practice, 30(6), 415–429. https://doi.org/10.1016/ 0965-8564(96)00004-3 Bekhor, S., Dobler, C., & Axhausen, K. W. (2011). Integration of activity-based and agent-based models: Case of Tel Aviv, Israel. Transportation Research Record, 2255, 38–47. https://doi.org/ 10.3141/2255-05 Van Den Berg, J., Guy, S. J., Lin, M., & Manocha, D. (2011). Reciprocal n-body collision avoidance. Springer Tracts Adv. Robotics, 70, 3–19. https://doi.org/10.1007/978-3-642-19457-3_1 Berry, A. (2019). The distributional effects of a carbon tax and its impact on fuel poverty: A microsimulation study in the French context. Energy Policy, 124, 81–94. https://doi.org/10.1016/ j.enpol.2018.09.021 Blue, V. J., & Adier, J. L. (2000). Cellular automata model of emergent collective bi-directional pedestrian dynamics. Artificial Life VII. The MIT Press.

The Social Digital Twin for Liveable Cities: A COVID-19 Case Study

95

Borysov, S. S., Rich, J., & Pereira, F. C. (2018). Scalable population synthesis with deep generative modelling. https://doi.org/10.1016/j.trc.2019.07.006 Boukerche, A., & Coutinho, R. W. L. (2019). Crowd management: The overlooked component of smart transportation systems. IEEE Communications Magazine, 57(4). https://doi.org/10.1109/ MCOM.2019.1800641 Delhoum, Y., Belaroussi, R., Dupin, F., & Zargayouna, M. (2020). Activity-based demand modeling for a future urban district. Sustainability, 12(14), 5821. https://doi.org/10.3390/su12145821 Edwards, K. L., & Clarke, G. P. (2009). The design and validation of a spatial microsimulation model of obesogenic environments for children in Leeds, UK: SimObesity. Social Science and Medicine, 69(7), 1127–1134. https://doi.org/10.1016/j.socscimed.2009.07.037 Ettema, D. F. (1996). Actvity-based travel demand modeling. European Union (2019). Harmonised European Time Use Surveys (HETUS) 2018 Guidelines. Gebruik telecomdata tegen corona kan alléén met wet | Autoriteit Persoonsgegevens. [Online]. Available: https://autoriteitpersoonsgegevens.nl/nl/nieuws/gebruik-telecomdata-tegen-corona-alleenmet-wet. Accessed 15 Dec 2021. GitHub (2021). sdv-dev/SDV: Synthetic data generation for tabular, relational and time series data. [Online]. Available: https://github.com/sdv-dev/SDV. Accessed: 19 Apr 2021. Hyrenius, H. (1965). Demographic simulation models with the aid of electronic computers. United Nations World Population Conference. Hyrenius, H., & Adolfsson, I. (1964). A fertility simulation model, pp 31-. Jacquard, A. (1967). La reproduction humaine en régime malthusien. Un modèle de simulation par la méthode de Monte-Carlo. Source, 22(5), 897–920. Jaklin, N., Kremyzas, A., & Geraerts, R. (2015). Adding sociality to virtual pedestrian groups. In Proceedings of the ACM Symposium on Virtual Reality Software and Technology VRST, vol. 13–15, pp. 163–172. https://doi.org/10.1145/2821592.2821597 Joint Research Centre (2021). Digitranscope the governance of digitally-transformed society. Kadaster (2021). Basisregistratie Grootschalige Topografie (BGT). Web. [Online]. Available: https:// docs.geostandaarden.nl/imgeo/catalogus/bgt/#topografie. Accessed: 04 Nov 2021 Dawson-Haggerty et al. (2021). mikedh/trimesh: Python library for loading and using triangular meshes. [Online]. Available: https://github.com/mikedh/trimesh. Accessed: 12 Nov 2021. Lawton, T. K. (1997) Activity and time use data for activity-based forecasting. In Activity-based travel forecasting conference: summary, recommendations and compendium of papers, vol. 103, p. 118 Ledoux, H., et al. (2021). 3dfier: Automatic reconstruction of 3D city models. Journal of Open Source Software, 6(57), 2866. https://doi.org/10.21105/joss.02866 Ledoux, H., Peters, R., Dukai, B., & Tom, C. (2021). geoflow3d/geoflow: flowchart tool for geospatial data processing. [Online]. Available: https://github.com/geoflow3d/geoflow. Accessed: 12 Nov 2021. Lehner, H., & Dorffner, L. (2020). Digital geoTwin Vienna: Towards a digital twin city as geodata hub. PFG—Journal of Photogrammetry Remote Sensing and Geoinformation Science, 88(1), 63–75. https://doi.org/10.1007/s41064-020-00101-4 Martínez, F., & Donoso, P. (2010). The MUSSA II land use auction equilibrium model. Advances in Spatial Science, 65, 99–113. Miller, E. (2018). The case for microsimulation frameworks for integrated urban models. Journal of Transport and Land Use, 11(1), 1025–1037. https://doi.org/10.5198/jtlu.2018.1257 Miller, E. J., Hunt, J. D., Abraham, J. E., & Salvini, P. A. (2004). Microsimulating urban systems. Computers, Environment and Urban Systems, 28(1–2), 9–44. https://doi.org/10.1016/S0198-971 5(02)00044-3 Miller, E. J., Kriger, D. S., & Hunt, J. D. (1999). Integrated urban models for simulation of transit and land use policies guidelines for implementation and use. Moussaïd, M., Perozo, N., Garnier, S., Helbing, D., & Theraulaz, G. (2010). The walking behaviour of pedestrian social groups and its impact on crowd dynamics. PLoS ONE, 5(4), e10047. https:// doi.org/10.1371/JOURNAL.PONE.0010047

96

C. Kuster et al.

Moussaïd, M. et al. (2016). Crowd behaviour during high-stress evacuations in an immersive virtual environment. Journal of the Royal Society Interface, 13(122). https://doi.org/10.1098/rsif.2016. 0414 Musse, S. R. (2001). Behavioral modeling of virtual human actors. RITA, 8(2), 125–141. Niu, X., & Qin, S. (2021). Integrating crowd-/service-sourcing into digital twin for advanced manufacturing service innovation. Advanced Engineering Informatics, 50. https://doi.org/10.1016/j.aei. 2021.101422 Quintana, D. S. (2020). A synthetic dataset primer for the biobehavioural sciences to promote reproducibility and hypothesis generation. Elife, 9. https://doi.org/10.7554/eLife.53275 Raes, L et al. (2021). DUET: A framework for building secure and trusted digital twins of smart cities. IEEE Internet Computing, 1–9. https://doi.org/10.1109/MIC.2021.3060962 Rephann, T. J., & Holm, E. (2004). Economic-demographic effects of immigration: results from a dynamic spatial microsimulation model. International Regional Science Review, 27(4), 379–410. https://doi.org/10.1177/0160017604267628 Reynolds, C. W. (1987). Flocks, herds and schools: A distributed behavioral model. ACM SIGGRAPH Computer Graphics, 21(4), 25–34. https://doi.org/10.1145/37402.37406 Ridley, J. C., & Sheps, M. C. (1966). An analytic simulation model of human reproduction with demographic and biological components. Population Studies (NY), 19(3), 297–310. https://doi. org/10.1080/00324728.1966.10406018 Rosenfield, A., Chingcuanco, F., & Miller, E. J. (2013). Agent-based housing market microsimulation for integrated land use, transportation, environment model system. Procedia Computer Science, 19, 841–846. https://doi.org/10.1016/j.procs.2013.06.112 Salim, A. (2018). Synthetic patient generation: A deep learning approach using variational autoencoders. Salvini, P., & Miller, E. J. (2005). ILUTE: An operational prototype of a comprehensive microsimulation model of urban systems. Santana, J. R., Sanchez, L., Sotres, P., Lanza, J., Llorente, T., & Munoz, L. (2020). A privacy-aware crowd management system for smart cities and smart buildings. IEEE Access, 8, 135394–135405. https://doi.org/10.1109/ACCESS.2020.3010609 Sindagi, V. A., & Patel, V. M. (2018). A survey of recent advances in CNN-based single image crowd counting and density estimation. Pattern Recognit. Lett., 107, 3–16. https://doi.org/10. 1016/j.patrec.2017.07.007 Snape, J., Van Den Berg, J., Guy, S. J., & Manocha, D. (2009). The hybrid reciprocal velocity obstacle, vol. 27599, pp. 1–11. Strauch, D. et al. (2004). Linking transport and land use planning: The microscopic dynamic simulation model ILUMASS. GeoDynamics, 295–312. https://doi.org/10.1201/978142003810 1-25 Van Toll, W. G., Cook, A. F., & Geraerts, R. (2012). A navigation mesh for dynamic environments †. Computer Animation and Virtual Worlds, 23, 535–546. https://doi.org/10.1002/cav.1468 The World Bank (2020). Urban development overview. The World Bank TRANSIMS (2009). {TR}ansportation {AN}alysis and {SIM}ulation {S}ystem. Los Alamos, NM. Unity (2021). Unity—Manual: Inner workings of the navigation system. Unity Documentation. [Online]. Available: https://docs.unity3d.com/Manual/nav-InnerWorkings.html. Accessed: 14 Dec 2021. Waddell, P. (2011). Integrated land use and transportation planning and modelling: Addressing challenges in research and practice. Transport Reviews, 31(2), 209–229. https://doi.org/10.1080/ 01441647.2010.525671 Wager, L. W., Orcutt, G. H., Greenberger, M., Korbel, J., & Rivlin, A. M. (1962). Microanalysis of socioeconomic systems: A simulation study. American Sociological Review. https://doi.org/10. 2307/2089823 White, G., Zink, A., Codecá, L., & Clarke, S. (2021). A digital twin smart city for citizen feedback. Cities, 110. https://doi.org/10.1016/j.cities.2020.103064

The Impact of Differing COVID-19 Mitigation Policies: Three Natural Experiments Using Difference-in-Difference Modelling Kingsley E. Haynes, Rajendra Kulkarni, Abu Siddique, and Meng-Hao Li

1 Background1 As the COVID-19 spread across the globe and became a pandemic, unlike much of the world where a central governing body has the authority to respond to health emergencies, in the U.S. with its federal governing structure, this became a 50 state response to fight the disease. Each state’s governor issued that state’s emergency response (https://www.nga.org/coronavirus-state-actions-all/) through a series of state level executive orders, spelling out in details a variety of action plans that her/his government had the constitutional authority to execute so that the state would be able to manage the growing threat of COVID-19. The main goals of this multitude of mainly non-pharmaceutical intervention plans were to be able 1. to slow the growth and spread of the disease, 2. to reduce the burden on the existing health care systems, 3. to cope with the growing health emergencies.

1 Please excuse some degree of overlap between this paper on the pandemic in county and county regions and our earlier reported study on the effects of the pandemic on specific states (10). Since counties are sub-divisions of states and states set policy but counties implement them, it is almost impossible to discuss one without the other.

K. E. Haynes (B) · R. Kulkarni · A. Siddique · M.-H. Li Schar School of Policy and Government, George Mason University, Arlington, VA 22201, USA e-mail: [email protected] R. Kulkarni e-mail: [email protected] A. Siddique e-mail: [email protected] M.-H. Li e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. G. Celbi¸s et al. (eds.), Pandemic and the City, Footprints of Regional Science, https://doi.org/10.1007/978-3-031-21983-2_7

97

98

K. E. Haynes et al.

Even though each state appears to follow a different path when it comes to implementation details as well as duration of these plans (https://en.wikipedia.org/wiki/U. S._state_and_local_government_responses_to_the_COVID-19_pandemic), the 50 states fall into two broad categories: states where governors issued “lockdown” or “stay-at-home” orders and the states that did not. There were just five states that did not require its residents to “stay-at-home,” namely, Arkansas, Iowa, South Dakota, North Dakota, and Nebraska. However, the effectiveness of “stay-at-home” orders in containing the spread of COVID-19 remains controversial. Some studies indicated that “stay-at-home” orders successfully limited the transmission and hospitalization of COVID-19. For example, China quarantined infected people and potentially infected people and restricted travel to and from affected areas (Ferguson et al., 2006; Kraemer et al., 2020; Maier & Brockmann, 2020; Tian et al., 2020). In the U.S., studies found that “stay-at-home” orders reduced COVID-19 cases and hospitalizations (Padalabalanarayanan et al., 2020; Sen et al., 2020). Nevertheless, for states that implemented “stay-at-home” orders, those with larger African American populations were likely to have higher COVID-19 case rates (Padalabalanarayanan et al., 2020). Additionally, using negative binomial regression with controls for sociodemographic and medical factors, research revealed that “stay-at-home” orders had no significant effects on the transmission of COVID-19 (Krishnamachari et al., 2021). One of the reasons to explain the divergent research findings is that research designs are often not sufficiently scaled to effectively evaluate the “stay-at-home” intervention. To bridge this research gap, we will look at the states that are neighbors but with different mitigation management policies and then turn to local border region differences between these neighboring states. With this approach, we can conduct nature experiments in two similar sociodemographic regions and compare COVID-19 transmission between regions with and without “stay-at-home” orders. In this paper, the “stay-at-home” has also been referred to as “shelter-in-place” or “shutdown” or “lockdown” or “safer-at-home”. For clarity and as a matter of convenience, the following phrases refer the stay-at-home states as “stay@home” states and those that did not are referred to as “not-stay@home” states. These circumstances—unfortunate as they are given the potential for worse health outcomes due to COVID-19 disease—present a unique opportunity to test the outcomes of a natural experiment. How effective were stay-at-home orders as a policy intervention measure? Do they result in slowing the growth of the disease computed in concrete terms? Is there a reduction in the rate of infections as well as a reduction in the total (or cumulative) infections at the end of this policy intervention stay-at-home interval? In our previous efforts (Haynes et al., 2022) we discussed the findings for the natural experiments with specific policy intervention analysis for the following pairs of neighboring states. 1. Iowa (IA) and Illinois (IL). 2. North and South Dakota (ND-SD) and Minnesota. 3. Arkansas (AR) and Mississippi (MS).

The Impact of Differing COVID-19 Mitigation Policies: Three Natural …

99

In our current efforts we briefly discuss the methodology and findings of the natural experiments with policy intervention analysis for border areas of the above pairings of states. We analyze how policy intervention at the level of border county regions affected the number of COVID-19 infections per 10,000 population. The border county regions are: A. Counties along the border between stay@home Illinois (IL) and not-stay@home Iowa (IA). B. Counties along the border between stay@home Minnesota (MN) and the two not-stay@home states, South Dakota (SD) and North Dakota (ND). C. Border county region between stay@home Mississippi (MS) and notstay@home Arkansas (AR). For convenience, the state and border county regions are also referred to as studyareas or study-regions. To determine the effect of the policy intervention measure on the number of infections in the study-regions during this period of policy intervention, we used a Difference-in-Differences assessment model as in Peter et al. (2017) and Lyu and Wehby (2020). The basic assumption is that neighboring study areas display specific trends in rates of infection relative to each other before a policy change in one of these study areas. This applies to local border county areas when they are part of pair of neighboring states. In other words, both sides of the border experiencing growth in the infection would not see any change unless one of the states adopts a policy change while the other does not. In such a scenario, the null hypothesis is to assume that policy intervention would have no effect on the outcome of the state’s local county border region rates. However, if the adjacent state’s border county regions with policy intervention show a statistically significant change then we reject the null hypothesis. We acknowledge at the state level even neighboring states exhibit more heterogeniety with differences in densities, metropolitan areas, economic structures and employment patterns. In an earlier study we used Difference-in-Difference analysis across all counties in all states that were being compared and in spite of high heterogeniety we found the policy differences were associated with differences in state covid infection outcomes (Haynes et al., 2022). This was an issue in our earlier state by state analysis (Haynes et al., 2022). This paper reports on side-by-side county border regions which are naturally more homogeneous than state-by-state comparisons. Given the previous state-wide stateby-state results and the greater heterogeneity of spatial unit inputs we hypothesize that with lower heterogeneity (greater homogeneity) the impact of the different policy impacts will be stronger. While the Diff-in Diff is a popular quasi experimental method to study the treatment effect among the population, we apply this approach in this paper on a spatial level—treated and non-treated counties. The validity of this Diff-in Diff model depends on the equal trend assumption. This was difficult to test in our unique case of COVID-19 pandemic since the NPI policy was implemented early in the pandemic in the USA and the growth of the pandemic was swift with little lead time. As a

100

K. E. Haynes et al.

result we could not observe a long time series to verify general trend assumptions. We have proceeded within these constraints.

2 The Diff-in-Diff Regression Model for COVID-19 Infections The description below applies to any two neighboring spatial units that share a common border.2 For convenience and without loss of generality, the explanation below is described in terms of two neighboring states A and B with a border area that consists of n counties such that n = Ca + Cb where Ca and Cb are counties in A and B respectively on that state’s border with the other. Both states have covid-19 infections in their respective counties. State B decides to implement “Stay-at-home” policy intervention to control infections for a duration T. Let InfP10K(t, i) represent the number of infections per 10K population in county i at time t. Infections, per 10K population, at time t in a spatial unit i can be expressed as a summation of an intercept, a spatial term, a temporal term and an interaction term between the spatial and temporal terms. The spatial term refers to a dummy variable C(i) that represents whether county i is in a stay@home state. Further C(i) = 1 occurs when county i is in a stay@home state, otherwise it is zero. The temporal term is a dummy variable D(t) that refers to whether the time variable is in the interval (duration T) or outside of it. Thus D(t) = 1 when the intervention policy is in effect and zero outside the interval. The interaction term is just a plain multiplication between the two dummy variables D(t) and C(i). After running the Diff-in-Diff regression model, the value of the coefficient of the interaction term will represent the number of infections per 10K population per day for the counties in the stay@home state where the “stay-at-home” NPI is used as a mitigation policy to reduce number of COVID-19 infections. Then the Diff-in-Diff regression model is represented by the following: I n f P10K (t, i ) = β0 + β1 D(t) + β2 C(i ) + β3 D(t) ∗ C(i )+ ∈

2

A Diff-in-Diff model does not require two entities to have a geographic border. However, for our current natural experiment, we are restricting our Diff-in-Diff model only to neighboring spatial units (states or counties) that share a geographic border.

The Impact of Differing COVID-19 Mitigation Policies: Three Natural …

101

where β0 is an intercept and ∈ is the error term and, β3 is the coefficient of the interaction term D(t) ∗ C(i ). The interaction term is 1 only for counties that are part of the policy intervention group and during the duration of the policy intervention. The Diff-in-Diff model was run on the cross-sectional data of daily covid-19 infections per 10K population for the respective border county regions as mentioned in the previous section. The model is run from the time (in days) when first case appeared in one of the study-areas or study-regions, through the start of the stayat-home order to the end of the stay-at-home order in that study-area/region. Below is a brief description of the data and the model output for each of these natural experiments.

3 Discussion and Results In this section we briefly outline discussion of the natural experiments for neighboring county border comparisons.3 The John Hopkins University “time-series confirmed COVID-19 _US” county data is used for the analyses (https://github.com/CSSEGISandData/COVID-19/tree/ master/csse_covid_19_data/csse_covid_19_time_series). The cumulative confirmed COVID-19 infections by the county covers the time interval from Jan 22, 2020 to July 31, 2020 for border-county regions and from Jan 22, 2020 to Aug 31, 2020 for neighboring states. The following steps are common to both datasets. The raw cumulative confirmed COVID-19 infections by the county was processed with the following steps. 1. A non-decreasing cumulative county level COVID-19 infections time series data4 was generated. 2. The data from step 1 was normalized with 2019 population estimates from the Census (https://www.census.gov/data/datasets/time-series/demo/popest/2010scounties-total.html) to generate non-decreasing cumulative county level COVID19 infections per 10,000 population time series. 3. The output of step 2 was further processed to compute daily COVID-19 infections per 10,000 population by county. 4. A subset from the output of step 3 was generated for the following states: Arkansas, Illinois, Iowa, North Dakota, South Dakota, Minnesota and Mississippi. 3

The JHU time series cumulative confirmed covid-19 cases by county level data was downloaded at two different dates. The Diff-in-Diff analyses for neighboring states has covid-19 data up to Aug 31, 2020 while the one reported here was used for border county regions has covid-19 data up-to Jul 31, 2020. 4 A time series with daily cumulative counts is expected to be a non-decreasing by date, i.e., the values either stay the same or increase with time as fresh counts are added. However, in the JHU time series data, the confirmed COVID-19 cases may show an aperiodic drop in the cumulative counts. This is part of noise to be managed (by 10 day cumulation levels) as the data is compiled continuously from various sources.

102

K. E. Haynes et al.

Step 4 data (daily infections per 10k population also referred to as crosssectional PInf10K) was used as input for the Diff-in-Diff analysis of the border area counties between these pairs of states. The pairs of neighboring state border area counties on each side of the common border are referred to as Study Area A and Study Area B where Study Areas B followed a “stay-at-home” NPI for time duration T while Study Area A did not follow this NPI. 5. The time series data for each study area covering duration T was serialized by date and further processed by adding time D(t), location C(i ) dummy variables and the interaction variable D(t) ∗ C(i). The time D(t) dummy variable signifies the time period of the stay-at-home or shutdown while the location of the variable C(i ) is set to 1 for counties with the lockdown policy and zero otherwise. 6. The resulting cross-sectional data for each border county region spans the time interval from the time first case was reported in either side of the border to the end of the lockdown period for one of the study areas. 7. The results obtained with Diff-in-Diff models are for each study area for 10day intervals as well as for the entire duration that spanned each study area’s “stay-at-home” NPI implementation. The following section presents a brief background on the natural experiments for each of the neighboring states the results of which are discussed elsewhere (Haynes et al., 2022). This is necessary as the states set the framework for county (sub-state) regions. We then focus on the discussion of Diff-in-Diff results for the border county regions for each of the pairs of the neighboring states.

4 Natural Experiments for the Neighboring State Comparisons For each of the following pairs of neighboring states, one of the states followed a “stay-at-home” equivalent NPI policy for a limited duration compared its neighbor(s) which did not enforce such a policy.

5 Stay@home Illinois (IL) and Not-stay@home Iowa (IA) There are total of 201 counties in this study area, of which 102 counties are in Illinois and Iowa has 99 (Map 1). Illinois Governor issued a “Stay-at-home” order (https://www2.illinois.gov/sites/gov/Documents/APPROVED%20-%20Coro navirus%20Disaster%20Proc%20WORD.pdf) that took effect on Mar 22, 2020 and was lifted (https://www2.illinois.gov/sites/gov/Documents/CoronavirusDisaste rProc-5-29-2020.pdf) on May 30, 2020; for a total duration of 71 days, while Iowa did not. Iowa recorded its first infections in Johnson county on Mar 09, 2020 while

The Impact of Differing COVID-19 Mitigation Policies: Three Natural …

103

Map 1 Illinois (IL) and Iowa (IA) counties

Cook county, IL recorded its first infection as early as Jan 24, 2020. The Diff-inDiff analyses includes only those infections that occurred between Mar 01 and Jun 20, 2020. Throughout this time period, Illinois continued to have higher number of infections per 10K population compared to Iowa (Fig. 1). For the entire shutdown interval (Mar 21 to May 30, 2020), the total number of confirmed cases in the Iowa grew from 5 to 19,244 while the confirmed cases in Illinois counties grew from just a few to 118,927. The corresponding data normalized by population for each side are 60.99 per 10K pop. for IA and 93.84 per 10K pop. for IL (Table 1). These numbers grew quickly for Iowa over the next couple of months, by the end of Aug 31, 2020, Iowa had more confirmed cases per 10K pop (206) to Illinois (187 per 10k pop.) (see Fig. 2).

104

K. E. Haynes et al.

Fig. 1 Statewide infections per 10K population in all of Illinois and Iowa counties during Illinois’ shelter-in-place of 71 days duration

Table 1 COVID-19 cumulative total and per 10K population infections in Iowa and Illinois

Iowa

Illinois

Pop 2019

31,55,070

1,26,71,821

Total cases May 30

19,244

1,18,917

Cases per 10K pop May 30

60.99

93.84

Total cases Aug 31

65,139

2,36,724

Cases per 10K pop Aug 31

206.46

186.81

6 Stay@home Minnesota (MN) and Not-stay@home North and South Dakota (ND-SD) There are total of 206 counties in this study area, of which 87 counties are in Minnesota; 66 in North Dakota and 53 in South Dakota, for a combined total of 119 counties between North and South Dakota (Map 2). Minnesota observed “Stay-at-home” policy (https://mn.gov/governor/assets/ 3a.%20EO%2020-20%20FINAL%20SIGNED%20Filed_tcm1055-425020.pdf) from Mar 25, 2020 to May 18, 2020 (https://mn.gov/governor/assets/EO%202056%20Final_tcm1055-433768.pdf); for a total duration of 53 days, while there was no “shutdown” in either of the neighboring states of North Dakota and South Dakota. Minnesota recorded its first infection in Ramsey County on Mar 06, 2020

The Impact of Differing COVID-19 Mitigation Policies: Three Natural …

105

Fig. 2 Statewide infections per 10K population in all of Illinois and Iowa counties from beginning of March 2020 to end of August 2020 during Illinois’ shelter-in-place of 71 days duration

Map 2 Counties in Minnesota (not_stay@home_states)

(a

stay@home_state),

North

and

South

Dakota

106

K. E. Haynes et al.

while North and South Dakota, each recorded one case on Mar 11, 2020. For the Diff-in-Diff analyses, only those infections that occurred between Mar 01 and Jun 20, 2020 are included. Throughout this time period, Minnesota continued to have a higher number of infections per 10K population compared to either North Dakota or South Dakota (Fig. 3). For the entire shutdown interval (Mar 27 to May 18, 2020), the total number of confirmed cases in the two Dakotas rose to 5,968 cases while Minnesota saw nearly three times that number (16,322). The corresponding data normalized by population for each side are 36.2 per 10K pop. for the two Dakotas and 29 per 10K pop. for Minnesota (Table 2). The total infections continued to grow over the next two months and by end of Aug 31, 2020; the two Dakotas had nearly 20 more cases per 10K population (153.8) compared to Minnesota (134.5 per 10k pop.) (Fig. 4).

Fig. 3 Statewide infections per 10K population in South and North Dakotas and Minnesota counties during Minnesota’s 53-day shelter-in-place shutdown duration

Table 2 Statewide COVID-19 cumulative total and per 10K population infections

South and North Dakotas

Minnesota

Pop 2019

16,46,721

56,39,632

Total cases May 18

5,958

16,372

Total case Aug 31

25,325

75,864

Cases per 10K pop May 18 36.2

29.0

Cases per 10K pop Aug 31 153.8

134.5

The Impact of Differing COVID-19 Mitigation Policies: Three Natural …

107

Fig. 4 Statewide infections per 10K population in all of South and North Dakota and Minnesota counties: before, during 53-day shutdown and after

7 Stay@home Mississippi (MS) and Not-stay@home Arkansas (AR) There are total of 157 counties in this study area, of which 82 counties are in Mississippi; 75 in Arkansas (Map 3). Mississippi had “Stay-at-home” policy from Apr 03, 2020 (https://www.sos.ms.gov/content/executiveorders/ExecutiveOrders/ 1466.pdf) to May 25, 2020 (https://mcusercontent.com/08cb3e52aa1308600f84d 49ea/files/e24a9045-8ab7-4aa0-bb23-bb49d342c816/Executive_Order_1477_S afer_at_Home.pdf); for a total duration of 53 days, while there was no “shutdown” in Arkansas. Mississippi recorded its first infection on Mar 12, 2020 and Arkansas recorded one case on Mar 11, 2020. For the Diff-in-Diff analyses, only those infections that occurred between Mar 01 and Jun 20, 2020 are used. Throughout this time period, Mississippi continued to have a higher number of infections per 10K population compared to Arkansas (Fig. 5). For the entire shutdown interval (Apr 03 to May 25, 2020), the total number of confirmed cases in Arkansas rose to 18,000 while Mississippi saw the numbers rise to 24,100. The corresponding data normalized by the population for each side are 60 per 10K pop. for Arkansas and 89 per 10K pop. for Mississippi. The total infections continued to grow over the next two months and by end of Aug 31, 2020; Mississippi had 73 per 10k more infections (279 per 10K pop) than Mississippi (206 per 10k Pop) (see Fig. 6 and Table 3).

108

K. E. Haynes et al.

Map 3 Counties in Mississippi (a Stay@home_state) and Arkansas (a Not_stay@home_state)

8 Natural Experiments for the Neighboring State Border Counties Region Comparisons This section has brief discussion of each of the 4 border region counties of neighboring states that followed opposing NPI policies to fight the COVID-19 infections in their respective states. • Counties along the border between stay@home Illinois (IL) and not-stay@home Iowa (IA). • Counties along the border between stay@home Minnesota (MN) and the two not-stay@home states, South Dakota (SD) and North Dakota (ND). • Border county region between stay@home Mississippi (MS) and not-stay@home Arkansas (AR).

The Impact of Differing COVID-19 Mitigation Policies: Three Natural …

109

Fig. 5 Statewide infections per 10K population in all of Mississippi and Arkansas counties during Mississippi’s shelter-in-place shutdown duration

9 Border County Region Between Stay@home Illinois (IL) and Not-stay@home Iowa (IA) Of the total 15 counties that share the border between Iowa and Illinois, there are 8 counties on the Iowa side and 7 belong to Illinois (Map 4). Clinton, Scott and Muscatine in Iowa and Whiteside, Rock Island and Mercer in Illinois are part of the quad-city MSA. These counties are indicated in the Map and in the table, with notation [M] for metropolitan, next to their names if appropriate. Illinois Governor Pritzker’s executive order (https://www2.illinois.gov/sites/gov/ Documents/APPROVED%20-%20Coronavirus%20Disaster%20Proc%20WORD. pdf) began the statewide shelter-in-place on March 22, 2020 and stayed in effect for next 71 days. It was lifted (https://www2.illinois.gov/sites/gov/Documents/Corona virusDisasterProc-5-29-2020.pdf) on May 30, 2020. The first case reported on Mar 17 was in Whiteside County, IL while one case each was reported on Mar 19, 2020 in Dubuque and Muscatine, IA. By Mar 21, Iowa border counties had five cases while Illinois border counties had one. The number of cases on the Iowa side of the border continued to increase compared to the Illinois side (Fig. 7). Table 4 shows the 2019 population with cumulative number of cases and with cases per 10K pop. for each of the counties until Illinois lifted its “stay-at-home” NPI order.

110

K. E. Haynes et al.

Fig. 6 Statewide Infections per 10K population in all of Mississippi and Arkansas counties before, during and after shutdown

Table 3 Statewide COVID-19 infections: cumulative total and per 10K population

Arkansas

Mississippi

Pop 2019

29,76,149

30,17,804

Total cases May 25

18,067

24,610

Cases per 10K pop May 25

59.87

82.69

Total cases Aug 31

62,156

83,046

Cases per 10K pop Aug 31

205.96

279.04

For the entire shutdown interval (Mar 21 to May 30, 2020), the total number of confirmed cases in the border counties of Iowa grew from 5 to 1,722 while in the Illinois border counties the increase was from 1 to 880. The corresponding data normalized by the population for each side are 38.5 per 10K pop. for IA and 30.7 per 10K pop for IL, i.e., nearly 8 more cases per 10K pop in IA than in IL (Table 5) border counties. This trend continued for the rest of the analysis period (Fig. 8 and Table 6). Table 7 shows output of the Diff-in-Diff model run at 10-day intervals as well for the entire shutdown duration of 71 days. Except for the first 10-day interval, the negative valued coefficient of Illinois infections (decrease) is statistically significant

The Impact of Differing COVID-19 Mitigation Policies: Three Natural …

111

Map 4 Iowa (IA)–Illinois (IL) border county region

indicating the stay-at-home policy intervention worked to keep the infections down on Illinois side of the border counties. The policy intervention for the entire shutdown (from March 22, 2020 to May 30, 2020, on the Illinois side had, −0.69 cases less per day per 10K population. For a total population of 286,986, Illinois border counties saw 1,401 less cases than without shutdown. These results confirm the findings from an earlier study (Lyu & Wehby, 2020) in which their Diff-in-Diff model shows that Iowa border counties saw an “excess” of 217 cases, which is the flip side of the lack of cases in Iowa border counties we show in our study. The actual counts are different for each study (Lyu & Wehby, 2020) and this report) due to the following reasons: the earlier report used five-day subintervals covering total of 30 days stay-at-home duration for their Diff-in-Diff model while we use 10-day subintervals covering the entire 71 days of stay-at-home duration. However, both studies confirm the finding that Stay-At-Home in this region led to reduction in cases.

112

K. E. Haynes et al.

Fig. 7 Infections per 10K population in border counties of IA and IL during the shutdown Table 4 IA–IL border county COVID-19 infections and per 10K infections at the end of stay-at-home duration

Pop2019

Infections 5/30

Carroll, IL

15,387

16

Hancock, IL

Infections P10K pop 10.40

19,104

17

8.90

Henderson, IL

7,331

8

10.91

Jo Daviess, IL

22,678

32

14.11

Mercer, IL [M]

16,434

Rock Island, IL [M]

1,47,546

17

10.34

651

44.12

Whiteside, IL [M]

58,498

139

23.76

Clinton, IA [M]

49,116

61

12.42

Des Moines, IA

40,325

62

15.38

Dubuque, IA

93,653

345

36.84

Jackson, IA

19,848

12

6.05

Lee, IA

35,862

25

6.97

Louisa, IA

11,387

343

301.22

Muscatine, IA [M]

42,745

556

130.07

358

21.67

Scott, IA [M]

1,65,224

The Impact of Differing COVID-19 Mitigation Policies: Three Natural … Table 5 IA–IL border county COVID-19 infections and per 10K infections at the end of stay-at-home duration

Pop2019

Infections 5/30

113 Infections P10K pop

IL border counties

2,86,978

880

30.7

IA border counties

4,58,160

1,762

38.5

Fig. 8 Infections per 10K population in border counties of IA and IL during the shutdown and beyond

10 Counties Along the Border Between Stay@home Minnesota (MN) and the Combined State Boundary of Non-stay at Home North Dakota (ND) and South Dakota (SD) Of the total 25 counties that share the border between Minnesota and the North & South Dakotas together, there are 13 counties in Minnesota and 12 counties in North and and South Dakotas (Map 5). Two counties in North Dakota (Cass & Richland) and two counties in Minnesota (Clay and Wilkin) are part of MSAs, designated in the notation as [M] in the map and the table. Clay and Polk counties in Minnesota as well Cass in ND and Minnehaha in SD have the largest populations. Other counties on both sides of the border have populations of 20,000 or less.

114

K. E. Haynes et al.

Table 6 Infections per 10K population for border counties from March 21, 2020 to Jul 30, 2020 Infections per 10K population Date

IA border counties

IL border counties

Ratio IA to IL infections

Mar 21, 2020

0.11

0.03

3.13

Mar 31, 2020

1.35

0.66

2.04

Apr 10, 2020

6.53

4.25

1.54

Apr 20, 2020

15.56

9.76

1.60

Apr 30, 2020

22.04

17.46

1.26

May 10, 2020

29.71

23.63

1.26

May 20, 2020

34.88

27.74

1.26

May 30, 2020

38.46

30.66

1.25

Jun 10, 2020

40.92

33.87

1.21

Jun 20, 2020

43.26

36.38

1.19

Jun 30, 2020

51.71

42.51

1.22

Jul 10, 2020

72.25

51.54

1.40

Jul 20, 2020

89.23

60.32

1.48

Jul 30, 2020

104.77

72.06

1.45

Table 7 Diff-in-Diff estimates for coefficient of infections for Illinois border counties at 10-day interval and the entire shutdown duration Coefficients of infections

Std err

t

P > |t|

95% conf interval

Mar 22–Mar 31

−0.0462

0.0272

−1.70

0.090

−0.0998

0.0073

Apr 01–Apr 10

−0.7224

0.2719

−2.66

0.008

−1.2578

−0.1870

Apr 11–Apr 20

−2.0322

0.8842

−2.30

0.022

−3.7736

−0.2908

Apr 21–Apr 30

−0.7022

0.3341

−2.10

0.037

−1.3601

−0.0442

May 01–May 10

−0.5220

0.1614

−3.23

0.001

−0.8397

−0.2042

May 11–May 20

−0.5131

0.1859

−2.76

0.006

−0.8792

−0.1469

May 21–May 30

−0.2750

0.1190

−2.31

0.022

−0.5094

−0.0407

Mar 22–May 30

−0.6876

0.1477

−4.66

0.000

−0.9774

−0.3978

Minnesota began its stay-at-home policy on March 27, 2020 (https://mn.gov/gov ernor/assets/3a.%20EO%2020-20%20FINAL%20SIGNED%20Filed_tcm1055425020.pdf) and ended it on May 18, 2020 (https://mn.gov/governor/assets/EO% 2020-56%20Final_tcm1055-433768.pdf). It was in effect for a total of 53 days. Minnehaha, SD was the first county to report 3 confirmed case on March 11, 2020, while Clay, MN reported its first case almost 9 days later, on Mar 20, 2020. By the time, the lockdown began on March 27, ND/SD border counties had 31 confirmed cases while the Minnesota side had 7 cases. The number of cases in both the North Dakota and South Dakota side of the border continued to increase compared to the Minnesota side (Fig. 9). Table 8 shows the

The Impact of Differing COVID-19 Mitigation Policies: Three Natural …

115

Map 5 North Dakota (ND)–South Dakota (SD) and Minnesota (MN) border county region ([M] = MSA)

2019 population, cumulative number of cases and cases per 10K pop. for each of the counties until Minnesota lifted its “stay-at-home” NPI order. For the entire shutdown interval (Mar 27 to May 18, 2020), the total number of confirmed cases in the border counties of ND/SD Dakotas grew from 31 to 4,728; while the Minnesota border counties saw an increase from 7 to just 453. The corresponding data on May 18, normalized by population on each side are 86.71 per 10K pop. for ND/SD and 26.05 per 10K pop. for MN (Table 9) border counties. This trend continued for the rest of the analysis period (Fig. 10 and Table 10). Table 11 shows the output of the Diff-in-Diff model at 10-day intervals as well for the entire shutdown duration of 53 days. Except for the first 10-day interval (Mar

116

K. E. Haynes et al.

Fig. 9 Infections per 10K population in border counties of ND-SD and MN during the shutdown

Table 8 ND-SD-MN border county COVID-19 infections and per 10K infections at the end of stay-at-home duration

Pop2019

Infections May 18, 2020

Inf P10K pop

MN border counties

1,70,464

444

26.05

SD-ND border counties

5,45,272

4,728

86.71

27 to Apr 06) and the fourth 10-day interval (Apr 27 to May 06), the negative valued coefficient of per day per 10K population infections saw a statistically significant decrease, indicating the stay-at-home policy intervention worked to keep the infections down on Minnesota side the border counties. Due to the policy intervention for the entire shutdown period (from March 27, 2020 to May 18, 2020), the Minnesota side had −0.37 cases less infections per day per 10K population. For a total population of 170,464 Minnesota border counties saw nearly 350 less cases than it would otherwise be likely to have occurred. To put this another way, without shutdown the number cases would likely be as high as 803 instead of 453.

The Impact of Differing COVID-19 Mitigation Policies: Three Natural … Table 9 ND-SD-MN border county COVID-19 infections and per 10K infections at the end of stay-at-home duration ([M] = MSA county)

County–State

Pop2019

Infections May 18, 2020

117 Inf P10K pop

64,222

310

48.27

Kittson–MN

4,298

1

2.33

Marshall–MN

9,336

8

8.57

Norman–MN

6,375

11

17.25

Clay–MN [M]

Polk–MN

31,364

56

17.85

Wilkin–MN [M]

6,207

12

19.33

Big Stone–MN

4,991

3

6.01

Lac qui Parle–MN

6,623

3

4.53

Lincoln–MN

5,639

4

7.09

Pipestone–MN

9,126

10

10.96

Rock–MN

9,315

20

21.47

Traverse–MN

3,259

3

9.21

Yellow Medicine–MN

9,709

3

3.09

Cass–ND [M]

1,75,249

1,173

66.93

71,083

317

44.60

Pembina–ND

7,069

6

8.49

Richland–ND [M]

16,353

9

5.50

Grand Forks–ND

Traill–ND

8,030

2

2.49

Walsh–ND

10,904

10

9.17

Brookings–SD

3.99

35,077

14

Deuel–SD

4,351

1

2.30

Grant–SD

7,052

9

12.76

Minnehaha–SD

1,93,134

3,150

163.10

Moody–SD

6,576

18

27.37

Roberts–SD

10,394

19

18.28

11 Border County Region Between Stay@home Mississippi (MS) and Not-stay@home Arkansas (AR) There are five counties on each side of the border between MS and AR (Map 6). Crittenden, AR and Desoto and Tunica counties in Mississippi are part of the Memphis, Tennessee MSA. These counties are indicated on the Map and in the table with the notation [M] next to their names.

118

K. E. Haynes et al.

Fig. 10 Infections per 10K population in border counties of ND&SD and MN Table 10 Infections per 10K population for border counties from March 21, 2020 to Jul 30, 2020 Date

MN border counties

ND-SD border counties

Ratio of ND-SD to MN border county infections

3–11

0

0.06

NA

3–18

0

0.11

NA

3–28

0.62

0.77

1.25

4–7

1.64

4.77

2.90

4–17

2.67

24.83

9.30

4–27

3.70

47.26

12.78

5–7

7.40

61.93

8.37

5–18

9.45

86.71

9.17

5–28

10.89

99.44

9.13

6–7

14.38

106.75

7.42

6–17

16.23

112.38

6.92

6–27

20.55

117.10

5.70

7–7

31.85

123.48

3.88

7–17

51.58

133.07

2.58

7–27

67.61

145.14

2.15

7–31

72.34

149.61

2.07

The Impact of Differing COVID-19 Mitigation Policies: Three Natural …

119

Table 11 Diff-in-diff estimates for coefficient of infections for Minnesota border counties at 10-day interval and the entire shutdown duration Coefficients

Std err

P > |t|

P

95% conf interval

Mar 27–Apr 06

−0.0687

0.0658

−1.04

0.297

−0.19797

0.06049

Apr 07–Apr 16

−0.4154

0.1516

−2.74

0.006

−0.71308

−0.11772

Apr 17–Apr 26

−0.4223

0.1663

−2.54

0.011

−0.7489

−0.0958

Apr 27–May 06

−0.2409

0.1468

−1.64

0.101

−0.5291

0.0474

May 07–May 18

−0.6516

0.1616

−4.03

0.000

−0.9690

−0.3341

Mar 27–May 18

−0.3676

0.06593

−5.58

0.000

−0.4969

−0.2383

Map 6 Arkansas (AR)–Mississippi (MS) border county region

Mississippi Governor Tate Reeves issued shelter-in-place executive order (https://www.sos.ms.gov/content/executiveorders/ExecutiveOrders/1466.pdf) that took effect on April 3, 2020 to end on Apr 19, 2020. It was extended twice, the

120

K. E. Haynes et al.

first time to last (https://mcusercontent.com/08cb3e52aa1308600f84d49ea/files/e24 a9045-8ab7-4aa0-bb23-bb49d342c816/Executive_Order_1477_Safer_at_Home. pdf) through May 11 and the second time (https://www.sos.ms.gov/content/execut iveorders/ExecutiveOrders/1480.pdf) to end on May 25, 2020. Mississippi counties were under lockdown from Apr 03 to May 25, 2020, for a total of 53 days. The first cases in the border regions were reported in three Mississippi counties (Bolivar, Desoto and Coahoma) on Mar 18, 2020. A single case was reported 4 days later, on Mar 22, 2020, in the Crittenden county on the Arkansas side of the border. In the beginning, the number of cases on Mississippi side continued to grow a bit more than on the Arkansas side of the border (Fig. 11). Table 12 shows population, cumulative number of cases and cases per 10K population for each of the counties. The table also includes the following data for the entire shutdown period (Apr 3 to May 25, 2020). The total number of confirmed cases in the border counties of Mississippi grew from 214 to 922, a 3.3% increase while Arkansas border counties saw an increase from 42 to 351, a 7.4% increase. The corresponding data normalized by population on each side are 36.5 per 10K pop. for AR and 31.6 per 10K pop. for MS, i.e., nearly 5 more cases per 10K pop on the AR side of the border than in MS. The higher number of cases per 10K pop. switched from the Mississippi side to the Arkansas side border around Apr 14, 2020 (Fig. 12) and this trend continued for the rest of the analysis period (Table 13). Table 14 shows output of the Diff-in-Diff model at 10-day intervals as well for the entire shutdown duration of 53 days. Out of the five 10-day intervals, two showed

Fig. 11 Infections per 10K population in border counties of AR and MS during the shutdown

The Impact of Differing COVID-19 Mitigation Policies: Three Natural … Table 12 AR-MS border county COVID-19 infections and per 10K infections at the end of stay-at-home duration

Y2019 Pop

Infections 5/25

121 Infections P10K pop

Chicot, AR

10,118

13

12.8484

Crittenden, AR [M]

47,955

293

61.0989

Desha, AR

11,361

19

16.7239

8,857

17

19.1939

Phillips, AR

17,782

9

5.0613

Bolivar, MS [M]

30,628

141

46.0363

Coahoma, MS

22,124

106

47.9118

DeSoto, MS [M] 1,84,945

474

25.6292 49.8339

Lee, AR

Tunica, MS [M]

9,632

48

Washington, MS

43,909

153

34.8448

Infections 5/25

Infections P10K pop

Y2019 Pop Arkansas border county region

96,073

351

36.5347

MS border county region

2,93,257

922

31.4400

Fig. 12 Infections per 10K population in border counties of AR and MS during the shutdown and beyond

122

K. E. Haynes et al.

Table 13 Infections per 10K population for border counties from March 21, 2020 to Jul 30, 2020 Infections per 10K population Date

AR border counties

MS border counties

Ratio AR to MS border county infections

March 18, 2020

0

0.21

–

April 3, 2020

4.4

7.3

0.60

April 19, 2020

16.7

16

1.04

May 11, 2020

25.8

22.3

1.16

May 25, 2020

36.5

31.7

1.15

June 25, 2020

160.1

66.9

2.39

July 25, 2020

270.8

184

1.47

July 31, 2020

309.3

216.6

1.43

Table 14 Diff-in-diff estimates for coefficient of infections for Mississippi border counties at 10day interval and the entire shutdown duration Apr 03–Apr 12

Coefficient

Std err

t

P > |t|

95% conf interval

−0.1387

0.1815

−0.76

0.45

−0.4960

0.2187

Apr 13–Apr 22

−0.0906

0.1736

−0.52

0.60

−0.4324

0.2512

Apr 23–May 02

−0.4538

0.1708

−2.66

0.01

−0.7902

−0.1175

May 03–May 12

−0.3444

0.1634

−2.11

0.04

−0.6662

−0.0226

May 13–May 25

−0.1163

0.1872

−0.62

0.54

−0.4848

0.2521

Apr 03–May 25

−0.1546

0.1152

−1.34

0.18

−0.3881

0.2001

a lower rate of infection coefficient for Mississippi to be statistically significant. It was −0.44 per 10K pop and −0.34 per 10K pop. between Apr 23-May 2 and May 3May 12, 2020, which translates to 131.97 and 99.71 less cases during these intervals for the Mississippi side. Although there were less cases per 10K population for the Mississippi border county region, the other time intervals do not show statistically significant coefficients. Therefore, one cannot infer whether these fewer cases were entirely due to the stay-at-home policy intervention.

12 Conclusions and Future Research Unlike much of the world’s nation states, the U.S. did not pursue a nationwide mitigation policy to fight the COVID-19 pandemic. Instead there appeared to be a fifty-state attempt to control the spread of the disease thereby centering the burden on state health-care systems (Haynes & Kulkarni, 2021). The mitigation strategies ranged from: strict stay-at-home orders to issuing loose guidelines and advisories to state residents. Even though there appears to be wide variations in each state’s

The Impact of Differing COVID-19 Mitigation Policies: Three Natural …

123

approach to COVID-19 mitigation strategies, these fifty states fall into two broad categories, 45 states that issued a limited duration stay-at-home executive orders under emergency powers, and five states that did not issue such orders. These two approaches can be seen as a natural experiment to fight COVID-19 pandemic and one could test the outcomes with a Difference-in-Difference (Diff-in-Diff) model applied to pair(s) of neighboring states and the border regions between these states where these two states differed in their mitigation strategies. Specifically, we applied the Diff-in-Diff model to three pairs of neighboring states and the boundary regions between these pairs of states. In the case of Iowa and Illinois, both the state level and border county region level, the 71-day “stay-at-home” mitigation strategy helped Illinois reduce the potential increase in COVID-19 infections by over 24,000 cases. Even at the border county region, the counties on the Illinois side of the border saw a net reduction of 152 cases. In the case of Minnesota and combined states of North & South Dakota, the diffin-diff results indicate that the more than 50-day shutdown did not help create a statistically significant reduction of state-wide cases for Minnesota. The outcome at the border county regions is mixed where only two discontinuous 10-day intervals out five 10-day intervals saw a statistically significant reduction in cases. For the diff-in-diff model applied to Arkansas and Mississippi, the 50-day plus Mississippi shutdown, in fact saw a statistically highly significant increase state-wide COVID-19 cases in Mississippi. Like Minnesota, the outcome at the border county region for Mississippi was somewhat mixed with two contiguous 10-day intervals saw a statistically significant reduction in COVID-19 cases in those border counties. Analyses of border counties between two neighboring states where one state had a non-pharmaceutical intervention (NPI) such as “stay-at-home” or “stay-inshelter” the border counties in that state appear to have reduced number of infections as compared to the border counties of its neighbor which did not. This is borne out by a statistically significant negative valued coefficient of infection per day per 10K population for the border counties with the NPI for the entire duration of that NPI. However, the picture is somewhat mixed for Diff-in-Diff model runs at smaller intervals of 10 days within the NPI duration. Some of these border counties are part of larger economically integrated county groups (MSAs) that span more than two neighboring states. The spillover effects of being part of bigger MSAs are not captured in these Diff-in-Diff models. In summary, the statewide results offer a mixed picture with Illinois showing statistically significant reduction in the number of covid-19 cases. On the other hand, Minnesota and Mississippi appear to have more infections during the “shelterin-place” duration, the exact opposite of how shutdown is supposed to mitigate these. Part of these variations may be that these states share borders with other states which may have non-uniform variations in the implementations of lockdown policies. Further, many of these states are host to MSAs that span two or more state boundaries where local, regional and statewide. The shelter policy works sometimes in some places but is not an across the board solution in the US with our highly variable implementation, density/urban variability and our high cross regions mobility patterns. In the absence of a national policy and

124

K. E. Haynes et al.

instead of each state on its own fighting the disease, a regional or a multi-jurisdictional cooperation with uniform regional policies may help in fighting infectious diseases such as COVID-19. In our future research we plan to study regional mitigation policies (Goolsbee et al., 2020) as a potential solution for COVID-19 like pandemics.

References County Population Totals: 2010–2019. Retrieved August 9, 2022, from https://www.census.gov/ data/datasets/time-series/demo/popest/2010s-counties-total.html Emergency Executive Order 20-20. Directing Minnesotans to stay at home. Minnesota Governor’s Office. Retrieved August 9, 2022, from https://mn.gov/governor/assets/3a.%20EO%2020-20% 20FINAL%20SIGNED%20Filed_tcm1055-425020.pdf Emergency Executive Order 20-56. Safely reopening Minnesota’s economy and ensuring safe non-work activities during the COVID-19 peacetime emergency. Minnesota Governor’s Office. Retrieved August 9, 2022, from https://mn.gov/governor/assets/EO%2020-56%20Final_tcm1 055-433768.pdf Ferguson, N. M., Cummings, D. A. T., Fraser, C., Cajka, J. C., Cooley, P. C., & Burke, D. S. (2006). Strategies for mitigating an influenza pandemic. Nature 448–452. pmid:16642006. Retrieved August 9, 2022 Goolsbee, A., Bei Luo, N., Nesbitt R., & Syverson C. (2020). COVID-19 lockdown policies at the state and local level. University of Chicago, Becker Friedman Institute for Economics Working Paper No. 2020-116, Sep 01. Retrieved August 8, 2020, from https://papers.ssrn.com/sol3/pap ers.cfm?abstract_id=3682144 Haynes, K. E., & Kulkarni, R. (2021). Modeling region based regimes for COVID-19 mitigation: An inverse Gompertz approach to coronavirus infections in the USA, New York, and New Jersey. Special Issue: Effects and Policies of Covid-19, Regional Science Policy & Practice, 13(S1), 4–17 (November) Haynes, K. E., Kulkarni, R., Li, M. H., & Siddique, A. B. (2022). Differences in state level impacts of COVID-19 policies. In M. Kawano, K. Kourtit, P. Nijkamp, & Y. Higano (Eds.), Theory and history in regional perspective. New frontiers in regional science (Vol. 56, pp. 415–431). Springer John Hopkins University. “Time Series Confirmed COVID-19 US” county data. Retrieved August 9, 2022, from https://github.com/CSSEGISandData/COVID-19/tree/master/csse_covid_19_data/ csse_covid_19_time_series Kraemer, M. U. G., Yang, C.-H., Gutierrez, B., Wu, C.-H., Klein, B., Pigott, D. M., et al. (2020). The effect of human mobility and control measures on the COVID-19 epidemic in China. Science. pmid:32213647. Retrieved August 9, 2022 Krishnamachari, B., Morris, A., Zastrow, D., Dsida, A., Harper, B., & Santella, A. J. (2021). The role of mask mandates, stay at home orders and school closure in curbing the COVID-19 pandemic prior to vaccination. American Journal of Infection Control, 49(8), 1036–1042. Lyu, W., & Wehby, G. L. (2020). Comparison of estimated rates of coronavirus disease 2019 (COVID-19) in border counties in Iowa without a stay-at-home order and border counties in Illinois with a stay-at-home order. Journal of the American Medical Association (Network Open), 3(5), e2011102. https://doi.org/10.1001/jamanetworkopen.2020.11102. https://jamane twork.com/journals/jamanetworkopen/fullarticle/2766229 Maier, B. F., & Brockmann, D. (2020). Effective containment explains subexponential growth in recent confirmed COVID-19 cases in China. Science. eabb4557. pmid:32269067 Padalabalanarayanan, S., Hanumanthu, V. S., & Sen, B. (2020). Association of state stay-at-home orders and state-level African American population with COVID-19 case rates. Journal of the American Medical Association (network Open), 3(10), e2026010–e2026010.

The Impact of Differing COVID-19 Mitigation Policies: Three Natural …

125

Peter, C., Katikireddi, S. V., Leyland, A., & Popham, P. (2017). Natural experiments: An overview of methods, approaches, and contributions to public health intervention research. Annual Review of Public Health, 38, 39–56. https://www.annualreviews.org/doi/full/10.1146/annurev-publhealth031816-044327 Sen, S., Karaca-Mandic, P., & Georgiou, A. (2020). Association of stay-at-home orders with COVID-19 hospitalizations in 4 states. Journal of the American Medical Association, 323(24), 2522–2524. State “Shelter-in-Place” and “Stay-at-Home” Orders. Retrieved August 9, 2022, from https://www. finra.org/rules-guidance/key-topics/covid-19/shelter-in-place State of Illinois Governor’s Office. Retrieved August 9, 2022, from https://www2.illinois.gov/sites/ gov/Documents/APPROVED%20-%20Coronavirus%20Disaster%20Proc%20WORD.pdf State of Illinois Governor’s Office. Gubernatorial disaster proclamation. Retrieved August 9, 2022, from https://www2.illinois.gov/sites/gov/Documents/CoronavirusDisasterProc-5-29-2020.pdf State of Mississippi Office of Governor. Executive order no 1480. https://www.sos.ms.gov/content/ executiveorders/ExecutiveOrders/1480.pdf State pf Mississippi Office of Governor. Executive order no 1466. https://www.sos.ms.gov/content/ executiveorders/ExecutiveOrders/1466.pdf State pf Mississippi Office of Governor. Executive order no 1477. Retrieved August 9, 2022, from https://mcusercontent.com/08cb3e52aa1308600f84d49ea/files/e24a9045-8ab7-4aa0-bb23bb49d342c816/Executive_Order_1477_Safer_at_Home.pdf Tian, H., Liu, Y., Li, Y., Wu, C.-H., Chen, B., Kraemer, M. U. G., et al. (2020). An investigation of transmission control measures during the first 50 days of the COVID-19 epidemic in China. Science. pmid:32234804. Retrieved August 8, 2022. U.S. state and local government responses to the COVID-19 pandemic. Retrieved June 18, 2020, from https://en.wikipedia.org/wiki/U.S._state_and_local_government_responses_to_the_ COVID-19_pandemic#cite_note-Coronavirus_Oklahoma-74

On the Association Between Income Inequality and COVID Spread: A View into Spanish Functional Urban Areas David Castells-Quintana

and Vicente Royuela

1 Introduction The declaration of the COVID-19 pandemic in March 2020 took place in a world with a significant trend of growing income inequality. Both the pandemic shock and the trend of increasing inequality have shown a significant impact on urban areas. Inequality and spatial segregation have worsened social discontent and have been especially manifested in large cities (Castells-Quintana & Royuela, 2021; CastellsQuintana et al., 2020). In this work, we analyze the association between income inequality and the expansion of the COVID-19 disease, by looking at several indicators of the disease: identifies cases, hospitalization, UCI and deaths. We consider Spain as a case study at two alternative geographical scales: municipalities and Functional Urban Areas (FUAs). We look at the first three waves of expansion (ending in February 2021) and consider different age cohorts to differentiate the role of inequality in different population groups. It is documented that the COVID-19 disease spreads faster in deprived communities, especially within cities and neighborhoods, which in turn may have reinforced spatial inequalities. Ahmed et al. (2020) suggest that more deprived populations lacking access to health services are left most vulnerable during crisis; they are more isolated in lockdowns due to lower connectivity; are more likely to have chronic conditions and subsequently higher mortality. Barnard et al. (2021) confirm such expectations and find for England that the pandemic has widened inequalities in premature mortality by area deprivation. Bhowmik et al. (2021) explain that higher income inequality is an indicator of large amounts of low-income workers who D. Castells-Quintana Department of Applied Economics, Univ Autonoma de Barcelona, Bellaterra, Spain D. Castells-Quintana · V. Royuela (B) AQR-IREA Research Group, Universidad de Barcelona, Barcelona, Spain e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. G. Celbi¸s et al. (eds.), Pandemic and the City, Footprints of Regional Science, https://doi.org/10.1007/978-3-031-21983-2_8

127

128

D. Castells-Quintana and V. Royuela

possibly need to continue their daily work despite the risk of the virus transmission. They find significant effects of income inequality in mortality rates in US counties. Similarly, Tan et al. (2021) also find a significant positive correlation between the Gini index and the spread of COVID-19 in US counties, which peaked in the summer months of 2020. Looking at countries worldwide, Von Chamier (2021) find that higher inequality is positively correlated with new infections. Similarly, Wildman (2021) finds a positive association for OECD countries. By means of cross section regressions, he finds that a 1% increase in the Gini coefficient is associated with 4% increase of infections and 5% increase in mortality. There is also evidence of faster spread of the disease in larger cities. While urbanization is a synonym of development and better living conditions, rapid urbanization process in places such as developing economies is, in many times, characterized by extreme poverty and poor quality institutions (Glaeser & Henderson, 2017), low quality of urban infrastructure (Castells-Quintana, 2017), bad health conditions and higher disease rates (Lillini & Vercelli, 2018). According to the epidemiology literature (Alirol et al., 2011; Neiderud, 2015) large and dense urban areas facilitate the transmission of viruses. Recently, González-Val and Sanz-Gracia (2021) analyzed, for 90 countries, the role played by urban agglomeration on the diffusion and intensity of the COVID-19. They find that more urbanized countries suffer a higher incidence of the virus, even after controlling for several development indicators. These authors’ reference Wheaton and Thompson (2020) looking at Massachusetts local areas, and Carozzi et al. (2020) and Geng et al. (2021) considering US local scales, who find similar results. In this paper, we focus on Spain. After Italy, Spain became the second epicenter in Europe by number of cases and deaths, due to the quick spread of the virus. Henríquez et al. (2020) describe the Spanish health system, organized via a national health system typology, providing universal healthcare through 13,122 primary care facilities of General Practitioners (GP), and 466 hospitals, mainly managed by regional departments (i.e., Autonomous Communities). Spain’s life expectancy is well above the average of OECD countries, although it is an ageing country, with the share of population over 65 exceeding other OECD countries, and with a significant rate of morbidity (62.3% of older than 65 report two or more chronic conditions). The current share of health expenditure over GDP (8.9%) is close to OECD figures (8.8%), being the per capita healthcare expenditure close to 3,500 USD (PPP) in 2018, while health and long-term care resources (doctors, nurses and beds per 1,000 inhabitants), and long-term care residences over the OECD average. The analysis of the spread of COVID-19 for Spain developed by Orea and Álvarez (2020) finds that the epidemic started in the most-populated provinces earlier than in the less-populated provinces. More-populated provinces are better connected with foreign countries, which facilitated importing the disease. They also find the effectiveness of the lockdown in preventing the propagation of the coronavirus both within and between provinces. Similarly, Amuedo-Dorantes et al. (2020) find that the adoption of a nationwide lockdown was more effective in limiting COVID-19 deaths in regions that, at the time of the lockdown, were at an earlier stage of the pandemic spread. Oto-Peralías (2020) looks at the correlation of the initial spread of the virus

On the Association Between Income Inequality and COVID Spread: …

129

and several climatic and socio-economic variables, finding negative correlations with temperature and distance to Madrid. To the best of our knowledge, none of these works devoted to the Spanish case looks at the potential role of inequality on the spread of the COVID-19. Besides, most of them develop the analysis at the provincial level, and few of them consider local areas or cities. We work with municipal data and also with the economic definition of a city, namely, Functional Urban Areas. In the next section, we describe the considered data for our study, with emphasis on the definition of inequality and showing the details of the COVID-19 data.

2 Data and Preliminary Evidence To investigate the association between inequality and COVID-19 mortality, we use official data on daily COVID-19 from the Spanish National Epidemiology Center (CNE), referring to: . Number of reported cases confirmed with a positive diagnostic test for active infection . Number of hospitalized cases . Number of cases admitted to ICU . Number of deaths This data is available for all municipalities with at least 1,000 inhabitants by the 1st of January 2020, and for different age cohorts (95). The dataset considered information until mid-April 2021, what allows for accounting for the first three full waves of the COVID pandemic, as can be seen in Fig. 1. We consider the start of the pandemic on any reported infection since the 1st of January 2020, and the end of the first three waves are assigned respectively to the 30th of June 2020, 30th of November 2020, and 28th of February 2021. The data for inequality comes is from the Household Income Distribution Atlas, published by the Spanish National Institute of Statistics (INE). The INE uses information from the Spanish Tax Agency (AEAT), aggregating the individual information by territory. For the construction of inequality measures, the AEAT adds the information for 204 fixed income lots per unit of consumption, from which they derive inequality indicators, such as the Gini index. Average income indicators are provided for municipalities with at least 100 inhabitants, while income distribution indicators are provided for municipalities with at least 500 inhabitants. Given that the INE discharge unit does not provide income distribution data for municipalities under the Common Tax Regime, information for the Autonomous Communities of the Basque Country and Navarra has not been considered. In this case, the measure of net income per inhabitant has been considered, as well as the Gini index calculated by the INE. Considering the restrictions in the minimum size for both COVID-19 data and for the Common Tax Regime, the final number of municipalities is restricted to 3,142 (out of 8,131 municipalities), summing over 96% of total population of the considered regions.

130

D. Castells-Quintana and V. Royuela

Fig. 1 Reported cases confirmed with a positive diagnostic. Source CNE. https://cnecovid.isciii. es/covid19/#ccaa

Besides the use of municipal information, we also consider cities as Functional Urban Areas, which go beyond administrative definitions. We use the definition offered by the OECD and the European Commission, based on the identification of a city and its area of influence. The procedure used by the OECD is based on four steps. First, the definition of city implies selecting basic spatial units (cells of approximately 1 km2 ) with a high population density, which is defined as a minimum of 1,500 inhabitants per km2 . Second, contiguous cells are grouped together, filling in empty spaces surrounded by other high-density cells. These groupings of cells are considered “urban center” (core) if they have at least 50,000 inhabitants. Third, the administrative definition is added, so that all municipalities with at least half of their population associated with the said urban center are assigned to the city, which should ensure that at least 50% of the population of the city lives in such urban centers and that no less than 75% of the population of the urban center lives in the city. Finally, the area of influence is established, which is made up of the municipalities in which a minimum of 15% of their employed residents are working in the city, being necessary that there be contiguity. The minimum size of the FUA can be restricted for convenience. Thus, in the definition of the OECD (2012) it established a minimum of 100,000 inhabitants. In the most recent definition (2019) the minimum has been set at 50,000 inhabitants. Following the more recent methodology, in 2019 the number of FUAs in Spain was restricted to 81, which involve a total of 1,260 municipalities. The largest functional urban area in Spain is Madrid, with 6.7 million inhabitants, followed by Barcelona, with 4.9 million, Valencia (1.7 million), Seville (1.5 million) and Bilbao (1 million). The rest do not reach one million inhabitants. The smallest FUAs are Elda (53 thousand inhabitants), Cuenca (60 thousand), Ávila and Linares (61 thousand). The

On the Association Between Income Inequality and COVID Spread: …

131

Fig. 2 Gini index 2017, Spanish functional urban areas. Source Castells-Quintana and Royuela (2021)

five (ten) largest FUAs in Spain account for 35% (42%) of the total population, while the set of 81 FUAs represents 71%. Between the 2004–07 period and the 2015–17 triennium, the largest FUAs have grown by 11%, well above the country as a whole (6%). The data of inequality for the FUAs follows Castells-Quintana and Royuela (2021), which considers the within and between municipal inequality for every FUA. As we do not have data for inequality for FUAs in the Basque Country and Navarra, our sample is restricted to 76 urban areas. Figure 2 displays the distribution of inequality for the Spanish FUAs in 2017. For our analysis, we also use additional controls, such as population size, average income, average household size and average age, all coming from INE. Figure 3, panels (a) to (d) display the association of the size of every FUA and the Gini index with two indicators of COVID-19 spread: infections and mortality. There is no clear pattern on the correlation, particularly for the measurement of inequality. Similarly, the correlation for all considered municipalities (see Table 1) is, at most, weak, with varying signs depending on the considered wave of the virus.

3 Empirical Analysis We estimate the impact of inequality on COVID incidence by means of a simple regression model (estimated either at the FUA or municipality level): Covid i = βGini i + γ Contr ols i + θ j + εi

132

D. Castells-Quintana and V. Royuela Mortality

Gini Index

Population

Infection Cases

Fig. 3 Scatterplots COVID-19 indicators and inequality and population size. FUAs

Table 1 Correlation of COVID-19 indicators with inequality and population (log of) municipalities Correlation of gini with …

1st wave

2nd wave

3rd wave

All waves

Infection cases

0.037

−0.029

−0.017

−0.023

Hospitalizations

0.038

−0.023

−0.028

−0.007

ICU

0.054

−0.011

−0.022

0.003

Deceased

0.000

−0.030

−0.008

−0.021

Correlation of population (log) with …

1st wave

Infection cases

0.086

0.035

0.001

0.034

Hospitalizations

0.066

−0.024

−0.050

−0.006

ICU

0.094

0.003

0.014

0.052

Deceased

0.002

−0.071

−0.102

−0.096

2nd wave

3rd wave

All waves

where . C ovi d i identifies cases, hospitalization, UCI or deaths in observation i (i.e., FUA or municipality), all per 100,000 inhabitants . Gini stands for the measure of inequality of the FUA or municipality . Controls include the average household size, average age, average income per household and the log of population . θ j are regional fixed effects (i.e., province fixed effects)

On the Association Between Income Inequality and COVID Spread: …

133

We estimate this model both for FUAs and municipalities, for the first three waves of the pandemic, and for different age cohorts. For brevity, we display the main results graphically. Figure 4 plots the standardized results of the parameters of the full model. Every color reports the results of the regression associated with one COVID-19 indicator, what implies that every graph shows the results of four alternative regressions. While the model for FUAs is not reporting significant results in any variable, probably due to the small amount of territorial units, the model for municipalities clearly shows that larger municipalities were more affected by the disease, although the indicator for mortality is not statistically significant. The parameter associated with average age reports significant and positive results, as expected, while household size display heterogeneous results. More affluent municipalities were less affected by COVID-19. Finally, the Gini index is mainly reporting a significant and negative parameter, what contrasts with the previous evidence reported by other works, for instance for the United States. Figure 5 displays the results of the regressions for municipalities that belong or nor to a FUA. As can be seen, the results are much more robust in urban municipalities than in those not belonging to any FUA (i.e., more rural), where the variability of the indicators is expected to be much larger. In any case, some differences arise, particularly in the influence of the average income per household, clearly negative in urban municipalities, and with varying parameters for non-urban ones depending on the COVID-19 indicator. In Fig. 6 we show results by wave. As for FUAs, we find again mostly nonsignificant results, with the exception of household size: municipalities with larger households were less affected. This result, though, is only visible in the first wave. As for municipalities, some variables report robust results, such as population size, almost always positive and significant, or average age. It is interesting to note the change in the sign of household size and average income per household. What remains quite persistent is the negative sign of the parameter associated to the Gini index: Municipalities (N=3,124)

Functional Urban Areas (N=76)

Note: Figures show point estimates from our regressions, with confidence intervals (at the 95% confidence level), for each factor in explain cases (blue), hospitalized (red), Intensive-Care (green) and deceased (yellow).

Fig. 4 Main results of the regressions. All waves

134

D. Castells-Quintana and V. Royuela Municipalities in FUAs (N=994)

Municipalities NOT in FUAs (N=1,888)

Note: Figures show point estimates from our regressions, with confidence intervals (at the 95% confidence level), for each factor in explain cases (blue), hospitalized (red), Intensive-Care (green) and deceased (yellow).

Fig. 5 Main results of the regressions. Municipalities, in and out of FUAs. All waves

more unequal municipalities were less severely affected by COVID-19 in the first wave, with less clear results in the second and third wave. As for COVID-19 indicators, the number of cases reports the more sensitive parameters, while the indicator for deceases is closer to zero in most regressions, probably due to the fact of being closer to personal characteristics of infected people than to social and territorial aspects. Finally, Fig. 7 shows the results of the parameter associated with the Gini index in the regressions with indicators by age cohort. The results are displayed for three alternative indicators, cases, ICU and deceased, and for the three considered waves. As expected, while younger cohorts were much less associated with contextual factors, middle age cohorts were more sensible to inequality, again with respect to the number of cases. The maximum value of the parameter is found for cohorts around 50 and 60 years old, lowering the influence for older cohorts, more likely to be more connected to personal than contextual circumstances.

4 Conclusions and Discussion In this work, we have looked at income inequality and COVID spread, focusing on Spanish municipalities and cities (considered as Functional Urban Areas). In line with recent literature, we find that larger cities (i.e., FUAs) have higher levels of inequality. In terms of COVID expansion, we find that larger municipalities, with lager household size, higher average age and lower average income had higher contagion rates, as expected. Controlling for income, we also find that municipalities with higher levels of inequality experienced lower contagion rates, in contrasts with the empirical evidence found for other countries. Regarding age cohorts, our estimates

On the Association Between Income Inequality and COVID Spread: … Functional Urban Areas

3rd wave

2nd wave

1st wave

Municipalities

135

Note: Figures show point estimates from our regressions, with confidence intervals (at the 95% confidence level), for each factor in explain cases (blue), hospitalized (red), Intensive-Care (green) and deceased (yellow).

Fig. 6 Main results of the regressions. First three waves

are more sensitive for middle-age cohorts, particularly for those between 50 and 60, for who contextual factors associated with inequality are more important. Our result for inequality may relate to social interaction. Areas with higher levels of income inequality tend to be more spatially and socially segregated. Thus, areas with higher levels of income inequality may have lower social interaction, which may have contributed to lower COVID spread. This idea is reinforced by the fact that our stronger results for inequality are found in non-FUA municipalities. Our findings come with important policy implications. They reflect the importance of socioeconomic and territorial variables in the expansion of the COVID-19 pandemic. In particular, based on our results, measures to control the pandemic

136

D. Castells-Quintana and V. Royuela Cases – All waves

1st wave (cases)

ICU– All waves

2nd wave (cases)

Deaths– All waves

3rd wave (cases)

Note: Figures show point estimates from our regressions, with confidence intervals (at the 95% confidence level), for the Gini coefficient.

Fig. 7 Regressions results for the Gini index by age cohort, indicator and wave

should pay attention to spatial differences not only in terms of size and income of municipalities, but also in terms of income inequality. However, the negative association between inequality levels and COVID spread should never be understood as signaling inequality as a good thing, but rather the opposite: policy makers aiming at higher social cohesion as well as lower contagion rates should consider potential trade-offs in the sound design of social and health policies. Finally, our analysis clearly calls for further research. It would be interesting to analyze the inequality-COVID spread relationship in dynamic terms and in other contexts different to the Spanish one. Also, it would be of high value added to study measures of social interaction and their association with inequality levels and COVID spread. In this line, the use of more detailed data, or the exploration of intra-city dynamics, could significantly help to better understand the relation of socio-economic dynamics and COVID spread, and better guide policy design.

On the Association Between Income Inequality and COVID Spread: …

137

Acknowledgements The authors thank the support of Ministerio de Ciencia e Innovación [grant number PID2020-118800GB-I00 MCIN/AEI/https://doi.org/10.13039/501100011033]. We also thank the comments received in the XLVI Reunión de Estudios Regionales (Madrid), in the 60th ERSA Virtual Congress (The Regional Science Academy Special Session on The Pandemic and the City) and by two anonymous referees.

References Ahmed, F., Ahmed, N., Pissarides, C., & Stiglitz, J. (2020). Why inequality could spread COVID-19. Lancet Public Health, 5(5), e240. https://doi.org/10.1016/S2468-2667(20)30085-2. Epub 2020 Apr 2 Alirol, E., Getaz, L., Stoll, B., Chappuis, F., & Loutan, L. (2011). Urbanisation and infectious diseases in a globalized world. Lancet Infectious Diseases, 11, 131–141. https://doi.org/10.1016/ S1473-3099(10)70223-1 Amuedo-Dorantes, C., Borra C., Rivera-Garrido, N., & Sevilla, A. (2020). Timing is everything when fighting a pandemic: COVID-19 mortality in Spain. IZA: Discussion paper series. Barnard, S., Fryers, P., Fitzpatrick, J., Fox, S., Waller, Z., Baker, A., Burton, P., Newton, J., Doyle, Y., & Goldblatt, P. (2021). Inequalities in excess premature mortality in England during the COVID-19 pandemic: A cross-sectional analysis of cumulative excess mortality by area deprivation and ethnicity. BMJ Open, 11(12), e052646 (2021). https://doi.org/10.1136/bmjopen-2021052646 Bhowmik, T., Tirtha, S. D., Iraganaboina, N. C., & Eluru, N. (2021). A comprehensive analysis of COVID-19 transmission and mortality rates at the county level in the United States considering socio-demographics, health indicators, mobility trends and health care infrastructure attributes. PLoS one, 2021, 16(4), e0249133 https://doi.org/10.1371/journal.pone.0249133. Carozzi, F., Provenzano, S., & Roth, S. (2020). Urban density and COVID-19. IZA discussion paper 13440. https://www.iza.org/publications/dp/13440/urban-density-and-covid-19 Castells-Quintana, D. (2017). Malthus living in a slum: Urban concentration, infrastructure and economic growth. Journal of Urban Economics, 98, 128–173. Castells-Quintana, D., & Royuela, V. (2021). Ciudades y desigualdad: una mirada a las áreas urbanas funcionales españolas, Cuadernos ICE, #920. Castells-Quintana, D., Royuela, V., & Veneri, P. (2020). Inequality and city size: An analysis for OECD functional urban areas. Paper in Regional Science, 99(4), 1045–1064. Geng, X., Gerges, F., Katul, G. G., Bou-Zeid, E., Nassif, H., & Boufadel, M. C. (2021). Population agglomeration is a harbinger of the spatial complexity of COVID-19. Chemical Engineering Journal, 420, 127702. https://doi.org/10.1016/j.cej.2020 Glaeser, E. L., & Henderson, J. V. (2017). Urban economics for the developing world: An introduction. Journal of Urban Economics, 98(1), 1–5. González-Val, R., & Sanz-Gracia, F. (2021) Urbanization and COVID-19 incidence: A cross-country investigation. Papers in Regional Science, in press. https://doi.org/10.1111/pirs.12647 Henríquez, J., Gonzalo-Almorox, E., García-Goñi, M., & Paolucci, F. (2020). The first months of the COVID-19 pandemic in Spain. Health Policy and Technology, 9(4), 560–574. Lillini, R., & Vercelli, M. (2018). The local socio-economic health deprivation index: Methods and results. Journal of Preventive Medicine and Hygiene, 59(2), 3–10. Neiderud, C. J. (2015). How urbanization affects the epidemiology of emerging infectious diseases. Infection Ecology & Epidemiology, 5(1), 27060. https://doi.org/10.3402/iee.v5.27060 Orea, L., & Álvarez, I. C. (2020). How effective has the Spanish lockdown been to battle COVID19? A spatial analysis of the coronavirus propagation across provinces. Documento de Trabajo FEDEA—2020/03. Oto-Peralías, D. (2020). Regional correlations of COVID-19 in Spain, mimeo.

138

D. Castells-Quintana and V. Royuela

Tan A. X., Hinman J. A., Abdel Magid H. S., Nelson L. M., & Odden M. C. (2021). Association between income inequality and county-level COVID-19 cases and deaths in the US. JAMA Network Open, 4(5), e218799. https://doi.org/10.1001/jamanetworkopen.2021.8799 von Chamier, P. (2021). Inequality, lockdown, and COVID-19: Unequal societies struggle to contain the virus. Original research from the Center on International Cooperation. Wheaton, W. C., & Thompson, A. K. (2020). Doubts about density: COVID-19 across cities and towns. https://ssrn.com/abstract=3586081 Wildman, J. (2021). COVID-19 and income inequality in OECD countries. The European Journal of Health Economics, 22, 455–462. https://doi.org/10.1007/s10198-021-01266-4

Urbanization Impact Arising from the Behavioral Shift of Citizens and Consumers in a Post-pandemic World Tannistha Maiti, Anwita Maiti, Biswajit Maiti, and Tarry Singh

1 Introduction Most of the world’s population is concentrated in cities largely due to the economic power they exert in the form of innovation, modern infrastructure, and sophisticated urban design. Maintaining these cities, provide access to markets and that requires access to talent, a need which the population fulfills. This same large concentration poses as an Achilles heel pushing cities to a tipping point posing as stressors for both man-made as well as natural disasters. While a lot of work has been done to understand the impacts of disasters as mentioned above, little has been understood from an urban design and redesign perspective. Most research in the past has focused on vulnerable and marginalized groups yet not much emphasis was on the modern urban aspects such as increased digital lifestyles and associated behavioral shifts in both consumer activities as well as citizen needs such as education and healthcare. This makes the current COVID-19 pandemic a massive opportunity to understand how this is impacting urban design and how can governments, cities and citizens prepare with mitigation strategies so as to minimize the impacts with intelligent urban redesign and significantly improve urban pandemic resilience (Fig. 1). Since the outbreak of COVID-19, the way governments have globally reacted and responded at national and municipal levels has received significant attention T. Maiti (B) · A. Maiti · T. Singh Deepkapha AI Research, Assen, The Netherlands e-mail: [email protected] A. Maiti e-mail: [email protected] T. Singh e-mail: [email protected] B. Maiti Indian Institute of Technology, Kharagpur, India e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. G. Celbi¸s et al. (eds.), Pandemic and the City, Footprints of Regional Science, https://doi.org/10.1007/978-3-031-21983-2_9

139

140

T. Maiti et al.

Fig. 1 Post-pandemic and impact on urban infrastructure. We have picked the post-pandemic impact on education, healthcare, economic and society. The factors for post-pandemic health care are Digital Psychiatry and telehealth. In economic impact we discuss simulation model and control strategies with economic projections. Societal impact includes social distancing and change in peoples behaviour

from the public and has been demanded from their leaders for better and robust urban infrastructures (Matthew and McDonald, 2006). While the discussion is on the wane, no concrete actions are being taken for the next pandemic and we must not forget that the current urban infrastructure has already been disrupted as shops and restaurants have closed and workers have fled professions that required close physical contact such as barbers, waiters etc. To this effect, the main aim of this systematic review is to understand the impacts of the pandemic on cities, specifically the urban design, from a social, economic, healthcare and education perspective and what major lessons can be drawn for post-COVID urban planning.

2 Socio-economic Impact 2.1 Social Distancing—How It Impacted Urbanization The COVID-19 pandemic had a major effect in the past two years. During the pandemic, cities remained major hotspots for the COVID-19 crisis. Because of dwindling economic activity and, the high rate of population, social interactions and face-to-face interactions were higher and the prompt outbreak of the virus is evident (Sharifi & Khavarian-Garmsir, 2020). In addition, huge populations live clustered

Urbanization Impact Arising from the Behavioral Shift of Citizens …

141

together within limited urban spaces (Glaeser, 2011). Therefore, social distancing was the most prevalent public health action that was imposed by most countries. It involved keeping a 1.5 m distance between people, which could prevent the spread of most respiratory infectious diseases produced by coughing, sneezing, or forced speaking (Qian & Jiang, 2020). To promote social distancing across Europe, South-East Asia and North America, large-scale restrictions on businesses and people’s movements were practiced (Unruh et al., 2022; Woskie et al., 2021). The social measures included closures of schools, closing workplaces, cancellation of public events, and restrictions on internal movement or stay-at-home policies (Brodeur et al., 2021). Prolonged implementation of social distancing had adverse effects on adults and children and has been stressful and overwhelming at times. People felt isolated and lonely, along with increased stress and anxiety, which caused a series of health concerns that included headaches, body pains, stomach problems, skin rashes, and chronic health and mental problems. Increased tobacco, alcohol, and other substances have also been observed (Roberts et al., 2021). The investigation by Meyer et al., (2020) reports that COVID-19 public health restrictions have significantly affected the pattern of physical activity, and an increase in sitting time and screen time has been observed amongst the population. In addition, the behavior of reduction of physical activity is associated with adverse mental health with increased stress and anxiety in individuals. The UK Biobank study highlights that maintaining or increasing physical activity (Singh et al., 2020) is of utmost importance. An inactive person has a 32% increased risk of hospitalization from COVID-19. Previous research indicates that preventing people from exercising was consistently associated with increased depressive and anxiety symptoms (Impact of COVID-19 and lockdown on mental health of children and adolescents: A narrative review with recommendations. Psychiatry Research, 2020) that those social interactions led to lower mortality and suicide rates, better physical health, and higher psychological well-being.

2.2 Post-pandemic Social Distancing Lift and Impact on Social Well-Being In the post-pandemic era, to alleviate the devastating consequences of the COVID-19 pandemic, cities need to find a balance between social distancing and social interactions. A theoretical framework as proposed by Askarizad and He (2022) represents the interactional correlation based on a few proposed variables (Fig. 2). As described earlier, social interactions led to depression and unconventional behavior where as social interactions maintains security and stability. A balance between the two states is obtained through psychological emotional interaction which leads to improved urban design (Fig. 2).

142

T. Maiti et al.

Fig. 2 Pandemic’s implications on urban design. Urban design has been modified by keeping a balance between social distancing and social interaction. The diagram highlights the positive and negative aspects of social distancing and social interaction respectively. The post-pandemic urban design incorporates at least two elements that are discussed here. Image modified from Askarizad and He (2022) impact on improving the built environment

During the pandemic increased focus was put on public spaces (Kane et al., 2021), to maintain a balance between social interactions and social distancing. As shown in Fig. 2 social distancing did put constraints on social behavior which affected psychological emotion interaction. However, social distancing proved to be the most effective to delay and flatten the spread of coronavirus (Matrajt & Leung, 2020). These constraints and opportunities generated the need for modification in urban design–cities’ rules and structures–and urban life (Urban mobility at a tipping point, 2015; Kane et al., 2021). Public spaces will play a key role in restarting cities after the COVID-19 pandemic by providing environments for community connection and social well-being (Innocent & Stevens, 2021). Post-pandemic urbanization or evolution in metropolises needs to balance social distancing and social interactions simultaneously. In addition, it should be as responsive as possible to improve urban resilience against similar pandemic perils in the future (Kane et al., 2021). The first aspect involved redesigning urban furniture. When people sit next to each other they engage in social interactions regardless of social distancing (Askarizad & He, 2022). Keeping this aspect in mind, arranging urban furniture in a way that provides opportunities for face-to-face contact can lead to increased social interactions along with the recommended social distance (Kane et al., 2021) proposed that furniture can be arranged in a grid system where every single grid would be equivalent to 60 cm reinstating social interactions which give up on social distancing. This also controls social loneliness and depression.

Urbanization Impact Arising from the Behavioral Shift of Citizens …

143

The second aspect of new urbanization concepts is the use of social media for the betterment of mental health where people feel connected and less lonely (Boursier et al., 2020). Social media’s impact is well pronounced in the urban sphere as they help in renewing connections and aid in playful forms of engagement. In the longterm reshaping of urban design, social media platforms can be beneficial (Wilkinson et al., 2020). These platforms can help adopt these playful designs through increased awareness and popularity (Agnihotri, 2020) of longstanding tactical urbanism initiatives. How social media has benefitted the mental well-being of people, has been discussed in the ‘Tele-health’ section.

2.3 Economic Impact During Pandemic World—Based on Epidemiological Models for Social Distancing Mitigation strategies during-pandemic period For this study, the incidental immunization of the population through low–risk groups by daily control of the level of social interaction. It is a strategy based on the adaptive triggering strategy suggested in (Ferguson et al., 2020) that we extend by various features. The ideas that led to this paper were generated during the WirVsVirus hackathon of the German federal government in the working group 1_044_flattenthecurve. As part of this mitigation strategy, we considered a discretized epidemiological model based on the standard SIR (S—susceptible, I—infected, R—recovered) model but extended by a control input k0 (n) ∈ [0.2, 1] modelling the level of social interaction in the population and a delay n d describing the time between the infection of an individual and the point of time when the disease is registered by the authorities. The model was based on the assumptions that all infected individuals were recorded, that an individual that has once gone through an infection is unsusceptible to a successive disease and that no medication or vaccine is available. We considered a family of models instead of a single model because the parameters of the model would be rather difficult to deduce from empirical data and the experimental controller, we propose here to be robust with respect to the choice of those parameters. With α describing how easily an infection is transmitted from an infected to a susceptible individual and β describing how fast an infected individual recovers, the equations for this family of models, indexed by i, are given by S(n + 1) = S(n) − αi k0 (n)S(n)I (n), I (n + 1) = I (n) + αi k0 (n)S(n)I (n) − βi I (n), R(n + 1) = R(n) + βi I (n),

(1)

144

T. Maiti et al.

Y (n + 1) = I (n − n d ),

n ≥ nd

f or n ≥ 0 with initial conditions, S(0) = S0 , I (0) = I0 ,

(2)

R(0) = 0, and Y (n) being the number of newly registered cases, which constitutes the output of the system that can be measured. Assume that a certain fraction of infected individuals given by κ requires intensive care and that the number N (n) of available intensive care units (ICU) is given as N n = N 0 + γ n,

(3)

corresponding to the situation of a given number of ICUs at the beginning of the epidemic and a linear increase in capacity over time, where γ defines the daily increase in capacity. N (n) will be relevant later when we define the control objective. Assuming that the population can be split into a high–risk and a low–risk sub–population with corresponding values for κ. Concerning the uncertainty of the control input, we will consider two variants. One, where the control input k0(n) can be precisely set and one, where it is subject to uncertainty reflecting the real world situation where social interaction will hardly follow the desired value in a precise manner. For this second case we assume that the effective level of social interaction will vary from the desired one by a uniformly distributed random variable in [−0.05, 0.05]. Target Definition and System Properties The target considered in this paper was to reach a sufficient level of immunity in the entire population to prevent a successive outbreak of the epidemic when the level of interaction k0(n) is set to 1 at the end of the intervention. And to reach this without exceeding the capacity of ICUs while finding an optimal trade-off between minimal time of the intervention and minimal reduction of social interactions, corresponding to the minimization of health–related, social, and economic burden that the epidemic itself and the intervention constitute. From Eq. (1) it follows directly that S(n) is monotonously decreasing, that I (n) locally follows an exponential-like growth for αk0(n)S(n) > β and an exponentiallike decay for αk0(n)S(n) < β and that R(n) is monotonously increasing. Exponential decay for the case of no intervention, i.e. k0 (n) = 1, n > n immunit y marks the point where the epidemic stops by itself. This effect is often referred to as group or herd immunity and is given for S(n) ≤ β. Note, that this does not mean that there

Urbanization Impact Arising from the Behavioral Shift of Citizens … Table 1 Description of model parameters

Parameter

145

Description

Nominal value

Range

α

Transmission coefficient

6.8 ×

±10%

β

Recovery coefficient

0.15

±10%

γ

Growth rate ICUs

100

–

κ

Fraction of cases requiring ICI

0.01

–

S0

Initial susceptible population

60.3 × 106

–

I0

Initial number of infected

1100

–

10–9

will be no further infections, but that further infections do not lead to an epidemic outbreak. The group is immune to an epidemic, not the individual to an infection. Simulation Model The implementation of Eq. (1) was straightforward, as the parameters are set and since the system has already processed a set of difference equations. The parameters for the model are shown in Table 1. Simulation data was used for the German region. The nominal growth and decay parameters α and β and their ranges were determined based on case numbers from Italy (see Italian Ministry of Health, 2020) and lead to a nominal basic reproduction number R0 = 2.73. The range of the parameters was chosen to be ± 10%, which reflects cases of R0 in the range of [2.48, 3.04]. Two interventions bringing the social interaction factor k0(n) down to 0.63 and 0.32, respectively, were identified, corresponding to the first restriction on mobility, closure of institutions, and the initiation of financial support measures between February 23 and February 29, 2020, and the lockdown of the entire country on March 11, 2020. System behavior without intervention Having identified a family of models in the previous section, the first question that should be answered is how the system behaves in case of no intervention except the isolation of the high–risk portion of the population. The result of this scenario is illustrated in Fig. 3. They show that depending on the model parameters, the peak of the epidemic is reached between 6 and 9 weeks after the outbreak and that the maximum number of cases requiring an ICU exceeds the available capacity of ICUs by a factor of approximately 5–9. The level of immunity within the low-risk population will have reached 90–95%. Economic Impact based on System Behavior without intervention See Fig. 4.

146

T. Maiti et al.

Fig. 3 Model predictions for all models for the low–risk subpopulation under the assumption that the high-risk sub-population is isolated, and no other intervention takes place, i.e. k0(n) = 1 for all n. All scenarios show a rapid transition through the epidemic resulting in herd or group immunity and a significant overload of the health care system

Fig. 4 Image generated based on no intervention in the economic impact, where the simulation model used data from Germany. The economic loss would amount to approximately e690Bn (Oliver et. al) according to the model

System behavior with adaptive triggering As an alternative to trying simple mitigation, Askarizad and He (2022) suggests using a simple 2–point controller with switch on and switch off thresholds. In this contribution, the levels between which the controller switches were chosen to be 70 and 30% of the normal level of social interaction. These values were chosen as they are relatively close to the two levels observed for the data from Italy and because they guarantee an increase and decrease, respectively, of the new case numbers even in case of uncertain parameterization. Figure 5 shows the results of such a strategy where the switch on and the switch off point were both chosen to be at 40% of the current ICU capacity.

Urbanization Impact Arising from the Behavioral Shift of Citizens …

147

Fig. 5 Predictions for all models for the low-risk sub-population under the assumption that the high-risk sub-population is isolated and a 2–point switching control is used, the switching point being at 40% of the ICU capacity

Fig. 6 Image generated based on optimized/adaptive triggering intervention where the simulation model used data from Germany. The model predicts that the economic loss would amount to approximately e340Bn

Economic Impact based on System Behavior with adaptive intervention See Fig. 6.

2.4 Economic Impact: How It Played Out in the Post-pandemic World While we are still at the tail end of COVID-19, the prospects for a speedier recovery have increased, especially with positive news about progress in vaccine production, approval, and deployment and a faster-than-expected global rebound in the latter half

148

T. Maiti et al.

Fig. 7 The November 2019 OECD Economic Outlook projections are extended into 2022 using the November 2019 estimates of the potential output growth rate for each economy in 2021

of 2020 and 2021 but more specifically 2022 [Fig. 7], but there are signs of increasing divergence in activity developments across sectors and economies. But risks remain as some strong signs of inflation have begun to emerge affecting consumer confidence. Financial market vulnerabilities persist, specifically in increased equity valuations, and strong house piece growth due to high demand in the face of short supply. Finally, the enormous cost of COVID-19 still looms on our heads. Our simulation model developed in March 2020 had projected an approximately plateaued EUR 340 billion worth of economic damage and this amount was quite close to the report from ifo forecast published in Feb 2022. For Germany, the costs of COVID-19 for 2020 and 2021 amounted to EUR 330 billion corresponding to a loss of 10% of 2019’s overall economic output. Future losses amounting from long covid, depression, longterm illnesses and impact to brick-and-mortar businesses due to hyper digitalization and shortfalls in education are yet to be considered.

3 Education Impact and Urban Redesign 3.1 Education in Pandemic: Then and Now The emergence of the Corona Virus disease (COVID-19) has led scientists to investigate its effect on varied aspects of society. School and university closures globally due to the COVID-19 have left one in five students out of school. According to UNESCO, by the end of April 2020, 186 countries have implemented nationwide

Urbanization Impact Arising from the Behavioral Shift of Citizens …

149

closures, affecting about 73.8% of the total enrolled learners (UNESCO, 2020). The demographic profile of the participants has little influence on the parameters of major concerns like mental health, social life, education, and so on. Therefore, the generalization of some of the events is possible worldwide. In this section, we discuss the importance of including different online modes of education for school children as well as in higher education. We discuss the infrastructures already available as we need better infrastructure to impart education in post pandemic’s new urban era. Coming to the Indian context, a nationwide lockdown was imposed on 24th March, 2020 resulting in the closure of all educational institutions. After a brief spell, the institutions reopened but the classes were conducted in a new format called ‘online mode’ classes. It was a big challenge for the teachers to reach the students and teach them in a manner of ‘off-line mode’ classes. During this period, the emergence of numerous software for managing ‘online mode’ classes was observed. This development is welcome as it will be useful for future planning of learning tools (Strengthening online learning when schools are closed: The role of families and teachers in supporting students during the COVID-19 crisis. OECD Policy Responses to Coronavirus (COVID-19), 24 September 2020; Muthuprasad et al., 2021). From the interaction with several teachers both from government-funded schools and top government-funded institutions in India, the difficulties and its types were elaborated by them are presented below. The internet connectivity issue is a major problem for rural students. Sometimes they need to go to a nearby town to fetch better internet facilities. All the students need to have their own electronic gadgets and a source of internet connectivity in order to attain online classes (Chatterjee et al., 2020). This is an economic barrier for many students. It has also been observed that some students belonging to rural locality must shift to nearby cities in search of better facilities for online classes. This migration from rural locality to city is commonly called urbanization, which happens to be a global problem and additionally, urbanization is one of the major social changes in the whole world. The sudden switch over to a new mode of teaching perplexed the students. The atmosphere, say, at home, is totally different from a normal classroom; there will be inquisitive peeping eyes; noise arising out of conversations with family members, and so on and so forth; the concentration required to listen to a class is lost. The interaction with the teacher, which was so spontaneous, became hesitant abruptly. Most of the students chose not to keep the camera on, and many students joined the class just for the sake of attendance (Castelli & Sarvary, 2021). However, at a later stage, some schools or institutions made appearances mandatory on the camera. Students who took laboratory courses had to depend on ‘virtual laboratory classes’, which invariably cannot give the feeling of hands-on experiments. Online examination was new to most of the students. Moreover, the students had to familiarize themselves with different methods which were adopted to evaluate the students’ performance. For school going children, the parent’s role has become important in the context of online learning and evaluation process. Therefore, the playground was never even. A student having learned parents will score better than a student of not-so-learned parents (Wadhwa, 2021).

150

T. Maiti et al.

Most likely, the worst sufferers were the research students, in particular, those who were involved in the laboratory experiments. A large number of studies have been reported about the mental health of the students due to the pandemic (Chaturvedi et al., 2021; Cost et al., 2022; Ganesan et al., 2019; Gupta & Agrawal, 2021; Maqsood et al., 2021; Rider et al., 2021; Schwartz et al., 2021). Friends are only visible on mobile screens, devoid of enjoying outdoor games, the boredom of captive life, all arising out of social distancing, mental illness has gradually crept within the students. Moreover, for the students of higher age groups, concern about a future career, in terms of higher studies, job opportunities, settlement in life and so on, have created additional mental stress. The abnormal increase in the use of social media by the students during lockdown compared to pre-lockdown period is also a point of concern (Werling et al., 2021). The pandemic had also posed substantial challenges to the teachers. They had to strive to develop appropriate online teaching methodology leaving aside conventional face-to-face teaching practice (Bashir et al., 2021; Kapasia et al., 2020; Tyagi & Malik, 2020). However, the charm of teaching is to get engrossed in the class through queries and its critical explanations were missing in online classes (Gutte, 2021). They were not able to identify weaker students, due to the lack of eye-to-eye contact. The classes sometimes become very monotonous as very few students do interact in the online classes. It happened to be difficult for them to ascertain if the important component of continuous evaluation—the assignments given were being executed independently by the students or not. Setting a proper question paper for major examinations had been a challenging task. All that was required was to ensure that the student should be able to independently answer the questions and upload the same within the stipulated time. For some students, internet connectivity could become a lacuna for scoring less marks. It was also interesting to note that for the same class, if scrolling over question paper is allowed in one examination, average marks scored by the students increase compared to another examination, where scrolling over the question paper was disallowed. It was further observed that for the same subject taught by the same teacher, overall marks scored by the students were more for the online class than the marks scored in the offline format. In general, students got higher grades last year in the online format than the grades they got in the pre-pandemic year. In recent times the spreading of the coronavirus has been declining. Most of the institutions call back the students in the respective campuses in a phased manner subject to maintaining proper covid protocol as per government directives. This is a welcome move for the students. As a matter of fact, students are demanding that on their campuses. This could be the beginning of a new era, where we might see a mixed format of teaching patterns with large classes that may continue to be held in online mode. Still, examinations may be held following earlier procedures. The experience gathered in the last two years of online teaching could be judiciously used for a better learning process for the students (Almahasees et al., 2021; Dhawan, 2020).

Urbanization Impact Arising from the Behavioral Shift of Citizens …

151

The role of students’ choice of courses will be important in the new format of the learning process. The courses will be chosen depending upon their need, convenience, affordability, and flexibility. It may so happen that to fulfill their choice of courses and to get quality and effective education, they need to migrate to urban areas. In general, there are some differences in the quality of education imparted by the institutions in urban and rural areas. The education in urban institutes is relatively more effective than that of rural areas. The reasons are availability and retentivity of good teachers (García & Weiss, 2019), availability of resources, availability of support staff etc. Therefore, it is interesting to observe how education is contributing to the cause of urbanization. What is important at this juncture is that, after two years of online mode of teaching, teachers have an additional responsibility of examining learning deficiencies, if any, perseveres in individual students. Their mental health should also be examined for requisite fitness for social mixing. Let us hope for new dawn where students can benefit from a new redesigned urban infrastructure.

4 Healthcare Impact and Urban Redesign The COVID-19 pandemic pointed toward the gaps and inefficiencies in health systems worldwide. To address immediate crises during the pandemic, health systems were operating at or above capacity (Ripp et al., 2020). The unprecedented pressure on healthcare had consequences for the general population which will continue long after the pandemic has subsided (Cheristanidis et al., 2021). This creates the need to restructure healthcare system with advanced technologies. Healthcare infrastructure has a direct correlation with urbanization both positive and negative (Shao et al., 2022). The positive factors of urbanization include better healthcare resources (Miao & Wu, 2016; Yang et al., 2013) and the negative factors include sedentary, stressful lifestyles (Turan & Besirli, 2008), unbalanced nutrition (Eckert & Kohler, 2014), air pollution (Diao et al., 2020) and so on. In post-pandemic era, COVID-19-related challenges involve evolving epidemic control and preventative measures and mass vaccine roll out (Beh & Fisher, 2022). These additional aspects of healthcare infrastructure will be incorporated into urban redesigning. These and more will create health demands that systems must prepare for. A whole-of-society and a whole-of-economy approach are needed in the health system worldwide reorganization (Lazard & Youngs, 2021). A few vital aspects such as patient-centered care, and integrated care models supported by digitalization are essential for reorganization and redesigning the healthcare system. To achieve this redesigning healthcare professionals also need to be trained to use digital health platforms. Baxter’s report mentions the following recommendations for the future of healthcare in Europe (Baxter, 2021) (a) Make the patient the point of care (b) Promote

152

T. Maiti et al.

integrated care pathways through community and home-based care and (c) Make the architecture of healthcare digital by default to support integrated care. Another novel way of managing health care is Stepped care (Sheehan et al., 2022). Stepped care is delivering care based on a patient’s needs (Bower & Gilbody, 2005). At first, the most effective and least resource-heavy treatment is provided to patients in need, gradually increasing in steps to resource-heavy treatment based on the patient’s needs. For such care, the healthcare systems require a well-designed and well-prepared care delivery that starts from screening to the overflow of mental illness that emerged during the pandemic (Bower & Gilbody, 2005). As part of stepped care telemedicine, mental health worker visits, delivery, and teaching care via technology platforms (Bower & Gilbody, 2005). All these aspects of health care need to be included in post-pandemic urbanization. In this section, we define the importance and use of (a) telehealth or telemedicine (b) digital psychiatry and (c) managing mental health through digital transformation.

4.1 Importance of Telehealth in the Post-pandemic Era Telemedicine, also referred to as telehealth or e-medicine, is the remote delivery of healthcare services, including exams and consultations, over the telecommunications infrastructure (Telemedicine, 2022). Telemedicine allows healthcare providers to evaluate, diagnose and treat patients without the need for an in-person visit (Telemedicine, 2022). Telehealth services are widely in use in night-time radiology coverage, urgent services, and mandated services (e.g., the delivery of health care services to prison inmates). The current coronavirus pandemic (COVID-19) has seen tremendous growth in telehealth services around the world (Megahed & Ghoneim, 2020). Mobile health is another concept that describes services supported by mobile communication devices, such as wireless patient monitoring devices, smartphones, personal digital assistants, and tablet computers (Weinstein et al., 2014). Although telehealth has been available for many decades, the COVID-19 experiences have reinstated raised awareness of telehealth in all spheres of society. At this time in the new pandemic era, many jurisdictions have considered the importance of telehealth (Bashshur et al., 2020). Modified and overall long-term sustainability of telehealth practice during the new normal includes (Weinstein et al., 2014): (a) developing a skilled workforce; (b) empowering consumers; (c) reforming funding; (d) improving the digital ecosystems; and (e) integrating telehealth into routine care.

4.2 Digital Psychiatry Another aspect of healthcare redesign that will be vital to urbanization is Digital psychiatry. Digital psychiatry assists patients through the ease of connecting with a

Urbanization Impact Arising from the Behavioral Shift of Citizens …

153

mental health specialist over the internet, through the web, or on mobile, during the pandemic era. Cost-effective deliverance of tele mental health care through virtual (videoconferencing) platforms has improved access to care for psychiatric patients. Evidence supports that digital psychiatry as a feasible platform improves outcomes and quality of life across various mental disorders (Bashshur et al., 2016). Because of the ease of internet access and people’s knowledge to use mobile apps, digital psychiatry is popular in the urban sphere. A rise in telepsychiatry has been felt recently, especially during COVID-19 times. For example, during COVID-19, the National Health Service in the UK facilitated the rapid ramp-up of virtual psychiatric care. Other countries such as Israel, Singapore, Korea, and Taiwan also managed the COVID-19 pandemic by responding rapidly and relying and responding heavily on technology (Salvador-Carulla et al., 2020). Hence, it will be very beneficial in post-pandemic urbanization redesign. However, the full potential of telepsychiatry and the potential use of virtual platforms is yet to be revealed for low and middle-income countries (Naeem et al., 2020). A lack of telehealth infrastructure often undermines the expansion of telepsychiatry in these contexts, national telehealth policies, data governance frameworks, and a lack of training and education on the use of telehealth technologies for the health workforce (Naeem et al., 2020). New urban design should be improved with such infrastructures. Torous et al., (2020) recommends the following improvements and better accessibility for digital psychiatry in post-pandemic urbanization: (a) trained clinicians to deliver virtual care. (b) Development of mobile applications powered by artificial intelligence (AI)— such applications have the potential to offer personalized real-time intervention and provide an immediate response to patients’ needs. (c) Real-time remote monitoring of health and real-time clinicians’ interventions— such a platform would lead to a more responsive mental health care system and improve patient outcomes in the post-pandemic world.

4.3 Managing Mental Health Through Digital Transformation The use of digital technologies can bridge social distancing, even while physical distancing measures are in place (Ferguson et al., 2020). In the Urban sphere, it has become increasingly popular because of the ease of access to internet facilities. Digital transformation includes—telemedicine, artificial intelligence-enabled medical devices, blockchain electronic health records, and many more (Srivastava et al., 2021). Digital transformations are entirely reshaping the interaction with health professionals in decision-making and treatment plans. There are around 4.66 billion active internet users (Statistics, 2022). Worldwide online activities in a virtual workspace where people can work or connect over video connection is a post-pandemic trend that can include mental health management.

154

T. Maiti et al.

The previous two sections highlight the development of infrastructure in the health system. This section is focused on other aspects of urbanization such as the workplace and schools. The following norms have been introduced as part of the Employee Engagement New Normal Workplace program (Yadav et al., 2020) to maintain social contact. Activities, where employers can ensure that each employee receives daily outreach video/voice during the workweek through a supervisor or buddy system have been introduced. Employers encourage groups to connect regularly, and individuals can be redirected to mental health support (Rossman, 2022). (Galea et al., 2020) mentions the importance of enhanced check-in functions a feature that can be included to provide regular contact with individuals and for people to share their well-being. Technology in mental health is particularly relevant for school children who are out of school during pandemics and still continue due to the occasional lockdown. To ensure that they have access to regular programmed work and are also in a safe environment and not subject to critical domestic violence and child abuse during isolation (Kluger, 2021) application of digital technologies is extremely necessary. AI-based mechanisms for surveillance, reporting, and intervention, particularly when it comes to domestic violence and child abuse, should be designed (Galea et al., 2020).

4.4 Summary As we can see, new healthcare alternatives have risen since the pandemic’s onset. While it has greatly benefitted the physical and mental well-being of people who, because of lockdown restrictions, couldn’t visit and be with their psychologists or psychiatrists face-to-face. However, it has only benefited the segment of the population that have the resource to avail internet and the education to use it. In short, telehealth couldn’t reach people belonging to the lower echelons of society socioeconomically. Because of its wise application, telehealth, and digital psychiatry, the government should ensure cheaper internet facilities and gadgets that the economically disadvantaged can afford. Also, the initiative must be taken to give primary education about using the internet for daily activities.

5 Conclusion During the writing of this paper, China’s largest city Shanghai with a population of 26 million people, is in lockdown with a new Omicron variant, so clearly, we are not in a post-pandemic world just yet. We foresee policy’s role not only in boosting the economy but also paying a major role continuing to play a major role in reinventing urban design in societal aspects to boost morale, help reduce stressors for citizens and consumers, provide financial and mental support for education so students can recover from those lost years studying from home and finally ensuring the urban

Urbanization Impact Arising from the Behavioral Shift of Citizens …

155

redesign offers economically affordable, technically easy-to-understand and superior servicing models for telemedicine to help patients and healthcare seekers receive the optimal QoS (Quality of Service) they need. While prudent financial instruments have prevented from an economic recession, we also foresee sustained enhanced structural reform efforts will be required in order to raise opportunities for displaced workers, students to accelerate economic activity, and foster resource reallocation towards industry verticals and work-related activities that must foster faster growth, increase resilience and contribute to environmental sustainability and urban redesign. The state of urban infrastructure has been irrevocably changed by the pandemic and will require significant adjusting, as well as reallocation of labor and capital, the extent of these, will be different for the sector. Some areas most affected by physical distancing requirements and associated changes in consumer preferences may be permanently smaller after the crisis. A lasting behavioral shift towards remote or hybrid working, severe reductions in business travel and the increasing digital products and services models and platforms, especially e-commerce, e-learning and telehealth could also change the mix of jobs available and the location of many workplaces which in turn will affect the city and urban design principles yet again. (Fig. 8). While the rapid and hurried shift towards the use of online platforms and remote working during the pandemic has underlined the possibilities provided by digital technologies, we foresee a new paradigm to emerge with respect to shifting humanto-human relationships in this hybrid world. Relationships and social interactions have always been the bedrock of cities. A redesigned urban architecture will play as a catalyst, serving as a focal point of that relationship building. While epidemiologists have painstakingly focused on reuniting the population through mathematical models, we need to rethink from social, linguistic, and cultural programs of creating spaces for communities to reunite them in spaces that will welcome people’s experiences so that they can also physically inhabit the city where they live.

Fig. 8 The post-pandemic world looks very different and that will impact the urban infrastructure even further. Source OECD ECONOMIC OUTLOOK, INTERIM REPORT MARCH 2021 (OECD Economic Outlook, 2022)

156

T. Maiti et al.

We must reimagine cities as a hospitable and welcoming place, and we must redesign the urban infrastructure to facilitate this change. The past two years have effectively disrupted human activities in cities, and we must use this moment of interruption to reflect back on how we can not only bring the herd immunity back to epidemiologically acceptable levels but also reunify humans and their pets on to the streets and ports, canals and bridges so they can rediscover nature and urban spaces.

References “Urban mobility at a tipping point.” McKinsey.com (2015, Sept 1st). Agnihotri, R. (2020). Social media, customer engagement, and sales organizations: A research agenda. Industrial Marketing Management, 90, 291–299. Almahasees, Z., Mohsen, K., Amin, M. O. (2021). Faculty’s and students’ perceptions of online learning during COVID-19. Frontiers in Education. https://doi.org/10.3389/feduc.2021.638470 Askarizad, R., & He, J. (2022). Post-pandemic urban design: The equilibrium between social distancing and social interactions within the built environment. Cities, 124, 103618. Bashir, A., Bashir, S., Rana, K., Lambert, P., Vernallis, A. (2021). Post-COVID-19 adaptations; the shifts towards online learning, hybrid course delivery and the implications for biosciences courses in the higher education setting. Frontiers in Education. https://doi.org/10.3389/feduc. 2021.711619 Bashshur, R. L., Shannon, G. W., Bashshur, N., & Yellowlees, P. M. (2016). The empirical evidence for telemedicine interventions in mental disorders. Telemedicine Journal and E-Health, 22, 87– 113. https://doi.org/10.1089/tmj.2015.0206 Bashshur, R. L., Doarn, C. R., Frenk, J. M., Kvedar, J. C., Shannon, G. W., & Woolliscroft, J. O. (2020). Beyond the COVID pandemic, telemedicine, and health care. Telemedicine and e-Health, 26(11), 1310–1313. Baxter reports third-quarter 2021 results. Baxter. (2021, October 28). Retrieved December 3, 2022, from https://www.baxter.com/baxter-newsroom/baxterreports-third-quarter-2021-results Beh, D. L., & Fisher, D. A. (2022). The global perspective on the COVID-19 pandemic. In COVID-19 pandemic in Singapore (pp. 41–95). World Scientific. Boursier, V., Gioia, F., Musetti, A., & Schimmenti, A. (2020). Facing loneliness and anxiety during the COVID-19 isolation: The role of excessive social media use in a sample of Italian adults. Frontiers in Psychiatry, 11, 586222. Bower, P., & Gilbody, S. (2005). Stepped care in psychological therapies: Access, effectiveness and efficiency: Narrative literature review. The British Journal of Psychiatry, 186(1), 11–17. Brodeur, A., Grigoryeva, I., & Kattan, L. (2021). Stay-at-home orders, social distancing, and trust. Journal of Population Economics, 34(4), 1321–1354. Castelli, F. R., & Sarvary, M. A. (2021). Why students do not turn on their video cameras during online classes and an equitable and inclusive plan to encourage them to do so. Ecology and Evolution, 11(8), 3565–3576. https://doi.org/10.1002/ece3.7123 Chatterjee, M., Saha, S., & Jha, M. (2020). Transitioning classes into electronic gadgets: Life of students in lockdown. Down to Earth, 21 August 2020. https://www.downtoearth.org.in/blog/hea lth/transitioning-classes-into-electronic-gadgets-life-of-students-in-lockdown-72962 Chaturvedi, K., Vishwakarma, D. K., Singh, N. (2021). COVID-19 and its impact on education, social life and mental health of students: A survey. Children and Youth Services Review, 121, 105866. https://doi.org/10.1016/j.childyouth.2020.105866. Cheristanidis, S., Kavvadas, D., Moustaklis, D., Kyriakidou, E., Batzou, D., Sidiropoulos, E., ... & Papamitsou, T. (2021). Psychological distress in primary healthcare workers during the COVID19 pandemic in Greece. Acta Medica Academica, 50(2).

Urbanization Impact Arising from the Behavioral Shift of Citizens …

157

Cost, K. T., Crosbie, J., Anagnostou, E., Birken, C. S., Charach, A., Monga, S., Kelley, E., Nicolson, R., Maguire, J. L., Burton, C. L., Schachar, R. J., Arnold, P. D., Korczak, D. J. (2022). Mostly worse, occasionally better: impact of COVID-19 pandemic on the mental health of Canadian children and adolescents. European Child & Adolescent Psychiatry. https://doi.org/10.1007/s00 787-021-01744-3 Dhawan, S. (2020). Online learning: A panacea in the time of COVID-19 crisis. Journal of Educational Technology Systems., 49(1), 5–22. https://doi.org/10.1177/0047239520934018 Diao, B., Ding, L., Zhang, Q., Na, J., & Cheng, J. (2020). Impact of urbanization on PM25-related health and economic loss in China 338 cities. International Journal of Environmental Research and Public Health, 17, 990. https://doi.org/10.3390/ijerph17030990 Eckert, S., & Kohler, S. (2014). Urbanization and health in developing countries: A systematic review. World Health Population, 15, 7–20. https://doi.org/10.12927/whp.2014.23722 Ferguson, N., et al. (2020). Impact of non-pharmaceutical interventions (NPIs) to reduce COVID19 mortality and healthcare demand. https://doi.org/10.25561/77482 Galea, S., Merchant, R. M., & Lurie, N. (2020). The mental health consequences of COVID-19 and physical distancing: The need for prevention and early intervention. JAMA Internal Medicine, 180(6), 817–818. Ganesan, B., Al-Jumaily, A., Fong, K.N.K., Prasad, P., Meena, S.K., Tong, R.K.-Y.: Impact of coronavirus disease 2019 (COVID-19) outbreak quarantine, isolation, and lockdown policies on mental health and suicide. 12, Article 565190. https://doi.org/10.3389/fpsyt.2021.565190 García, E., Weiss, E. (2019). The teacher shortage is real, large and growing, and worse than we thought. The first report in ‘The Perfect Storm in the Teacher Labor Market’ series, March 26, 2019. https://www.epi.org/publication/the-teacher-shortage-is-real-large-and-growing-andworse-than-we-thought-the-first-report-in-the-perfect-storm-in-the-teacher-labor-market-series/ Glaeser, E. (2011). Triumph of the city: How urban spaces make us human. Pan Macmillan. Gupta, R., & Agrawal, R. (2021). Are the concerns destroying mental health of college students?: A qualitative analysis portraying experiences amidst COVID-19 ambiguities. Analyses of Social Issues and Public Policy, 2021, 1–19. https://doi.org/10.1111/asap.12232 Gutte, N. (2021). What are the biggest challenges facing online education today? Hurix.com, 10 December, 2021. https://www.hurix.com/what-are-the-biggest-challenges-facing-online-educat ion-today/ https://www.statista.com/statistics/617136/digital-population-worldwide/ https://www.techtarget.com/searchhealthit/definition/telemedicine#:~:text=Telemedicine%2C% 20also%20referred%20to%20as,for%20an%20in%2Dperson%20visit Husain, M. O., Gratzer, D., Husain, M. I., & Naeem, F. (2021). Mental illness in the post-pandemic world: Digital psychiatry and the future. Frontiers in Psychology, 12. Impact of COVID-19 and lockdown on mental health of children and adolescents: A narrative review with recommendations. Psychiatry Research, 293, 113429. https://doi.org/10.1016/j.psy chres.2020.113429 Innocent, T., & Stevens, Q. (2021). Urban play as catalyst for social wellbeing post-pandemic. Frontiers in Computer Science, 3, 634145. Innocent, T., & Stevens, Q. (2021). Urban play as catalyst for social wellbeing post-pandemic. Frontiers in Computer Science, 16. Italian Ministry of Health. (2020). Novel coronavirus. Retrieved April 5, 2020, from http://www. salute.gov.it Kane, G. C., Nanda, R., Phillips, A., & Copulsky, J. (2021). Redesigning the post-pandemic workplace. MIT Sloan Management Review, 62(3), 12–14. Kapasia, N., Paul, P., Roy, A., Saha, J., Zaveri, A., Mallick, R., Das, B. B. P., Chouhan, P. (2020). Impact of lockdown on learning status of undergraduate and postgraduate students during COVID19 pandemic in West Bengal, India, Children and Youth Services Review, 116. https://doi.org/10. 1016/j.childyouth.2020.105194 Kluger, J. (2021). Domestic violence is a pandemic within the COVID-19 pandemic. Time Magazine, 3.

158

T. Maiti et al.

Lazard, O., & Youngs, R. (2021). The EU and climate security: Toward ecological diplomacy. Carnegie Europe, 12. Maqsood, A., Abbas, J., Rehman, G., & Mubeen, R. (2021). The paradigm shift for educational system continuance in the advent of COVID-19 pandemic: mental health challenges and reflections. Current Research in Behavioral Sciences, 2, 100011. https://doi.org/10.1016/j.crbeha.2020. 100011 Matrajt, L., & Leung, T. (2020). Evaluating the effectiveness of social distancing interventions to delay or flatten the epidemic curve of coronavirus disease. Emerging Infectious Diseases, 26(8), 1740. Matthias an der Heiden and Udo Buchholz. Modellierung von Beispielszenarien der SARS-CoV2-Epidemie 2020 in Deutschland. 2020. https://doi.org/10.25646/6571.2 Megahed, N., & Ghoneim, E. (2020). Antivirus-built environment: Lessons learned from Covid-19 pandemic. Sustainable Cities and Society, 61, Article 102350. https://doi.org/10.1016/j.scs.2020. 102350 Meyer, J., McDowell, C., Lansing, J., Brower, C., Smith, L., Tully, M., & Herring, M. (2020). Erratum: Meyer, J., et al. Changes in physical activity and sedentary behavior in response to COVID-19 and their associations with mental health in 3052 US adults. International Journal of Environmental Research and Public Health, 17(18), 6469. International Journal of Environmental Research and Public Health, 17(19), 6949. Miao, J., & Wu, X. (2016). Urbanization, socioeconomic status and health disparity in China. Health & Place, 42, 87–95. https://doi.org/10.1016/j.healthplace.2016.09.008 Muthuprasad, T., Aiswarya, S., Aditya, K. S., Jha, G. K. (2021). Students’ perception and preference for online education in India during COVID-19 pandemic. Social Sciences & Humanities Open, 3(1). https://doi.org/10.1016/j.ssaho.2020.100101 Naeem, F., Husain, M. O., Husain, M. I., & Javed, A. (2020). Digital psychiatry in low- and middleincome countries post-COVID-19: Opportunities, challenges, and solutions. Indian Journal of Psychiatry, 62, S380–S382. https://doi.org/10.4103/psychiatry.IndianJPsychiatry_843_20 OECD Economic Outlook. (2022). Interim report March 2022: economic and social impacts and policy implications of the war in Ukraine. OECD Publishing, Paris. https://doi.org/10.1787/418 1d61b-en Oliver, M., Schwerk, A., Singh, T., Maiti, T. Intelligent unlocking of a population through active steering back to immunity in an epidemic amidst locally unstable delay systems (currently in pre-print draft). Press Release, ifo Instituut. (2020). Coronavirus pandemic caused EUR 330 billion in economic losses for Germany. https://www.ifo.de/en/node/68067 Qian, M., & Jiang, J. (2020). COVID-19 and social distancing. Journal of Public Health, 1–3. Rider, E. A., Ansari, E., Varrin, P. H., Sparrow, J. (2021). Mental health and well-being of children and adolescents during the covid-19 pandemic. BMJ, 374(1730). https://doi.org/10.1136/bmj. n1730 Ripp, J., Peccoralo, L., & Charney, D. (2020). Attending to the emotional well-being of the health care workforce in a New York City health system during the COVID-19 pandemic. Academic Medicine. Roberts, A., Rogers, J., Mason, R., Siriwardena, A. N., Hogue, T., Whitley, G. A., & Law, G. R. (2021). Alcohol and other substance use during the COVID-19 pandemic: A systematic review. Drug and Alcohol Dependence, 229. Rossman, A. (2022). Community directed exploration of expressive therapy applications to combat barriers to mental health for first responders. Salvador-Carulla, L., Rosenberg, S., Mendoza, J., Tabatabaei-Jafari, H., Pandemic-Mental Health International, N. (2020). Rapid response to crisis: Health system lessons from the active period of COVID-19. Health Policy and Technology, 9, 578–586. https://doi.org/10.1016/j.hlpt.2020. 08.011

Urbanization Impact Arising from the Behavioral Shift of Citizens …

159

Schwartz, K. D., Exner-Cortens, D., McMorris, C. A., Makarenko, E., Arnold, P., Van Bavel, M., Williams, S., Canfield, R. (2021). COVID-19 and student well-being: stress and mental health during return-to-school. Canadian Journal of School Psychology, 36(2), 166–185. Shao, Q., Tao, R., & Luca, M. M. (2022). The effect of urbanization on health care expenditure: Evidence from China. Frontiers in Public Health, 10. Sharifi, A., & Khavarian-Garmsir, A. (2020). The COVID-19 pandemic: Impacts on cities and major lessons for urban planning, design, and management. Science of the Total Environment, 749, Article 142391. https://doi.org/10.1016/j.scitotenv.2020.142391 Sharifi, A., & Khavarian-Garmsir, A. R. (2020). The COVID-19 pandemic: Impacts on cities and major lessons for urban planning, design, and management. Science of the Total Environment, 749, 142391. https://doi.org/10.1016/j.scitotenv.2020.142391 Sheehan, K. A., Schulz-Quach, C., Ruttan, L. A., MacGillivray, L., McKay, M. S., Seto, A., ... & Berkhout, S. G. (2022). “Don’t Just Study our Distress, Do Something”: Implementing and evaluating a modified stepped-care model for health care worker mental health during the COVID19 pandemic. The Canadian Journal of Psychiatry, 07067437221111372. Singh, S., Roy, D., Sinha, K., Parveen, S., Sharma, G., Joshi, G. Srivastava, S., Singh, S. V., Singh, R. B., & Shukla, H. K. (2021). Digital transformation of healthcare: A blockchain study. International Journal of Innovative Science, Engineering & Technology, 8(5). Strengthening online learning when schools are closed: The role of families and teachers in supporting students during the COVID-19 crisis. OECD Policy Responses to Coronavirus (COVID-19), 24 September 2020. https://www.oecd.org/coronavirus/policy-responses/strengthe ning-online-learning-when-schools-are-closed-the-role-of-families-and-teachers-in-supportingstudents-during-the-covid-19-crisis-c4ecba6c/ Torous, J., Jan, M. K., Rauseo-Ricupero, N., & Firth, J. (2020). Digital mental health and COVID19: Using technology today to accelerate the curve on access and quality tomorrow. JMIR Mental Health, 7, e18848. https://doi.org/10.2196/18848 Turan, M. T., & Besirli, A. (2008). Impacts of urbanization process on mental health. Anatolian Journal of Psychiatry, 9, 238–243. https://doi.org/10.5801/ncn.v14i2.585 Tyagi, H., Malik, S. (2020). Online teaching in Delhi-NCR schools in India during Covid-19 pandemic. Indian Journal of Science and Technology, 13(38), 4036–4054. https://doi.org/10. 17485/IJST/v13i38.1613 UNESCO (2020). COVID-19 educational disruption and response (2020). https://en.unesco.org/ themes/educationemergencies/coronavirus-school-closures Unruh, L., Allin, S., Marchildon, G., Burke, S., Barry, S., Siersbaek, R., ... & Williams, G. A. (2022). A comparison of 2020 health policy responses to the COVID-19 pandemic in Canada, Ireland, the United Kingdom and the United States of America. Health Policy, 126(5), 427–437. Wadhwa, W. (2021). Covid-19 and mental health: Are children ready to go back to school? Retrieved September, 24, 2021, from https://www.ideasforindia.in/topics/human-development/covid-19and-mental-health-are-children-ready-to-go-back-to-school.html Weinstein, R. S., Lopez, A. M., Joseph, B. A., Erps, K. A., Holcomb, M., Barker, G. P., & Krupinski, E. A. (2014). Telemedicine, telehealth, and mobile health applications that work: Opportunities and barriers. The American Journal of Medicine, 127(3), 183–187. Werling, M., Walitz, S., Grünblatt, E., Drechslera, R. (2021). Media use before, during and after COVID-19 lockdown according to parents in a clinically referred sample in child and adolescent psychiatry: Results of an online survey in Switzerland. Comprehensive Psychiatry, 109. https:// doi.org/10.1016/j.comppsych.2021.152260 Wilkinson, A., Ali, H., Bedford, J., Boonyabancha, S., Connolly, C., Conteh, A., ... & Whittaker, L. (2020). Local response in health emergencies: Key considerations for addressing the COVID-19 pandemic in informal urban settlements. Environment and Urbanization, 095624782092284. Woskie, L. R., Hennessy, J., Espinosa, V., Tsai, T. C., Vispute, S., Jacobson, B. H., ... & Gabrilovich, E. (2021). Early social distancing policies in Europe, changes in mobility & COVID-19 case trajectories: Insights from Spring 2020. PLoS One, 16(6), e0253071.

160

T. Maiti et al.

Yadav, S., Dubey, A., Chawla, P., & Jha, S. (2020). Employee engagement strategy for employees working in virtual environment in the IT industry. Yang, G., Wang, Y., Zeng, Y., Gao, G. F., Liang, X., Zhou, M., et al. (2013). Rapid health transition in China, 1990–2010: Findings from the global burden of disease study 2010. Lancet, 381, 1987–2015. https://doi.org/10.1016/S0140-6736(13)61097-1

The Socioeconomic and Labor Market Effects of Pandemics on Cities and Local Economies

What Happened After SARS in 2003? The Economic Impacts of a Pandemic Ilan Noy, Yasuyuki Sawada, Nguyen Doan, and Canh Phuc Nguyen

1 Introduction The current COVID-19 pandemic has already had a catastrophic economic impact across the world, but despite a rapidly growing literature, the likely overall impacts of this pandemic are still largely unknown (Altig et al., 2020; Baker et al., 2020; Beach et al., 2022; Doan & Noy, 2021; Hailu, 2020). It therefore may be instructive to revisit the economic impacts of previous disease outbreaks.1 Such information may also have significant policy implications because of the government’s decisions 1 The economic impacts of disease outbreaks have been getting sporadic interest from economists for many years, though most past research efforts have been directed at understanding the economic impact of non-infectious (or non-epidemic) diseases, and health more broadly (Baldwin & Weisbrod, 1974; Gillies et al., 1996; Weisbrod et al., 1974). For example, HIV/AIDS (Barnett et al., 2000; Dixon et al., 2002; Kabajulizi & Ncube, 2017), avian influenza H5N1 (Bloom et al., 2005), and dengue (Castañeda-Orjuela et al., 2012). The one epidemic that has received more research attention is HIV/AIDS, especially within the context of African development, but also in high-prevalence countries or regions elsewhere. Dixon et al. (2002) find that the spread of AIDS led to reductions of labour supply, productivity, exports and overall economic development in Africa in the 1990s. More recently, Kabajulizi and Ncube (2017) evaluate the transition of the management of AIDS into a chronic condition requiring investment in continuing treatment, and investigate the impact of these fiscal costs on Uganda’s economic development.

I. Noy (B) Victoria University of Wellington, Wellington, New Zealand e-mail: [email protected] Y. Sawada University of Tokyo, Bunkyo City, Japan N. Doan University of Finance-Marketing, Ho Chi Minh City, Vietnam C. P. Nguyen University of Economics, Ho Chi Minh City, Vietnam © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. G. Celbi¸s et al. (eds.), Pandemic and the City, Footprints of Regional Science, https://doi.org/10.1007/978-3-031-21983-2_10

163

164

I. Noy et al.

over timing and extent of reopening depend critically on the likely future path of the pandemic-economic curve. Recent studies have examined the evidence about the economic impact of the 1918–19 pandemic flu to better understand the potential economic impacts of COVID-19 (Almond, 2006; Barro et al., 2020; Beach et al., 2018; Boberg-Fazlic et al., 2021; Guimbeau et al., 2020; Helgertz & Bengtsson, 2019; Karlsson et al., 2014; Noy et al., 2020). Starting from Almond (2006), several papers have investigated the long-term impact of exposure to an epidemic on in-utero human development by focusing specifically on the 1918–19 influenza pandemic and the cohort that was born immediately after it (Beach et al., 2018; Boberg-Fazlic et al., 2021; Guimbeau et al., 2020). Karlsson et al. (2014) focus on Sweden’s experience with the 1918–19 influenza to describe in more detail the impact of the pandemic on macroeconomic outcomes, while Noy et al. (2020) do the same for Japan’s experience with the 1918– 19 event, and Jinjarak et al. (2022) on the 1968 Influenza pandemic. There is also a growing literature identifying the economic impacts of the 2014 Ebola epidemic in West Africa (Campante et al., 2020; Kostova et al., 2019; Maffioli, 2021). For a survey of this literature, see Noy and Uher (2021). Others have attempted to estimate the economic impact of epidemics by looking at a panel of country-level macroeconomic data together with a historical record of past epidemics. For example, Jordà et al. (2022) used the rates of return on assets with data going back to the fourteenth century to study medium and long run impacts of pandemics. They find that the macroeconomic after-effects of pandemics can persist for decades. Barro et al. (2020) use cross-country comparisons in the aftermath of the 1918–19 influenza to identify declines in GDP and consumption of 6–8%. Overall, motivated by the COVID-19 crisis, these recent papers suggest that the adverse economic impacts of pandemics may persist for a long period of time.2 The SARS epidemic in 2003 is of particular interest, given the genetic and symptomatic similarities between the two coronaviruses, and the ways in which the affected economies reacted (at least initially). Some previous studies have modelled the impacts of SARS on the affected Asian economies. For instance, in a widely cited paper, Lee and McKibbin (2004) concluded from their model that SARS had caused 2.63, 1.05, 0.49, and 0.47 percentage point decline in annual GDP for the most heavily affected countries of Hong Kong, China, Taiwan, and Singapore, respectively. Based on a computable general equilibrium model for Asia–Pacific, they modelled the direct and indirect economic impacts of SARS rather than focusing only on the affected industries such as healthcare, tourism, or retail service sector as other studies have done (Chou et al., 2004; Siu et al., 2004). In contrast with their approach, here we examine post-pandemic data, rather than rely on semi-real-time structural modelling as is done by Lee and McKibbin (2004). Our study makes an important contribution by introducing two novelties in reexamining the economic impacts of SARS. First, we apply a modern statistical method, Synthetic Control (Abadie, 2021; Abadie & Gardeazabal, 2003; Abadie 2

While the literature on the economic impacts of COVID-19 has been extensive already, we do not survey this literature here. For a literature review, see Brodeur et al. (2020).

What Happened After SARS in 2003? The Economic Impacts …

165

et al., 2010). This allows us to rigorously estimate the counterfactual growth trajectory for the affected economies, provinces, and cities, without the epidemic. We also utilise a more recent alternative Synthetic Control approach which aims to improve the pretreatment fit of the counterfactual—the Augmented Synthetic Control introduced by Ben-Michael et al. (2021). Second, we employ both conventional macroeconomic data and night-time light luminosity data from satellites that were not available in the immediate aftermath of the epidemic when most previous research was done. More specifically, we examine the impacts of SARS on the affected Asian economies—China, Hong Kong, Singapore, and Taiwan. In China, we also examine the impact of SARS more locally, focusing particularly on Beijing, Guangdong, Hebei, and Shanxi, since these were the provincial-level divisions that were affected the most.3 We find primarily negative effects of the SARS epidemic on economic growth. However, we find that such effects were very short-lived for Hong Kong and Taiwan; the adverse economic impacts of SARS lasted only during the immediate epidemic quarter given its relatively limited spread to other countries and the affected countries’ abilities to stop its spread quickly. Yet, we also find that more localized impacts persisted somewhat longer: When we apply the synthetic control method (SCM) to night-time lights (NTL) data from Beijing and Guangdong, the adverse effect at the local level appears to be longer lasting (this is not the case for the more lightly impacted Hebei and Shanxi). The findings are not entirely consistent with the widely held optimistic view of a rapid recovery after the SARS pandemic, as we show that SARS did lead to statistically observable declines in economic activity in the most heavily affected provincial economies. It seems that national economies have indeed bounced back quickly, but more locally affected areas have taken longer to recover. This result, we note, is not that unusual when one examines sudden-onset disasters and their impact on economic activity. National (or spatially large) areas seem to recover very quickly from a localised event, but the locality affected—by the earthquake, hurricane, or flooding event—does seem to take much longer to recover.4 The paper proceeds as follows: The next section summarizes the background to the economic impacts of 2003 SARS. Section 3 presents the data and methodology. Section 4 discusses the results. Section 5 concludes with some policy discussions and directions for future research.

3

Strictly speaking, Guangdong, Hebei, and Shanxi are provinces, while Beijing is a municipality. But all are one administrative step below the central government (a province-level administrative division). To simplify, we call all province-level subdivisions (provinces, autonomous regions, municipalities) “provinces”. 4 For a review of these findings about sudden-onset events, see Noy and duPont (2018).

166

I. Noy et al.

2 The 2003 SARS Outbreak The most significant presence of SARS was registered in four Asian economies: China, Hong Kong, Singapore, and Taiwan. Once the pandemic became known, the number of international visitors fell precipitously in these economies. Brahmbhatt and Dutta (2008) estimated that the GDP loss amounted to US $13 billion. By all descriptive accounts, these losses did not affect any of these national economies for more than two quarters and even the most heavily affected countries were already growing rapidly by Q3 2003. The observed affects were distributed unequally across sectors; disproportionately affecting tourism, leisure, and transport, especially airlines. In Hong Kong, international visitor arrivals dropped by 65% on the previous year’s figure during April 2003 (Cooperation, 2004). Airlines, and specifically the city’s carrier—Cathay Pacific—cancelled over 45% of their scheduled flights during the epidemic’s peak, and their monthly passenger rate fell by 80% (Noy & Shields, 2019). Notably, cross-border trade, and especially the Hong Kong–China movement of goods, continued without significant disruption. Even the stock market reaction was comparatively mild, with the Hong Kong Seng Index dropping by 1.78% between March 12 and April 30. The main channel of impact during the SARS epidemic was the behavioural change of millions of individuals (Noy & Shields, 2019). Indeed, public opinion surveys at the height of the epidemic reveal that 23% of respondents in Hong Kong, for example, thought that they were either very or somewhat likely to become infected with SARS, which was dramatically incommensurate with the eventual infection rate of only 0.0026% (Leung et al., 2004). Similar exaggerated perceptions were recorded in Taiwan where 74% of survey respondents rated the likelihood of death following SARS contraction as 4-or-5 on a 5-point scale (Liu et al., 2005), while the actual case-fatality rate was 10% (still an order of magnitude higher than COVID19). Disproportionate risk assessments were even found in places hardly affected by the epidemic, such as the U.S. where 16% of survey respondents felt that they or their family were ‘somewhat’ or ‘very likely’ to get infected with SARS in the next 12 months (Brahmbhatt & Dutta, 2008). The economic consequences of an epidemic can usually be delineated into direct and indirect impacts. Direct impacts include lost income and output due to death and symptomatic illness as well as increased healthcare costs, whereas indirect costs arise, specifically in this case, from aggregate behavioural changes driven by the public’s perception of the epidemic outbreak or by government directives.5 Because there was relatively limited mortality and morbidity associated with SARS, its economic analysis differs from some other notable epidemics. Typically, economic losses in such epidemics as HIV/AIDs in the 1980s and 1990s or the influenza of 1918–19 were first, and maybe foremost, measured via the cost of illness and death and the loss of income associated with that mortality and morbidity. This cannot be the basis for an evaluation of the economic impacts of SARS as such an approach will severely under-estimate its costs. 5

Yet, the difference between the two is often very hard to disentangle (Katafuchi et al., 2021).

What Happened After SARS in 2003? The Economic Impacts …

167

3 Data and Methodology 3.1 Data We construct quarterly GDP per capita data using quarterly GDP series (in 2010 USD) taken from the Global Economic Monitor (GEM) of the World Bank. Data on imports and exports are extracted from the International Financial Statistics (IFS) of the International Monetary Fund (IMF) and the Organization of Economic Cooperation and Development (OECD) statistics. Unemployment rate data are from the International Labour Organization (ILO).6 Our data covers the period from the first quarter of 1999 to the fourth quarter of 2006. Quarterly GDP data from GEM are available for 94 countries. We exclude 31 countries in which quarterly GDP is missing for any quarter during this period (Q1, 1999 to Q4, 2006). Also, we drop 9 countries as data of all other predictors are not available in the pre-intervention period (Q1, 1999 to Q4, 2002). We assume that the treated group includes countries with more than 100 cases of SARS. We drop Canada from the sample, though it was hit by SARS, since only one Canadian city (Toronto) was meaningfully impacted. After all this, our sample includes 4 treated countries (China, Hong Kong, Singapore, and Taiwan) and 49 control countries. Table 1 presents the list of countries used for the cross-country/economy analysis. We use two alternative data sources for China’s provincial GDP growth. The first source is the official statistics, provided by National Bureau of Statistics of China. Quarterly GDP of China’s provinces are not available for long enough to run the synthetic control algorithm. Hence, we use annual GDP series from the National Bureau of Statistics of China. Data for covariates includes investment in fixed assets and household spending. We obtain data from 1993, the first year in which data is available, to 2013. As before, the pre-intervention period is before the year of 2003 (T0 = 2002). The second one is constructed from night-light remote sensing data in 1993–2003 by aggregating data from the Defense Meteorological Program Operational Line-Scan System (DMSP-OLS) of US National Oceanic and Atmospheric Administration (NOAA).7 The sensor in the DMSP-OLS data is not sensitive enough for bright light, so in densely populated urban areas it always registers the highest reading possible throughout the city. Hence, the data is aggregated to the provincial level. Night-time light data has previously been considered as a useful proxy for regional economic activity (Chen & Nordhaus, 2011; Henderson et al., 2012) and have been specifically preferred in the Chinese context (Clark et al., 2020). The SARS epidemic emerged in Guangdong in November 2002. Though the epidemic spread to 26 provinces, there were more than 2,500 and 1,500 cases reported in Beijing and Guangdong, respectively; this is significantly more than 6

Unemployment rate for Indonesia, Brazil, India, and Vietnam are only available annually. We use Version 4 DMSP-OLS Night-time Lights Time Series. The satellite night-light data is available in annual frequency from 1992 to 2013. Visible Infrared Imaging Radiometer Suite (VIIRS), a newer night-light data, is available at higher resolution and frequency and can be fruitfully used for city-level analysis. Yet, the data is only available after 2012.

7

168

I. Noy et al.

Table 1 List of the treated and synthetic countries Treated group

Control group

China

Argentina

Germany

Latvia

Romania

Hong Kong

Australia

Greece

Lithuania

Slovenia

Singapore

Austria

Hungary

Luxembourg

South Africa

Taiwan

Belgium

Iceland

Malaysia

Spain

Brazil

India

Mexico

Sweden

Bulgaria

Indonesia

Morocco

Switzerland

Chile

Ireland

Netherlands

Thailand

Cyprus

Israel

New Zealand

Turkey

Denmark

Italy

Norway

United Kingdom

Ecuador

Japan

Philippines

United States

Estonia

Kazakhstan

Poland

Uruguay

Finland

Korea

Portugal

Vietnam

France Notes Data from GEM includes 94 countries. We drop 31 countries with missing quarterly GDP: Albania, Algeria, Armenia, Bahrain, Belarus, Bosnia and Herzegovina, Colombia, Croatia, Georgia, Ghana, Guatemala, Honduras, Kenya, Kuwait, Malta, Mauritius, Moldova, Mongolia, Mozambique, Nicaragua, Nigeria, Oman, Panama, Peru, Qatar, Saudi Arabia, Sri Lanka, Tajikistan, Tunisia, Ukraine, Uzbekistan. We exclude 9 countries with missing predictors data: Bolivia, Botswana, Czech Republic, Costa Rica, El Salvador, Jordan, Paraguay, Serbia, Slovakia

all the other provinces combined. The number of cases of SARS was highest in Beijing and Guangdong, followed by Shanxi and Hebei. We set Beijing, Guangdong, Hebei, Shanxi as treated units. Table 2 presents the series of 30 provincial-level administrative regions including 4 treated and 26 control units.8

3.2 Methodology We employ the synthetic control methodology (SCM), originally developed in Abadie and Gardeazabal (2003), to identify the impact of SARS. The SCM has previously been used in estimating the impact of extreme sudden-onset shocks in many settings (e.g.,Cavallo et al., 2013; duPont et al., 2015). The approach involves the calculation of a synthetic counterfactual, i.e., observation of the SARS-affected region without the epidemic, by weighting the average of all countries in the donor pool that have not been directly or were marginally affected by the ‘treatment’ of the SARS 2002–2003 epidemic. The methodology also allows for the inclusion of additional regressors, in order to improve the pre-shock fit between the treated observations (pre-treatment) and the control observations (also pre-treatment). As 8

We also drop Chongqing as data on its population, one of the predictors, is missing before 1997.

What Happened After SARS in 2003? The Economic Impacts …

169

Table 2 List of the treated and synthetic provinces of China Treated group

Control group

Beijing

Anhui

Henan

Inner Mongolia

Tianjin

Guangdong

Fujian

Hubei

Ningxia

Xinjiang

Hebei

Gansu

Hunan

Qinghai

Tibet

Shanxi

Guangxi

Jiangsu

Shaanxi

Yunnan

Guizhou

Jiangxi

Shandong

Zhejiang

Hainan

Jilin

Shanghai

Heilongjiang

Liaoning

Sichuan

Notes Data from National Bureau of Statistics of China Tables includes 31 first-level administrative divisions. We drop Chongqing as missing data on population. Municipalities include: Beijing, Shanghai, Tianjin. Autonomous regions include: Guangxi, Inner Mongolia, Ningxia, Xinjiang, Tibet. Others are provinces

such, as long as the pre-shock fit is determined to be a good one, the method allows a comparison between the post-shock counterfactual (that represents the how the treated observations would have fared without the shock) and the actual observations for the affected economies/provinces/cities. This counterfactual-actual gap then represents the impact of the shock. As a test of the statistical significance of this counterfactual-actual gap, Abadie and his co-authors developed a placebo procedure that allows one to compare the gap for the treated unit with gaps estimated for placebo units (units that were actually not affected by the shock). Following closely Abadie et al. (2010), let Yit be the GDP per capita for unit i (economy/province) at time t (quarterly/yearly). We set i = I for the treated economy/province, where 1 ≤ i ≤ N . Then Y I t is the outcome for provinces/countries exposed to the epidemic at time t. In the cross-economy analysis, as the SARS epidemic started in Q1, 2003 (with the very first patients diagnosed in November 2002), we set T0 = Q4, 2002, where 1 ≤ T 0 ≤ T . Our sample includes 32 time periods, in which there are 16 pre- and 16 post-intervention periods. The economic impact of SARS can be observed by the quarterly GDP per capita for economy i at quarter t, where t ≥ T0 + 1. The assumption is that the epidemic had no effect on the outcome before the event, i.e., Y I t = Yit , where 1 ≤ i ≤ N and t ≤ T0 .9 An alternative strategy is to focus on heterogenous impact at lower administrative levels by using disaggregated data from China. As before, the pre-intervention period is before the year of 2003, T0 = 2002. The sample of China provinces includes 21 time periods with 10 pre- and 11 post-intervention periods. In this within-country analysis, we can only use the yearly GDP and nightlights (NTL) data, so there is a trade-off between the spatial and temporal dimensions.

9

See further details in Abadie et al. (2010).

170

I. Noy et al.

To calculate p-value, we implement the approach developed by Galiani and Quistorff (2017) to get a distribution of “in-place” placebo effects. This non-parametric statistical significance test does not impose any distributional assumption on the errors (placebo effects are as large as the estimates for treated unit). Let α it be the estimated effect for a particular post-treatment period of unit i (economy/province) at time t (quarterly/yearly), where { 1≤i ≤N } and t > T0 . The distribution of placebos the two-sided p-value for the treated unit is α IPtL = α it : i /= I . We(|then| compute |) | | | |) ∑ i/= I 1 |||α IPtL |||≥||α I t || (| | | . as: p − value = Pr |α IPtL | ≥ |α I t | = NPL Λ

Λ

Λ

Λ

Λ

Λ

Λ

4 Results 4.1 The Economic Impacts of SARS: Evidence from Asian Countries Figure 1 presents the actual evolution of GDP per capita in the four affected countries, compared against the synthetic counterfactual (what that evolution would have been had the 2002–03 event not occurred). It shows a short-term decrease of GDP per capita in the outbreak period of SARS in Hong Kong, Singapore, and Taiwan. The short-term decline is less apparent in China, where only a few provinces were affected by the virus (not unlike what happened in 2020 with COVID-19). With the synthetic algorithm, we are not able to provide an ideal counterfactual (synthetic) for the post 2003 period, since the economies of the affected countries grew unusually rapidly in this period when compared to other (control) countries, and also when compared to their historical experience. We therefore suspect that the deviation from the estimated counterfactual likely was driven by non-SARS factors. One may argue that SARS helped the affected economies, for example, with the stimuli that some countries adopted post-epidemic. In order to stimulate the economy and housing market after the pandemic, Taiwan, for instance, lowered interest rates substantially after SARS. Nonetheless, this argument is not very persuasive given the long-run increases in GDP growth in the treated countries in the post 2003 period. In Fig. 2, we present the placebo results. All treated countries fail to pass the placebo test, as the gap line (the gap between the actual and the counterfactual) is clearly on the upper bound from the distribution of placebo gaps. The placebos for many control units also show a dip, indicating the violation of stable unit treatment value assumption (SUTVA) possibly due to spill-over effects of the pandemic.10 For Hong Kong and Taiwan, we only find a short-time negative effect in the

10

The assumption requires the observed outcome of one particular unit to be independent from other units.

What Happened After SARS in 2003? The Economic Impacts …

171

Fig. 1 Synthetic analysis for quarterly GDP per capita: China, Hong Kong, Singapore, Taiwan. Notes The dashed line indicates the SARS outbreak. The graphs compare the logarithm of quarterly GDP per capita in a treated economy with the synthetic counterfactual (of the same economy without SARS). The overall period includes 32 quarters. The pre-intervention period is Q1-1999 to Q4-2002, and post-intervention period is Q1-2003 to Q4-2006. We stop the post-intervention period in Q4-2006, as we aim to differentiate the effect of the global financial crisis of 2007–2008

SARS outbreak (Q2, 2003).11 Taken together, the results indicate that indeed these economies bounced back quickly after the epidemic, and that after a fairly modest decline in economic activity in the immediate aftermath of the pandemic. One explanation for the observed rapid take-off of the affected countries in the early 2000s, is the emergence of China as a dominant trading partner for these countries, after China had joined the World Trade Organization (WTO) in 2001. The other three economies were all heavily reliant on China’s trade and were (and still are) to some extent entrepôt economies linked to China’s trade. In principle, what the synthetic algorithm captures is the impact of the 2002–2003 period. These simultaneous events pose a challenge to extract the pure economic impact of SARS.

11 Table A1 in the appendix shows the estimated effects of SARS on quarterly GDP per capita of China, Hong Kong, Singapore, Taiwan for each post-treatment period. For Hong Kong and Taiwan, the point estimates in Q2, 2003 are negative but not statistically significant.

172

I. Noy et al.

Fig. 2 Placebo tests of synthetic analysis for quarterly GDP per capita: China, Hong Kong, Singapore, Taiwan. Notes The dashed line indicates the SARS outbreak. The graphs show the difference between the logarithm of quarterly GDP per capita of a treated economy and its synthetic counterfactual, the dark line. The grey lines are the difference of the logarithm of quarterly GDP per capita of each economy in the donor pool and its counterfactual. The synthetic control of each economy is a weighted average of all other countries excluding the treated economy: China, Hong Kong, Singapore, Taiwan

4.2 The Economic Impacts of SARS: Evidence from Chinese Provinces For the synthetic control analysis for GDP per capita in Beijing, Guangdong, Hebei, and Shanxi, the predictors include household spending and investment per capita. The goodness of fit over the pre-intervention period and the balance for all predictors indicate a plausible pool of control units (though, in Appendix Figure A1, even for the pre-intervention period, the synthetic is not able to replicate the trend of Beijing. The placebo tests in Appendix Figure A2 indicate there is little evidence that annual per capita income for Beijing and Guangdong was affected by the epidemic.) The lack of clear results in the analysis for GDP per capita is puzzling, but it may be attributed to measurement errors in macroeconomic data. Indeed Clark et al. (2020) questions the quality of China’s official regional GDP statistics. Using nighttime lights to compute the optimal weights for a battery of economic activities, Clark et al. (2020) argue that China’s actual GDP growth may be higher than in official reports. Given the absence of satisfactory macroeconomic data, we use night-time light data as an alternative proxy for economic activity. We identify the economic impacts by examining changes in night light around the SARS period in 2003. We run the synthetic control algorithm with the average per-province annual normalized

What Happened After SARS in 2003? The Economic Impacts …

173

night-light data in Beijing, Guangdong, Hebei, Shanxi. Predictors include: population density, household spending, investment per capita, and the one-year lag of GDP per capita. The results, in Fig. 3, do suggest a noticeable decline in economic activity during the SARS period in the two affected areas that are most associated with SARS, Beijing and Guangdong. As described earlier, the actual provincial per capita GDP data, which might be perceived as less reliable, does not corroborate that. The normalized NTL per capita declines by 0.033 and 0.004 points (20 and 30%) in Beijing and Guangdong, respectively, against the counterfactual rapid growth that the synthetic model predicts. We also ran the synthetic control algorithm for Shanxi and Hebei, two neighbouring provinces that were also affected by the epidemic. Shanxi and Hebei also show a drop in night lights, though one that is significantly smaller. This differentiation between Beijing and Guangdong on the one side, and Shanxi and Hebei on the other, is likely because the case load of SARS in those to later regions was much smaller. This is, clearly, just a conjecture that the smaller caseload is the explanation for the milder reaction, especially since even in the higher SARS exposed places, the per capita exposure of people to the virus was small, and the likely policy reaction (lockdowns, shutdowns, etc.) was therefore necessarily milder as well. Figure 4 provides the placebo effects for all other provinces. The validity of our results will be in doubt should we find many placebos (SARS-unaffected) provinces with similarly negative effects. But the results in Fig. 4 seem to confirm our findings of a negative impact of SARS on Guangdong and Beijing. In Table 3, we present the point estimates of effect on NTL and the two-sided p-value for each post-treatment period. The point estimates for Beijing are statistically different from the placebo in

Fig. 3 Synthetic analysis for annual NTL per capita: Beijing, Guangdong, Hebei, Shanxi. Notes The dashed line indicates the SARS outbreak. The graphs compare the annual normalized NTL per capita in a treated province with the synthetic counterfactual (of the same province without SARS). The overall period includes 21 quarters. The pre-intervention period is 1993 to 2002, and post-intervention period is 2003 to 2013

174

I. Noy et al.

Fig. 4 Placebo tests of synthetic analysis for annual NTL per capita: Beijing, Guangdong, Hebei, Shanxi. Notes The dashed line indicates the SARS outbreak. The graphs show the difference between annual normalized NTL per capita of a treated province and its synthetic counterfactual, the dark line. The grey lines are the difference of annual normalized NTL per capita of each province in the donor pool and its counterfactual. The synthetic control of each province is a weighted average of all other province excluding the treated province: Beijing, Guangdong, Hebei, Shanxi

all post-treatment periods. This provides further support for our finding of a negative impact of the epidemic in the most heavily affected Chinese city.

5 Robustness Checks For a traditional SCM, the critical assumption is that the synthetic algorithm can replicate the trajectories of the treated units in the pre-intervention period. There is a growing literature that develops methods to improve the ability or accuracy of the SCM algorithm to replicate accurately pre-intervention data given the multiple possibilities for how to do so (Ben-Michael et al., 2021; Robbins et al., 2017), or alternatively relax the assumptions required for the original SCM (Doudchenko & Imbens, 2016; Powell, 2018). We use the Augmented Synthetic Control Method developed by Ben-Michael et al., (2021); it is a derivative of SCM which seeks to improve the pre-intervention fit of the synthetic counterfactual to the factual time series of the treated units and then implement a bias correction. As discussed in Ben-Michael et al. (2021), the augmented estimator is a weighting estimator to adjust the SCM weights and then de-bias the original SCM estimate. The bias in the pre-intervention outcomes between the treated units and the synthetic estimated by an outcome model, m, is bias m . ASCM adds a bias term to the original Λ

What Happened After SARS in 2003? The Economic Impacts …

175

Table 3 Synthetic control estimates of the effects of SARS on normalized annual NTL per capita: Beijing, Guangdong, Hebei, Shanxi

Overall Effect

(1)

(2)

(3)

(4)

Beijing

Guangdong

Hebei

Shanxi

−0.033**

−0.004

−0.001

−0.002

(0.029)

(0.196)

(0.436)

(0.342)

−0.011***

−0.001

−0.002

−0.003

(0.000)

(0.360)

(0.240)

(0.160)

2004

−0.017**

−0.002

−0.001

0.000

(0.040)

(0.200)

(0.400)

(0.720)

2005

−0.019**

−0.003

−0.001

−0.001

(0.040)

(0.120)

(0.320)

(0.240)

2003

−0.025**

−0.004

−0.001

−0.001

(0.040)

(0.200)

(0.320)

(0.320)

2007

−0.037**

−0.004

−0.001

−0.003

(0.040)

(0.160)

(0.400)

(0.160)

2008

−0.040***

−0.004

−0.001

−0.002

(0.000)

(0.200)

(0.400)

(0.200)

2006

−0.037**

−0.003

0.001

−0.002

(0.040)

(0.160)

(0.480)

(0.280)

2010

−0.048**

−0.007

−0.001

−0.003

(0.040)

(0.160)

(0.760)

(0.360)

2011

−0.043***

−0.005

−0.001

−0.003

(0.000)

(0.200)

(0.520)

(0.360)

2012

−0.040*

−0.006

−0.002

−0.001

(0.080)

(0.200)

(0.520)

(0.600)

2009

2013

−0.047***

−0.006

−0.002

−0.002

(0.000)

(0.200)

(0.440)

(0.360)

Notes P-values are in parentheses. ***p < 0.01, **p < 0.05, *p < 0.10 Λ

Λ Λ

SCM estimator: Yit ASC M = Yit SC M + bias m .12 Using a ridge-regularized linear regression as the outcome model, ASCM replicates the pre-intervention trajectory more closely. In this framework, the treated units may be outside the convex hull of control units. Thus, the Ridge ASCM allows non-negative weights to improve the SCM pre-intervention fit, but penalizing extrapolation from the convex hull. When the original SCM provides a good pre-intervention fit, Ridge ASCM and the original SCM approach the same weights. Applying the ASCM analysis, as described above, we find a short-term drop in GDP per capita in China and Hong Kong (Fig. 5). This result is similar to what 12

See specification details in Ben-Michael et al. (2021).

176

I. Noy et al.

we described above. Interestingly, Hong Kong appears to experience a longer-term decrease of 0.5% in GDP per capita on average. However, confidence intervals are too wide to reject a null effect. In the China provincial analysis, Fig. 6 shows negative effects on NTL per capita in Beijing and Guangdong. In Fig. 6, the synthetic closely replicate the mean trajectories in Beijing in the pre-treatment period. In the post-intervention period, the observed difference between annual normalized NTL per capita of Beijing and its synthetic counterfactual (dark line) is constantly less than zero. As consistent with previous results, on average, the post-treatment treated and synthetic units imply a reduction in NTL per capita for Beijing of around 0.031 points (20%). Guangdong appears to experience a sharp decrease in NTL per capita on average followed by a recovery after 2006.

Fig. 5 Difference in quarterly GDP per capita using ASCM: China, Hong Kong, Singapore, Taiwan. Notes The vertical dashed line indicates the SARS outbreak. The graphs show the difference between the logarithm of quarterly GDP per capita of a treated economy and its synthetic counterfactual, the dark line. The grey area represents the 95% confidence interval. The synthetic control of each economy is a weighted average of all other countries excluding the treated economy: China, Hong Kong, Singapore, Taiwan

What Happened After SARS in 2003? The Economic Impacts …

177

Fig. 6 Difference in annual NTL per capita using ASCM: Beijing, Guangdong, Hebei, Shanxi. Notes The vertical dashed line indicates the SARS outbreak. The graphs show the difference between annual normalized NTL per capita of a treated province and its synthetic counterfactual, the dark line. The grey area represents the 95% confidence interval. The synthetic control of each province is a weighted average of all other provinces excluding the treated province: Beijing, Guangdong, Hebei, Shanxi

6 Conclusions For insights about longer-term economic prospects in a post COVID-19 world, we turn to the experience from previous pandemic outbreaks. We revisit the economic impacts of SARS in 2003. We quantify the effect of the SARS epidemic on the economic growth in Asian countries and Chinese provincial-level administrative regions by measuring the economic growth with GDP and night-time lights. In this study, we use the synthetic control methods, to create pre- and post-intervention comparisons between observed and counterfactual outcomes. There is a short-term negative effect of the epidemic on GDP per capita in Q2, 2003 in the heavily affected East Asian economies (Hong Kong, Taiwan, Singapore, and China). Beyond that, in the aftermath of the epidemic’s peak, we do not find any observable negative impact on economic activity at the national level even for the first post-pandemic quarter. We conjecture that any plausible mild negative impact would have been masked by the economic boom that followed China becoming a full member of the WTO. We have no evidence to support this view, however, but the sustained increase in economic

178

I. Noy et al.

activity that lasts for more than a decade is unlikely to have been originated by this very brief interlude of a spreading and unknown epidemic. Using night-light data, we find some evidence of a more persistent negative effect of SARS in Beijing and Guangdong, the two most heavily affected Chinese provinces. Even there, however, it is challenging to determine the depth of this effect because the data series includes a time of dramatic change in the Chinese economy including the WTO accession in 2001 and eventually the global financial crisis in 2008. Another limitation of the within-country analysis is that while we observe a negative effect in Beijing and Guangdong, there are potential spill-over effects on other provinces. The SCM may thus provide a lower bound of the adverse effect. These estimations of the economic impacts of the SARS epidemic indicate that, fortunately, the national economies proved to be quite resilient to this temporary but large shock. This also suggests that, while country-level macroeconomic impact assessment will be useful for the central government to identify overall impacts, such macro-analysis may mask some local-level impacts if localities were affected differently. Indeed, as has been found in some research papers investigating the impact of natural hazard disasters, there is little evidence that these events pose a significant macroeconomic risk (Cavallo et al., 2013). This is especially true when disaster risk reduction policies are managed well and the risk is reduced rapidly. This was clearly not the case for COVID-19, so none of this implies much to our understanding of the current post-pandemic. It is also worthwhile to acknowledge that while, in the aggregate, these lack of aggregate impact is welcome, it may mask distributional changes or a change in the composition of income between groups. In short, it very well may be the case that some sectors gain, and some sectors (and people) lose from this pandemic. Therefore, it is imperative to conduct disaggregated local assessments of impacts of a pandemic in addition to any macroeconomic analysis. While we disaggregated data to the provincial level, data can be disaggregated to even much smaller spatial units than that. This further disaggregation can shed light on some of the likely mechanisms that led to these findings. Indeed, more detailed microeconomic data will shed more light on the possible mechanisms behind any observed decline as well as its duration. Such an analysis is part of our future research agenda.

What Happened After SARS in 2003? The Economic Impacts …

179

Appendix

Fig. A1 Synthetic Analysis for Annual GDP per capita: Beijing, Guangdong, Hebei, Shanxi. Notes: The dashed line indicates the SARS outbreak. The graphs compare the logarithm of annual GDP per capita in a treated province with the synthetic counterfactual (of the same province without SARS). The overall period includes 21 quarters. The pre-intervention period is 1993 to 2002, and the post-intervention period is 2003 to 2013

180

I. Noy et al.

Fig. A2 Placebo tests of Synthetic Analysis for Annual GDP per capita: Beijing, Guangdong, Hebei, Shanxi. Notes The dashed line indicates the SARS outbreak. The graphs show the difference between the logarithm of annual GDP per capita of a treated province and its synthetic counterfactual, the dark line. The grey lines are the difference of the logarithm of annual GDP per capita of each province in the donor pool and its counterfactual. The synthetic control of each province is a weighted average of all other provinces excluding the treated province: Beijing, Guangdong, Hebei, Shanxi

What Happened After SARS in 2003? The Economic Impacts …

181

Fig. A3 Difference in Annual GDP per capita using ASCM: Beijing, Guangdong, Hebei, Shanxi. Notes The vertical dashed line indicates the SARS outbreak. The graphs show the difference between the logarithm of annual GDP per capita of a treated province and its synthetic counterfactual, the dark line. The grey area represents the 95% confidence interval. The synthetic control of each province is a weighted average of all other provinces excluding the treated province: Beijing, Guangdong, Hebei, Shanxi

Table A1 Synthetic control estimates of the effects of SARS on log quarterly GDP per capita: China, Hong Kong, Singapore, Taiwan

Overall effect Q1 2003 Q2 2003 Q3 2003

(1)

(2)

(3)

(4)

China

Hong Kong

Singapore

Taiwan

0.183**

0.071

0.102

0.051

(0.016)

(0.271)

(0.144)

(0.379)

0.129**

0.020

0.033

0.030

(0.042)

(0.479)

(0.375)

(0.417)

0.113**

−0.018

0.011

−0.015

(0.042)

(0.583)

(0.667)

(0.604)

0.128**

0.023

0.055

0.017

(0.021)

(0.521)

(0.250)

(0.604) (continued)

182

I. Noy et al.

Table A1 (continued)

Q4 2003 Q1 2004 Q2 2004 Q3 2004 Q4 2004 Q1 2005 Q2 2005 Q3 2005 Q4 2005 Q1 2006 Q2 2006 Q3 2006 Q4 2006

(1)

(2)

(3)

(4)

China

Hong Kong

Singapore

Taiwan

0.113*

0.018

0.073

0.042

(0.063)

(0.625)

(0.208)

(0.333)

0.159**

0.048

0.089

0.052

(0.021)

(0.292)

(0.104)

(0.271)

0.155**

0.049

0.091*

0.042

(0.021)

(0.292)

(0.083)

(0.333)

0.160**

0.044

0.085

0.034

(0.021)

(0.333)

(0.146)

(0.479)

0.164**

0.057

0.103*

0.039

(0.021)

(0.313)

(0.063)

(0.438)

0.192***

0.076

0.093

0.055

(0.000)

(0.229)

(0.125)

(0.354)

0.187***

0.098

0.111*

0.065

(0.000)

(0.146)

(0.063)

(0.292)

0.206***

0.099

0.125*

0.046

(0.000)

(0.125)

(0.063)

(0.438)

0.215***

0.100

0.150**

0.080

(0.000)

(0.146)

(0.042)

(0.250)

0.236***

0.129*

0.141**

0.081

(0.000)

(0.063)

(0.042)

(0.271)

0.255***

0.133**

0.147**

0.083

(0.000)

(0.042)

(0.021)

(0.292)

0.253***

0.128*

0.150**

0.071

(0.000)

(0.063)

(0.042)

(0.396)

0.265***

0.133*

0.176**

0.093

(0.000)

(0.083)

(0.021)

(0.292)

Notes P-values are in parentheses. ***p < 0.01, **p < 0.05, *p < 0.10

Table A2 Synthetic control estimates of the effects of SARS on log annual GDP per capita: Beijing, Guangdong, Hebei, Shanxi

Overall effect

(1)

(2)

(3)

(4)

Beijing

Guangdong

Hebei

Shanxi

0.087

−0.085

−0.011

0.074

(0.564)

(0.606)

(0.848)

(0.606) (continued)

What Happened After SARS in 2003? The Economic Impacts …

183

Table A2 (continued)

2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013

(1)

(2)

(3)

(4)

Beijing

Guangdong

Hebei

Shanxi

0.149

−0.023

0.034

0.117

(0.125)

(0.750)

(0.667)

(0.250)

0.153

0.007

0.043

0.138

(0.125)

(0.917)

(0.667)

(0.167)

0.112

−0.010

0.013

0.078

(0.375)

(0.958)

(0.917)

(0.542)

0.124

0.006

−0.011

0.065

(0.375)

(0.958)

(0.917)

(0.625)

0.147

0.004

−0.003

0.090

(0.250)

(0.958)

(0.958)

(0.542)

0.076

−0.062

−0.016

0.109

(0.667)

(0.708)

(0.917)

(0.542)

0.062

−0.102

−0.005

0.053

(0.750)

(0.458)

(1.000)

(0.750)

0.037

−0.147

−0.037

0.044

(0.833)

(0.292)

(0.833)

(0.792)

0.022

−0.178

−0.019

0.076

(0.917)

(0.250)

(0.917)

(0.583)

0.029

−0.216

−0.046

0.048

(0.917)

(0.208)

(0.875)

(0.875)

0.041

−0.213

−0.078

−0.001

(0.875)

(0.208)

(0.667)

(1.000)

Notes P-values are in parentheses. ***p < 0.01, **p < 0.05, *p < 0.10 Table A3 Predictors of synthetic analysis for China, Hong Kong, Singapore, Taiwan China

Synthetic

Imports (% of GDP)

0.102

0.145

Exports (% of GDP)

0.114

0.156

Unemployment rate

3.825

5.797

Hong Kong

Synthetic

Imports (% of GDP)

1.322

0.440

Exports (% of GDP)

1.264

0.495

Unemployment rate

5.879

6.062

Singapore

Synthetic

0.913

0.424

Imports (% of GDP)

(continued)

184

I. Noy et al.

Table A3 (continued) China

Synthetic

Exports (% of GDP)

0.953

0.462

Unemployment rate

3.925

6.419

Taiwan

Synthetic

Imports (% of GDP)

0.405

0.405

Exports (% of GDP)

0.449

0.457

Unemployment rate

3.915

5.005

Table A4 Predictors of synthetic analysis for Beijing, Guangdong, Hebei, Shanxi (1)

(2)

Normalized NTL per capita

Log annual GDP per capita

Beijing

Synthetic

Beijing

Synthetic

Household spending

823.232

687.717

823.232

814.648

Investment per capita

956.397

766.78

1008.266

1008.135

Population density

762.460

1106.989

Log GDP per capita (t-1)

7.583

7.392

Guangdong

Synthetic

Guangdong

Synthetic

Household spending

553.627

431.955

553.627

533.942

Investment per capita

421.760

420.909

438.319

495.559

Population density

411.489

400.554

Log GDP per capita (t-1)

7.117

6.946

Hebei

Synthetic

Hebei

Synthetic

Household spending

258.593

280.597

258.593

282.922

Investment per capita

244.809

244.810

259.586

259.097

Population density

345.406

381.182

Log GDP per capita (t-1)

6.505

6.490

Shanxi

Synthetic

Shanxi

Synthetic

Household spending

227.511

269.111

227.511

233.423

Investment per capita

168.945

175.902

175.321

192.225

Population density

200.453

201.781

Log GDP per capita (t-1)

6.259

6.257

What Happened After SARS in 2003? The Economic Impacts … Table A5 Weights of synthetic analysis for China, Hong Kong, Singapore, Taiwan Weights in synthetic China India

0.147

Kazakhstan

0.434

Korea

0.029

Vietnam

0.390 Weights in Synthetic Hong Kong

Canada

0.628

Malaysia

0.357

Norway

0.015 Weights in Synthetic Singapore

Denmark

0.011

Israel

0.141

Malaysia

0.262

Switzerland

0.585 Weights in Synthetic Taiwan

Malaysia

0.462

Slovenia

0.451

Sweden

0.084

Switzerland

0.003

Table A6 Weights of synthetic analysis for Beijing, Guangdong, Hebei, Shanxi Weights in synthetic Beijing Liaoning

0.059

Shanghai

0.503

Tianjin

0.438 Weights in synthetic Guangdong

Liaoning

0.641

Shanghai

0.074

Tianjin

0.285 Weights in synthetic Hebei

Henan

0.277

Shandong

0.383

Xinjiang

0.340 Weights in synthetic Shanxi

Henan

0.762

Xinjiang

0.238

185

186

I. Noy et al.

References Abadie, A. (2021). Using synthetic controls: Feasibility, data requirements, and methodological aspects. Journal of Economic Literature, 59(2), 391–425. Abadie, A., & Gardeazabal, J. (2003). The economic costs of conflict: A case study of the Basque Country. American Economic Review, 93(1), 113–132. Abadie, A., Diamond, A., & Hainmueller, J. (2010). Synthetic control methods for comparative case studies: Estimating the effect of California’s tobacco control program. Journal of the American Statistical Association, 105(490), 493–505. 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. Altig, D., Baker, S., Barrero, J. M., Bloom, N., Bunn, P., Chen, S., Davis, S. J., Leather, J., Meyer, B., & Mihaylov, E. (2020). Economic uncertainty before and during the COVID-19 pandemic. Journal of Public Economics, 191, 104274. APEC, Asia-Pacific Economic Cooperation. (2004). Commentaries & Analyses—The Economic Impact of SARS on the APEC Region. Asia-Pacific Biotech News, 8 (01), 33–38. Baker, S. R, Bloom, N., Davis, S. J., & Terry, S. J. (2020). Covid-induced economic uncertainty. National Bureau of Economic Research. Baldwin, R. E., & Weisbrod, B. A. (1974). Disease and labor productivity. Economic Development and Cultural Change, 22(3), 414–435. Barnett, T., Whiteside, A., Khodakevich, L., Kruglov, Y., & Steshenko, V. (2000). The HIV/AIDS epidemic in Ukraine: Its potential social and economic impact. Social Science & Medicine, 51(9), 1387–1403. Barro, R. J., Ursúa, J. F., & Weng, J. (2020). The coronavirus and the great influenza pandemic: Lessons from the “spanish flu” for the coronavirus’s potential effects on mortality and economic activity. National Bureau of Economic Research. Beach, B., Clay, K., & Saavedra, M. H. (2022). The 1918 influenza pandemic and its lessons for COVID-19. Journal of Economic Literature., 60(1), 41–84. Beach, B., Ferrie, J. P., & Saavedra, M. H. (2018). Fetal shock or selection? The 1918 influenza pandemic and human capital development. National Bureau of Economic Research. Ben-Michael, E., Feller, A., & Rothstein, J. (2021). The augmented synthetic control method. Journal of the American Statistical Association, 116(536), 1789–1803. Bloom, E., De Wit, V., & Carangal-San Jose, M. J. (2005). Potential economic impact of an avian flu pandemic on Asia. ERD Policy Brief – Asian Development Bank. Boberg-Fazlic, N., Ivets, M., Karlsson, M., & Nilsson, T. (2021). Disease and fertility: Evidence from the 1918 influenza pandemic in Sweden. Economics and Human Biology, 43, 101020. Brahmbhatt, M., & Dutta, A. (2008). On SARS type economic effects during infectious disease outbreaks. World Bank. Brodeur, A., Gray, D. M., Islam, A., & Bhuiyan, S. (2020). A literature review of the economics of COVID-19. Campante, F. R., Depetris-Chauvin, E., & Durante, R. (2020). The virus of fear: The political impact of ebola in the U.S. National Bureau of Economic Research. Castañeda-Orjuela, C., Díaz, H., Alvis-Guzman, N., Olarte, A., Rodriguez, H., Camargo, G., & De la Hoz-Restrepo, F. (2012). "Burden of disease and economic impact of dengue and severe dengue in Colombia, 2011. Value in Health Regional Issues, 1(2), 123–128. Cavallo, E., Galiani, S., Noy, I., & Pantano, J. (2013). Catastrophic natural disasters and economic growth. Review of Economics and Statistics., 95(5), 1549–1561. Chen, X., & Nordhaus, W. D. (2011). Using luminosity data as a proxy for economic statistics. Proceedings of the National Academy of Sciences., 108(21), 8589–8594. Chou, J., Kuo, N.-F., & Peng, S.-L. (2004). Potential impacts of the SARS outbreak on Taiwan’s economy. Asian Economic Papers., 3(1), 84–99. Clark, H., Pinkovskiy, M., & Sala-i-Martin, X. (2020). China’s GDP growth may be understated. China Economic Review, 62(C).

What Happened After SARS in 2003? The Economic Impacts …

187

Dixon, S., McDonald, S., & Roberts, J. (2002). The impact of HIV and AIDS on Africa’s economic development. British Medical Journal, 324(7331), 232–234. Doan, N., & Noy, I. (2021). A comprehensive measure of lifeyears lost due to COVID-19 in 2020: A comparison across countries and with past disasters. Global Policy, 12(4), 553–561. Doudchenko, N., & Imbens, G. W. (2016). Balancing, regression, difference-in-differences and synthetic control methods: A synthesis. National Bureau of Economic Research. duPont, W., Noy, I., Okuyama, Y., & Sawada, Y. (2015). The long-run socio-economic consequences of a large disaster: The 1995 earthquake in Kobe. PLoS ONE, 10(10), e0138714. Galiani, S., & Quistorff, B. (2017). The synth_runner package: Utilities to automate synthetic control estimation using synth. The Stata Journal., 17(4), 834–849. Gillies, P., Tolley, K., & Wolstenholme, J. (1996). Is AIDS a disease of poverty? AIDS Care, 8(3), 351–364. Guimbeau, A., Menon, N., & Musacchio, A. (2020). The brazilian bombshell? the long-term impact of the 1918 influenza pandemic the south american way. National Bureau of Economic Research. Hailu, G. (2020). Economic thoughts on COVID-19 for Canadian food processors. Canadian Journal of Agricultural Economics., 68(2), 163–169. Helgertz, J., & Bengtsson, T. (2019). The long-lasting influenza: The impact of fetal stress during the 1918 influenza pandemic on socioeconomic attainment and health in Sweden, 1968–2012. Demography, 56(4), 1389–1425. Henderson, J. V., Storeygard, A., & Weil, D. N. (2012). Measuring economic growth from outer space. American Economic Review., 102(2), 994–1028. Jinjarak, Y., Noy, I., & Ta, Q. (2022). Pandemics and economic growth: Evidence from the 1968 H3N2 Influenza. Economics of Disasters and Climate Change, 6(1), 73–93. Jordà, Ò., Singh, S. R., & Taylor, A. M. (2022). Longer-run economic consequences of pandemics. Review of Economics and Statistics, 104(1), 166–175. Kabajulizi, J., & Ncube, M. (2017). Financing HIV/AIDS responses in Africa: Impact evidence from Uganda. Journal of Policy Modeling, 39(5), 843–860. Karlsson, M., Nilsson, T., & Pichler, S. (2014). The impact of the 1918 Spanish flu epidemic on economic performance in Sweden: An investigation into the consequences of an extraordinary mortality shock. Journal of Health Economics, 36, 1–19. Katafuchi, Y., Kurita, K., & Managi, S. (2021). COVID-19 with stigma: Theory and evidence from mobility data. Economics of Disasters and Climate Change, 5(1), 71–95. Kostova, D., Cassell, C. H., Redd, J. T., Williams, D. E., Singh, T., Martel, L. D., & Bunnell, R. E. (2019). Long-distance effects of epidemics: Assessing the link between the 2014 West Africa Ebola outbreak and US exports and employment. Health Economics, 28(11), 1248–1261. Lee, J.-W., & McKibbin, W. J. (2004). Globalization and disease: The case of SARS. Asian Economic Papers, 3(1), 113–131. Leung, G. M., Quah, S., Ho, L.-M., Ho, S.-Y., Hedley, A. J., Lee, H.-P., & Lam, T.-H. (2004). A tale of two cities: Community psychobehavioral surveillance and related impact on outbreak control in Hong Kong and Singapore during the severe acute respiratory syndrome epidemic. Infection Control and Hospital Epidemiology, 25(12), 1033–1041. Liu, J.-T., Hammitt, J. K., Wang, J.-D., & Tsou, M.-W. (2005). Valuation of the risk of SARS in Taiwan. Health Economics, 14(1), 83–91. Maffioli, E. M. (2021). The political economy of health epidemics: Evidence from the Ebola outbreak. Journal of Development Economics, 151, 102651. Noy, I., & duPont, W. (2018). The long-term consequences of disasters: What do we know, and what we still don’t. International Review of Environmental and Resource Economics, 12(4), 325–354. Noy, I., & Shields, S. (2019). The 2003 severe acute respiratory syndrome epidemic: A retroactive examination of economic costs. Asian Development Bank Economics Working Paper 591. Noy, I., & Uher, T. (2021). Economic consequences of pre-COVID-19 epidemics: A literature review. Handbook on the Economics of Disasters, forthcoming. Noy, I., Okubo, T., & Strobl, E. (2020). The Japanese textile sector and the influenza pandemic of 1918–1920. CESifo Working Paper 8651.

188

I. Noy et al.

Powell, D. (2018). Imperfect synthetic controls: Did the Massachusetts health care reform save lives? SSRN 3192710. Robbins, M. W., Saunders, J., & Kilmer, B. (2017). A framework for synthetic control methods with high-dimensional, micro-level data: Evaluating a neighborhood-specific crime intervention. Journal of the American Statistical Association., 112(517), 109–126. Siu, A., & Wong, Y. C. (2004). Economic impact of SARS: The case of Hong Kong. Asian Economic Papers, 3(1), 62–83. Weisbrod, B. A., Andreano, R. L., Baldwin, R. E., Epstein, E. H., Kelley, A. C., & Helminiak, T. W. (1974). Disease and economic development: The impact of parasitic diseases in St. Lucia. International Journal of Social Economics, 1(1), 111–117.

Industrial Composition, Remote Working and Mobility Changes in Canada and the US During the COVID-19 Pandemic: A SHAP Value Analysis of XGBoost Predictions Mehmet Güney Celbi¸s, Cem Özgüzel, Karima Kourtit, and Peter Nijkamp JEL Classifications C81 · J6 · J40 · J21 · R11

1 Introduction The worldwide corona crisis has exerted profound impacts on the functioning of economies, not only on their productivity but also on the distribution of their wellbeing. In this context, a new type of inequality became manifest as the COVID-19 pandemic broke out: inequality in the exposure to circumstances with a high risk of COVID-19 infection. Various types of government stringency policies have been implemented all over the world, which reduced individual mobility and interaction, and hence lowered the risk of infection for many people (see, for example, Chernozhukov et al., 2021). Despite the presence of such policies, the inequality in infection risk was exacerbated consciously in some cases, as considerably large groups of people opposed reducing their social interactions due to reasons driven by cultural attributes and their low trust in political actors (Barbieri & Bonini, 2021; Bargain & Aminjonov, 2020). On the other hand, a major driver of the disparities in mobility was due to local labour market characteristics, as shown by Gauvin et al. (2021) and Caselli et al. (2022) in the context of the drastic pandemic effects M. G. Celbi¸s (B) Department of Economics, Yeditepe University, Istanbul 34755, Turkey e-mail: [email protected] United Nations University-Maastricht Economic and Social Research Institute on Innovation and Technology (UNU-MERIT), 6211 Maastricht, The Netherlands C. Özgüzel Sorbonne Economics Center, Maison des Sciences Économiques, 106–112 bd de l’Hôpital, 75647 Paris Cedex 13, France K. Kourtit · P. Nijkamp Faculty of Management, Open University, 6419 Heerlen, The Netherlands Centre for European Studies, Alexandru Ioan Cuza University, 700506 Ia¸si, Romania © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. G. Celbi¸s et al. (eds.), Pandemic and the City, Footprints of Regional Science, https://doi.org/10.1007/978-3-031-21983-2_11

189

190

M. G. Celbi¸s et al.

in Italy in early 2020. Through the spatial analysis of Italian provinces, Caselli et al. (2022) have found that locations with higher number of persons working in jobs with fixed-term contracts, and jobs that allow for flexible arrangements, in addition to those with high disease exposure, experienced more cautious rises in mobility upon the relaxation of stringency policies. The unmistakable differences in the infection risks of individuals based on their occupations, which became apparent immediately in the initial stages of the pandemic, have been rapidly substantiated scientifically by a plethora of research output. For instance, Almagro and Orane-Hutchinson (2022) observed that an individual’s occupation was a stronger determinant of early COVID19 transmission in New York City, compared with other individual demographic and socio-economic characteristics. Ascani et al. (2021) highlight how, in Italy during early 2020, the workers had higher mobility in those areas which had more persons working in jobs within sectors defined as “essential." Furthermore, the transition to online education in many parts of the world has resulted in decreased mobility in places with higher shares of students (Casalone et al., 2021). Remote working potential emerged as a prominent feature in relation to the reduction of mobility and the avoidance of circumstances with high infection risk (Barbieri et al., 2021). However, as underlined by Crowley and Doran (2020) and Bonacini et al. (2020), further inequality is embedded in individual’s ability to work remotely in labour markets, and income support measures may mitigate this inequality to a certain extent. Similarly, cities and urban areas benefit from higher opportunities for remote work as they have a higher share of occupations that are amenable to remote work (Ozguzel et al., 2020). The present study focuses on the mobility changes in workplaces in USA states and Canadian provinces during 2020 by applying a sequential machine learning algorithm using explanatory features from pre-pandemic periods. As a result, the study identifies how a large set of regional socio-economic attributes are related to workplace mobility outcomes. Upon the application of the machine learning algorithm, we further explore our findings using interpretable machine learning approaches, particularly through the use of Shapley Value-based procedures. The remainder of this chapter is organised as follows. Section 2 presents the variables included in our analysis and discusses their relevance to workplace mobility. Section 3 outlines the empirical implementation of our sequential machine learning model. The results of the model are discussed in several subsections in Sect. 4. Finally, there is a concluding discussion in Sect. 5.

2 Variables, Data Sources and Theoretical Relevance This chapter uses a novel source of subnational level data produced in real-time which is the Google-maps lifestyle data for NUTS2 regions in Canada and the United States, disaggregated into distinct location types to measure mobility changes (such as workplaces). The empirical analysis in this study also matches the information from NUTS2 units with regional-level data, including regional industrial structure, local

Industrial Composition, Remote Working and Mobility Changes …

191

population characteristics and the intensity of teleworkability, in order to evaluate their effect on changes in mobility during 2020. The variables, along with their abbreviations as presented in our analytical output, are described below.

2.1 Google Mobility Data We measured the changes in the mobility patterns in NUTS2 regions of Canada and the United States during 2020 through the usage of data with daily frequency obtained from the Global Mobility Report provided by Google (Google LLC, 2021).1 The data concern information on individual mobility to various places, such as grocery stores, parks, retail, transit stations, or workplaces at the regional level. For this analysis, we focus on people’s mobility related to places of work (‘Workplaces’). Following the approach taken in OECD (2021) publication on remote working adoption, we use these data to compute the percentage change in the mobility of individuals within a region with respect to the median value for the corresponding day of the week in the period from 3 January 2020 to 3 February 2020 (i.e. the baseline period). After controlling for sample selection biases and seasonality effects, values close to zero (within ± 10%) imply ‘return to normality’, whereas large negative and positive values (outside ± 10%) imply ‘below normality’ and ‘above normality’ respectively (OECD, 2021). A descriptive look at the spatial mobility patterns is shown in Fig. 1, where regions with larger colours experienced larger drops in mobility.

2.2 Regional Characteristics Throughout this paper, we use several variables drawn from the recent COVID-19 literature. Including these variables is essential for measuring the contribution of local factors which affect local mobility. These factors determine the resilience and vulnerability of regions to COVID-19, and the mobility needs to ensure the continuation of economic activity. This section presents these variables and explains their potential effects on local mobility and the underlying data used in their construction. Local industrial structure: Workplace mobility patterns can be affected by industries. Industries may not operate due to disruptions in their activities as a result of border closures, interruptions to international trade, or demand shocks. Regions which may have a high share of industries affected by the demand shock may see a stronger drop in the workplace-related mobility. Moreover, some industries may be more suitable for remote work than others. Regions which have a higher share of industries which can continue functioning remotely would see less work-related mobility. The vulnerability of industries to the demand shock or their capacity to transition to telework would affect the intensity of workrelated mobility. To account for such differences, we measure the composition of 1

Google LLC (2021).

192

M. G. Celbi¸s et al.

local industries, using regional employment at the place of work in 2019 for ten main industry categories according to ISIC rev 4. The data is collected from the OECD regional database. The abbreviations used for the industries are as follows InfoComm, SciTechAdm, FinIns, AgriForFish, IndEnerg, Estate, TradeFoodAcc, Construct, PubEduHealth and OtherServ. Unemployment rate: Workplace-related mobility only concerns those who work. Regions that have a high unemployment rate would expect less work-related mobility. We collected unemployment rates in 2019 from the OECD regional database. Total population in the region: The risk of spreading the virus and the mobility needs might be higher in more populous regions (Stier et al., 2020). We account for these differences by including a measure which captures the total population in the region in 2019 obtained from the OECD regional database. Doctors’s rate: Medical resources such as hospital beds and number of doctors per inhabitants are crucial for managing health crises but differ substantially across regions. We measure the strength of the regional health system capacity through the number of doctors relative to the total population. The measure captures the number of doctors per 1,000 inhabitants in 2018 obtained from the OECD Regional Database. Elderly population: Larger shares of the elderly population (e.g. aged 65 or above) are particularly vulnerable to the virus (Kashnitsky & Aburto 2020; Ramírez et al., 2022). Regions which may have a larger elderly population are more vulnerable to the health consequences of the virus. Moreover, individuals who live with older family members or are responsible for their care might prefer working from home to avoid contamination. Therefore, we measure the share of elderly population in the region in 2019 obtained from the OECD Regional Database. Excess Mortality: Some regions have suffered a higher number of deaths due to the pandemic than others. People living in regions with a higher number of deaths would be more likely to decrease their mobility and remain at home. Given the difficulties in measuring human deaths associated with the COVID-19 pandemic across and within countries, the concept of excess mortality has been proposed. ‘Excess mortality’ is defined as the percentage increase in the cumulative number of deaths (from all causes) between the period of February to June 2020 with respect to the average number of deaths in the same period in 2018 and 2019. The data come from OECD (2020). Education: Socio-economic characteristics such as educational attainment are shown to be factors that influence the spread of COVID-19 and related deaths (Brandily et al., 2021; Ramírez et al., 2022). People with higher socio-economic status tend to have higher awareness about COVID-19 (Ramírez et al., 2022; Zhong et al., 2020) and understand, trust, and be able to follow expert advice to cope with the pandemic (Ajzenman et al., 2020; Ramírez et al., 2022). Therefore, it is possible that higher-educated individuals are more likely to reduce their mobility to avoid their exposure to the pandemic. To capture differences in the educational composition of the regional population, we account for the share of the population aged 25 to 64 with tertiary education in 2019 as defined in the OECD Regional Database from which this data is obtained. Remote work: Workers employed in occupations that are amenable to remote work can reduce their work-related mobility. Consequently, regions with a larger

Industrial Composition, Remote Working and Mobility Changes …

193

share of workers who can work remotely would see sharper drops in worker mobility. We assessed the remote work potential as the share of jobs amenable to remote work in 2019, as defined by Dingel and Neiman (2020) and adapted to regions in the US and Canada by OECD (2020). In Fig. 2, the regions with higher remote working potential are represented by darker colours. The patterns in Figs. 1 and 2 are quite similar; the two figures hint that larger workplace mobility reductions took place in regions with high remote working potential. Union density: More organised workers may pressure employers to implement better precautions in the workplace to limit the spread of COVID-19. These measures can include fewer workers on-site or a higher share of remote work, which may reduce workplace mobility. We use data on the union density measured as the number of individuals who are a member of a union as a proportion of employees in 2019 obtained from the OECD/AIAS ICTWSS Database. Adjusted bargaining (or union) coverage rate: Workers covered by collective bargaining or who are members of a union can exert better pressure on their employers to take the necessary precautions to minimise their exposure to the virus. These precautions may lower workplace-related mobility. To capture this effect, we include a measure that corresponds to the number of employees covered by a collective agreement as a proportion of the number of eligible employees equipped (i.e., the total number of employees minus the number of employees legally excluded from the right to bargain) in 2019 obtained from the OECD/AIAS ICTWSS Database. Geographical coordinates: We obtained the longitude and latitudes of the centroid of each region from Google Maps in order to calculate the Euclidean distancebased spatial weight matrix.

2.3 Government Response to the Pandemic In addition to regional characteristics, confinement and social distancing measures played a crucial role in local mobility. Since the beginning of the COVID-19 pandemic, countries have been applying various types of containment measures and travel restrictions, while encouraging remote work (Ramírez et al., 2022). These policy measures tend to reduce social interactions and decrease individual mobility within and across regions, including daily home-to-work commuting activities, and therefore, slowing-down the spread of the virus (Ramírez et al., 2022). It is therefore plausible that the effectiveness of these measures can be proxied through the change in mobility between different periods (Glaeser et al., 2022; Pan et al., 2020; Ramírez et al., 2022). To capture government response, we use three indicators constructed by Hale et al. (2021). We take the daily average of these three indices (explained in detail below) at the regional level for 2020. The below presented variable definitions come from the COVID-19 Government Response Tracker by Hale et al. (2021). Stringency index: This index records the strictness of ‘lockdown style’ policies that primarily restrict people⣙s behaviour. It is the average of containment and closure policy scores leading to restrictions in movement (e.g. school closures). Overall government response index: This index records how the response of governments has varied over all indicators in the database, becoming stronger or weaker throughout the outbreak. It is calculated using all ordinal indicators.

194

M. G. Celbi¸s et al.

Fig. 1 Change in workplace mobility

Economic support index: This index records measures such as income support and debt relief. It is calculated using all ordinal economic policies indicators.

3 Empirical Application We adopt a two-step approach in our exploration of the regional economic and demographic determinants of workplace mobility. Firstly, our machine learning predictions are generated by an Extreme Gradient Boosting (XGBoost) model. XGBoost, developed by Chen and Guestrin (2016), algorithmically enhances the Gradient Boosting Machine (GBM) approach formulated by Friedman (2001,2002; Friedman et al., 2001). As the dependent variable, change in workplace mobility is a quantitative feature, the GBM model in our case is based on boosted regression trees.2 The following outline of our prediction model follows Friedman’s above-mentioned GBM which has been highly influential in the machine learning area, as explained in detail in Friedman et al., (2001); Friedman (2001) and Friedman (2002). The technique creates an ensemble of sequential regression trees t = 1, ..., T where each regression tree t generates revised predictions, using the training data, based on the errors made by the previous tree t − 1. When the outcome feature is quantitative (as opposed to a classification context), the very first tree is simply the average of the dependent variable values in the training data. This value F(X ) for the initial prediction of workplace mobility yˆi,1 , which is the same for every US state and Canadian province, is computed as 2

The regression tree technique is developed by Breiman et al., (1984). For more details on regression trees, see Breiman et al., (1984), James et al., (2013) and Friedman (2001). A recent application on regional European data can be found in Celbi¸s et al., (2021).

Industrial Composition, Remote Working and Mobility Changes …

195

Fig. 2 Remote working potential

yˆi,1 = arg min

N ∑

f (X )

L(yi , f (X ))

(1)

i=1

where the feature space is X = x1 , ..., x K and k = 1, ..., K , and X i denotes the vector of input values for region i in the training data (n = 1, ..., N ). Generally, a squared-error loss function is used for regression tree predictions. A convenient loss function can be used as a monotonically transformed squared-error case such )2 ∑n 1 ( yˆi − F(X i ) . In the sequence of trees, each tree t produces new as L = n1 i=1 2 predictions. This is done by acquiring the residuals from the preceding tree t − 1: [ eit = −

∂ L(yi , f (X i )) ∂ f (X i )

] f or i = 1, ..., N

(2)

F=Ft−1

In other words, Eq. 2 evaluates the expression with respect to the prediction of the previous tree (Ft−1 = yˆi,t−1 ). These residuals eit are then predicted by the regression tree t which generates nodes r and terminal nodes r¯ , each consisting of observations X i ∈ Dr¯ , where Dr¯ denotes the subset of the observations in the training data that end up in the terminal node r¯ , and each tree t has a number of terminal nodes denoted by |t|. As a result, the regression tree groups observations in X i into terminal regions r¯ and predicts a common output value e¯ r¯ ,t for each r¯ of the tree t that applies to all X i ∈ Dr¯ , as follows:

196

M. G. Celbi¸s et al.

er¯ t = arg min e

∑

L(yi , Ft−1 (X i ) + e)

(3)

xi ∈Dr¯ t

Therefore, the regression tree predicts how off-target a given group of observations were predicted by t − 1. Finally, the prediction for X i ∈ Dr¯ can be updated in the next iteration t. However, correcting each X i ∈ Dr¯ by adding er¯ t could itself cause errors in prediction, as the regression tree that predicts the residuals is not necessarily foolproof. An incremental learning structure is then implemented by allowing tree t to learn a fraction from the errors of t − 1 instead of going for a “full correction." The share of learning is given by the “learning rate" α, also called the shrinkage parameter, and 0 < α < 1. The predictions of tree t are then subject to the following revision where 1 is the indicator function: yˆit = Ft−1 (X i ) + αer¯ t 1(X i ∈ Ar¯ t )

(4)

The number of times Eq. 4 will be applied (i.e. the number of iterations T ) can be determined by an early stopping rule, which is often the number of trees after which no further prediction improvements are achieved. Regularisation and cross-validated pruning can be further introduced to the GBM through the use of the XGBoost algorithm developed by Chen and Guestrin (2016). Furthermore, Friedman (2002) introduced the stochastic gradient boosting machine where, at each iteration, a random subsample of the training data is selected without replacement. The introduction of stochasticity has been shown to improve predictions, and XGBoost expands this feature further by also allowing for a random selection of a subset of variables Z ⊂ X at each node in each t (Adam-Bourdarios et al., 2015; Chen & Guestrin 2016; Chen et al., 2015). We conduct a grid search to identify the best parameter values (i.e. those yielding the minimum test error estimate) and use the following values: α = 0.05, Z = 0.5, the share of observations randomly drawn at each iteration = 0.5, maximum depth = 5, minimum size of a node = 5 and T = 5000. The test data root mean squared error of the XGBoost predictions is 5.62. Following the XGBoost predictions performed in this step, we proceed to calculate the Shapley Additive Explanations (SHAP) values. The use of SHAP values is based on the Shapley value calculation in cooperative game theory introduced by Shapley (1953), and their use is adapted to the machine learning framework by Lundberg and Lee (Lundberg & Lee 2017). We accordingly define the SHAP value value v for feature xk for a given US or Canadian region i as: φv =

∑ |S|!(|X | − |S| − 1)! [g(S ∪ {v}) − g(S)] |X |! S⊆X \{v}

(5)

Equation 5 indicates that a subset of features S ⊂ X for region i is held constant, and the remaining feature values are replaced by those of a randomly selected region j. The term g(S), as shown in Eq. 6, is the deviation of the expected value of the mobility level of region i (subscript i is omitted)—over multiple versions where

Industrial Composition, Remote Working and Mobility Changes …

197

all features, except those in S, are replaced by the feature values of a newly and randomly drawn j—from the average predicted value in the training data E(F(X )) (Lundberg & Lee, 2017; Molnar, 2019). In Eq. 5, the variable value v pertains to region i (like those in S) as opposed to g(S). Therefore, the term in square brackets in Eq. 5 calculates the aforementioned deviation with and without the actual value of the variable of interest for region i.3 The deviation, which itself is calculated over multiple random replacements, is, in turn, performed for all possible combinations that can yield a subset S ⊆ Z \{v} and the average is calculated—hence the inverse of the combination term as shown in the fraction in Eq. 5 (Lundberg & Lee, 2017; Molnar, 2019): ∫ (6) g(S) = F(x1 , . . . , xk )d Px ∈S / − E(F(X )) Since the above implementation can hardly be run by computers due to the high number of random sampling steps for a high number of combinations, an approximation is given by (Štrumbelj & Kononenko, 2013), as shown in Eq. 7: φ¯ v =

M ] 1 ∑[ ∗ ∗ F ( yˆ (i )m ˆ (i )m +v ) − F ( y −v ) M m=1

(7)

where, for region i, the prediction F ∗ ( yˆ (i)m +v ) is computed through replacing a randomly determined subset of the feature values of i by those of j, which is, as earlier mentioned, another data instance that is randomly sampled from the training dataset. This prediction is performed twice: by keeping the true variable value v of region i as indicated by the subscript +v, and by replacing it with the corresponding value of region j as well, indicated by −v, and then the difference between the two versions of the prediction is calculated (Molnar, 2019; Štrumbelj & Kononenko, 2013). The difference is computed in this manner M = 1, . . . M times, and the SHAP value approximation is given by the average value φ¯ v (Molnar, 2019; Štrumbelj & Kononenko, 2013). For performing the SHAP value computations on the XGBoost predictions, we use the R routine SHAPforxgboost developed by Liu and Just (2020). SHAPforxgboost is also able to compute the SHAP importances by summing up, in absolute values, φ¯ v for all regions i, as also shown explicitly by Molnar (2019): Φv =

n ∑

|φi,v |

(8)

i=1

Finally, we introduce spatially lagged dependent and explanatory features to our analysis in order to explore the potential existence of spatial processes. The spatial 3 In fact, j may be equal to i in some random draws, and therefore for that specific instance in the integration, the deviation may potentially equal zero (Molnar, 2019). However, as the probability of drawing the same observation twice in the same iteration is low, this possibility can be neglected.

198

M. G. Celbi¸s et al.

dimension is introduced into the data used by the machine learning models through the use of a spatial weight matrix W , as done in Spatial Autoregressive Models (Anselin, 1988) in spatial econometric studies. W is an N × N matrix with elements wi j being equal to the inverse Euclidean distance between the cenotrids of regions i and j, where wi j = 0 if i = j, with i = 1, . . . , N and j = 1, . . . N . Spatially lagged features are denoted by adding ‘W’ to the beginning of the name of a given nonlagged feature xk . The resulting SHAP value and interaction findings are discussed in the next section.

4 Empirical Results The SHAP values for all variables with a non-zero Φv are plotted in Fig. 3, also ordered by the Φv values. Each dot represents a USA state or Canadian province. Darker colours represent the regions with higher remote working potential and are positioned in the row pertaining to a variable based on their SHAP values, shown on the x-axis. We discuss the main results under five categories.

4.1 Work and Industries As expected, regional remote working potential is a strong predictor of workplace mobility changes. In regions with high remote working potential, the contribution of this variable has been in the form of a negative deviation from the average change in workplace mobility, which is negative. In other words, higher remote working potential is associated with lower workplace mobility. A cluster of regions in light colours, i.e., those with low remote working potential, are positioned on the far right, indicating that the deviation of workplace mobility from the average value in the training data has been positive. As a result, our findings suggest that populations in regions with higher remote working potential were able to reduce their workplace mobility more. A result similar to that for RemoteWork is also observable for the InfoComm variable. Above-average reductions in workplace mobility are predicted for regions with a higher share of information and communication sector. The same relationship with the outcome variable is also indicated by the SciTechAdm, albeit with a lower absolute SHAP value. On the other hand, the degree of prominence of the agriculture, forestry and fishing activities, alongside the real estate and financial and insurance sectors, is associated with higher SHAP values. We also observe that higher reductions in workplace mobility took place in regions with higher numbers of doctors per person. Furthermore, the Shapley values for IndEnerg hint a nonlinear relationship, whereas for TradeFoodAcc the effect seems unclear. A low industrial share of the construction sector and high shares of employment in public administration, education, and health are suggested to lead to lower workplace mobility, albeit that

Industrial Composition, Remote Working and Mobility Changes …

199

the SHAP importances of these features are relatively low. The interpretation of the SHAP values of the unemployment rate is not straightforward as individuals who frequent workplaces are those who are employed. However, the results show that high rates of unemployment contributed to lower predictions of workplace mobility. The relationship could be explained better through the use of individual level data, as done by Celbi¸s et al. (2022), who find that elderly persons along with individuals working in the construction, energy supply, transportation/communication, community/personal services and manufacturing sectors have higher probabilities of becoming unemployed in locations where there are higher decreases in workplace mobility. Finally, workplace mobility decreased more in those regions with higher shares of persons working in other service sectors. A further look at these relationships outlined above is presented through the use of the interaction plots in Fig. 4, 5, 6, 7 and discussed in Sect. 4.5.

4.2 Policies Stringency policies have been effective in reducing workplace mobility. However, a stricter stringency level did not lead to a decrease in mobility, as it took place for regions with high remote working potential. That said, the effect of Stringency is the strongest among the other policy variables. High government support ranks immediately below Stringency in terms of its SHAP importance variable, but its contribution to the reduction of workplace mobility has been more than that of Stringency for several regions. The SHAP value plots for Support also suggest that these policies affect workplace mobility only above a certain level of intensity. Similarly, a high government response index led to lower predicted regional workplace mobility. Therefore, all three policy variables have played an effective mobility-reducing role in US and Canadian regions.

4.3 Demographics Among the variables related to demographics, the elderly share exhibits the highest SHAP importance. Regions with high shares of persons above 65 years old deviated positively from the average change in workplace mobility. Workplaces are arguably more populated with persons younger than 65 years old. This implies that regions with a larger share of individuals of working age experienced higher drops in workplace mobility. On the other hand, the population variable and the share of persons with tertiary education seem to be related to workplace mobility in a nonlinear way, as regions with higher SHAP values contributed to the prediction of both higher and lower than average mobility changes. Regarding the impact of the pandemic on the regional population, regions with higher values of the feature Deaths_Ratio experienced higher reductions in regional mobility in workplaces.

200

M. G. Celbi¸s et al.

Fig. 3 XGBoost SHAP values

4.4 Spatial Effects The spatially lagged explanatory variables, including the spatial lag of the outcome feature, denoted by W y in our figures, are also presented in Fig. 3. It should be noted that the spatial units in our data (US states and Canadian provinces) in general have very large surface areas. It is well known that spatial effects are less observable if spatial units are large. This may be the reason why we find spatial effects to be weaker than direct effects, as observable in our SHAP value results. W y has a very low SHAP importance level. That suggests that regions close to those with high reductions in mobility have themselves higher mobility predictions than average and can therefore be neglected. In the next section, we explore potential spatial effects further using a SHAP interactions plot, both for W y and for spatially lagged explanatory variables.

4.5 Interactions The interactions among SHAP values are visualised in Fig. 4, 5, 6, 7 where each region is represented by a circle. Panel (A) of Fig. 4 shows that for regions with higher remote working potential, RemoteWork contributed more strongly, towards predictions that were lower than the average prediction of workplace mobility changes. This

Industrial Composition, Remote Working and Mobility Changes …

Fig. 4 XGBoost SHAP interaction values: labour, demography and policy

Fig. 5 XGBoost SHAP interaction values: industries

201

202

Fig. 6 XGBoost SHAP interaction values: industries (cont’d)

Fig. 7 XGBoost SHAP Interaction Values: Spatial Effects

M. G. Celbi¸s et al.

Industrial Composition, Remote Working and Mobility Changes …

203

effect becomes particularly clear after around a RemoteWork value of around 32%. Each circle is coloured based on the lagged value of the change in workplace mobility of the corresponding region, and a clear pattern is not present. In other words, the effect of remote working potential in a region does not depend on the level of workplace mobility in its surroundings. In fact, this type of spatial dependence is also not observed for the other main explanatory spatially lagged and non-lagged variables of interest that we discuss in this section, in accordance with our findings presented in Fig. 3. The government support index begins to demonstrate a mobility-reducing effect in workplaces for values above about 50, as shown in panel (B) of Fig. 4. However, while most probably being beneficial regarding other welfare-related outcomes, the mobility-reducing effect of Support stabilises for values above around 60. In other words, for mobility, in a region where the government support index is around 80, this policy instrument contributed to lowering workplace mobility the same as in other regions where the index is lower than this value (but higher than around 60). As for RemoteWork, the absence of clear clusters of similarly coloured circles (regions) suggest that the effect of Support in a region does not depend on the changes in workplace mobility in surrounding regions. Similar interpretations can be made on the effect of stringency policies: as shown on the y-axis of the plot in panel (D), the deviation from the average prediction due to the values of this variable is very similar to that generated by similar values of Support. In addition to regional policy indices, we observe that increases in Deaths_Ratio are associated with lower mobility. Turning to industry-specific effects, in Figs. 5 and 6 we present how sectoral composition of regions contributed to the prediction of the changes in workplace mobility, together with the feature UnempRate. Panel (A) of Fig. 5 suggests that, in regions with an agriculture, forestry and fishing industry, a share of up to about 3% AgriForFish contributed to the prediction of more than average reductions in workplace mobility. But, in regions where AgriForFish was above around 3%, drops in workplace mobility have been predicted to be less than average, as shown by the positive SHAP values for this variable on the y-axis. Similar results are observed for TradeFoodAcc and FinIns, as well, as is shown in panels B and D of Fig. 6, where the positive effect is above around 25% for the former and 5% for the latter. For employment in other service industries, as shown in panel (C) of Fig. 6, the effect is less clear most probably because the “other" category lumps together many different sectors. However, a mobility-reducing pattern can be seen for regions with higher shares of employment in these areas. We also observe, in panel (D) of the same figure that workplace mobility has dropped more than average in regions with unemployment rates higher than around 4%. In regions with higher shares of persons working in the information and communication and science & technology sectors features (panels B and D in Fig. 5), InfoComm and SciTechAdm contributed to the prediction of higher mobility reductions for values more than around 1.5% for the former and around 15% for the latter. It is likely that these findings are due to the fact that individuals working in these sectors work exclusively in the areas classified under the workplaces cate-

204

M. G. Celbi¸s et al.

gory and could adopt flexible working practices more easily. On the other hand, for the industry and energy sector, we observe in panel (C) of Fig. 5 that shares higher than around 7% contribute positively to the prediction of mobility changes compared to the average. In other words, mobility decreases are more mild for regions with higher IndEnerg values. However, this finding does not apply for regions with IndEnerg values greater than about 11%, as no effect is observed for these locations. This heterogeneity may be due to the kind of industry and energy sectors present in a region, and could be understood with further disaggregated industrial data. As discussed earlier, we have been unable to observe spatial relationships through the inclusion of lagged dependent and explanatory variables. The SHAP values of selected spatially lagged variables are plotted against their non-lagged counterparts (RemoteWork, Support, InfoComm and Stringency) in Fig. 7. None of the plots show clear patterns of increase or decrease in the effect of the spatial lag of a given variable, as the value of the non-lagged variable changes. In other words, we do not observe whether, say, an increase in remote work potential is associated with the potential for remote working in surrounding regions.

5 Conclusion Particularly during the heaviest stages of the pandemic, reducing individual mobility was a crucial policy target all over the world (Borkowski et al., 2021; Chang et al., 2020; Engle et al., 2020). This chapter has focused on Canadian provinces and USA states with the aim of understanding the reasons behind spatial differences in regional mobility reductions on workplaces. The first analytical tool employed to answer the research question was a sequential boosted tree machine learning algorithm, more specifically, extreme gradient boosting (XGBoost). The second tool used for interpreting the XGBoost model was a Shapley value-based method, namely SHAP (Shapley Additive Explanations). Using a data set that includes regional socio-economic data alongside with spatial policy variables, we observed that higher remote working potential has been the main factor leading to regional reductions in workplace mobility. We categorised our explanatory features under four categories: Work and Industries, Policies, Demographics, Spatial Effects and Interactions, the latter examining various potential interactions, particularly between the main explanatory features and spatial effects. In addition to the role of regional remote working potential, our results also showed how the share of certain industries in a region have impacted workplace mobility. While some industries had nonlinear effects, we observed that the higher shares of the information and communication sector in addition to the science and technology sector had mobility-reducing effects, as opposed to the food, agriculture and energy industries. Regarding spatial interdependencies, our algorithmic approach detected only negligible effects. This outcome is most probably due to the fact that the spatial units examined in this chapter are large areas, and, as a result spatial processes may be

Industrial Composition, Remote Working and Mobility Changes …

205

clouded. However, using smaller spatial units may be useful in capturing spatial effects through tree-based machine learning methods, as suggested by our application. Finally, our results suggest that policies targeted towards reducing mobility in workplaces, have been useful as intended, while having less influence in workplace mobility reductions compared with regional remote working potential and industrial characteristics.

References Adam-Bourdarios, C., Cowan, G., Germain, C., Guyon, I., Kégl, B., & Rousseau, D. (2015). The Higgs boson machine learning challenge. In NIPS 2014 workshop on high-energy physics and machine learning, pp. 19–55. Ajzenman, N., Cavalcanti, T., & Mata, D. D. (2020). More than words: Leaders’ speech and risky behavior during a pandemic. SSRN Electronic Journal. Almagro, M., & Orane-Hutchinson, A. (2022). JUE insight: The determinants of the differential exposure to COVID-19 in New York city and their evolution over time. Journal of Urban Economics, 127, 103293. Anselin, L. (1988). Lagrange multiplier test diagnostics for spatial dependence and spatial heterogeneity. Geographical Analysis, 20(1), 1–17. Ascani, A., Faggian, A., Montresor, S., & Palma, A. (2021). Mobility in times of pandemics: Evidence on the spread of COVID19 in Italy’s labour market areas. Structural Change and Economic Dynamics, 58, 444–454. Barbieri, P. N., & Bonini, B. (2021). Political orientation and adherence to social distancing during the COVID-19 pandemic in italy. Economia Politica, 38(2), 483–504. Barbieri, T., Basso, G., & Scicchitano, S. (2021). Italian workers at risk during the COVID-19 epidemic. Italian Economic Journal, 8(1), 175–195. Bargain, O., & Aminjonov, U. (2020). Trust and compliance to public health policies in times of COVID-19. Journal of Public Economics, 192, 104316. Bonacini, L., Gallo, G., & Scicchitano, S. (2020). Working from home and income inequality: risks of a ‘new normal’ with COVID-19. Journal of Population Economics, 34(1), 303–360. Borkowski, P., Ja˙zd˙zewska-Gutta, M., & Szmelter-Jarosz, A. (2021). Lockdowned: Everyday mobility changes in response to COVID-19. Journal of Transport Geography, 90, 102906. Brandily, P., Brébion, C., Briole, S., & Khoury, L. (2021). A poorly understood disease? the impact of COVID-19 on the income gradient in mortality over the course of the pandemic. European Economic Review, 140, 103923. Breiman, L., Friedman, J. H., Olshen, R. A., & Stone, C. J. (1984). Classification and regression trees. Wadsworth and Brooks. Casalone, G., Michelangeli, A., Östh, J., & Türk, U. (2021). The effect of lockdown on students performance: A comparative study between Sweden, Italy, and Turkey. University of Milan Bicocca Department of Economics, Management and Statistics Working Paper, (473). Caselli, M., Fracasso, A., & Scicchitano, S. (2022). From the lockdown to the new normal: Individual mobility and local labor market characteristics following the COVID-19 pandemic in Italy. Journal of Population Economics. Celbi¸s, M. G., Hang Wong, P., Kourtit, K., & Nijkamp, P. (2022). Impacts of the COVID-19 outbreak on older-age cohorts in european labor markets: A machine learning exploration of vulnerable groups. Regional Science Policy & Practice.

206

M. G. Celbi¸s et al.

Celbi¸s, M. G., Wong, P.-H., Kourtit, K., & Nijkamp, P. (2021). Innovativeness, work flexibility, and place characteristics: A spatial econometric and machine learning approach. Sustainability, 13(23), 13426. 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, 589(7840), 82–87. Chen, T., & Guestrin, C. (2016). xgboost: A scalable tree boosting system. In Proceedings of the 22nd ACM SIGKDD international conference on knowledge discovery and data mining, pp. 785–794. Chen, T., He, T., Benesty, M., Khotilovich, V., & Tang, Y. (2015). xgboost: Extreme gradient boosting. R package version 0.4-2, 1-4. Chernozhukov, V., Kasahara, H., & Schrimpf, P. (2021). Causal impact of masks, policies, behavior on early COVID-19 pandemic in the U.S. Journal of Econometrics, 220(1):23–62. Crowley, F., & Doran, J. (2020). COVID-19, occupational social distancing and remote working potential: An occupation, sector and regional perspective. Regional Science Policy & Practice, 12(6), 1211–1234. Dingel, J. I., & Neiman, B. (2020). How many jobs can be done at home? Journal of Public Economics, 189, 104235. Engle, S., Stromme, J., & Zhou, A. (2020). Staying at home: Mobility effects of COVID-19. SSRN Electronic Journal. Friedman, J., Hastie, T., & Tibshirani, R. (2001). The elements of statistical learning (Vol. 1). Springer. Friedman, J. H. (2001). Greedy function approximation: A gradient boosting machine. Annals of Statistics, 5, 1189–1232. Friedman, J. H. (2002). Stochastic gradient boosting. Computational Statistics & Data Analysis, 38(4), 367–378. Gauvin, L., Bajardi, P., Pepe, E., Lake, B., Privitera, F., & Tizzoni, M. (2021). Socio-economic determinants of mobility responses during the first wave of COVID-19 in Italy: from provinces to neighbourhoods. Journal of The Royal Society Interface, 18(181), 20210092. Glaeser, E. L., Gorback, C., & Redding, S. J. (2022). JUE insight: How much does COVID-19 increase with mobility? Evidence from New York and four other U.S. cities. Journal of Urban Economics, 127, 103292. Google LLC (2021). Google COVID-19 community mobility reports. www.google.com/covid19/ mobility/ Accessed 12 Nov 2021. Hale, T., Angrist, N., Goldszmidt, R., Kira, B., Petherick, A., Phillips, T., Webster, S., CameronBlake, E., Hallas, L., Majumdar, S., & Tatlow, H. (2021). A global panel database of pandemic policies (Oxford COVID-19 Government Response Tracker). Nature Human Behaviour, 5(4), 529–538. James, G., Witten, D., Hastie, T., & Tibshirani, R. (2013). An introduction to statistical learning (Vol. 112). Springer. Kashnitsky, I., & Aburto, J. M. (2020). COVID-19 in unequally ageing European regions. World Development, 136, 105170. Liu, Y., & Just, A. (2020). SHAPforxgboost: SHAP plots for ‘XGBoost’. R package version 0.1.0. Lundberg, S. M., & Lee, S.-I. (2017). A unified approach to interpreting model predictions. In Proceedings of the 31st international conference on neural information processing systems, (pp. 4768–4777). Molnar, C. (2019). Interpretable Machine Learning. https://christophm.github.io/interpretable-mlbook/. OECD (2020). OECD regions and cities at a glance 2020. OECD. OECD (2021). Implications of remote working adoption on place based policies. Ozguzel, C., Veneri, P., & Ahrend, R. (2020). Capacity for remote working can affect shutdowns’ costs differently across places. OECD Policy Responses to Coronavirus (COVID-19).

Industrial Composition, Remote Working and Mobility Changes …

207

OECD, Paris. http://www.oecd.org/coronavirus/policy-responses/capacity-forremote-workingcan-affect-lockdown-costs-differently-acrossplaces-0e85740e/ Pan, Y., Darzi, A., Kabiri, A., Zhao, G., Luo, W., Xiong, C., & Zhang, L. (2020). Quantifying human mobility behaviour changes during the COVID-19 outbreak in the United States. Scientific Reports, 10(1), 1–9. Ramírez, M. D., Veneri, P., & Lembcke, A. C. (2022). Where did it hit harder? Understanding the geography of excess mortality during the COVID-19 pandemic. Journal of Regional Science, 62(3), 889–908. Shapley, L. S. (1953). A value for n-person games. In H. Kuhn & A. Tucker (Eds.), Contributions to the theory of games II, annals of mathematics studies (Vol. 28, pp. 307–317). Princeton University Press. Stier, A. J., Berman, M. G., & Bettencourt, L. M. A. (2020). Covid-19 attack rate increases with city size. Štrumbelj, E., & Kononenko, I. (2013). Explaining prediction models and individual predictions with feature contributions. Knowledge and Information Systems, 41(3), 647–665. Zhong, B.-L., Luo, W., Li, H.-M., Zhang, Q.-Q., Liu, X.-G., Li, W.-T., & Li, Y. (2020). Knowledge, attitudes, and practices towards COVID-19 among Chinese residents during the rapid rise period of the COVID-19 outbreak: A quick online cross-sectional survey. International Journal of Biological Sciences, 16(10), 1745–1752.

The Need for New Types of Data and Applications, and Existing Challenges in Analysing the Effects of COVID-19 on Cities

Problems with Recording the Spread of COVID-19 in Developing Countries: Evidence from a Phone Survey in Indonesia Budy P. Resosudarmo, Rus’an Nasrudin, Pyan A. Muchtar, Usep Nugraha, and Anna Falentina

1 Introduction Not much attention was given when the Wuhan Municipal Health Commission in China identified a novel coronavirus in late December 2019 or even when the Chinese government closed down a wet market in Wuhan on 1 January 2020 due to a potential disease outbreak. At that time, approximately 260 people were identified as infected in China. The infection was originally named 2019-nCoV and later became widely known as coronavirus disease 2019 or COVID-19. COVID-19 started to attract global attention when, at the end of January 2020, the World Health Organisation (WHO) announced that there were approximately 7,700 confirmed COVID-19 cases globally. Among them, approximately 170 people passed away, and the disease spread to 18 other countries. Since then, cases of COVID-19 spread rapidly around the world, severely impacting human health and the global economy. By end of April 2020, it was reported that approximately 3.1 million people were affected and approximately 217 thousand people had died because of the virus. By June 2020, almost no country had been left untouched, due to the severity of the virus and its ability to rapidly spread between humans. Since then, the world quickly realised that they were facing a major crisis (Baldwin & di Mauro, 2020). Developed countries, which have better health facilities than developing countries, quickly developed testing, trace, and treat (3T) systems. The 3T system is an effort B. P. Resosudarmo (B) Arndt-Corden Department of Economics, Australian National University, Canberra, Australia e-mail: [email protected] R. Nasrudin · U. Nugraha Faculty of Economics and Business, Universitas Indonesia, Depok City, Indonesia P. A. Muchtar Economic Research Institute for ASEAN and East Asia, Jakarta, Indonesia U. Nugraha · A. Falentina Statistics Indonesia, Jakarta, Indonesia © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. G. Celbi¸s et al. (eds.), Pandemic and the City, Footprints of Regional Science, https://doi.org/10.1007/978-3-031-21983-2_12

211

212

B. P. Resosudarmo et al.

to detect the presence of the COVID-19 virus by conducting as many as possible COVID-19 tests within any community, provide as soon as a possible comprehensive treatment to patients diagnosed with positive COVID-19, and trace contacts, mostly using mobile applications, who have interacted with those patients. In such case, developed countries—i.e., several European countries and the United States of America (USA)—could at least accurately predict the magnitude of the pandemic and monitor the spread of the pandemic. Governments in those countries could, therefore, promptly and appropriately respond to the pandemic, e.g., by implementing a policy of restricting people’s mobility and social activities in areas with high confirmed COVID-19 cases. On the other hand, most developing countries did not have the capabilities in implementing a solid 3T system. Most of them could not conduct the COVID-19 test as many as those in developed countries (Olivia et al., 2020). As a result, governments’ announcements on the figures of COVID-19 cases have been underestimated and the actual size of the pandemic has become difficult to determine. This has led to inappropriate policy responses, such as no strict restrictions on travelling or on gathering in crowded places (Resosudarmo et al., 2021). At least in the first year of the pandemic, in many developing countries, solid prediction of the spread and magnitude of the pandemic was not available. The lack of clarity about the size of the COVID-19 pandemic can cause some further problems. First, this condition can cause public panic and conflicts (Mietzner, 2020). Second, fluctuations of the outbreaks could frequently happen, i.e., the cases of India’s drastic increases of infection cases on September 2020 and in May 2021, or of Thailand’s increase of infection cases on August 2021.1 Third, the pandemic may become endemic in several areas within developing countries due to the inability of those countries to locally eliminate the virus (Herrero & Madzokere, 2021). To overcome the uncertainty of the actual spread and size of COVID-19 in developing countries, many non-governmental agencies/individuals conducted various methods to gather data on the spread of the pandemic as an alternative to government announced figures. Most of these methods expect that the actual numbers are much larger than those announced by their governments (Barro et al., 2021; Chandra, 2013; Resosudarmo & Irhamni, 2021). This paper aims to assess these data gathering and estimation methods conducted by non-governmental agencies/individuals to be able to quickly assess the spread and the magnitude of the pandemic, with Indonesia as the case study. This paper deals with the question of whether these alternative methods produce reliable quality estimates and so could be an alternative estimate for government formal figures on the spread of COVID-19. This paper evaluates the reliability and usefulness of a phone survey approach by carefully conducting rapid phone surveys in Jakarta and Yogyakarta on our own in the mid 2020. We evaluate the results of our phone survey to understand the quality of rapid data gathering and estimation method with phone survey approach in predicting the spread of COVID-19. We also compare our results with those based on other methods whenever possible. However, our conclusion will remain subjective. We 1

WHO Coronavirus (COVID-19) Dashboard: https://covid19.who.int.

Problems with Recording the Spread of COVID-19 in Developing …

213

admit that a phone survey is not a substitute for comprehensive COVID-19 testing for precisely estimating the spread of the pandemic in a given country/region. Indonesia is used as the case study for the following reason. Indonesia is not that far from China and has a relatively strong economic connection with China. Large number of exchange business trips happened daily between the two countries. Consequently, Indonesia is not an exception; rather, it seems highly susceptible to the spread of COVID. Although initially there was optimism among Indonesians that COVID19 would not spread out much in the country due to its tropical weather of the country, the Indonesian government reported the first COVID-19 case found in the country on 2 March 2020 (Kompas, 2020). Since then, more tests have been conducted; these tests initially have been mostly in Java and later throughout the country. Cases of COVID-19 infection detected have since then rapidly grown. Figure 1 shows the average 7 days new cases detected since the first case was announced by the Indonesian government till end of August 2020. As mentioned before, we argued that these numbers might be underestimate the actual spread of the pandemic. However, these numbers can illustrate how the pandemic has spread out throughout the country. It is concerning that number of new cases detected has not been flattened at least till the end of August 2020 and, although it cannot be seen in Fig. 1, the spread of the cases has gone throughout the country. By 31 August 2020, there were at least approximately 174 thousand confirmed cases in the country. Observing what has happened in Indonesia will be an important lesson for other developing countries, particularly those in Southeast Asia.

Fig. 1 Government’s report on the spread of COVID-19 in 2020. Source Our World in DataUniversity of Oxford

214

B. P. Resosudarmo et al.

Furthermore, Indonesia is the fourth most populous country in the world, after China, India and the USA. Hence, what has happened in Indonesia matters for a not insubstantial portion of the world’s population. Important to note as well is that other populous countries are continental countries, while Indonesia is a large archipelago. Indonesian case in managing COVID-19 hence will be an interesting case to observe because it illustrates the global ability in managing the pandemic. The next section reviews the uncertainty situation in Indonesia during the first semester and the Indonesian government’s initial response in managing the pandemic. This section is followed by a section reviewing several private attempts to estimate the actual spread of the virus in the country. We then describe our own phone survey to monitor the spread of the virus and the results of our estimates. This paper is then ended with a final remark. Our general conclusion would be that, although the reliability of a phone survey could be variable, a phone survey might be able to provide a better estimate than those announced by the government in developing countries where the capabilities of conducting 3T system are limited.

2 Indonesia During the First Semester of the Pandemic In the early period of the pandemic outbreak, there are two main questions related to Indonesia’s public health sector: (1) how effective Indonesia could implement public health policies to curb the spread of the pandemic, and (2) how prepared Indonesia’s health sector is in treating massive number of patients affected by COVID-19. On the public health sector, it was quickly realised that the country’s ability to detect and to control the spread of the pandemic has been limited. The country was unable to quickly increase the number of testing and make the results quickly available, particularly off Java-Bali islands. For example, the number of tests per 1,000 head of population in Indonesia by mid-August 2020 was only 2–3 per cent of those numbers in developed countries such as Australia and the United States. The number of tests per capita in Indonesia so far has also been smaller than those in India, the Philippines, Vietnam, or Thailand (Fig. 2). This limitation to timely detect the COVID-19 cases created uncertainty on the actual spread of the virus. It seemed, however, the spread of the pandemic has been much larger than the situation formally announced by the government. Strict restriction of human movements to avoid the spread of the pandemic seemed to be almost impossible, given a large proportion of the population’s livelihood depends on their daily activities. Furthermore, in the early period of the outbreak, the Eid al-Fitr holiday was approaching, which most Indonesians have a tradition to return to their hometowns. This created more speculative predictions that there would be an explosion of COVID-19 cases in several areas of the country, particularly in Jakarta and some other areas in Java Island. Ability of the country’s health facilities to treat COVID-19 patients was also in question. Table 1 shows Indonesia’s health sector facilities when the pandemic

Problems with Recording the Spread of COVID-19 in Developing …

215

Total Test per Thousand People

250 200 150 100 50 0

Fig. 2 Number of COVID-19 test in several countries by mid-August 2020. Source Our World in Data-University of Oxford

entered the country in 2019 in comparison with the situation in other countries. Despite the improvement of Indonesia’s health infrastructure, the national number of health facilities per population had been low compared to advanced countries, such as Australia, the United States, Japan, and EU countries. Indonesia’s health facilities might be comparable with India and the Philippines, but below Thailand and Vietnam. Even in comparison with India and the Philippines, Indonesia is in a disadvantaged situation. Indonesia might have more hospital beds (health physical capital), but not medical workers (health human capitals) which could be more crucial in treating patients. Table 1 Health facilities in Indonesia and in other countries Country

Nurses and midwives (per 1,000 people)

Hospital beds (per 1,000 people)

Physician (per 1,000 people)

Australia

12.6

3.8

3.7

Indonesia

2.4

1.2

0.4

India

1.7

0.7

0.9

Latin America and Caribbean

5.1

2.2

2.3

Euro area

9.8

6.2

3.9

United States

14.5

2.9

2.6

Vietnam

1.4

2.6

0.8

Thailand

2.8

2.1

0.8

Philippines Japan

4.9 12.2

1

0.6

13.4

2.4

Source World Development Indicators (WDI), World Bank (2020)

216

B. P. Resosudarmo et al.

Important to note as well is that Indonesia is still struggling in managing several endemics such as tuberculosis, dengue fever and malaria infections. Furthermore, which might be true for some other developing countries, these health facilities are not equally distributed throughout Indonesia. The facilities, such as the number of medical doctors, have been, in general, limited outside Java and Bali Islands. The overall condition of health facilities in Indonesia, hence, indicates that the country has limited ability to prevent the outbreak of COVID-19 pandemic. Government responses to the pandemic have been, nevertheless, relatively quick. Few days after the first confirmed cases were found, a new task force to manage the spread of the pandemic was formed. The government also quickly enacted a regulation for large social distancing (31 March 2020). However, Indonesia has never strictly imposed the social distancing policy as well has never fully restricted the movement of people within and across regions. On 31 March 2020, the President signed the Government Regulation in Lieu of Law (Peraturan Pemerintah Pangganti Undang-Undang or Perppu) No. 1/2020 on State Financial Policy and Stability of Financial Systems in Mitigating the COVID19 Pandemic and/or Other Threats to the National Economy and/or the Financial System Stability. The Law allows the government to have a budget deficit of more than 3% from 2020 to 2022, to seek other revenues to fund this deficit, and to revise the 2020 national government budget to be able to provide better responses to the pandemic. On 3 April 2020, the President issued the Presidential Regulation (Peraturan Presiden or Perpres) No. 54/2020 to set up a budget for COVID-19 public health and national economic recovery (PEN) programs. By signing Perpres No. 72/2020 on 24 June 2020, the government increased this budget allocation to be as high as IDR 695.20 trillion. The general objectives of the COVID-19 public health and PEN program would be: (1) to control the spread of COVID-19 pandemic as well as providing better treatment for COVID-19 affected patients by, among others, providing support to medical workers and COVID-19 patients; (2) to soften the impact of the pandemic on poor and vulnerable groups by providing social assistances and preserving jobs; and (3) to be able to implement countercyclical policies to reverse the economic downturn as well as increasing the resilience of the economy and providing foundations to jump start the economy when the pandemic is manageable by, among others, supporting enterprises, particularly micro, small and medium enterprises (MSMEs). The COVID-19 and PEN programs consist of (1) public health program (IDR 87.55 trillion), (2) social assistance program (IDR 203.9 trillion), (3) sectoral and regional program (IDR 106.11 trillion), MSME incentive program (IDR 123.46 trillion), enterprise incentive program (IDR 120.61 trillion) and corporate financing program (IDR 53.57 trillion). By mid-July, a total of IDR133.93 trillion (or approximately 19.3% of the total COVID-19 related program budget) has been disbursed. Given implementation period of many programs are for until September (or 6 months since April) this year, this rate of disbursement is slow.

Problems with Recording the Spread of COVID-19 in Developing …

217

Disbursement of funding for social assistance has reached approximately 38%. The social assistance programs also disbursed the most funding so far (IDR77.42 trillion). Most of the mechanisms to distribute this fund are existing mechanisms such as PKH and Kartu Sembako. This is the main reason for being able to smoothly distribute the funding, in general. By July 2020, approximately 19 million households participated in the Kartu Sembako. Approximately 10 million households among those Kartu Sembako holders are also likely to receive PKH packages, i.e., those having pregnant women, children aged 0–6 years old and school-aged children. For Bansos-Sembako in Greater Jakarta area, by mid-July, there have been approximately 1.8 million households receiving this. For Bansos-Tunai in outside Greater Jakarta areas, approximately 8.7 million households have received this program from April until June this year. Implementation of electricity subsidy has also seemed to work well. More than 45% of the budget for the electricity subsidy has been spent as of mid-July 2020, which is on track as the remaining budget is for the allocation from July to September 2020. The concern is the disbursement of the public health budget. Only approximately 7% of the budget (Olivia et al., 2020) for this program (approximately IDR 6.32 trillion) has been disbursed by mid-July. More than 70% of the budget (IDR 65.8 trillion) is allocated to expenses for COVID-19 disease mitigation and prevention. Unfortunately, the budget process to allocate the funding to appropriate ministries has been slow. Only in July 2020 that Rp. 31.8 trillion have been included in the budget execution list (DIPA) of Ministry of Health (IDR 23.8 trillion) and Ministry of Defence (Rp. 8.1 trillion), which means that the budget is ready to be spent. Unlike the other sub-programs under health sector which have more specific disbursement mechanism and clear target of potential beneficiaries, the details on the activities for budget allocation for expenses for handling COVID-19 sub-program is yet to be updated and made publicly available. Without any comprehensive strategies on how to spend such health budget allocation, it is even more challenging to timely meet the needs of public health demand to address the existing COVID-19 cases and contain its spread. It may not necessarily mean that there is a decline in the abilities of the country’s public health sector in managing a pandemic, since the regular public health budget should still be disbursed as usual. However, it does mean that the ability of the public health sector in managing the pandemic may not be as high as expected. The largest budget for the COVID-19 public health program is for COVID-19 disease mitigation and prevention, which mostly provides medical equipment (IDR 65.8 trillion). Budget disbursement for this activity has yet to happen. Using the budget allocated to them, the COVID-19 response acceleration task force, however, has been able to provide some COVID-19 medical equipment and distributed them throughout the country. The most recent figures reported that the task force has distributed the following medical equipment related to the handling of COVID-19 patients; approximately 4.7 million Personal Protective Equipment (PPE) coveralls, 22 million surgical masks and 1.1 million rapid testing kits, throughout the country. Budget disbursements for MSMEs incentive and enterprise incentive programs by mid-July have also been slow, i.e., approximately 24.4% (IDR 30.15 trillion) and

218

B. P. Resosudarmo et al.

11.3% (IDR 13.64 trillion), respectively, from their total budget. As consequence, the resilience of the country’s MSMEs in facing the current pandemic impact might not improve as much as expected. Hence, it is particularly important to monitor the implementation progress of these programs. Incentives to support enterprises by relaxing their tax obligations, in general, have been progressively implemented. For example, although the level of budget implementation for the income tax relief (PPh 21 DTP) programs by 15 July has only been approximately 3% of its target, approximately 108 thousand enterprises have received benefits from this program. Majority of beneficiaries are small enterprises. Implementation of the programs to support enterprises by relaxing their credit obligation, meanwhile, has yet to be effective. The fund disbursement through regional governments has been slow. Coordination with regional government typically takes time. Channelling funding through sectoral ministries should be much faster. Unfortunately, this is not the case so far. By midJuly, IDR 6.40 trillion of the sectoral and regional programs have been disbursed (6.03% of its total budget). No budget for the corporate financing has been disbursed by mid-July. Slow implementation of corporate finance programs is not that alarming. It needs careful preparation to ensure its effectiveness. Given the inability of the Indonesian government in increasing the number of COVID-19 tests, in quickly disbursing budget to control the spread of the virus and the weak public health facilities, finding a reliable estimate of the spread of the virus became urgent. This was the reason that several private agencies took their own initiative to conduct various methods of data gathering and estimation to predict the spread of COVID-19 in the country.

3 Private Initiatives of Estimating the Spread of COVID-19 3.1 Digital COVID-19 Information Collection Platforms The first method of quick data gathering implemented in Indonesia has been a digital information collection platform. Since the beginning of the pandemic around March 2020, there have been voluntary initiatives by private agencies/citizens to collect reports from individuals or organisations on COVID-19 cases in their areas via digital platforms. The most prominent of these platforms are Kawal COVID-19 and Lapor COVID-19. These platforms gathered public attention as they successfully present decent and user-friendly visualization of COVID-19 related data and information. Their presence as a source of information was important during the early pandemic given the chaotic information flow among citizens, particularly in the online environment on the spread of COVID-19. Lapor COVID-19 built a citizen reporting platform that works as a place to share information about incidents related to COVID-19 which are not covered by the

Problems with Recording the Spread of COVID-19 in Developing …

219

government. The website uses a crowdsourcing approach that involves the participation of citizens in recording COVID-19 numbers, including death outside the hospital and death during self-isolation. It also gathers and publishes citizens’ report on COVID-19 issues, such as health protocol violation, stigma related to COVID19, public services on COVID-19 cases, and government social assistance programs. Citizens can easily report these issues through WhatsApp and Telegram-based chatbots which are active 24/7. Lapor COVID-19 does not provide an estimate on the number of COVID-19 cases; it does, however, provide some indications on where COVID-19 has been spread out. Figure 3 shows an example of a citizen’s ‘mapstory’ report. It shows a case of a person who tested positive but could not get any response from the COVID-19 hotline in Sukabumi, West Java. This type of report makes people aware that this formally uncounted case could happen in several places. Kawal COVID-19 started as a web-based platform that serves as a source of information from Indonesian citizen volunteers, consisting of health practitioners, academics, and professionals. The site provides some real-time statistics, such as the number of active cases, death rate, and cure rate. It also educates people about health-related information and guides them to fight against hoaxes. The data and information presented in the platform is an aggregation of various data taken from government sources, health NGOs, and the media that have been published, both in Indonesia and outside Indonesia. On the cases of COVID-19, this platform mostly, however, recorded those who have been formally confirmed as COVID-19 patient. Hence, Kawal COVID-19 does not provide an alternative estimate of COVID-19 cases, which is supposed to include those who have not formally confirmed to be infected by COVID-19, but are sick. Kawal COVID-19 reveals the latest formal counts faster than the government typically does. An example of information can be seen in Fig. 4. It shows the total cases of COVID-19 infection, death and recovered by August 2020.

Translation: 28 April 2021 Hello, good afternoon. I hope this message reaches you. I am ****, a citizen of Sukabumi district. On 19th of April, I tested positive of COVID-19. I am in self-isolation and my company covers my medical needs. I tried to call the district COVID-19 hotline which I got from the government website (Pilkobar) but did not get any response. I have also registered myself on the website for a free swab test, but the schedule has not been provided. I am still waiting for any notification. …

Fig. 3 An example of a citizen’s ‘Mapstory’ report at the Lapor COVID-19. Source Lapor COVID19, retrieved from https://laporcovid19.org/data/mapstory

220

B. P. Resosudarmo et al. 4,000

200,000

160,000 3,000 120,000 2,000 80,000 1,000 40,000

0 2-Mar-20

0 2-Apr-20

New case

2-May-20 Total case

2-Jun-20 Death

2-Jul-20

2-Aug-20

Recovered

Active case

Note: The right axis is only for new case, others use the left axis

Fig. 4 Cases of COVID-19 reported by the Kawal COVID-19. Note The right axis is only for new case, others use the left axis. Source Kawal COVID-19 database, retrieved from https://datastudio. google.com/u/0/reporting/fda876a7-3eb2-4080-92e8-679c93d6d1bd/page/3cjTB

There have been other digital COVID-19 information gathering platforms established in Indonesia collecting publicly available data and presenting them for a specific purpose. Among others are Kawalcorona.com specializing in collecting information at provincial levels, zicare.id specializing in information on inpatient status, and Drone Emprit specializing in analysing COVID-19 sentiments in Twitter. Despite providing convenient information, these platforms have not inclusively accessible for all citizens. Smartphones are widely available in Indonesia, but more than 50% of the population still have no access to this technology, particularly in rural area (Falentina et al., 2021). Even for someone who has access, some may still find it difficult to interact with a bot in Whatsapp/Telegram. There is no local land-based office where a citizen can report in-person. Hence, the main issue with this technology in collecting information would be a selection bias issue; there are certain characteristics of individuals who would participate in the system. The results from this method would be bias or reliable for covering some cases only (Solanki et al., 2020; Wu et al., 2020). They mostly cover data and information about cases in metropolitan areas, such as Jakarta, Surabaya, and Yogyakarta.

3.2 Online Survey Approach One of the earliest household surveys conducted related to COVID-19 in Indonesia is an online survey, i.e., making a list of questions available on a webpage and directing candidates for respondents to the website. Instead of phone survey, some organisations employ online survey approach. An example of this approach is an

Problems with Recording the Spread of COVID-19 in Developing …

221

online survey conducted the J-PAL Southeast Asia using the Google Survey Platform to monitor the economic impact of the pandemic.2 In this survey, they utilise a convenient sampling—a type of non-probability sampling method where the sample is taken from a group of people easy to contact or to reach—among Gmail users in Indonesia. The Statistics Indonesia took a similar approach to understand people’s behaviour during the pandemic and the implementation of health protocol (BPS, 2020). Based on information from the respondents in their previous surveys, Statistics Indonesia invited these respondents to visit the website containing their online survey. For the period of 7–14 September 2020, they can get approximately 90 thousand people filling out their online survey. Online survey is the cheapest and the most convenient approach to collect data. The time span needed to complete the survey is much less than other traditional approaches. It also allows for quick analysis and quick policy response. However, as the data collection lies on voluntary responses, the sample may not sufficiently represent the targeted population. Most online surveys do not use rigorously designed sampling weights. More importantly, it may suffer from low accountability and low accuracy (Solanki et al., 2020; Wu et al., 2020).

3.3 Rapid Phone Survey Approach Another approach available in Indonesia in providing alternative information on the spread of COVID-19 for government announcements is by conducting a rapid phone survey asking people whether they are sick or not. The choice of phone survey was due to the need to follow the protocol for conducting safe and fast data collection during the COVID-19 pandemic. A number of research organisations and academic institutions have shifted all of their field data collection to remote approaches, primarily rapid phone surveys. Phone surveys bear no risk of disease infection spread which was the main reason it was used for data collection during Ebola crisis in West Africa (Etang & Himelein, 2020), and hence it suits the condition of COVID-19 pandemic. Compared to field survey, rapid phone survey is cheaper as it does not require transportation and accommodation expenses.3 Surveyors can conveniently move from one interview to another without travelling and thus save time. Phone surveys are often used in epidemiology research because it has some advantages, including largescale accessibility, rapid data collection (especially with the integration of computerassisted systems), quality control, anonymity, and flexibility (Safdar et al., 2016). Furthermore, the centralised aspect of phone-based data collecting enables quick

2

See: https://www.povertyactionlab.org/blog/11-10-20/monitoring-social-impact-covid-19-pan demic-indonesia. 3 The interview cost to obtain one respondent for in-person interview is approximately 10–30 USD in Indonesia, while phone interview cost is approximately 5–10 USD.

222

B. P. Resosudarmo et al.

identification and correction of errors, real-time interaction with primary researchers, and faster data entry (Dillon, 2012). Nevertheless, phone surveys have some disadvantages, such as lack of visual materials, time constraints, wariness, and inattentiveness (Safdar et al., 2016). Sometimes telephone calls are perceived as telemarketing, spam, or even scam, causing potential respondents to respond negatively. It can also be challenging to design an effective interview because the questions need to be short and precise for easy comprehension. Ideally, a phone interview should be less than 30 min. Another major issue is that phone surveys are prone to considerable nonresponse, especially for mobile phone samples in panel surveys (Fuchs, 2012). The list of potential respondents is frequently unreliable as many people change their phone numbers or they do not want to participate in another interview. Other criticism of phone survey concerns its methodological implications of survey error design, such as coverage (Busse & Fuchs, 2012), sampling issues (Wolter et al., 2010) and measurement error (Vehovar et al., 2010). Several phone surveys are conducted in Indonesia on COVID-19-related topics, including older people’s wellbeing (Komazawa et al., 2021) on 3,430 respondents. The study found that along with the decline in economic conditions, the physical and mental health of some older people during the COVID-19 pandemic has deteriorated. The World Bank conducted 5 rounds of phone-based surveys over 4,000 households to track the impact of COVID-19 on their welfare (Purnamasari & Sjahrir, 2020). One of the findings from the latest round (March 2021) is that while recovery in employment continued, the prevalence of food insecurity remains unchanged since July 2020. The World Bank also run another call survey to study gender-based violence during the pandemic (Halim et al., 2020). Lembaga Survey Indonesia (LSI) (2021) managed to collect over 1,200 respondents through phone interviews which gathered information as follows. The majority, 66.5%, feel very/sufficiently aware about COVID-19 and they feel that the virus is a threat to the economy. Around 40.5% of the respondents think that they are very/quite likely infected with the virus. The majority, 90.3%, know that the government has started a vaccination program. Furthermore, the majority, 84.9%, strongly/agree with the COVID-19 vaccine program for the community. Although much information was collected, these surveys do not estimate the number of COVID-19 cases and its spread. Regarding the phone survey, the quality of information they gathered is questionable, as rejection rates have been exceedingly high (approximately 80–90%) and respondents tend to be not as serious, because they felt like being directly interviewed in responding to the questions. This paper is evaluating the effectiveness of the phone survey approach and compares the results with those from the econometric approach.

3.4 Econometric Cross-Country Approach Several agencies are collecting global data on the number of tests, confirmed COVID19 cases, deaths due to COVID-19, and other COVID-related measures in various

Problems with Recording the Spread of COVID-19 in Developing …

223

countries in the world. Among the popular ones are those that have been conducted by John Hopkin University Corona Virus Research Centre, Google Cloud’s COVID-19 Public Dataset Program, and Our World in Data-University of Oxford. These global datasets are publicly available, allowing academics, NGOs, and government staff to use them for research purposes or policymaking. If the estimation is properly conducted, this approach could also provide an alternative estimate of the cases of COVID-19 in a country. An example of this work is by Resosudarmo and Irhamni (2021) who use data provided by the Our World in Data-University of Oxford on number of tests and of detected cases for countries around the world. Controlling for country- and day-fixed effects, there appears to be a strong relationship between the number of tests and the number of cases detected. The details are as follows: Using a daily observation data on total cases and tests in approximately 100 countries from 1 January to 24 August 2020, the following relation is estimated to reveal the true size of the pandemic: ( ) ( ) ln h i,d = σ.ln ti,d + θi + θd + εi,d where h i,d is total daily case in country i on day d, ti,d is the total daily test in country i on day d. These variables are specified in logarithm unit. θi is the country-fixed effect representing time-invariant characteristics of each country. θd is the day-fixed effect representing daily global condition during the observation. Parameter of interest is σ, which is the elasticity of a person being infected per testing. Table 2 shows the regression result. Models (1) and (2) are the results of an ordinary least square (OLS) estimation method and models (3) and (4) are from a fixed effects method. Models 2 and 4 use the number of total test-lagging 14 days. The models are preferred as there has been some lag between the COVID-19 test and the result in many countries at that time. Among models (2) and (4), model (4) is preferable since it strictly controls time-invariant characteristics of each country and daily global condition during the observation. The model indicates that a one percent increase in the total number of tests on any given day increases the total number of confirmed COVID-19 cases by approximately 0.34%. Utilising the model (4), Resosudarmo and Irhamni then estimated the cases of COVID-19 in Indonesia as follows. It was recorded that on 20 August 2020 in Indonesia, the total number of tests per thousand people was 4.1 tests and the total number of confirmed COVID-19 cases was about 145,000. If by that date the number of tests per thousand people in Indonesia was approximately 210 per thousand people instead, or about the same as those in the United States (210 tests) or United Kingdom (183 tests), it is estimated that the number of confirmed COVID-19 cases would be 560,000, approximately 4 times the actual confirmed cases. The estimated number means that 0.2 of the population is infected. It is not an overestimation if compared to the United States and the United Kingdom, which were approximately 1.7 and 0.5% of their populations, respectively. The number of COVID-19 cases provided by Resosudarmo and Irhamni (2021), hence, could indicate the true spread of COVID-19 in Indonesia.

224

B. P. Resosudarmo et al.

Table 2 Estimation results OLS (1) Total tests

Fixed effects (2)

0.908***

(3) 1.296***

(0.006)

(0.011) 0.700***

Total test-lagging 14 days

(4)

0.343***

(0.008)

(0.010)

With control variables

Yes

Yes

No

No

Date dummies

No

No

Yes

Yes

Country fixed effect

No

No

Yes

Yes

Observations

7,291

6,768

10,243

9,603

R-squared

0.882

0.824

0.890

0.688

# of country

55

59

83

104

Note *, **, and *** denotes statistical significance at 10%, 5%, and 1%. Numbers in the brackets are standard errors. Total tests are in million tests. Total cases are in thousand cases. Both total tests and cases are calculated based on their 7-day moving average numbers. Control variables are 2019 population, 2019 population density, 2019 portion of urban population, 2019 (PPP) current GDP per capita, 2017 diabetes prevalence, 2016 portion of smokers and 2017 physicians per thousand population Source Resosudarmo and Irhamni (2021)

No systematic data, however, is available yet at subnational level. A collection of subnational case studies in Resosudarmo et al. (2021) indicate that the spreads of COVID across subnational levels in Indonesia vary a great deal. There are regions in Indonesia, such as some remote districts in the Papua Province where no COVID-19 case was detected till end of August 2020. While the number of COVID-19 detected in Jakarta Province reached approximately 37 thousand by the end of August 2020.

4 Our Rapid Phone Survey 4.1 Survey Areas We conducted a phone survey to analyse the potential size of COVID-19’s spread in two of Indonesia’s major urban areas: Jakarta City and Yogyakarta Province. Jakarta is the capital City of the country and Yogyakarta is another large City in the centre of the Java Island. Table 3 provides general characteristics of Jakarta City and Yogyakarta Province. Although Yogyakarta Province covers a large area, its population is much smaller than Jakarta City. Jakarta City has a much highly dense population. Jakarta City is a much business-oriented place where living cost is among the highest in the country. Yogyakarta Province is a more student-oriented area where many schools have been established over there since the country became independent

Problems with Recording the Spread of COVID-19 in Developing …

225

Table 3 Characteristics of Jakarta City and Yogyakarta Province 2019 Population Population density Land area

Indonesia

Jakarta City

Yogyakarta Province

Unit

266.9

10.5

3.9

Million persons

0.1

15.9

1.2

Thousand persons/km2

1,919.4

0.7

3.2

Thousand km2

Source Statistics Indonesia

and living cost is among the lowest in the country. Social capital is expected to be stronger in Yogyakarta Province then than in Jakarta City. By 9 November 2020, there have been approximately 113 thousand formally reported cases and formally reported casualties reached approximately 2.4 thousand in Jakarta City. Death rate due to COVID-19 has been approximately 2.1% in this City; lower than the national situation of 3.3%. The PCR (polymerase chain reaction) test rate in Jakarta has been much higher than that in other places, covering approximately 127.6 thousand tests per million persons, while the rate is approximately 11.5 thousand for the national average. In Yogyakarta, by 9 November 2020, there have been approximately 4.3 thousand formally reported cases and formally reported casualties reached approximately 105. Death rate due to COVID-19 has been approximately 2.5% in this Province, which is lower than the national situation. The PCR test rate in Yogyakarta Province has been much lower than in Jakarta, but slightly higher than the national average.

4.2 Survey Implementation A small team was established at the Faculty of Economics and Business, Universitas Indonesia, in April 2020 to implement our survey. The survey was performed from 1 May to 30 June 2020 in Jakarta City and from 14 June to 17 August 2020 in Yogyakarta Province. Our sampling frames originated from the sampling frame of the 2018 flood survey and one of micro and small enterprises in the 2016 BPS Economics Census for the Jakarta region. As for Yogyakarta, the sampling frame used is only from the latter. These sampling frames also contains information on sample weights for each respondent constructed by the statistical agency of Indonesia (BPS), also known as Statistics Indonesia. We then randomly took as many as 4,600 and 3,600 respondents for Jakarta City and Yogyakarta Province, respectively as our sampling frames. The 15 survey enumerators for the Jakarta region were mainly undergraduate students from Universitas Indonesia (UI) and Universitas Negeri Jakarta (UNJ). While the 15 survey enumerators for Yogyakarta region were mainly undergraduate students from Universitas Gadjah Mada (UGM). Prior to conducting the survey, enumerators attended training sessions to master the survey questionnaire, interview

226

B. P. Resosudarmo et al.

mechanisms as well as survey mobile application. The training was performed online through Zoom facilities. We divided the questionnaire into nine sections. The first section aims to capture the characteristics of respondents with five items, including age, gender, size of household, and born in the City. The second section is the core section for measuring the potential size of the COVID-19 pandemic. It consists of seven question items, i.e., information about patients under surveillance, people under surveillance, and possible contracted household members. The third section seeks information about the impact of mental health during the pandemic, seen with the level of anxiety described in three simple questions. Sections four to seven contain social distancing behaviour, expenditure, food insecurity, and social assistance questions. Section eight, consisting of six questions, is the section that collects information on family members who died due to the pandemic. The last section aims to collect information on working status and characteristics before and after the pandemic occurred. In total, there were 40 questions that the respondents had to answer in an estimated 15-min interview. Computer-Assisted Personal Interviewing (CAPI) technology, the Survey Solutions developed by the World Bank, was used for the enumeration process. Together with the use of the telephone as a medium for conducting interviews, enumeration can be carried out more cheaper and easier than in-person field surveys. Especially during the pandemic, this approach reduces the risk of infection because the enumerators do not need to meet directly with respondents. Phone surveys also have advantages in terms of the cost because they require lower costs than field surveys. On average, approximately USD15 per fully interviewed respondent. It covers the cost of the enumerator and the cost of respondent time and phone use. Another advantage of using telephone surveys is that there are almost no costs for respondents who fail to be interviewed. Despite the convenience of phone surveys, this approach has a drawback of a low response rate. This is reflected in the small number of respondents who want to be interviewed compared to its target sample (Table 4). Out of 4,600 targeted samples in Jakarta City, only 342 received the call and agreed to be interviewed. 3,559 respondents could not be contacted, and 699 refused to be interviewed.4 High rejection rates also occurred in Yogyakarta Province. The rates are about 67 and 51% in Jakarta and Yogyakarta, respectively.5 The ultimate step in preparing our dataset is to create a frequency weight for our observations. The demography of our samples in both cities is quite comparable to the statistics obtained from the National Socioeconomic Survey (SUSENAS) 2020. See the numbers in the bracket in Table 4. The only minor differences are that our samples have a larger household size, a higher representation of females in Jakarta City and 4

The case is similar to Fuchs (2012)’s attrition issue in which he loses phone survey samples from 1,451 in 2009 to 208 in 2010. 5 For comparison with other survey in Indonesia, the response rate in LSI (2021) study is 16.05% (1,200 respondents out of 7,477 target) and the response rate in Komazawa et al. (2021) is 73.54% (2,574 respondents out of 3,500 target).

Problems with Recording the Spread of COVID-19 in Developing … Table 4 Enumeration results Original list

227

Jakarta City

Yogyakarta Province

4,600

3,600

Reasons Cannot be connected

3,559

2,794

Refused to be interviewed

699

409

Interviewed

342

397

Percentage of female (%)

42.7 [32.4]

43.1 [43.5]

Average age (year)

45.8 [44.7]

46.1 [49.3]

Average size of household (person)

4.6 [3.88]

4.3 [3.80]

Percentage of migrant household (%)

55.6 [54.8]

29.5 [23.7]

Characteristics

Source Authors’ own survey Note The numbers in the brackets are the population weighted statistics from the National Socioeconomic Survey (SUSENAS) March 2020 of household members who work in self-employment activities

migrants in Yogyakarta Province. Moreover, the respondent’s characteristics interviewed in both Jakarta City and Yogyakarta Province are similar. Judging from the percentage of female respondents, the average age of the respondents, and the average household size, the results are not much different between the two regions (Table 4). Regarding the characteristics, a vast difference between respondents in both regions is the percentage of migrant households, where the percentage of respondents from migrant households in Jakarta is 55.6%, whereas in Yogyakarta is only 29.5%. As with these characteristics, however, we do not know whether the respondents were successfully interviewed or not. The last step of preparing our dataset is to adjust our sampling weight considering they could not be contacted and rejected to be interviewed.

5 Tracing COVID-19 Cases and Impacts 5.1 Principle of the Heuristic Algorithm To estimate the potential number of COVID-19 cases, we use four questions related to ‘being exposed to the COVID-19’ or ‘having symptoms of COVID-19 infected’. The first question (Q1) looks at the presence of confirmed COVID-19 patients in the respondent’s family:

228

B. P. Resosudarmo et al.

Is there anyone, including yourself, in this house who is/was stated by a medical doctor as patient under surveillance (PDP) or a person stated by a medical doctor as ‘very possible to be infected by the Coronavirus’? If yes, how many of them?

The second question (Q2) sees people suspected of having COVID-19 in the respondent’s family: Is there anyone, including yourself, in this house who is/was stated by a medical doctor as people under surveillance or a person stated by a medical doctor as ‘possibly infected by the Coronavirus’? If yes, how many of them

Question three (Q3) identifies the emergence of COVID-19 symptoms in the respondent’s family: “Since March this year, is there anybody, including yourself, in this house who ever got illness with following symptoms: fever-high temperature or cough or breathing difficulties, but not being a patient under surveillance?” If yes, how many of you?

Meanwhile, the last question (Q4) asked whether the respondent has possible contact with a COVID-19 patient: Have you, or anybody living in your house, ever in contact with a person stated by a medical doctor as infected by Coronavirus?

We differentiate respondent’s status regarding COVID-19 infection into three, i.e., confirmed COVID-19, very possibly contracted COVID-19, and maybe contracted COVID-19 cases. The number of confirmed COVID-19 cases represents the accuracy of our phone survey in replicating government formal number of COVID-19 cases. Very possible contracted COVID-19 and maybe contracted COVID-19 cases are our predictions on the true number of COVID-19 cases in Jakarta City and Yogyakarta Province. A heuristic algorithm based on respondent’s answers on question one until question four is developed to determine the number of confirmed COVID-19, very possible contracted COVID-19, and maybe contracted COVID-19 cases, as follows: . Respondents are categorised into confirmed COVID-19 cases if their answer is “yes” to question one (Q1 = yes). We then calculate how family members are in this category. . Respondents are categorised into possible COVID-19 cases (excluding those who are confirmed infected) if they meet the following three conditions. First, people under surveillance who met patient COVID-19 but not the patient, when respondents answered “yes” on questions two and four but answered “no” on question one (Q2 = yes and Q4 = yes but Q1 = no). Second, people under surveillance with COVID-19 symptoms but not a patient, when respondents answered “yes” on questions two and three but answered “no” on question one (Q2 = yes and Q3 = yes but Q1 = no). Third, people with COVID-19 symptoms who met patient COVID-19 but not the patient, when respondents answered “yes” on questions three and four but answered “no” on question one (Q3 = yes and Q4 = yes but Q1 = no). We then calculate how many family members are in each of the above categories.

Problems with Recording the Spread of COVID-19 in Developing …

229

. Furthermore, if the respondent answered “yes” on question two or question three but answered “no” on question one, he was categorised as maybe contracted COVID-19 cases (Q2 = yes or Q3 = yes but Q1 = no). We then calculate how many family members are in this situation. The rules applied in the algorithm were developed based on several interviews conducted with medical doctors at the Universitas Indonesia and at several health centres in Depok and Jakarta cities. We run this heuristic algorithm on our datasets from our phone surveys in Jakarta City and Yogyakarta Province. We use our sample weight when calculating each prediction of COVID-19 cases.

5.2 Estimated Cases of COVID-19 and Discussion As announced by the Indonesian government by end of August 2020, the formal number of contracted COVID-19 cases in Jakarta City and Yogyakarta Province were approximately 0.34% of Jakarta City’s population and approximately 0.03% of Yogyakarta Province’s population. Meanwhile, in large cities in developed countries the number of COVID-19 tests announced was much larger than the population affected by COVID-19 by end of August 2020—i.e., 2.83% in New York City and 1.93% in Los Angeles (Our World in Data-University of Oxford). Information from these modern cities indicates that the number of COVID-19 infected cases announced by the Indonesian government might be lower than the actual number of COVID-19 cases in Jakarta City and Yogyakarta Province. The main reason for this is the small number of COVID-19 tests per population in the country. Table 5 depicts the estimated results of COVID-19 cases in Jakarta City and Yogyakarta Province using our phone survey. In Jakarta, the results predict that 0.009% of the population were confirmed positive for COVID-19, 0.5% were very possibly contracted COVID-19 cases, and 18.5% maybe contracted COVID-19 cases by July 2020. In Yogyakarta Province, our results show that 0.2% of the population were predicted as confirmed COVID-19 cases, 0.5% of very possible contracted COVID-19 cases and 19.7% maybe contracted COVID-19 cases. Important to note that the consistency between the numbers of very possible contracted COVID-19 and maybe contracted COVID-19 cases in Jakarta City and Table 5 Estimated cases of COVID-19 (in %)

Type of COVID-19 Jakarta City cases

Yogyakarta Province

Confirmed to be contracted

0.009

0.233

Very possible contracted

0.498

0.536

18.508

19.679

Maybe contracted

Source Authors’ owned calculation

230

B. P. Resosudarmo et al.

Yogyakarta Province. By also considering the numbers in modern cities such as New York and Los Angeles, we can expect that the true size of COVID-19 cases in Jakarta City and Yogyakarta Province would be 0.5% of their population by July– August 2020 (about 6 months after the first case found). For Jakarta City, this prediction would be almost twice the number announced by the government; whereas for Yogyakarta Province, this prediction would be more than ten times the government’s number. On the other hand, our predictions on confirmed COVID-19 cases were not able to trace government’s announced number of confirmed COVID-19 cases in both Jakarta City and Yogyakarta Province. Our number is too low in Jakarta City, but too high in Yogyakarta Province. These results indicate the low, but rather varied, quality of a phone survey. The first reason for this low but varied reliability of a phone survey is due to the large rejection and uncontactable cases. The second reason is that people are less likely to reveal their true condition if they were positive COVID-19. There has been a negative social stigma on those infected (WHO, 2020).

5.3 Impacts of the Pandemic Aside from its reliability and its short length, phone surveys could gather some particularly useful information in detecting consequences of COVID-19 pandemic on societies. In our phone survey, we collected some of this information. First, our phone survey also tries to estimate the impact of income changes felt by households during the pandemic. We asked for changes in average income before and after the COVID-19 pandemic emerged. Households in Jakarta and Yogyakarta reported a decrease in income or no change in income. In both regions, the median change in income experienced by households is an income decrease of 0.5–2 million rupiahs (Fig. 5). Slightly a bit more than 10% of the population in both cities reported that their incomes do not change due to the pandemic and few of them claimed an increase in income. Jakarta City

Yogyakarta Province

Fig. 5 Impacts of COVID-19 pandemic on household income. Source Authors’ own calculation

Problems with Recording the Spread of COVID-19 in Developing … Jakarta City

231

Yogyakarta Province

Fig. 6 Household food insecurity experience. Source Authors’ owned calculation

Considering the average household incomes in Jakarta City and Yogyakarta Province, these changes in household income before and after the COVID-19 pandemic occurred seem to be reasonable. During the pandemic, Indonesian people tend to be more open when providing income reduction information but could be reluctant in providing precise information regarding COVID-19 infection. Second, it is interesting that our phone survey managed to collect information related to the food insecurity impact of the pandemic. We asked whether due to the pandemic respondents had to eat less than previous week before being interviewed. Our survey predicts that more than 40% of Jakarta City’s population, in various degrees, had to eat less in the previous week before being interviewed due to the pandemic. The number is slightly lower in Yogyakarta (Fig. 6). The results related to food insecurity experiences are consistent with the predictions of household income reduction. Hence, the information would be reliable and could indicate the severeness of the pandemic impacts on the society. The third information collected by our phone survey is whether respondents feel disturbed due to the COVID-19 pandemic. Our results indicate that the majority of population in Jakarta City and Yogyakarta Province feel somewhat or very disturbed due to the pandemic. The concern is that the proportion of population that feels very disturbed by the pandemic were more than 30% in Jakarta City and more than 20% in Yogyakarta Province. If these numbers are reliable, this indicates the seriousness of mental health issues caused by the pandemic in those cities (Fig. 7). This information—which is related to the impacts of the pandemic on household incomes, food insecurity experience and feeling disturbed by the pandemic— is certainly useful for supporting countries’ governments in developing policies softening the impact of pandemic on their population.

232

B. P. Resosudarmo et al. Jakarta City

Yogyakarta Province

Fig. 7 Feeling disturbed due the COVID-19 pandemic. Source Authors’ owned calculation

6 Sensitivity Analysis on the Estimated Cases of COVID-19 To understand whether the results of our phone survey is robust, we conduct a sensitivity analysis toward the estimated cases of COVID-19 from this survey. The procedure for conducting the sensitivity analysis is as follows. First, we randomly take as many as 20% of the total observations in our data. Second, we add these randomly selected observations into our original data. Then, we compute the heuristic procedure in estimating cases of COVID-19 as mentioned in Sect. 5.2. We iterate this process for five hundred times. The results are as can be seen in Table 6. As shown in Table 6, the 95% confident intervals for all estimations—i.e., Confirmed COVID-19 cases, Very possible contracted COVID-19 cases, and Maybe contracted COVID-19 cases—are relatively narrow. This indicates that the estimates are robust. Table 6 Estimated cases of COVID-19 from our phone survey (in %) Type of COVID-19 cases

Mean

[95% confident interval]

Jakarta City Confirmed COVID-19 cases

0.009

0.008

0.009

0.492

0.478

0.506

18.498

18.463

18.534

Confirmed COVID-19 cases

0.234

0.230

0.239

Very possible contracted COVID-19 cases

0.539

0.526

0.552

19.679

19.645

19.713

Very possible contracted COVID-19 cases Maybe contracted COVID-19 cases Yogyakarta Province

Maybe contracted COVID-19 cases Source Authors’ own calculation

Problems with Recording the Spread of COVID-19 in Developing …

233

7 Conclusion Due to limited health facilities and fiscal capabilities, developing countries typically have serious issues managing an outbreak of a pandemic. In this case, the fundamental issue is their inability to conduct enough tests, let alone implementing proper tracing and treatment, as part of the proper 3T (testing, trace, and treat) system to control the pandemic. Government formal announcements about the spread of a pandemic would therefore be not accurate. While the econometric approach as conducted by Resosudarmo and Irhamni (2021) might provide a reliable estimate of the magnitude of the pandemic, systematic data at subnational level are not available. Digital information collection through mobile phone applications and rapid phone survey seems two alternative methods in obtaining quick data on estimating the spread and magnitude of a pandemic in a developing country, where a robust public health testing, trace, and treatment system is not available. These two methods are inexpensive, mainly because there is no need for transportation and accommodation costs to obtain the information. However, the main issue with digital information collection platforms is that participants providing information could be self-selected, resulting in bias or incomplete information. Samples or participants of a phone survey, on the other hand, theoretically could be purely random representing a defined population. Statistically, a phone survey could be more reliable than a digital information collection platform. However, how reliable the results from a phone survey remain debatable. By conducting our own phone survey in Jakarta City and Yogyakarta Province, this paper is observing a reliability and usefulness of a phone survey approach directly asking respondents whether they are infected or sick and the impacts of the pandemic on their livelihoods. The main results of our observation would be as follows. First, on the issue of providing a better estimate on the magnitude of a pandemic, with developing a careful algorithm, a phone survey might be able to provide a better estimate than those announced by the government in developing countries where capabilities of conducting 3T system are limited. Furthermore, the estimate of the magnitude of a pandemic could also be relatively robust. For example, our phone survey reveals that the magnitude of the pandemic in Jakarta City and Yogyakarta Province could be as big as infecting 0.5% of their population by July–August 2020 (about six months after the first case was found). Second, the reliability of a phone survey in predicting the magnitude of a pandemic, however, could be low but varied. To resolve this issue, proper strategies to significantly reduce cases of rejection and uncontactable respondents and to ensure that respondents honestly reveal their situation related to being infected by COVID-19 should be implemented. Third, phone surveys could be used to collect important relatively reliable information related to the impact of the pandemic on societies’ livelihoods. This information could be very useful to develop policies for softening the impact of the pandemic. Our phone survey indicates that median population in Jakarta City and Yogyakarta Province experienced a decline in their incomes by around 0.5–2 million

234

B. P. Resosudarmo et al.

rupiahs. More than 40% of Jakarta City’s population, in various degrees, have ever to eat less last week, before being interviewed, due to the pandemic. The number is slightly lower in Yogyakarta. Finally, our phone survey discloses a possibility of mental health issue due to the pandemic among the population in Jakarta City and Yogyakarta Province. The overall conclusion of this paper is that private agencies/individuals should be encouraged to conduct proper phone surveys to provide alternative better information on the status of a pandemic and its livelihood impact. Furthermore, it is important to encourage these private agencies/individuals to make the data they collected publicly available. Acknowledgements The authors would like to acknowledge the support from the Arndt-Corden Department of Economics at the Australian National University for providing some funding to conduct the phone survey and from the Statistics Indonesia for the list of phone numbers in Jakarta and Yogyakarta. We also would like to thank the two reviewers for their suggestions and comments. All mistakes are, nevertheless, the authors’ responsibility.

References Badan Pusat Statistik (BPS). (2020). Prilaku Masyarakat di Masa Pandemi COVID-19: Hasil survey 7–14 September 2020. BPS. Baldwin, R., & di Mauro, B. W. (2020). Mitigating the COVID economic crisis: Act fast and do whatever it takes. Centre for Economic Policy Research (Issue July). https://voxeu.org/content/ mitigating-covid-economic-crisis-act-fast-and-do-whatever-it-takes Barro, R. J., Ursua, J. F., & Weng, J. (2021). The Coronavirus and the great influenza epidemic— Lessons from the “Spanish Flu” for the Coronavirus’s potential effects on mortality and economic activity. CESifo working paper no. 8166. https://doi.org/10.2139/ssrn.3556305 Busse, B., & Fuchs, M. (2012). The components of landline telephone survey coverage bias. The relative importance of no-phone and mobile-only populations. Quality & Quantity, 46(4), 1209– 1225. Chandra, S. (2013). Mortality from the influenza pandemic of 1918–19 in Indonesia. Population Studies, 67(2), 185–193. https://doi.org/10.1080/00324728.2012.754486 Dillon, B. (2012). Using mobile phones to collect panel data in developing countries. Journal of International Development, 24(4), 518–527. Falentina, A. T., Resosudarmo, B. P., Darmawan, D., & Sulistyaningrum, E. (2021). Digitalisation and the performance of micro and small enterprises in Yogyakarta, Indonesia. Bulletin of Indonesian Economic Studies, 57(3), 343–369. https://doi.org/10.1080/00074918.2020.1803210 Fuchs, M. (2012). Nonresponse and panel attrition in a mobile phone panel survey. Paper presented at the Federal Committee on Statistical Methodology Research Conference in Washington, USA, 10–12 January. https://nces.ed.gov/FCSM/pdf/Fuchs_2012FCSM_III-B.pdf Halim, D., Can, E. R., & Perova, E. (2020). What factors exacerbate and mitigate the risk of gender-based violence during COVID-19?: Insights from a phone survey in Indonesia. World Bank. https://openknowledge.worldbank.org/bitstream/handle/10986/35007/What-Factors-Exa cerbate-and-Mitigate-the-Risk-of-Gender-Based-Violence-During-COVID-19-Insights-From-aPhone-Survey-in-Indonesia.pdf?sequence=1&isAllowed=y Herrero, L., & Madzokere, E. (2021). COVID will likely shift from pandemic to endemic—But what does that mean? The Convesation.Com, 20 September.

Problems with Recording the Spread of COVID-19 in Developing …

235

Komazawa, O., Suriastini, N. W., Wijayanti, I. Y., & Maliki, K. D. (2021). Older people and COVID19 in Indonesia. BAPPENAS. https://www.eria.org/research/older-people-and-covid-19-in-ind onesia/ Kompas. (2020). Standing together against COVID-19. Kompas, 03 March 2020. https://www.kom pas.id/baca/english/2020/03/03/standing-together-against-covid-19 Lembaga Survei Indonesia (LSI). (2021). Rilis hasil survei nasional: Sikap publik terhadap vaksin dan program vaksin pemerintah. LSI. https://www.lsi.or.id/post/rilis-survei-lsi-terkait-vaksin-18juli-2021 Mietzner, M. (2020). Populist anti-scientism, religious polarisation, and institutionalised corruption: How Indonesia’s democratic decline shaped its COVID-19 response. Journal of Current Southeast Asian Affairs, 39(2), 227–249. https://doi.org/10.1177/1868103420935561 Olivia, S., Gibson, J., & Nasrudin, R. (2020). Indonesia in the time of Covid-19. Bulletin of Indonesian Economic Studies, 56(2), 143–174. https://doi.org/10.1080/00074918.2020.1798581 Purnamasari, R., & Sjahrir, B. S. (2020). High-frequency monitoring of socio-economic impact of COVID-19 on households in Indonesia. In: ANU Indonesia Project Global Seminar Series (Vol. 30). https://www.bps.go.id/publication/2021/08/02/29234b08faa4910dee5279af/perilakumasyarakat-pada-masa-ppkm-darurat--hasil-survei-perilaku-masyarakat--pada-masa-pandemicovid-19--periode-13 Resosudarmo, B. P., & Irhamni, M. (2021). Consequences of the COVID-19 pandemic on human capital development. In B. Lewis & F. Witoelar (Eds.), Economic Dimensions of Covid-19 in Indonesia (pp. 170–189). https://doi.org/10.1355/9789814951463-013 Resosudarmo, B. P., Mulyaningsih, T., Priyarsono, D. S., Pratomo, D., & Yusuf, A. (Eds.). (2021). Regional perspectives of COVID-19 in Indonesia. IRSA Press. Safdar, N., Abbo, L. M., Knobloch, M. J., & Seo, S. K. (2016). Research methods in healthcare epidemiology: Survey and qualitative research. Infection Control & Hospital Epidemiology, 37(11), 1272–1277. Solanki, H. K., Gopal, P. G., & Rath, R. S. (2020). Common pitfalls in using online platforms for data collection in COVID times and its implications. Nepal Journal of Epidemiology, 10(4), 930–932. https://doi.org/10.3126/nje.v10i4.31614 Vehovar, V., Berzelak, N., & Manfreda, K. L. (2010). Mobile phones in an environment of competing survey modes: Applying metric for evaluation of costs and errors. Social Science Computer Review, 28(3), 303–318. WHO. (2020). Social stigma associated with COVID-19: A guide to preventing and addressing (Issue February). https://www.who.int/docs/default-source/coronaviruse/covid19-stigma-guide. pdf Wolter, K. M., Smith, P., & Blumberg, S. J. (2010). Statistical foundations of cell-phone surveys. Survey Methodology, 36(2), 203–215. Wu, J., Wang, J., Nicholas, S., Maitland, E., & Fan, Q. (2020). Application of big data technology for COVID-19 prevention and control in China: Lessons and recommendations. Journal of Medical Internet Research, 22(10), e21980. https://doi.org/10.2196/21980

Pandemic Regional Recovery Index: An Adaptable Tool for Decision-Making on Regions J. Irving , K. Waters , T. Clower , and W. Rifkin

1 Introduction The COVID-19 pandemic seemed to bring the world to a halt in early 2020. As the virus spread and the impacts were being experienced firsthand across the world, governments instituted lockdowns to control the spread of the virus and keep their populations safe. The side effect of these lockdowns, however, was the grinding halt of much economic activity. Though many industries adapted quickly (online retail sales, professional services), others continued to struggle. These differences in industry performance impact regional economies differently. Agglomeration ensures that clusters of economic activity form, benefiting from co-location. These agglomeration economies often result in regions becoming specialized in a handful of industries. In economically good times, the co-location of economic activity and the ensuing specializations can result in productivity gains. These productivity gains can make communities in these regions more economically competitive. However, in recent years, it has become painfully obvious that a disaster, such as a hurricane or wildfire, or the recent pandemic, can show the flip side of industrial co-location and specialization. Those communities that are the most reliant on economic activity that is heavily impacted by a given disruption—such as travel and entertainment—have endured the worst economic consequences, and that compounds the human costs of the pandemic. Despite the massive disruption brought by the virus, the world appears to be on course to move into a recovery phase. That is shown in Fig. 1. Substantial downturns J. Irving University of South Australia, UniSA Business, Mawson Lakes, SA 5000, Australia K. Waters (B) · T. Clower Schar School of Policy and Government, George Mason University, Fairfax, VA, USA e-mail: [email protected] W. Rifkin College of Human and Social Futures, University of Newcastle, Newcastle, NSW, Australia © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. G. Celbi¸s et al. (eds.), Pandemic and the City, Footprints of Regional Science, https://doi.org/10.1007/978-3-031-21983-2_13

237

238

J. Irving et al.

Fig. 1 Changes in GDP across 6 countries and the OECD (Source OECD, 2022)

in gross domestic product (GDP) were seen in the first two quarters of 2020. That was followed by rises in the third quarter, as industries adjusted and government expenditures started to get traction. By the fourth quarter of 2020, shifts in GDP moderated in some but not all countries that we analysed at a regional level for this chapter. Even as economic activity appears to head toward previous levels at a national scale, the damage for regions and communities that rely on certain dominant industries could well last for years. To mitigate such long-term effects, it is useful to get a sense, early during a disruption, of where policy attention should focus. COVID19 has shown that policy makers have been hitherto unequipped to understand their regional resilience capacities. To help them, we acknowledge that for tools to be most effective, they need to be bespoke and mature models. However, such models need significant resources—and time to develop—beyond the scope of any rapid response tool. That time and resources were not available to regional economists in the midst of the COVID pandemic, and that is likely to be the case in other such disruptions. One is reminded of the story of the young Californian couple looking for dinner in an isolated town in rural Alaska that had just 12 inhabitants. They asked the innkeeper what was on the menu for dinner. The innkeeper said, “Steak.” The couple inquired about less meat-focused options. The innkeeper looked at them incredulously and replied that it was steak that is what he had thawed. In this sense of limited resources, the model presented here suggests how to cook that steak. It is not the best dinner that a regional economist could cook, but it is what is on the menu given what is readily at hand. In this sense, we have aimed to capture the strength of a region’s local economic structure in positioning it to recover from the COVID-19 pandemic as elaborated in our methodology.

Pandemic Regional Recovery Index: An Adaptable Tool …

239

To help to identify those regions that might need additional government help for economic recovery, six countries were investigated in this study. The goal was to build a tool to help policy makers guide resources to the communities as impacts are becoming evident in order to help those that are likely to struggle to recover due to their industry mix. By use of the term “tool”, we are referring to the suite of data that was found to be available on a reasonably consistent basis across several countries, the focus on a limited number of salient inputs to enable rapid modelling, and use of professional experience in assessing key sectors, media coverage, and certain data sets that governments make available in the near term on regional outcomes (such as monthly unemployment figures) to assemble, assess and refine the specific model for the given disruption. As a tool, like an adjustable spanner, the model offered here is meant to be adapted and adjusted. The application of this COVID tool in 6 countries was undertaken to highlight possible commonalities and differences across developed nations. Discussion is provided here of regions that seem more vulnerable or less vulnerable according to the index calculations with some confidence in relation to the authors’ countries— the US and Australia. Mention of regional variations in other countries analysed is necessarily cursory at this point, suggestive rather than definitive. Other chapters in this volume provide more in-depth stories and analytical results on specific cities and regions. One can imagine that further analysis of how regions fared in the countries that we looked at will continue to emerge, in a similar way to how studies of the longterm effects of the Global Financial Crisis continue to surface. The reader is welcome to assess the accuracy of our initial assessment of their country and is invited to adapt the index based on their expert insight into the economy of the country or countries that they study. Comparing the tool’s findings to case-specific results suggests where novel policies, business strategies or unique economic structures can make a meaningful difference. At a higher level, development of this COVID-focused economic policy tool suggests how such tools can be developed and quickly adapted for other types of future global disruptions. While the index presented here has been developed to be specific to COVID’s economic impacts where the policy responses involved social isolation, the general approach—of using a limited slice of data and insight into economic impacts from the specific type of disruption—can be generalised. Other types of disruptions, like the heat wave and drought in Europe in the summer of 2022 or flooding in Australia during the southern autumn of 2022, would require use of other factors. For example, a heat wave could be modelled as having greater impacts on industries related to use of electricity due to heightened demand for air conditioning, with flooding disruption being relevant to industries in the agricultural supply chain and construction. In sum, this chapter is not presenting a definitive analysis of COVID’s impact on different regions in 6 different countries. Rather, it is illustrating how to undertake a relatively rapid, iterative analysis based on readily available data combined with insight into the nature of regional economies of a target country.

240

J. Irving et al.

The literature review that follows addresses regional resilience, generally, such as in relation to natural disasters or the Global Financial Crisis, before focusing specifically on pandemic resilience and recovery. We then provide a brief description of the data used and challenges related to data consistency across countries, specifically related to spatial granularity. Following that is a discussion of the metric employed as the recovery assessment tool. We then describe the application of the tool in the United States and reveal how it tested against timely unemployment data. Application of the tool to each of 5 additional countries is then discussed. This discussion focuses on regions with the highest and lowest index figures to enable an initial assessment of the extent to which the model is making sense. Finally, we draw some conclusions about how the COVID situation differs from more typical natural disasters and from the Global Financial Crisis, and we suggest that this pandemic index points to a need for flexible assessment tools that can be developed and refined as impacts unfold.

2 Literature Review An overview of studies on regional resilience, including studies centred around disasters and the 2008 Global Financial Crisis (GFC), is usefully supplemented by a review of emerging research on economic vulnerability and economic recovery from the COVID-19 pandemic. This later set of studies, predominantly published in 2020 and 2021, seem to focus on identifying what regions or which industries have been impacted and what policy levers might speed economic recovery. These various studies tend to differ in scope—national versus regional—and in terms of what factors are analysed and at what level of granularity—either by geography or industry sub-sector.

2.1 Regional Resilience The resilience of regions has become a greater preoccupation of regional studies since the onset of the GFC in 2007 (Stanickova & Melecký, 2018). A region’s resilience to natural disasters (see Cutter et al., 2010; Manyena et al., 2019) has been studied along with economic crises such as the GFC (see Giannakis & Bruggeman, 2017; Rose, 2017) and more recent health crises, such as the COVID-19 global pandemic (see Brada et al., 2021; Menoni & Schwarze, 2020). The global effects of such events can be seen to have increased over the years as national market interdependencies are at historical highs, which makes regional economies more exposed than ever to external events and disturbances (Xiao et al., 2018). While regional resilience has received renewed interest, it remains relatively underdeveloped (Boschma, 2015) with no definitive methodology to measure it, nor understanding as to what its determinants are (Stanickova & Melecký, 2018). Broadly, however, the concepts of “engineering resilience”, “ecological resilience”,

Pandemic Regional Recovery Index: An Adaptable Tool …

241

and “adaptive resilience” are well understood as described by Martin and Sunley (2015). Engineering resilience is a system’s ability to return to a theoretical equilibrium after a deviation from that equilibrium. Ecological resilience is a system’s ability to absorb pressure while remaining qualitatively unchanged. Adaptive resilience is a system’s ability to adapt to external forces in order for the system to survive in some manner. Additionally, evolutionary economic geography frameworks for examining regional resilience have gained prominence in the last decade or so (see Boschma, 2015; Pike et al., 2010) including coupling with concepts of social resilience (Argent, 2021) and institutional resilience (Halonen, 2019). Rather than understanding resilience in terms of equilibria or simply adapting to external forces, evolutionary frameworks perceive resilience as an ongoing capacity for economic path creation (Boschma, 2015), that is, its ability to evolve and survive long into the future. For the purpose of this literature review, a two-part categorisation of resilience will be employed to be consistent with research published since the onset of COVID19. The concepts will be referred to as ‘absorptive’ and ‘adaptive’ capacities or in terms of ‘vulnerability’ and ‘recovery’, respectively. Absorptive capacities in the early COVID-19 literature refers to a mix of engineering and ecological resilience, while adaptive capacities refer to a mix of adaptive resilience and some concepts of long-term prosperity as suggested in an evolutionary framework. The significant narrative that emerges from these two concepts throughout the resilience literature is that the tools that are developed to determine regional resilience are being variously adapted across scenarios and circumstances. What is increasingly consistent is a focus on informing wholistic policy to bolster regional economies in the face of economic shocks borne from a variety of crises. The tools employed across resilience studies, welcome as they are, should also be approached with caution, the literature suggests. It notes such analyses are perhaps better viewed as presenting tools to inform and facilitate discussion and engage communities rather than dictate policy remedies (Melecký, 2017; Melecký & Stanickova, 2015b in Stanickova & Melecký, 2018).

2.1.1

Disaster Resilience

Within disaster resilience literature the permutations of resilience have evolved in recent years. They range from a focus on a return to previous conditions, to more current conceptions of resilience as being adaptive, the ability to transform to new conditions (see Manyena et al., 2019; Robinson & Carson, 2016). This recent literature lends itself more acutely to our development of the recovery index presented in this chapter. For example, the framework employed for COVID-19 resilience by Drzeniek et al. (2020) bases part of its index on a region’s absorptive capacity as well as its adaptive economic capacity to recover from the pandemic. This framework follows the example set by Manyena et al. (2019) in the field of disaster resilience to examine both of these capacities.

242

J. Irving et al.

In an earlier paper by Cutter et al. (2010), considerations toward impediments to rebound are also explored. They developed a composite index in their study using indicators for social, economic, institutional, and infrastructural resilience, as well as community capital. The first three of these indicators are very much reflected in the recent work on COVID-19 resilience, with the other two expressed to a lesser extent (see Davradakis et al., 2020; Deloitte, 2021; Drzeniek et al., 2020; Pemandu Associates, 2021; Zandi et al., 2020). Other types of crises, aside from natural disasters—economic crises such as the Global Financial Crisis, or health crises such as the current COVID-19 pandemic— have received significant attention only in more recent times. This prior lack of attention is consistent with Rose’s (2017) assertion that the bulk of literature on regional resilience has mostly been concerned with a region’s capacity to withstand the economic shocks associated with natural disasters and that work is wanting to make it more relevant to regional studies as a whole and a ‘broader set of future disasters.’ The likely long-term impacts of shutting down entire sectors of the economy during the pandemic, and the recent memory of the GFC from which many countries were just emerging, suggest that the next generation of resilience literature should incorporate assessments of risk and planning for economic resilience when faced with an array of future global crises (Djalante et al., 2020; Menoni & Schwarze, 2020; Nicola et al., 2020).

2.1.2

The 2007–2009 Global Financial Crisis

The GFC is recognised as the worst global economic crisis since the Great Depression. In no less than a decade after countries began to emerge from their GFC recessions, the global economy experienced yet another unprecedented shock from the COVID-19 health crisis. While these economic downturns differ in their causes, their similarities are notable, particularly in relation to their global extent geographically and the number of industry sectors affected. This global influence suggests that useful insights for development of a COVID-recovery index can be gained from regional resilience literature on the GFC. The relationship between regional resilience to the GFC and their economic structures is examined by Xiao et al. (2018). They measure resilience in terms of a maintained or boosted rate of the entry of new industry specializations before, during and after the onset of the crisis. Their methodology builds on the literature of related and unrelated varieties of industries (Boschma, 2017), which have respectively been linked positively to economic growth and absorptive capacity for external economic disturbances (Frenken et al., 2007). Based on their analysis of European sub-regions, Xiao et al. (2018) find that related and unrelated variety have a positive effect on regional ability to withstand shock and maintain or increase entries into new industryspecializations, that is, to have been resilient to the external shocks of the GFC. A more recent study by Brada et al. (2021) develops a COVID-19 resilience index informed by the employment patterns during recovery of non-agricultural industries

Pandemic Regional Recovery Index: An Adaptable Tool …

243

from the GFC. Their work simulates these employment patterns in a COVID-19 recovery of Central and Eastern European countries, at the regional level. They conclude that, similar to the GFC, COVID-19 economic impacts will be long lasting even if some areas experience a quicker recovery. For our development of the pandemic recovery index, we decided not to feature related and unrelated variety, as cited in this literature, in the interests of speed, simplicity, and availability of data across our target OECD countries (obtaining consistent data has been a challenge nonetheless). For similar reasons, we have not undertaken to estimate the length of time for regions to demonstrate a recovery from the COVID-19 pandemic. Notwithstanding, we aim to capture the strength of a region’s local economic structure in positioning it to recover from the COVID-19 pandemic through means elucidated below, in the presentation of our methodology. A brief look at the frameworks presented within the regional resilience literature illustrates a range of resilience indicators, an array of baseline and recovery measures, and approaches to view resilience as a combination of absorptive and adaptive capacities.

2.2 Review of COVID-19 Indices Recent literature on the economic effects of COVID-19 deserves attention, particularly where it has an index component and includes analysis of economic indicators. We have searched for studies using the key terms pandemic, index, economic, recovery, vulnerability and COVID-19. As the pandemic is a global disaster with unique economic impacts, drawing from studies that looked at economic resilience to this crisis in particular was considered important even though they lack the academic peer review of the research above. The review of these studies is explored as ‘vulnerability’—absorptive capacities—and ‘recovery’—adaptive capacities—of regions.

2.2.1

Vulnerability

Economic vulnerability, or absorptive resilience, of a country to COVID-19 was considered by Davradakis et al. (2020) to be affected by the country’s economic structure and its exposure to and ability to respond to shock. Dependencies on global value chains, tourism, remittances and fuel, metal and ore exports as a percent of their GDP are used as sample indicators of a country’s degree of economic diversity by Davradakis et al. (2020). Remittances are seen as a particularly important consideration as they have been used in the past to help economies respond to previous crises (Davradakis et al., 2020). Their analysis suggested that low-income economies, like Afghanistan and Barbados, would experience a significant impact but that high-income countries, like Australia and South Korea, would be less affected.

244

J. Irving et al.

The vulnerability index produced by Davradakis et al. (2020) is comparative between nations and as such their sample indicators centre on international economic measures. These measures, however, do not find themselves repeated in regional analysis within a single country. Baum and Mitchell (2020), for example, examined vulnerability within Australia at the regional level. They used different sample indicators based on regional labour economics, specifically looking at employment vulnerability. The indicators used are: • the proportion of people employed in construction, mining, manufacturing, selected retail, selected wholesale, accommodation and food services, air transport, financial services, property operations and real estate operations; • the proportion of people without post-school qualifications; and • the proportion of employed persons working part time. Exploring the differences between the sets of sample indicators used by the two studies highlights the importance of considering the scale and scope of the research. Davradakis et al. (2020) have produced their research to highlight countries on a global scale that will be in need of assistance from the European Investment Bank. On the other hand, Baum and Mitchell’s (2020) research is focussed on providing insight for local governments. Because of these fundamental differences in intent, the former’s set of indicators is found to be inadequate on a national scale because it ignores regional differences within countries. At the same time, the latter’s indicators are too narrowly constrained to the key characteristics of the Australian labour pool, without much effort evident on how these factors might be considered in an international comparison.

2.2.2

Recovery

Most research on recovery reviewed here determines the measure of ‘recovery’ as a return to baseline metrics prior to COVID-19 (see Deloitte, 2021; Morelli, 2020; Pemandu Associates, 2021; Zandi et al., 2020). Alternatively, Drzeniek et al. (2020) investigates how countries can thrive after the virus is contained, which is very similar to the measure of resilience proposed by Xiao et al. (2018), who included emerging growth paths following the GFC as an essential component of their analysis. These two approaches set out an interesting distinction, which affects the spatial scales and types of economic indicators used in these studies. The studies that look at recovery as a return of a selection of indicators to preCOVID-19 levels provide useful insight into the optimum timing for governments to transition from strategies aimed at mitigating negative impacts to strategies for (re-)growth (Deloitte, 2021; Morelli, 2020; Pemandu Associates, 2021; Zandi et al., 2020). Generally, though, these studies do not suggest what government growth strategies might actually be effective. Because of the diagnostic trend of this research, some emerging similarities exist among the studies:

Pandemic Regional Recovery Index: An Adaptable Tool …

245

– The geographic scales are large, with the exception of Morelli (2020); – The values of indicators are continuously changing; and – The research does not explore how well a jurisdiction might rebound. However, studies that explore the preconditions for a post COVID-19 recovery, that is, adaptive resilience, do offer prescriptive elements to their research. Drzeniek et al. (2020) explore the measures of the size, strength, and adaptability of a country’s labour market, the presence of a digital economy, education, financial system robustness, and the strength of health systems as indicators of both absorptive and adaptive resilience—that is, their ability to cope and rebound from COVID-19. These measures are different to the other research cited here by their inherent goal to identify those countries with the capacity to rebound, not just a measure of who is recovering. It is this feature of the work of Drzeniek et al. (2020) that enables it to support the production of recommendations to governments on strategies for recovery and rebuilding such as their recommendations to target digital skills in the workforce and information technology infrastructure (p. 22). These recent studies have been done with a relatively fast turnaround in order to influence policy and strategy responses in 2020, rather than longer-term plans to develop resilience to future shocks. These practical considerations mean that the studies have been shaped by the availability of data; there had not been time to survey and generate new sources of data, particularly not at scale. This constraint has also guided our own analysis.

2.3 Granularity of Data Which level of jurisdiction is employed in vulnerability and recovery indices is not just a factor of who the intended audience is, but is influenced by limitations of the scale of data that is readily available and data that is common across all selected jurisdictions. However, commonality of regional data types and granularity may be seen readily within a single country, or continent in the case of Europe, and it is increasingly seen where efforts in that direction have occurred—in the OECD and European Union. The diagnostic research on recovery that is discussed earlier tends to use large scale geographical and economic data because that is what is readily available when aiming to provide a method and analysis that can be applied at the state or national level. This scale is understandable as the financial or regulatory resources required to have a significant impact in a sector or region are more likely available at the national level. As the targeted geographical scale becomes smaller, the sample indicators become more idiosyncratic as in the case of Morelli’s (2020) case study of NYC. This specificity results in a methodology that is not necessarily transferable to other jurisdictions as data sets available in NYC tend to differ from those available for mega cities in other countries.

246

J. Irving et al.

Less demanding levels of data granularity may allow one’s net to be cast wide, but it also provides little insight to finer administrative levels in the regions being assessed. That is, the broader jurisdiction’s recovery—or capacity for recovery—may differ from that of one of its sub-regions. By the same token, aiming for a greater level of granularity across relatively small geographical areas can be difficult due to the lack of data. Using a finer geography can also lead to issues in comparative studies because of confounding jurisdictional effects within states and countries. In other words, comparing regions from one country to a region in another country can lead to misleading conclusions about their resilience with respect to each other. Baum and Mitchell (2020) address these two issues by (a) using historical data and (b) comparing regions only within Australia. Using historical data limits the capacity of the research to use emerging metrics from the period of the COVID-19 pandemic. However, it enables utilising granular and reliable data, such as from census counts. Secondly, by only comparing regions within the same country and not regions across different countries, the authors avoid having to account for any jurisdictional effects when transferring their model, such as a reliance on overseas remittances—a key economic indicator for the vulnerability used by Davradakis et al. (2020), which would have effects on regional capacities to recover. Zandi et al. (2020) similarly pursue a finer comparative analysis than other studies by only looking at the U.S. and its constituent states. Zandi et al., however, use up-to-date sample indicators, which would prevent them from looking at a finer geographical level. For example, they employ data on the number of seated restaurant diners from OpenTable and Unemployment Insurance Weekly Claims from the U.S. Department of Labor, which are not available at the regional level with the exception of selected cities. While having readily available indicators can be convenient, there is necessarily a trade-off that Zandi et al. (2020) makes in that their index does not provide any insight into which regions within states may be closer to a recovery to pre-COVID-19 levels. Certain studies like Morelli (2020) and Zandi et al. (2020) employ several economic indicators like the number of seated restaurant diners as proxies for recovery to indicate that society has gone back to ‘normal’ (Zandi et al., 2020). However, the world has experienced a global shift in its social, industrial and economic landscape, so what ‘normal’ is has to be understood differently and reassessed; what a recovered region is may indeed not be reflected by a return of one or two indicators to pre-COVID-19 levels. Several changes that represent this new normal can be observed regardless of what level of granularity is employed in analysis. For example, the transitions that regions are undergoing and have undergone in regard to shifts in market demands (Van Huellen & Asante-Poku, 2020) and confidence in global value chains (Gao & Ren, 2020) are relatively predictable due to the mobility stasis caused by the pandemic. That said, general trends are not necessarily telling us about the shifts in—or needs of—particular regions relative to other regions. Such considerations from the literature on regional resilience and response to the COVID-19 pandemic have helped to guide our focus on the development of a tool that can be employed in multiple countries. One aim was to provide rapid turnaround in

Pandemic Regional Recovery Index: An Adaptable Tool …

247

the analysis by relying on existing data sets, as noted above. Nevertheless, the regional scale of resolution was essential, with national, state and regional governments being the target audiences. The focus on rapid turnaround limited which aspects of industry employment could be scrutinised across multiple countries, due to data availability and consistency. These constraints underline the importance of being flexible in developing the index, and that flexibility may be the key lesson here. The index described below provides some revealing insights into the potential capability of certain regions to recover from economic shutdowns associated with COVID-19. However, the greater lesson may be in finding a sweet spot between high granularity and broad availability of data and in testing potential indicators to assess their capability to differentiate among regions in a way that is meaningful for decision-makers. The index developed in this chapter focuses on factors that can be characterised from more readily available data on local employment by industry sub-sectors. Factors used in our analysis are location quotients and shift-share factors. Location quotients provide an easy measure of the concentration of unaffected or growing sub-sectors, which can characterise a location’s resilience to the pandemic. The shiftshare factor expresses the strength of a local economy over time, and it is a marker of fertile ground for future growth. The modelling provides insight for policy making that should be considered alongside other academic insights into regional resilience.

2.4 Recovery Index To identify regions that may experience prolonged economic pain, we offer here a recovery index comprising three components and approximately 11 industries, depending on data availability. The industries selected include: food manufacturing; non-metallic mineral product manufacturing; infrastructure construction; warehousing; professional and technical services; information services; financial services; insurance carriers; public administration; air transportation; and travel lodging. In practice, the 11 industries are often aggregated from sub-sectors. This step ensures that the metric is as comparable as possible across countries. Industries were identified to be vulnerable based on experience with modelling regional industry trajectories in the consultancy work of two of the authors and by monitoring media coverage in the US and Australia. It is possible that additional vulnerable industries exist in these settings that did not receive much attention, but where outcomes are more evident after the COVID recovery. Those results become part of the knowledge base for subsequent modelling efforts. The focus here is on a method of analysis that yields insights quickly—as impacts are beginning to unfold, rather than on identifying this one model as a definitive one for assessing all pandemic impacts. Clearly, such major disruptions deserve concerted investigation after the fact. To capture reliance on what were expected to be the more resilient industries, we begin by calculating location quotients for the industries and geographies of interest.

248

J. Irving et al.

Formally, L Q i,g,t

) ( Σ ei,g,t / i ei,g,t ) = (Σ Σ Σ g ei,g,t / g i ei,g,t

(1)

where e is the level of employment in resilient industries, i, in geography, g, for the most recent period, and t, for which we have data. For the precise, more resilient industries used, the authors are providing supplemental online material. Next, we calculate the regional component of a shift-share analysis as ( RS i,g,t = ei,g,t−1 ∗

ei,g,t ei,g,t−1

Σ −Σ

g ei,g,t

g ei,g,t−1

) (2)

which is the regional employment in the starting period, t − 1, scaled by industry growth in the region less national growth in the industry. This step is meant to capture the regional competitiveness of the region’s industry over the recent past. Finally, sum the regional shift-share components as a ratio of the local employment in those industries for a shift-share factor. Σ RS i,g,t SS F g,t = Σi (3) i ei,g,t We then aggregate the regional shift-share factors and subtract an interaction between the location quotients of regional employment in air transport and accommodation. This interaction is included based on direct observations and communications with business leaders, which suggested that there would likely be differential recovery patterns between business and leisure travel. Leisure travel, as exhibited by bookings at caravan parks and camping facilities, showed surges in demand relatively early in the pandemic economic recovery. Businesses, on the other hand, embraced remote meetings through teleconferencing activities, especially for intra-corporate gatherings. However, there is no distinction in government employment data differentiating between leisure travel and business travel sectors, but there is a difference in average trip spending as is commonly found in visitor travel spending surveys gathered by tourism bureaus in most industrialized nations. Business travellers are more likely to travel to their destination by air transportation. In our judgment, that has been borne out in recovery trends, business travel will be slower to recover, and the best way to identify those communities with relatively higher dependence on business travel is by using the multiplicative (interaction) term between lodging (indicator for travel industry) and air transportation (biased toward business travel) as our best proxy measure for this sub-sector. We did not foresee the need to sub-divide other sectors in this same way for the purposes of this analysis. So no other interaction terms have been used in the index calculation. However, a more in-depth analysis of outcomes post-COVID might suggest other interaction components to consider in the future.

Pandemic Regional Recovery Index: An Adaptable Tool …

Rg,t =

Σ

L Q i,g,t + SS F g,t − (L Q air,g,t ∗ L Q acc,g,t )

249

(4)

i

The first two portions of the raw recovery index, R, represent regional strengths with regard to COVID-19, and the second portion represents regional economic weaknesses. The raw index is normalized by dividing by the average raw score, R, and multiplied by 100 for the reported index, I.

2.4.1

United States Results

To begin, we calculate the recovery index using private data from JobsEQ for 933 metropolitan and micropolitan statistical areas in the United States. Calculations are based on private data as publicly available data are often masked for key industries, such as air transportation. The data used are a 3-month average from the 3rd quarter of 2015 and the 3rd quarter of 2020. The data are aggregated to the CBSA level as defined by the U.S. Office of Management and Budget for 2015. This vintage of definition was chosen to align with Unemployment Statistics provided by the U.S. Bureau of Labour Statistics. The top and bottom CBSAs in the U.S. by recovery index are reported in Table 1. The CBSA with the highest recovery index is Sweetwater, TX, a town of 10,000 residents—40% of Hispanic background—and home to a rattlesnake roundup. It is in the northern part of the state on the north–south route between Dallas and El Paso. Sweetwater would not be seen as a wealthy community that was riding out the storm, as per capita and household income are both less than two-thirds of the national average. The biggest employer is manufacturing, with a new gypsum production facility announced in late 2020. Manufacturing employs 13-percent of the workforce, double the national average, with retail, educational services, healthcare/social services, agriculture and transport logistics all at 10-percent of the workforce or just above, according to the website of the town’s economic development organization (https://www.sweetwatertexas.net/strategic-advantages/ key-industries/—31/5/21). The county, on the edge of the Permian Basin, has seen significant development of oil and gas as well as in the newly emerging industry of windfarms and solar energy. Sweetwater’s relative economic resilience can be attributed to an array of industries and agriculture that were not vulnerable to COVID-19 shutdowns. This economic structure is similar to that in the second-and third-ranked communities, Owatonna, Minnesota and Dodge City, Kansas. The CBSAs with the worst recovery indices are Kahului, HI; Las Vegas, NV; Ketchikan, AK; Kapaa, HI; and Atlantic City, NJ. These lowest-rated localities all have a strong focus on tourism and/or convention travel, with several being reliant on gambling specifically. As an illustrative example of a lesser-known region, Ketchikan, Alaska, is a major tourist destination and a major cruise ship port. In 2020, Ketchikan had a location quotient (LQ) of 3.56 for accommodation and 7.37 for air transportation. While Ketchikan had location quotients over 3 for both food manufacturing and

250

J. Irving et al.

Table 1 U.S. recovery index—top and bottom metropolitan statistical areas Rank

Metropolitan statistical areas

Index

1

Sweetwater, TX (45,020)

406.2

2

Owatonna, MN (36,940)

391.4

3

Dodge City, KS (19,980)

386.5

4

Storm Lake, IA (44,740)

353.2

5

Vineland-Bridgeton, NJ (47,220)

331.6

…

…

…

929

Atlantic City-Hammonton, NJ (12,100)

−69.2

930

Kapaa, HI (28,180)

−142.3

931

Ketchikan, AK (28,540)

−173.2

932

Las Vegas-Henderson-Paradise, NV (29,820)

−188.6

933

Kahului-Wailuku-Lahaina, HI (27,980)

−237.6

public administration, the region’s economy has LQs below one for every resilient industry except for financial services, for which the index was almost exactly 1. In early 2021, the city council considered major layoffs as a result of lost revenue and a corresponding budget shortfall (Stone, 2021). Overall, that the index indicates that regional economies reliant on tourism will do worse is intuitive; hence, our use of the air travel and accommodation industries factors in the calculation of the index (Fig. 2). To examine the index more thoroughly, we compare the index against yearover-year unemployment change for months prior to and following the onset of the COVID-19 pandemic. Pearson correlation coefficients comparing CBSA recovery indices to year-over-year change indicate that the recovery index is highly correlated with changes in unemployment from 2019 to 2020. While the correlation varies over time, it remains reasonably strong. This consistency provides strong evidence that the index is performing well (Table 2). Some might well ask if the index correlates so well with unemployment figures, why do the modelling and not just track unemployment? The index has forecasting issues, whereas unemployment figures are not. Additionally, unemployment can become a problematic measure during such disruptions. For example, employers were paid by the federal government in Australia to keep staff on their payrolls. In other instances, staffs were put on furlough or on reduced hours, not fully employed but not necessarily looking for a new job, either.

2.5 International Results Given that the index performs well on U.S. data, we apply the index in 5 additional countries. The countries are Great Britain, Australia, New Zealand, Norway and

Pandemic Regional Recovery Index: An Adaptable Tool …

251

Fig. 2 Recovery index for U.S. metropolitan areas

Table 2 Correlation between recovery index and year-over-year change in unemployment

Year-over-year change in unemployment Pearson’s correlation September 2019–September 2020

−0.420 (0.000)

October 2019–October 2020

−0.441 (0.000)

November 2019–November 2020

−0.363 (0.000)

(p-value in parentheses)

Ireland. The target geographical area applied for all countries was the OECD (2020, p. 126) defined Territorial Level 3 (TL3). In the case of Great Britain, the chosen geography was smaller to better reflect the local administrative divisions. Additionally, labour force data across all counties is achieved minimally at the International Standard Industrial Classification (ISIC) 2-digit level, which is reasonably comparable to the U.S. NAICS 3-digit level used above. The country-specific terminology for the geographies and industrial classifications is outlined in each section.

2.5.1

Great Britain

The British data we used was publicly available from the Office for National Statistics’ Business Register and Employment Survey dataset. The geography reflects

252

J. Irving et al.

Table 3 Great Britain recovery index—top and bottom local unitary authorities/counties

Rank

Local authorities: county/unitary

Index

1

City of London

296.8

2

Clackmannanshire

275.7

3

Stirling

266.3

4

North Tyneside

250.4

5

Tower Hamlets

231.8

…

…

…

201

Manchester

2.9

202

Kensington and Chelsea

0.2

203

West Sussex

−36.6

204

Hillingdon

−132.5

205

Hounslow

−381.3

Local Authorities: County/Unitary and the employment data were pulled at NACE level 2. The survey data used were collected from businesses for the third quarters of 2015 and 2018. The dataset did not include data from Northern Ireland for our particular data selection, and it is therefore excluded in the rankings. Consequently, Great Britain, rather than the United Kingdom, was analysed for the rankings. The top and bottom performing local authorities are listed in Table 3. The local authority with the highest recovery index is the City of London. The authorities with the lowest recovery indices are Hounslow; Hillingdon; West Sussex; Kensington and Chelsea; and Manchester. Of the lowest-ranked areas, only a small number are from the north of England. While the north has been considered to be economically challenged in recent decades, the indicator suggests that Clackmannanshire and Stirling are doing quite well. Additionally, numerous regions from the south may struggle with recovery, including Wales. A similar number are from various parts of London, including suburbs near Heathrow and Luton airports. These results, and the quality of data available in Great Britain, suggest an opportunity for more detailed analysis that could provide fruitful for refining this model and developing other disaster indices.

2.5.2

Australia

For the Australian rankings, we used publicly available data from the Australian Bureau of Statistics (ABS) at the Statistical Area Level 4 (SA4) geography. SA4s are designed to represent labour sheds and for the output of Labour Force Survey data (ABS 2021). The employment data used in the recovery index was taken from the 2011 and 2016 ABS Censuses of Population and Housing at the Australian and New Zealand Standard Industrial Classification (ANZSIC) 2-digit level.

Pandemic Regional Recovery Index: An Adaptable Tool … Table 4 Australia recovery index—top and bottom statistical areas

Rank SA4

253 Recovery index

1

Sydney—North Sydney and Hornsby 154.0

2

Sydney—Blacktown

150.4

3

Melbourne—West

141.4

4

Sydney—Eastern Suburbs

140.7

5

Sydney—Inner West

138.6

…

…

…

82

Far West and Orana

74.2

83

Mandurah

73.9

84

Queensland—Outback

66.7

85

South Australia—Outback

65.2

86

Cairns

34.6

The top and bottom performing SA4s by recovery index are listed in Table 4. Four out of the five top performers are within the greater Sydney metropolitan area and position three is taken by Melbourne—West. The SA4s that have the lowest recovery indices are Far West and Orana, Mandurah, Queensland—Outback (where ‘Outback’ means remote from towns of any size), South Australia—Outback and Cairns (which is also in Queensland but on the coast near the Great Barrier Reef). Examining the components of the index, it is clear to see why Queensland— Outback and South Australia—Outback may have the most difficulty recovering. Queensland—Outback has only two location quotients greater than one among resilient industries, Civil Engineering and Administrative Services. South Australia—Outback shares with Cairns a single location quotient greater than one among the resilient industries, Civil Engineering and Administrative Service. More concerning, all three have location quotients greater than 2 in accommodation. Cairns is particularly ill-situated for the current pandemic, having a location quotient of 2.45 in air transport. Their tourism economy based on international travellers exemplifies impacts in the tourism industry across the country, where an absence of international travellers has created major declines in hotel prices (Carruthers, 2021).

2.5.3

New Zealand

For New Zealand, we used publicly available data from Statistics New Zealand (Stats NZ) at the Regional Council geography. Regional Councils are the top tier of local government in New Zealand. The employment data used in the recovery index was taken from the 2013 and 2018 Census datasets at the ANZSIC 2-digit level. The top and bottom performing Regional Councils by recovery index are listed in Table 5. Expectedly the highly gentrified Regional Councils have performed the best in the index including West Coast, Auckland and Wellington in the top three

254

J. Irving et al.

Table 5 New Zealand recovery index—top and bottom regions

Rank

Region

Index

1

Auckland

142.5

2

Wellington

126.6

3

Canterbury

111.7

4

West Coast

106.4

5

Hawkes Bay

101.9

…

…

…

12

Marlborough

91.0

13

Manawatu-Wanganui

90.7

14

Tasman

88.2

15

Gisborne

83.7

16

Nelson

83.0

positions. The bottom three positions in the index are held by the Tasman, Gisborne, and Nelson regions. Among the bottom three, only the Tasman Region and the Nelson Region had LQ’s greater than 1 in accommodation and air transport. The Gisborne region has relatively low shares of the detracting industries. However, all three had the lowest average LQs for the resilient industries, suggesting that all three have very little economic activity to sustain them. Despite this deficit, the Gisborne Region surprisingly increased economic output between December 2019 and December 2020, with stable visitor spending from January 2020 to January 2021 due to a specific reliance on intranational tourism (Rae, 2021; Thornber, 2021). Such contradictions highlight the need for local context. The low ranking for the Tasman Region and the Nelson region is worrying given that these areas were among the lowest output regions prior to the pandemic (Satherley, 2021).

2.5.4

Norway

The Norwegian data we used was publicly available from the Statistics Norway dataset 11,687. The geography used was counties, or fylker, which are the top level of local government in Norway. They also represent the Nomenclature of Territorial Units for Statistics Level 3 (NUTS3) geography. The NUTS system defines Level 3 regions as having a minimum population of 150,000 and a maximum population of 800,000 with a preference to reflect existing administrative divisions, which “supports the availability of data and the implementation capacity of policy” (Eurostat, 2021). Employment data were pulled for these administrative areas at the industry classification NACE level 2. The survey data used were collected from businesses for the fourth quarters of 2014 and 2019. The Norwegian recovery index rankings are presented in Table 6. The top three fylker in the recovery index are Oslo, Viken and Trøndelag, while the bottom three

Pandemic Regional Recovery Index: An Adaptable Tool … Table 6 Norway recovery index—top and bottom regions

255

Rank

Region

Index

1

Oslo (F-03)

149.4

2

Viken (F-30)

129.1

3

Trøndelag (F-50)

106.7

4

Vestfold og Telemark (F-38)

102.7

5

Vestland (F-46)

96.7

6

Nordland (F-18)

90.4

7

Innlandet (F-34)

88.2

8

Troms og Finnmark (F-54)

85.5

9

Rogaland (F-11)

84.6

10

Møre og Romsdal (F-15)

83.6

11

Agder (F-42)

83.0

positions in the index are Agder, Møre og Romsdal and Rogaland. In Norway, while the three lowest-ranked regions were not reliant on accommodation or air transport, they have the lowest average location quotients for the industries that we determined to be resilient. The Agder Region’s largest city, Kristiansand, is relatively small at just over 60,000 people. Given the small population, it has particularly low LQs in Warehousing and Storage (0.03), Professional and technical services (0.47), and Information Services (0.58).

2.5.5

Ireland

The data on the Republic of Ireland for the recovery index comes from publicly available data from the Central Statistics Office (CSO) in Ireland at the Nomenclature of Territorial Units for Statistics Level 3 (NUTS3) geography. The NUTS system defines Level 3 regions as having a minimum population of 150,000 and a maximum population of 800,000 with a preference to reflect existing administrative divisions which “supports the availability of data and the implementation capacity of policy” (Eurostat, 2021). The Irish employment data for employment were pulled for the NUTS3 regions at the Statistical Classification of Economic Activities in the European Community (NACE) level 2 and finer in some cases from the 2011 and 2016 Censuses. NACE level 2 data is equivalent to the target ISIC 2-digit level of our analyses. The Irish NUTS3 regional rankings are presented in Table 7. In total there are only eight NUTS3 regions in Ireland. The recovery index places the South-East, Mid-West and Dublin regions in the top three positions to best recover from the pandemic and the West, South-West and Midland regions in the bottom three positions to recover. Ireland is seen to have a relatively low disparity between its richest and poorest regions (OECD, 2018). The Border, South-West and Western regions are recognised to be falling behind economically, in general, and their recovery index values are low

256 Table 7 Ireland recovery index—top and bottom regions

J. Irving et al. Rank

Regional authority

Recovery index

1

Mid-West

116.4

2

South-East

115.8

3

Mid-East

105.9

4

Dublin

103.8

5

Midland

96.9

6

Border

95.5

7

South-West

90.5

8

West

75.1

in the analysis here. They are also assessed as having a deficit of access to services and education. The Western region lags all Irish regions in our index due to the fact that it has a high location quotient in Lodging, and only has two resilient industries, Fund Management and Civil Engineering, with location quotients greater than 1. The index is supported by research by the Irish department of finance, which identified the city of Galway, as the most at-risk city in Ireland (Kelly, 2020). We can see across these countries that the Recovery Index distinguishes between regions, differentiating strongly between those at the top—expected to be more resilient—and those at the bottom—expected to be having a harder time in recovering from COVID-19. The ranking of regions seems to correspond with what one might expect, based on local knowledge of the history and economy of each region. That intuition is confirmed by viewing location quotients for key industries tracked by the Recovery Index. Across countries, the Index is identifying strengths in urban capitals in European countries as well as in New Zealand and Australia. That is not the case in the US, though, with the top ranked localities being rural towns with populations of 10,000–25,000 residents and a base of manufacturing and agriculture.

2.6 Conclusion This work sought to develop an economic recovery tool specifically for regional economies impacted by economic fall-out of COVID-19. The tool was informed by prior work, particularly in relation to absorptive and adaptive regional resilience, while making worthwhile adjustments. The tool was initially applied to the U.S. context, where it was tested for effectiveness. The tool performed reasonably well in explaining the recovery in terms of declining unemployment. The tool was then applied to 5 additional countries—the UK (Great Britain), Australia, New Zealand, Norway, and Ireland. While these countries have less timely data, anecdotal evidence for a number of the countries—e.g., media coverage on the economic performance of certain regions—suggests that the tool performs well.

Pandemic Regional Recovery Index: An Adaptable Tool …

257

The tool appears to fill a gap in the literature on economic recovery from disasters and the more recent publications on COVID-19 impacts. The tool combines regional level granularity to inform policy and decision-making by local, state and federal governments. At the same time, it appears in this initial analysis to be revealing when applied in different countries that are affected by the same pandemic during the same time period, something that more traditional disaster analyses do not do. In this sense, it could be seen as part of a trend in analysing broad scale impacts at local levels, ranging from the Global Financial Crisis to the impacts of climate change. While the tool has identified regions that may need additional help in forging an economic recovery from COVID-19, there have been several lessons learned when applying the same tool to a range of countries. First, it is impossible to apply exactly the same tool given differences in categorization of industries. In many respects, this difference in taxonomies makes it difficult to apply lessons from some contexts, such as the U.S., to other contexts, such as Norway or New Zealand. That suggests that insights developed about how to diversify a regional economy to make it more resilient in the face of a pandemic may be tough to translate in a credible way. On this front, gleaning lessons from one country that could be applied to another, substantial benefits could be gained if data were more harmonized among countries. That could contribute to revealing lessons that can be more easily transferred, with better outcomes for all in the face of future global crises. The second notable lesson is that, of course, regional particularities matter. While regions reliant on international tourism have been decimated en masse, there is an exception in New Zealand. The Gisborne Region increased tourism year-over-year, which can be seen to reflect a focus on domestic tourism. That provides an interesting example that can help other such regions to retool. Similarly, in Australia, certain regions that relied more on international visitors than others were harder hit in the wake of the pandemic and have been slower to recover. Even once international borders open, they will await the recovery of overseas economies on whose visitors they rely. Regional economies focused on tourism can thus work to rebalance between proportions of local to international customers. The type of economic activity may not need to change, but the customer base may have to. One can argue that such dynamics apply to international supply chains, more generally—with regions likely to benefit from balancing international supply chain partners with domestic supply chain partners. There are several areas of future work that seem to emerge from the development of the present index. First, there is a need for national datasets to further harmonize with one another. If datasets are further harmonized, lessons learned regionally can be more quickly applied internationally. Second, the index developed here can be applied and tested in additional regions. Again, data were the constraint here, with these six countries being the only ones in the OECD with suitable data that is readily available. Additional applications and testing could lead to additional refinements to the index. These data constraints inhibit the analysis of potentially valuable factors. For example, ignoring related and unrelated variety added simplicity, but it can detract from the accuracy of the model’s results. For example, this work has shown that

258

J. Irving et al.

greater variety among related industries is correlated with regional economic growth (Frenken et al., 2007) in two ways. First, the present analysis might be tied to related variety by labour migrating from an origin industry, one that experiences a large number of layoffs due to the pandemic, into a related industry. However, it is possible that such related industries can experience similar rates of employment loss, which would make them unable to absorb excess labor. Second, industries that might have been experiencing growth, bolstering a region’s economy, may have lost extra-regional sources of labour as nations closed their borders. Since this analysis was performed in 2020, labour shortages emerged during 2021 and 2022 that have constrained the speed of economic recovery in some localities. That has been particularly evident in Australia, where immigration is a key source of skilled labor. It would be expected in countries in the European Union that also depend on migrant labor. Unfortunately, data on international migration is not sufficiently detailed to enable formulating meaningful, actionable policy measures. These factors would benefit from further investigation, to assess the extent to which related variety exacerbates or helps to relieve labour shortages. Finally, a family of indices similar to the one proposed here could be developed and maintained to apply in the case of future disasters, or other major changes, such as due to increasing economic and social effects of climate change, conflict or pulses in migration of refugees. Developing and maintaining a family of indices for international disasters, such as pandemics and financial crises, would allow regional policy makers to access information and work to help their regions recover quickly in the event of a disaster. Any organization maintaining such a set of indices could also work to socialize best practices. Conducting analysis in advance of such disasters could be particularly worthwhile in the development of an array of policy levers. Funding This work was supported by the Regional Studies Associated COVID-19 Grant Scheme. Disclosure Statement The authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript; or in the decision to publish the results.

Data Availability All data used in this study will be made available upon request to the authors.

References Argent, N. (2021). Adaptive capacities and social resilience on Kangaroo Island: Beyond the staples trap. Geographical Research, 59(3), 378–393. Australian Bureau of Statistics (ABS). (2021). Australian Statistical Geography Standard (ASGS). Retrieved March 4, 2021, from https://www.abs.gov.au/websitedbs/d3310114.nsf/home/austra lian+statistical+geography+standard+(asgs) Baum, S., & Mitchell, W. (2020). Employment vulnerability index 3.0. Centre of Full Employment and Equity.

Pandemic Regional Recovery Index: An Adaptable Tool …

259

Boschma, R. (2015). Towards an evolutionary perspective on regional resilience. Regional Studies, 49(5), 733–751. Boschma, R. (2017). Relatedness as driver behind regional diversification: A research agenda. Regional Studies, 51(3), 351–364. Brada, J. C., Gajewski, P., & Kutan, A. M. (2021). Economic resiliency and recovery, lessons from the financial crisis for the COVID-19 pandemic: A regional perspective from Central and Eastern Europe. International Review of Financial Analysis, 74. Carruthers, F. (2021). As Cairns hotels dip below $99, NSW coastline cashes in. Financial Review. Retrieved March 26, 2020, from https://www.afr.com/companies/tourism/as-cairns-hotels-dipbelow-99-nsw-coastline-cashes-in-20210307-p578h7 Cutter, S. L., Burton, C. G., & Emrich, C. T. (2010). Disaster resilience indicators for benchmarking baseline conditions. 7(1), article 51. Davradakis, E., Zwart, S., Marchitto, B., & Santos, R. (2020). The EIB COVID-19 economic vulnerability index—An analysis of countries outside the European Union. European Investment Bank, Economics Department. Deloitte. (2021). COVID-19 Economic recovery dashboard. Retrieved February 8, 2021, from https://www2.deloitte.com/ca/en/pages/about-deloitte/articles/covid-dashboard.html Djalante, R., Lassa, J., Setiamarga, D., Sudjatma, A., Indrawan, M., Haryanto, B., Mahfud, C., Sinapoy, M. S., Djalante, S., Rafliana, I., Gunawan, L. A., Surtiari, G. A. K., & Warsilah, H. (2020). Review and analysis of current responses to COVID-19 in Indonesia: Period of January to March 2020. Progress in Disaster Science, 6. Drzeniek, M., Tambourgi, S., & Marchese, I. (2020). Recover, (re)build: Exploring country recovery capacity post-COVID. Horizon Group. Eurostat. (2021). NUTS—Nomenclature of territorial units for statistics: Principles and characteristics. Retrieved March 4, 2021, from https://ec.europa.eu/eurostat/web/nuts/principles-and-cha racteristics Frenken, K., Van Oort, F., & Verburg, T. (2007). Related variety, unrelated variety and regional economic growth. Regional Studies, 41(5), 685–697. Gao, H., & Ren, M. (2020). Beyond travel, a trans-Tasman bubble is an opportunity for Australia and NZ to reduce dependence on China. Retrieved March 26, 2021, from https://theconversation.com/beyond-travel-a-trans-tasman-bubble-is-an-opportunityfor-australia-and-nz-to-reduce-dependence-on-china-137062 Giannakis, E., & Bruggeman, A. (2017). Determinants of regional resilience to economic crisis: A European perspective. European Planning Studies, 25(8), 1394–1415. Halonen, M. E. (2019). The long-term adaptation of a resource periphery as narrated by local policy-makers in Lieksa. Fennia, 197(1). Kelly, B. (2020). Galway is most at risk city to economic hit from Covid-19. Galway Daily. Retrieved March 26, 2020, from https://www.galwaydaily.com/news/galway-city-is-most-at-risk-city-toeconomic-hit-from-covid-19/ Melecký, L. (2017). Meta-Analysis of compsite indices in the evaluation of European Union Territory. In Conference Proceedings: 20th International Colloquium on Regional Sciences. Brno: Masaryk University. https://doi.org/10.5817/CZ.MUNI.P210-8587-2017-4. Available through https://www.researchgate.net. Manyena, B., Machingura, F., & O’Keefe, P. (2019). Disaster Resilience Integrated Framework for Transformation (DRIFT): A new approach to theorising and operationalising resilience. World Development, 123. Martin, R., & Sunley, P. (2015). On the notion of regional economic resilience: Conceptualization and explanation. Journal of Economic Geography, 15(1), 1–42. Menoni, S., & Schwarze, R. (2020). Recovery during a crisis: Facing the challenges of risk assessment and resilience management of COVID-19. Environment Systems & Decisions, 40, 189–198. Morelli, A. (2020). New York City economic recovery index methodology. Newstex.

260

J. Irving et al.

Nicola, M., Alsafi, Z., Sohrabi, C., Kerwan, A., Al-Jabir, A., Iosifidis, C., Agha, M., & Agha, R. (2020). The socio-economic implications of the coronavirus pandemic (COVID-19): A review. International Journal of Surgery, 78, 185–193. OECD. (2018). OECD regional statistics (database). Retrieved August 16, 2022, from https://doi. org/10.1787/region-data-en OECD. (2020). Regions and cities at a Glance 2020. OECD Publishing. Retrieved August 16, 2022, from https://doi.org/10.1787/959d5ba0-en OECD. (2022). Quarterly National Accounts. OECD National Accounts Statistics (database). Retrieved August 11, 2022, from https://doi.org/10.1787/data-00017-en Pemandu Associates. (2021). Global COVID-19 dashboard. Retrieved February 9, 2021 from https://covid19.pemandu.org/about Pike, A., Dawley, S., & Tomaney, J. (2010). Resilience, adaptation and adaptability. Cambridge Journal of Regions, Economy and Society, 3(1), 59–70. Rae, S. (2021). Blow to regional economies becoming starker, stats show. Otago Daily Times. Retrieved March 26, 2020 from https://www.odt.co.nz/business/blow-regional-economies-bec oming-starker-stats-show Robinson, G. M., & Carson, D. A. (2016). Resilient communities: Transitions, pathways and resourcefulness. The Geographical Journal, 182(2), 114–122. Rose, A. (2017). Economic resilience in regional science: Research needs and future applications. In R. Jackson & P. Schaeffer (Eds.), Regional research frontiers—Vol. 1: Innovations, regional growth and migration (pp. 245–264). Springer International Publishing. Satherley, D. (2021). New figures reveal the most and least productive regions of New Zealand, Newshub. Retrieved March 26, 2020 from https://www.newshub.co.nz/home/money/2021/03/ new-figures-reveal-the-most-and-least-productive-regions-of-new-zealand.html Stanickova, M., & Melecký, L. (2018). Understanding of resilience in the context of regional development using composite index approach: The case of European Union NUTS-2 regions. Regional Studies, Regional Science, 5(1), 231–254. Stone, E. (2021). Ketchikan City Council to consider layoffs as officials reckon with cruise shipsized hole in city budget, KTO. Retrieved March 25, 2021 from https://www.ktoo.org/2021/03/ 18/ketchikan-city-council-to-consider-layoffs-as-officials-reckon-with-cruise-ship-sized-holein-city-budget/ Thornber, L. (2021). Domestic tourism: The biggest winners and losers a year after the Covid-19 pandemic closed borders, Stuff . Retrieved March 26, 2020 from https://www.stuff.co.nz/travel/ news/300259014/domestic-tourism-the-biggest-winners-and-losers-a-year-after-the-covid19pandemic-closed-borders Van Huellen, S., & Asante-Poku, N. A. (2020). How commodity exporting countries like Ghana have been hit by COVID-19. Retrieved March 26, 2021 from https://theconversation.com/howcommodity-exporting-countries-like-ghana-have-been-hit-by-covid-19-155799 Xiao, J., Boschma, R., & Andersson, M. P. (2018). Resilience in the European Union: The effect of the 2008 crisis on the ability of regions in Europe to develop new industrial specializations. Industrial and Corporate Change, 27(1), 15–47. Zandi, M., DeAntonio, D., & Colyar, M. (2020). Moody’s analytics & CNN business back-to-normal index, Moody’s analytics.

The Geography of Daily Urban Spatial Mobility During COVID: The Example of Stockholm in 2020 and 2021 Ian Shuttleworth, Marina Toger, Umut Türk, and John Östh

1 Introduction This chapter explores how the spatial mobility of Stockholm’s residents around the city changed at the peak of the COVID-19 Pandemic in 2020 and early 2021 compared with a 2019 baseline. This study first introduces the development of the pandemic by comparing lockdown levels (Covid Government Response Tracker, 2022) with aggregate mobility (Google Mobility Reports, 2021) and then completes the picture by continuing with an analysis of finer-grained mobile-phone based mobility data, joining socioeconomic and contextual data ecologically. Stockholm is a useful and important case study. Sweden followed its own exceptional (by EU standards) public health policy from the outset of the pandemic. Under the leadership of the Public Health Agency (PHA), headed by Anders Tegnell, Swedish policy sought to protect the risk groups and the national healthcare system by reducing the spatial mobility of the population and large gatherings. However, unlike most other European countries, it aimed only to mitigate the virus’ effects rather than suppress it completely (Granberg et al., 2021) and, what is particularly important, operated chiefly through nudge and guidance, rather than by compulsion. This was because the PHA was sceptical of the need for (and effectiveness of lockdowns) given the perceived nature of Swedish society and its dispersed population geography (Pierre, 2020). Understanding how the city’s population responded when given greater personal discretion I. Shuttleworth (B) QUB, Belfast, UK e-mail: [email protected] M. Toger University of Uppsala, Uppsala, Sweden U. Türk Abdullah Gül University, Kayseri, Turkey J. Östh Oslo Metropolitan University, Oslo, Norway © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. G. Celbi¸s et al. (eds.), Pandemic and the City, Footprints of Regional Science, https://doi.org/10.1007/978-3-031-21983-2_14

261

262

I. Shuttleworth et al.

in a national context different to cities in many other countries is therefore useful in understanding more about how urban populations behave in stressful times.

2 Background There is an emerging literature on population responses to public health restrictions (for example, Almlöf et al., 2021; Järv et al., 2021; Santamaria et al., 2020), and findings that the better-off are more able to modify their behaviour than those in lowerpaid essential and public-facing work (Dingel & Neiman, 2020). For instance, the work of Coven and Gupta (2020) reveals mobility-related disparities between richer and poorer neighbourhoods in New York, where—according to their findings—lowincome, black, and Hispanic neighbourhoods exhibited higher work-place mobility during the first months of the pandemic. Similarly, Lou et al. (2020) show that lowincome communities were 46–54% more mobile than high-income communities in the US (see also Kim & Kwan, 2021). Also, populations with worse socioeconomic conditions did not have the same opportunity to reduce their mobilities despite lockdown policies in Bogota, Colombia (Dueñas et al., 2021). A major reason for these disparities is unequal access to remote work opportunities by occupation type and education level (Mongey et al., 2021). In line with this, Pullano et al. (2020) document that while general mobility decreased as much as 65% in France, mobility reduction was not homogeneous across different sectors and socioeconomic groups. Associated with this is evidence that visible minorities and the poor faced higher incidences of infection and death in Sweden (Drefahl et al., 2020). However, there are questions about how far these social and spatial patterns are generalizable. This is because geographical and political context matters and because the morphology and social geography of cities shape daily spatial activity in unpredictable and complex ways (Östh et al., 2018). A good example of this is the mobilephone-based work of Xu et al. (2018) who found that phones from high-income areas had a wider spatial range in Boston, Massachusetts than those from lowerincome areas, whereas the opposite was the case in Singapore. The reason for this, of course, was to do with where high- and low-income areas were located relative to opportunities and facilities. In a study of mobile phone activity spaces in the greater Stockholm region before and during the pandemic, revealed that areas with high share of lower educated workers displayed long but thin activity spaces that are consistent with a commuting pattern that includes few or no additional destinations. In higher educated areas, smaller rounder activity spaces became more common during the pandemic. The results indicate that labour market and personal attributes (being able to work from home or not) affected and affect mobility behaviour (Toger et al., 2021). Adding more information on Stockholm therefore adds to the evidence base in which all cities, to a greater or lesser degree, are unique. With these broad rationales in mind, the chapter aims to deal with two interrelated spatial mobility themes which it discusses sequentially. The first concerns population interaction; in what parts of the city was there the greatest interaction of people from

The Geography of Daily Urban Spatial Mobility During COVID: The …

263

other parts before and during the pandemic? Did social interaction, as measured by spatial propinquity, decrease during the pandemic? The second is about the effectiveness of public health guidance; did people change their spatial behaviour? In what parts of the city were the greatest changes observed? Were the changes long lasting? Before we attempt to answer these questions, the next section of the chapter sets the Swedish scene in more detail and demonstrates how and why the analysis goes further than Google Mobility reports. Following this, the data, metrics, and methods used in the analysis are discussed. We then continue to the main findings as they relate to the themes and questions are considered and we conclude with some wider reflections on the research.

3 Setting the Swedish Scene As COVID-19 spread across the world in Spring 2020 there were no medical or pharmaceutical treatments, such as vaccines or effective medicines, available to health professionals and policy makers to combat the disease. Therefore, the response of first resort by governments was to limit the spread of the virus. Since it is spread through interpersonal contact, this required spatial mobility to be restricted (or cease completely) and interpersonal contacts to be minimised (or stopped altogether). In practice this meant state restrictions on personal freedom and on the spatial mobility of people at a variety of spatial scales ranging from the two-metre distancing rule at the personal level, through restrictions on the size of social gatherings for households, to instructions or advice to work from home where possible, and the closure of education at the societal scale. Governments across Europe took similar actions to regulate the behaviour of their populations although, of course with variations in the time, mode, and severity of the application of their rules. France, Italy, and Ireland were quick off the mark in implementing lockdowns, the UK was comparatively slow in moving to restrictions but then instituted a harsh lockdown in late March, whereas Sweden started restrictions in good time but operated them through persuasion and advice, in contrast to most other European countries (Andersson & Aylott, 2020). This reflected varying national traditions of governance and the variety of ways in which states engaged with their populations, although Sweden was considered an outlier (Warren et al., 2021). It would be wrong to say, however, that the Swedish response was completely laissez faire, as some restrictions were recommended, and their course through 2020 are clearly summarised by Ludvigsson (2020). As elsewhere, the measures in Sweden concerned the regulation of spatial mobility and social interaction. During the First Wave of the pandemic on March 10th 2020 the advice was to limit visits to healthcare and elderly care; March 11th 2020 saw meetings with more than 500 people cancelled; from March 16th 2020 people aged more than 70 were told to limit close social contacts and workers were advised to work from home if possible. From March 19th 2020, the recommendation was to avoid unnecessary travel in Sweden and at the end of March, gatherings of 50 or more

264

I. Shuttleworth et al.

people were banned. In common with much of Europe, restrictions eased in Summer 2020 (travel restrictions within Sweden were lifted on June 13th 2020 and again on July 1st 2020) but November and December 2020 saw spatial mobility restrictions returning after a quiet Summer with two speeches by the Swedish Prime Minister in November 2020 highlighting the dangers of COVID and then the reintroduction of travel restrictions by the PHA. It is possible to trace the evolution of policy restrictions and corresponding changes in spatial mobility in Figs. 1, 2, and 3. Figure 1, drawing on the University of Oxford international database of governmental COVID responses, outlines them in a summary index (Covid Response Tracker, 2022). This resource, in its own words, aims “to track and compare policy responses around the world, rigorously and consistently.” The rapid onset of restrictions in March 2020 is very apparent as is their summer easing and then their reintroduction in late 2020. Autumn 2021 shows a reduction in restrictions, down to their level in March 2021. Figures 2 and 3 show changes from baseline in the two major domains of work and home, respectively, for Stockholm County and Stockholm Municipality. These are two of the spatial units for which Google Mobility data are available in Sweden—see Map 1—and they can usefully begin the exploration of changing spatial behaviour at the scale of whole cities or, indeed, entire societies. These data are freely available to download and are based on location histories with some random perturbations to preserve privacy. They have been compiled and aggregated by Google and show “…how visitors to (or time spent in) categorized places change compared to our baseline days. A baseline day represents a normal value for that day of the week. The baseline day is the median value from the 5-week period Jan 3–Feb 6, 2020” (Google Mobility Reports, 2021). Immediately noticeable in Figs. 2 and 3 are the far larger changes for workplace than for residence, and the greater amplitude of the workplace changes for Stockholm County than for Stockholm Municipality—the most central part of the city. This hints at differences in behaviour within urban areas, perhaps because of the geographies 80 70

stringency index

60 50 40 30 20 10

8/31/21

9/21/21

8/10/21

7/20/21

6/8/21

Fig. 1 The Sweden COVID stringency index. Data source University of Oxford

6/29/21

5/18/21

4/6/21

4/27/21

3/16/21

2/2/21

2/23/21

1/12/21

12/1/20

12/22/20

11/10/20

10/20/20

9/8/20

9/29/20

8/18/20

7/7/20

7/28/20

6/16/20

5/5/20

5/26/20

4/14/20

3/3/20

3/24/20

2/11/20

1/21/20

0

The Geography of Daily Urban Spatial Mobility During COVID: The …

265

10

mobility at workplaces

0 -10 -20 -30 -40 -50 -60

Work Stockholm Municipality

9/21/21

8/31/21

8/10/21

7/20/21

6/8/21

6/29/21

5/18/21

4/6/21

4/27/21

3/16/21

2/2/21

2/23/21

1/12/21

12/1/20

12/22/20

10/20/20

11/10/20

9/8/20

9/29/20

8/18/20

7/7/20

7/28/20

6/16/20

5/5/20

5/26/20

4/14/20

3/3/20

3/24/20

2/11/20

1/21/20

-70

Work Stockholm County

Fig. 2 7-day moving average—workplace change from baseline, Stockholm County and Stockholm Municipality. Data source Google Mobility Reports

mobility at residential neighbourhoods

20 18 16 14 12 10 8 6 4 2

Residence Stockholm Municipality

9/21/21

8/31/21

8/10/21

7/20/21

6/29/21

6/8/21

5/18/21

4/27/21

4/6/21

3/16/21

2/23/21

2/2/21

1/12/21

12/1/20

12/22/20

11/10/20

10/20/20

9/29/20

9/8/20

8/18/20

7/28/20

7/7/20

6/16/20

5/26/20

5/5/20

4/14/20

3/24/20

3/3/20

2/11/20

1/21/20

0

Residence Stockholm County

Fig. 3 7-day moving average—residential change from baseline, Stockholm County and Stockholm Municipality. Data source Google Mobility Reports

of opportunity densities (see also Dahlberg et al., 2020). In distinction, the deviation from baseline for residence is very similar for the residents of the most central part of the city compared with those from other parts. This again suggests that spatial behaviour is complex and context dependent, but more so for some populations and some activity domains than others. As workplace visits decreased, residential time increased, although there was not a one-for-one match. The initial shock of the onset of the pandemic can be observed in March and April 2020 and then a general growth in workplace activity as restrictions eased in later 2020. What is also notable is the large effect of seasonal festivals, especially Christmas, but also to a lesser extent Swedish Midsommar holiday in June. The changes in behaviour also seem long lasting.

266

I. Shuttleworth et al.

Map 1 The location and extent of the study area (dashed black line, left) in Sweden (left) and the areas of Stockholm municipality (blue polygon, right) and Stockholm County (blue line, right) and the study area (dashed black line, right)

Google Mobility data can only partially deal with some of the questions raised at the start of the chapter but in some other ways go beyond the MIND data. They deal, for instance, with other domains (for example, retail and transit) which are not considered in this chapter for a more general overview of mobility at the scale of cities and societies. However, there are limitations in the answers they can provide to the chapter’s questions. First, the municipality geography is too large and coarse to capture the often local, social geographies of cities, which influence spatial mobility through the social composition of areas and their locational context. Second, to answer our questions fully, other metrics are needed beyond the deviation from baseline provided by Google—indeed, there are also questions about the appropriate baseline from which to measure change, an issue we discuss in our differencing approach in the next section.

The Geography of Daily Urban Spatial Mobility During COVID: The …

267

4 The MIND Mobile Phone Data, Metrics, and Urban Data 4.1 The MIND Data We use the MIND dataset located at Uppsala University, which contains pseudonymised phone-usage records of 10–20% of the Swedish population from one major mobile phone network (MPN) provider. This is an extended Call Detail Records (xCDR) dataset that includes silent handovers. This means there are records 24/7 whilst the phone is switched on and not just when it is in use. Its research use is guided by an ethics application approval. A selection of phones (which had spent the night in Stockholm) was made for the third Thursday of the months January, March, and October in 2019, 2020 and 2021 to make a panel. In addition, Easter Thursdays were also included in the panel to examine changes in spatial mobility outside normal working days. To study and compare changes in urban spatial mobility patterns over time, detailed records depicting location and time are needed. The accuracy of the data is dependent on the density of GSM1 antennas, and the densest distribution of antennas are found in or near the metropolitan areas. It is thus especially suitable for this Stockholm analysis.

4.2 The Spatial Mobility Metrics In the extended CDR dataset, there is a registered link between each phone and an antenna providing service. If there are several antennas in the vicinity, the phone will switch between antennas based on strength of signal, kind of service, and availability of the antenna, hence the need for duration weighting. For each day, the origin–destination distance was therefore calculated by using the duration-weighted connections to the different antennas within specific time-frames (00:30–07:20 for origin, and 11.00–12.00 and 13.00–15.00 for destination) with origin and destination coordinates at a 1 km square grid resolution. These represent place of home (origin) and day activity (destination). The OD distance variable was calculated as each phone’s Cartesian distance between the estimated coordinates of origin and destination. Two more mobility metrics were used to show the mobility changes qualitatively: inflow and number of destinations per origin (inflow metric here is a similar mobility indicator as inwards movement in Santamaria et al., 2020), aggregated here from an Origin–Destination Matrix of mobile phones by square km grid units). Only active phones that are in the area during both night and day will have coordinates that make it possible to place a phone in a km2 unit of origin and destination. Just the phones that were estimated to have an origin within the 50 km threshold from the Stockholm Central Station were included in the analysis to exclude cross-city mobility. To measure the impact of COVID-induced change, dates were selected in 2019—a 1

Global System for Mobile Communications.

268

I. Shuttleworth et al.

non-COVID year—and compared with their equivalents in 2020 and 2021. The third Thursdays of January, March, and October were chosen along with the Thursday before Good Friday. Thursdays were selected as being the least variable of weekdays (Toger et al., 2020). Different months were chosen because there are spatial mobility differences seasonally. Thus, the best baseline comparator is not the last five weeks previous to the onset of COVID regulations—as in Google Mobility Reports—but the equivalent month and day the year before. Dates in October were selected because Autumn 2020—as seen above—saw a lessening of COVID rules in Sweden. Finally, Easter was chosen precisely because it was a holiday time. Populations behave differently over holidays than on normal weekdays and this allowed this study to explore how behaviour was modified on these dates. The analysis is mainly descriptive and focusses on first differences, e.g., between January 2019 (non-COVID) and January 2021 (COVID), as well as on aggregated patterns of movement around the city.

4.3 Urban Morphology and Social Geography For each phone-populated km2 unit in the dataset, additional data was appended describing its relationship to the city centre. In this case, Cartesian distance between the estimated origin and the main rail station in Stockholm was saved as DistOJV and used to indicate the level of centrality of residence of each phone-user. This was a proxy for the density of opportunities and services with centrally placed residents having greater nearby opportunities in contrast with their greater sparsity in more peripheral areas. To describe area socio-demographic composition, a k-nearest neighbour (KNN) approach was implemented in which the population composition among the 500 nearest neighbours from each km2 unit midpoint was used. The data, which are unrelated to the phone dataset, come from the population register PLACE that is available to researchers at Uppsala University. PLACE is derived from the Swedish Population Register. It has geocoded information about socio-demographics and geography at a 100 m square grid resolution. It is the major element of the Swedish population data system and is collected for official statistical purposes. The latest available version was for 2017. However, social and demographic change is not speedy enough to make these data invalid for 2019 which was used as the base for the analysis. The following variables were created: Highedu 500, describes the share of individuals with a university degree, and therefore likely having greater opportunities to telecommute; Rich 500, describes the number of rich people (using an EU measure of poverty; Eurostat, 2021); and VM 500, describing the share of individuals born in Africa, Asia or Latin-America. VM, or Visible Minority groups, have in Sweden been observed to have greater COVID19-related mortality rates compared to the average population. These area types were grouped into quintiles (e.g., very close distance to 500 visual minority people, close distance, medium, far, and very far or very high shares, high shares, average shares, low shares, and very low shares) to explore the importance of social context on mobility behaviour. The

The Geography of Daily Urban Spatial Mobility During COVID: The …

269

variables and the socio-demographic variables listed above were all created using EquiPop (Östh, 2014; Östh & Türk, 2020). For each phone, a crude measure of origin is calculated as the duration-weighted average position between 00:30 and 07:20 for each of the days. Coordinates and durations are derived from location data of the antennas to which the phone is connected within this night timeframe.

5 Results At this juncture, it is worthwhile restating in summary the chapter’s questions before attempting to answer them with the analysis. In brief, where was there the greatest interaction in the city before and during the pandemic? Did interaction within the city decrease during COVID and, if so, where? Did people change their spatial behaviour during the pandemic? Were the changes greater in some areas than in others, for example in places with ‘at risk’ populations such as visual minorities? Were the changes long-lasting? The first of these questions, about interaction within the city, are answered in Map 2a–c. These show the number of origins per cell destination on January 17th 2019, January 16th 2020, and January 14th 2021 respectively, the first two non-COVID, the third a COVID month. The reds indicate destinations with more origins, the redder the greater the number. For all three months, the centre of the city is the place with most origins. This means that it is the central focus of flows from other parts of the city— thus is the key area where people from different locations are brought potentially into close spatial propinquity and danger of the transmission of infections. There are few differences between the non-COVID months of January 2019 and January 2020. However, the COVID month of January 2021 looks different with smaller highorigin areas in the city centre, and fewer high-origin areas in general throughout the map. This is evidence for COVID-induced changes in behaviour which limit spatial mobility and hence the probability of interaction within the city centre. The second set of questions are answered by the data visualised in Map 3a and b which show first differences between Thursdays in March 2019 and 2020, and March 2019 and 2021, respectively. Red coloured locations had a positive difference (increase) and blue colour—negative difference (decrease). The bluer the colour, the larger the decrease in inflows in terms of the number. On these weekdays, the largest falls are in the city centre and also along transit routes. The areas (red) with gains are widely distributed but sparse, tending to lie in more peripheral and rural areas. These patterns match to prior expectations, which would have suggested decreases in commuting and workplaces especially in central areas. Map 4a, b, with comparable colour schemes, show how behaviour changes during religious and holiday festivals comparing Easter Thursday 2019 with Easter Thursday 2020 and Easter Thursday 2021. There are two observations to make in comparison with the weekdays in Map 2. First, in comparison with the pre-COVID Easter of 2019, the city centre experienced decreases at Easter 2020 and Easter 2021. Second, relative to Easter 2019, the Stockholm Archipelago received more inflows in 2020 and 2021. This is

270

I. Shuttleworth et al.

Map 2 Comparison of number of origins per destination between Thursdays in January in 2019 (a), January 2020 (b) and January 2021 (c). Number of origins can give an indicator of diversity of locations from which people originate at each destination and thus can affect the possibility of spread of the virus. Colours range from blue (one origin) to red (many origins), with roads and central urban locations (Stockholm and Uppsala) expectedly having more origins than less central areas. Many factors can affect number of origins, such as weather, but still there is a considerable reduction in the number of origins even in the central areas of Stockholm and Uppsala

especially so for Easter Thursday 2021. This might be the result of one-off factors such as weather but might also indicate restriction fatigue and people seeking green spaces and water. More generally, holidays differ from weekdays in aggregate spatial mobility patterns with more activity in waterside and park locations. Figure 4 answers the question as to whether people changed their spatial behaviour during the pandemic. Of course, this question has been partially answered by the Google Mobility Report data, and also by Maps 1, 2, and 3 but this figure deals with a different outcome. The histograms show the distribution of individual origin– destination distances on a log 10 scale for two non-COVID January days in 2019 and 2020, compared with COVID-hit January 2021. The two non-COVID days have similar distance distributions. There are two clusters—one for longer distances, the other very local. These reflect the spatial scale of urban opportunities which shape daily activity spaces; work, school, university, college, or access to other activities fall into certain distance ranges. In January 2021, the two clusters remain—the spatial opportunity structure remains the same. However, it is readily apparent that the frequency distribution changed in January 2021 in comparison with earlier years. There are fewer trips in the longer distance group, more in the shorter group. What this means is that COVID did not result in a decrease in mean origin–destination distances for everyone. Instead, the mean reduction was the result of more people staying very local with fewer people travelling around the city as people switched between two states in a bimodal distribution. There are basically two mobility states, ‘near’ and ‘far’, and during the pandemic more switched to ‘near’, although some stayed in ‘far’. What did not happen was everyone shifting nearer to home into a new

The Geography of Daily Urban Spatial Mobility During COVID: The …

271

Map 3 Inflow first difference at each destination between a 2020-March-26 to 2019-March-28 and b 2021-March-11 to 2019-March-28

Map 4 Inflow first difference at each destination between a 2020-April-09 to 2019-April-18 and b 2021-April-01 to 2019-April-18

272

I. Shuttleworth et al.

‘intermediate’ category. Figures 5 through 7 now explore how the bimodal origin– destination distributions changed through time in different social contexts that are hypothesised to relate either to ability to work from home or vulnerability to COVID. Figure 5 begins with higher education for selected Thursdays in March (March 28th 2019, March 26th 2020, March 11th 2021). These graphs show that there is a clear gradient in spatial behaviour with greater frequencies in the mobile cluster

Fig. 4 Origin–destination distance histograms on January 2019 (top left), January 2020 (top right), January 2021 (bottom)

Fig. 5 Distribution of distances between origin and destination by higher education March 2019, March 2020, and March 2021

The Geography of Daily Urban Spatial Mobility During COVID: The …

273

on the right of the x-axis as the share of the highly educated in the k = 500 neighbourhood increases. In all area types, there is a marked decrease in this more mobile cluster when the COVID months of March 2020 and March 2021 are compared with March 2019 but it is greater in the quintiles with more highly educated people. This is suggestive of a greater scope for residents of these areas for homeworking. In all places, however, the proportion of very local/stationary people grows with a symmetry between the near and far clusters in the quintile with the smallest share of highly-educated people, whereas in the more highly educated quintiles longer moves remain dominant despite their large falls. In Fig. 6, the distributions for visible minorities are presented, starting in the top left pane with places that have more minorities within their neighbourhood. The proportion of longer moves increases as areas have fewer minorities, except for the least dense category in the bottom-right pane. In all places, the arrival of COVID leads to a shift in mobility frequencies to the left to low/no mobility cluster. However, the proportionate decrease in the more mobile cluster is greatest in places with fewer visual minorities. The picture is one of general decreases but less so in dense visible minority neighbourhoods. Figure 7 considers spatial mobility in the context of different distances (in 10 km bands) from Stockholm city centre. These patterns are intriguing. Once again, there is a general switch of mobilities from right to left, from the mobile to the least mobile cluster. This is larger for the distance bands less than 20 km and smaller for the bands between 20 and 50 km. It looks like this is a reflection of the structure of opportunities in the city and the patterns of commuting to its centre with a decline in mobility to the city centre from neighbourhoods within this 20 km band—something also supported by the earlier maps of the geography of mobility change.

Fig. 6 Distribution of distances between origin and destination by distance to k-th nearest visible minority March 2019, March 2020, and March 2021

274

I. Shuttleworth et al.

Fig. 7 Distribution of distances between origin and destination by centrality (distance from Stockholm Central Station March 2019, March 2020, March 2021

We now turn our attention to mobility changes during a holiday, considering it in relation to the share of highly educated, visible minorities, and to distance band from the city centre respectively in Figs. 8, 9, and 10. Easter was chosen as the holiday because at the time of writing and analysis there had been two COVID Easters and only one COVID Christmas. In these figures it is very apparent that behaviour does not change over the holiday period in comparison with the non-holidays days that we analyse; people remain at or near to home in the COVID Easters more than in pre-COVID days. Further analysis reveals (not shown here), that Easter mobility is biased towards peripheral parts of the urban area and the water and green places of the Stockholm Archipelago. This is, in fact, the sort of pattern that would be expected given the recreational rather than the work use of space.

6 Conclusion Before considering the questions, the chapter began with, it is worthwhile considering its limitations, and also its wider background. The analysis is based on a repeated cross-section of phones that were active in Stockholm before and during the COVID pandemic. It is longitudinal in this sense, but not longitudinal in that individual phones were followed through time in a fully longitudinal design. Therefore, some phones may have left Stockholm during the pandemic, for example to second homes, but these are not observed; the chapter is concerned only with spatial mobility within the city itself and not the fortunes of residents who might have left it for longer or shorter periods during COVID times. Two additional and important points are made by this chapter beyond the questions set earlier. The first is that the rules introduced by the Swedish government led to

The Geography of Daily Urban Spatial Mobility During COVID: The …

275

Fig. 8 Distribution of distances between origin and destination by higher education Easter 2019, 2020, and 2021

Fig. 9 Distribution of distances between origin and destination by visible minority Easter 2019, 2020, and 2021

measurable changes in spatial mobility during the COVID pandemic. As indicated earlier, these relied on advice, discretion, and personal responsibility and tended to avoid the compulsory and wide-ranging restrictions introduced in other European states. Of course, these similar compulsory controls, if introduced in Sweden, might have led to even greater mobility decreases—we cannot observe this counterfactual in the analysis—but we do see a clear policy effect. More might have been done, however; as noted above, Stockholm had one of the highest COVID mortality rates of any European capital city and Fig. 2 is useful in understanding why this might be the case. De Castro et al. (2021), saw in Spain, that a reduction in the radius of gyration to less than 70% of its pre-pandemic level hampered the spread of COVID.

276

I. Shuttleworth et al.

Fig. 10 Distribution of distances between origin and destination by distance band from city centre Easter 2019, 2020, and 2021

If this threshold applies to origin–destination distances then, in Stockholm County, the value for work mobility is only met in the Spring of 2020 and then from January to March 2021. On the other hand, Stockholm Municipality remains below this limit for far longer. The evidence is thus mixed—more central parts of the urban area record bigger mobility falls, something also observed in the MIND phone data— with significant mobility decreases observed in neighbourhoods within 20 km of the city centre. The second main point is that the average decline observed in origin–destination mobility was not caused by a general change to more spatially restricted patterns close to home by all phones. Instead, it is caused by fewer phones (people) following prepandemic movement patterns with an associated increase in the frequency staying at or near home. Some phones, albeit fewer, range over similar distances before and during the pandemic and whilst we have no evidence for the causes of this, nor on the individual demographics of the people using these phones, it is hypothesised that some people were unable or unwilling to stay in or close to home. They accessed facilities, services, and opportunities and it is this fixed geography— the location of schools, shops, and workplaces are relatively stable and did not change through COVID—that shapes the continuity in the frequency distribution of origin–destination distances. We now deal with the questions that were set. Where was the greatest spatial interaction in the city? The place where there was the greatest interaction in Stockholm, whether in terms of numbers of phones, or numbers of areas contributing flows, was the city centre. This was true before and during the pandemic, especially on weekdays (but less so at Easter) with the major qualification that the flow of phones to central Stockholm fell during the pandemic as did the number of contributing places. Stockholm is monocentric and this remains, although the pandemic has made this less so.

The Geography of Daily Urban Spatial Mobility During COVID: The …

277

Were the changes long lasting? The evidence presented suggests that they are but variable across the Stockholm region. Google Mobility data shows this is true for Stockholm Municipality, but that for Stockholm County decreases in mobility were smaller and more variable. This is supported by the analysis of the MIND phone data, where the greatest decreases are within 20 km of the city centre—although they are general within a 50 km radius of the city centre. The results also indicate that behaviour changes persist through weekdays at different times of the year and over holidays like Easter too. Finally, were there differences in changes in origin–destination distances by socioeconomic characteristics? The answer to this question is qualified. Shifts in the frequency distribution of origin–destination moves happened in all areas. There was no place type where the frequencies remained the same. The biggest proportionate falls, however, were in places that had sparse visible minority populations and that had high shares of highly educated people. Therefore, to some degree, differentials were apparent in reaction to COVID advice. Given what is known about the differential health impact of COVID, with higher incidences for visible minorities and for the socially deprived (Drefahl et al., 2020), we question whether these differentials were epidemiologically significant. Again, we cannot conclusively answer this question with the MIND data, but the place variations that were observed directs attention to the possibility that spatial behaviour may have been important.

References Almlöf, E., Rubensson, I., Cebecauer, M., & Jenelius, E. (2021). Who continued travelling by public transport during COVID-19? Socioeconomic factors explaining travel behaviour in Stockholm 2020 based on smart card data. European Transport Research Review, 13(1), 31. https://doi.org/ 10.1186/s12544-021-00488-0 Andersson, S., & Aylott, N. (2020). Sweden and coronavirus: Unexceptional exceptionalism. Social Sciences, 9(12), 232. Coven, J., & Gupta, A. (2020). Disparities in mobility responses to COVID-19. Unpublished Manuscript. https://static1.squarespace.com/static/56086d00e4b0fb7874bc2d42,5,158 9583893816. Covid Government Response Tracker. (2022). Blavatnik School of Government, University of Oxford. Retrieved June 21, 2022, from https://www.bsg.ox.ac.uk/research/research-projects/ covid-19-government-response-tracker Dahlberg, M., Edin, P. A., Grönqvist, E., Lyhagen, J., Östh, J., Siretskiy, A., & Toger, M. (2020). Effects of the COVID-19 pandemic on population mobility under mild policies: Causal evidence from Sweden. arXiv:2004.09087 de Castro, A. H., Mateo, D., Bayer, J., Barrios, I. (2021). Radius of Gyration as predictor of COVID-19 deaths trend with three-weeks offset. medRxiv. Dingel, J. I., & Neiman, B. (2020). How many jobs can be done at home? Journal of Public Economics, 189, 104235. https://doi.org/10.1016/j.jpubeco.2020.104235 Drefahl, S., Wallace, M., Mussino, E., Aradhya, S., Kolk, M., Brandén, M., Malmberg, B., & Andersson, G. (2020). A population-based cohort study of socio-demographic risk factors for COVID-19 deaths in Sweden. Nature Communications, 11(1), 1–7. Dueñas, M., Campi, M., & Olmos, L. E. (2021). Changes in mobility and socioeconomic conditions during the COVID-19 outbreak. Humanities and Social Sciences Communications, 8(1), 1–10.

278

I. Shuttleworth et al.

Eurostat. (2021). Living conditions in Europe—Poverty and social exclusion. Retrieved February 3, 2022, from https://ec.europa.eu/eurostat/statistics-explained/index.php?title=Living_condit ions_in_Europe_-_poverty_and_social_exclusion/ Google Mobility Reports. (2021). Retrieved October 10, 2021, from https://www.google.com/COV ID19/mobility Granberg, M., Rönnblom, M., Padden, M., Tangnäs, J., & Öjehag, A. (2021). Debate: COVID-19 and Sweden’s exceptionalism—A spotlight on the cracks in the social fabric of a mature welfare state. Public Money & Management, 41(3), 223–224. Järv, O., Tominga, A., Müürisepp, K., & Silm, S. (2021). The impact of COVID-19 on daily lives of transnational people based on smartphone data: Estonians in Finland. Journal of Location Based Services, 1–29. https://doi.org/10.1080/17489725.2021.1887526 Kim, J., & Kwan, M. P. (2021). The impact of the COVID-19 pandemic on people’s mobility: A longitudinal study of the US from March to September of 2020. Journal of Transport Geography, 93, 103039. Lou, J., Shen, X., & Niemeier, D. (2020). Are stay-at-home orders more difficult to follow for low-income groups? Journal of Transport Geography, 89, 102894. Ludvigsson, J. F. (2020). The first eight months of Sweden’s COVID-19 strategy and the key actions and actors that were involved. Acta Paediatrica, 109(12), 2459–2471. Mongey, S., Pilossoph, L., & Weinberg, A. (2021). Which workers bear the burden of social distancing? The Journal of Economic Inequality, 19(3), 509–526. Östh, J. (2014). Introducing the EquiPop software: An application for the calculation of knearest neighbour contexts/neighbourhoods. Nedladdningsbart via: http://equipop.kultgeog.uu. se/Tutorial/Introducing%20EquiPop.pdf Östh, J., Shuttleworth, I., & Niedomysl, T. (2018). Spatial and temporal patterns of economic segregation in Sweden’s metropolitan areas: A mobility approach. Environment and Planning a: Economy and Space, 50(4), 809–825. Östh, J., & Türk, U. (2020). Integrating infrastructure and accessibility in measures of bespoke neighbourhoods. Edward Elgar Publishing. Pierre, J. (2020). Nudges against pandemics: Sweden’s COVID-19 containment strategy in perspective. Policy and Society, 39(3), 478–493. Pullano, G., Valdano, E., Scarpa, N., Rubrichi, S., & Colizza, V. (2020). Evaluating the effect of demographic factors, socioeconomic factors, and risk aversion on mobility during the COVID-19 epidemic in France under lockdown: A population-based study. The Lancet Digital Health, 2(12), e638–e649. Santamaria, C., Sermi, F., Spyratos, S., Iacus, S. M., Annunziato, A., Tarchi, D., & Vespe, M. (2020). Measuring the impact of COVID-19 confinement measures on human mobility using mobile positioning data. A European regional analysis. Safety Science, 132, 104925. https://doi. org/10.1016/j.ssci.2020.104925 Toger, M., Shuttleworth, I., & Östh, J. (2020). How average is average? Temporal patterns in human behaviour as measured by mobile phone data--or why chose Thursdays. arXiv:2005.00137 Toger, M., Kourtit, K., Nijkamp, P., & Östh, J. (2021). Mobility during the covid-19 pandemic: A data-driven time-geographic analysis of health-induced mobility changes. Sustainability, 13(7), 4027. University of Oxford. (2021). COVID-19 government response tracker. Retrieved October 7, 2021, from https://www.bsg.ox.ac.uk/research/research-projects/COVID-19-government-response-tra cker Warren, G. W., Lofstedt, R., & Wardman, J. K. (2021). COVID-19: The winter lockdown strategy in five European nations. Journal of Risk Research, 24(3–4), 267–293. Xu, Y., Belyi, A., Bojic, I., & Ratti, C. (2018). Human mobility and socioeconomic status: Analysis of Singapore and Boston. Computers, Environment and Urban Systems, 72, 51.

Social Justice, Digitalization, and Health and Well-Being in the Pandemic City Laurie A. Schintler and Connie L. McNeely

1 Introduction Pandemics have always been the “enemy of urban life.”1 Considering the crucial role of cities in regions and the international arena (Glaeser et al., 2020), pandemic disruption is a grand challenge and a local and global policy priority, with particular reference to economic and social impacts in cities and urban networks (OECD, 2020). A pandemic is a “an epidemic occurring worldwide, or over a very wide area, crossing international boundaries and usually affecting a large number of people,” and when pandemics sweep through urban settlements and systems, “they upend critical structures, such as health systems and medical treatments, economic life, socioeconomic class structures and race relations, fundamental institutional arrangements, communities, and everyday family life” (Pew Research Center, 2021). Accordingly, the health and well-being of the denizens of a city affected by pandemics represent an essential benchmark and objective for urban planning and performance. This issue becomes especially pertinent when recognizing that pandemics can be understood as natural disasters and social crises that lay bare and expand what often are already existing social justice challenges. In reference to cities, social justice concerns empowering oppressed and marginalized groups, eliminating barriers to access, achievement, and development at the local or community level (Coates & Heitzeg, 2008; Coates & Williams, 2007). Accordingly, in an urban context, conceptions of social justice underscore the inequalities and “uneven distribution of goods and burdens, opportunities and resources, found in most of the world’s cities,” bringing into question social determinants of health—i.e., “conditions in the environments where people are born, live, learn, work, play, worship, and age that affect 1

https://www.city-journal.org/cities-and-pandemics-have-long-history.

L. A. Schintler (B) · C. L. McNeely George Mason University, Arlington, VA, USA e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. G. Celbi¸s et al. (eds.), Pandemic and the City, Footprints of Regional Science, https://doi.org/10.1007/978-3-031-21983-2_15

279

280

L. A. Schintler and C. L. McNeely

a wide range of health, functioning, and quality-of-life outcomes and risks.”2 To this point today, the COVID-19 pandemic has emerged as a crisis and a social justice challenge for cities and their populations, operating “at the intersection between different dimensions of disadvantage and marginalization” (Haase, 2020). Indeed, the “novel coronavirus” has ravaged cities worldwide, bringing to “the fore the issue of urban vulnerability to pandemics” (Sharifi et al., 2020), expanding the lens on social justice in general. Inequalities in relation to COVID-19 outcomes are a contemporary example in a long history of health disparities and social justice concerns during pandemics. Pandemics over time have spread in keeping with urbanization processes and “cities have systematically metamorphosed in response to threats posed to health and other kinds of security” (Lai et al., 2020). Moreover, technological developments have been linked to the distribution of disease over time and, relative to this vision—especially given today’s increasingly “smart” cities—disparities in health point to broader social and digital asymmetries and inequities, with evidence indicating that structural inequality is a critical determinant of disease incidence and socioeconomic fallout (Barr et al., 2020; McNeely et al., 2020; Schintler & McNeely, 2019, 2020; Yaya et al., 2020). Indeed, the COVID-19 pandemic has occurred in this time of increased digitalization and automation, with associated health disparities and inequalities noted as reflections of broader references to socio-technological divides. More to the point, these divides are linked to asymmetries and inequities within and across cities, affecting people with particular health, social, and mobility needs, especially in under-served and under-privileged areas. Of particular note are also issues such as artificial intelligence (AI) and algorithmic applications as critical considerations in determining health outcomes, especially given their centrality in planning crisis responses in cities everywhere (Schintler & McNeely, 2022a, 2022b). Health itself is a “state of complete physical, mental, and social wellbeing” (WHO, 2008), encapsulating conditions involving not only physical and mental health, but also wider contextual factors, conditions, and relations. In this sense, health as a social justice issue can be framed relative to an overarching “culture of health” (Chandra et al., 2016), encompassing demographic, socioeconomic, and locational factors associated with different populations as significant predictors of disease transmission and effects. Against this backdrop, a broader understanding of the consequences of pandemic crises today, in both the short and long term, must reflect consideration of expanding digital engagement and effects as central to global and local societal processes. Accordingly, this research examines how digitalization and automation for pandemics affect public health and differentiating structural dynamics and relations in cities in the face of such crises. Given increasing calls for metrics beyond assessing AI and algorithmic performance to consider broader impacts on communities and society (Schiff et al., 2020; Stray, 2020; Musikanski et al., 2020; Dankwa-Mullan et al., 2021), an essential purpose here is to delineate measures that capture social justice, digitalization, and health and well-being in the pandemic city. To that end, 2

https://health.gov/healthypeople/objectives-and-data/social-determinants-health.

Social Justice, Digitalization, and Health and Well-Being …

281

after a brief discussion of the urban context in relation to pandemics as background, prominent theoretical and analytical issues relevant to research in this area are critically examined. This serves as the basis for a more integrated and in-depth exploration of AI, pandemics, and social justice, providing the foundations for metrics specification. As such, this research contributes to a cross-cutting literature on natural disasters and pandemics and on socio-technological interactions in terms of public health and well-being for promoting social justice in cities today and in the future.

2 Contextual Background Across time, pandemics have erupted as “natural disasters” that ravaged humanity worldwide (McNeill, 1998; Piret & Boivin, 2021). The spread of infectious diseases and the increased risk of pandemics are primarily due to human activities and environmental effects resulting from urbanization, land use, and climate changes (Piret & Boivin, 2021). Urban areas not only bear the brunt of pandemics, they are a catalyst of disease transmission and related problems in the first place. Population densities render cities targets for infectious disease, with concentrations of people providing ideal environments for contagion (Wong & Li, 2020). Moreover, cities are complex nuclei of human activity, including “services, government, education, commerce, markets, and finance” (Rybski & González, 2022), making them particularly vulnerable to the systemic dynamics and effects of a pandemic (Schweitzer, 2020). These systemic effects are further exacerbated given that cities are connected globally through vast and intricate transportation networks, acting as global conduits and transit points and tourist and business destinations (Glaeser, 2022; Reggiani et al., 2021). Compounding matters, the “space–time convergence” through technological developments in transportation across urban areas has made cities increasingly susceptible to disease transmission. That is, “with more and quicker global travel between city hubs, local outbreaks turn more easily into epidemics that can turn into global pandemics” (Martínez & Short, 2021). Thus, for example, within weeks of the first recorded case in Wuhan, China, COVID-19 had seeded and spread around the world—a world in which almost 60 percent of the population lives in cities (cf. Martínez & Short, 2021). Note that, in many cities, it was the more affluent global travelers, the rich cosmopolitans, who brought the disease (Martínez & Short, 2021). In Houston in the United States, for instance, “the virus came with international travelers,” first impacting the “white, affluent neighborhoods before moving onto the lower-income, minority communities” (Long et al., 2020; Martínez & Short, 2021). Similar scenarios were played out in cities across the world (Mooney et al., 2020). With urbanization driven in part by promises (or hopes) that urban centers offer better opportunities, jobs, quality of life, and access to better services and goods (Martínez & Short, 2021), people often flock to cities searching for economic and social advantage. However, pandemics typically operate counter to such promises. For example, the COVID-19 crisis saw dramatic shocks to the labor market in its first

282

L. A. Schintler and C. L. McNeely

few months, with skyrocketing unemployment rates—and with the greatest impact on already economically disadvantaged individuals. Moreover, in cities everywhere, infection rates of certain diseases are disproportionately high among socially disadvantaged and underserved groups, negatively impacting health and well-being and trapping associated individuals and communities in cycles of illness and poverty. Albeit with varying implications, this situation repeats in one way or another across urban areas within and across different countries (McNeely et al., 2020). Again, disadvantaged populations—e.g., the poor and racial/ethnic minorities, who are often disproportionately vulnerable to and negatively impacted by pandemics (Wade, 2020)—tend to be concentrated in cities (Adler et al., 2020). In this regard, for instance, residents of impoverished slums and informal settlements are especially prone to infectious disease outbreaks because of crowded and unsanitary conditions (Wahba et al., 2020), underscoring the social determinants of health. Also, whether referencing smallpox outbreaks, the bubonic plague, the 1918 influenza pandemic, more recent HIV/AIDS, SARS, or Ebola pandemics, and now COVID-19 and related variants, a long history of discrimination, blame, and violence escalated against people from marginalized groups has been evidenced during rampant disease outbreaks (Dionne & Turkmen, 2020). Along with explicit effects on health, pandemics have presented major social, economic, and financial consequences throughout the ages. Pandemic-instigated crises mark the diverse and profoundly interconnected character of societal health and well-being within and across cities and city networks.

3 Theoretical Issues and Analytical Perspective Public health has been described as “the science and art of preventing disease, prolonging life, and promoting health through the organized efforts and informed choices of society, organizations, public and private communities, and individuals” (Winslow, 1920). It is aimed at “fulfilling society’s interest in assuring conditions in which people can be healthy” (IOM, 2002). Accordingly, understanding health as a public issue requires looking beyond questions of medical care and of illness and disease as such. Although scientific research is continually being undertaken to combat disease, cities and the people within them continue to face uncertain times in relation to pandemics, reactions to which constitute an analytical project aimed at challenging social, political, and economic forces that determine societal injustices and disparities. A more critical and expansive perspective is required to better understand related morbidity and mortality rates and patterns, especially from the perspective of social justice, and to plan interventions accordingly to engage the particular challenges that pandemics present in urban settings (cf. McNeely & Schintler, 2020, p. 346). To that end, an integrated and encompassing “culture of health” perspective is engaged here as a lens through which to view the systemic nature of pandemics in the context of cities and related networks. In basic terms, a culture of health is “a set

Social Justice, Digitalization, and Health and Well-Being …

283

of social ideas and practices that promote healthy individuals, households, neighborhoods, communities, states, and nations. These shared ideas and practices are key factors for whether and how sectors work together to improve health environments” (Chandra et al., 2016, p.3). As such, a culture of health turns on social determinants of health, providing a basis for assessing relative pandemic effects as a social justice issue. Taking cues from history, city sustainability, resilience, and growth can be considered against the backdrop of pandemic processes. Pandemics have been characterized by initial establishment with high random effects, exponential infection growth, a peak, and subsequent die-down phases (which may be repeated, with surges and new forms). Noting this pattern, public health in relation to pandemic processes in city contexts involves interactive and organizing features that take place before (prevention), during (containment and mitigation via distancing and separation), and after (contingency planning and countermeasures for offsetting future risks) the general threat and impact (cf. Lai et al., 2020). More to the point, operating under conditions of risk and disturbance, cities are tasked with developing resilience and capacities to absorb or adapt to change and perturbations (Schintler & McNeely, 2022a, 2022b). Thus, this discussion is centered around responses to pandemics relative to institutional capacity strengthening, encompassing contributive and determinant technologies, tools, products, communication, organizations, and practices that together constitute understandings, enactments, and underlying supports and resources for healthy outcomes and societal well-being—a culture of health. Culture in this sense may be understood in terms of institutional and organizational properties that are open to change (Ormrod, 2003), and creating a culture of health would mean making health a shared value, fostering cross-sector collaboration to improve well-being, creating healthier and more equitable communities, and strengthening the integration of health services and systems.3 A critical feature in this regard is legitimacy, which is essential to ensuring that institutional and organizational structures and practices are meaningful and ultimately trustworthy (Suchman, 1995a, 1995b). As a cultural trait, legitimacy concerns the norms, values, and institutional imperatives that dictate how human activities and interactions should be conducted in pragmatic and moral terms. Of particular note in reference to pandemic relations and effects, legitimacy also operates as an organizational or systems-level phenomenon, shaped by context-specific values based on different experiences and circumstances within a given city. Keep in mind that cities involve social interactions, structures, and relations and, therefore, are held to certain values, expectations, and norms on which public trust depends, embedded in a kind of social contract between the city and its constituents. This contract outlines an ethical and moral responsibility of the city to ensure that its residents receive certain provisions and are treated in a just and fair manner. However, this is not a straightforward situation. It calls for transforming dominant and encompassing patterns of stratification and mobility that generate and reflect societal disparities; it calls for the disruption of institutionalized 3

https://www.rwjf.org/en/cultureofhealth/taking-action.html.

284

L. A. Schintler and C. L. McNeely

societal patterns of haves and have-nots; it calls for a rewriting of the social contract and realizing ideals of equity and well-being in practice and application…. However, the societal and institutional will to implement such changes will take a deep and abiding commitment beyond anything seen thus far. (Schintler & McNeely, 2020, p. 156)

Recognizing these conditions, the research here embraces a multi-faceted perspective on social, economic, cultural, and technological antecedents and effects related to key indicators and determinants of health and well-being and of related disparities and gaps, as laid bare in observance of pandemic outcomes. A culture of health is engaged as an interpretive platform for defining and analyzing related issues emphasizing questions of fair and equitable public and individual health, i.e., of social justice.

3.1 Social Justice and Social Determinants of Health As mentioned, cities encompass a diverse array of individuals and groups with differing capacities for achieving healthy outcomes and well-being (Acosta et al. Acosta et al., 2015); as in society in general, they are marked by disparities. In cities around the world, there are communities with “insufficient access to jobs, adequate transit, safe and affordable housing, parks and open space, healthy food options, or quality education— the necessary conditions and opportunities to fully thrive. This lack of opportunity is particularly evident in the disparities that exist in health status and health outcomes” (cf. NASEM, 2017 p. ix). When pandemic factors are added to the mix, putting additional pressure on already strained conditions, structural relations, dynamics, and outcomes can make for even starker divisions. Social determinants of health and well-being are varied and far-reaching, including for example the socioeconomic and built/physical environment, housing, employment, income/wealth, safety and security, and access to health-related systems, services, and activities (Davis et al., 2016; Lavizzo-Mourey, 2014). Accordingly, considering broad social, political, and economic impacts, focus here is on health disparities and the severe effects of pandemics on some populations—particularly already disadvantaged and underserved ones—and cities relative to others. COVID19, for example, has impacted the economy, education, workforce, and more, indicating its significance for broader questions of sustainable development, societal well-being, and social justice on a global level (Seshaiyer & McNeely, 2020). Beyond direct problems of public health and healthcare access, pandemics raise questions that touch upon all aspects of urban life, including housing, sanitation and hygiene, education access, public space, social and economic contacts, transportation, and social inequalities (Martínez & Short, 2021). In this regard, health equity, encompassing notions of social justice, is a principal reflection point for addressing well-being in cities in response to pandemics and other crises, indicating capacities to address and adapt to changing health conditions and needs. Pandemics can be viewed as providing opportunities for building safe and healthy public spaces, but, as history has shown, the application and realization of such opportunities are highly uneven—or unlikely.

Social Justice, Digitalization, and Health and Well-Being …

285

Cities are sites of deepening inequality, reinforced by pandemics which, in relative terms, might present a “minor inconvenience” to the wealthy, whereas poorer individuals can face economic collapse and the exacerbation of pre-existing disparities (Martínez & Short, 2021). As shown in Fig. 1, social determinants of health include a range of interrelated factors, including, for example, education, the built/physical environment, housing, employment, income and wealth, safety and security, transportation, technology, and access to health-related systems and services (Davis et al., 2016; NASEM, 2017). Such determinants are themselves shaped by inequalities attending the distribution of resources and power at local, national, and global levels of analysis, leading to health inequities across different areas and groups of people (Yaya et al., 2020; Barr et al., 2020; IOM, 2006; NASEM, 2017). Moreover, a diversity of actors, sectors, settings, and stakeholders interact and operate in the urban context to shape related outcomes (WHO, 2008; NASEM, 2017), adding layers of complexity to the factors that determine health disparities (McNeely et al., 2020). In a world characterized by inequality and inequity, pandemics such as COVID-19 have exacerbated the marginalization of already disadvantaged groups. In reference to social justice, vulnerable communities comprise population subgroups of, for example, low social and economic status and specific ethnic or racial minority group designations, and stigmatization and marginalization concerns are crucial aspects of “emerging infectious disease for which there is greater uncertainty and greater fear and underlying anxiety among the public” (Dionne & Turkmen, 2020). Factors associated with disproportionately poor health outcomes—e.g., lower levels of education, digital divides, lower incomes, lower quality care, high risk jobs, inadequate and unsafe housing, and living near environmental hazards (Thomas, 2022)—are especially rampant among under-served and disadvantaged populations, which is

Fig. 1 Social determinants and outcomes of health (Modified from Solar & Irwin, 2010)

286

L. A. Schintler and C. L. McNeely

often the case for many racial and ethnic minorities. Reflecting societal bias and socioeconomic differentials, such situations target and predispose these individuals and groups to higher pandemic morbidity and mortality rates. Thus, for example, as in prior pandemics, the health, social, and economic impact of COVID-19 has been greatest on already disadvantaged and vulnerable populations, “aggravating an already very fragile situation characterized by these grave inequalities,” such that racism has emerged as “a monumental COVID-19 challenge” (Yaya et al., 2020). Often lacking healthcare insurance coverage and adequate healthcare access, racial fault lines are marked by persistent health disparities, illness incidence, and death (Williams et al., 2019; Yaya et al., 2020), which are exacerbated in the face of pandemic conditions. Accordingly, pandemic effects and outcomes must be identified and understood within related historical, political, economic, and social contexts. Focusing on moral and pragmatic legitimacy concerning issues of bias and discrimination that mobilize against social justice can have potentially profound implications for pandemic-induced health disparities and challenges to well-being. Social justice speaks to how people are treated, the distribution of resources, and access to services and opportunities, which are crucial considerations in determining health and other pandemic-related outcomes (Soken-Huberty, 2021). At its most basic, social justice is grounded in principles of access, equity, participation, diversity, and human rights (KSUPA, 2020; Soken-Huberty, 2021). It can apply to different levels of analysis, whether “a city, a state, or a continent, or span the entire world” (Culp, 2018). As an institutional construct, it concerns changing, eliminating, or re-designing the supports that maintain exploitative systems of discrimination and oppression (Coates & Williams, 2007). Looking to social justice in terms of legitimacy in urban relations involves casting a wider net in reference to promoting health and preventing and treating disease—and, ultimately, building trust in the city relative to pandemic effects and social determinants of health. Pandemics reveal differences in healthcare access and resources, constituting issues of social justice as fundamental determinants of health (McNeely & Schintler, 2020). They pose substantial challenges to progress toward sustainability, relative poverty reduction, and better governance (Seshaiyer & McNeely, 2020). For example, “cities in lowerincome countries are less prepared to face the heath crisis and economic aftermath of pandemics. The population in lower-income countries have higher malnutrition rates, present higher rates of co-morbidities, and the health system does not receive the same level of resourcing” (Martínez & Short, 2021). World poverty is “vastly concentrated in urban areas” and, without specific countermanding policies and efforts, similar tales can be expected both across and within the stratified realities of urban domains (Martínez & Short, 2021). Pandemics aggravate such deep-rooted problems, flying in the face of social justice claims. Arguably, “health inequalities that are avoidable and are not avoided are unjust. Putting them right is a matter of social justice” (Marmot, 2017). Involving sociospatial processes that discriminate against, marginalize, and exclude some members of society (Harvey, 2008; Heynen et al., 2018), social justice involves morality and human rights, where a right is an “obligation embedded in some social or institutional context where expectation has a moral force” (Smith, 1994, p. 36). While multiple

Social Justice, Digitalization, and Health and Well-Being …

287

disadvantages may vary in specifics relative to socio-spatial and temporal factors, their common ground is found in the intersectional disadvantage that inequality, poverty, and attendant features indicated in urban public health and social crises. Intersectional disadvantage is a global phenomenon and is the “face” of the urban pandemic crisis (cf. Haase, 2020; Paremoer et al., 2021). It affects seasonal workers in Europe and India, precarious workers who lost their jobs in the US, people living in crowded and poor areas without access to clean water, space for distancing or access to open/green space in the midst or at the fringes of our cities, refugees and people without a secure legal status, homeless people, etc. (Haase, 2020).

Even more than past pandemics, COVID-19 has exposed long-standing systemic and structural drivers of health disparities, e.g., adverse working conditions, growing economic gaps and inequalities, and related political and institutional processes. Such issues have an intersectional relationship with, for example, race, class, ethnicity, migrant status, gender, education level, and other factors that interact relative to social dynamics and relations, operating to exacerbate existing social vulnerabilities and undermine health (cf. Haase, 2020; Paremoer et al., 2021)—in other words, that drive systemic inequalities in health outcomes. Although typically invoked as a matter of fairness and ethical practice in relation to improving the life chances and living conditions of historically marginalized populations, it more fundamentally is aimed at the general well-being of the society and of all individuals and groups. Still, conflicting ideas about fairness and the allocation of resources, and about power and the application of rights, may be highly contested and denote divisive tensions attached to implementing social justice and bringing about a more just and overall beneficent society. With pandemics, social justice and related processes are brought into sharp relief, emphasizing areas of critical need and inequity (McNeely & Schintler, 2020). A culture of health by definition incorporates a social justice approach applied to social determinants of health, including contributive and affective technologies, tools, products, communication, organizations, and practices that together constitute the habitus and praxis that underlie supports and resources for health and societal well-being in the urban context. Indeed, against this backdrop in today’s world, the COVID-19 pandemic is accelerating the trend toward smart cities (Pietras et al., 2021). In this sense, technology and digitalization as social determinants of health and of social justice can provide an analytical leverage point for understanding and gaining insights on pandemic effects in a world of increasingly smart cities.

3.2 Digitalization and Pandemics In today’s technology-driven, digitalizing, and urbanizing world, public health remains a key priority for governments everywhere. Digitalization itself is a multi-vector process of technological transition that affects all spheres of public

288

L. A. Schintler and C. L. McNeely

life (Mikheev et al., 2021). Indeed, digitalization has been disruptive, with societal impacts and new challenges in personal and institutional arenas—e.g., social networking and government transparency—changing expectations, values, and needs (Mikheev et al., 2021). Societal issues such as privacy and data protection, especially in health and education sectors, have invoked both technical and ethical concerns. In regard to cities and pandemics in particular, change in the characterization of networks as increasingly digitalized relations in addition to or in replacement of physical connections can affect epidemiological outcomes and patterns. Also, the nature of work in some fields and positions has been affected by technological developments to the effect that physical presence in the workplace is less required or expected. This trend was occurring before COVID-19 erupted, but has increased more quickly and in more areas due to pandemic effects, possibly lessening density impacts over time for some workers. Still some caveats exist even in related fields. For example, “manufacturing can be offshored” and “basic information processing can be done almost anywhere with an internet connection. However, turning information into knowledge and ideas into products is best done in dense cities” (Martínez & Short, 2021). Digitalization and automation historically have contributed to variations across cities and regions in wages, productivity, and growth and development (Acemoglu & Restrepo, 2017; Florida & Mellander, 2016). Given varying demographics, occupational and industrial compositions, socioeconomic characteristics, and other relevant characteristics in different areas, growing automation is likely to reflect inequalities in pandemic response capacities and effects. Related conditions can be viewed relative to health literacy, especially in relation to culture of health concerns. Health literacy—concerning capacities for effective and efficient uses of information and other resources in health and healthcare—reflects a critical disconnect between healthful and non-healthful conditions. Thus, for example, even with technological access to information (e.g., computers and the Internet), people may be “limited by a knowledge gap that separates them from any meaningful understanding of issues shared” (cf. Erdiaw-Kwasie & Alam, 2016). Differences in related skills can vary in any case, but individuals from more privileged backgrounds tend to be more able to access and be in positions to reap potential benefits of technological developments, relative to those from more disadvantaged backgrounds (cf. Hargittai & Jennrich, 2016). In this regard, health disparities are indicators of broader sociotechnological divides linked to societal asymmetries and inequities in health and other social conditions in under-served and under-privileged communities and areas (McNeely & Schintler, 2020). In reference to cities, technology typically is framed in instrumental terms, expressed relative to fulfilling functional applications for different purposes in society, regardless of social or cultural context (Ahlborg et al., 2019). As such, technologies can be depicted relative to public health for preventing, managing, containing, and recovering from pandemics and addressing the distinct challenges they face in the urban context. More specifically, technologies have played a vital role in disease surveillance, social distancing, health communications, and epidemiological forecasting in and across urban areas, both intentionally and fortuitously. For

Social Justice, Digitalization, and Health and Well-Being …

289

instance, in the 1918–1920 influenza pandemic, the telephone provided the means for remote communication (including distance education) for some of the people in quarantine (McCracken, 2020). Pandemics themselves can be framed as engines of innovation, catalyzing new technologies and related applications in the interest of containing and mitigating the spread of disease and detrimental effects on people and places. For example, epidemiological cartography was born out of the cholera pandemic and efforts to map the spatio-temporal prevalence of the disease, which also led to the discovery that cholera was caused by exposure to contaminated water (Shiode et al., 2015). In more recent pandemics, such as the 2013 SARS outbreak and Ebola, digital technologies, platforms, and artifacts have played a crucial role in the fight against them (Tan et al., 2022). It is possible that related technological developments can be used to improve outcomes depending on application goals and means. Health capacities can be directly linked to smart-city sensor technologies that offer surveillance and contact tracing, which are key to determining risk and fighting pandemics and other diseases, enabling rapid response by individuals and policy makers. Technological changes have made possible many of the features defined within a culture of health and can facilitate the realization of a vision of general well-being (Schintler & McNeely, 2020). However, framed in more intrinsic terms and noting social justice concerns in the urban context, technological and social interactions are fundamental factors affecting individual and community capabilities within a city to survive or adapt in the face of various stressors or disturbances such as a pandemic (Irani & Rahnamayiezekavat, 2021; Spaans & Waterhout, 2017). From this perspective, cities and city networks are conceived as increasingly integrated sociospatio-technological systems around institutionalized rules that affect the units within them. Accordingly, the decisions made and resources mobilized in cities are shaped, at least in part, relative to structures and strategies prescribed in the institutional context and wider system. In this sense, the city itself is an institutional and systemic actor (McNeely & Schintler, 2020). Smart cities in particular engage technologies that raise various legitimacy questions and ethical issues concerning privacy, datafication, surveillance, profiling, social sorting, and anticipatory governance, with significant consequences for how residents are treated—and can work to reproduce and reinforce inequalities among them (Kitchin et al., 2019). Smart cities rely on digital networked technologies and the production and analysis of big data (Kitchin et al., 2019), and smart city technology data and analytics can directly and indirectly affect the everyday lives of people. For example, they are applied to create data profiles that are used to socially sort and redline populations and places, “selecting out certain categories to receive a preferential status and marginalizing and excluding others” (Kitchin et al., 2019). Related views on smart cities emphasize the “people side” of technology and features such as “community capacity, social and human capital, knowledge inclusion, participation, social innovation, and social equity” (Arafah & Winarso, 2017). This points to a more systemic perspective, looking to how socio-technological interactions can “alleviate and/or reproduce complex and persistent problems” (McPhearson et al., 2021). Indeed, “as it becomes more ingrained in human life, technology itself will

290

L. A. Schintler and C. L. McNeely

rewrite traditional notions of ethics and social contract,” resulting in “fresh challenges for policymakers and technologists alike” (Saran, 2017).

4 Artificial Intelligence, Pandemics, and Social Justice in the City As part of the trend toward smart cities, Artificial Intelligence (AI) techniques, including machine learning, robotics, computer vision, and natural language processing coupled with big data, are essential components of the urban “health informatics toolkit,” including for pandemic management (Leslie et al., 2021). AI generally refers to “an artifact able to acquire information on the surrounding environment and make sense of it, in order to act rationally and autonomously even in uncertain situations” (Cugurullo, 2020). While the technology is certainly not new, the current generation of AI is more sophisticated in form and function, with greater capabilities for learning, perception, decision-making, prediction, automatic knowledge extraction and pattern recognition, interactive communication, and logical reasoning (Vinuesa et al., 2020). In light of these capacities, AI techniques are well-suited to address the challenges and tasks that come part and parcel with a pandemic (Schintler, 2021a). Machine learning can make extrapolations, glean insights, and quickly discern and anticipate patterns, anomalies, and other features from massive data points. Thus, it is an effective tool for “intelligent” disease surveillance, as well as onthe-fly forecasting. New developments in computer vision, including facial recognition systems, enable savvy forms of surveillance—e.g., detecting infected individuals and enforcing pandemic policies and interventions based on images of “faces in the crowd.” Also, innovations in natural language processing and AI voice recognition give robots—in both physical and virtual guises—the ability to communicate with humans, augmenting human-to-human interaction simultaneously. AI automation then enables various “tele-activities”—i.e., tele-working, tele-health, tele-education, and tele-services—facilitating business continuity and physical distancing even in densely populated cities. However, while AI has an instrumental role in anticipating and controlling pandemic disruptions in cities, it also comes with ethical downsides and dangers that can have detrimental effects on the well-being of urban communities, bringing into focus the intrinsic aspects of the technology. In general, AI has the propensity to produce unfair, erroneous, and unexplainable outcomes and decisions, violate privacy expectations, and instigate social isolation and polarization (Leslie, 2019). In this regard, marginalized and vulnerable individuals and groups tend to be disproportionately negatively affected (Obermeyer et al., 2019; Schintler & McNeely, 2021). Referring to Fig. 2, the application of AI to pandemics in cities can contribute to and amplify health and socioeconomic inequities and disparities—and social injustice more broadly, especially vis-a-vis its impact on and interaction with the social determinants of health (Dankwa-Mullan et al., 2021). As such, the roots of

Social Justice, Digitalization, and Health and Well-Being …

291

Fig. 2 AI-driven health and well-being in the pandemic era. Source Authors (Original)

health inequality and inequity in the era of AI may be more aptly viewed as “moral determinants of health.”4

4.1 Social Justice Implications and Provocations Unlike most prior major public health crises, responses to COVID-19 have been largely mediated by AI digital technologies. Shortly after the virus was detected in China, Boston Children’s Hospital and the Canadian company Bluedot deployed an AI system with health news and social media data to predict “cities at most risk if people were to travel,” forecasting the spread of the novel coronavirus “days before the WHO and weeks before the rest of the world caught up” (Chakravorti, 2022). Since then, AI has been on the frontline in the war against the pandemic, employed in various ways to help contain the crisis and its adverse effects on urban communities (OECD, 2020). Accordingly, the COVID pandemic is a useful focal point for examining social justice implications and provocations of AI-driven urban pandemic management. This issue is pertinent to pandemics more generally, considering that AI will inevitably continue to be a public health strategy. Real-time situational awareness about current and imminent threats and conditions is imperative for effective response to pandemic developments (Zwitter & Gstrein, 2020), especially in urban areas where the disease contagion tends to be rampant. 4

Although this term was introduced by Berwick (2020), it is used in in a different way here.

292

L. A. Schintler and C. L. McNeely

Although AI surveillance and big data can help fulfill this need, their use in this context raises various privacy concerns that are incongruent with social justice principles. Privacy, or the right to “freedom from interference or intrusion”,5 is considered a fundamental human right and moral entitlement. More to the point, it encapsulates multiple dimensions relating to “freedom from outside interference with personal decisions; seclusion, solitude, and bodily integrity; confidentiality, anonymity, data protection, the secrecy of personal facts; and limits on the use of a person’s name, likeness, or other attributes of identity.”6 However, AI can be used in ways that challenge and violate these freedoms and rights. Of course, AI surveillance was pervasive in cities well before the COVID-19 pandemic (Kitchin et al., 2019), and thus the idea that one can “vanish in the crowd” in an urban space has long been a mere illusion, as have privacy concerns in relation to the use of AI in this context (Schintler & McNeely, 2020). However, a pandemic opens the door for more excessive and invasive surveillance, which is often justified along the lines of promoting and preserving public safety and the urgent need to control the unrelenting spread of the disease, along with other related interests such as international security and economic well-being (Newlands et al., 2020). In fact, the coronavirus pandemic has led to profound changes in the character of surveillance, as well as the analytical tools behind it, raising privacy concerns across the board (Bentotahewa et al., 2021). Some of the most consequential innovations are those that use machine learning and automated decision making to, for example, “parse people’s digital footprints, identify those who are potentially infected, trace their contacts, and enforce social distancing” (Calvo et al., 2020a). Thus, the privacy dangers of surveillance now go beyond the data that are collected, but also the AI analytics running behind the scenes to process the data and make decisions based on it. The COVID-19 crisis has also expanded the scope of surveillance: it is no longer just about what people do or where they go but also how they feel emotionally and function biologically; surveillance has shifted from “over the skin” to “under the skin” (Harari, 2020). Automated thermal detectors, which have been used extensively in the pandemic for screening individuals for fever (Brzezinski et al., 2021), are a prime example of this new form of surveillance—i.e., “disease espionage.” However, at what point does it go too far? What is next? For instance, could ingestible sensors for monitoring the gut microbiome be used in the future to flag individuals who may be especially prone to contagious diseases? Who would have access to that information? How would that information be used? The systemic nature of pandemics and urban areas themselves complicates matters further regarding AI surveillance and privacy. Indeed, COVID-19 has highlighted just how intermingled public health is with other functional components of cities— e.g., economic, social and environmental systems, as well as how connected cities are locally and globally. Thus, siloed and insulated approaches to urban pandemic management are not practical or effective. Instead, what is needed is a “wholesystem” strategy, calling for confronting the pandemic crisis in a holistic manner 5 6

https://iapp.org/about/what-is-privacy/. http://livewithai.org/ai-surveillance-human-rights-privacy/.

Social Justice, Digitalization, and Health and Well-Being …

293

using systems thinking, methods, and practice.7 However, this strategy requires data and information sharing within and across sectors, organizations, and cities—and, ultimately, centralized databases, which are especially vulnerable to security and privacy breaches. Moreover, integrating different data sources makes it easier to “connect-the dots” about people and communities (including neighborhoods). For example, data collected through digital contact tracing could “be paired with existing public and corporate datasets to reveal intimate details of people’s private lives, including their political leanings, sexual orientation, gender identity, religious beliefs, and whether they receive specialized forms of health care” (Shabaz & Funk, 2020). The inclusion of geospatial data compounds the issue, as sensitive information about individuals and aspects of their lives can be revealed based on location choices and related activities in space (Schintler & Chen, 2017), even when the underlying data is course-grained or anonymous (Schintler & Fischer, 2018). “All data with location stamps (which is most of today’s collected data) is potentially very sensitive and we should all make more informed decisions on who we share it with,” doing it in a way that provides “adequate guarantees to preserve privacy” (Ratti, in Malewar, 2018). Possibilities for the repurposing AI surveillance data during and after a pandemic raises additional privacy concerns, as well as possibilities for analytical misdirection. In particular, the use of digital tools for monitoring people and the spread of disease can lead to “surveillance creep,” where data initially collected for public health functions is later used for other purposes (e.g., marketing, policing, political targeting, etc.), a situation which flies in the face of “widely held commitments to privacy, autonomy, and civil liberties” (Leslie, 2020). For example, digital contact tracing “apps” and other forms of surveillance used for tracking the spread of disease through a population may be later exploited for law enforcement purposes. Also, such activities can lead to “control creep,” whereby the social control apparatus used to advance public health (e.g., enforcement of social distancing policies) “progressively expands and penetrates (or ‘creeps’) into different social arenas” in response to imagined or manufactured fears regarding security or health and well-being (Innes, 2001). Additionally, the privacy risks tied to using AI for urban pandemic surveillance tend to differentially impact certain vulnerable or disadvantaged segments of the population, as previously discussed, more negatively than others (Li et al., 2018), which has further implications for social justice. A case in point is the pandemicinduced use of “panoptic technologies” in the workforce, which has “ballooned in an unprecedented fashion” during the COVID-19 crisis and is causing “a reconfiguration of power relationships in professional settings” (Aloisi & De Stefano, 2021). In a major disease outbreak, “essential workers”—“those who conduct a range of operations and services that are typically essential to continue critical infrastructure operations,”8 and who tend to be minorities, foreign born, and low-wage earners 7

https://www.publichealthscotland.scot/our-organisation/about-public-health-scotland/suppor ting-whole-system-approaches/applying-a-whole-system-approach/. 8 https://www.ncsl.org/research/labor-and-employment/covid-19-essential-workers-in-the-states. aspx.

294

L. A. Schintler and C. L. McNeely

in the first place (Lancet, 2020)—are generally employed in occupations that use aggressive forms of surveillance in the workplace, such as manufacturing and warehousing.9 In fact, “invasive and exploitative workplace surveillance” is becoming the “new normal” for employees in such jobs, especially when they have weak collective bargaining power and “a lack of legal protections or regulatory restrictions on these behaviors.”10 At the same time, teleworking, which now tends to involve a better off and more educated slice of the workforce (Marshall et al., 2021), is also subject to surveillance, albeit in a “cyber cloak.” For instance, in COVID-19, companies have resorted to the use of AI software for monitoring teleworkers’ Internet and computer activities, such as “the time spent online, the number of keystrokes on the keyboard, and the list of websites visited” (Aloisi and De Stafano 2021). However, these workers tend to be digitally literate, with “the ability to apply strategies for individual online privacy regulation and data protection” (Li et al., 2018). They also fare better than the aforementioned essential workers in the pandemic in terms of overall health and well-being (Marshall et al., 2021). Such circumstances are contributing to a growing polarization between teleworkers and essential workers in terms of key dimensions of the social determinants of health, potentially adding to the scope, scale, and speed of long-standing economic and social inequality and vulnerabilities (Gilmore et al., 2010). On top of potential injustices and compromised rights, AI surveillance can have psychological consequences, which can degrade a person’s sense of “volition and choice” and lead to an evasion of (aversion to) situations where there is believed to be surveillance—e.g., avoiding testing because of the use of contact tracing apps that track movements of individuals (Calvo et al., 2020a). In such circumstances, minorities and the poor, who may experience less access to healthcare and higher pandemic morbidity and mortality rates (Leslie et al., 2021; McNeely et al., 2020), are particularly vulnerable. Moreover, not only do such groups face greater challenges regarding privacy, but they are also more likely to opt out of digital opportunities for fear of privacy and other personal violations (Li et al., 2018). The propensity for AI to produce biased and discriminatory outcomes and decisions raises further concerns in relation to social justice (Calvo et al., 2020a). Algorithmic bias can “emanate from unrepresentative or incomplete training data or the reliance on flawed information that reflects historical inequalities” (Leslie et al., 2021), or from the context (including social rules and values) being analyzed not being adequately understood or captured in the model in the first place” (Schintler & McNeely, 2022a, 2022b). In fact, as in the point of repurposed data mentioned above, while an AI system may perform well in one context, if applied to another situation, it can produce “transfer context bias,” with erroneous (and potentially dangerous) outcomes (Ahlborg et al., 2019). 9

https://observer.com/2020/06/amazon-artificial-intelligence-monitor-warehouse-worker-socialdistancing-coronavirus. 10 https://www.ncsl.org/research/labor-and-employment/covid-19-essential-workers-in-the-states. aspx.

Social Justice, Digitalization, and Health and Well-Being …

295

Big data, which AI models rely on for training, testing, and validation, tends to be fraught with an assortment of imperfections and biases, stemming from technical issues as well as broader societal factors and dynamics, such as the digital divides (Schintler & Fischer, 2018). In pandemics, there are often profound gaps and disparities in the production and consumption of public health data. For example, COVID19 testing remained sparse in many low-income and minority neighborhoods, and demographic information was missing in much of the data (Servick, 2020). Such blemishes and imperfections get “baked” into AI systems, especially in the absence of appropriate strategies for de-biasing data or selecting algorithms in the first place. As previously highlighted, a pandemic is not an equal opportunity event by any stretch of the imagination (Stiglitz, 2020). Indeed, “pandemics are not strictly a microbial phenomenon; they follow the contours of the society that the contagion encounters” (Moss & Metcalf, 2020), disproportionately impacting already disadvantaged and impoverished populations, particularly in urban areas, as has been the case with COVID-19 (McNeely et al., 2020). The combination of the disproportionate impact of a pandemic like COVID-19 on vulnerable communities and the socio-technological determinants of algorithmic bias and discrimination can deliver a “brutal triple punch” in terms of social justice (Leslie et al., 2021). This raises a compelling question (cf. Leslie et al., 2021): Does AI stand for “Automating Inequality” in the pandemic? Indeed, while AI has been central to the fight against the novel coronavirus, “the existing issues raised by the use of the technology persist, as the technology, like the pandemic itself, highlights and threatens to exacerbate existing social inequalities,” working against social justice principles (AAAS, 2021). AI has discriminatory effects as well and, in this regard, historically vulnerable populations and marginalized groups are the most susceptible to such harms (Akselrod, 2021). Algorithmic profiling, or the “systematic and purposeful recording and classification of data related to individuals” (Büchi et al., 2020), which is typically used for allocating scarce resources and can result in “social sorting and other discriminatory outcomes” (Mann & Matzner, 2019). In a pandemic, given that people are “being tracked and traced at different scales of spatial and temporal resolution,” they are especially prone to “geo-targeted profiling and social sorting,” which is a concern that has been raised about smart cities more generally (Kitchin, 2016). Further, this can translate to an inequitable allocation of healthcare resources and other basic needs (Haas, 2020), compounding already existing differentials within and across cities (Schintler et al. 2020). In other words, AI for pandemic management can “solidify and amplify societal discrimination,” further deepening health and social inequities. A complicating factor is the increasing involvement of “Big Tech” in urban planning and pandemic management. While the private sector has long played a hand in urban public affairs, although to varying degrees in different countries, technology companies and firms become even more involved in a public health emergency like a pandemic, given that they often have relevant data and technological resources, tools, and skills to mobilize quickly. However, commercial interests, biases, motivations, and values tend to align more with profit maximation than social objectives and principles (Schintler, 2021b; Choi and McNeely, 2022). Accordingly, the AI systems

296

L. A. Schintler and C. L. McNeely

they develop may not be programmed with the social determinants of health or social justice in mind. Furthermore, they are often opaque for proprietary reasons, raising issues related to access and accountability in this context. In general, a pandemic puts a lot of power in the hands of big technology companies, which can fuel and perpetuate social and economic inequalities in cities, regions, and countries. Finally, AI automation in a pandemic crisis raises even more issues regarding social justice in the urban context. COVID-19 has led to the unprecedented utilization of automated solutions for urban pandemic management. For example, AI-enabled robotics (e.g., “chatbots,” virtual assistants, and droids) have been instrumental in facilitating “virtual learning among students and scholars, online purchasing of merchandise and contactless payments, and delivery of goods and services (Renu, 2021). However, one of the dangers of AI robotics is that it can lead to “new forms of automated social control.” In particular, automation tends to reduce human-to-human interaction and to manage, monitor, and control movement, which is typically the case in pandemic applications (Chen et al., 2020). Automation also can contribute to technological unemployment, a problem that bubbles to the surface in every period of disruptive technological change. In fact, there is already evidence that automation tied to COVID-19 is exacerbating job loss and economic inequality (Stiglitz, 2020), which has obvious implications for the social determinants and distribution of health and well-being.

5 Critical Imperatives in Metrics Development As alluded to in the previous section, without placing equity, justice, and ethics as central concerns for AI in urban pandemic management and planning, such tools may exacerbate disparate health outcomes and social and economic inequities. As such, appropriate governance tactics and institutional arrangements—e.g., standards, policies, and legal and regularity frameworks, along with directed technical solutions—are needed, especially to ensure that AI promotes rather than undermines principles of social justice. To help inform the formulation, implementation, and assessment of governance strategies, as well as the ongoing use and deployment of AI for managing pandemics, a significant priority is the development of reliable and relevant metrics and evaluation benchmarks. Existing solutions are inadequate in that they fail to address the complexities and dynamics of AI for pandemics in the context of cities and systems of cities, including social justice ramifications relative to the social determinants of health (Fig. 2). However, addressing this problem poses an array of issues and challenges (Calvo et al., 2020b), necessitating novel approaches and “outside of the box” thinking. To that end, ten imperatives for and reflections on the design and utilization of metrics for gauging the social justice implications of AI for urban pandemic management are indicated. First, metrics are needed that go beyond measuring technocratic aspects of AI for pandemic management. In general, discussions about the benefits and downsides of AI for urban disruptions and crises such as those stemming from pandemics

Social Justice, Digitalization, and Health and Well-Being …

297

are narrowly framed in instrumental terms, considering primarily engineering and operational issues such as speed of recovery and shock absorption (Schintler & McNeely, 2022a, 2022b). That is, the legitimacy and evaluation of AI in this context are based on whether it solves a problem, rather than whether it is the right thing to do. Thus, it is important to draw upon not only metrics that capture the accuracy of AI systems and, in an automation context, effects on business continuity and related activities, but also those that address them in terms of equity and justice. Although the reliability of AI may also have implications for fairness, it needs to be translated accordingly. Second, a consistent and agreed upon set of metrics is needed for measuring the fairness of AI, particularly when applied in an urban pandemic context. While there is a plethora of metrics for assessing algorithmic fairness, they capture different forms and manifestations of bias and discrimination and have the propensity to produce contradictory and conflicting outcomes (Mehrabi et al., 2021). Such metrics generally encapsulate only aspects of distributive fairness, emphasizing the equitable allocation of resources, whereas procedural fairness, or the “perceived fairness of the process that leads to the outcome” (Tomašev et al., 2020), may be an equally important consideration. Of course, all of this raises a thorny philosophical question: What is meant by equity and fairness in the first place? While that may speak to the moral dimensions of the issue, in more pragmatic terms, what does fairness mean in a pandemic, especially given that public health often trumps rights and liberties in such situations? Third, a way is needed to convert the immediate output of an AI system to higherorder effects on social and economic inequities, gaps, and other aspects of social justice in cities, also considering how they vary spatially within an urban community (e.g., by neighborhood). In fact, this has been raised as a critical priority for AI development and evaluation in general, as mentioned earlier. Metrics currently used to assess AI tend to reflect principally the computed results of an algorithmic system, ignoring the broader real-world context in which the tools were developed and are used and positioned. Also important is the assessment of repurposed applications in public health, particularly vis-a-vis to disparities tied to privacy breaches and potential human rights violations. To address these issues, methods such as simulation and agent-based modeling may be useful for anticipating related effects, as well as Geographic Information Systems (GIS) for mapping outcomes to people and places within a city. Fourth, the ability is needed to measure not only short-term effects of AI on cities (e.g., during a pandemic) but also longer-term implications (i.e., after a pandemic subsides). Such metrics must capture both fast dynamics and slow dynamics. Yet, this is challenging for a few reasons. One issue is that AI technology advances and morphs rapidly, especially during a pandemic, as witnessed with COVID-19. Thus, it is critical to be able to foresee how AI innovations and their applications to urban pandemics might evolve over time, which may require the use of technology visioning sessions with stakeholders and analysts such as computer scientists, public health experts, and urban planners and managers. Another challenge is that a pandemic is highly fluid, with circumstances and priorities constantly changing. It is essential to

298

L. A. Schintler and C. L. McNeely

ensure that the metrics are flexible and adaptive and, in this way (dare we say), semior fully-automated approaches for measuring the impact of AI on social justice may be in order. Fifth, metrics are needed that are grounded in a complex, integrated frame of reference (i.e., whole-systems perspective) to characterize the systemic aspects of AI for urban pandemic management. Indeed, as emphasized, the association between AI and social justice is multi-dimensional and dynamic, shaped by and shaping social determinants of health. As aforementioned, there are multi-level and hierarchical effects and patterns of interaction that play out in cities and systems of cities regionally and globally. However, while metrics that capture varying levels of spatial and temporal resolution and stratifications may be desirable, care should be taken to avoid Modifiable Aerial Unit (MAU) and Modifiable Temporal Unit (MTU) problems,11 as they can lead to deceptive measurements and consequences. Sixth, appropriate metrics are needed for use in all stages of the AI lifecycle, from the inception of systems to their final use and application, particularly to ensure that health disparity concerns and related social justice matters (and values) are properly reflected in the process (Dankwa-Mullan et al., 2021). This, of course, requires involving various stakeholders, e.g., Big Tech entities, as well as engaging multiple disciplinary perspectives and analytical approaches for breadth and depth of insight and contributions. Seventh, metrics and strategies are needed for assessing tradeoffs. As highlighted, AI for urban pandemic management is fraught with dilemmas, especially between public health and social justice objectives—e.g., disease containment versus privacy in AI surveillance. There also are other tradeoffs present with AI—e.g., accuracy versus fair and equal treatment, personalization versus solidarity and citizenship, quality and efficiency of services versus privacy and informational autonomy, and convenience versus self-actualization and dignity (Whittlestone et al., 2019). Efficiency versus resilience is yet another critical issue, particularly for urban disruptions such as pandemics (Schintler & McNeely, 2022a, 2022b). However, it is not only important to have ways to measure tradeoffs but also to resolve them given that stakeholders in a pandemic tend to have conflicting values, expectations, and priorities. Methods, such as multicriteria optimization or Delphi techniques (Dodgson et al., 2009; Linstone & Turoff, 1975), may be usefu. Eighth, metrics are needed to provide unbiased and reliable measures of the effects of different governance strategies and technical solutions on social justice in urban communities. They should be designed for use in policy analyses and impact studies, including artificial intelligence impact assessments (Calvo et al., 2020b). Even technical governance strategies, such as Explainable AI (XAI), which are designed to inform the public about the inner workings of AI algorithms for enhancing transparency and accountability, have their own ethical and social implications. Accordingly, metrics should be designed and used to evaluate such systems as well (Rosenfeld, 2021). 11

MAU and MTU generally refers to the propensity for different outcomes at varying levels of spatial and temporal aggregation, respectively.

Social Justice, Digitalization, and Health and Well-Being …

299

Ninth, metrics are needed that can measure trust and legitimacy since, at the end of the day, they are the name of the game when it comes to urban resilience relative to disruptions such as pandemics (Schintler & McNeely, 2022a, 2022b). Legitimacy is a “generalized perception or assumption that the actions of an entity are desirable, proper, or appropriate within some socially constructed system of norms, values, beliefs, and definitions” (Suchman, 1995a, 1995b). There are wide variations in value systems and cultures within and across cities (e.g., expectations of privacy), including in a pandemic (Mayer et al., 2020). In this regard then, approaches and tools are needed for understanding the values, preferences, and attitudes of residents of and visitors to a city—i.e., the “soul of the city” (Kourtit et al., 2021)—especially in terms of urban pandemic disruptions. Tenth, in all of this, and ultimately, metrics are needed that are sensitive to and account for contextual variations, as appropriate. While there are generalizable properties and processes that hold across cities and city systems, the metrics should reflect the nuances involved in specific urban contexts in relation to AI development in response to pandemics.

6 Conclusion Regarding policy development and culture of health aims, there can be a cautious optimism that progress can be made both in improving the health of the worst-off in society and narrowing health inequalities (Marmot, 2017). However, policy consequences of pandemic-related marginalization and effects beyond the health arena, especially when pandemics have been politicized to further particular interests and aims—e.g., immigration restrictions, workforce retrenchment and exclusion, attitudinal underpinning, etc. (Dionne & Turkmen, 2020). Pandemics may increase urban poverty (especially in the global south) and reinforce already pronounced economic and social inequalities and inequities, but one also might argue that they could provide opportunities for transformation and betterment of governance structures and practices, provision of public services, data use, and community engagement—i.e., “there will be losses, but there also are opportunities to rethink cities” (Martínez & Short, 2021). It is in this regard that governance and the institutional and political will to realize a culture of health and related practices in keeping with social justice principles might represent practicable goals and policy priorities in cities today. Frankly, to overcome the social crises and social justice challenges posed by pandemics—and with particular reference to COVID-19 today—requires confronting the underlying structures and mechanisms leading to inequitable outcomes in today’s cities, and re-thinking the social and justice benchmarks of current urban sustainability and resilience debates and strategies (Haase, 2020). “It is not only the case that the poorest and most excluded are hardest hit by the pandemic; in some respects, the response to the crisis also comes at the expense of those with the lowest incomes, the worst housing conditions, and the least access to high-quality open/green space, health insurance, healthy food, etc.” and “forecasts suggest we

300

L. A. Schintler and C. L. McNeely

can expect a further increase in disadvantages for those who are most vulnerable and hardest hit by the crisis in a social respect.” Again, “this polarization is concentrated in cities” (Haase, 2020). Looking to the future, there is a need to re-imagine and re-think the city in a world where globalization, along with other factors and challenges such as climate change, may foster even more pandemics (Martínez & Short, 2021). Emphasized in the discussion here has been the need for comprehensive and contextualized research on critical socio-cultural, organizational, institutional, and ethical constraints and catalysts, looking beyond typical foci on narrow technical and market concerns. Pandemics must be positioned relative to broad societal effects and health disparities, providing an approach for understanding the pressing and determinant social issues attending them, with an eye to questions of equity (Stabile, 2019, 2020). On the one hand, structural relations and social conditions and dynamics are instrumental in the increase of morbidity and mortality rates. On the other hand, infectious diseases themselves can contribute to and intensify such conditions, resulting in patterns of disadvantage and health disparities (McNeely et al., 2020; Schintler et al., 2020). This systemic and mutually reinforcing relationship marks critical terms for understanding pandemics as natural disasters and social phenomena relative to questions of public health and societal well-being in cities around the world.

References AAAS. (2021). Artificial intelligence and COVID-19: Applications and impact assessment. Report prepared by Ilana Harrus and Jessica Wyndham under the auspices of the AAAS Scientific Responsibility, Human Rights and Law Program. Acemoglu, D., & Restrepo, P. (2017). “Low-skill and high-skill automation.” National Bureau of Economic Research, No. w24119. Acosta, J. D., M.D. Whitley, May, L. W., Dubowitz, T., Williams, M. V., & Chandra, A. (2015). Stakeholder perspectives on a culture of health. Rand. Adler, P., Florida, R., & Hartt, M. (2020). Mega regions and pandemics. Tiidschrift voor economische en sociale geografie, 111(3), 465–481. Ahlborg, H., Ruiz-Mercado, I., Molander, S., & Masera, O. (2019). Bringing technology into socialecological systems research: Motivations for a socio-technical-ecological systems approach. Sustainability, 11(7), 2009. Akselrod, O. (2021, July 13). How artificial intelligence can deepen racial and economic inequities. ACLU. https://www.aclu.org/news/privacy-technology/how-artificial-intelligence-can-deepenracial-and-economicinequities. Aloisi, A., & De Stefano, V. (2021). Essential jobs, remote work and digital surveillance: addressing the COVID-19 pandemic panopticon. International Labour Review. Arafah, Y., & Winarso, H. (2017). Redefining smart city concept with resilience approach. IOP Conference Series: Earth and Environmental Science, 70, 012065. https://doi.org/10.1088/17551315/70/1/012065 Barr, C., Kommenda, N., McIntyre, N„ Ethnic Minorities Dying of COVID-19 at Higher Rate, Analysis Shows.” The Guardian, 22 April 2020. https://www.theguardian.com/world/2020/apr/ 22/racial-inequality-in-britain-found-a-risk-factor-for-covid-19 Bentotahewa, V., Hewage, C., & Williams, J. (2021). Solutions to Big Data privacy and security challenges associated with COVID-19 surveillance systems. Frontiers in big Data, 4.

Social Justice, Digitalization, and Health and Well-Being …

301

Berwick, D. M. (2020). The moral determinants of health. JAMA, 324(3), 225–226. Brzezinski, R. Y., Rabin, N., Lewis, N., Peled, R., Kerpel, A., Tsur, A. M., ... & Hoffer, O. (2021). Automated processing of thermal imaging to detect COVID-19. Scientific Reports, 11(1), 1–10. Büchi, M., Fosch-Villaronga, E., Lutz, C., Tamò-Larrieux, A., Velidi, S., & Viljoen, S. (2020). The chilling effects of algorithmic profiling: Mapping the issues. Computer Law & Security Review, 36, 105367. Calvo, R. A., Deterding, S., & Ryan, R. M. (2020a). Health surveillance during covid-19 pandemic. Bmj, 369. Calvo, R. A., Peters, D., & Cave, S. (2020b). Advancing impact assessment for intelligent systems. Nature Machine Intelligence, 2(2), 89–91. Chakravorti, B. (2022, March 17). Why AI failed to live up to its potential during the pandemic. Harvard Business Review. https://www.hbr.org/2022/03/why-ai-failed-to-live-up-to-its potential-during-the-pandemic. Chandra, A., Acosta, J. D., Carman, K. G, Dubowitz, T., Leviton, L., Martin, L. T., Miller, C., Nelson, C., Orleans, T., Tait, M., Trujillo, M. D., Towe, V. L., Yeung, D., Plough, A. L. (2016). Building a national culture of health: Background, action model, measures and next steps. Santa Monica, CA: Rand. Chen, B., Marvin, S., & While, A. (2020). Containing COVID-19 in China: AI and the robotic restructuring of future cities. Dialogues in Human Geography, 10(2), 238–241. Choi, Y.J., & McNeely, C.L. (2022). The dialectic of changing corporate masks: From profit maximizers to predators to socially responsible global leaders. In Research Handbook on the Sociology of Organizations (pp. 509–527). Edward Elgar Publishing. Coates, R., & Heitzeg, N. (2008). Micro-level social justice projects, pedagogy, and democratic movements: introduction. American Behavioral Scientist, 51(5), 595–598. Coates, R. M., & Williams, J. E. (2007). Introduction. American Behavioral Scientist, 51(4), 491– 494. Cugurullo, F. (2020). Urban artificial intelligence: From automation to autonomy in the smart city. Frontiers in Sustainable Cities, 2, 38. Culp, J. (2018). Social and global justice: Models of development and theories of justice. In Drydyk, J., Keleher, L. (Eds.), Handbook of development ethics (pp. 103–115). Routledge. Dankwa-Mullan, I., Scheufele, E. L., Matheny, M. E., Quintana, Y., Chapman, W. W., Jackson, G., & South, B. R. (2021). A proposed framework on integrating health equity and racial justice into the artificial intelligence development lifecycle. Journal of Health Care for the Poor and Underserved, 32(2), 300–317. Davis, R. A., Savannah, S., Harding, M., Macaysa, A., Parks, L. F., Aboelata, M.,... Haar, W. (2016). Countering the production of health inequities: A framework of emerging systems to achieve an equitable culture of health. Extended Summary. Prevention Institute. https://www.preventionin stitute.org/publications/countering-production-health-inequities-extended-summary Dionne, K., & Turkmen, F. (2020). The Politics of pandemic othering: putting COVID-19 in global and historical context. International Organization, 74(S1), E213–E230. do Carmo Barriga, A., Martins, A. F., Simões, M. J., & Faustino, D. (2020). The COVID19 pandemic: Yet another catalyst for governmental mass surveillance?. Social Sciences & Humanities Open, 2(1), 10009. Dodgson, J. S., Spackman, M., Pearman, A., & Phillips, L. D. (2009). Multi-criteria analysis: a manual. Erdiaw-Kwasie, M. O., & Alam, K. (2016). Towards understanding digital divide in rural partnerships and development: a framework and evidence from rural Australia. Journal of Rural Studies, 43, 214–224. Florida, R., & Mellander, C. (2016). The geography of inequality: Difference and determinants of wage and income inequality across US metros. Regional Studies, 50(1), 79–92. Gilmore, B., Ndejjo, R., Tchetchia, A., De Claro, V., Mago, E., Lopes, C., & Bhattacharyya, S. (2010). Community engagement for COVID-19 prevention and control: A rapid evidence synthesis. BMJ Global Health, 5(10), e003188. https://doi.org/10.1136/bmjgh-2020-003188

302

L. A. Schintler and C. L. McNeely

Glaeser, E. L. (2022). Urban resilience. Urban Studies, 59(1), 3–35. Glaeser, E., Kourtit, K., & Nijkamp, P. (Eds.). (2020). Urban empires: Cities as global rulers in the new urban world. Routledge. Haas, B. (2020). UCB_Playbook_R10_V2_spreads2.pdf (berkeley.edu). Haase, A. (2020). Covid-19 as a social crisis and justice challenge for cities. Frontiers in Sociology, 5, 583638. https://doi.org/10.3389/fsoc.2020.583638 Harari, Y.N. (2020). Surveillance is getting under our skin—and that should alarm us | Science and Technology. Jazeera. Hargittai, E., Jennrich, K. (2016). The online participation divide. In M. I. Lloyd & L. A. Friedland (Eds.), The communication crisis in America and how to fix it, (pp. 199–213). Palgrave Macmillan. Harvey, D. (2008). The right to the city. New Left Review, 53, 23–40. Heynen, N., Aiello, D., Keegan, C., & Luke, N. (2018). The enduring struggle for social justice and the city. Annals of the American Association of Geographers, 108(2), 301–316. Innes, M (2001). Innes: Control Creep (socresonline.org.uk). Institute of Medicine (IOM). (2002). The future of the public’s health in the 21st Century. National Academies Press. Institute of Medicine (IOM). (2006). Genes, behavior, and the environment: Moving beyond the nature/nurture debate. National Academies Press. Irani, M., & Rahnamayiezekavat, P. (2021). An overview of urban resilience: Dimensions, components, and approaches. Acta Sci. Pol. Administratio Locorum, 20(4), 305–322. Kent State University Public Administration (KSUPA). 2020 (30 July). The five principles of social justice. KSUPA Blog. https://onlinedegrees.kent.edu/political-science/master-of-public-admini stration/community/five-principles-of-social-justice Kitchin, R. (2016). The ethics of smart cities and urban science. Philosophical Transactions of the Royal Society a: Mathematical, Physical and Engineering Sciences, 374(2083), 20160115. Kitchin, R., Cardullo, P., & Di Feliciantonio, C. (2019). Citizenship, justice, and the right to the smart city. https://osf.io/preprints/socarxiv/b8aq5. Kourtit, K., Nijkamp, P., & Wahlström, M. H. (2021). How to make cities the home of people–a ‘soul and body’ analysis of urban attractiveness. Land Use Policy, 111, 104734. Lai, K. Y., Webster, C., Kumari, S., & Sarkar, C. (2020). The nature of cities and the COVID-19 pandemic. Current Opinion in Environmental Sustainability, 46, 27–31. Lancet, T. (2020). The plight of essential workers during the COVID-19 pandemic. Lancet (london, England), 395(10237), 1587. Lavizzo-Mourey, R. (2014). Building a culture of health: President’s message. Robert Wood Johnson Foundation. http://www.rwjf.org/en/about-rwjf/annual-reports/presidents-message-2014.html. Leslie, D. (2019). Understanding artificial intelligence ethics and safety. arXiv preprint arXiv:1906. 05684. Leslie, D. (2020). Tackling COVID-19 through responsible innovation: Five steps in the right direction. arXiv preprint arXiv:2008:06755. Leslie, D., Mazumder, A., Peppin, A., Wolters, M. K., & Hagerty, A. (2021). Does “AI” stand for augmenting inequality in the era of covid-19 healthcare?. bmj, 372. Li, X., Chen, W., & Straubhaar, J. D. (2018). Privacy at the margins| concerns, skills, and activities: multilayered privacy issues in disadvantaged urban communities. International Journal of Communication, 12, 22. Linstone, H. A., & Turoff, M. (Eds.). (1975). The delphi method (pp. 3–12). Addison-Wesley. Long, S. W., Olsen, R. J., & Christiansen, P. A. (2020). Molecular architecture of early dissemination and massive second wave of the SARS-Cov-2 virus in a major metropolitan Area. mBio. https:// doi.org/10.1101/2020.09.22.20199125v3. Malewar, A. (2018). The privacy risks of compiling mobility data—Tech Explorist. Mann, M., & Matzner, T. (2019). Challenging algorithmic profiling: The limits of data protection and anti-discrimination in responding to emergent discrimination. Big Data & Society, 6(2), 2053951719895805.

Social Justice, Digitalization, and Health and Well-Being …

303

Marmot, M. (2017). Social justice, epidemiology, and health inequalities. European Journal of Epidemiology, 32, 537–546. Marshall et al. (2021). Working from home during the pandemic. https://www.census.gov/library/ stories/2021/03/working-from-home-during-the-pandemic.html. Martínez, L., & Short, J. R. (2021). The pandemic city: Urban issues in the time of COVID-19. Sustainability, 13, 3295. https://doi.org/10.3390/su13063295 Mayer, J. D., Schintler, L. A., & Bledsoe, S. (2020). Culture, freedom, and the spread of Covid-19: Do some societies and political systems have national anti-bodies? World Medical & Health Policy, 12(4), 498–511. McCracken, H. (2020, April 3). How the telephone failed its big test during the 1918’s Spanish flu pandemic. https:/www.fastcompany.com/90484820/how=1918s-spanish-flu-outbreak-crushedthe-u-s-telephone-system. McNeely, C. L., & Schintler, L. A. (2020). The pandemic challenge: reflections on the social justice dynamic. World Medical and Health Policy, 12(4), 344–346. McNeely, C. L., Schintler, L. A., & Stabile, B. (2020). Social determinants and COVID-19 disparities: Differential pandemic effects and dynamics. World Medical and Health Policy, 12(3), 206–217. McNeill, W. H. (1998). Plagues and peoples. Anchor. McPhearson, T., Raymond, C. M., Gulsrud, N., Coles, A. N., Fagerholm, N., Nagatsu, M., Olafsson, A. S., Soininen, N., & Vierikko, K. (2021). Radical changes are needed for transformations to a good Anthropocene. npj Urban Sustainability, 1(5). https://doi.org/10.1038/s42949-021-000 17-x Mehrabi, N., Morstatter, F., Saxena, N., Lerman, K., & Galstyan, A. (2021). A survey on bias and fairness in machine learning. ACM Computing Surveys (CSUR), 54(6), 1–35. Mikheev, A., Serkina, Y., & Vasyaey, A. (2021). Current trends in the digital transformation of higher education institutions in Russia. Education and Information Technologies, 26(4), 4537–4551. Mooney, C., Achenbach, J., Fox, J. (2020). Massive genetic study shows coronavirus mutating and potentially evolving Amid Rapid U.S. spread. https://www.washingtonpost.com/health/2020/09/ 23/houstoncoronavirus-mutations/?arc404=true Moss, E., & Metcalf, J. (2020). High tech, high risk: Tech ethics lessons for the COVID-19 pandemic response. Patterns, 1(7), 100102. Musikanski, L., Rakova, B., Bradbury, J., Phillips, R., & Manson, M. (2020). Artificial intelligence and community well-being: A proposal for an emerging area of research. International Journal of Community Well-Being, 3(1), 39–55. National Academies of Sciences, Engineering, and Medicine (NASEM). (2017). Communities in action: Pathways to health equity. National Academies Press. Newlands, G., Lutz, C., Tamò-Larrieux, A., Villaronga, E. F., Harasgama, R., & Scheitlin, G. (2020). Innovation under pressure: Implications for data privacy during the Covid-19 pandemic. Big Data & Society, 7(2), 2053951720976680. Obermeyer, Z., Powers, B., Vogeli, C., & Mullainathan, S. (2019). Dissecting racial bias in an algorithm used to manage the health of populations. Science, 366(6464), 447–453. OECD (2020). Using artificial intelligence to help combat COVID-19 (oecd.org). Ormrod, S. (2003). Organizational culture in health service policy and research. Policy and Politics, 31(2), 227–237. Paremoer, L., Nandi, S., Serag, H., & Baum, F. (2021). Covid-19 pandemic and the social determinants of health. BMJ, 372, n129. https://doi.org/10.1136/bmj.n129 Pew Research Center. (2021). Experts say the ‘New Normal’ in 2025 will be far more tech-driven, presenting more big challenges. https://www.pewresearch.org/internet/2021/02/18/experts-saythe-new-normal-in-2025-will-be-far-more-tech-driven-presenting-more-big-challenges Pietras, S., Marks, A., & Cameron, G. (2021, June 14). Mobilizing smart city growth for a resilient future. Thematic Research Network on Data and Statistics. https://www.sdsntrends.org/blog/ 2021/mobilizing-smart-city-growth-for-a-resilient-future?

304

L. A. Schintler and C. L. McNeely

Piret, J., & Boivin, G. (2021). Pandemics throughout history. Frontiers in Microbiology, 11, Article 631736. https://doi.org/10.3389/fmicb.2020.631736 Reggiani, A., Schintler, L. A., Czamanski, D., & Patuelli, R. (2021). Reflections on entropy, complexity and spatial dynamics: The rebirth of theory? Edward Elgar Publishing. Renu, N. (2021). Technological advancement in the era of COVID-19. SAGE Open Medicine, 9, 20503121211000910. Rosenfeld, A. (2021, May). Better metrics for evaluating explainable artificial intelligence. In Proceedings of the 20th International Conference on Autonomous Agents and Multiagent Systems (pp. 45–50). Rybski, D., & González, M. C. (2022). Cities as complex systems—Collection overview. PLoS ONE, 17(2), e0262964. Saran, S. (2017). Technology and transformation. CyFy Journal, 4, 7–10. Schiff, D., Ayesh, A., Musikanski, L., & Havens, J. C. (2020, October). IEEE 7010: A new standard for assessing the well-being implications of artificial intelligence. In 2020 IEEE International Conference on Systems, Man, and Cybernetics (SMC) (pp. 2746–2753). IEEE. Schintler, L. A. (2021a). “COVID-19 pandemic” encyclopedia of big data. Springer. Schintler, L. A. (2021b). “Policy Analytics” encyclopedia of big data. Springer. Schintler, L. A., Kulkarni, R., McNeely, C. L., & Haynes, K. E. (2020a). Environmental and occupational exposure to toxic industrial chemicals and COVID-19: An exploratory analysis of U.S. counties. https://ssrn.com/abstract=3640797 Schintler, L. A., & Chen, Z. (Eds.). (2017). Big data for regional science. Taylor & Francis Limited. Schintler, L. A., & Fischer, M. M. (2018). Big data and regional science: Opportunities, challenges, and directions for future research. Schintler, L. A., & McNeely, C. L. (2022a). Artificial intelligence, institutions, and resilience: Prospects and provocations for cities. Schintler, L. A., & McNeely, C. L. (2019). Rural-urban digital asymmetries in the emerging sociotechnological landscape. In P. Nijkamp, K. Kourtit, & A. Kocornik-Mina (Eds.), Sustainable villages and green landscapes in the new urban world (pp. 167–197). Shaker Publishing. Schintler, L. A., & McNeely, C. L. (2020). Mobilizing a culture of health in the era of smart transportation and automation. World Medical and Health Policy, 12(2), 137–162. Schintler, L. A., & McNeely, C. L. (2022b). Artificial intelligence, institutions, and resilience: Prospects and provocations for cities. Journal of Urban Management. Schintler, L. A., & McNeely, C. L. (2021). Vulnerability, encyclopedia of big data. Schintler, L. A., Wu, J., & McNeely, C. L. (2020b). Regional health disparities, surge capacity, and impacts on COVID-19: Critical insights and lessons from China. World Medical & Health Policy, 12(4), 487–497. Schweitzer, J. (2020). Systemic-risks_COVID-FINAL.pdf (sra.org) Servick, K. (2020). Huge hole’ in COVID-19 testing data makes it harder to study racial disparities. Science Magazine. Seshaiyer, P., & McNeely, C. L. (2020). Challenges and opportunities from COVID-19 for global sustainable development. World Medical and Health Policy, 12(4), 443–453. Shabaz, A., & Funk, A. (2020). Freedom on the Net 2019: The crisis of social media. Freedom House. https://freedomhouse.org/sites/default/files/2019-11/11042019_Report_FH_ FOTN_2019_final_Public_Download.pdf. Sharifi, A., & Khavarian-Garmsir, A. R. (2020). The COVID-19 pandemic: Impacts on cities and major lessons for urban planning, design, and management. Science of the Total Environment, 749, 142391. Sharifi, A., Kawakubo, S., & Milovidova, A. (2020). Urban sustainability assessment tools: Toward integrating smart city indicators. In Urban systems design (pp. 345–372). Shiode, N., Shiode, S., Rod-Thatcher, E., Rana, S., & Vinten-Johansen, P. (2015). The mortality rates and the space-time patterns of John Snow’s cholera epidemic map. International Journal of Health Geographics, 14(1), 1–15. Article Number 21. https://doi.org/10.1186/s12942-0150011-y

Social Justice, Digitalization, and Health and Well-Being …

305

Smith, D. M. (1994). Geography and social justice. Blackwell. Soken-Huberty, E. (2021). Four principles of social justice. https://www.humanrightscareers.com/ issues/four-principles-of-social-justice Solar, O., & Irwin, A. (2010). A conceptual framework for action on the social determinants of health. WHO Document Production Services. Spaans, M., & Waterhout, B. (2017). Building up resilience in cities worldwide: Rotterdam as participant in the 100 resilient cities programme. Cities, 61, 109–116. Stabile, B. (2019). Continuing streams of policy consciousness to advance global health. World Medical and Health Policy, 11(2), 124–126. Stabile, B. (2020). The persisting importance of rhetoric and equity in health policy and outcomes. World Medical and Health Policy, 12(2), 86–89. Stiglitz, J. (2020). COVID-19 and global inequality—IMF F&D. Stray, J. (2020). Aligning AI optimization to community well-being. International Journal of Community Well-Being, 3(4), 443–463. Suchman, M. C. (1995a). Managing legitimacy: Strategic and institutional approaches. Academy of Management Review, 20(3), 571–610. Suchman, M. (1995b). Managing legitimacy: Strategic and institutional approaches. Academy of Management Review, 20(3), 57l–610. Tan, C. W., Chia, W. N., Zhu, F., Young, B. E., Chantasrisawad, N., Hwa, S. H., et al. (2022). SARS-CoV-2 Omicron variant emerged under immune selection. Nature Microbiology, 7(11), 1756–1761. https://doi.org/10.1038/s41564-022-01246-1 Thomas, L. (2022). The Intersectional Environmentalist: How to dismantle systems of oppression to protect people+planet. Voracious. Tomašev, N., Cornebise, J., Hutter, F., Mohamed, S., Picciariello, A., Connelly, B., ... & Clopath, C. (2020). AI for social good: Unlocking the opportunity for positive impact. Nature Communications, 11(1), 1–6. Vinuesa, R., Azizpour, H., Leite, I., Balaam, M., Dignum, V., Domisch, S., ... & Nerini, F. F. (2020). The role of artificial intelligence in achieving the sustainable development goals. Nature communications, 11(1), 1–10. Wade, L. (2020). An unequal blow. Wahba, S et al., (2020). Cities are on the front lines of COVID-19 (worldbank.org) Whittlestone, J., Nyrup, R., Alexandrova, A., Dihal, K., & Cave, S. (2019). Ethical and societal implications of algorithms, data, and artificial intelligence: A roadmap for research. Nuffield Foundation. Williams, D. R., Lawrence, J. A., & Davis, B. A. (2019). Racism and health: Evidence and needed research. Annual Review of Public Health, 40, 105–125. Winslow, C. E. A. (1920). The untilled fields of public health. Science, 51(1306), 23–33. Wong, D. W., & Li, Y. (2020). Spreading of COVID-19: Density matters. PLoS ONE, 15(12), e0242398. World Health Organization (WHO). (2008). Closing the gap in a generation: Health equity through action on the social determinants of health. WHO. Yaya, S., Yeboah, H., Charles, C. H., Otu, A., & Labonte, R. (2020). Ethnic and racial disparities in COVID-19-related deaths: counting the trees, hiding the forest. BMJ Global Health. https://doi. org/10.1136/bmjgh-2020-002913 Zwitter, A., & Gstrein, O. J. (2020). Big data, privacy and COVID-19–learning from humanitarian expertise in data protection. Journal of International Humanitarian Action, 5(1), 1–7.