The Philosophy and Methods of Political Science 9780333786949, 9780333945063, 9781403904461, 9781403904478

Citation preview

Series Editors: B. Guy Peters, Jon Pierre and Gerry Stoker Political science today is a dynamic discipline. Its substance, theory and methods have all changed radically in recent decades. It is much expanded in range and scope and in the variety of new perspectives – and new variants of old ones – that it encompasses. The sheer volume of work being published, and the increasing degree of its specialization, however, make it difficult for political scientists to maintain a clear grasp of the state of debate beyond their own particular subdisciplines. The Political Analysis series is intended to provide a channel for different parts of the discipline to talk to one another and to new generations of students. Our aim is to publish books that provide introductions to, and exemplars of, the best work in various areas of the discipline. Written in an accessible style, they provide a ‘launching-pad’ for students and others seeking a clear grasp of the key methodological, theoretical and empirical issues, and the main areas of debate, in the complex and fragmented world of political science. A particular priority is to facilitate intellectual exchange between academic communities in different parts of the world. Although frequently addressing the same intellectual issues, research agendas and literatures in North America, Europe and elsewhere have often tended to develop in relative isolation from one another. This series is designed to provide a framework for dialogue and debate which, rather than advocacy of one regional approach or another, is the key to progress. The series reflects our view that the core values of political science should be coherent and logically constructed theory, matched by carefully constructed and exhaustive empirical investigation. The key challenge is to ensure quality and integrity in what is produced rather than to constrain diversity in methods and approaches. The series is intended as a showcase for the best of political science in all its variety, and demonstrates how nurturing that variety can further improve the discipline.

Political Analysis Series Series Standing Order ISBN 978–0–333–78694–9 hardback Series Standing Order ISBN 978–0–333–94506–3 paperback (outside North America only) You can receive future titles in this series as they are published by placing a standing order. Please contact your bookseller or, in the case of difficulty, write to us at the address below with your name and address, the title of the series and one of the ISBNs quoted above. Customer Services Department, Macmillan Distribution Ltd, Houndmills, Basingstoke, Hampshire, RG21 6XS, UK

Series Editors: B. Guy Peters, Jon Pierre and Gerry Stoker Editorial Advisory Group: Frank R. Baumgartner, Donatella Della Porta, Scott Fritzen, Robert E. Goodin, Colin Hay, Alan M. Jacobs, Eliza W.Y. Lee, Jonathon W. Moses, Craig Parsons, Mitchell A. Seligson and Margit Tavits. Published David Beetham The Legitimation of Power (2nd edition) Peter Burnham, Karin Gilland Lutz, Wyn Grant and Zig Layton-Henry Research Methods in Politics (2nd edition) Lina Eriksson Rational Choice Theory: Potential and Limits Jean Grugel and Matthew Louis Bishop Democratization: A Critical Introduction (2nd edition) Colin Hay Political Analysis Colin Hay, Michael Lister and David Marsh (eds) The State:Theories and Issues Andrew Hindmoor and Brad Taylor Rational Choice (2nd edition) Vivien Lowndes and Mark Roberts Why Institutions Matter David Marsh and Gerry Stoker (eds) Theory and Methods in Political Science (3rd edition) Ioannis Papadopoulos Democracy in Crisis? Politics, Governance and Policy B. Guy Peters Strategies for Comparative Research in Political Science Jon Pierre and B. Guy Peters Governance, Politics and the State Heather Savigny and Lee Marsden Doing Political Science and International Relations

Rudra Sil and Peter J. Katzenstein Beyond Paradigms: Analytic Eclecticism in the Study of World Politics Martin J. Smith Power and the State Gerry Stoker, B. Guy Peters and Jon Pierre (eds) The Relevance of Political Science Cees van der Eijk and Mark Franklin Elections and Voters Keith Dowding The Philosophy and Methods of Political Science

Forthcoming Alan Finlayson and James Martin Interpretive Political Analysis: A Critical Introduction Colin Hay Globalization and the State Johanna Kantola and Emanuela Lombardo Gender and Political Analysis William Maloney and Jan van Deth Political Participation and Democratic Politics David Marsh Political Behaviour Karen Mossberger and Mark Cassell The Policy Process: Ideas, Interests and Institutions Dimiter Toshkov Research Design in Political Science

The Philosophy and Methods of Political Science Keith Dowding

This page intentionally left blank

The Philosophy and Methods of Political Science Keith Dowding

© Keith Dowding 2016 All rights reserved. No reproduction, copy or transmission of this publication may be made without written permission. No portion of this publication may be reproduced, copied or transmitted save with written permission or in accordance with the provisions of the Copyright, Designs and Patents Act 1988, or under the terms of any licence permitting limited copying issued by the Copyright Licensing Agency, Saffron House, 6–10 Kirby Street, London EC1N 8TS. Any person who does any unauthorized act in relation to this publication may be liable to criminal prosecution and civil claims for damages. The author has asserted his right to be identified as the author of this work in accordance with the Copyright, Designs and Patents Act 1988. First published 2016 by PALGRAVE Palgrave in the UK is an imprint of Macmillan Publishers Limited, registered in England, company number 785998, of 4 Crinan Street, London, N1 9XW. Palgrave Macmillan in the US is a division of St Martin’s Press LLC, 175 Fifth Avenue, New York, NY 10010. Palgrave is a global imprint of the above companies and is represented throughout the world. Palgrave® and Macmillan® are registered trademarks in the United States, the United Kingdom, Europe and other countries. ISBN 978–1–4039–0446–1 hardback ISBN 978–1–4039–0447–8 paperback This book is printed on paper suitable for recycling and made from fully managed and sustained forest sources. Logging, pulping and manufacturing processes are expected to conform to the environmental regulations of the country of origin. A catalogue record for this book is available from the British Library. A catalog record for this book is available from the Library of Congress.

For Steven Kennedy Conferences will not be the same without you

This page intentionally left blank

Contents

List of Illustrative Materials

xii

Preface and Acknowledgements

xiv

1

Introduction

1

2

Isms

9

2.1 2.2 2.3 2.4 2.5 2.6 2.7 2.8 2.9 2.10 2.11 2.12 2.13 2.14 3

4

Introduction Realism and anti-realism Nominalism, idealism Empiricism and positivism Instrumentalism, conventionalism, pragmatism Essentialism, objectivism Critical theory Postmodernism Interpretivism Critical realism Naturalism Constructivism Realism and relativism in ethics A few isms not in Figure 2.1

9 10 14 15 18 19 21 22 23 24 26 27 30 32

What Is an Explanation?

36

3.1 3.2 3.3 3.4 3.5 3.6 3.7

36 37 45 50 55 60 67

Introduction Language and the world The attempt to produce a model of explanation Proximate and ultimate; type and token Description, causation and understanding Generalizations, laws and mechanisms Conclusion

What Is a Theory?

68

4.1 4.2 4.3 4.4

68 72 79 88

Introduction Organizing perspectives Explanatory theories or models Explanatory theory: non-formal models

ix

x

Contents 4.5 4.6 4.7

5

6

7

8

9

Mechanisms and constraints Cumulative and non-cumulative research Conclusion

94 98 100

Hypotheses and Theory Testing

102

5.1 5.2 5.3 5.4 5.5 5.6

102 106 107 116 128 130

Introduction How evidence bears on theories: preliminaries Confirmation, induction and theory Hempel’s paradox and Popper’s falsifiability Concepts are theory-laden Conclusion

Narratives, Mechanisms and Causation

133

6.1 6.2 6.3 6.4 6.5

133 134 138 151 157

Introduction Causation as narrative Dichotomies in causal accounts Case studies and causation: process tracing Conclusion

Methods and Methodologies

160

7.1 7.2 7.3 7.4 7.5 7.6 7.7 7.8

160 163 166 169 170 174 181 188

Introduction Qualitative and quantitative research Data access and research transparency Policy-oriented research Different methods: institutional-structural Different methods: behavioural Different methods: interpretative Conclusion

Concepts and Conceptual Analysis

189

8.1 8.2 8.3 8.4 8.5 8.6 8.7

189 191 201 203 206 208 212

Introduction Principles of conceptual analysis The existence of concepts Natural kinds Measurement Some principles of classification Contests and contestability of concepts

Analytic Political Philosophy

213

9.1 9.2 9.3

213 216 223

Introduction Interpreting the work of dead people Moral problems and grand theory

Contents 9.4 9.5

Conceptual analysis, thought experiments and intuition pumps Conclusion

10 Political Science as a Vocation 10.1 10.2 10.3 10.4

Political science Belief and opinion Surface and structure The profession

xi

224 240 243 243 244 245 250

References

253

Index

275

List of Illustrative Materials

Boxes 2.1 2.2 2.3 3.1 3.2 3.3 3.4 3.5 3.6 3.7 3.8 3.9 3.10 4.1 4.2 4.3 4.4 4.5 4.6 4.7 4.8 5.1 5.2 5.3 5.4 5.5 6.1 6.2 6.3 6.4 7.1 8.1

Discussion of Figure 2.1 How to handle isms Two laws Metaphysical and epistemic relationships What is prediction? Explanation and prediction DN models of explanation Failures of DN explanation DN and reasons for action Inductive-Statistical explanation Type and token Proximate and ultimate explanation Mechanisms and generalizations What are theories? Types of theories (applied to politics) Kuhn on paradigms What is a model? False assumptions Empirical content The inversion strategy The specification problem Theory not induction: spotted bodies and striped tails Affirming the consequent Eddington’s test of general relativity Popper and verisimilitude Hypotheses Causes of effects and effects of causes Binary causal oppositions Uses of process tracing Experimental methods and process tracing Objective and subjective Concepts should be as primitive as possible

xii

11 12 13 40 44 44 46 47 47 49 52 54 62 70 72 74 80 82 86 93 96 109 112 114 124 127 137 138 153 154 164 192

List of Illustrative Materials 9.1 9.2 9.3 9.4 9.5 9.6 9.7

Rawls and Northern Ireland Interpreting in the sea of verbiage Joining the objects Train Two toy game thought experiments Surgeons and shootings Trolleys and bridges

xiii 214 222 229 230 231 232 234

Figures 2.1 9.1 9.2

Illustrative map of isms Stylized relationship of political philosophy to philosophy and political science ‘Outliers’

10 214 237

Table 4.1

Non-formal models of urban politics

91

Preface and Acknowledgements

This book has been a long time coming. Steven Kennedy first approached me to write a book following an article of mine almost 15 years ago. He asked for a punchy book on how to go about political science research that explained my underlying philosophical views. I signed up with some reluctance as I felt I was engaged in too many other projects. But Steven could be very persuasive and somehow got me to agree. I did not start the book for many years. At first some of the ideas appeared in methodology courses I gave for MSc and PhD students at the London School of Economics. When I came to the Research School of Social Sciences at the Australian National University, there was no classroom tuition for PhD students in politics and IR at all. John Dryzek and I (later with John Ravenhill) ran an intensive course. This has expanded over the years to a suite of five courses with many teachers, and I have continued to expand and develop an account of some philosophy to underpin the more practical advice on research design and professional development that we also offer. The book is not designed to give a comprehensive view of all the differing approaches to the philosophical underpinnings of different methods. It presents, as Steven Kennedy first requested, my views. Perhaps they are rather heretical in places, but my students find the ideas here useful. Over time the lectures developed into proto-chapters, and eventually, after more prompting from Steven, I started to write the book. It was written in three relatively short bursts over several years, undergoing substantial revisions in response to Steven Kennedy and other readers. My views have changed over the long gestation and writing, and continue to do so. Whilst I might not make different claims if I started another draft, I would certainly make some points rather differently, omit some material and write more about others. But one has to stop some time. The book took so long from contract to final version that my editor, Steven Kennedy, retired from Palgrave. It is partly for that reason, and in recognition of his persistence in extracting the book from me, that I dedicate it to him. Political science conferences will no longer be the same without Steven’s presence in the publishers’ exhibition. However, his replacement, Stephen Wenham, in his quieter way, is just as persistent, and very efficiently helped me to complete the task. I also thank Cathy Tingle for some careful copyediting.

xiv

Preface and Acknowledgements

xv

An early draft of the book was discussed at the Australian Political Studies Association Quantitative Methods Research Group’s inaugural meeting in Melbourne in December 2012, where what I say about causation was picked over and not much liked. I have included some new material as a result but have not changed my position much. I would like to thank the contributors to that discussion and all those others who have discussed aspects of the book, commented on chapters or, in some cases, on one or other of the complete drafts of the book. They include William Bosworth, Brett Calcock, Richard Carney, Mike Dalvean, Ben Goldsmith, Al Hajek, Andrew Hindmoor, Yusaku Horiuchi, Peter John, Aaron Martin, Mike Miller, Anne Phillips and Brad Taylor. Anne Gelling copyedited each chapter as I went through several drafts. Her comments often go far beyond simple copyediting queries to substantial discussion of issues and I thank her once again not only for that contribution, but for everything she does that enables me to spend longer on my academic life than other material needs would else allow.

This page intentionally left blank

Chapter 1

Introduction

This is a book on the philosophy of political science. It could have been a book on the philosophy of social science, but to cover the broader remit of this discipline in the way I have tried to do for political science would have involved my learning a lot more about what goes on in other social sciences, such as economics, sociology, history (yes, history is a social science) and particularly psychology and social psychology, of which I know less. Nevertheless, I hope it might be of some interest to social scientists in disciplines beyond political science. It is not a methods book as such. It does not try to teach any political science methods, although I discuss various methods and what one might achieve with them, and make some recommendations as to how to go about studying some issues. The principal aim of the book is to examine some philosophical issues through the lens of political science methodology. In order to do that, I need to cover a little ground on relevant methods and discuss their justification; for a full grounding in methods, the reader will need to consult the texts referred to in this book. There are many textbooks on the philosophy of social science and many more books on social scientific methods – both general ones and those dealing with specific methods. Why, then, another one? Well, Steven Kennedy of Palgrave Macmillan pressed me to write this book with the enthusiasm for which he is famous, for one reason; for another, I think there is a disjunction between texts on the philosophy of social science and politics methods texts. The problem with some books on the philosophy of social science is that their authors are philosophers who do not engage in empirical social science. These books tend to be rather general, and whilst they cover important issues in philosophy they do not make it clear why these issues should matter to working social scientists. Does it make any difference to a legislative studies scholar whether scientific realism can be justified? If not, why should that issue be covered in a book on the philosophy of political science? On the other hand, when political scientists tackle philosophical questions they do not always realize the technical issues involved, and make what philosophers consider to be rather naive inferences and blunders. Furthermore, what political scientists say about philosophy tends often to be a generation or two out of date. While I am by no means a philosopher, I have some training in the discipline, and have tried to ensure the claims

1

2 The Philosophy and Methods of Political Science made here have some philosophical pedigree. I am experienced in using many different empirical methods, both qualitative and quantitative. I try to expose the reader to philosophical arguments that are relevant to highlevel political science research – while attempting to avoid the dual pitfalls of philosophical naivety and ignorance of empirical practicalities. Doubtless some people will think I have failed in both regards. This book is an exploration of the philosophy of social science and a defence of mainstream empirical analysis. I make no bones about the fact that I think that some anti-empirical tracts are based on a general ignorance of both empirical techniques and philosophy. Several recent books and articles suggest that political scientists need to be aware of their ontological and epistemological commitments prior to embarking upon research. The authors of these works discuss ‘ontological and epistemological commitments’ as though they are preferences that one simply picks or chooses from the menu (that they thoughtfully set out for the reader). They suggest that these commitments cannot be empirically examined, but then seem to infer from that claim that they cannot be examined at all. That is not true: one can examine the logic, the coherence and the honesty of ‘ontological commitments’. In other words, do these commitments make sense? Are the writers consistent in their claims? And, just as importantly, do they live their lives as though they believe the ontological and epistemological claims they profess in their academic writings, or are their academic publications simply ‘cheap talk’ – a tool to further their careers, but not something they live by? In short, are they academic hypocrites? Chapter 10 of this book explains why I think academic honesty matters. Another aspect of this book that contrasts with many others is the fact that I use a lot of examples drawn from many parts of political science and public administration. Some merely illustrate a point I am trying to make. Others are discussed in some depth because I think that the philosophical issues I examine require them to be explained in some detail. I also draw on a lot of examples from outside of political science; some readers will think too many. However, they are used to show complexities that exist even in simple cases before we turn to examples from political science that are usually more complex still. Philosophers like to use the simplest possible examples to illustrate just how problematic they can be. I draw extensively upon some of my own empirical research in this book. I do so because I am intimately aware of what my colleagues and I were trying to achieve in that research. I know why we designed it as we did, what mistakes we made, how we could have done better. Textbooks often set out the grounds of good research design, but the reality is always messier than the ideal. One sometimes realizes as the research is being conducted that some of the design needs to be altered, and indeed – being brutally honest – how some of the cracks need to be plastered over in order for it to be published.

Introduction

3

This book is directed at undergraduate and graduate students (though I hope academic colleagues can get something from it too), and both need to be aware of the strength of the evidence in the work they read. Doctoral students also have to get into the publishing business and should think about compromises they might need to make whilst keeping their work honest. This book also differs from many others on the philosophy or theory of social sciences in that they are often structured around ‘isms’ – positivism, realism, constructivism, and so on. The common book plan is to look at each ism, define it, critique it, and then move on to the next one; until finally we get to the favoured ism, which is defined and defended rather than criticized. I believe this standard textbook technique (not only in this field, but in others) is pernicious, and responsible for the poverty of much social science. Dealing with issues in this manner is kind of ad hominem: the ism, rather than the arguments around the issues being discussed, becomes the target. I have (at least) four problems with this way of presenting material: • Often those labelled as, say, ‘positivists’ or ‘relativists’ have greater disagreement with others within the same ism than with those labelled with a different ism. Rudolf Carnap, Carl Hempel, Ernest Nagel and Hans Reichenbach can all be justifiably termed positivists, but they had some very different ideas. Karl Popper, laughably, is labelled a positivist by some academics, although he – in his usual modest manner – saw himself as positivism’s greatest critic. (In fact, everyone can be described, with some justification, as a positivist; but the label is now usually applied to identify people with logical positivism, currently rather out of fashion.) Likewise, Paul Feyerabend and Michel Foucault can both be regarded as ‘relativists’, but their views on the conduct of research are far removed from each other. • Partly for the above reason, I simply do not always recognize the labels as they are defined – what someone defines as positivism, or realism, or critical theory, say, just isn’t described in the manner I would describe it. My point here is not that the author has got the definition and account of the ism wrong, whereas I (of course!) have got it right. Sometimes I simply do not see how a particular writer can be allocated that particular label (given how the ism was defined); or how anyone in that ism can be thought of as having those beliefs. But my point is rather that the author has chosen to focus upon certain aspects of the thoughts of writers within the ism rather than others. They have chosen to chop up the world differently from the way I would slice it. • I find it offensive when lesser writers (and great writers rarely write textbooks or books like this one, getting on, instead, with their more important work) dismiss some of the world’s greatest thinkers with rather puny critiques of the particular ism into which the lesser writer pigeonholes them. I sometimes read these critiques and ask myself, ‘Does this writer really

4 The Philosophy and Methods of Political Science think that Karl Popper or Willard van Orman Quine or … was so stupid that they would be floored by that criticism?’ • Finally, given that isms are not ‘natural kinds’ or even coherent theories, the whole intellectual enterprise seems on shaky foundations from the start. One creates a way of cataloguing the world into isms, for each of which no person, living or dead, would assent to every aspect as defined, then proceeds to knock them down by demonstrating their logical incoherence or contradictions (in the best texts) or dismiss them with poorly thought-out critiques (in the worst). In this sense the entire enterprise seems to me to be ad hominem, a form of critique that is informally fallacious. You dismiss a writer or a claim by associating them with the ism you have criticized, without having to do the hard work of proper analysis and contextualization. This bizarre way of behaving seems, boa-like, to have constricted the entire discipline of international relations (IR). A few years ago I sat on an interview panel for a post in an IR department. Each candidate at their presentation, and then again at the interview, labelled themselves in terms of an ism: ‘I am a neo-realist’; ‘I am a constructivist with some realist leanings’; ‘My work is informed by post-constructivism’, and so on. It was noticeable that each wanted to hedge their ismistic bets a little, lest they commit some faux pas and upset a member of the panel. IR seemed to be engaged in some battle where self-identifying colours must be nailed to the mast prior to any discussion of topic or research. In fact, a strong swing away from this form of thinking is now under way in IR (Sil and Katzenstein 2010). What is doubly confusing for a person like me, who reads a bit of philosophy, is that in IR what is meant by ‘realism’ and ‘constructivism’ (the two main contenders, it seems) has, at best, a glancing connection with what those terms mean in the philosophy of (social) science. Indeed, in philosophy, as I explain in Chapter 2, in a strong sense many modern realists are constructivists. (Constructivism in analytic political philosophy, by the way, has only a tangential relationship to constructivism in science or in IR.) So this book is not designed in that way. I am afraid, though, that I am unable to avoid mentioning isms, in part because I need to engage with the literature out there. Indeed, in Chapter 2 I discuss isms in much the way I have criticized others for, largely due to pressure from my students, who have suffered the rambling lectures on which this book is based and who have demanded I provide a roadmap to help them understand the course reading. But I do so uneasily. I try to map out the grounds of the isms as I see them. Whilst I demarcate specific views that I attach to those isms, I also point out how one can coherently hold various of these positions simultaneously (or at least aspects of these various positions simultaneously).

Introduction

5

I am indeed an ismistic pluralist. Rather than dismissing any ism (or, to be honest, virtually any ism), I embrace them all, choosing those bits I like and discarding the rest. Some of my colleagues insist I am a positivist, though from that label what they then infer I believe is often mystifying to me; but for myself I am happy to be a positivist. I self-identify as a realist, a constructivist, a naturalist, an interpretivist and an objectivist; I have been greatly influenced by Karl Popper (though I am not a Popperian, finding too many problems with his method when it comes to ‘theory’ or what I call ‘model’ testing: see Chapter 5); I am weakly relativist in certain contexts; reductionist in explanation; foundationalist with regard to theoretical models, but holistic with regard to putting them together; an instrumentalist and pragmatist. All of these positions are underwritten for me by Darwinian evolution, which must underpin the natural world and the way in which we perceive it. And, of course, if it underpins both the natural world and the way we see that world, it must underpin the social world, our science and our philosophy of science. So, along with strong or radical relativism, one of the few isms that I am not prepared to encompass within my set of beliefs is creationism  – intelligent design (ID), as it has now been rebranded – and its ilk (and creationism is the only other game in town when it comes to our planet’s biological heritage). Any epistemology or ontology we adopt needs to take account of Darwinian evolution and its effects on our social world and the ways in which we form our beliefs. I will explain in Chapter 2 in what sense I can identify with all these isms. I will also explain in other places why I am a methodological pluralist, believing in the worth of quantitative and qualitative analysis; survey- and interview-based evidence; interpretive shadowing; discourse analysis; good old-fashioned archival work; formal modelling and inductive data dredging; large-n analysis and individual case studies. I also believe in the value of straightforward description (Gerring 2012a). What I do not believe, however, is that each method is equally good at answering each and every question. Some questions can only be answered by certain methods, whilst other methods are more efficient at answering some questions. Perhaps the most important lesson of all to learn in social scientific research design is the constraints upon what each method can explain. Indeed it might be that the best way of thinking about any particular method of study is less what questions it can be used to address than what questions it cannot be used to answer. In other words, what assumptions do we need to hold in order to reach the inferences we have made? What logical constraints are there upon those inferences and what are the problems with our assumptions? Many techniques in the social sciences have been developed because people have interrogated extant techniques specifically with regard to their constraints and failings, and through that process developed new techniques. (A word here to the eager student about to embark on a PhD. If you insist on

6 The Philosophy and Methods of Political Science writing about the difficult and important questions you would like to answer, then you need to learn the techniques that can enable you to address those questions. If you insist on using only certain techniques, then be sure to ask questions that those methods can answer.) Mixed-method research is now fashionable, but when engaging upon it one must be sure what each method is adding to the analysis. That is, whilst mixed methods might enable one to get a handle on different aspects of the research problem, they will not necessarily all support each other with regard to the aspects each method addresses. One of the most controversial aspects of my account is how social science should handle causation. Some scholars (Quine 1960, for example) hold that all explanation is reduction to causes. A great deal of modern empirical science is about how we capture causal processes – various statistical techniques such as structural equations; the experimental turn in political science; process tracing; the unification of the logic of political science explanation in King et al. (1994). Indeed King et al., despite writing sensibly about descriptive inference, seem to assume that all explanation is causal. In that regard I am completely heretical and think far too much is made of trying to demonstrate causation. It is not that I think causation is unimportant, but I do not believe that demonstrating causation is the only useful thing to do, nor always the most important aspect of any science. The notion of ‘cause’ has almost completely disappeared from theoretical physics; has no role in the building-blocks of chemistry; it is not at all clear that we can represent ‘natural selection’ in evolutionary biology in any ordinary sense of ‘cause’ (Matthen and Ariew 2009); and an awful lot of good social science simply has not pinned down causation. I believe that noting structural conditions and identity relationships is still important. Furthermore, philosophers have struggled with the whole notion of causation for hundreds of years. Indeed, people’s views of it vary according to their education. Lawyers (and most philosophers) tend to look at causation in terms of ‘but for’ conditions (see Amsel et al. 1991), whereas scientists look at causation probabilistically (see Chapter 6). This is not a minor issue. It has big implications for our understanding of the world, for explanation, for research design, and indeed for the conduct of law in the courtroom (where often what scientists would consider the best evidence available is deemed inadmissible). One of the big divides in political science, between quantitative and qualitative researchers, mirrors the divide over the nature of causation, with quantitative scholars utilizing a probabilistic account of causation, and qualitative ones a ‘but for’ account (see Goertz and Mahoney 2012). I discuss this issue in Chapter 6, making no claim to analyse causation in any original manner. Physicists are not so interested in causation, engineers are; biologists not so much, medical practitioners and environmentalists are; chemists not,

Introduction

7

pharmacists are. We become interested in causation when we want to intervene in the world. One way of demarcating the divide between political science and policy studies/public administration is by how far we are interested in causation in our analysis. For example, as political scientists we should be content with noting, say, the functional relationships between characteristics of polities, their environment and historical situation and, say, the rise and success of insurgency. Only if we want to intervene and help or hinder insurgency (in other words, become policy analysts), should we care about the precise causal effect of any aspect of those relationships and what will happen if we intervene to alter some of them. As far as I am concerned, narrative history is data (and a good yarn) and I do not think historians should worry too much about whether they correctly identified the ‘causes’ of any particular historical event, such as the First World War, the rise of Hitler or the weakness of contemporary US presidents. The fact that political scientists and historians do worry (too much) about these matters is a fault of methods courses that concentrate excessively on pinning down causation. If physicists, chemists and biologists do not agonize, why should we? Abandoning our fixation on causation might lead to greater precision and point in our models, as I discuss in Chapter 6. I am not saying these causal questions should not be asked, but they should be framed in a more general sense. We cannot prevent the rise of Hitler, but we might be able to intervene to stop the rise of murderous dictators. As policy analysts, we should not be concerned, as such, with the cause of the rise of Hitler, but rather the cause of the rise of dictators of which Hitler is one data point. It is the general question that holds our interest because we are interested in the broader policy issue. What could have prevented Hitler’s rise might be a fun question, but I am not sure it is good history. In this example, good history is narrating Hitler’s rise. Indeed, even if we can plausibly argue that there is a unique event without which Hitler would not have risen (a ‘but for’ condition), this is not much help to the broader policy issue. The broader policy issue concerns the conditions that facilitate the rise of dictators, not the chance (‘but for’) events that enable given ones to arise in specific cases. In that sense we are interested in the structural features around causation just as much as, if not more than, specific causes of unique events. This statement already assumes some aspects of the analysis of causation with which some might want to take issue. But it also reveals why I am not so concerned with pinning down causation as some of my colleagues. My sense from discussion with historians is that they are uncomfortable with my suggestion that they should be content with the narrative rather than pinning down causes. They think that it somehow downgrades their subject and makes it non-explanatory. But they can only think like this if they feel that all explanation is reduction to causes, and they are required to

8 The Philosophy and Methods of Political Science give causal explanations; getting out of that discursive trap would be a real freedom, I think. However, my view of the role of causation in social science is subversive and whilst it will appear several times later in this book, I will endeavour to ensure that it does not overly colour my characterization of good political science. This book is an introductory text designed for high-level undergraduates, who are increasingly subject to research methods courses, and for higherdegree research students. I have taught parts of it at the Australian National University and some time ago at the London School of Economics. Some of it is quite difficult to understand and appreciate. I have tried to make everything as simple and readable as possible, but I myself find some topics difficult to get on top of; and that is, I think, because they are inherently difficult to grapple with. Any student (indeed any person) who does not find the issues difficult to comprehend has probably missed the point. I tend to think there are quite a few social theorists around who do miss the point on a regular and almost systematic basis. One of the things I sometimes say to students (at both graduate and undergraduate level) who complain about the difficulty of the material on my various courses is, ‘Of course it is: this is a university.’

Chapter 2

Isms

2.1 Introduction I do not like isms; I would much prefer not to have this chapter in the book at all. However, my students have insisted that it is necessary, since they feel they need some guidance in their reading of what others write. And in truth I often mention isms myself in this book. Isms can be useful shorthand; and even with the best of intentions it is hard to break out of the discourse of our community. One reason I am hesitant to discuss isms is that I have no confidence that philosophers or social scientists will agree with the way in which I demarcate them in this chapter – if only because I do not fully agree with the way anyone I have read has defined them. And I do not want, nor do I want readers, to get involved in a debate over the ‘true nature’ of the isms and what they ‘really’ entail. Just keep in mind that isms exist only in the sense in which we construct them and so, given that different people construct them very differently, we have a discursive problem. Isms are doctrines, theories or practices that are thought to have a distinctive character, but the suffix tends to be used disparagingly. The problem with using isms when doing academic work, as opposed to propagandizing, is that most isms mean rather different things to different people. Each and every ism is a summary statement of a set of inconsistent claims. The claims are inconsistent since writers who are pigeonholed under any given ism have different views about key aspects of the subject. (Of course no author is completely consistent either, especially over a whole career, despite our best effort.) But within any ism there will be controversy and disagreement. Since the claims that come under any ism are so inconsistent, there cannot be any ‘true nature’ of any ism. ‘Truth’ is relative to each person who defines the isms – where inconsistency rules, so does some form of relativism in meaning, which entails there can be no resolution. This chapter is designed to explain how I think of these things and help you understand some of my comments in the more important parts of the book about studying political science. I try to put each ism in the best light, though I am also critical of some of the inferences drawn by those using one ism or another. I map the relationship of the isms as I see them in Figure 2.1; though, because of the wide variety of writers involved, the figure is rather

9

10 The Philosophy and Methods of Political Science Figure 2.1

Illustrative map of isms

REALISM Strong

ANTI-REALISM Weak

Weak

Strong

Non-(Weak) Relativism

Relativism

Nominalism Idealism Empiricism Positivism (Comte)

(Logical Positivism) Instrumentalism Pragmatism Conventionalism

Essentialism Objectivism

Critical Realism

Structuralism/ Post-structuralism Critical Theory Interpretivism

(Radical Scepticism) Postmodernism

Naturalism Constructivism [Realism (science)

Relativism (morals)]

complex and I am not sure how useful it really is. Nevertheless, I will refer to it several times in the chapter. Box 2.1 discusses Figure 2.1 in terms of its general layout – though even this is somewhat arbitrary. Box 2.2 offers some advice about how to handle isms.

2.2 Realism and anti-realism Scientific realism is the thesis that there is a world that exists independently of us. That belief gives rise to the claim that any proposition about the world has a truth-value that is dependent (at least to a large degree) upon the way the world is. The strong version says that every proposition about the world (including the social and political worlds) is either true or false. Propositions might come in degrees of truth and falsity. A complex proposition might include some true elements and some false ones. Or a vague proposition might be vaguely or partially or ‘sort of’ true. But the idea is that any proposition can be judged in terms of its truth-value and we can interrogate complex or vague propositions in order to approach more precise statements with clearer truth-values. We can think of this relationship between propositions and the world as a constraint. The way the world is constrains the set of propositions that

Isms

Box 2.1

11

Discussion of Figure 2.1

Here I give an overview of Figure 2.1, but the full explanation of how I see all these isms fitting together is contained in Chapter 2 itself. The two general categories are realism and anti-realism. First, note that here ‘realism’ refers to ‘scientific realism’ and not ‘realism’ as used in international relations. Scientific realism can be opposed to nominalism, idealism or phenomenalism, or to relativism; I use a blanket term ‘anti-realism’ which just means opposed to realism. Scientific realism is a term I use to cover the idea that there is a world that is beyond our perception of it. I suggest that scientific realism might be held strongly or weakly. I take weak realists to believe that there are things outside of our perception of them, but many of our explanatory concepts, theories and so on are ideas we impose on that reality. For example, if you think the idea of causation is something imposed on the universe outside of the object that we see subject to cause, then you would be weakly realist. Strong realists are prepared to see everything as part of the real world – critical realists, for example, see themselves opposed to the ‘positivism’ of finding statistical correlations (imposed on the world) by saying there are real causal mechanisms and we need to discover that underlying reality. Nevertheless, scientific realists are constructivists in the sense that at the base level what is in the real world is what we can measure with our instruments. Anti-realism is simply anything that is opposed to realism. To some extent, in Figure 2.1 I equate that with relativism – weak relativism or strong relativism, though I hedge my bets by calling weak anti-realism ‘Non-(Weak) Relativism’ since some anti-realists would deny being relativists. Postmodernism seems to deny all reality and suggest that everything we see we construct. Idealism is the thought that we cannot get a hold on any world external to our perception of it (even if there is one). I put most categories in ‘weak relativism’ since many writers within those traditions do not deny a world outside of our perception as such, but argue that everything we perceive is our version of that reality. Another way of thinking about realism and anti-realism that is defended in the chapter is that realism holds that all propositions have a truth-value – they are either true or false. And it is something in the world that makes them true or false. Anti-realists can be thought of as those who think that at least some propositions are neither true nor false – weak relativists hold there is some third value, ‘neither true nor false’, and strong relativists hold that some propositions can be both true and false. Note that constructivism covers the whole spectrum – some constructivists (in international relations for example) seem to be strong relativists – but a new brand of realism says ‘of course we construct all our scientific categories, but that does not make them any less real’. The issue with any brand of constructivism – discussed at length in this chapter – is whether there are any constraints on what we can construct, and if so, how strong they are. The stronger the constraints we think for any propositions, concepts, models, theories etc that we can construct the stronger is our commitment to realism. Note that one might be strongly realist when it comes to the empirical world, and strongly relativist when it comes to morality; some logical positivists denied that moral statements had any meaning beyond preference revelation.

12 The Philosophy and Methods of Political Science we can meaningfully and normatively assert. In Chapter 3, we shall see that many of the major theories of meaning develop their accounts in terms of the truth conditions of propositions. I argue that these accounts are normative as well as empirical, and that both the normative and the empirical are constrained by the nature of the environment we inhabit. (I use the term ‘proposition’, rather than ‘statement’ or ‘sentence’ – philosophers debate which of these expressions are truth bearers. This has some implications for realism/ anti-realism, but need not concern us overmuch.) Anti-realism is the antithesis of realism. In Figure 2.1 strong anti-realism is identified with relativism. Anti-realists largely hold that some propositions might be thought not to have a truth-value. I say ‘largely’ because some weak anti-realists plausibly deny they are relativists – indeed weak realism and weak anti-realism in some forms are hardly, if at all, distinguishable. One can deny that some propositions have a truth-value (or, more correctly, that they have a bivalent truth-value) in one of two ways. First, some propositions might be neither true nor false; they have a third (or nth) value. Second, one might hold that some propositions are both true and false. The first claim denies the ‘law of excluded middle’ (more precisely, perhaps, the ‘principle of bivalence’); whilst the second denies the ‘law of non-contradiction’ (see Box 2.3). The difference between these two laws is important in understanding claims about the way in which we see the world. Denying the law of excluded middle can be consistent with many forms of (weak) scientific realism. For example, one might think that there are truths of the matter over empirical issues, but not over moral ones, where the ‘truth-values’ of normative claims might take on a different meaning. Or one might agree that many empirical claims are too vague to sustain a truth-value, and might be better considered to hold a third or fourth value – say ‘useful’ or ‘not so useful’, and so on. We

Box 2.2

How to handle isms

Use isms only as a shorthand for a particular way of looking at the world and do not use them to critique others. You might want to defend an ism. You might want to suggest that ‘interpretivism’ is the right way of looking at the world; do so in a way that does not attack an erstwhile rival ism, say ‘positivism’ for getting it wrong. Indeed, rather than attack ‘positivism’, try to take what seem to be strong points in favour of ‘positivism’ and show how they are compatible with ‘interpretivism’. Never attack a particular argument, or a particular writer, on the grounds that it is or they are a ‘—ismist’; that is simply a form of ad hominem. Rather, attack what you think is logically, conceptually, empirically or morally, wrong with their argument or claims. Be precise when critiquing and use the general form of ‘isms’ only for very general purposes.

Isms

13

Box 2.3 Two laws Law of non-contradiction: any proposition, p, cannot be both true and false at the same time. Law of excluded middle (principle of bivalence): all propositions, p, are either true or false; there is no other value. Strong relativists are committed to the first claim; weak relativists, or weak anti-realists, to the second.

might believe that statements about ‘objects’ and ‘events’ have a bivalent truth-value, but theories or concepts do not, since theories and concepts are imposed upon the objects and events by us. One way of thinking about the strength of a person’s realism is the degree to which they are convinced that objects, events, models or theories are ‘real’. So some people think that objects have a reality that models or theories do not. I go all the way in thinking of everything as a candidate for reality, as I cannot see any reason not to do so. However, I also think that real things come in different forms or types of existence (though they all ‘really exist’). I admit I have no good argument for thinking reality does not come in degrees, whilst thinking there are different types of existence, other than that I want to be able to place propositions about fictional figures in truth tables, so making them real; but do not believe they exist in the sense that people around me exist. You might cut ‘reality’ and ‘existence’ another way, but you would have to convince me that your way matters enough to make it worth my while to argue with you. How far I believe any particular example of one of these things is real is, of course, a different matter. Denying the law of non-contradiction is a more serious matter. Logically any conclusion can be derived from a contradiction; thus if a proposition is both true and false, then any other proposition can be derived from it. So if we think there is a class of propositions that are both true and false, then from those we can derive any conclusion whatsoever. We come across this problem in explanatory contexts in Chapter 3 and with regard to testing hypotheses in Chapter 5. So if you believe that some proposition is both true and false, then you are led to radical or strong relativism where any claim or proposition goes. In order to protect realism or weak anti-realism, we need to dismiss the denial of the law of non-contradiction for any meaningful proposition. Or at the very least we might want to protect our propositions about the world from any contradictory proposition we hold by bracketing it off somehow so it does not infect any other beliefs.

14 The Philosophy and Methods of Political Science For example, the Christian doctrine of the Trinity of God in three persons is a contradiction. How can God be both one and indivisible and also three? Either this contradiction needs to be explained as not really being contradictory (thus theologians make distinctions between the one and the three, such as personhood being understood not just as individual but as community; distinguishing between substance and essence; or seeing the Trinity as three separate persons in essence but not in purpose, and so on); or the contradictory proposition needs to be bracketed away from all other beliefs so that it cannot infect them. The Trinity in this case becomes part of faith, one of the mysteries of Our Lord, and of no relevance to any other belief claim. I have suggested that isms are problematic because any ism is bound to be a summary statement of a set of inconsistent claims. Inconsistencies and problems in isms are, of course, the reason why introductory books are often planned around them. They become easy targets for the authors to shoot down before moving on to their own favoured position. I do not want to adopt that strategy, and therefore need to bracket them off. We should not let isms infect our beliefs, but rather regard the ism as simply the label stuck on to a group of people with some overlapping but inconsistent views. We should also recognize that the isms themselves overlap, and a consistent view might be formed using elements from many different isms. At best we use isms as a shorthand expression to cover some general ideas, but the real work of putting together a consistent account needs to be done elsewhere. So for me an ism is merely a label, and labels do not have to have any precise meaning – though the way they are applied might affect the objects to which they refer.

2.3 Nominalism, idealism Traditionally in philosophy, realism is opposed to nominalism, the latter being the thought that universals – the properties shared by all individuals of a certain class, such as black hair for the class of ‘black-haired people’ – have no existence beyond thought and are mere names for that which does exist, namely each person in the set of black-haired people. A nominalist does not deny that some things are real – realists in this sense are simply stronger realists than nominalists in the way I am defining realism. For that reason I have regarded nominalism as a form of weak realism in Figure 2.1 – the nominalist category is a little different from most of the others. A stronger contrast comes from opposing realism to idealism (or, in a less Kantian version, phenomenalism). Immanuel Kant distinguished categories of realism. Transcendental realists believe that the nature and existence of objects are wholly independent of us; the empirical realist believes we can perceive the real objects and gain knowledge of them. Kant also believed

Isms

15

that perception gives us knowledge only of appearances and not reality, so the empirical realist must also be a transcendental idealist – that is, material objects only consist in their appearances to us. This simple argument has caused all sorts of problems for philosophers over the years. It rests on the claim that might at first seem obvious: that ‘perception can only give us knowledge of appearances’. However, before we make too much of it, we need to interrogate the terms ‘perception’ and ‘appearances’ and ask what the term ‘only’ is doing there; more on this below.

2.4 Empiricism and positivism The philosophical categories of nominalism and idealism do not usually play a large role in methodological books on the social sciences. Empiricism and positivism do. In much modern literature on the philosophy of social science they are terms of abuse. Empiricism is the ad hominem insult used if someone makes a claim that has little or no theoretical background. To call someone an empiricist is to suggest they are atheoretical or their claims have no theoretical warrant. I argue in Chapter 3 that all empirical claims are backed by some theory: they cannot but be; but I will also suggest that it is not such a crime, sometimes, to conduct research that is backed by no explicit theory, nor designed to test any set of hypotheses. That will be my defence of ‘empiricism’. It is a defence that suggests we do not always need to conduct theoryheavy research, but should be aware that any empirical claim we make has to have some theory underlying it, even if that theory is something we take for granted. And the critics of empiricism are correct that perhaps we take for granted some things we should not. I have made empiricism a weak realist claim in Figure 2.1: since empiricists are thought to be atheoretical, they are likely to think that theories, laws or mechanisms that are used to predict and explain empirical evidence are not themselves real. Positivism, however, is the real enemy of much social theory. Usually, critics mean not positivism, but some form of logical or empirical positivism. Positivism is a thesis that demarcates the empirical world from the normative, and takes some claims about the world as given. Critics claim that we cannot demarcate the empirical world from the normative. They argue that every claim about the world (certainly any claim about the social world, but for many also about the natural world) binds us to some normative commitments. This is surely correct, but we have to be careful about our precise understanding of ‘normative commitment’. At the very least, we see the world in certain sorts of ways and not others. When walking through a wood we avoid bumping into trees: demarcating trees from air around them is thus important to us. Similarly, we try to avoid eating items that we believe will cause us harm or discomfort, and seek out food we find flavoursome.

16 The Philosophy and Methods of Political Science Again, this shows we have normative commitments in the way in which we demarcate the world. (And the food example demonstrates that there are ways of demarcating the world that affect us as a species, as well as aspects that might be due to more individual tastes deriving from personal characteristics and upbringing.) We also think that some people – cabinet ministers or rich industrialists – have greater effects on politics than other people. We consider the relative impacts people have on political outcomes because we think there is something normatively interesting about those relative impacts; or we avoid such questions because we do not want to consider the relative impacts, again a normative decision. Our perception of the world is normative because it depends upon making distinctions that matter to us. So much: so trivial. What do these sorts of normative commitments tell us about more developed ones – theories about the good life or about how the social world should be organized? They are not irrelevant – what is good to eat and what is not ought to inform public policy – but they do not tell us very much. In other words, whilst the way we view the empirical world already involves normative commitments, we should not make too much of that fact. Again, the sorts of questions that social scientists find interesting tell us something about what is important to them. That might reveal something about their ideology, but does not, on its own, tell us anything about the veracity of their findings. The methods that people use, and the kinds of evidence they find persuasive, might also tell us something about their normative commitments. Again, on its own that does not tell us anything about the veracity of their findings. Of course, if the methods adopted and evidence used do not lead to the claims they make, that tells us something about the veracity of the findings. And this book is about such methods and what we can reasonably infer from different methods. Labelling some approach as ‘positivist’, however, does not really help the interrogation of such findings. Nor does it follow from the obvious claim that all empirical representations of the world are normative in the weak sense. I have identified that we cannot demarcate the empirical and the normative in a stronger sense of normative – a sense that involves rival moral theories, or theories of justice, or theories about how society should be organized. It has to be demonstrated that a claim – that, say, 3 per cent of the variance in participation rates in collective action can be explained genetically – is determined by a specific moral theory, rather than relying upon a separate evidential basis. Of course, one might not want there to be a genetic basis to variance in participation rates in collective action, any more than one likes the fact that eating too much sweet food tends to lead to obesity or drinking too much alcohol to cirrhosis of the liver. All three might have public policy implications, and therefore getting them right, no matter what one’s other moral commitments are, should be important to caring people.

Isms

17

But is evidential basis normative? Yes, but only in the weak sense I have identified. It needs to be shown that certain sorts of evidence are moralized and not simply normative. This is often asserted but rarely demonstrated. I explore this issue further in Chapter 5 (Section 5.5 where I give an example of how concepts are theory-laden and the problems this creates for empirically testing hypotheses drawn from rival models). I think everyone is a positivist in the weakest sense of the term. The original label was applied to Comte for his argument that the laws of nature, such as the law of gravity (‘every point mass attracts every other point mass with a force directly proportional to the product of their masses and inversely proportional to the square of the distance between them’), have a different logical form from the laws of humankind, such as legislation against speeding (‘the penalty for speeding is £110 and three points on your licence’). I’ve yet to meet anyone who disagrees with this positivist claim. In that sense we are all positivists. However, anti-positivist social theorists often deny that there are any social laws (or ‘universal generalizations’, ‘law-like generalizations’, or simply ‘generalizations’) that apply to humankind, and the claim that there are such social laws (to be juxtaposed logically to human legislation) is a positivist claim. I discuss the nature of social laws in various places (notably Chapters 4, 5 and 6), but the belief that they exist is not specifically a logically or empirically positivist view. Karl Popper held that there were social scientific laws and he was certainly no logical or empirical positivist (see Chapter 5); whilst critical realists who see themselves as anti-positivist also believe in social laws, or what they prefer to regard as social mechanisms (just so there is no doubt they are not positivists). I discuss the distinction between a mechanism and a law (and argue that whilst we are interested in mechanisms in social explanation, they do at base rely upon generalizations). In fact, we make generalizations all the time: all collective nouns are generalizations. Generalizations are ‘law-like’ in so far as they are descriptions and natural laws are descriptions. Rather than juxtapose natural science with social scientific laws, we should be clearer about what any generalization is trying to describe (see Chapters 3–6). I therefore make a distinction between law-like or invariant generalizations and empirical generalizations. Positivism is compatible with a host of other claims, both realist and antirealist. Indeed many of the arguments used against positivism in the social sciences are directed at logical positivists who (in the main) were anti-realist. Logical positivists held a verification theory of meaning, which entails that metaphysical statements that cannot be empirically verified in principle are meaningless, and knowledge is based on the a priori (what can be known without reference to experience beyond that necessary for understanding the terms of a proposition) and sense-data (what we directly perceive). We have no warrant for considering the existence of anything outside of our experience

18 The Philosophy and Methods of Political Science of it. What makes logical positivism anathema for many people, however, is the logical positivists’ claim that because moral statements can neither be known a priori (though Kant had a stab at that) nor empirically verified, they merely express the preferences or desires or views of the speaker (the ‘boo–hurrah theory’ of morality). It is that distinction between the empirical and the normative that underlies much of the anti-positivist rhetoric. I do not think any serious moral or social theorist (that is, one publishing in refereed academic journals) born after about 1930 has maintained that logical positivist stance; it is a windmill against which we need no longer tilt.

2.5 Instrumentalism, conventionalism, pragmatism Instrumentalism, conventionalism and pragmatism are philosophy of science isms that concern attitudes towards the nature of laws. Instrumentalists hold that scientific laws are not descriptions of reality but instruments that enable prediction. The doctrine grew up in the face of the verification theory of meaning, since laws cannot be directly inspected and their meaning is generated simply by their usefulness in making predictions. Later instrumentalists were more struck by the Quinean argument that theories are always underdetermined by the data: that is, that there might always be rival theories that extant data do not allow us to choose between. Conventionalism is closely related to instrumentalism. It claims that theories and laws are accepted by convention, and there might be different, but non-rival theories that would work equally well. In part, conventionalism is a thesis about whether or not theories can be proved or disproved, corroborated or falsified, or whether they are chosen and discarded by convention depending upon how useful they are. Realists will agree that many of our theories are conventional in the sense that their specific form is chosen by convention (there might be non-rival but different theories), but they maintain that scientific principles and theories are not simply useful conventions but are also true. The world constrains what theories we can maintain in the face of our probing and testing. Pragmatism is a more general notion of conventionalism that takes its stance from the argument that all attempts to define reality and pin down the truth and meaning – to distinguish the analytic from the synthetic; the necessary from the contingent, and so on – have failed. Rather than retreating to full-blown scepticism or relativism, pragmatists maintain that truth and meaning are specific to social practices and are interpretations we give to our world (Rorty 1980, 2007). We hold certain things to be true, or give meanings to our words, simply to enable us to live our lives. What we take out of instrumentalism and pragmatism is indeed that laws are instruments to enable prediction; and that our understanding of the

Isms

19

world is based on the meanings we attach and the truths we understand from our words and theories. What keeps pragmatism from full-blown relativism, however, might make it compatible with full-blown realism. I have placed it on the weak anti-realist side in Figure 2.1, since that is where I think most self-identifying pragmatists would place themselves, but it can easily slide over to the right of the figure to strong relativism.

2.6 Essentialism, objectivism Essentialism is about meaning, naming and necessity. Quine (1953: 22) pithily summed it up: ‘meaning is what essence becomes when it is divorced from the object of reference and wedded to the word’. How far we are essentialists is somewhat important when it comes to thinking about concepts and thence to designing research. Essentialism might be thought to be important in how far we think we are constrained by the world. Can one be a person without a body? At what level of violence do we say that a polity is unstable? Does Napoleon have to be the person who marched across Europe or could that have been some other French ruler? What is the difference between insurgency and civil war? Virtually all research design requires one to make some decisions about the denotation of objects and events. The essentialist problem is that whatever decision the researcher makes, a clever philosopher armed with counterfactual examples can suggest the definition must be wrong. We have to make conceptual distinctions; language is not possible without them. Often the correct response to the clever philosopher is: ‘So what?’ The empirical researcher might respond: ‘Rather than giving me an example that shows my definition might be problematic, give me an example that shows it is problematic for the empirical case I am studying.’ Or: ‘Give me an example that matters and I might take your argument seriously; otherwise go back to your navel gazing.’ For example, political scientists and economists often use social indicators to judge relative welfare across countries, or across groups of people within countries. Such indicators are sometimes criticized using an example that shows that, even if a person scored more highly on that indicator than another person, we would not necessarily judge that person to have higher welfare. In other words, that indicator is not essential to welfare. Perhaps the philosopher might construct an example that shows that some people are better off if they leave school early: thus using the average age of leaving school as an indicator of social welfare in a country is problematic. However, the empirical researcher can reply that such a counterfactual example is not enough to demonstrate a problem with their research. Whilst it might be true that some people might be better off leaving school early, in order to demonstrate

20 The Philosophy and Methods of Political Science that that statistic is problematic for a country, one needs to show that there is bias across countries. That is, it needs to be shown that there are more such people who are better off leaving school early in some countries than others, or in some groups than others. Only if the distribution of such people varies across the cases is there a bias in the estimations (Dowding 2009). So the philosopher must demonstrate that a systematic bias exists across all the cases for his counterexample to have any relevance. He needs to demonstrate some systematic reason why the indicator is inappropriate for the cases given, not simply use a specific example to suggest it might be. It is hard not to be an essentialist to some degree (Gelman 2009); and almost impossible not to agree that essentialism is false when faced with a barrage of examples and arguments. Essentialism gets to the heart of what is necessary and what is not; and to a host of levels of necessity (logical and natural, de re and de dicto, to name just a couple of distinctions). I will try to avoid these specific issues. How far one is an essentialist, we might say, depends on how great one thinks the constraints are upon conceptual analysis. One might be tempted to think that the easier it is to define something, the more it has an ‘essential nature’, and that might tell us about the degree of its reality. Yet it is probably easier to define ‘prime minister’ than ‘politician’ and ‘politician’ than ‘Conservative’, but is it the case that ‘politicians’ are less real than ‘prime ministers’, or a person’s Conservatism less real than their leg? We will leave it simply that essentialism is tied to necessity and meaning (see Chapter 8). Objectivism is usually counterposed to subjectivism; it would be wrong, however, to place objectivism under realism and subjectivism under antirealism. Putting objectivism under realism is pretty safe, but the distinction is orthogonal to realism/anti-realism so I omit subjectivism from Figure 2.1 altogether. Objectivism is the idea that judgements are either objective – they refer to objects – or subjective – they refer to subjects: that is, to people. ‘The prime minister is a member of the Labour Party’ is an objective statement; ‘the prime minister is a buffoon’ a subjective one; ‘most Conservative voters think the prime minister is a buffoon’ an objective one. It is standard, especially in the seminar room, to label a claim ‘subjective’ if there seems to be little evidence for it. It makes the claim a ‘preference’ and, when generalized, asserts that all claims are mere opinions or desires rather than having some stronger evidential basis. This shifts the nature of the term ‘subjectivism’ from its original reference towards relativism, a fate subjectivism does not deserve. If all propositions were mere opinions, then any claim is relative to the person who makes it. The subjectivist move is applied in ethics where subjectivism suggests that all morality is simply the expressed desires of people (the boo–hurrah theory again); objectivism in morality is then the view that ethical values concern objects completely outside of people and their attitudes. Both views seem equally implausible. A moral objectivism that sees morality as something rationally defendable seems more defensible, whilst realism in moral matters

Isms

21

also does not have to completely divorce value from the people who hold those values. The relationship between objectivism and essentialism concerns the degree to which objects can be thought to have essences. The extent to which any object can be divorced from any way of describing it defines the degree of essentialism of objects. If there are no constraints on ways of describing it, then all definitions would be subjective in the sense that they would express only the views of the definer.

2.7 Critical theory Critical theory is not, in name at least, an ism. However, it underlies some of the other isms and is worth considering here. Critical theory has an interesting place in relation to the realist/anti-realist divide I have specified. It might seem to be anti-realist. Originally it was a development of Marxism that was critical of mainstream philosophical and scientific thought. It critiqued the idea that science was value-free, suggesting that all schemes of thought operate within ideologies and that closed theoretical systems need to be examined from without. This seems to suggest anti-realism. What makes critical theory ‘critical’ is the idea that no system – including itself – is without critique, together with the idea that the end point of social theory should be action to change society (for the better). So the underlying motivation within critical theory is that social scientists should be aiming to overcome exploitation and domination; and that claim implies some idea of transcendent value. The realism of critical theory becomes more apparent in its second generation, inspired by Jürgen Habermas. His great contribution is the analysis of communication and its role in deliberation. The idea of communicative competence is that free and open communication can overcome distorted ideological discourse through deliberative practices (Habermas 1984, 1998). Critical theory in this sense involves using insights from analytic philosophy to reflect upon the norms and practices of discourse, understanding their biases and rules to help design institutional processes to reform discursive practice. Critical theory tries to engage with the experience of people, but also tries to uncover hidden structures and suggests that knowledge is power. Again the end product is supposed to be action. In these latter elements critical theory is realist; and in the sense of trying to find hidden structures it is strongly realist. Many who use critical theory, however, are more taken with the critical aspects – critiques of mainstream philosophy and science (especially social science) – and with the idea that theories are partial and ideologically driven. Thus what people often derive from critical theory in many approaches is an anti-realist and often strongly relativist bent. (As a further consequence of expending so much effort on the

22 The Philosophy and Methods of Political Science critical, critical theorists tend to come up rather short on the ‘action’ end, which once was supposed to be the point of critical theory.)

2.8 Postmodernism Postmodernism is strongly relativist. It begins with the claim that reality is a social construction. Realists need not disagree with that. However, postmodernists go further and suggest that our language is no more than socially agreed-upon gestures that have no denotation or reference to anything external to the discourse itself. They argue that people are born into communities with already agreed-upon meanings; they utilize and thereby reproduce those meanings themselves; language has no capacity to mirror or represent objects. The reason why postmodernists think denotation or reference is impossible seems to be because once a name or definite description is uttered it already has, through that utterance, connotation or sense beyond denotation. Thus words have more than a referential label to objects. I am not sure this is entirely true. I think a grunt and pointing to an object could lead someone else to, say, pass that object to me, without our agreeing on what the object is. However, even if the claim is true, I cannot see how one can infer from the fact that words have connotations or sense that they cannot also have denotations or reference. Indeed, postmodernist thought seems dominated by exclusivity. Either there is truth or there is not; either an object is real or it is not; words have either denotation or connotation but not both; either language is a form of domination or it is not. Postmodernists do not countenance the idea that some claims might be true and some not; and some might be partially true or hold some third value. Or that reality or existence might come in different forms (so a fictional character’s reality is somewhat different from that of a living person; or ‘rights’ might exist in a different sense from the desk that my feet are under); or that terms might have sense and reference; or that language might sometimes be used to dominate and sometimes not; or that there might be a continuum to domination, with some aspects of language leading to strong domination, some weak, some not at all, and so on. Postmodernism claims there is no final umpire to rule on truth claims. Its practitioners point out that what we believe is true has varied over time and varies across communities. But this does not show there is no truth. It may be easier to concentrate upon those views of the world that have changed than think about what has not. For example, whilst what tastes good depends upon individual palate – which might depend upon the food one has eaten previously, which is strongly culturally determined – the basic ingredients do not change; and poisons remain poisons. In other words, whilst preferences can vary greatly, they are constrained by something.

Isms

23

I am impatient with postmodernism’s supposedly radical but to me obvious propositions – reality is socially constructed; what we think to be true has changed over time – leading to conclusions which do not follow; therefore there is no reality, there is no truth. The riposte that I am using the traditions of the Enlightenment and the canons of analytical logic does not move me, since as far as any sense can be made of any claims whatsoever, whether ancient, modern or postmodern, the necessary thought process requires some basis in reasoning that approximates these canons. I consider postmodernists to be hypocrites who make verbal claims, but make no attempt to live their lives by them, preferring the comforts provided by the Enlightenment and its canons of reasoning. What can be said for postmodernism is that it made a lot more sense in its original form as applied to modernist architecture; and that it has brought to the surface various ideas about domination through language, especially in the form of discourse analysis to which it is closely related. Unfortunately, many of the lessons are wrapped up in a language that makes them harder to grasp than they need be, perhaps because the writers’ fear of analytic philosophy leads them to celebrate obscurity (either language is clear or it is not; analytic arguments are clear, so postmodern ones must be obscure).

2.9 Interpretivism Interpretivism comes in many forms. To some degree it derives from hermeneutics and phenomenology, and is closely associated with postmodern and post-structuralist philosophies. It is often contrasted with behaviouralism and positivism, though some interpretivist approaches are compatible with many elements of the latter. The basic idea of interpretivism is that the study of human society needs to proceed from within: within the researcher who must interpret the actions of subjects and, as far as possible, from within the subjects and their society. Understanding social life is about meanings, and the meanings the interpretivist privileges are the meanings of the subjects. The intentionalist branch of interpretivism begins by examining the beliefs, desires or intentions of subjects. This is perfectly compatible with versions of behaviouralism that give a role for such objects. Other versions of interpretivism concentrate upon social meanings – traditions or ways of thinking within societies that lead individuals to take on the meanings they do. There is a clear link here to structuralism and post-structuralism. Interpretative theories of government or ‘governance’ (which signals emphasis away from formal institutions) are narratives. They concentrate upon networks of groups who might have their own ways of thinking or distinctive traditions or explanations of the same historical events or operations of institutions. Sometimes interpretivists point out problems or contradictions

24 The Philosophy and Methods of Political Science in such traditions, or argue that clashes of traditions within cultures or polities lead to governance problems. Interpretivism in this form is relativist, since it maintains that traditions are different from each other and none can be said to bear the ‘correct’ narrative of the same events. If the argument is that no narrative is either true or false, but each has a separate or third truth-value, then interpretivism with regard to narratives is weakly relativist. If interpretation or narrative in this sense is separate from causal explanation, then interpretivism is compatible with positivist, empirical, behaviourist and realist research. (Or it is compatible if the causal claims within such interpretivism are restricted to the effects of the different narratives upon people’s behaviour.) If it is not seen as a separate activity, however, and such narrations are seen as equally true and rival, then interpretivism is strongly relativist. Interpretivism in the first sense provides valuable explanations; indeed it provides the last chance for humanist explanation. Interpretivism in the stronger sense, however, is nonsensical; the conclusions do not follow from the premises. Many interpretivists seem to swing from the weak to the strong interpretation.

2.10 Critical realism Critical realists define themselves against both relativists of the sort I have just been discussing, and positivists. As realists, they accept the laws of non-contradiction and excluded middle, and are strongly realist in holding that theoretical entities are just as real (if not more real) than the surface properties of social life (Bhaskar 1979; Sayer 2000). These theoretical entities are those structures and mechanisms that lead people and societies to certain ends. Their critique of positivism is a critique of the (early) methods of quantitative scholars in their interpretation of correlations between independent and dependent variables in linear equations. They argue that any causal claim made must be theorized and any causal relations must be due to structures or mechanisms that underlie the phenomenon to be explained. When those critiques of standard regression techniques and causal claims were first made by critical realists in the 1970s, there might have been some justification for them. However, in the last 40 years quantitative scholars have worked incessantly to critique and improve their methods precisely to demonstrate more clearly underlying causal processes, and to model the structures underlying their data. In that sense, positivist quantitative scholars are all now realist. They are not, however, all critical realists. Whilst the basic thesis of critical realism is that we can only make sense of cognitive practices such as science by assuming they are about something that exists independently of us, critical realists are also determined to demonstrate scepticism about any current knowledge. Hence they note

Isms

25

that (a) particular truth claims at a particular time might be false; (b) there is reflexivity about the condition of knowing about something – theorizing is itself a form of social knowledge; (c) surface appearances can be misleading (hence the need to look at underlying structures); (d) overcoming misleading appearances means that current beliefs are always open to later correction. An important element of critical realism is that mechanisms gain veracity when people find them independently of each other. For the critical realist transcendental arguments are important for demonstrating the reality of social structures. An example of such a transcendental argument is: We accept p. For p to be the case there must be c. Hence c must be the case, since given p, then c. An example of the transcendental argument for structures is: Person j is poor. For someone to be poor there must be a structure of inequality (‘poor’ is a relative term). Hence there is a structure of inequality. What makes critical realists ‘critical’? In part, it is the reflexivity of any specific claim to reality. However, they adopt the term ‘critical’ in line with the idea of critical theory that the point of social science is to uncover the relations of exploitation and dominance that underlie social processes. They are critics as well as analysts of the society they study. Thus the other element of critical realism is an emancipatory critique of society. So, for the example given, once the structures of inequality are demonstrated, we can work to overcome them. (One might accept these kinds of argument without believing the corollary. For example, one might think there is a natural ‘power law’ of inequality entailing that inequality could never be overcome. Again though, demonstrating that there seems to be a power law is not the same as showing it cannot be overcome; and since the amount of inequality varies across societies, any extant level of inequality is demonstrably not a necessary feature of society.) How far the ‘action’ element is actually taken seriously by critical realists in their work is open to question. Critical realists are opposed to ‘reductionism’, which posits a hierarchy of levels of reality, such as: social science psychology physiology/anatomy

26 The Philosophy and Methods of Political Science organic chemistry/biology physical chemistry physics. They argue that whilst there might be lower mechanisms, these cannot explain what goes on higher up the chain. So, for example, whilst physiology can explain why people can speak, it cannot predict what language they will speak. Furthermore, higher levels causally affect lower levels, since the organization of society might affect future physiology. Critical realism is explicitly committed necessarily to emancipatory politics. Why is it necessarily committed? Because truth is preferable to falsity. Critical realists believe that the surface reality of capitalism seems to justify its form, but once we see how the real structures emerge, then we find that the surface reality hides patterns of domination and exploitation. This might be true – though I am not sure why it is necessarily true, and that would be my only criticism of critical realism – but without the emancipatory element, critical realism is simply a specific form of realism.

2.11 Naturalism Naturalism can be thought of as the thesis that everything is natural – there is nothing that is supernatural. Another way of thinking about it is that explanation of everything is unified. This does not necessarily imply that all explanation can be reducible to, say, physics, but does deny there are any sui generis subjects that involve special methods or argument. The most obvious subject that is denied naturalistic status is ethics. Naturalism sits easily with realism, although some who seem to be naturalist have tended towards an anti-realist idealist framework – some logical positivists, for example, or even David Hume at times. (Hume also provided a non-naturalist account of value, but then seemed to slide morality back to naturalism with moral psychology.) Essentially, in metaphysical terms, naturalists think that anything that science comes up with – including theoretical entities – must be fitted into our naturalist framework of the universe. In the main, naturalism is an epistemological thesis about how we should go about gaining knowledge. In response to the sceptical challenge to justify our beliefs, naturalist epistemology suggests that maybe we cannot ‘ultimately’ defeat the sceptic, but in practical terms we must accept what we have learned through science so far. In one sense, the sceptical challenge is similar to the argument that we cannot escape the context of our society, so we cannot claim universal beliefs. This invites the pragmatic response that, whilst it is true that we cannot escape our context, our context still leads us to make universal claims.

Isms

27

Ethical naturalism is the idea that (1) ethical terms are definable in nonethical natural terms, or (2) ethical conclusions can be derived from nonethical premises, or even (3) that ethical concepts are natural ones. The first was famously attacked by G. E. Moore (1903) as committing the naturalistic fallacy. In one sense, Moore must be right – good cannot mean, for example, pleasure or the sentence ‘it might not be much fun, but it would be good’ would be contradictory. Nevertheless, ethical terms might be entailed by a basket of natural ones, which slides into the second account. Any account of morality, such as Hume’s or those who see morality constrained by Darwinian selection processes (that is, that moral conventions must be evolutionarily stable strategies – see my comments on Binmore below), will be naturalistic accounts. Naturalism in science is virtually always realist (hence I have put it on the realist side of Figure 2.1), but, as noted below, naturalism in ethics might be realist or anti-realist with regard to ethical concepts, yet realist in the explanation of ethical terms.

2.12 Constructivism I’ve placed constructivism all the way from strong relativism to strong realism. In the social sciences, constructivism is a term, used widely in international relations (IR), that shares some ideas with scientific constructivism but whose heritage is rather different. Proponents of IR constructivism (as I will term it) claim it is based on ideas from critical theory, shares some ideas with post-structuralism and postmodernism, but engages more with empirical international relations than any of those approaches. Critics from those approaches claim IR constructivism buys into positivism, whilst critics from the other major IR camp – ‘realists’, ‘neo-realists’ or ‘neoliberals’ – suggest that either it is too naive or its major insights can be encompassed with their own realist paradigm. It should be noted again that ‘realism’ in international relations has only tangential relationship to scientific realism as discussed in this chapter and as it appears in Figure 2.1. Here, however, I will concentrate upon constructivism in the philosophy of science. Traditionally constructivism in the philosophy of science is juxtaposed with realism, hence the anti-realist side in Figure 2.1. The idea of constructivism is that to assert there is a given object with certain properties is to assert that one knows how to discover or construct such an object. The thesis was originally applied to mathematics: objects within mathematics are composed of our constructions and, say, real numbers are constructions whose properties follow the rules that we have also constructed. The thesis has some purchase over realism in mathematics, for it denies the existence of some objects that are assumed by realist mathematicians. For example, the proof that certain sets lack a given property does not prove there is a set that

28 The Philosophy and Methods of Political Science has that property, unless one provides a method for constructing such a set. Mathematical constructivists also reject the law of excluded middle. Beyond mathematics, constructivists maintain that many objects of science are also constructed, because many of them are seen to have the properties they do simply because it enables prediction. For example, in order to maintain certain claims in particle physics, particles are theoretically constructed, and then predictions are made as to what observational evidence would be required to confirm their existence. Both the theory of the particles and the observational evidence (and of course the physical machinery to collect the observations!) are constructed. We can see this is far from logical empiricism or positivism and the idea of ‘sense-data’ that are directly perceived by us. Some constructivists in physics see certain similarities between these processes in the natural sciences and those described by relativists with regard to social constructions; hence constructivism is often viewed as anti-realist. However, it is not necessarily so. I suggested above that the Kantian claim that ‘perception can only give us knowledge of appearances’ rather than reality, entailing that the empirical realist must also be a transcendental idealist, has caused all sorts of problems for philosophers. Now is the time to interrogate that Kantian claim, particularly the terms ‘perception’ and ‘appearance’, and ask what the term ‘only’ is doing there. The claim rests on the idea that if there is a world external to us, there must be something more to it than what we perceive. Why should we think that to be so? One obvious answer is that we have discovered that there are properties of objects that we have not previously perceived. The invention of microscopes, electron microscopes, infrared cameras, and so on has demonstrated all sorts of properties that we cannot see with our own eyes. We know that objects can appear differently in different lights, that the play of the sun on hot deserts or tarmac can give the appearance of water when closer inspection reveals none, and so on. Thus we know there is more than we directly apprehend with our eyes. Similarly, instruments have shown that there is more to sound than we directly apprehend with our ears. We know that some animals have much greater sensitivity to certain surfaces than we do, so there are feelings that are possible beyond those of our own sensors. Thus the properties we apprehend at any given moment with whatever instruments we have available to us are not all the properties that are apprehensible with other instruments. We can say ‘appearances can lie’ because of these facts, and phenomena such as mirages. However, can we conclude that perception can thus give us knowledge only of appearances? Consider the mirage. Do we say that there are two appearances, neither of which is ‘real’, or do we privilege one over the other? For a thirsty person stuck in the desert, that is not a serious question (or rather, the answer is so serious you do not bother asking the question). Nevertheless, do we conclude

Isms

29

that water is not just its appearance? It is what it looks like, and what it feels like going down the throat. The mirage is a mirage because one of the appearances is missing. Water can also be described by other properties, ones that require instruments of perception other than those we are born with. And it is H2O, which we know through another ‘appearance’. We might say that water is ‘really’ all its appearances. Why do we think there should be anything else to reality than all its appearances? Just because we know the world has appearances that once we did not apprehend, and believe it might well have appearances that we do not yet know of, and might even have appearances we shall never know of, why should we think it is more than assembled appearances? One answer is that some appearances are false – the mirage water, for example. That is simply the light effect of heat rising from a dark surface into a brighter one, which looks to us like water. However, we can know this is a mirage by comparing observations of the ground from different viewpoints, and can provide an explanation of how the mirage operates through the way our perception works. So the mirage is real – it is a real mirage. Another way of thinking about this issue is to ask why we should privilege the appearances that we perceive with the senses nature has given us over the appearances we can perceive with instruments or through modelling. After all, many of the appearances we gain by physical instruments are visualized through that machinery by mathematical manipulation. Water being H2O is something we know through measurement given theory about that measurement. Our knowledge of many aspects of our society – evidence about mass beliefs gleaned through surveys; analysis of the voting patterns of different groups; underlying rates of inflation, and so on – are estimates based on the mathematical manipulation of data. Some of this manipulation is straightforward descriptive statistics; some uses more complex regression, maximum likelihood or other techniques, to estimate what is considered to be underlying reality. And why not? We perceive realities through our instruments, whether physical or virtual. Other realities might be entirely theoretical. One example is the centre of gravity – the centre of the mass of any object in the presence of a gravitational field. There is nothing to be seen there to mark the centre of gravity, but it can be perceived in a sense that enables predictions about the relative movement of large bodies. That is the reality of the centre of gravity (Dennett 1998). We can call such things abstract objects, and we can deny them existence in the sense we admit the existence of tables and chairs. But why privilege the senses that nature has given us? One form of realism is to allow anything that can be measured to count as real. (See Chapter 8, Section 8.5 for what I mean by ‘measured’ here.) So even mirages are real – they are real mirages. And we can predict under what conditions mirages will be seen. Of course we make ‘perceptual errors’. We think we see something when we do not. We make errors when performing mathematical calculations, producing

30 The Philosophy and Methods of Political Science mistaken beliefs about underlying reality. But the fact we can accept there are errors shows us that there is something that is not an error. Errors occur when we do not follow the rules or procedures associated with the task we are doing. Grasping the nature of error when considering appearance and reality is not easy, but I shall argue repeatedly it has something to do with inconsistency and non-replicability. So we have interrogated the ideas of ‘perception’ and ‘appearance’. Perception can be whatever we do to find out about the world. It is not untheorized (to think it is would be to return to the logical positivist’s notion of ‘sense-data’), whilst appearance is whatever we glean through our perceptual devices. But why must we think that reality is different from these appearances? Why can perception ‘only’ reveal appearance? The obvious answer to the question ‘What is the real shape of a cat?’ is ‘It depends on what angle you are looking at it from.’ We do not need to think that cats have ‘real shapes’ other than the shapes we see – they can all be real (even if some are real distortions through a lens). A stick appears bent if held in water, but is really straight. The bent appearance is a real bent appearance; what we mean by ‘really straight’ is that we consider as real the straightness of sticks when not immersed in water. That is the conventional meaning of ‘straight stick’, but the fact it is conventional makes it no less really a straight stick. We can also do more careful analysis and measure various sticks, again using conventional techniques and meanings, to see which is the straightest. In this notion of realism the term ‘real’ can drop out of most of the sentences above where it appears. ‘Real’ is used relative to something else – an ‘appearance’ or an ‘error’, which marks the fact that what is being attributed or denied ‘reality’ is relative to some standard use of what we think the meaning of the term is. The scientific (and indeed our ordinary) understanding of reality in this sense is how well what we accord reality allows us to navigate through that aspect of life with which we are engaged. That is the underlying sense of ‘prediction’ as discussed in Chapter 3 (see Box 3.2).

2.13 Realism and relativism in ethics At the bottom of Figure 2.1 in square brackets I have represented the entire spectrum for realism and relativism in ethics. As we have seen, one might be a strong realist over empirical matters, but a strong relativist when it comes to morality. One might believe that the world completely constrains our description of the empirical world: there are straightforward truths of the matter about the empirical world whether or not we can discover all those truths. At the same time one might believe that there are no such truths in the world about ethical issues. One might believe that the empirical world is to be discovered, but that our moral universe is entirely created by us and moral

Isms

31

statements simply reflect our preferences or desires. Or one might think that we are constrained in our ethical views, not by the physical world as such but by the social world; so our ethics are simply conventions that our cultural groups have stumbled upon or follow because of the teachings of some past wise or inspirational leader. Different communities might therefore have completely different conventions or rules governing their ethical attitudes and behaviour. Since morality is simply a set of conventions or rules that are created by communities, we can have no warrant for claiming any specific rule or convention is the right one, or better than any other. Morality is strongly relative to culture and community. Moral realists need to deny the relativism that follows from this plausible view that ethical rules are simply conventions created within different communities. Moral objectivists maintain that moral rules exist outside of us. Adherents do not usually defend their religion on the grounds that a wise or inspirational leader offered guidance that became conventions for that group; rather, it is claimed that religious rules are revelations from God passed to the world through his prophets. Here morality, like other empirical reality, is external to people. It is also possible to be a secular objectivist. Here the ‘objectivity’ proceeds not from a sentient being outside of humanity, but from humanity itself. However, the objectivity is the full and complete rational consideration of how people should lead their lives. It might partly be based on ‘intuition’ (see Chapter 9) – some objectivists suggest that moral intuition gives one a direct insight into morality. Others say that objectivity comes about by considering how we can promote non-contradictory, coherent and consistent morality. For example, Kant’s categorical imperatives are supposed to be rules that we have to accept, since we cannot all will their contrary. If morality is universal and we cannot all will the rules’ contrary, but can will the categorical imperatives, then these must form an objective base for morality. There are other moral realist positions that are not objectivist in this manner. One might maintain that ethical rules are real in the sense that one can run a (bivalent) truth table over any action to see if it conforms to the convention of a given society. Thus ethical claims are true relative to the community rule. Of course, this type of ‘realism’ is compatible with what earlier I termed relativism. But we can as a community examine our rules and conventions, and change them as we find contradictions or problems, or refine them as circumstances change, thus bringing in some objectivist constraint upon the set of rules. We might also suggest that the empirical world constrains the set of possible conventions. Some moral rules might be inefficient and civilizations that have inefficient rules in relation to others with which they clash must either change their rules or see their culture decline. Natural selection and evolution will thus ensure that there are empirical constraints upon the set of moral rules.

32 The Philosophy and Methods of Political Science In very general terms this final realism in morality is one that I maintain. Ethical principles exist to the extent that we think and act as though they do. Thus someone has a right to, say, their car, just so far as everyone in their society recognizes that right and (by and large) respects it (Dowding and van Hees 2003: 288–90). In that sense I call myself a moral realist. However, underlining how clumsy these titles are, the position I have just elucidated is very similar to that suggested by Ken Binmore (1994, 1998 and 2005) in his books on justice. Whereas I call myself a moral realist, Binmore calls himself a moral relativist. And why not – he argues (just as I do) that many moral rules are simply conventions, equilibrium strategies in the game of life that differ across communities. They are relative and culture-bound. That is fine for my realism too, since the real moralities are culture-bound by what people accept, but I concentrate upon the fact that there are constraints on what is acceptable, and that what is acceptable might change and converge over time as globalization occurs. So how come I call myself a realist and Binmore labels his beliefs relativism? We might simply define ourselves against our targets. I might call myself a realist since my targets (my ‘enemies’) are relativists who are struck by the fact that morality is culture-specific and argue that therefore we cannot criticize other cultures’ morality, since we will always do so from inside our own. I do not think that claim follows from the culture-boundedness of moral conventions. Binmore’s targets are what he sees as realists who want to universalize their own views to the rest of the world (following Bentham, he calls them ipse dixitists). I think, in fact, that is the function of isms more generally. We define them by what they are not. They can be helpful shorthand, especially for what one is opposed to. Often we define ourselves, and our views, in terms of what we are against, rather than or as much as what we are for. However, when one gets down to the fine detail, we often have as much in common with some of those labelled as being in the opposing camp as we have with those counted as within our own camp. It is the detail that is important.

2.14 A few isms not in Figure 2.1 Finally, I round up some isms that were excluded from Figure 2.1 because to a large degree they are orthogonal to the realism/anti-realism divide.

Reductionism Reductionism is again one of those isms often employed to damn an explanation. The claim is usually that one set of phenomena is being reduced to another set in an illegitimate manner (see the discussion of critical realism in

Isms

33

2.10). In one sense, all explanation is reduction. If you want to explain something, especially something highly complex, you need to simplify and break the phenomenon down to constituent parts to see how these fit together and how they contribute to the explanation you are attempting. The charge of reductionism, however, is directed to a shift in the nature of the explanation. So, for example, shifting the explanation of the reasons why someone votes as they do from their attitudes understood in terms of mental states to an understanding in terms of their brain states would be such a reduction. The claim of ‘reductionism’ is the charge that it is illegitimate to such an explanation to reduce mental states to brain states. So reductionism has several meanings, one of which is trivial. Explaining any proposition by others requires some reduction. But that is not really the way ‘reductionism’ is used as a charge. The critique of reductionism is that we reduce explanation from one sphere to explanation in another. In order for that to be a critique, however, one must show that it is illegitimate to make that shift. One must demonstrate that something is lost in the explanation by such reduction. How can one show that? The most obvious way would be to show that it is harder to predict outcomes by such a reduction. So reducing social explanations to physical ones is a reductionist strategy that would (certainly with our current technology) make it much harder to predict what is going to happen socially. As long as it is more efficient (easier) to predict voting by trying to measure mental states than brain states, then we should stick with mental states when explaining voting behaviour. (Or we might omit both and explain voting outcomes from objects without bothering with mental or brain states.) It is clear that (a) since explanation always involves simplifying, a form of reductionism is an advantage; (b) reducing from one sphere to another might sometimes bring clear explanatory advantages and clearly sometimes does not – but one must substantiate the charge of ‘reductionism’ as a critique by demonstrating what is lost; (c) explanation as cause is a specific form of reductionism, discussed in Chapter 3. I see the charge of reductionism as mere abuse from those who have run out of other things to say. That is not to claim, in my example above, that reducing mental states to brain states in an explanation of voting behaviour would be a useful thing to do. But its utility or disutility as an explanation is something that must stand or fall on grounds other than the supposed reductionism; one must show that something important to explanation is lost in the reduction, in the manner of critical realism above.

Foundationalism and coherentism Foundationalism is related to reductionism in the sense that it is a thesis about building explanations up from a base that cannot be further analysed.

34 The Philosophy and Methods of Political Science These basic propositions are ones that are indubitable; from them we build up more complex propositions about the world. In many versions these basic propositions are equated with sense-data. Foundationalism is often associated with ‘correspondence theories of truth’, where basic atomic propositions correspond to simple or logical atoms in the world. How realist foundationalists are might depend on how far they see complex propositions as real. One might be a realist with regard to the atomic propositions and conventionalist with regard to more complex propositions or theories built up from them, for example. Coherentism is juxtaposed with foundationalism. Coherentists argue that there are no indubitable foundations or atomic propositions that stand on their own. Instead all our beliefs are theory-laden and as such each proposition is affected by each of the others. They argue that changing our beliefs in one area is likely to reverberate through our whole system of beliefs. These reverberations will affect the meaning of propositions, sometimes right down to our most basic beliefs. Most modern accounts of this coherentist view look back to Quine’s (1953, 1960) attack upon the notions of the a priori and analyticity, where he suggests that our knowledge is a ‘man-made fabric which impinges on experience only along the edges’, and further that ‘no particular experiences are linked with any particular statements in the interior of the field, except indirectly through considerations of equilibrium, affecting the field as a whole’ (Quine 1953: 42). Knowledge might be seen as a balloon that expands with our science: propositions at the edge of the balloon always revisable; ones in the centre less likely to be revised, or less affected by the expansion, but affected just the same. Quine’s target was the verificationist theory of meaning. Coherentism appeals to anti-realists because it suggests that all knowledge is in flux and there is nothing empirical or secure that ties human experience to the world. But anti-realism does not follow from coherentism, and Quine himself was a realist with regard to scientific knowledge (though something of a ‘boo–hurrah’ theorist over ethics). Another possibility is to combine elements of coherentism and foundationalism. One might be sceptical of foundationalism as a theory of knowledge because one is sceptical about the possibility of there being indubitable ‘atomic propositions’, yet be foundationalist within certain domains. To take the foundations analogy seriously, we might think of different areas of knowledge, such as mathematics, physics, political science, as structures that are supported by foundations. Those foundations support the structure because they are embedded within a larger structure (the earth), and it is that which provides the theorized underpinning for the basic accepted ‘indubitable’ propositions that found these areas of thought. Thus when we conduct research in any given area we can accept as given our knowledge base in other areas, even though we acknowledge that, in theory, changes at the surface of our structures affect other parts of our knowledge.

Isms

35

Holism and individualism Holism is the thesis that the properties of any individual are determined by the relations they bear to other individuals within the complex whole. In that sense it is connected to coherentism. So in a theory of meaning, the meanings of any words or sentences are thought to depend upon the complex set of relationships between the whole set of words and potential sentences within a language. It is usually thought to follow, especially in a theory of meaning, that as the whole changes (new words or sentences are formed), then the meaning of existing words and sentences might also subtly shift. It is the identification of the individual with the whole that characterizes holism and not simply the fact that some relational properties can only be defined in relation to the whole and not any individual. Within social and political science a major debate once raged over whether we should be methodological individualists or holists. The methodological individualists suggested that all social explanation could be reduced to the interaction of individual agents under the thesis that only individuals act. Methodological holists suggested that we cannot understand human motivation if we take it outside the social environment within which people operate. Holists tended to see the operation of universal processes or laws upon human aggregates; individualists were more concerned with the ability of human agents to have freedom of action. Over time, however, individualism became associated with methods that sought generalizations and mechanisms that explained social processes, concentrating upon the constraints under which individuals act. I think this methodological debate is now largely dead, with both sides recognizing the importance of structural constraints on human action, and the ability of individual agents to try to overcome social processes (List and Spiekermann 2013). (There is a lot of social theory about the precise way we view this, though by and large it is entirely unempirical; and it seems to me the rival camps are all saying (much) the same thing (Dowding 2008a).) Everyone now accepts that humans are social animals and that the environment in which we act has a strong effect upon our attitudes and beliefs. There is no particular relationship between holism and either a realist or anti-realist stance, though most individualists would privilege agents over other social forces as ‘real’.

Chapter 3

What Is an Explanation?

3.1 Introduction The place to begin thinking about methods of enquiry is with consideration of the nature of explanation. Research is designed to explain things; that means it is designed to answer questions. We cannot think about what constitutes explanations without thinking about the sorts of questions we ask. Some questions are about causal processes, some about which conjectures or theories are correct and which false, some about what best characterizes an object or event. Sometimes one can answer a question by giving a description. Description is an important part of political science, just as important as theory. Too many undergraduate textbooks in political science are short on description and long on theory. Writing and reading theory is much more fun than carefully describing and understanding institutions and political processes. But without close description, without precise understanding of institutions and processes, we cannot come to clear judgements about grand theories of politics. But does description actually explain anything? If explanation provides more than description, what more does it provide? What exactly is an explanation? We all seem to recognize an explanation when we see it, even if we don’t all agree on what is a good one. That is, we recognize the logical form of explanations even when we disagree over how good particular ones proffered are. So we might think that it ought to be easy to specify precisely the logical form of an explanation. It is not. In fact we might suspect that ‘explanation’ does not have a logical form. There are many different kinds of explanation; more importantly, what constitutes an explanation is always relative to the person who asks the question that the proffered explanation tries to answer. Thus the worth of an explanation depends upon the knowledge, the interests and the competence of the questioner (Achinstein 1981, 1983). Since these need to be built into an account of explanation, endowing explanation with a set logical form that is not itself trivial might be impossible. I will examine one attempt to provide the logical form of scientific explanations: the Deductive-Nomological (DN) model of explanation. The DN model fails quite spectacularly to do the job it was supposed to do and now is rarely discussed in detail. However, seeing why it fails is instructive. Moreover, the desiderata that Carl Hempel (1965, 1966) had in mind in

36

What Is an Explanation?

37

setting it up are interesting in themselves. My account of the DN model, its failures and the desiderata closely follow Peter Achinstein’s account in his The Nature of Explanation (1983), which remains one of the best books on the subject. First, however, I shall say a little about language and our desire for explanation. Why do we want explanations?

3.2 Language and the world What is obvious is that explanations are couched in words and sentences. At the very least, an explanation of something in the universe is a verbal representation of something about the universe (even if it were the ‘universe of words’). Before we can try to get to grips with what an explanation is, we need to think about language and its relationship with the world. Words might be seen as labels stuck on objects in the world – we name one object ‘Keith’, another ‘Fido’. In that sense they merely refer to those objects, and could be thought to be atheoretical. This might be one way of thinking about reference or denotation: there is a correspondence of a word to a thing, and the word simply refers to the thing. Nouns are often thought of in this way: Keith is a human and Fido a dog. The nouns ‘human’ and ‘dog’ are labels given to (natural) objects that enable us to describe and navigate the world more easily. However, nouns are not entirely atheoretical (hence nor are proper names). We have an understanding of what is meant by the noun and that understanding can change over time as we investigate the objects in question. The problem with thinking about reference simply as words attached like labels to objects is that it assumes that we have already demarcated the objects. We know what the object is that has the label stuck on it. But ‘to know what the object is’ is to assume we can describe it, and description is more than mere label. Descriptions are already theorized. Whilst the word ‘dog’ does not seem to be a very theoretical term, we do have understandings of what objects are dogs and what are not, and those understandings are based upon the web of other beliefs we have about the universe of non-dogs. Some things are dogs and some things are not, and some non-dogs are closer to dogs than other non-dogs. All animals are closer in some sense to ‘dogs’ than plants, and some animals are more dog-like than other animals. So the word ‘dog’ is to some extent a theoretical term, even if it is not as theoretical as the expression ‘discourse’. Indeed, different people have different ‘official’ ways of categorizing dogs depending on whether they are vets, dog-show organizers, hunters or whatever (Hofstadter and Sander 2013: 241), and we have related categories of foxes, coyotes, wolves, dingoes, and so on. The ‘theory’ underlying the specific categorization will affect the categories demarcated. (Though theorized descriptions do not fully determine the extension of a term: we can mistakenly think that a wolf is a dog without having defective descriptions.)

38 The Philosophy and Methods of Political Science Few terms in political science are as easily defined as ‘dog’. We might think terms such as ‘prime minister’, ‘president’ or ‘chief executive’ are straightforward, but soon realize that whilst we have a general idea of what each of these means, defining them (with necessary and sufficient conditions) so they cover all cases is problematic, since political systems vary across countries and change within countries over time. Other examples in political science are even more problematic. Is the European Union a federation? What are the necessary and sufficient conditions for something to be federal? In what sense is the British Labour Party the same party now as it was in 1945? How do we analyse identity of political objects through time? We can define terms within formal models more easily, but with regard to ministers or chief executives, such formal definitions do not always translate directly into people in actual systems – more obviously, theoretical concepts also often do not translate directly into actual entities in our world (see Chapter 8). It is for these kinds of reasons that philosophers of language moved away from constructing theories of meaning in terms of words and towards sentences or propositions – though Kripke from the 1970s started to reverse this trend (see, for example, Morris 2007; or Soames 2003a, 2003b). The issue is whether we try to construct a theory of language from the meaning of the words contained within it or develop theories of meaning for a language that give meaning to the words contained within it. The first line of thought leads us to distinguish sense and reference or connotation and denotation, the second, whilst not relinquishing that distinction, leads people to argue that reference (the thing in the world to which the proposition is applied) and sense (the meaning of the proposition) are so closely combined that one must be explained in terms of the other. To go beyond words into sentences or quasi-sentences (meaningful chunks of language) should not commit one to the view that we cannot understand any word unless we understand the whole of language (though some philosophers of language seem to think so). It is true that with this more holistic picture of meaning discovering new implications about certain concepts might affect our understandings of other concepts and theories. A shift in meaning in one part of our language might entail shifts in meanings in other parts, reverberating throughout the language and affecting our understanding in fields far distant from that of the original change. We might, though, expect these reverberations to fade with distance. In a famous article (and many later works) Quine (1953, 1960) challenges the analytic/synthetic distinction, suggesting that there are no terms (words or sentences) whose meaning depends solely on the meaning of other terms. Empirical evidence often changes the meaning of our sentences. We can accept that as theoretically possible, without really giving up the idea that some phrases are analytic, if only because some terms are so close in their reference that we cannot imagine any empirical evidence that would directly change the meaning of an expression without also directly changing the meaning of its synonym. In other words, some heavy-duty theory would have

What Is an Explanation?

39

to come along to enable us to split two synonyms by the empirical evidence. Without such theory, we cannot imagine what it might be. Philosophers can worry about such theoretical possibilities; empirical researchers can leave the worries to the philosophers (Strawson and Grice 1956 suggested as much). Many modern philosophers of language have in fact abandoned worrying about such theoretical possibilities, leaving Quine behind. Kripke’s (1972, 1980) work on reference has had an enormous influence on the philosophy of language in recent years, returning to the idea that words are labels that fix reference with descriptions (usually ostensive) that are not necessary and sufficient conditions for determining its extension generally. Words that refer to fixed properties or objects across our description of possibilities are called ‘rigid designators’ (Kripke 1980). Of course, we can fix anything as a point of comparison across possible worlds, but those that are fixed across all possibilities are ones that are necessary. So what is logically necessary is rigidly designated. Metaphysically necessary truths are described by rigid designators. But are there any characteristics of objects that are necessary to them? The idea of analyticity is that it captures what is necessary in a definition. In the oft-used example, ‘bachelors are unmarried males’, the property of being an unmarried male is, if the sentence is analytic, a necessary property of being a bachelor. Now Quine, as we have seen, attacks the analytic– synthetic distinction on the grounds that it is possible our understandings of any term might change. One response might be: well, it is true in theory that understandings might change, but if a relationship is really necessary then in fact they will not. Unfortunately, such a simple response is modally problematic, for analyticity is a relationship between terms, and not a relationship between things. Kripke suggests a historical (or what is usually referred to as a causal) account of reference. In this, the proper name is the label first stuck on to some object that is then historically passed down. For proper names, like Adolf Hitler, the name is attached to the person who was baptised thusly, who then served on the Western Front, went to jail, wrote Mein Kampf, became Chancellor, led his country to war and set in motion the Holocaust. None of these is a necessary feature of Hitler. He might have been killed on the Western Front; and to the extent that the rest of the history would have occurred, it would have occurred as it did through the actions of others. What rigidly designates the man who did do all those things is his baptism. If the term ‘bachelor’ is to take on new meanings, then something else has to change, perhaps our understanding of gender or marriage. The term ‘bachelor’ can take on a new meaning and we can see how that changes historically as new understandings of terms within its original definition change. We might say the same for terms such as cabinet minister, which might originally have a specific legal definition, but takes on new meanings as it is applied more widely to countries with different legal systems. The term ‘federation’

40 The Philosophy and Methods of Political Science as applied to political entities has a specific first understanding, but that changes as systems of government evolve over time, so that we consider something like the European Union a federation even though it does not conform to the original definition. In this sense there are no qualities that rigidly designate the term. Other concepts might have rigid designators. Water has the chemical structure H2O and that is a necessary feature of water. It rigidly designates water, because the object that was given the name ‘water’ has that structure (even though that was not known at the time it was so named). Water has to have that structure, otherwise it would not have the features it does in the natural world. Gold has the atomic structure of 79. Again it has to have that number or it would not have the features it does in the real world. So H2O rigidly designates water and atomic number 79 rigidly designates gold across possible worlds. For Kripke, this makes ‘water is H2O’ a necessary a posteriori truth. Necessary, because it has to be the case in all metaphysical worlds, but a posteriori because we have to know something about the world in order to know it is true. (See below and Box 3.1.)

Box 3.1

Metaphysical and epistemic relationships

Metaphysical

Epistemic

Candidate examples

Necessary

Contingent

Water is H2O

Contingent

Necessary

‘I am here now’

Necessary

Necessary

1+1= 2

Contingent

Contingent

Australia’s GDP in 2013 was $1560bn

A priori

=

epistemic necessity

A posteriori

=

epistemic contingency

The Kripkean revolution was to prise apart the (metaphysically) necessary– contingent distinction from the a priori–a posteriori distinction, and in doing so further clarify distinctions glossed over by the analytic–synthetic distinction. Using these distinctions helps us to: 1 Discover that identities (which is a descriptive inference) are explanatory. After all, identity statements are not causal. Thus description is a legitimate and important form of social scientific explanation: indeed, just as important as making causal inferences. 2 Understand the relationship of laws to mechanisms, and both to correlations of various strengths. 3 Understand the nature of reality. 4 Fix the relationship of conceptual analysis to empirical enquiry – the two are not separate. 5 Appreciate why the historical understanding of terms is important to contemporary use, even as the meaning of terms and phrases changes over time.

What Is an Explanation?

41

Could something precisely like water, that was not H2O, exist in some logically possible world? Perhaps: but only if many other features of the universe changed too. Since Hume (1742/1978: 32), philosophers have argued that if something is conceivable it is metaphysically possible. If something is conceivable it is possible. However, philosophers tend to give themselves an easy time of it. Hume wrote that we can imagine a mountain made of gold, but not a mountain without a valley (he meant a valley without surrounding hills). However, scientifically we want our metaphysics to underlie our physics (Ladyman and Ross, 2007). If we imagine a substance like water but slightly different, chemically speaking, then we have to ask what the chemistry would look like. Is it composed of elements in our universe? If not, what would the existence of those elements mean for our elements? Scientific conceivability needs a lot more work. When it comes to mountains made of gold, could this happen naturally? What conditions would have to obtain and what would have to change on our planet for those conditions to obtain? A mountain on earth made of gold would not, given gold’s malleability, be very stable, so conceiving it would require imagining it in a specific form or changing other elemental features of our universe. Scientific conceivability is not the easy game philosophers normally indulge in. Scientific conceivability is particularly important for the human sciences. It is easy to conceive of social utopia, but difficult to precisely specify what conditions need to obtain for them to exist. Are there unintended consequences that naturally follow from those conditions, and how robust are they against noncompliant behaviour? In the natural sciences we have generally accepted strong laws (what are described below as invariant generalizations). In the social sciences, such invariant generalizations are lacking, and those generalizations we have established are not so widely accepted by all social scientists. Hence it is far more difficult to place limits on conceivability. Where those limits are placed is a major topic in political philosophy (see Chapter 9). The importance of rigid designation in the historical account of reference is that the name of a thing is that which it is given by some initial attempt at demarcating the reference – some reference-fixing event – that has subsequently been passed down. Whatever that thing is that was so named will have certain features. Some of those features will be contingent, some might be necessary; and it is the necessary ones that qualify it as an object that could be rigidly designated. It might be the case that a thing historically named will have no features that rigidly designate it: all its features are contingent. It might be the case that none of its features is strictly necessary. They are neither logically nor naturally necessary. However, in order to conceive of them differently, we still have to change so much else in the actual world that we can treat those features as though they were rigid designators. One way in which we might think about this problem is in terms of invariance. A strict natural law is invariant. It holds at all times and places in the

42 The Philosophy and Methods of Political Science universe. Such laws are usually identity relationships. Gravity is the relationship described by its conditions. Some invariant laws, however, are considered to be causal rather than designators of identity relationships. Many relationships are not invariant, but do not vary much. The problem for the social sciences is that many of the (law-like) relationships it discovers are not strictly invariant. In that sense they are not necessary, and hence the relationship does not rigidly designate a feature of the items on either side of the equation. For example, ‘democracies do not go to war with each other’ is not an invariant relationship on our planet. But it is a strong relationship (at least for political science) (Russett 1993; Kim and Rousseau 2005). For that reason, ‘not going to war with another political system that shares its essential features’ cannot be a necessary defining condition of the term ‘democracy’. That feature cannot be a rigid designator of ‘democracy’. This does not imply that there are no necessary conditions of democracy, just that this is not one of them. However, there might be no such rigid designators. Under the historical theory of reference, ‘democracy’ might be defined as ‘rule by the people’ by the meaning of ‘demos’ from its first application to political systems in Ancient Greece. However, ‘rule’, ‘by’ and ‘the people’ might be considered too imprecise to be considered as rigid designators to give us an easy application to subsequent political systems to enable us to determine some necessary characteristics for democracies in the manner that H2O and atomic number 79 give us for water and gold. I return to these examples in Sections 3.6, 4.5 and Chapter 6 where I consider laws, generalizations and mechanisms. And again in Chapters 8 and 9 on conceptual analysis, where I will return to Kripke’s account and explain a little more about how the historical theory of reference should affect how we go about conceptual analysis. What I want to take from our sojourn in the philosophy of language here is that the account we take will have an important effect on our attitudes to methodological issues. What Kripke did was to distinguish metaphysical and epistemic necessity, breaking apart the identity of the analytic–synthetic, a priori–a posteriori and necessary–contingent distinctions. (See Box 3.1.) We can make use of Kripke’s argument to demonstrate that discovering identities is explanatory. It also shows us that causal explanation is not the only form of important explanation there is. (I shall return to these points several times in this and later chapters.) It can help us to understand the relationship of laws and mechanisms (as I argue in this chapter and in Chapters 4 and 6) and to comprehend the nature of reality and the relationship of conceptual analysis to that reality (Wendt 1999 utilizes this argument in this manner; my use of it is very different from his). Finally, it can help us appreciate why the historical understanding of terms is important to contemporary use, something I take up in Chapters 8 and 9 (especially). Thinking about the relationship of language and the universe like this gives credence both to ourselves as perceptual beings and to the world around us. It implies that the world around us places constraints upon how we can view

What Is an Explanation?

43

it. Kripke’s argument suggests meaning is inevitably determined by the actual world we inhabit – meaning can never be purely abstract, hence the necessity of some empirical grounding in argument. The nature of those constraints deserves further enquiry, but at its most basic it means we can have a language that helps capture the world. The world can be thought of as a set of events, some of which continuously persist and so are called objects; these persisting events become objects because of our construction. After we have labelled a set of objects by a term, we can further analyse that set and discover new aspects, following Quine’s (1953: 22) dictum that ‘meaning is what essence becomes when it is divorced from the object of reference and wedded to the word’. What we are doing is finding patterns in the world. Those that help us predict are the real patterns (Dennett 1998). Not everyone will find the same patterns, especially as we look to more theorized explanations. But though these patterns are different, they are not contrary or contradictory to our own. Two social scientists might describe a set of human events in very different ways. We must not jump to the conclusion, frequently arrived at (as I shall detail in later chapters, especially 4 and 5), that their descriptions are contradictory. They might even explain a set of events differently, but again we must not hastily assume that two different explanations of the same set of events are contradictory. Acemoglu and Robinson, for example, claim that their argument over the rise and decline of nations is rival to that of Jared Diamond. I am not sure it is, since both can work together and the explanations are directed at slightly different objects (Acemoglu and Robinson 2012; Diamond 1998). One explains the success of civilizations broadly in terms of geography, the other in terms of institutions; but the institutional factors and the geographical factors are not inconsistent with each other. And the objects, civilizations and nations, are not quite the same. Diamond’s is a broader brushstroke, Acemoglu and Robinson’s more detailed. We find patterns in the universe so we can predict what is going to happen, minimizing surprise. Without predictability we would constantly be surprised; we would never know what was happening around us. Predictability is the basis of our understanding of the world. Constructing patterns in the universe is, in an important sense, an exercise in prediction. It has been said that prediction and explanation are two sides of the same coin. I am not sure I understand that analogy, but I am sure that explanation requires prediction in the sense that any claim that ‘X’ explains ‘Y’ means that under the relevant circumstances X will always explain Y. That is, we can predict that if X satisfactorily explains Y, then, when conditions are identical, X will always satisfactorily explain Y. In an important sense, then, explanation is the reduction of surprise. (We should note immediately that ‘X satisfactorily explains Y’ might be probabilistic: X will satisfactorily explain Y when X and Y are present, but X being present does not necessarily mean Y is, even when conditions are identical. Rather Y will be present when X is, when conditions are identical with some probability p.)

44 The Philosophy and Methods of Political Science

Box 3.2 What is prediction? Prediction is the claim that if you think X explains Y under conditions w, then you are predicting Y given X under conditions w. And this holds even where, say, the relationship of X and Y given w is probabilistic. It does not follow that if you predict Y given X under conditions w that you have explained Y by X. Prediction is not the same as forecasting. The predictive conditions might never obtain in the future. Nevertheless, forecasting or predicting future events provides the underlying motivation for seeking explanations.

So prediction and explanation are inextricably bound, a claim of ‘positivism’ often derided. However, note the decidedly non-positivist twist. Prediction is a normative constraint upon explanation. Or, in other words, explanations that are non-predictive are not much use to anyone. If you don’t believe that, just think of someone claiming that X explains Y, but also proclaiming, ‘not that I think X will always explain Y; under identical conditions not-X could explain Y equally well’. Since conditions are always identical to

Box 3.3

Explanation and prediction

Prediction is defined in the main text. I also place it in a box since it is so important for my argument and is often misunderstood. Explanation requires prediction in the sense that any claim that ‘X’ explains ‘Y’ means that under the relevant circumstances X will always explain Y. That is, we can predict that if X satisfactorily explains Y, then, when conditions are identical, X will always satisfactorily explain Y. In an important sense, explanation is the reduction of surprise. Where ‘X satisfactorily explaining Y’ is a probabilistic relationship, then X will satisfactorily explain Y when X and Y are present; but X being present does not necessarily mean Y is, even when conditions are identical. Rather Y will be present when X is, when conditions are identical with some probability p. Prediction does not mean prophecy. After all, a probabilistic relationship means only a prophecy with some probability, and if the probability is low then it is not much of a prophecy. (Though one with very high probability is not much of a prophecy either.) Successful prophecies are not always part of an explanation, however; though predictive, they have a more complex relationship with explanation. So in this sense prediction is necessary for explanation but not sufficient, and that is one reason why it is hard to grasp what explanation is. What are the other conditions? In part they are psychological. Though predictions, as I use the term, and prophecies are different, the normative force of prediction within explanation is its utility in prophecy, our ability to see what is happening and going to happen in the world. Thus the necessarily predictive element in any explanation is a normative element of explanation.

What Is an Explanation?

45

themselves, the person would be claiming that both X and not-X explains Y in the case at hand. Such claims are sometimes made. It is not unknown for identically psychotic behaviour to be explained by completely opposite relationships with the subject’s mother – which just goes to show that mothers can do nothing right for psychotic children.

3.3 The attempt to produce a model of explanation From the 1930s to the 1960s or so, philosophers of science attempted to produce a formal model of what an explanation looks like: the highlights of such attempts were the Deductive-Nomonological (DN) and InductiveStatistical (IS) models of Carl Hempel and Paul Oppenheim (Hempel 1965). Both IS and DN models have the same structure. Their premises each contain statements of two types: (1) initial conditions C, and (2) law-like generalizations L. The only difference between the two is that the laws in a DN explanation are universal generalizations, whereas those in IS explanations are statistical generalizations. These models of explanation ultimately failed, due to numerous counterexamples that either looked like explanations, but did not fit either DN or IS form, or fitted the forms, but did not look like explanations. Others have written at length about the models’ failure; here I want to highlight the desiderata that Hempel felt an explanation should produce. Achinstein (1983) suggests there were two strong motivations behind the DN and IS models. First, to ensure that an explanation is something that cannot be known analytically: that is, relying solely upon the meanings of the words. I will return to that issue below, since at the end of the day we surely want our theories to be analytic. Second, that it should have a ‘No-Entailment-by-Singular-Sentence’ (NES) requirement. What is good about the NES requirement? Achinstein adduces three answers: 1 It precludes self-explanations, such as those where the sentence of what is to be explained is part of the explanatory sentence. So the explanation of ‘this metal expanded’ cannot be ‘this metal was heated and expanded’; the latter entails the former. 2 Second, the DN proponents want to emphasize general laws, since science seems to be the development of such general laws. Laws provide the link between singular explanatory sentences and the singular sentence of what is to be explained. 3 The NES requirement ensures that explanatory connectives such as ‘explains’ or ‘because’ or ‘due’ are removed from the model of explanation. The aim is to explain explanation and including such terms as primitives in the model would be circular.

46 The Philosophy and Methods of Political Science

Box 3.4

DN models of explanation

Syllogism All men are mortal

Major premise

Socrates is a man

Minor premise

Therefore Socrates is mortal

Conclusion

An example of a DN covering-law model (taken from Achinstein) is: DN covering-law model Any metal that is heated expands

Covering law

This metal was heated

Singular statement

Therefore This metal expanded

Inference or prediction

Note: I have inverted the normal way of writing the DN covering law to make the analogy to a syllogism more obvious.

The DN model, as we see in Box 3.4, has a similar structure to a simple syllogism with the covering law playing the role of the major premise and the singular statement playing the role of the minor. What matters, however, is that the DN model looks as though it provides an explanation, since it resembles a regular deduction. The deduction is based upon an empirical law which itself then allows the inferred conclusion. (For more on this, see Clarke and Primo (2012).) In the same way as the syllogism explains why Socrates is mortal (because he is a man and all men are mortal), the covering law explains the expansion of metal (because it was heated and metal expands on heating). Of course, we can always ask ‘why does metal expand when it is heated?’, as we can ask why humans always eventually die. For the metal example, the answer would involve a discussion of the nature of heat, the sub-atomic structure of metals, the effect of heat on particles. (As we see later, this move is associated with finding mechanisms, though, as I also suggest, laws underlie mechanisms.) Such further questions immediately become more theoretical; indeed the term ‘heat’ is not without problems in science, where some usages imply entropy and others do not. This need not detain us, since all that matters is that these further explanations, according to Hempel, would also take the DN form. However, there are numerous problems with the DN model as an account of explanation.

What Is an Explanation?

Box 3.5

47

Failures of DN explanation

Two well-known examples of purported explanations that fit the DN model, but fail to be explanatory: Barometers and storms The barometer falling leads us to predict that a storm is coming. We assume the storm explains the falling barometer. But logically, we can predict a barometer is falling because a storm is coming, whereas we do not think the barometer falling causes the storm to come. However, the logical form of the inferences is identical. Pendulums and their lengths We can explain the period of a pendulum in terms of the length of the pendulum together with the law of simple periodic motion. But one can just as easily explain the length of a pendulum in terms of its period in accord with the same law. Our intuitions tell us that the first is explanatory, but the second is not.

The DN model fails because there are examples of DN covering laws that do not seem to explain anything; there are examples that seem to explain but do not fit the covering-law model; and in the social sciences it is difficult to find any covering laws. An example of the first is in Box 3.5 and of the second in Box 3.6. Furthermore, the final desideratum is problematic because explanations always explain for someone. A single sentence may be sufficient explanation for one person who has the relevant background information, but inadequate for another. I examine this issue in Section 3.6 below; here I want to briefly consider the NES requirement.

Box 3.6

DN and reasons for action

We often explain the actions of people by giving their reasons for action, but the DN covering law does not seem to work for that. We might say that John voted for the Conservative Party because he thinks that it would make the best government for the country. Whether or not that is a good explanation, it does seem to be one. However, it is difficult to fit such reasons-for-action explanations into a DN model in an illuminating manner. How does this overcome the NES problem? DN covering-law model for voting People vote for the party they think best John thinks the Conservative Party is best

Covering law Singular statement

Therefore John votes Conservative Party

Inference or prediction

Note: Again I have inverted the normal way of writing the DN model.

48 The Philosophy and Methods of Political Science The thought underlying the NES requirement, I believe, is that we need to fit our explanations into a more general structure. An entailment does not look like an explanation, since the explanatory sentence is too close to the one we are trying to explain. However, a sentence S that merely purports to explain some Y cannot just do so arbitrarily. This is the predictive element I keep stressing. The fact that S always explains Y under the relevant circumstances must be maintained (even if the relationship between the process referred to in S and the Y to be explained is probabilistic), and that fact requires some reference to other sentences. A universal law will do, but where universal laws are not in evidence we require something else. Most of what I discuss below is that ‘something else’. Since the DN model of explanation requires deterministic laws to generate the deduction, but many generalizations, especially in the social sciences, do not seem that deterministic, the IS model was generated. This tries to follow the same deductive pattern, but turns out to be even more hopeless at providing a satisfactory model of explanation (see Box 3.7). One aspect of the generalizations in these examples not discussed enough is the level at which we generalize. I have followed a standard example in using ‘metal expands when heated’. However, this is not a baseline scientific generalization, but rather one derived from baseline generalizations. The surface phenomenon that we are generally interested in might not, in social science, be subject to generalizations. Perhaps the major problem with these examples of generalizations is that they concern surface phenomena and not deeper structures. They are empirical generalizations that in the natural sciences might strictly hold because they follow directly from underlying laws. In the social sciences any such empirical generalizations will not follow directly from underlying laws. Rather they are predicted by the mechanisms that provide explanations of the surface empirical generalizations. Underlying these mechanisms might be some laws. I shall return to this topic in Chapter 6 (and see also Section 3.6 below), where I discuss generalizations in terms of invariance, and connect these up to mechanisms. An important aspect of Achinstein’s account here is that explanation is something that transfers meaning from one person to another. In other words, contrary to the positivist account, explanation here is not a purely logical concept. Is it, then, a causal concept? In the sense that S always explains Y to an individual with the right background, then S is a causal notion. What it causes is a psychological phenomenon that we call an explanation. It is the meaning of the sentence S, its propositional content, together with other propositional content accepted by the individual, that leads to S being explanatory. If we want to find the logical form of explanations,

What Is an Explanation?

Box 3.7

49

Inductive-Statistical explanation

Hempel recognized that not all explanation can be of the DN form – some is inductive and statistical – but he tried to force explanation into a similar model. An example of an IS explanation is: The Republicans outspent the Democrats by almost 2:1. The political party that outspends the other by large amounts virtually always wins elections. The Republicans won the election. The statistical relationship allows us to infer that the Republicans won the election because they spent more. However, Hempel was aware that such claims are problematic. He thought DN explanations were always better than IS explanations. First, because the deductive relationship between premises and conclusion maximizes the predictive value of the explanation, hence he thought IS arguments explanatory only to the extent that they approximated DN explanations: only IS explanations with high probability over the event to be explained were explanatory. Secondly, Hempel thought explanation is something that should be understood fundamentally in terms of its logical form. For a DN explanation to work (to be an explanation), its premises have to be true; in the same way that a valid syllogism has a true conclusion only if its premises are true. However, IS explanations do not have the correct logical form, since the statistical relationship can be altered by new information (the probability is conditional on the information we have). So, for example, The Republicans have been in power for 20 years. The party that has been in power for over 16 years almost always loses. The Republicans lost the election. This is also an IS, and the probabilities associated with the two rival IS explanations might combine to give a new IS. That might have a prediction at odds with (though with lower probability) than the first IS. This means it is always possible that any supposed IS explanation, even if the premises are true, would fail to predict the effect and so have no explanatory significance. Furthermore, it is simply not true that only statistical inferences of high probability count as explanatory. We know drinking even moderate amounts of alcohol increases the probability of crashing when driving a car, but it is not the case that one is more likely to crash than not to crash if one has had a few drinks; though we might explain George’s crash as the result of his reduced reaction time to the kangaroo leaping in front of his car due to his consumption of two glasses of wine. The problem with a probabilistic law is that it is only a statistical correlation rather than a description of what is to be explained. Following von Wright (1971), we might suggest it is better not to say that the IS explains, only that it justifies certain expectations.

50 The Philosophy and Methods of Political Science we have to recognize the entire set of propositions together with the propensity of the individual to put them together. This takes us into the nature of belief formation and the assessment of evidence. I shall discuss this in more detail in Chapter 5.

3.4 Proximate and ultimate; type and token Before examining explanations in political science, I need to draw attention to an important distinction used in evolutionary explanation: that between proximate and ultimate explanation (sometimes expressed as proximate and ultimate causation) (Mayr 1961; Ariew 2003). It is a distinction that should be used more in the social sciences, particularly as it can illuminate the different explanatory claims made in quantitative and qualitative methods. First, though, I will introduce another, related, distinction: between type and token. Again, in the social sciences we often give explanations of types, and sometimes of tokens. Or we use type explanation as part of a token explanation. A clearer recognition of this distinction will also illuminate the different explanatory claims that are made in different contexts.

Type and token A type is a given class that is made up of many token examples. A token is an example of a given class of items. Julia Gillard and David Cameron are both token examples of the class of prime ministers; George Bush and Barack Obama are both token examples of the class of US presidents (and of course the broader class of presidents). If we are examining the power of prime ministers versus those of presidents, we will couch our explanation in terms of type. We will compare and contrast the structural and institutional constraints upon presidents and upon prime ministers. We might illustrate our argument with some token examples, but the explanation would be couched at the level of type, because it is the differences in type in which we are interested. On the other hand, we might try to explain why Barack Obama had to compromise so greatly in attempting reform of the US health care system. Here we are trying to explain the constraints upon a token president. We would examine specific historical events, perhaps noting particular contingent factors. Types would still enter into the explanation of the token event ‘Obama’s health care reform’: we would note constraints upon Obama that apply to all US presidents. Nevertheless, the object of the explanation is a token rather than a type.

What Is an Explanation?

51

Another example would be a comparison of the token events of Obama’s health care reform and Cameron’s Conservative–Liberal coalition government’s attempt to reform the British National Health Service. Again, whilst the objects of explanation, the problems faced by a British coalition prime minister and a US president in a specific reform of health care, are token examples, our explanation is likely to be couched broadly in type terms. This is because the explanation of their rather different difficulties would involve contrasting the institutional and structural conditions the two leaders faced in their respective countries. We might note, in the case study comparison, certain contingent events that caused problems or enabled reform; and we might note aspects of the specific personalities involved that proved important. For example, it might be that mistakes made by Andrew Lansley, at one time Cameron’s minister of health, might be important as part of our explanation of Cameron’s troubles, and we might tie those mistakes to his (token) personal characteristics rather than to his type ‘minister of health’. Whilst we would, of course, examine the token processes that occurred in our case studies, the bulk, the oomph so to speak, of the explanation would be type explanation. With type explanations we are often explaining in terms of a set of subjunctives. For example, groups of people are frequently powerless because they face some sort of collective action problem. There are a host of ways to overcome collective action problems. Some people might do so thanks to the intervention of some outside agency; others through the actions of some subgroup within the larger group; others by using institutional procedures originally designed for some other purpose. We might be able to give some structural or institutional reasons why some solutions are achieved under some circumstances and others in different circumstances. However, for a general class of collective action problems we might simply have a set of subjunctives. The type explanation remains general. When we study a particular group, we might give historically contingent reasons why this group overcame its collective action problem as it did. The token explanation bolsters or ‘fills in’ the type explanation. Case studies are tokens. Each variable that is entered into a spreadsheet is also a token, but the inferences derived from their manipulation are types. As such, those type explanations can only be read back into any particular case from which a variable is drawn to the extent that the variable enters into the type explanation. We tend to think that the probabilities (or variance) that we associate with such type explanations apply to any given token case, but only as conditioned on the total number of cases. Conditioned on the particular token case from which a given variable is drawn, the probability (and variance) might be dramatically different because of the particular conditions of that token case. The probability and variance only apply to the type (as estimated from the sample of token cases) and not to each token.

52 The Philosophy and Methods of Political Science

Box 3.8 Type and token A type is a class made of many tokens. A token is an example of a given class. Barack Obama is a token example of the class of US presidents, and of the class ‘Democrat US presidents’. Explanations of token events will involve other token events and types. Explanations of types may be given entirely in terms of types, though token examples will be used as data or illustrations. Most explanation in the social sciences is of types. Most explanation in history is of tokens. The type–token distinction might be used to distinguish history from social science.

If we keep in mind the difference between token and type explanations, we might reduce tension between claims made by those who use case studies and those who do large-n quantitative analysis. I would suggest that most explanation in political science is type explanation. Most explanation in history is token explanation. This is the key difference between history and political science (not, as was once famously discussed in Oxford University’s politics faculty, the date at which history ends and politics starts). I also believe that is why we should keep theory to a minimum in historical accounts. We cannot do so entirely (as my quote from Gibbon on p. 135) demonstrates (but see what I say about ‘theory’ in Chapter 4). I shall return to the type–token distinction several times over the course of this book.

Proximate and ultimate The distinction between proximate and ultimate explanation is strongly related to the type–token difference. This distinction is well known in biology, introduced by Ernst Mayr in 1961; though I do not claim my use is identical to that of Mayr, and should point out that there is some dispute about its usage in biology. André Ariew (2003: 557–8) suggests that ‘what distinguishes evolutionary explanations from “proximate” or individuallevel causal explanations is that the former type of explanation is a statistical population-level explanation’. One way of thinking about it is through a fauna example. We can explain why Marty the zebra is eaten by Alex the lion by the manner in which the hunting lionesses caught Marty. The lionesses watched the zebra herd; when they started the chase Marty, caught on the edge, stumbled; then unluckily, when he zigzagged his run he cut left just as a lioness also cut left. Had he cut right, he might just have escaped, for the lioness was at the very edge of her stamina. That is the proximate (token) causal history of why chunks of Marty ended up in Alex’s stomach.

What Is an Explanation?

53

We can contrast this with an ultimate explanation of why some zebras are caught by lions whilst others are not. It will be couched in statistical terms. We might note, for example, that zebras at the edge of the herd at the time the hunters pounce are more likely to be caught; that zebras which zigzag are more likely to escape than those that run straighter; that certain individual characteristics, such as age (probably a non-linear relationship, with younger and older zebras more likely to be caught as they are slightly slower), size, sex, and so on, might also contribute. An unlucky stumble might also feature, as it did in our story of Marty – although it might not, because it is too unusual an occurrence to enter in the statistical explanation as anything more than noise. (That is, such events ought to feature in type explanations, but might not due to data restrictions.) Now we can note that the ultimate explanation makes use of a set of types of zebra – type in terms of age, strength, propensity to zigzag when running, likelihood of being at the edge of the herd. But again, note, these ‘types’ might not be made up of individuated zebras such as Marty, though other types might be composed of biologically separate individuals. Marty began life as a young zebra and died, shall we say, as an old one. Thus he went through several types that might be important for ultimate explanation. (Zebras do not change sex, so that feature of Marty will remain constant.) Other features might be statistically related to biological individuals. For example, there might be some ‘propensity to zigzag’, causing Marty to zigzag more than others; or it might simply be something that all zebras sometimes do and sometimes do not. The former might have a genetic component, or perhaps environmental ones: young zebras zigzag according to the pattern of their mothers, to whom they stick closely when growing up. The propensity to stumble might be a feature of some zebras (clumsy ones) or the probability of stumbling might be equal for all zebras. We can see, therefore, that whilst type explanation is associated with ultimate explanation, proximate explanation is not identical to token explanation. We tell the story of why Marty was caught; he is a token example of many aspects of the type characteristics that give the statistical propensity to be caught, but those characteristics might not feature in the proximate explanation of his being caught in quite the same relationship as they do in ultimate explanation. The type explanation might explain the likelihood of why Marty and not Mary was caught, but the stumble might explain why it was Marty and not his friend Norman who was actually caught that day. In some of my work I examine why ministers resign (Berlinski et al. 2012) and have studied the proximate explanations of resignations (Dowding and Kang 1998; Dowding and Lewis 2012; Dowding et al. 2012). For proximate explanation of specific types of resignation, notably those that occur after a call to resign, we categorized non-resignations (where a minister does not resign following a call to do so) and resignations (where a minister does go

54 The Philosophy and Methods of Political Science

Box 3.9

Proximate and ultimate explanation

Proximate means coming immediately before (or immediately after) some element in a relation (such as a chain of causation or reasoning). Ultimate means complete or beyond all others. Proximate explanation is the immediate explanation of some incident or event (dependent variable), usually specifying particular events (independent variables) that occurred prior to the occurrence of the dependent variable. The independent variables satisfy as an explanation, since we can understand how they led to the incidence of the dependent variable. The narration makes sense. Ultimate explanation provides those conditions (independent variables) that are associated with events of the kind we are explaining (dependent variables). Once an ultimate explanation is fully specified, there is nothing more to say to explain either the full set of dependent variables or any particular example of them. Any further explanation of the dependent variable in a proximate explanation could be added into the set of independent variables for the ultimate explanation. Approximate means ‘very near’ or ‘nearly resembling’. It obviously has the same root as proximate, but is contrasted with ‘precise’ or ‘exact’. A proximate explanation can be fully specified and so is not ‘approximate’, though what it cannot constrain is the random or chance element that is provided in ultimate explanations. Given all the independent variables in the proximate explanation, what is the chance that the dependent event would occur? We cannot know from the proximate evidence itself. We can only constrain the chance element within the ultimate statistical explanation. Note that whilst ‘ultimate explanation’ does imply nothing more can be said, any example of ultimate explanation is likely to be underspecified: we can always add to or say more about extant ultimate explanations.

within a few days of the call). We find that ministers never (or very rarely, with some generous coding) resign over problems or mistakes made by their departments; they are far more likely to resign when they have been personally involved in some scandal. In other words, we can categorize certain types of event as being more likely to lead to a resignation than others. Some types of events are more likely to be proximate causes of resignation than others. In further, more quantitative, work we have uncovered other factors involved in the probability of ministers resigning, including a call for resignation; previous calls for resignation; age (a non-linear relationship); sex; experience; popularity of the government; and the cumulative number of calls for resignation for ministers in the cabinet overall (Berlinski et al. 2007, 2010 and 2012). These are ultimate causes of ministerial resignation. Now, if one examines the reasons prime ministers give in interviews or their memoirs for why they sacked (or accepted the resignation of) certain ministers, they rarely refer to some items that feature in the ultimate explanation; those they do mention tend to appear in code. So the prime minister

What Is an Explanation?

55

might say, ‘Unfortunately, while he was a good minister, he had been involved in too many controversies [too many previous calls for resignation] and had had a good run [age].’ If they mention the popularity of the government or cumulative calls, it will be even more circumspectly: ‘The government was facing difficulties and we needed to relaunch our programme.’ Some reasons that feature in the type explanation, such as the minister’s sex and experience, do not feature at all. And, of course, the qualitative evidence could not estimate how big a difference these factors should make in any individual case. There might also be proximate reasons in some cases that do not feature in the ultimate explanation at all. There might be evidence that a prime minister hung on to one minister because he was personally close to them, and was quick to jettison another because of personal animosity. Such features could be coded in a statistical model, of course, if one were confident of being able to code accurately. Sometimes, however, qualitative evidence in token explanations is more efficient than quantitative evidence. In other words, the quantitative evidence is too difficult or expensive to collect, or perhaps is so marginal in ultimate explanation overall that it would disappear amongst other noise.

3.5 Description, causation and understanding Some claim that we have a continuum with description at one end and explanation at the other (von Wright 1971). We do not ordinarily think of description as being explanatory. Indeed, students are often told that they need some theory in order to answer their research question, not just description. However, physical laws are descriptions of phenomena and they feature strongly in some models of explanation, as we see below. A story is a description of events, and we often believe we understand why something occurred because of the narrative we have been told. There is something that relates the laws and described events to the object of explanation. This is what we think of as causation (see more on causation in Chapter 6). So, for example, one physical law is that metals expand when heated (in fact virtually everything does). The explanation of why this token piece of metal expanded is because it was heated. It was the heating that caused the expansion. The sentence ‘it was heated’ stands as an explanation for why ‘this piece of metal expanded’ because of the law ‘metal expands when heated’. But the law is simply a description of what happens when metal is heated. It is type description. Metal is the sort of thing that expands when heated, but that type description stands as an explanation of a token example of it. In this sense descriptions are explanations. They are explanations since, if it is indeed a law that all metals expand when heated, we should expect (predict) ‘this piece of metal’ to expand when heated.

56 The Philosophy and Methods of Political Science We think of earlier parts of a story as an explanation of some of its later events because we assume a causal process. In J. K. Rowling’s Harry Potter series, Severus Snape’s ambiguous feelings about Harry derive from his hatred of Harry’s bullying father and his love for Harry’s mother, Lily. Snape worked against Voldemort, whom he once followed, becoming Dumbledore’s most trusted follower and a member of the Order of the Phoenix tasked with protecting Harry, because Voldemort murdered Lily. Earlier events (that we do not read about until near the end) explain later events (that we read about earlier). The narration is explanatory and satisfying because it illuminates actions by placing them in context. And we can empathize with Snape when we gain an understanding of his behaviour that earlier in our reading was missing. When we learn about Snape’s childhood relationship with Harry’s mother, we can understand his complex emotions. Descriptions stand as explanations because they stand in a causal or, when we discuss human action, in a rational or reasoned relationship to that being explained. There are many theories of causation. I will discuss different aspects (and in relation to reasons and interpretations) in Chapter 6 and more briefly below; but what we see as causal we view thus because it is a pattern in the data that we discern because it is predictive. Natural laws are patterns in the data described in type terms that stand as predictions for token examples of those patterns. It is because the token example conforms to the pattern we have previously discerned that the pattern (and type-level description) now stands as an explanation. Fully determinative laws explain because they fully predict. Reasons are predictive because they are also patterns in the data and thus can stand for explanations of behaviour. The reason that someone offers as explanation of their behaviour stands as such for me because I recognize it as a reason for that behaviour. I can only recognize it if I recognize that it could stand as a reason for me to behave in that way (I empathize with it) or see that it is a reason often offered for such behaviour (it fits the pattern I have discerned in others) even if I cannot imagine it standing as a reason for me to behave in that manner. In both ways, reasons are predictive in some sense as well as explanatory. A proffered reason for some behaviour that does not fit either cannot stand as a reason for that behaviour. When such a ‘reason’ is offered we usually create a background to make it reasonable. Most of psychoanalysis is precisely that process (Davidson 2004). Reasons are not, of course, as fully determinative as laws, but then not all patterns we perceive are fully determinative. Indeed most patterns we discern only predict probabilistically. Once we learn of Snape’s relationship with Lily, we understand his actions. Had we read about that relationship first (rather than merely knowing Snape hated Harry’s father), when it came to later events, such as Snape’s muttering when Harry is playing Quidditch, we would have been more likely to guess (predict) that he was muttering not a spell to knock Harry off his

What Is an Explanation?

57

broomstick, but a counterspell to save him. Earlier events would have given us grounds for assuming a different set of reasons for action, and hence a different set of actions. Are reasons causes? One can think so (Davidson 1980), but more generally reasons can be thought of as interpretations of behaviour. Old versions of interpretivism did not assume that reasons are causes; indeed, that is the distinction to be made between positivist social science and hermeneutical methods (von Wright 1971), though some modern interpretivists seem to assume interpretivist explanation is causal (Bevir and Rhodes 2003). This matters to the extent to which interpretivist explanation is rival to (other forms of) causal explanation. This is an important change in emphasis, since if interpretivism is not making causal claims, then it is not a rival explanation to quantitative evidence. If it is making causal claims then, potentially, it is a rival explanation. King et al. (1994) also assume that interpretive evidence is making causal claims, probably because they assume all explanation is causal. I think of cause as a property we impose on the universe in order to make sense of it. It is simply a pattern. It is not, of course, the only pattern we so impose. That is not to say that causation is not real – the patterning we see as causation is a real pattern – but what makes a real pattern is that it is predictive. Our view of causation is an important tool in our predictive capacity. The different ways in which causation is defined in the literature reveal different ways in which it is used in prediction; and I think of the best analysis of causation simply in terms of its predictive capacity. Some of those ways of thinking about causation bear a closer relationship to token and proximate explanation, others to type and ultimate explanation. However, I do not think we need try to fit all explanation into the straitjacket of causation. Functional explanation is not causal, but good functional explanation surely explains (Cohen 1976, 1982; Enc 1979; Sober 1984: ch. 2, 1993: 82–6); indeed, invoking functions in an explanation also invokes mechanisms. Different models of legislative committees emphasize their informational, distributive and partisan components (Shepsle and Weingast 1981; Weingast and Marshall 1988; Krehbiel et al. 1991). These different functions will have causal effects upon policy outcomes, but we can also provide explanations in terms of those functions of how committees operate, are sustained and affect policy stability that are not straightforwardly causal. And often in political science all we have by way of evidence are functional relationships and correlations. These, I will argue, are explanatory even if we cannot defend them as causal. Descriptions are also explanatory when they identify identity relationships that were not previously realized and which can then lead to new predictions or inferences. Completely invariant generalizations are descriptions and of course they play an important role in explanation.

58 The Philosophy and Methods of Political Science In the social sciences especially, but also in the philosophy of science, ‘mechanisms’ have become a popular exemplary for explanation opposed to generalizations. One reason is that there appear to be few strict laws governing human behaviour, and those that do seem to exist are best stated in terms of statistical propensities: they are highly variable. The problem for explanation – especially token and proximate explanation – is that statistical accounts are often psychologically unsatisfactory as explanations. This is more of a problem for humans than it is for explanation, but it is a problem nonetheless. A real mechanism is a feature of the world, a process by which an outcome occurs. We model mechanisms by connecting up certain structural features of society together with behavioural assumptions to produce empirical generalizations. A mechanism is a narration that makes sense of the data. That narration can be formalized into a model with strict predictions (see Chapter 5). The narration fills the gap between a propensity for something to occur and its actual occurrence. Hence specifying a mechanism satisfies our craving for explanation where statistical accounts leave us wanting more. When formalized into a model, the outcomes deductively follow from the premises. With less formal accounts the outcomes do not strictly follow, but they make sense. With a strict law (an invariant generalization) there is no gap in this sense and hence no felt need to fill it with a narration. (Though even with laws we often want to know why the relationship holds, so, for example, we explain the process (the mechanism) explaining empirical generalizations.) Below (Section 3.6) and again in Chapters 4 and 6 where I consider mechanisms at some length, I argue that invariant generalizations underpin mechanisms, but that we use mechanisms in explanations without considering those underpinnings. The mechanisms then help to explain the empirical (and more variable) generalizations that we see in our surface social phenomena. It is important to recognize that description can be explanatory and that we can make significant inferences about the world from it. What is a descriptive inference? King et al. (1994: 8) define it as ‘using observations about the world to learn about other unobserved facts’. In revealed-preference analysis, an agent choosing an apple over an orange means the agent weakly prefers apples to oranges; similarly, the same agent choosing an orange over a banana means he weakly prefers oranges to bananas. The principles of revealed-preference analysis allow us to infer that the agent weakly prefers apples to bananas, an inference from observed to unobserved behaviour. Moreover, the unobserved behaviour might never be observed if the agent never faces the last opportunity set. (Weak preference means that the agent either strictly prefers x to y or is indifferent between them. So the inference only allows us to suggest that the agent will not strictly prefer bananas to oranges.) King et al. (1994: 34) later add that another aspect of descriptive

What Is an Explanation?

59

inference is distinguishing what is systematic about the observed facts from what is non-systematic. Of course, that is precisely the inference we make when we use a deterministic law to explain a token event along the lines of the DN model of explanation. Often fundamental laws are not causal but descriptive. King et al.’s (1994) chapter on ‘Descriptive Inference’ is largely about applying the principles of logic, or what they call the methods of scientific inference, to descriptions about the world, just as they apply to causal claims. The chapter makes clear their commitment to the worth of good descriptive analysis – ‘Good description is better than bad explanation’ (45) – the utility of careful descriptions within case studies, and the fact that description is important to scientific endeavour. That said, they seem to contrast description and explanation (see also p. 75, n. 1, where they are more explicit); but to the extent that laws are descriptions, and the extent to which they explain token events, then descriptions often are explanations. In part, I think King et al. (1994) contrast description and explanation because they equate explanation with causation. But, as I have asserted, not all explanation is causal. Functional explanation and interpretative explanation need not be causal, and many fundamental laws are essentially descriptive. They can still stand as explanations. King et al. (1994: 43) also make the important point that all analysis requires simplification, pointing out ‘the difference between the amount of complexity in the world and that in the thickest of descriptions is still vastly larger than the difference between this thickest of descriptions and the most abstract quantitative or formal analysis’. What is important in any simplification is choosing the relevant aspects for the questions one is posing. The simplification process occurs with data collection and analysis, with description and with interpretation. In all cases we have to choose, either consciously or unconsciously, what to put in and what to leave out. We include those elements of the world we consider relevant and exclude others as not relevant. What we consider relevant is determined by the manner in which we pattern the data before us. One problem for token narrative accounts is that the decision process of what to leave out is less systematic than when data are collected according to a specific scheme. Hence the probability of bias is much greater in qualitative than quantitative analysis. I return to bias in Chapter 6 (under the title of the specification problem). Deductive inferences are explanatory simply because the rules of inference lead them to be so. But many inferences or arguments are not strictly deductive. What I have been calling a narrative explains if only because it suggests a causal process. Can we say more about narratives or arguments as explanations than this? Not much. There are many textbooks on what makes a good argument. Many of these operate by explaining fallacious reasoning. Formally, fallacious reasoning is simply that which breaks the rules of

60 The Philosophy and Methods of Political Science deduction. Informal fallacies fail other criteria. One standard text gives five conditions of good argument, but one of those conditions is the relevance of the premises to the conclusion (Damer 2005). Yet specifying normative criteria for relevance is problematic. Relevance is something that we recognize when we see it. Sperber and Wilson (1986) is one of the few works that really try to grapple with relevance. Their principle of relevance is a ‘generalisation about ostensive communication’ (159), which really tells us no more than that communication or language is about goal-directed behaviour, and what is relevant helps us to achieve our goals (that is, is predictive). We recognize some evidence is relevant to some conclusion, because evidence like that has helped us predict outcomes. Often the appeal of arguments that are informally fallacious is that they might have some relevance to past prediction. Ad hominem arguments are informally fallacious on the grounds of irrelevance. You cannot show an argument is invalid by attacking the speaker. However, if past evidence suggests that the speaker is unreliable, ad hominem is appealing. If our reason to accept the premises is on the basis of the speaker’s say-so, then perhaps his past unreliability is not irrelevant. However, if we have independent evidence on the premises, his reliability on past cases is not relevant. His reliability is only relevant for addressing the reliability of the premises. The evidence for premises then becomes the subject, and the speaker’s assertion of them, and his reasons for asserting them, become relevant to answering that question. Who pays for research should be irrelevant to the assessment of evidence for conclusions of that research. However, natural and social scientists are required to state who funded their research, because in any scientific endeavour a certain amount of trust in the research design, coding, collecting and analysing data is required. Demanding we declare our interests is not an ad hominem requirement.

3.6 Generalizations, laws and mechanisms In the literature on the philosophy of science we sometimes find a distinction made between empirical generalizations and law-like generalizations (or laws). Karl Popper (1972b: Appendix *x) explains the distinction by referring to an extinct New Zealand bird: the moa. Let us say that no moa ever lived for more than 50 years. There is an empirical generalization: ‘Moas do not live for more than 50 years.’ It is a true generalization, since no moa ever did live for more than, say, 48 years and two months. However, it might not stand as a law, since it could have been possible for a moa to live beyond 50 years; it just so happened that none did. But it might be a law that no moa could live for 200 years. If it were true that moas could live for more

What Is an Explanation?

61

than 50 years but not as long as 200 years, there will be reasons for that. These might have to do, say, with the physical deterioration of major organs given the basal metabolic rate (BMR) (the energy an animal expends at rest) of moas. Given moas are extinct, we cannot directly discover their BMR. However, we can infer it by indirect means – the nature of BMR for creatures like moas, the physical deterioration of major organs in creatures like moas with that rate of BMR, and so on. In other words, we discover a law governing the lifespan of creatures based on their physical make-up – including size and lifestyle – and their BMR. Our discovery constitutes a law rather than a simple empirical generalization by the narration. Once applied to moas, that law provides the reason why no moa could live for 200 years. But, note, it is the generalization that underpins the narration. James Woodward and Christopher Hitchcock suggest that a better way of viewing generalizations is in terms of their invariance (Woodward and Hitchcock 2003; Hitchcock and Woodward 2003). They say that a ‘relationship is invariant if it continues to hold, or rather would continue to hold, in the presence of a certain range of changes’ (Woodward and Hitchcock 2003: 7). The changes that matter are those that involve the variables in the generalization. Woodward and Hitchcock illustrate this with the idea of interventions in experimental situations. The relevant counterfactuals that determine invariance are those that we see as we manipulate the conditions under which the generalization holds. So a relationship R between two (sets of) variables X and Y is invariant, as it holds where the value of X is varied due to some intervention. It follows in their account that the greater the range of such interventions that leaves R intact the greater the invariance, and the greater the invariance the deeper the explanation. For many surface phenomena, generalizations that we might find are not very invariant: they are not law-like. One test (‘the inversion strategy’ introduced in Chapter 4 and returned to in Chapters 5 and 9) is not only whether we have found the generalization breaking down, but whether or not we can imagine it doing so. Where we imagine the generalization breaking down, we might be thinking about deeper underlying generalizations that lead, under certain conditions, to this one. We often think about these instances in terms of mechanisms. I will argue that deeper generalizations often underlie mechanisms, but in political science the relevant explanation can be satisfactorily given by specifying the mechanism. To return to Popper’s example, it is simply not possible for moas to live to 200 given their BMR. For that reason, Popper thought of physical laws as naturally necessary (or he called them ‘physical necessity’) features of life. If something is necessary, then it is highly predictive. Where we do not have natural necessity but just probabilities, we are left with the gap mentioned above. Sometimes those probabilities might be so high that we do not feel the need to fill the gap. But sometimes they are low enough that we do feel such

62 The Philosophy and Methods of Political Science

Box 3.10

Mechanisms and generalizations

Mechanisms and Generalizations Empirical Generalizations (variant)

Surface Properties

Mechanism (narration or model)

Mixture of Properties

Law-like Generalizations (invariant)

Subsurface Properties

A confusion in the philosophy of science that is particularly pertinent to the  social sciences is between empirical generalizations and law-like ones. In the figure above, I place such generalizations into two categories, split by the notion of a mechanism. A mechanism is a standard term used to describe explanations in the social sciences. It can be a strict model with strict predictions (as in my veto-player example, below) or a non-formal model or narration without such predictions. Empirical generalizations describe the surface properties of the social or political world, whilst law-like generalizations describe theoretical relationships that help us to explain those generalizations. An example of an empirical generalization might be that presidential systems have greater policy stability over time than parliamentary systems. An example of a law-like generalization might be ‘the more veto players in a system the greater the policy stability’. It is law-like because it follows from the meaning of ‘veto player’ within the formal model, and, given assumptions

→

a need. Hence we produce a narration that tries to fill the gap. Often we are trying to satisfy an urge to explain why something happened in this and not that token example, given the probabilities we have discovered with typelevel ultimate explanation. Marty the zebra copped it because he stumbled. That satisfies us better than the explanation that both Norman and Marty had precisely the same probability of being Alex’s dinner that day (as did several other zebras) and given that some zebra was caught that day, Marty happened to be the one. People tend not to like those sorts of explanations, especially when Marty is our mate. One of the problems for attempts to sue companies for negligence is that negligence leads to bad outcomes with higher probabilities. It

What Is an Explanation?

63

→

about policy formation that are held constant across systems, the generalization is logically implied by the model. More veto players entails more policy stability. The generalization is thus invariant. (Note that invariance does not mean that a system with more veto players cannot, at a particular time, show less stability than one with fewer veto players. It only entails having more stability over time. How much more will depend on specific assumptions in the model.) The empirical generalization has much more variance, for two reasons. First, whilst presidential systems tend to have more veto players than parliamentary systems, this is not always the case. So examples of parliamentary systems that are more stable than presidential systems can be explained by the same mechanism that is underlain by the invariant law-like generalizations. Second, even when a particular presidential system has more veto players than a particular parliamentary system, say, other things might lead the generalization not to be true of those two systems. The model or mechanism that leads from the lawlike generalization to the empirical generalization might not apply or some particular assumption of it might not apply. If the mechanism is a good one, it should also explain the variations in the empirical generalizations. It should explain why high-veto-player parliamentary systems are stable (clearly it does) and also, if not explain, at least give us a route to explaining counterexamples where specifics of the mechanism do not apply. The figure has two sorts of generalizations, empirical (variant) and law-like (invariant). In fact, any given mechanism might be underpinned by a set of generalizations (which operate together) or a series of generalizations (some underpinning others). These subsurface law-like generalizations might not all share the same degree of invariance, but will all be largely invariant. (How much a law-like generalization is a law will depend on its degree of invariance. A completely invariant generalization is, in Kripkean terms, metaphysically necessary.) Empirical generalizations will not share the same degree of invariance. In the social sciences few, if any, will be invariant. In the natural sciences some empirical generalizations might be highly invariant, which is one reason why the two categories I identify have not always been generally recognized. Finally, note that at the most reduced subsurface level, generalizations are only probabilistic. At the level of quanta everything is probabilistic.

does not establish that the death of Marty the production-line assistant from the particular cancer he had was determined by the actions of his employer when Norman worked alongside Marty but did not get cancer. Lawyers tend to see causation in terms of ‘but for’ conditions, whereas scientists tend to see it in terms of probabilities (see Chapter 6). But everyone can fall into the ‘but for’ trap notion of causation. Only recently (and more to counter the propaganda of climate-change deniers) have climatologists felt able to claim that ‘this drought’ or ‘this storm’ was caused by climate change, usually suggesting instead that climate change makes extreme events in this region more likely. Climate-change deniers use the ‘but for’ idea of causation as the bottom line against probabilistic causal claims, aided by scientists using the

64 The Philosophy and Methods of Political Science ‘but for’ condition when discussing token examples. The climatologist might claim that climate change increased the probability of our suffering ‘this drought’ by, say, 90 per cent which – given the presentation of weather forecasts (by the better weather channels) in terms of probabilities – is perfectly understandable for the general public, and a higher probability than is often given for occurring token storms. Some people write that mechanisms, narrations or ‘process tracing’ (discussed further in Chapters 4 and 6) fill in the black box of explanation. The problem with that claim is the assumption that there is a black box that needs to be filled. Sometimes we can say more about the relationship of independent variables and dependent ones. We can explain why there are limits to the natural life of animals. We can explain why some types of people are more likely to get cancer than other types. But there need be no explanation of why one person gets cancer and another does not given they are of the same type. They each had the same chance. There is no black box that needs to be filled. If drawing the winning lottery ticket really is random, then someone is going to win, and there need be no more explanation for their winning than that theirs was the chosen ticket. To be sure, one might examine the tickets in the tumbler, and follow their course to see how one was picked out. But this does not change the fact that two tickets had an equal chance of winning. Moreover, the appropriate answer as to why one lottery ticket rather than another wins is the statistical one, not the process-tracing one. The same might be true of some social phenomenon. We can process trace, we can point to some ‘but for’ conditions, but the ultimate answer is the statistical one, not the process-tracing one. Deeper explanations involve greater invariance. However, detailed narratives of token events pick out ‘but for’ conditions across equivalent sets of potential descriptions. As we shall see in Chapter 6 (Section 6.3), chasing down equivalence sets of potential explanations will end up with a probabilistic explanation because the most detailed equivalence statements are at the quantum level. My point is that there need be no more explanation of why Marty the zebra got eaten or Marty the worker got cancer when Norman did not, other than they both had the same chance and one of them got it. Trying to tie a particular token example to a statistical explanation is simply a category mistake. We should not be trying to explain this token storm or that one, just the propensity for storms in the region. For that reason, I think we should be content with narrative histories that are backed with strong evidence. They do not answer queries about type, but they do provide us with satisfying descriptions as explanations of what happened in the token history that is narrated. (I explain this in a little more detail when I discuss causation and the specification problem or ‘narrative fallacy’ in Chapter 6.) The mechanisms that are most satisfying to us are structural ones that explain differing probabilities. They go below the surface of empirical

What Is an Explanation?

65

phenomena to explain patterns in the data. A good example of such a structural explanation is contained in George Tsebelis’s (2002) account of veto players in policy stability. To understand the importance of some explanations, we need to place them in the context of the problem the writer was trying to solve, particularly in the light of previous literature. This is so with Tsebelis’s account. Students of comparative politics have noted many different features of political systems with relation to policy output. One is that over time parliamentary systems tend to have less policy stability than presidential systems. However, there are counterexamples, and furthermore, at times, presidential systems manage to bring about radical policy reform, whilst parliamentary systems seem hidebound. There is also variation in policy stability across parliamentary systems. Prior to Tsebelis, comparative scholars tried to systematize these differences within typologies, bringing out subcategories of parliamentary systems (single-party versus multi-party governments, for example) or defining some systems with features of both categories as ‘semi-presidential’. Country experts provided detailed narrative historical reasons why their countries did not fit the general trend or why they did not fit the general trend at specific moments in history. In other words, the explanations were all rather unsatisfactorily ad hoc. Tsebelis suggests we look beneath the surface of constitutional and institutional features of different forms of government. He defines two theoretical concepts – ‘agenda setters’ and ‘veto players’ – within a general (gametheoretic) model which explains the general trend of presidential and parliamentary systems, token exceptions to those trends, and times when those trends do not occur for sets of countries. An agenda setter is an agent who can present take-it-or-leave-it proposals to other players in the political game. A veto player is an individual or collective agent who can authoritatively reject any proposal that comes before it. We can distinguish institutional veto players – such as parliaments, presidents, law courts – and social veto players – such as political parties or professional groups – whose cooperation is required to ensure that policies are implemented. Tsebelis argues that the greater the number of veto players within a political system, the greater its policy stability and the less radical policy change will tend to be. Presidential systems tend to have more veto players than parliamentary ones, hence their greater policy stability. However, there are exceptions. Some parliamentary systems have large numbers of veto players, either institutional or, more often, social. Radical policy change can occur even in systems with many veto players, if they all agree on a specific course of action; hence radical policy change can occur in any system. Tsebelis and others have provided empirical support for the thesis: generally speaking, presidential systems show greater policy stability than parliamentary ones because they have greater numbers of both institutional and social veto players.

66 The Philosophy and Methods of Political Science The beauty of Tsebelis’s model is its simplicity. With a few simple, theoretical concepts and relatively simple straightforward game-theoretical and spatial models, he produces clear-cut hypotheses. Some empirical work is needed to correlate the theoretical actors with actual actors in different systems, but once that is done, we have testable hypotheses. He produces a mechanism that is important because it looks below the surface phenomenon. He cuts through what had become a seemingly endless set of ad hoc ideas about policy stability. His models are structural. Agenda setters and veto players are defined clearly and their relationships mapped in terms of their types of preferences to produce statistical ultimate-type explanation of policy stability. We note that applying the veto-player model to a token example adds little by way of explanation to its narrative history. All it would add is a description of certain actors as being ‘agenda setters’ or ‘veto players’ that is probably completely otiose to the narrative-description explanation of events there. (And rarely are there actual agents who are, for example, agenda setters. Rarely are proposals truly take it or leave it.) Often there is no one-to-one correspondence between biological individuals and agents in theoretical models, but the models still pick out structural features of reality (see Ross 2005, 2014 for an account of this feature of models in the social sciences). Statistical propensities associated with ultimate explanation do not add to narratives, because they rely upon assumptions about the distribution of preferences among actors. But when we narrate a token history we can see the actual preferences; hence the ultimate explanation adds nothing to that narration. The ultimate explanation is about type – why presidential systems have greater policy stability than parliamentary ones. It is not supposed to explain why there was stability in system X at time t1, but radical change in system X at time t2. A proximate explanation of those features should be consistent with the ultimate explanation. We can represent the reason for stability in system X at time t1 in terms of veto players vetoing change, but saying ‘the radical legislation was struck down by the veto-player Supreme Court’ adds nothing to the explanation that ‘the radical legislation was struck down by the Supreme Court’. Or saying ‘the veto-players president, Republican Senate majority, Democrat House majority and Supreme Court all shared the same views on the need for the radical policy change Z’ adds nothing to ‘the president, Republican Senate majority, Democrat House majority and Supreme Court all shared the same views on the need for the radical policy change Z’. They might be veto players, but the theoretical concept does its work in the comparative and probabilistic setting, not in individual case studies. It is the role of structural explanations, mechanisms and ultimate explanations to explain across settings rather within them,

What Is an Explanation?

67

though they can frame the particular questions and evidence we look for in specific case studies. I will return to the discussion of mechanisms in Chapters 4 (Section 4.5) and 6 (Sections 6.3 and 6.4) in the context of causation. Some argue that to specify causation we have to provide mechanisms. I will argue that causation needs to be defined separately from mechanism and that where we demand finer and finer details in the form of a mechanism as process tracing to pin down causation (what I term ‘bump-bump causation’), we are actually demanding universal laws. Mechanisms rely upon a prior understanding of causation or upon law-like generalizations in order to satisfy our explanatory demands, unless, as I have suggested with the veto-player example, they provide a structural narrative at the level of type that enables us to partition the systematic from the non-systematic elements in a token narration of some particular outcome.

3.7 Conclusion I’ve argued that we are pattern-finding creatures and that we find patterns in order to reduce surprise. We want to predict what is happening around us, and our language, our descriptions, laws, mechanisms and models of the world enable that prediction. We are satisfied that we have explained something when we are confident in our predictions and feel the descriptions enable us to understand processes of the world. The social world does not have many invariant laws and those that we might find act as constraints on or encouragements (‘structural suggestions’) to the actions of agents. As constraints or encouragements, social structures can be seen clearly at the aggregate level and form the basis of ultimate explanations, but are often harder to see at the individual level within proximate explanations. I have introduced the categories of type and token, proximate and ultimate explanations; and argued that explanation must always take account of the interests and knowledge of the questioner in order to provide satisfactory explanation. I have also argued, however, that it does not follow that explanation has to be an explanation for someone; that it does not entail that anything can stand as an explanation; or that explanations, whilst directed at subjects, are not objective in any reasonable manner. Prediction, as I have used the term, provides the underlying element of that objectivity.

Chapter 4

What Is a Theory?

4.1 Introduction Students are often taught that in order to have an explanation you have to have a theory. A standard critique of PhD students’ initial proposals is: ‘What is the theory you are testing?’ In some ways, this is a rather old-fashioned question. At least one theory in the social sciences, ‘complexity theory’, suggests that we should not try to test theories, but rather let the data talk. One aspect of the movement towards ‘big data’ is developing description from which inductive inferences can be made, rather than theory testing. There is still much controversy about whether and how much big data affects our methods (something I look at in Chapters 5 and 7), and how far it is inconsistent with the other major movement in political science, the experimental turn (Grimmer 2015; Monroe et al. 2015). The experimental turn provides strong criteria on generating causal inferences, though any causal claim depends upon some underlying theory. When it comes to examining particular events, a good detailed description can be much more fruitful and enlightening than a half-assed attempt at testing some complex theory. Of course, the description needs to be original; we do not want the same thing described over and again. Work can be question- or problem-driven rather than theory-testing. Half-assed testing is what often happens when essentially descriptive analyses are used to ‘test’ theory. Rarely can theory be tested by a case study, however detailed (see Chapter 5 and Chapter 6, Section 6.4). In truth, the term ‘theory’ has many different meanings in the social sciences. Consider the theories that are discussed in Dryzek and Dunleavy (2009): pluralism, elite theory, Marxism, market liberalism, neo-pluralism and governance, feminism, environmentalism, conservatism, postmodernism and globalization. Many of these have very different logics; even what it is they are trying to explain varies. Whilst we use terms like rational choice theory, and some of the isms of Chapter 2 are considered theories, we also have specific formal models that are given the epithet ‘theory’, and empirical generalizations such as the theory of democratic peace are often described as theories. Some of the theories in Dryzek and Dunleavy (2009) are fairly specific arguments about the way in which the state operates – pluralism, for example, is a claim that power in the modern state is held by different competing

68

What Is a Theory?

69

elites. Power-elite theory asserts that there is a single power elite that controls the state. Now, to some extent, these two theories are simply claims about the way the state operates. They do not, in themselves, imply any particular method of studying power in society. However, they are associated with specific methods of study. Classical pluralism is associated with behaviouralism  – studying what actors do in an untheorized manner, surfacing the important issues by examining what appears in newspapers, interviewing key actors, and so on. The power-elite theory is associated with reputational analysis – interviewing people and asking who the key actors are, then interviewing those key actors and triangulating. Critics of both theories often argue that the picture of the state each promotes is dependent upon the specific methods adopted. We see therefore that many theories of the state are both descriptions of the power structure of states and accounts of the mechanisms that keep those structures in place, but may also include particular methods of how we develop those theories. Other theories of the state are more directly methodological. Feminist theories of the state also concentrate upon aspects of power relations between actors, notably gender relations, and (as Dryzek and Dunleavy suggest) one could have a feminist version of pluralism or power elitism, with gendered relations between sets of elites in the first and within the elite in the second, and between elites and non-elites in both. Similarly, Marxism can be seen both as description of what goes on in the state and as prescribed methods or theories about how to view relations (which involve, for example, the labour theory of value; substructure–superstructure relations; the perpetuation of an ideology, and so on). The theories of governance and globalization are different again, being accounts of changes in the ruling structures of states. So in these grand theories of the state we have various ways in which the term ‘theory’ is used. Some people claim that those who hold those different theories have different ‘ontological and epistemological commitments’ (Marsh and Furlong 2002). However, it is perfectly possible for two people to agree on the nature of existence and the methods for discovering how the world is, but disagree about the nature of the power structure, because of differing assessments of the evidence, different expectations about what is possible or their different normative views. If someone believes that the state is governed by elites but power ought to be more evenly spread, then presumably they think the state could be elite-driven or pluralist. In the main, different ‘ontological and epistemological commitments’ simply means different ways of assessing evidence given expectations about what is possible, together with different normative views. Even in the philosophy of science, what constitutes a ‘theory’ varies. Karl Popper admits he does not use the term consistently. He opens Chapter 3 of The Logic of Scientific Discovery with the statement ‘Scientific theories are universal statements’ (Popper 1972b: 59), but goes on to suggest that

70 The Philosophy and Methods of Political Science

Box 4.1 What are theories? Theory (Perspectival) Organizing Perspectives

Methodology

Normative Perspectives

Theory (Explanatory)

Formal Models

Mechanisms

Informal Models (Incl. Typologies, Classifications)

Predictions or Hypotheses

Empirical Generalizations

Predictions or Hypotheses

Many different sorts of things are called theories. A purported generalization might be termed a theory, so might a mechanism. Hypotheses are sometimes described as theories, whilst at a more general level what I call organizing perspectives or methodologies, such as rational choice theory or postmodernism, are called theories. There is also a host of theories, from theories of justice to theories about the way in which the state operates, which are highly normative as well as (purportedly) descriptive. I break these down into Theory (explanatory) and Theory (perspectival). The latter is simply to note that we often have different perspectives on the world and how to go about explaining it; these different perspectives are sometimes called theories. The text discusses the nature of these perspectives.

→

a theory is rather more than that. Sometimes, though, he uses the term to mean ‘hypothesis’. Elsewhere we see the term ‘theory’ standing for different methodological processes – ‘economic theory’, ‘rational choice theory’, ‘sociological theory’, ‘discourse theory’, and so on. If you read psychology or socio-psychology journals you will find that virtually every paper presents a new theory of something. Sometimes these constitute hypotheses about behaviour, but often they are simply conjectures about different psychological processes underlying behaviour. For me, theory is a general term that can stand for any type of explanatory claim or conjecture, well formulated or not. In this book ‘theory’ is a generic term that can stand for any of the following (in alphabetical order initially): conjecture, framework, generalization, hypothesis, mechanism, methodology, model, organizing perspective, paradigm, many of the isms such as historical institutionalism, Marxism, ‘realism’ and so on, or for more overtly normative theories such as libertarianism or utilitarianism. I shall discuss different forms of theory, interlocking subsets of the general term ‘theory’ for

What Is a Theory?

71

→

Theory (explanatory) is more complex. Over the course of Chapters 4 to 6 I discuss the relationship of generalizations to mechanisms. Generalizations can be empirical or law-like, and how far they are law-like depends upon how invariant the generalization turns out to be. Mechanisms are less invariant, but often, as we specify the mechanism more precisely, what we specify becomes more invariant. Models can be seen as descriptions of the purported mechanisms. Some claim that models need not represent anything: as they make predictions they are useful in explanation. In the text (and see Box 4.5) I argue that models predict best when their structure represents real mechanisms or structures in the world. Formal models, often mathematical, but certainly deductive, produce strict predictions or hypotheses. These will correspond to some of the generalizations (laws or empirical generalizations) that underlie the mechanisms. But, note, the assumptions of formal models will also correspond to generalizations that underpin mechanisms. Informal models also lead to predictions. How clear the predictions are will depend on how deductive the informal model is, but informal models often do not produce clear predictions, and are not clear about the precise mechanisms that are supposed to underlie those predictions. We might say formal models are deductive and can be judged by the criteria of formal logic or mathematics. Informal models vary in how informal they are and how far they can be judged by the criteria of logic; that is to say, it is less clear how justified are the inferences that are drawn from them. Again, the assumptions of informal models should correspond to some generalizations, but the problem with informal models is that the assumptions are not all clearly specified (hence it is harder to judge them by the criteria of logic). To the extent they produce predictions, these will also be generalizations (laws or empirical generalizations).

which I will stipulate conditions of application to demarcate them from one another. In keeping with my general line on concepts, however, at the margins some of these forms of theory merge together. Many theories are complex and have elements that fit into my different categories, but that does not make the categories any less helpful in navigating our way. Taken together, these categories are intended to cover most of the important ways in which ‘theory’ is used in political science. I think there are three broad cases of theory relevant to political science. First, perspectival theories, that lead us to examine the world in certain ways; second, explanatory theories, that try to explain the world; and third, normative theories, that try to say how the social world should be organized (I leave these more or less alone until Chapters 8 and 9). We ought nevertheless to recognize that some theories cross these boundaries. Many of the ‘theories of the state’ mentioned above can be seen as both explanatory and normative, whilst some explanatory models sit firmly within a given organizing perspective (such as rational choice models).

72 The Philosophy and Methods of Political Science

Box 4.2 Types of theories (applied to politics) Type

Description

Examples

Perspectival theories

Specific ways of looking at the world Question generating Tend to utilize specific methods pertinent to answering questions they generate

Rational choice theory, discourse analysis, complexity theory

Explanatory theories

Models that try to explain types of outcome

Principal–agent, vetoplayer democratic peace, network models

Normative theories

Worked-through accounts of how the social world should be organized

Utilitarianism, Rawls’s theory of justice

Mix of all three

Takes a specific perspective and a normative stance and claims to explain the way society operates

Institutionalism, pluralism marxism, feminism

4.2 Organizing perspectives I will begin by briefly examining perspectival theories. Some of these are fairly straightforward methodological approaches to answering questions. For example, both rational choice theory and discourse analysis are methodological approaches to answering questions about political processes. They are often seen as rivals, but I do not think this is necessarily the case. There are also those who consider them theories that can be ‘tested’. Again, I do not think this is the case. Many other accounts of perspectival theories have been offered. Thomas Kuhn’s (1962, 1970) account of paradigms could be seen in this light; as could Imre Lakatos’s (1978) account of research programmes. I am going to use the phrase ‘organizing perspective’ as a generic term. I take it from Greenleaf (1983), though I use it more in the manner of Andrew Gamble (1990: 405). I choose to use ‘organizing perspective’, largely because it is not in much current use, and I do not want to associate what I say about perspectival theory too closely with the work of Kuhn or Lakatos. I do not use Lakatos’s notion of a research programme, since I do not think the idea of the core and the auxiliary hypotheses very helpful or accurate, at least not for all perspectival theories. And whilst Kuhn has had an enormous influence in certain areas of the philosophy of social science, this influence is largely baleful, mostly because the term ‘paradigm’ is so ambiguous. In an early review of Kuhn’s book, Dudley Shapere (1964: 385) points out that the term paradigm: covers a range of factors in scientific development including or somehow involving laws and theories, models, standards, and methods (both

What Is a Theory?

73

theoretical and instrumental), vague intuitions, explicit or implicit metaphysical beliefs (or prejudices). In short, anything that allows science to accomplish anything can be a part of (or somehow involved in) a paradigm. Margaret Masterman demonstrates that Kuhn uses the term paradigm in ‘not less than twenty-one different senses, possibly more not less’ (Masterman 1970: 61). Some of these ways of using paradigm clearly mean explanatory theory in the form of models (as I define the term below) (see Box 4.3: Kuhn on paradigms). What I mean by organizing perspective is a subclass of the myriad ways in which Kuhn uses the term paradigm. Masterman divides Kuhn’s notion of paradigm into three different groups, metaphysical, sociological and artefact or construct paradigms. Of these, it is the metaphysical that is the most widely used in subsequent social theory. Indeed in one of these metaphysical forms Kuhn describes a paradigm as an ‘organizing principle’ that can govern perception itself. It is in this sense that much subsequent social theory takes scientific paradigms and takes a relativist reading. That is, many take this to mean there are different incommensurable ways of looking at the same evidence, depending on one’s organizing perspective. In fact, that is not Kuhn’s own reading (see, for example, Kuhn 1970, 1977), and owes more to another historian of science, Paul Feyerabend (1975, 1978). Relativists think that people have different ontologies and epistemologies and there is nothing further to say (though they usually insist on saying more). Where we see our perceptions governing how we navigate the world we will think there are preferable ways of looking at the world, ones that are more predictive. Another reading is realist, that different perspectives give people different views of the same object. Kuhn’s general line is that at any given moment there is a paradigm that dominates in science, and that paradigm constitutes ‘normal science’. He argues that paradoxes or problems emerge as the paradigm fails to predict and explain certain phenomena. At that point some scientists think out of the box, and a new paradigm arises. During these revolutionary phases two rival paradigms might grip science until a new one prevails, and normal science continues under the new paradigm. In other words, one paradigm replaces another because it is superior: it enables us to answer questions that earlier paradigms could not. But this does not describe the way in which different organizing perspectives often operate in social science. Rather, different organizing perspectives operate simultaneously. And they cannot overthrow each other since they are not themselves subject to empirical test. Explanatory models that utilize that perspective are open to empirical test, and if models or explanations derived from a particular perspective keep failing or do not help us to predict what is happening around us, the particular theoretical perspective might fall into disuse, but it is not strictly falsified because it is not being tested. The change is

74 The Philosophy and Methods of Political Science

Box 4.3

Kuhn on paradigms

Kuhn uses the term paradigm in ‘not less than twenty-one different senses, possibly more, not less’ (Masterman 1970: 61): 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21

a universally recognized scientific achievement a myth a philosophy or constellation of questions a textbook or classic work a whole tradition a scientific achievement an analogy a successful metaphysical speculation an accepted device in common law or pattern of findings or model a source of tools a standard illustration a device or type of instrumentation an anomalous pack of cards a machine-tool factory a gestalt figure that can be seen in two ways a set of political institutions a standard applied to quasi-metaphysics an organizing principle which can govern perception itself a general epistemological viewpoint a new way of seeing something that defines a broad sweep of reality

→

gradual as some methods simply get replaced by others. In this sense of paradigm the use of game-theoretic methods in economics and political science have enabled new understandings but they have been squarely built upon older methods as an extension rather than a revolution (Ross 2014: 195). A rather different way of using Kuhn’s notion of a paradigm also seems to describe some battles in social science. Here rival paradigms try to explain the world, but it is not possible for one to overthrow the other because the proponents of each do not accept the assumptions, methods or means of assessing evidence that exist in the rival paradigm. As a description of what seems to be going on at the separate tables of political science (as Almond (1988) once described it), this has some validity. However, this radically relativist approach to social science does not represent the sort of progress we have made in understanding the social world. Often the different perspectives are not empirical rivals, because they are not asking the same questions; and, to the extent that they are rivals, they are normative rivals. Proponents of each think the other is asking irrelevant, uninteresting or unanswerable questions; or are driven by ideology to look

What Is a Theory?

75

→

Masterman suggests they fall into three categories: Metaphysical (1, 2, 8, 11, 17, 18, 20, 21) = metaphysical paradigms – what critics refer to Sociological (1, 6, 9, 16) = sociological paradigms Artefact or construct paradigms (4, 7, 10, 12; grammatical 15) She does not say what 3, 5, 13 14, 19 are. Kuhn (1977: 460) says ‘Though neither … [Masterman] nor I think the situation so desperate as those divergences suggest, clarification is obviously called for’ and introduces two important senses of paradigm: 1 Global – embracing all the shared commitments of a group which he called a ‘disciplinary matrix’. ‘“Disciplinary” because the common possession of the practitioners of a professional discipline; “matrix” because it is composed of ordered elements of various sorts, each requiring further specification’ (Kuhn 1977: 463). 2 Local – an important sort of commitment and a subset of (1) which he calls exemplars. He goes on to say that symbolic generalizations are formal elements of a matrix and models are what provide the group with preferred analogies or, when deeply held, with an ontology. At one extreme they are heuristic, at the other they are a metaphysical commitment. Exemplars are concrete solutions to problems accepted by the group as paradigmatic. All three need to be understood in order to comprehend how a scientific community operates. A change in any one of these will affect the validation of research.

for answers that fit a pre-ordained view of the world. In my view, two explanatory theories are only true rivals if they produce different predictions from the same data, in the sense that if one prediction is true the other is false. I say more about this in Chapter 5. So the explanatory theories (or what I will be calling models) that are informed by different perspectives are often not empirical rivals. Kuhn’s use of the term ‘paradigm’ is just too loose, and his account shifts too much between what looks like radical relativism and something that looks more realist, to be of much help. So I see an organizing perspective as a way of looking at the social world that structures discussion and generates sets of questions. The questions are often very different in nature, and though at times the answers offered by different approaches might appear to be rival, we need to demonstrate that they do in fact produce different predictions. We need to show that the explanatory theories (models) within different perspectival theories are actually rival. We can only do that if there are clear empirical consequences that are, at least in principle, testable against each other. In that sense the models derived from within one perspective will drive out those from the other perspective, which

76 The Philosophy and Methods of Political Science will fall into disuse. But if the models do not produce rival predictions, then we have no warrant to claim that the perspectives are rivals. Whilst organizing perspectives can govern the way in which we perceive a problem, it is the stance from which we view it that is different. Taking the perception metaphor literally, if one looks at the same object from a different perspective it will appear different. A cat seen from above does not appear precisely the same as from below, or from inside out, but those observations are not rival. This is relativity rather than relativism. In the sense I have been using the terms, discourse theory and rational choice theory are different organizing perspectives. Discourse theory is the idea that our experience is largely written for us by a multitude of different, often conflicting, discourses. The social world, including our identities, is constructed by those discourses. As a methodology, it explains how we can come to know the social world, and it includes a set of methods for studying it. The underlying thesis of discourse analysis is that social reality is produced through discourse – verbal, written, symbolic – and also through behaviour. How is discourse understood? In one standard text it is defined as ‘an interrelated set of texts, and the practices of their production, dissemination, and reception, that brings an object into being’ (Phillips and Hardy 2002: 3). Whilst discourse analysis uses techniques familiar to many other qualitative analyses, such as studying primary source documents, interviews and ethnological practices of studying actors ‘up close and personal’, it wants to describe the meanings of interactions in terms of specific discourses. In that sense, discourse theory is an organizing perspective. It cannot be refuted, since its claim is that our identities are contained in the meanings that we gain through the discourses of life. Put in that manner, the claim is analytic. It must be true that the meaning we ascribe to the (social) world is contained in our discourse. The underlying claim of discourse theory is, I take it, axiomatic. And that is why it is not falsifiable. However that does not mean that discourse analysis as an organizing perspective is not scientific. Some claims contained within discourse theory might be falsifiable. For example, it might not be true, perhaps, that different discourses always conflict: that is, it is not necessary or logically true that different discourses conflict (for example, if Habermas’s (1984) ideal speech situation could be generated, then perhaps discourses would not ultimately conflict). However, the basic discourse claim is surely true and few would doubt it; but I take it that the specific claim that discourses always conflict is falsifiable. It is falsifiable in the sense that the claim can certainly be supported (or corroborated, in Popper’s language) by finding discourses that conflict. Moreover, one might have rival discourse analyses of the same issues. As I argued above, if they are truly rival then there must be claims made in one analysis that conflict with claims in another. Hence the hypotheses drawn from one analysis can be pitted against those drawn from another.

What Is a Theory?

77

Only if it is not possible to assess any claims made from discourse analysis can we say it is unscientific. If that were so we would have no warrant for claiming the analysis was explanatory; it would not be predictive. The interest and importance of discourse theory are derived from the questions the technique generates and the answers that emerge from them. Rational choice theory is also an organizing perspective. As a programme of study, it makes certain assumptions: technically that in order to interpret human behaviour the preferences we presume agents hold have to abide by certain assumptions, and given those assumptions we are able to explain human action by the interrelationship of those agents seen as sets of preferences and the social structures within which they operate. Rational choice theory then uses a set of techniques, notably social choice and game theory (both also organizing perspectives on this account), to model social situations. As an organizing perspective, rational choice theory is not falsifiable, but many hypotheses drawn from its models are. One can falsify one rational choice model and support a rival one or one game theoretic model and not another without falsifying rational choice or game theoretic methodology itself. I think it is more problematic to claim that rational choice theory is not refutable in the way that discourse analysis is not refutable. There is some debate about the nature of some of the basic assumptions surrounding the interpretation of preferences. I will weasel my way out of that difficulty by suggesting there might be rival rational choice perspectives that utilize slightly different sets of assumptions about the characterization of preferences. If we systematically falsify models drawn from one perspective in favour of models drawn from another perspective, we will reject the former. And we might care to call that rejection a falsification. Nevertheless, as a perspective, some form of rational choice analysis is not refutable, only hypotheses drawn from specific models, which then tend to falsify some models in favour of others. So organizing perspectives cannot be falsified, but hypotheses, implications, predictions, conclusions or claims drawn from within models can. Organizing perspectives cannot be pitted against each other in terms of the veracity of their conclusions. For example, it is sometimes suggested that ‘rational choice’ produces predictions which are false relative to other approaches (Green and Shapiro 1994; Udehn 1996), but those who argue thus pick their cases carefully and ignore the fact that those same rival conclusions can be derived from a rational choice model competing with the one they chose to discuss. I take it that – much of the rhetoric deployed in academic articles notwithstanding – discourse analysis and rational choice methods are not generally rivals when it comes to explaining the social world, since the questions they address are rather different. In most of the clashes I read, I see one specific analysis using discourse methods pitted against a particular conclusion drawn from a specific rational choice model. I can

78 The Philosophy and Methods of Political Science usually see how another conclusion might be reached from a model based on different assumptions or by using a somewhat different discourse approach. One must be clear what one can conclude about an organizing perspective by examining a restricted set of models or frameworks drawn from it. There is good and bad discourse analysis, good and bad rational choice modelling; one cannot demonstrate much about either approach from examples taken in isolation. The lesson I would like the student to draw is to worry less about condemning whole theoretical perspectives and more about specific arguments or models that pertain to the specific issue they are concerned with. A particular explanation might fail because its purported explanation is wrong; or it might fail to address the specific question that the student is working on. One is a direct rival, the other merely shows there is room for more work to be done. Rational choice theory, discourse analysis, critical theory, evolutionary theory, social choice theory, game theory, even postmodernism, do not, each as a separate body of literature, produce unified conclusions. Rational choice models compete against each other over the same phenomenon. Discourse analysts dispute with each other the correct analysis of some phenomena. Critical theorists debate amongst themselves, and so on. These organizing perspectives carry on their own internal debates about the best techniques, evidential criteria and specific explanations of different phenomena. The failure, however absolute, of one argument or model within a technique does not demolish the organizing perspective (the ‘theory’). This fact alone makes irrelevant hosts of books and articles that attack a whole organizing perspective (‘rational choice theory’, ‘behaviouralism’, ‘discourse theory’) on the basis of problems with some examples drawn from that perspective. The fact that an organizing perspective cannot be falsified does not mean that it cannot be criticized. Organizing perspectives are ways of looking at the world, ways of generating questions, and usually include a specific set of techniques and methods (in some cases these techniques and methods can be large in number, broad in scope, and utilized quite happily by other organizing perspectives). They can be criticized for generating uninteresting or dull questions; or for missing important aspects, even the most important aspects, of the phenomenon being addressed. They might be criticized on normative grounds because the very techniques adopted do not allow for some questions to be asked. If a given organizing perspective were the only perspective adopted, for example, and it simply did not generate some questions, then some important social phenomenon or aspect of the subject of study might be ignored. To paraphrase a famous expression in political science, ‘Organizing perspectives organize some questions into political study whilst organizing others out.’ In that sense, organizing perspectives might be ideological in that the questions they organize out of the study might just be those that enable normative critique of the political or social structure. Rational choice theory

What Is a Theory?

79

was once criticized for being ideological, though its perspective has now been adopted by writers of so many different ideological viewpoints that the charge can no longer hold. Sometimes, of course, a view is just wrong; being proved wrong does not mean the proof is ideological, no matter how much we dislike its conclusion. What we might agree is that working within any given organizing perspective tends to cant questions in specific ways and with specific normative spins. It might be that some organizing perspectives simply cannot countenance certain types of questions, though this claim would need to be demonstrated, not simply asserted. I am not sure that it is true, and have never seen it demonstrated. But certainly working within a perspective tends to generate questions of a certain kind because of its outlook on the social world. I do not think that it necessarily, as is often claimed, determines normative direction, but it might bias the type of questions and answers expected. One might think of organizing perspectives as methodologies with particular methods of study, and that would be perfectly reasonable. However, I think that adopting a particular organizing perspective to generate questions does not entail specific methods or methodologies to answer them. Rational choice theory is associated with particular analytic and mathematical techniques, but discourse theory can equally well use logic or mathematical reasoning. Qualitative historical descriptions in the form of analytic narratives have been framed by rational choice reasoning and stricter hypotheses drawn and tested using quantitative data and statistical methods. We might agree on the rational choice organizing perspective, but disagree over the appropriateness of qualitative or quantitative methodology then adopted; or agree over the quantitative methodology for testing, but disagree over the particular statistical method used to analyse the data. Much discourse analysis has examined text in a selective and qualitative fashion, picking on specific examples and highlighting changes in the meaning of words or phrases over time. Or it has told historical narratives in which understandings of key elements of social life have changed over time. It can also use dedicated computer software to systematically analyse changing usage within contexts or to examine how specific issues are framed in totally different ways over time. No specific qualitative or quantitative methods are entailed by discourse analysis. Hence my emphasis on the organizing perspective category as distinct from specific methodologies and methods used.

4.3 Explanatory theories or models Explanatory theories are those that logically produce predictions or hypotheses that in principle, and usually in practice, are testable. I use the term ‘model’ to stand for such explanatory theories. A model of something is a representation of that thing. A good model is isomorphic to that which it

80 The Philosophy and Methods of Political Science represents in the relevant aspects. Models are usually simplified versions of the things they represent, eliminating aspects that are not important for the use to which the model is being put. So two models, non-isomorphic to each other, might apply equally well to something if they are representing and helping us to understand different aspects of that thing. For example, we might use a principal–agent model to represent one aspect of public agency and a network model to represent another aspect of it. The two models might be completely different, but each may explain different aspects of bureaucrat behaviour within that public agency. Or we might consider different models of legislative politics that each attempt to capture different aspects of legislative behaviour. Explanatory theories, on this view, model the world they are trying to explain. They represent it in a fairly straightforward sense. In the literature you will find typologies of different types of scientific models and far more formal definitions than I offer. Some make useful distinctions in how models are used. Fudging models are used for heuristic purposes; that is, to help computation and cognitive purchase. Some models are used to pick out the explanatorily important or salient features of some process; whilst forecasting models take little account of mechanism or causal process, but are used simply for their powers of prophecy. They might not be considered explanatory, though if they do forecast accurately, miracles aside, we should expect to be able to produce some further model, that is explanatory, of how they perform. I do not make much use of these distinctions, but talk more generally about the utility of models for explanation. Generally speaking in political science, models are sets of statements related formally or analytically to generate testable hypotheses. Models are deductive and have a set of assumptions

Box 4.4 What is a model? A model of something is a representation of that thing. A good model is isomorphic to that which it represents in the relevant aspects. Models are usually simplified versions of the things they represent, eliminating aspects that are not important for the use to which the model is being put. Models are deductive and have a set of assumptions, which together logically generate a set of conclusions or logical implications, usually called hypotheses or predictions. Strictly, what we get from the model is the logical implication. How we apply it to the world is the hypothesis or prediction (‘from this implication, given the abstractness of the model, we expect that …’). It is the hypotheses that are empirically tested. From the results of those tests, we can make judgements about whether we think the model is true or false. Two models, non-isomorphic to each other, might apply equally well to something if they are representing and helping us to understand different aspects of that thing.

What Is a Theory?

81

which together logically generate a set of conclusions or logical implications, usually called hypotheses or predictions. Strictly, what we get from the model is the logical implication. How we apply it to the world is the hypothesis or prediction (‘from this implication, given the abstractness of the model, we expect that …’). As I discuss in Chapter 5, it is the hypotheses that are empirically tested. From the results of those tests we can make judgements about whether we think the model is true or false. Some people will say that models (or theories when used in the explanatory sense) cannot be true or false, merely more or less useful. It might not matter too much whether we think of models as more or less useful, or more or less true, though personally I can see no error in predicating truth on them. Clarke and Primo’s (2012) book on models in political science announce that models cannot be true or false because they are objects. A coffee mug is an object, they say, and it cannot be true or false. But then the sentence you are currently reading on the page or screen is also an object. Does that mean sentences cannot be true or false? The referent of a sentence is its corresponding proposition, and it is the referent of a proposition that has the truth-value. Here the proposition can be thought of as the meaning behind the sentence. To the extent that any object takes on meanings, it can take on a truth-value. We might say of a purported coffee mug that it is a false coffee mug if in fact one cannot fill it with coffee (say there is some hidden glass covering the opening). Oh, but then it is not a coffee mug! But that is merely to say that its representation as a coffee mug (for some amusing jape) is false. And models are explicitly representations that carry meanings for that which they represent. Clarke and Primo (2012) usefully draw an analogy between semantic models and maps. But we easily predicate truth-values to maps precisely because, like language, they are signals carrying meaning. If you ask me where the town centre is and I point south, but when you walk in that direction you find yourself leaving town, was my signal not false? We use maps for such directional signals and they can mislead. The British Ordnance Survey deliberately placed minor errors on their maps to catch out commercial companies reproducing them without authorization (and once took legal action against the Automobile Association for so doing). Can we really not predicate ‘falsity’ to those deliberate errors, and ‘truth’ to correct representations? Sometimes it is suggested that models share features with the world by ‘similarity’, but is ‘similarity’ not similar to ‘correspondence’? There is a long tradition of correspondence theories of truth. Of course, we rate models by their usefulness, and we could stick with calling them ‘useful’ rather than ‘true’, but those who hold pragmatic theories of truth might say their degree of truth depends on their degree of utility. Nothing might hang on this verbal dispute, but I will stick with thinking that models can take on truth-values, and I will do so because of my realism. In formal political science, following a famous article by Milton Friedman (1953), it is often claimed that good

82 The Philosophy and Methods of Political Science

Box 4.5

False assumptions

Formal models often utilize assumptions that do not fully reflect reality; rather they are simplifications. But the better those assumptions reflect reality, the better the model will be. We can say that where the assumptions do not fully reflect reality they are not completely correct. Sometimes they are called ‘false assumptions’. In a famous essay Milton Friedman (1953) defended economic models based on ‘false assumptions’. He made this claim because he thought economics was only really concerned with prediction. He illustrated his argument by suggesting we can model how trees grow their leaves and branches ‘as though’ they mathematically work out the movement of the sun; or we can model pool or snooker players ‘as though’ they mathematically calculate all the angles. The problem for his argument is that neither of these examples utilizes false assumptions. Trees do work out where the sun is moving; and pool players do calculate the mathematics of the angles of their shots. Of course, neither does so consciously. It is a common mistake in the social sciences to think that the reality of agency requires us to model agents’ conscious thoughts in order to explain their behaviour. But if we can predict their behaviour better without such considerations, then we can explain it better without such considerations. If you find it hard to believe that pool players do in fact work out the mathematics of their shots, consider the following. You find what appears to be a living organism snuffling around in the grass, doing what unintelligent herbivores tend to do. However, it seems to have growing on its back a keyboard with numbers and symbols. You discover that if you use the keyboard to type in mathematical questions, the creature produces from a hole in its side paper-like

→

models can have false assumptions because their aim is to predict. But unless models predict by miracle, they predict because of their features and we can understand how they predict by those features. Box 4.5 explains why I think Friedman is simply wrong. Clarke and Primo (2012) also disagree that good models can have false assumptions, saying that assumptions cannot be true or false because models cannot be true or false. However, I like to think that the utility of any given model depends directly on the truth of its assumptions, though on the truth of those assumptions as they are isomorphic to the structure of the phenomenon to be explained. This is particularly important where we consider models as representing mechanisms as I do below. Perhaps I just want explanations to be true ones, and not simply representations about which neither truth nor falsity can be predicated. (Forecasting models are ones judged by their ability to predict the future in some regard. That might be because they are explanatory, but it might be that they make no claim to explain the phenomenon being predicted; they might, for example, correlate sets of phenomena that are caused by something else.)

What Is a Theory?

83

→

threads that have printed on them the correct answers. Does the creature work out the mathematics or does it act ‘as though’ it does? Does the calculator on your computer work out the arithmetic you set it or does it just act ‘as though’ it does? Pool players do not produce their mathematics on paper. Rather, the mathematics we can produce on screen or on paper to calculate the angles of their shots is a representation of what they do. (And our mathematics can represent their play when they make the shot and when they miss, though it will not represent, necessarily, what they consciously intended to do.) Even though the player does not represent the shot in that manner contemporaneously with making it, that does not mean the representation is not a representation of the behaviour. We are not calculating with false assumptions, unless we do assume that the player makes that contemporaneous conscious representation. But we have no need to do so. Similarly, if we model a firm’s reaction to demand, we need not assume that the reaction we model also has contemporaneous conscious form in the minds of the firm’s managers. What they are doing and what they think they are doing are not irrelevant to what happens, but nor are they necessarily the best representation of the firm’s reaction. We might only be interested in prediction, in which case we might not care about the relationship of the model to the world. However, if a model consistently produces hypotheses that stand up to empirical scrutiny it will do so because it relies upon some mechanism that either more closely reflects reality than the structures and assumptions of rival models that produce hypotheses that do not stand up so well to empirical scrutiny, or because it relies upon some features that are correlated with such mechanisms. We have no warrant for thinking otherwise.

We can think of verbal models analogously to non-verbal ones. A toy car represents a real car and shares some features with it; these features might enable us to understand features of real cars. For example, a toy car with turning wheels at the front much like a real car can be used to explain why it is easier to drive into a parking space and reverse out, and why one can reverse into parallel parking spaces that one cannot drive into. Of course, using a toy car to demonstrate these facts could be replaced by geometric mathematical representation and explanation. We might think the latter is a more fundamental explanation; though it would only be an explanation for a person who could follow the mathematics. If not, the use of the physical toy car would provide a better demonstration and illustration. The mathematical representation can be thought to be superior overall, since it instantiates a method that can be used for other models, tying them together through the mathematical form. So models are simplified representations of the world that are designed to abstract (some) important features so that we may examine some of their causal effects in theory and try to examine them more closely in reality. We begin to model when we try to put together data (by which I mean any

84 The Philosophy and Methods of Political Science reasonably systematic information about the world) into a set of propositions by which we relate the data together. And this is just as true of formal mathematical models as of more descriptive ones. Any formal model of society is supposed to model aspects of that society, and it is thinking about those aspects that leads the modeller to shape the model in the way he or she does. We start by collecting information (even unconsciously) about the world and conjecturing about how it fits together in terms of identity and causal relationships. In other words, we start to make descriptive and causal inferences. The former is about using observations of the world to draw conclusions about non-observed features of the world. The latter is about explaining what we often think of as the causes of features of the world. We begin to model when we fit these together more precisely. I am going to draw a distinction between formal models, which produce clear testable hypotheses, and non-formal ones (that might be called ‘frameworks’), which do not. We can draw hypotheses out of non-formal models, but their implications are not always so clear. Models become formal when we represent features of the descriptive frameworks by symbols that we are able to manipulate in order to deductively draw conclusions. When formalizing models we face hard choices. We cannot include all the complexity of frameworks, nor the complexity of the full descriptions of reality, let alone the complexity of the world itself. We have to simplify and even to assume that relationships between aspects of the world are not as we know them really to be. We are forced to back our hunches and lay out the descriptive and causal inferences we think are important, in ways that can be inspected, analysed and tested by others. (Note, what we think of as important inferences are not necessarily those we think are the most important overall, but perhaps those that have been ignored or attributed less importance in the past than we think they deserve.) Thus we produce models with definite predictions that we can then test in one way or another, using data gathered from the actual world. Modelling does not have to be formal, though formal modelling is preferable where possible, because it enables us to see more clearly where to look for descriptive and causal inferences, and to test them. Of course, some sections of the discipline of political science do not see formal modelling as preferable. Becky Morton (1999: 41–2) says: One source of puzzlement among formal modelers is the belief by some in political science that nonformal models make less restrictive assumptions than formal models and that, because of this, nonformal models have advantages over formal models in empirical study. Imprecision is argued to have an advantage over precision because ambiguity is assumed to be more general and flexible than exactness … Because formal modelers and non-formal modelers look at nonformal models differently, there is

What Is a Theory?

85

misunderstanding over which is ‘better’. There are two ways to look at nonformal models: as unexpressed formal models or as loose frameworks for empirical analysis. A formal model has a number of stated assumptions. These may be very difficult to empirically evaluate or may be known not to fully reflect reality, but stating them at least allows for empirical evaluation of both the assumptions and what should occur were they fully reflective of the world. The model will provide the elements of the world to the extent those assumptions do represent the world. The implications are deductively produced and hence, if the deduction is formally correct, are known to be valid. Thus one element of debate or conflict is not open to issue. Another theorist may produce a rival model with different assumptions, and different implications. The derivation of the implications is clear in each case, and the two models may be pitted against each other in empirical tests. Of course, translating the formal model into a statistical test, or testing through the systematic collection of qualitative evidence, is not straightforward. Formal models are determinate. They produce predictions. If a descriptive model is to explain anything, it too must produce predictions. Appropriate statistical techniques may analyse the relationships between the variables to test the models’ predictions, but will always have unmeasurable, exogenous elements. Which model produces predictions most in accord with reality may be open to dispute. This may be because both models have captured some correct descriptive and causal aspects of reality. We may then attempt to capture the weight of the variables in the two models – a difficult task, but at least we have a clearer idea of what we are looking for. When modelling either formally or non-formally, we recognize that many factors are left out, but the model is produced in order to derive predictions from those factors that are put in. Many factors, sometimes those recognized to be the most important causes, may not be included in the model, but the model is designed to look at the implications of some factors that are causally efficacious. Decisions have to be made about what to leave out and what to include. Something is always left out. Without oxygen on this planet there would be no policy process, but I have never seen oxygen mentioned in an explanation of any policy outcome. Critics of formalization often believe it simplifies too much. But, as King et al. (1994: 43) point out, ‘the difference between the amount of complexity in the world and that in the thickest of descriptions is still vastly larger than the difference between the thickest of descriptions and the most abstract quantitative or formal analysis’ (emphasis removed). We simplify in order to help our understanding. For explanatory purposes, putting in everything is not only impossible but otiose. King et al. (1994:

86 The Philosophy and Methods of Political Science 29–31) use the idea of ‘leverage’. The more you can explain with the least, the more leverage you have. A complex story explaining a single event uses a host of explanatory variables to explain little – its leverage is low. Qualitative work generally has low leverage; we should try to give it as much as possible by simplifying, picking out key variables and thinking of ways to demonstrate their importance in simple tests. That means we often need to look for tests that are comparative in that they go beyond the case we are studying to similar cases in other policy areas or in the same policy areas in other countries or contexts. By keeping the ideas simple, the fact that other cases differ substantially in other ways matters less. If we can show, for example, that an annual review in a policy area, such as British agriculture, tends to make the policy community tighter, we can make some descriptive and causal inferences. This may be so even though the ‘tightness’ of different policy networks differs widely across the set of cases that have an annual review. All we need to show is that in all (or most) cases where annual reviews were introduced the policy community got ‘tighter’ than in similar communities that did not introduce annual reviews in similar periods. In other words, we show (a statistically significant) correlation between annual reviews and policy community ‘tightness’. Non-formal models have no advantages over formal models if they are simply unexpressed formal models – though until we are capable of expressing something formally, an informal account can be preferable to nothing. As Morton (1999) argues, non-formal models can be very important in providing ideas for empirical analysis. Her discussion is almost entirely concerned with statistical modelling, where researchers have some ideas about empirical relationships but, wishing to avoid being tied down to explicit assumptions, they choose data to analyse using ‘assumption-free’ statistical techniques. Interpreting the results is then an inductive technique and further testing of the claims of that induction, which then produce a non-formal model, a claimed explanation is required. Qualitative descriptive modelling is a form

Box 4.6

Empirical content

King et al.’s (1994) notion of leverage is similar to Popper’s idea of empirical content (Popper 1972a: esp. 119–30; 1983: 233–55; 1989: esp. 217–20, 385–8). According to Popper, the empirical content of a model increases as what is inconsistent with the model increases: thus the more evidence consistent with the model, the lower its empirical content. Hence, according to Popper (1989: 218), ‘if growth of knowledge means we operate with theories of increasing content, it must also mean we operate with theories of decreasing probability’. This entails the need to produce models with more definite, and thus falsifiable, predictions. Vague accounts allow for more evidence to be compatible with them.

What Is a Theory?

87

of loose framework for empirical research. Such ‘theories’ are deductive, but not as tightly argued as formal ones. Formal models constitute what we might think of as the classical tradition of scientific ‘theories’ that can be pitted against each other using empirical evidence. In order to be thus compared they need to produce rival hypotheses. Not all models that are different are rival. Furthermore, when we pit models against each other we need to agree not only over what constitutes empirical evidence for or against each hypothesis, but also over the theoretical terms used in each model. As Popper amongst many others recognizes, our concepts are theoretically laden, and sometimes what appear to be identical concepts are understood quite differently in different models. In Chapter 5 (Section 5.5), I show how two purportedly rival models that seem to produce opposing predictions are actually difficult to test against each other since they conceptualize an important feature of the spatial dimension of politics, whether in traditional left–right or in n-dimensional space, quite differently. Formal modellers can spend a lot of time trying to show something we know happens can occur with assumptions they like to utilize. For example, much work has gone into examining rational voter turnout. The collective action problem is that often it is not in people’s self-interest to take part in collective acts that benefit everyone. A stark example is voting: since one person’s vote is unlikely to change the outcome of an election, and unless voting brings positive benefits in itself, it is not in anyone’s interest to bother to vote. A lot of work has been directed at showing that it is worthwhile for people to vote (Blais 2000; Dowding 2005; Aldrich et al. 2011). Some explores the direct benefits of voting; some highlights the fact that we vote to bring benefits to the community not to ourselves, to make a difference other than ‘changing the outcome of the election’; some re-examines the probabilities of having an important effect, and so on. One cannot help feeling that part of this effort is normative. Political scientists tend to be democratic and think that voting is a good thing; they dislike the idea that it is not ‘rational’, so they try to rationalize it. Since relatively high turnout was thought to be inconsistent with ‘rational choice’ assumptions by some proponents of that method, this has been used by opponents to bash the technique – leading proponents to work ever harder to defend it. We can note here that such attempts to formally model voting do not themselves provide much by way of testable hypotheses with regard to the ‘rationality’ of voting. (Indeed most of the effort goes in the opposite direction: trying to produce a formal model that predicts the turnout we actually get.) The effort to justify the ‘rationality of voting’ is vainglorious in itself, and decidedly normative. Nevertheless some results have emerged from this work that are testable hypotheses: for example, that as the costs of voting go up, turnout declines; as the importance of the election rises, so does turnout

88 The Philosophy and Methods of Political Science (importance here can be measured both by the budget of the relevant level of government and by the ideological gap between the contenders); those who are less likely to be affected by the result (such as those intending to move jurisdictions for local elections) are less likely to vote, though if they care about the area (as shown by social capital indicators) they will still vote, and so on. Such results are important, not because they show some support for one or other theory about human nature or for versions of ‘rational choice theory’, but because they are of interest to those who want to encourage turnout or to discover why it is falling. Formal modelling, then, produces predictions; but if two different models produce the same predictions we have no reason for preferring one to the other, unless the internal workings of one correspond better to mechanisms in the world. How interesting those predictions are is a separate question. The issues here are how far we test hypotheses drawn from models against the evidence, and whether we can only test hypotheses against rival hypotheses implied by rival models. This issue deserves further discussion, which I shall defer to Chapter 5. There is a further important aspect of formal models, raised by Clarke and Primo (2012). They claim, on Duhem–Quine grounds, that models cannot, in fact, be tested. (See again in Chapter 5.) Nevertheless, they argue, models are useful for providing insights into the behaviour of agents in the world. Parties fail to locate at the median voter, as implied in spatial models following Downs (1957), yet such models have taught us much about candidate competition. The fact that voting is not straightforwardly instrumentally rational leads us to ask why people vote, and what features make voting instrumentally rational. Formal models of legislative behaviour might not accurately predict what legislators do, but may generate questions about what legislators are doing; whilst games such as simple prisoners’ dilemmas games focus attention upon what features of humans and their interaction are more likely to lead people to cooperate with each other.

4.4 Explanatory theory: non-formal models Whereas a formal model is a set of statements related formally or analytically to generate testable hypotheses or predictions, a non-formal model is not so precisely specified. It is a set of statements about the world, sometimes a classification, from which we can draw some descriptive and causal inferences. Some non-formal models are more analytic and precise than others. Albert Hirschman’s (1970) argument concerning the relationship between exit and voice is not a fully specified formal model, but he does draw plausible hypotheses from it. He argues that when the quality of a product declines, consumers have a choice of voicing their complaint to the producer or exiting

What Is a Theory?

89

to another supplier. He suggests that various factors will come into play in the consumer’s decision. He points out that voice is a nuanced response, whereas exit is a crude binary choice. His general argument is that, as the costs of exit decrease relative to the costs of voice, consumers are more likely to exit. He then asks whether it is good for public provision to make exit easier and suggests that often it is not. Put this way, the argument is simple, if not simplistic. Its strength comes from its application and the fact it was directed at an issue that at the time seemed problematic. Economists are inclined to think competition is beneficial to society as it tends to improve product quality. Hirschman, an economist, was struck by the fact that the railways in Nigeria grew worse as competition in the form of other forms of transport developed. According to standard economic theory, they should have improved. He postulated that the richest and most educated rail users exited to road transport and took their voices with them. Without their complaints, the quality of the rail service declined. Hirschman argued that we can see the same process occurring in many forms of largely collective service provision. If we make it easier for the richest people to privately educate their children or use private health care, then important voices that keep state education and health provision effective will disappear, leading to a decline in quality. Thus he argues that making exit easier will reduce voice and reduce quality. These seem fairly clear hypotheses. Hirschman’s framework is made more complex by his addition of a third concept, ‘loyalty’. He suggests that some people might not want to exit from the railway, state schools or public health provision because they are loyal to them. The problem for empirically testing Hirschman – and it has proved a problem over 40 or more years – comes with conceptualizing these three concepts, measuring the ease of exit, the amount of competition, and the fact that exit and voice are not strict alternatives (one can do both). Nevertheless, Hirschman’s non-formal model has been modified and empirical attempts made to test the relationships he hypothesizes (Dowding et al. 2000; Dowding and John 2012). Where models are not so formally specified they are not testable as formal models. To the extent they provide predictions (though not perhaps logically implied or entailed as in the case of formal models), then they can be tested and can be considered explanatory theories. Some non-formal models are not really testable at all; they are not explanatory theories, but rather perspectival theories. They generate questions because they frame the world by an argument. One such non-formal model that has been enormously influential in the UK is the Rhodes network model. This model suggests that we cannot understand the policy process without understanding the networks of actors involved. So it provides a perspective from which to ask questions about the policy process in different domains. It does not produce (many) testable hypotheses and does not really stand as rival to

90 The Philosophy and Methods of Political Science other similar approaches such as Richardson and Jordan’s (1979) policy network/policy community framework, the similar framework of Wilks and Wright (1987), or the advocacy coalition framework of Paul Sabatier (Sabatier and Jenkins-Smith 1999) and others (for description and critique, see Dowding 1995a). Some predictions and explanations have emerged from these approaches, however. One of the problems of non-formal models is that they tend to create unnecessary debate. Often, so-called rival models simply portion out the world in different ways, giving the same phenomena different names. As I argued in Chapter 3 (and I shall return to this in Chapters 7 and 9), it is perfectly possible to have different but non-competing ways of conceptualizing the world, where concepts have different but non-competing extensions. If the frameworks are non-competing, then any dispute over terms and the usage of concepts is purely semantic. The disputes are merely verbal (Chalmers 2011). Non-formal models might compete in two ways: in a ‘soft’ sense relevant to their role as an organizing perspective or in the ‘hard’ sense of generating rival hypotheses. They compete as perspectival theories to the extent that one or other might be preferable in terms of the nature and interest of the questions it generates about the policy process. One might lead to more insights into more of the questions we find interesting. They compete as explanatory theories or models to the extent that they produce competing hypotheses about some aspect of the policy process. In Table 4.1, I classify several non-formal models used to explain the policy process in urban politics. We can see that many of them are difficult to distinguish with regard to central questions about the policy process. Where differences exist, this is because they have been applied to different situations and are therefore not rival. The growth-machine model, for example, is intended to answer a specific question about the dominance of development in urban politics (Molotch 1976). The different approaches use different methods. For example, the advocacy coalition framework (ACF) uses both network analysis and specially designed surveys to gauge changing perceptions. But are they rivals? Just because two entries in a box in a given column are different, does not make them rivals. The policy community (PC) and issue network (IN) approaches have ‘low’ and ‘high’ respectively in the box for ‘number of members’ here, but that is because they are applied to different cases. They are not rival explanations of the same case. ACF could be applied to the same empirical case study as a policy network approach (using either of or both PC and IN views). Does that make ACF a rival to the other two? It is not clear. In part it depends on whether they draw different descriptive inferences, produce different causal predictions from the same cases, and the nature of the unstated assumptions. Either way, the answer lies not in a theoretical consideration of any two frameworks, but rather in examination of the elements that make up

91

Resources of actors Number of members Continuity of membership Policy styles across nations and sectors Levels of government

Rules of interaction Type of interest

Linkages between actors Degree of integration Values of actors

Unequal

High

Low

Vary

Several

Equal

Low

High

Vary

One

Distinct

Shared

Distinct

Shared

Not agreed

Variable

High

Agreed

Problematic

Issue networks

Close

Policy communities

Several

Same

High

Low

Unequal

Shared

Agreed

Shared

Close institutional High

Urban corporatism

Several

Same

High

Low

Professions dominate Unequal

Professional values dominate Agreed

Variable

Variable

Professionally dominated sub-system

Several

Vary

Variable

Variable

Unequal

Distinct

Agreed

Distinct

Variable

Variable

Advocacy coalition framework

Table 4.1 Non-formal models of urban politics

One

Some variation

High

Low

Developers dominate Unequal

Not agreed

High for developers; low for others Developers dominate

Finance dominates

Growth machine model

One

Vary

Variable

Low

Unequal

Shared

Agreed

Shared

Close formal and informal High

Regime theory

92 The Philosophy and Methods of Political Science the explanation in each: that is, the entries down the left-hand side of the table. These are the universal features to which, in the final analysis, each supposedly rival model may be reduced. It is on these elements we should concentrate our attention, and by doing so leave aside questions about the ‘best’ explanations. The lesson to be drawn from this comparison is that non-formal models often need to be treated differently from formal models. With formal models we can draw predictions or hypotheses sufficiently precise to see whether and to what extent those models are rivals. Rarely can we produce such strict hypotheses from non-formal models. We can find evidence that is consistent with a non-formal model without producing interesting hypotheses for complete testing. (I explain the difference between ‘testing’ hypotheses and finding evidence ‘consistent with’ them in Chapter 5.) Too often researchers who utilize non-formal models attack, criticize or otherwise deprecate other non-formal models, whilst sparing their own. Unfortunately the evidence rarely decides between the so-called rivals. Furthermore, seldom, if ever, does the author demonstrate that their model produces hypotheses that are inconsistent with hypotheses drawn from other non-formal models. They do not show their purported ‘theory’ provides a better explanation than the rival ones they have knocked down earlier. That is because often they are not explanatory theories but perspectival ones. As such they provide a way of looking at the particular cases the author addresses. They frame questions, suggest avenues to explore and provide a structural base for the study. The less deductive theories are, the less they can withstand the rigours of the hard sciences that many social scientists aspire to; and we should not pretend that they do. We should treat them as what they are: useful devices to help us think about aspects of the social or political world. If a non-formal model is carefully specified, like Hirschman’s, then hypotheses can be drawn out and tested. Usually, however, hypotheses are merely consistent with the model. Often they do not strictly follow from the framework, but derive from general discussion and seem reasonable. To see how interesting and testable any hypothesis is, one can adopt the inversion strategy: take the hypothesis and invert it. Then ask two questions. First, is there anyone who holds the inverted hypothesis? If not, the hypothesis is trivial and virtually any evidence that you are likely to find will be consistent with it. The model might be useful, but not for generating that prediction. Second, if the inverted hypothesis seems as reasonable as the original, would one test it in the same way that one would test the original or would one need different evidence? If the latter is the case, then the inference should be changed from a specific to a more generalized claim. Thus the hypothesis ‘Local community organizations help communities overcome their common-pool problems’ has the contrary ‘Local community organizations prevent communities overcoming common-pool problems’.

What Is a Theory?

93

Box 4.7 The inversion strategy Take the hypothesis and invert it. Then ask two questions. First, is there anyone who holds the inverted hypothesis? If not, the hypothesis is trivial and virtually any evidence that you are likely to find will be consistent with it. The model might be useful, but not for generating that prediction. Second, if the inverted hypothesis seems as reasonable as the original, would one test it in the same way that one would test the original or would one need different evidence? If the latter is the case, then the inference should be changed from a specific to a more generalized claim. Rather than suggesting a hypothesis, the researcher should turn it into a question.

Who holds the latter? I seriously doubt if anyone does. But the hypothesis ‘Local community organizations are better at helping communities overcome their common-pool problems than externally driven solutions’ has the contrary ‘Local community organizations are not so good at helping communities overcome common-pool problems as externally driven solutions’, which is not trivial and has been held by many. Indeed this has been debated for over 40 years in development studies, and probably concluded only recently by the work of Elinor Ostrom (1990, 2001 and 2005) within her Institutional Analysis and Development (IAD) framework. The relevant importance of local versus outside influence on solving common-pool or collective action problems depends on a set of variables, including the precise nature of the problem, long-standing institutional rules (history and conventions) and other environmental factors. Given the cumulative nature of research in this area, any hypotheses drawn should examine elements of Ostrom’s framework to see the relative importance within the cases the researcher is studying. The researcher should always list the hypotheses contrary to those she is interested in. When hypotheses that are contrary to those drawn from a model seem equally plausible, it is preferable to ask a question than propose a hypothesis. So here the better expression is: ‘Under what conditions do community organizations better help communities overcome their common-pool problems?’ A non-formal model is often better at suggesting questions to direct one’s research than suggesting specific hypotheses. Sometimes the inverse is so unlikely or even ludicrous that there is hardly any point in trying to corroborate or confirm the initial hypothesis, as it will be consistent with any reasonable theory you can think of. The inversion strategy is also a neat way of puncturing the rhetoric of politicians. ‘Now is not the time for cowardice!’ thundered the politician. ‘Just when is the time for cowardice?’ asked the late Simon Hoggart, political correspondent of the Guardian.

94 The Philosophy and Methods of Political Science

4.5 Mechanisms and constraints Different models might produce the same predictions. If their point is merely prediction (as Friedman suggested), we would have no reason for preferring one to the other. However, if the models can be differentiated, it is through their working parts. They might have different assumptions or they might have different mechanics – describe a different process. With formal models the assumptions and the mechanics are easier to observe and interrogate, but even with informal models we might tease out differences in assumptions or process. With non-formal models the process is a narrative that tells us how we get to the outcomes via the process. Formal models are also interpreted as a narrative. They have to be: all explanation is couched narratively. The two main ways that models can be interpreted as providing explanations are as modelling a mechanism – the causal process by which outcomes occur – or as constraints upon the set of outcomes. For example, equilibrium solutions in game theory suggest constraints upon the set of expected outcomes. Outcomes will conform to the equilibrium or, in complex games, one of a set of equilibriums. They may not specify the particular process or mechanism by which an outcome is obtained. Models can also describe the mechanism by which an outcome occurs and models with the same predictions can describe different mechanisms. We can also test models by looking at evidence for one or another mechanism. In Chapter 3 I said that a real mechanism is a feature of the world, or a process by which an outcome occurs, and that we model mechanisms by connecting up certain structural features of society together with behavioural assumptions to produce empirical generalizations. I also suggested a mechanism is a narration that makes sense of the data and that can be formalized into a model with strict predictions. Reading that, you might have thought this is not one claim but several. Is a mechanism a feature of the world or a narration? Is it something that makes sense of the data, some description, or is it causal process? Does it describe a structure or causal process, or are these the same? The term mechanism is used in the social sciences in all of these senses. We certainly model mechanisms, and the idea is that the working parts of models correspond to something in the world, but that is still something that we pick out as being explanatory for us about the outcome. We can also think of models as picking out structural features that constrain options. Veto-player theory is about constraints on change, and how different institutional structures constrain the possibilities for policy change. Those features are certainly part of the mechanism through which policy change occurs, but that mechanism might not be considered directly causal of a token change. In Chapter 6 I examine the different ways in which causation

What Is a Theory?

95

is considered in social science; the take-home message here is that what we consider to be causal depends on the relevant research question. Veto-player theory might be causal for type explanation, but regarded as background structural conditions for explanation of token outcomes. Explanation involves narrative; causal explanations involve narrative. Indeed, in an important sense, a narrative, as contrasted with simple description, makes a causal claim. A narrative has the structure of X led to Y, and thus to Z. It is not simply X, then Y, then Z. (History might be one damn thing after another, but the way it is told suggests a logical causal sequence.) Mechanisms are described in sentences, and work as explanations by picking out a structure that is invoked in a causal process, or are themselves narratives of causal process. And we can formalize a mechanism as a model to provide strict implications or predictions. Informal models often also seek to provide predictions and describe mechanisms. I suggest that the most satisfying mechanisms are those that provide structures, often institutional features of society at a subsurface level that explain differing probabilities of outcomes across those structures. They fill the gap between a propensity for something to occur and its actual occurrence. Thus specifying mechanisms satisfies our craving for explanation where statistical accounts leave us wanting more. We might accept simple functional relationships as laws or indeed see them as identity statements, but generally we want something to fill that gap. Where there is an invariant generalization, there is no gap and hence no felt need to fill it with a narration. In politics our empirical generalizations are usually highly variant, so we always have that urge to fill the gap with a narration. We have to be careful with narrations. We are apt to fall for what Nassim Nicholas Taleb (2007) calls the ‘narrative fallacy’, more generally known as ‘the specification problem’ (Dawes 1996). As Dawes explains, the structure of a story is a single sequence of events. These events are generally connected by a set of hypothesized causal links. During the story some evidence for these links might be provided – often rationalizations of behaviour by reasons given by those features of the context the narrator has chosen to emphasize. The problem is one of bias. Without comparisons between the narrated case and others, one cannot see if the causal links adduced in this case are valid. One way of trying to overcome the fact that the narration is only one sequence is to consider counterfactuals; that is, one tries to imagine what might have happened had the circumstances been slightly different. The problem is that we can construct manifold such counterfactuals and it is hard to specify which ones are relevant. Dawes argues that we should only countenance counterfactuals that are themselves based upon statistical generalizations. In his example of an investor gambling wildly on the stock market, basing his reason on the fact he was facing bankruptcy anyway might

96 The Philosophy and Methods of Political Science be justified by the attested generalization that people are usually risk averse except when facing a sure loss (Dawes 2001: ch. 7). Even here, however, the specification problem arises, since there are other pertinent generalizations such as one’s pride in not borrowing money, positive views about financial risks, and so on, which might be equally pertinent to the gambler’s decisions. Only in large-n studies under carefully controlled conditions might we be able to estimate the relative importance of these issues in like cases. In other words the purported cause in the specific case is plausible only given evidence from the large-n case. Again I consider this in more detail in Chapters 5 and 6. So how can we avoid bias in such narrations? We cannot. All we can do is tell a plausible story backed by the best evidence, laying out our sources so others can later reanalyse and look for obvious bias. A really tough-minded empiricist might say that the real problem is our insistence on seeing stories as explanations when all they really are is data. Stories explain nothing: they simply set out data that can be used to provide explanations in large-n studies. I am afraid, however, that the hard-nosed empiricist will be a hypocrite, in the sense I have been using that term; for they will not live their life that way. They will certainly listen to stories and act on them; and will undoubtedly narrate their own life as though the rationalizations they give for their behaviour really are causal. We live by stories; the palaeontologist Stephen Jay Gould once described humans as ‘primates who tell stories’. It is through stories that we casually pattern and understand the world. Our casual patterns can easily be mistaken for causal ones. Such mistakes are rife, but if they were not often authentic we

Box 4.8 The specification problem The specification problem, or narrative fallacy, should lead us to recognize that single narrated case studies only provide plausible accounts or stories of the causal mechanisms that operate. Whilst we can narrate a plausible set of proximate causes to some outcome, the nature of the counterfactual claims involved in any such causal claim allows for several rival narrations. Whilst we can provide evidence that might back any one of them, such evidence will be selected on the dependent variable – it will be selected because it provides such evidence for that particular narration. Whilst the evidence can make the claim plausible it can do no more. Furthermore, such a narration only provides evidence to the extent that it is backed up by large-n evidence. That is, we know from previous similar cases that such things occur. We must thus treat all plausible stories as no more than that, and they can always be trumped by more controlled evidence to the contrary. For this reason, some questions can only be answered by large-n statistical analysis or by controlled experimental evidence. Note that the specification problem can also apply to causal interpretations of simple regressions.

What Is a Theory?

97

would not be so successful as a storytelling primate. Nevertheless, when conducting social science we should distinguish the plausible stories we tell all the time from stronger, more scientific, evidence we evince from careful study. Sometimes plausible stories are all the evidence we can have, and often our narrations support and are supported by generalizations from other sources. We also need to recognize that our interpretations of quantitative evidence are themselves a form of narration. Quantitative evidence also has qualitative interpretation. Nevertheless, the specification problem or narrative fallacy always looms large, and we ought to treat all plausible stories as no more than that, recognizing that they are always trumped by contrary quantitative evidence – if indeed that quantitative evidence is contrary. It is for these reasons that some questions can only be answered by large-n statistical analysis or by controlled experimental evidence. It is a truism that one can learn from history; but actually we can only really learn through the eyes of the modeller, since with so many variables we have to decide which ones to apply to new situations. Qualitative scholars may say that some events are unique: they cannot be studied quantitatively but only understood in all their detail. If that is true, then we can learn nothing from those unique events about future ones, since such a history could only provide lessons for a future event equally and identically complex in all its details. If the uniqueness is to be taken seriously we can learn nothing from qualitative history. But, of course, we believe we can learn from past events. We abstract from them the important variables that applied to those situations and which we think apply to a current situation. If we are right, and we can learn a lesson about a current event from a past one, then we are already performing a quantitative analysis: we are making assumptions about correlations, about causal relationships. We cannot help but perform these calculations: that is a large part of what pattern finding is all about. However, if we can carry out those calculations more formally in well-specified statistical models then we can be more confident in them. The narrative fallacy occurs when we narrate a set of events as though they are causal, because we have to do so in order to make sense of them. But we have no real justification for such narrations beyond the lessons we have learned about sets of similar events, the correlations and causal processes we have seen previously. The more individual a set of events truly is, the less confidence we should have in our narration. Modelling can help such narrations. Deirdre (then Donald) McCloskey in a classic article demonstrated that the standard story told about the failure of the economy in Victorian Britain must be false (McCloskey 1970). The accepted view among historians at the time was that the British economy failed in the late nineteenth century by losing out to the United States and eventually Germany. Failure was typically defined as output growing too slowly due to sluggish demand, inept entrepreneurship causing stagnating

98 The Philosophy and Methods of Political Science productivity, and too much investment overseas because of the imperfection of capital markets. Using a straightforward production function model (exploring the inputs of capital, labour and total factor productivity, with the residual derived typically from a Cobb–Douglas production function), McCloskey demonstrated that the British economy grew as rapidly as the supply side would allow. In other words, by looking only at aggregate demand historians had missed the point. Viewed from the new perspective, the USA and Germany were bound to overtake the UK because they had larger productive potentials on the supply side. By using generalizations derived both theoretically (formally modelled) and empirically (from other situations) and applying them to the events under study, McCloskey demonstrated that the previous narration was faulty.

4.6 Cumulative and non-cumulative research An expert is usually defined as someone whose judgement, through experience and learning, is trusted. But some expert judgements have been shown to be little better than random (Gawel and Godden 2008; Hodgson 2008, 2009). In their joint work on expert intuition Gary Klein and Daniel Kahneman realized that expert intuition worked best in environments which are sufficiently regular that the agent can learn the regularities and receive regular feedback on their judgements (Kahneman 2011: ch. 22). In other words, when pattern recognition is easy and you can check whether your predictions turn out to be correct, expertise can build up. Some aspects of political life are regular enough to enable expertise to develop – the study of elections in stable environments allows reasonable forecasting ability (LewisBeck 2005); but political experts are notoriously poor at predicting dramatic changes in political regimes precisely because the environment is not regular. The stochastic nature of politics often makes it hard to judge just how wrong political experts are in their judgements (Tetlock 2005). Some research is cumulative and some not. Cumulative research comes about when the environment – that is, the background conditions and assumptions – is stable and when it is relatively easy to check the results. Mathematical modelling in political science is a cumulative enterprise, since the methods are largely agreed by all practitioners and it is relatively easy to check whether those results have, in the first instance, validity and, in the second, application to the real world of politics. In other areas of the discipline acceptable methods are less secure and where predictions are vague it is less easy to utilize feedback to check claims. Note, though, that cumulative research need not be interesting, nor even pertinent. Whilst formal models can build on each other, it does not follow that they teach us any more about politics. One can accumulate an awful lot of trivia. Nevertheless we should

What Is a Theory?

99

try to work in cumulative areas of the discipline so that our knowledge can grow. Otherwise political scientists will become experts on politics much as wine tasters are on wine. They know what they like and people will listen to them to learn those preferences. One area of cumulative research in political science is the effects of electoral systems on party systems. For a while scholars became embroiled over ‘Duverger’s Law’, Maurice Duverger’s (1954) hypothesis that single-member plurality districts will tend to generate two-party competition, whilst more proportional systems would generate multi-party competition. Duverger suggested a mechanical effect that translated votes into seat shares, and a psychological effect that led voters to choose major parties, reinforcing the mechanical effect. Initially there was some debate over the mechanics of these electoral system effects relative to cultural or historical elements (Wildavsky 1963; Riker 1976, 1982a), then discussion of the workings and counterexamples (Grofman and Lijphart 1994; Gaines 1999; Diwakar 2007; Grofman et al. 2009). There was a great deal of conceptual analysis of how we measure the number of parties and the proportionality of an electoral system (Taagepera and Shugart 1989; Dunleavy and Boucek 2003; Taagepera 2007; Grofman and Kline 2011). Whilst there is still disagreement over these issues, the debate has progressed beyond simple ways of trying to confirm Duverger’s empirical regularity in its various guises. The effects of the electoral system in terms of proportionality and district magnitude (the number of representatives elected to the same legislature from any district or constituency) affect the nature of the party system; as does whether the system is parliamentary or presidential (Jones 1995; Persson and Tabellini 2005). However, the relationship is complex. It is mediated through the strategic considerations of voters and representatives, whose strategies are affected by other political and social factors such as social heterogeneity, and the political history and the culture of a country. Rather than seeking straightforward empirical regularities, work in this area now looks for mechanisms through which the effects are mediated. As in all social behaviour, our knowledge, both theoretically derived and from working within such mechanisms, can then affect the subject; political actors themselves design electoral systems for their own advantage (Colomer 2005). This cumulative research programme, by no means complete, has progressed from descriptive inference – Duverger’s original conjecture – through comparative analysis, with claims of confirmation and disconfirmation of an empirical generalization; conceptual analysis and redefinition, more measurement and claims of tests; debate over those concepts and further measurement; and sets of case studies examining the specific workings of token actors. It has involved game theory, historical analysis, case-study analysis, conceptual analysis and much debate. The debate has utilized theoretical concepts such as the ‘effective number of parties’ (the number of parties weighted by

100 The Philosophy and Methods of Political Science the relative strength) and ‘proportionality’ measured by theorized indices and the actual number of parties and relationships between votes and parties in legislatures. The measures lie under the surface in order to try to understand better the surface reality we see. (It is for that reason that some of the debate over the theoretical concepts might never be satisfactorily resolved, though in truth rival measures make little difference to conclusions.) The impact of electoral laws on political parties, voter behaviour and party systems is an ideal example of the cumulative way in which political scientists go about their work. Definite conclusions have not yet been reached, but we know a lot more about the interactive effects of electoral systems, parties and voters than we once did (Grofman 2006). Furthermore, debate has moved on (in the best work anyway) from considering whether or not Duverger’s empirical generalization has been confirmed, to examining the deeper forces at work, and thus to attaining more invariant generalizations about the effects of electoral systems on parties and on voters.

4.7 Conclusion The term ‘theory’ is used in many different ways. I have distinguished perspectival theories from explanatory ones. The former consist largely of organizing perspectives that lead us to ask questions in certain sorts of ways. The organizing perspectives are not falsifiable themselves, but within them they have explanatory theories or models that produce predictions that are. We can suggest models are falsifiable to the extent their predictions are so; and perspectival theories are more or less interesting in the sense that the models contained within them produce interesting and testable predictions. Formal models are theoretically closed and produce strict predictions. Non-formal models are not closed and hence contrary predictions are often compatible with non-formal models. The harder it is to produce rival predictions from non-formal models the less explanatory they are. To the extent that nonformal models do not produce predictions they are perspectival rather than explanatory, useful for leading us to examine the social world in one way rather than another. Models differ in their internal workings as well as in predictions. These are intended to specify mechanisms that causally explain outcomes. Some general models, rather than specifying the precise mechanism that determines token outcomes, rather describe structures that constrain the class of outcomes. Predictions need not be deterministic. Multiple equilibriums suggest classes of potential outcomes. Indeed few models in the social sciences produce point predictions, hence few fully determine outcomes, and thus do not comprehensively describe causation. Depending on how we view causation, we might see models that determine types of outcomes as producing causal

What Is a Theory?

101

explanation, or we might sooner see them as the background structural conditions through which the token causal explanation is narrated. Formal models can demonstrate theoretically that purported explanations are invalid. The importance of formalizing models and explanations as much as possible is that it allows research to be cumulative. Cumulative work often goes down side-alleys, but as researchers build on previous work, modifying models, collecting more data, redefining concepts and re-examining explanations, a stronger and better picture emerges. This can occur where models are not fully formalized, but if it is not clear how models diverge in their predictions, their concepts and their internal workings, research does not build but rather becomes a series of verbal standoffs.

Chapter 5

Hypotheses and Theory Testing

5.1 Introduction In Chapter 4 I distinguished many different ways in which the term ‘theory’ is used, dividing them into two main forms: ‘perspectival’ and ‘explanatory’. Explanatory theory, if it is at all useful, must produce predictions in the sense in which I defined them. My definition of a prediction is broad, but when we have formal models we have more precise predictions as deduced explicitly from a model. Non-formal models are suggestive of such hypotheses. In this chapter, when I use the term ‘theory’ I mean model: for strict hypothesis testing, a formal model; for ‘suggestive hypothesis testing’, non-formal models. Where it matters whether we have a formal model I will use the term ‘formal model’. Remember that models are supposed to demonstrate the working parts of mechanisms. Laws or ‘law-like generalizations’ are the invariant aspects of mechanisms (subject to quantum effects at the highest granularity – see discussion in Chapter 6). Mechanisms tend to produce certain types of outcomes, so the outcomes of mechanisms (the predictions of the model that specify them) are empirical generalizations. Empirical generalizations are not invariant (or as invariant) to the extent that (1) the conditions of the model do not fully obtain (the mechanism is subject to other forces not contained in the model of it) or (2) many models produce multiple equilibriums so predictions themselves are probabilistic. In this chapter I will eschew many of the complications of interpreting testing hypotheses. We have enough work to do understanding what counts as a test under different versions of the proper scientific method. So this chapter is about empirically testing theories. Some think that simply means seeing whether or not a purported generalization holds. Not all are clear about whether the generalizations of which they write are law-like or empirical. I will discuss the issues concerning interpretation of testing generalizations. I see such testing as examining the predictions or hypotheses that are deduced formally or informally from a model and then collecting and analysing data to confirm (corroborate) or disconfirm (falsify) the hypotheses that then reflect on the quality of the model. A slightly different process occurs when we consider machine-learning techniques, ‘big data’ and

102

Hypotheses and Theory Testing

103

complexity theory, though I will argue that the logic of explanation is not as different there as often imagined. In Chapter 6 I will look again at ‘testing’ models, particularly in relation to certain assumptions made in quantitative modelling touched upon in this chapter. In the philosophical literature there is a great deal of discussion and controversy about the logic of this testing process, which I will not attempt to sort out here, though I will draw attention to misunderstandings prevalent in many texts in the social sciences. In fact, I will steer a rather dangerous course, for it seems to me that some of the philosophical or logical issues that surround the confirmation/corroboration issue, whilst important philosophically, are not of much consequence to most empirical researchers and, where they are, they concern our interpretation of what we find rather than how we go about finding it. And so I suggest empirical researchers should not worry too much about them; I will do some of your worrying for you in this chapter. Others are of some consequence, however. Rather than concentrating solely upon the problems of confirmation and corroboration, I will try to draw lessons for the empirical researcher from what different philosophers were trying to achieve. More controversially, I will suggest that empirical researchers should keep in mind some distinctions that will enhance their work, although they cannot ultimately be philosophically justified. However, those philosophical considerations, I will argue, should be weighed pragmatically and not sceptically. Indeed, unless we put a stop on scepticism somewhere, we are forced to conclude that we can have no knowledge of anything. The working political scientist perhaps need not be that bothered about some of the philosophical issues discussed here. We have an intuitive sense of what constitutes predictions drawn from models or, less formally, from non-formal models and can suggest what evidence would tend to support those hypotheses. One only needs to worry when facing sceptical challenges and this book is designed to help respond to some of them. In empirical contexts it is rare that the testing of hypotheses is challenged on philosophical grounds; but when this occurs, that philosophical background is often not well understood. The major challenge is usually based on some variant of the Duhem–Quine holistic thesis that empirical disconfirmation can always be challenged by shifting understanding of the theory (Quine 1953, 1960; Duhem 1954). This philosophical point is a good one; but there are plausible and implausible shifts in theory. I will discuss these with examples. A bigger problem, perhaps, is the misuse of induction and the relationship between inductive confirmation and deductions from models. The problem of induction has bedevilled philosophical debate since David Hume (1742/1978). Hume argued that we have no rational reason for relying on induction, though we use it all the time in our everyday lives and in science. (Indeed the pattern finding which I suggest is the key element of all description, prediction and explanation relies upon induction. The very use

104 The Philosophy and Methods of Political Science of naming objects and processes is an inductive one.) For anyone wanting to defend the rationality of science (as opposed to mysticism or simple assertion), the problem looms large. The big debate (confirmation versus corroboration) is over whether Bayesian confirmation theory is justified or whether we need Popperian falsification. Popper grandly announced that he had solved the problem of induction, not as such, but simply by showing we do not use it; rather, we deduce and then corroborate or falsify. The majority verdict is that Popper failed to demonstrate this: corroboration does require induction. Indeed, as I have suggested several times in this book, a lot of our ideas are induced from what we see. The very patterning of the data could be seen as induced for our predictive purposes. Nevertheless, in political science Popper’s view of good scientific method is not well understood. Most accounts of his idea of falsifiability in secondary literature are more like the confirmation theories he was criticizing than Popper’s own ideas. The fact that Popper’s own ideas fail (and arguably, in the end, collapse into confirmation) is not an excuse for the confusion. What is important in the account of falsifiability is what Popper was trying to achieve. The important lesson I want to draw from Popper’s work is that you cannot beat something with nothing. If the evidence does not support a hypothesis drawn from a model (or explanatory theory) then, unless you have something to replace it with, you do not give up the theory yet. The more reasonable action is to modify the model to encompass the negative result you found. That process is dangerous, of course, and is precisely where proponents of the Duhem–Quine holism thesis criticize scientific method. To some extent, the Duhem–Quine thesis demonstrates the superiority of formal models over non-formal ones. When a model is well specified, altering it to enable a changed hypothesis requires deductive work that might produce new (or other modified) hypotheses. This gives the researcher something new to test and we can see clearly how they are responding to the evidence. For less formalized models, what is being modified and what effects that has on other hypotheses and potential tests is less clear. With non-formal models, encompassing changed hypotheses is relatively easy and might involve merely shrugging the shoulders. Contrary to what some researchers seem to think, having less formal models is not an advantage, and does not avoid the Duhem–Quine holism problem of theory testing. Quite the opposite: the less formal the account, the more there is for Duhem–Quine to feed upon. A second lesson I want to draw from Popper, though he is certainly not alone in pressing it, is that evidence is theory-laden. It is for that reason he claims we cannot test an explanatory theory or model against the evidence, but can only test such theories against each other given the evidence. In fact, I think this teaches us about the abuse of induction, a well-known abuse: assuming a process will continue just because it has in the past. The turkey

Hypotheses and Theory Testing

105

thinks it will be fed today, because it has been fed every day prior to this; but today it is killed for Christmas dinner. Understanding the process (why people feed turkeys) can lead us to the inference that one day the turkey will be not fed but killed. Theory helps us interpret the evidence (the turkey might ask itself: ‘Why do these guys keep on feeding us? What’s in it for them?’). Theory here does not solve the problem of induction, but it can help explain the turkey’s mistake. I am suggesting merely that explanatory theory can constrain the manner in which we use induction, not solve the problem altogether. Hume’s problem is more fundamental. He asks if we can prove that what we have observed happening in the past will continue in the future. This includes fundamental beliefs such as that objects will continue to be identifiable as they have in the past; that our concepts will not change even as we dare speak their name; that natural laws (law-like generalizations) or mechanisms will not alter at some point in time, and so on. It is no good saying that the evidence that has piled up from the past gives us no reason for not thinking that the universe will continue as before, any more than the turkeys’ evidence that they would continue to be fed, since that evidence relies upon induction. Unlike the turkey, we could have a theory that it will not, but what is the empirical support for the theory? The turkeys could have evidence about what has happened at Christmas in the past, but again, why should that continue as before? Fundamentally, the problem of induction is simply that it is not deduction. The deep problem of induction is something that practical researchers do not need to concern themselves with. Leave it to the philosophers. What should concern practical researchers is the theoretical justification of the inductions they use to generate aspects of the explanatory theory or model they are testing. Nevertheless, they should be aware of the formal problems of testing hypotheses. First, there are important differences between finding evidence that is consistent with a hypothesis and one that actually confirms or corroborates it. Second, we should be aware of the real problems that exist, not just of some shadowy difficulties alluded to in the literature. Third, it is important to understand that some questions can only be answered by certain types of evidence and not by evidence of a different nature. Some questions can only be answered by quantitative evidence. Case studies can suggest mechanisms and processes in the token example they are about. They can suggest proximate causes, though they suffer from the specification problem (introduced in Section 4.5; see Box 4.8). Case studies only provide a data point for type and ultimate explanation. If you are searching for ultimate type explanations, you need to go beyond single case studies. For some questions you need to design the appropriate research. If you only want to do qualitative research, then you need to restrict the questions that you set out.

106 The Philosophy and Methods of Political Science

5.2 How evidence bears on theories: preliminaries I want to suggest that there are three ways in which we can think of evidence bearing on hypotheses and thereby models, though if interrogated closely these distinctions might not be sustainable. The evidence might (1) confirm (disconfirm) the hypothesis; (2) corroborate (falsify) the hypothesis; or (3) be consistent (inconsistent) with the hypothesis. The following definitions are provisional, since the account given here will be made more complex later in this chapter. Note that the antithesis of each of the three ways in which evidence bears on hypotheses is not always symmetrical. 1 Evidence that confirms a hypothesis increases the probability that the model from which the hypothesis is drawn is true. Evidence that disconfirms a hypothesis decreases the probability that the model is true. Disconfirmatory evidence demonstrates that a general law or invariant generalization is false. (I shall put caveats on that below, so consider this the ‘pure case’.) 2 Evidence that corroborates a hypothesis increases the odds that the model from which the hypothesis is drawn is true relative to a rival model that produces a contrary hypothesis. Evidence that falsifies a hypothesis decreases the odds that the model from which the hypothesis is drawn is true relative to a rival model that produces a contrary hypothesis. (I use the term ‘odds’ to signal that the probability is relative to some other model, not the overall probability that the model is true. I will put a caveat on ‘odds’ below and discuss whether corroboration is really different from confirmation.) This second category is Popperian. Some readers might be startled to read that, according to Popper, falsifying a hypothesis does not falsify a model (or ‘theory’) (although many scholars have attributed this view to him). Whilst Popper sometimes discusses falsifying as though it shows something to be false, he argues that nothing is certain. Furthermore, his formal definition of corroboration and its antithesis, falsifiability, clearly shows that he thinks falsifying only ever reduces the odds of one theory (model) being true relative to another (Popper 1989: Addenda 1–3, 385–96; 1983: 217–55). He is adamant that these odds cannot be reduced to a probability that a model is true and that they are different from probabilities in accounts of confirmation. I shall discuss the serious problems with his account below. 3 Evidence that is consistent with a hypothesis does not increase the probability that the model from which it is drawn is true. It has no (practical) effect on the probability that a model is true or false. Evidence that is inconsistent with a hypothesis does show that the model from which it is drawn must be false (in the ‘pure case’). Evidence consistent with a hypothesis does not really test it. It shows that the world is consistent with

Hypotheses and Theory Testing

107

the hypothesis but does not really do anything else. However, if evidence is inconsistent with a hypothesis, we might think the model from which it is drawn must be false. The asymmetry I point to here might make the idea of ‘consistent with’ without ‘confirming’ nonsense. Strictly or logically speaking, the idea of ‘consistency’ that is not confirmation is indeed problematic. I will explain later why I want to add this third way of thinking about examining hypotheses.

5.3 Confirmation, induction and theory Confirmation is associated with positivism and the DN model of explanation we briefly reviewed in Chapter 3. The DN model of explanation of a singular event has the form of a deduction with a general law, a singular statement and a prediction or hypothesis. However, confirmation of the generalization is inductive. Confirming the generalization means here that the probability that it is true has increased. Whilst the hypothesis is deduced from the generalization (and singular statement), confirming the generalization increases the probability it is true through induction. So if the generalization is ‘all swans are white’ and we find a bird that is a swan and is white, that evidence confirms the generalization, increasing the probability that it is true. One of the problems of confirmation here is that it only takes one swan to be black for the generalization to be shown to be false. Each positive case slightly increases the probability; any negative one destroys it completely. Furthermore, we can never be sure that there are a finite n number of swans to be found, or that we have found them all, so we are never certain the generalization is true. Our belief in it might approach certainty, but will never attain it. There has been a great deal of philosophical debate about the rationality of induction: that is, whether or not we can really rationally believe such generalizations based on incomplete evidence. There are a number of solutions, though none seems satisfactory (Howson (2000) is a good survey of solutions and their problems). Yet induction is something we do all the time. We find patterns in the data that we expect to continue in the future. The whole basis of conceptual naming is that the pattern so named is a continuing one, unless it is explicitly time-dependent. Some justifications of induction are based precisely upon such Darwinian accounts. Induction is justified because it works. However, whilst the Darwinian account can explain why we do it, and can justify our continued use on those grounds – ‘it works’ – it does not answer Hume’s problem of whether we are fully justified in using it. The problem of induction is whether we are justified in considering the patterns we find will continue, however long they have continued in the past. Past success can only justify our belief in future success because it increases the probability

108 The Philosophy and Methods of Political Science by confirmation, but confirmation cannot justify the confirmatory process; induction cannot justify induction. The turkeys expect to be fed each day, but one day they are killed for Christmas dinner. The pattern they expected to find was ended. Evolution does not help here. The turkey evolved and survived (with some help from humans), whilst the dinosaurs, the most successful complex creatures in terms of longevity on the planet thus far died out; possibly because their inductive inferences did not see Christmas coming in the form of a comet. Success in evolutionary terms is only relative to other species; science aims at more than that. The biggest problem for the empirical (or logical) positivist and the DN model is that generalizations are, somehow, only supposed to be empirical. We are not supposed to generate them theoretically, but to build them up from our observations. The generalization is simply an empirical claim based on the evidence. However, when I provided an initial definition of confirmation above, I did so by suggesting that the evidence confirms the model from which the hypothesis is drawn and is not simply a generalization. In that sense I did not set it up as an empirical positivist would. A model is more than a generalization. It has theoretical coherence. It deductively narrates some reasons why the hypothesis must be so. And that is one way we can try to avoid the problem of induction. We do not simply induce generalizations or patterns from the data. Rather, our model of the data gives greater credence to those patterns through its theoretical form. Science does not (or should not) claim that all swans are white because every swan we have seen is white, but because of some reason (model, mechanism, set of inferences) why all swans must be white. If we have a theory that says why a given type of bird must have plumage of a given colour, then we have more than simple past evidence to induce the generalization that all swans are white. The example of ‘all swans are white’ is not a good one here (which is why it is a favourite example of the problem). It is difficult to think of a theory of why birds of a particular type must have plumage of a certain colour. However, there are theoretical reasons why birds’ plumage might vary in colour under different circumstances. There are theoretical reasons why birds of prey tend to have light-coloured under-plumage, whilst ground feeders are often speckled. A better example of non-contingency is striped and spotted animal fur. There are developmental reasons why spotted animals can have striped tails, but striped animals cannot have spotted tails. The generalization ‘on planet earth no striped animal has a spotted tail’ is not derived from simple observation but from theory. It is based on a mathematical model of pigmentation through chemical reaction, forming stripes or spots at different stages in foetal development. It is possible, then, to see that we can produce generalizations about animal colour that are driven by theory as well as based upon prior empirical evidence. So we might imagine there could be a theory about why all swans

Hypotheses and Theory Testing

109

Box 5.1 Theory not induction: spotted bodies and striped tails Some animals, such as the cheetah and the leopard, have spotted bodies and striped tails. However, no animal has a striped body and a spotted tail. This might be thought to be a simple empirical generalization – if we keep searching, one day we will find a counterexample. However, theory suggests not. The mathematical model of pattern formation indicates that animal coat patterns depend only on the size and shape of the region where they are formed. Thus patterns are dependent upon the size of the embryo at the moment different chemicals react and form patterns on the skin through the stimulation of melanin. Given the stage at which tails are formed and their shape, the mathematics shows that it is possible to have spotted bodies and striped tails, but not striped bodies and spotted tails. In fact the model is hard to test since the chemical substances are difficult to detect experimentally. Nevertheless, the empirical generalization is not produced simply through induction but theoretically. (For more on this, see Murray 1989.)

are white. In that case, if we find a black swan, then we have to relinquish those theoretical reasons alongside the generalization. This might give us a choice. We might give up the generalization (and the model that supports it) or suggest that, despite appearances, this black bird is not a swan, even though it looks like one, because theoretically it cannot be so. In much scientific analysis this is exactly what happens. Empirical evidence that seems to show a model to be false is used to reinterpret some of the concepts within that model. Think of another case. Say your model predicts the (naively) strict generalization that ‘all members of the working class vote for socialist parties’. You can confirm this each time you find a working-class voter who does indeed vote socialist. Conversely, if you find a worker who votes for a conservative party, then you disconfirm the hypothesis. You prove it is false. However, you might probe further into why that worker voted conservative and try to save your generalization. Maybe the worker made a mistake: she meant to vote socialist but put her cross in the wrong box. Your model assumptions do not allow for mistakes. So you modify your generalization to ‘all working-class people intend to vote socialist’. Or perhaps the description ‘voted for the conservative party’ mis-describes what she was doing, since her actions are best described as ‘voting for the candidate she is related to’, and your model does not allow for that. Or perhaps you find this worker is also a team leader and hence a supervisor of other workers, and so is also a manager. And your model does not allow for voters to be both workers (under one description) and managers (under another). Once you have a model with assumptions, then the precise conceptualization of ‘worker’ and of ‘voting for party x’

110 The Philosophy and Methods of Political Science might mean that we can argue that evidence that seems to disconfirm the model does not. This makes some people sceptical about such scientific testing, because it seems evidence can ‘always’ be reinterpreted to save models. But although this should lead to healthy scepticism about any particular claim or model, it should not lead to scepticism of modelling in general. Some shifts of definitions are reasonable and in the spirit of the original model; others are not reasonable, or change the model so much that it moves closer to the claims of critics of its original form than to its defenders. Such shifts might be seen as special pleading by those who do not want to admit to being wrong. Some shifting of definitions and categories can be justified and some not. All such shifts change the theory – we hope in a more plausible direction – and we get closer to determining the real patterns in the world. Each claim needs to be examined on its merits, rather than discarding altogether the model–testing– changing method. In all fields some research is superior to others. The upshot is that testing hypotheses becomes much messier when you recognize the relationship of theory to the inference. It does not follow, however (as we see below), that one can never disconfirm. The prediction that 58.6 per cent of working-class voters vote socialist is confirmed if precisely 58.6 per cent do indeed vote socialist, and disconfirmed otherwise. Disconfirmation means that the hypothesis, and therefore the model, is false. Confirming only increases the probability it is true, disconfirming demonstrates that it is false. Any doubts about disconfirmation could only enter at the stage of querying the veracity of the evidence. Despite the problem of theory complicating the issue, confirmation and disconfirmation seem relatively easy with such point predictions. However, there are virtually no point predictions in political science. Rein Taagepera (2008) makes a strong case that we should look for point predictions. For him, most statistical analysis in political science examines how variables interact and is thus descriptive. To turn them into ‘why’ questions, we need to model the processes formally. Only answers to ‘why’ questions are truly explanatory. These models will place parameters on potential outcomes and reduce the set of expected possibilities, ideally to point predictions. There is much to learn from Taagepera, but I do not think that description should be juxtaposed with explanation quite so starkly. And I fear that whilst there might be some mechanistic processes, including the relationship of electoral systems and party systems that Taagepera has studied so carefully that can be modelled to give predictions with restricted scope, in some areas of social science the parameters will always be broader than Taagepera allows. What we can do is to chase down conceptually to try to reduce the scope of our analysis. Take the simple prediction that ‘working-class voters are more likely to vote socialist than non-working-class voters’. This hypothesis also looks

Hypotheses and Theory Testing

111

relatively easy to confirm. Even if only 4.5 per cent of working-class voters vote socialist, as long as that is more than the proportion of the non-workingclass category then the hypothesis is confirmed. Every time evidence emerges that working-class voters are more likely to vote for socialist parties than nonworking-class voters, the probability that the hypothesis is true is confirmed, increasing the probability that it is true. The problem is with the claim ‘every time evidence emerges’, since great debate can occur over that evidence: ‘What does “working class” mean?’ ‘What constitutes “socialist” parties?’ ‘How do we handle the special cases [such as the examples above]?’ And so on. Even if we confirm the hypothesis that more working-class people vote socialist than non-working-class, if very few vote socialist anyway, that hypothesis might be considered uninteresting. You can confirm all sorts of trivial claims. It might be more interesting to discover why the 4.5 per cent of working-class people who vote socialist do so, rather than the fact that working-class people have a greater tendency to vote socialist than other voters. For example, you might find that virtually all those who vote socialist are workers in heavy industry or blue-collar public-sector workers. You then need to investigate what underlies that finding, which might be much more interesting than any claim about ‘working-class’ voters. We might conclude that we are patterning the world of people the wrong way. It is not ‘working class’ and ‘non-working class’ but ‘heavy industry’ and ‘blue collar public sector’ that are the interesting categories with regard to voting behaviour. This might tell us something about changes in the working class (beyond simply that it is no longer an interesting category), but it is the new categories that we should concentrate upon. Again, the evidence reacts with the theory to change the model and make it more interesting. Indeed in such a case we should be interested in the mechanism that underlies why ‘heavy industry’ or ‘blue collar public sector’ is now important where ‘working class’ once was. Is the mechanism a new one? Or is the mechanism that once performed for working class now performing only for the new category? What has changed about people to make the mechanism work only for these new categories? Once we have worked these things out, we can then perhaps identify what factors should drive our patterning of voters across all places and times, classifications that might only partially correlate with more everyday surface descriptions of people. This sort of theory development belies the quantitative–qualitative distinction in social research. Evidence that leads to theory development often relies upon our understanding of the mechanism that is thought to generate the hypothesis in its original form, as modified (which might rely on new thoughts on the mechanism) or as applied to both. Mechanisms are interpretations or narratives that form explanatory bridges between theories, hypotheses and evidence. Sometimes, as we shall see, as studies develop along these lines, critics complain that disconfirmation goes out of the window. As the results come

112 The Philosophy and Methods of Political Science in, the definition of the important dependent variables shifts to save the hypothesis. This can be a problem, but is not universally so. Shifting the definition of the important dependent variables to save the hypothesis is good science if that shift reflects the important underlying patterns, if it reveals the mechanisms that produce the outcomes we witness. Of course, these shifts are often messy, and much theoretical as well as empirical debate lingers over any such modification, but models need to change with the evidence. We should not be embarrassed about that. I return to some of these issues in Chapter 8 when I discuss conceptual analysis. In a recent book Clarke and Primo (2012) argue that we cannot test formal models by using data analysed in empirical models. Part of their argument is a variant of the Duhem–Quine thesis that evidence is itself theorized and we can save models by changing other assumptions. So the data used to test hypotheses from formal models are themselves interpreted through the assumptions of empirical models. That is of course true, but this philosophical doubt needs to be considered in light of the practical reality that all evidence, including simple visual evidence, suffers from this problem; however, the dubiousness of our assumptions when interpreting the data varies massively across empirical models. To be sure, we need to assess evidence very carefully; but that, I take it, is part and parcel of empirical political science. Another aspect of Clarke and Primo’s argument is that it is a logical mistake to think you can test a model by showing evidence that is consistent with the hypotheses drawn from it. They argue that to confirm a model M by showing that hypothesis H drawn from it is supported by the evidence is to commit the fallacy of affirming the consequent. (See Box 5.2). Evidence that confirms hypothesis H supports model M to no greater extent than it supports any model M’ from which hypothesis H might be drawn; and there might be an infinite number of such models. In other words, all tests of hypotheses are only consistent with models from which the hypotheses are drawn and have no effect on the probability that they are true. This is the problem of induction again, in the sense that affirming the consequent is a deductive fallacy. Clarke and Primo’s challenging inferences on the basis of their deductive failure is simply telling us that inductions are not deductions. Box 5.2 Affirming the consequent If P then Q Q therefore P If McTavish is English, then he is British. McTavish is British, therefore he is English. But McTavish, as his name suggests, is a Scot, and so he is not English.

Hypotheses and Theory Testing

113

The difference between confirmation and consistency involves considerations of the relevance of reasons for testing M (rather than M’), and that is related to the Duhem–Quine thesis. I return to this issue below where I challenge the relevance of Duhem–Quine for empirical researchers. We noted above an asymmetry in that a single case logically can completely disconfirm a generalization and so throw doubt on a model predicting it. However, it can only increase the probability that a generalization is true without confirming it fully. In reality the world is not that simple. Supposed disconfirmations might be problematic, because of (a) measurement problems and (b) the fact we can shift the concepts involved in a theory. Some people claim, however, that a single case can prove a theory. They assume that the case will confirm a specific point prediction drawn from a theory and that the case is such that, if it applies there, it will apply to every case with those characteristics. It assumes strict universal laws. One case that is often invoked by political scientists to show that confirmatory case studies are possible (for example, van Evera 1997: 66–7; Gerring 2007: 117–18) is Eddington’s astronomical observation confirming Einstein’s general theory of relativity. The story in these texts claims that only Einstein’s theory predicted that gravity would bend light by a specific amount and this was shown at a full eclipse. The test showed Einstein was right, as the measurements got the predicted result. The crucial test confirmed the theory. It did, in the sense that those who believed Einstein’s theory maintained the test demonstrated it, and subsequent tests supported the theory. However, the details of the crucial test case are somewhat more ambiguous (see Box 5.3), and illustrate the difficulties of interpreting empirical evidence, even with point predictions in the natural sciences. We can see from this so-called crucial case study the problems that are involved in measurement and interpretation of those measurements. But what is important here – and which is rarely found in political science – is that Einstein’s predictions were point predictions; they were made prior to collecting evidence; they were perhaps surprising; and they were backed by a substantial body of theory that unified other findings and included many other predictions. We also note that the law-like generalization of the deflection of light is a prediction generated from a theory that produces many other predictions, some known prior to the theory being formulated (leading and helping Einstein’s intuitions), others emerging from the theory itself. Whilst we can imagine confirmatory crucial case studies in the sense we can imagine surprising point predictions that a case study could show, thereby making the model seem (more) plausible, in reality measurement error and controversies over interpretation of observations (whether they are subject to mathematical or statistical manipulation or not) will always ensure that such case studies simply provide evidence for one side or the other. The problems are much greater when we do not have point predictions, and

114 The Philosophy and Methods of Political Science

Box 5.3

Eddington’s test of general relativity

In 1911 Einstein predicted that light would undergo a deflection of around 0.86 arc seconds when it passed the sun (Einstein 1923/1911). Had the 1914 attempt to measure it by an eclipse seen in Russia taken place, Einstein’s theory as it existed at that time might have been disconfirmed. However, the First World War intervened. By 1915, Einstein had formulated new equations predicting a deflection of 1.7 arc seconds (Einstein 1961/1916). Arthur Eddington set up two teams to measure the deflection in the eclipse of May 1919, one in northern Brazil and one led by Eddington himself in Principe, a small island off the Atlantic coast of Africa. The Brazil team had two instruments: one measured a deflection of 1.98 arc seconds, the other 0.86 but with a higher margin of error. Because of poor conditions, Eddington’s own plates showed few stars; he performed a series of complex calculations to extract some data that seemed to show a deflection of 1.6 arc seconds. Eddington discarded the lower reading from Brazil and averaged out the other two readings to announce confirmation of Einstein’s prediction of 1.7 arc seconds. Not exactly a pristine crucial case study, and the results proved controversial (Isaacson 2007: ch. 11). Indeed it is not unrealistic to suggest that only those who already believed general relativity on other grounds accepted Eddington’s results. Few sceptics were convinced. Of course, the predictions of general relativity theory (these and others) have been confirmed on numerous occasions subsequently. The astronomical readings only appear as an example of a crucial test case in retrospect when the rather dubious machinations to achieve confirmation were later supported by other readings. We might say that the history of crucial case studies is written by the winners. Furthermore, in this case, unlike many in the social sciences, whilst the measuring itself is difficult, what is being measured should remain stable. The generalization is invariant and law-like.

are considering empirical, not law-like, generalizations, as we generally are in the social sciences. Crucial case studies cannot decisively confirm in the social sciences. The same is true of case studies falsifying claims. Going to Australia and finding a black swan might have seemed to be crucial, but only once one confirms the bird is rightly called a swan. In political science Arend Lijphart’s (1968) study of the Netherlands is often cited as a crucial case study, falsifying the claim that cross-cutting cleavages are required for political stability. The Netherlands has a stable political system, yet also has reinforcing social cleavages. Lijphart’s own theory is that elite accommodation can overcome the mechanism that leads to the need for cross-cutting cleavages. Perhaps; though such elite accommodation did not last for long in other cases such as the Lebanon, and arguably no longer holds so well in the Netherlands today (Pennings and Keman 2008).

Hypotheses and Theory Testing

115

This is the lesson I draw from Popper’s (failed) account of falsifiability. Complex models have many facets, and we cannot be sure from a single case that we have really shown that a model is wrong. The Netherlands might have found a way of dealing with its reinforcing cleavages, at least for a time; and from where we stand now those cleavages do not look very deep. The Lebanon also managed a degree of stability for a while with similar institutions, but their cleavages remained profound and the failure of those arrangements deepened them. The original theory might need reinterpreting in the light of Lijphart’s work on the Netherlands, but it still seems fairly robust. With deep reinforcing cleavages it is difficult for a polity to remain stable, though some institutional arrangements might help counter the effects. Whilst institutional arrangements might be able to mitigate threats to political stability, we should expect greater stability in systems with cross-cutting cleavages. The importance is in the story, the narrative, or the mechanism that explains relative stability. The empirical generalization and how it might be sustained or countered is what follows from the mechanism, and the evidence we should be collecting needs to be directed at examining the mechanism not the generalization. I once conducted what I thought might prove to be a crucial case study of two rival theories of the state (Dowding and Christiansen 1994). Pluralism (roughly) claims that government consults legitimate group organizations and government policy balances out rival groups based on strength of interests (Dahl 1961a, 1961b, 1982). The relative autonomy of the state thesis (roughly) argues that governments only take account of group interests that are already in line with the state’s own interests (Nordlinger 1981). Amnesty International (British Section) had offices in central London. On the second floor was the office dealing with human rights abroad; the third-floor office handled human rights in the UK. The foreign section had direct and swift access to the British Foreign Secretary, was regularly consulted on humanrights abuses abroad, and was involved in training British diplomats on human rights. The home section was never consulted, was not even on the list of organizations receiving press releases about human rights, or penal policy, and so on, in the UK, whilst requests to talk to the Home Secretary were often refused. The fact that the same organization received completely different treatment by the same government because different departments in that government have completely different interests with regard to human rights suggests that the relative state autonomy thesis was corroborated. However, the awkward word ‘legitimate’ appears in the pluralist thesis. At first I considered that since the example concerned the same organization, if one part of it was considered legitimate, then so should the other be. However, I really needed to delve further and see what the British people thought of the way Amnesty International was treated, precisely what ‘state interests’ meant in

116 The Philosophy and Methods of Political Science the two theories, and so on. It was a nice and suggestive case study supporting relative state autonomy, but certainly not a crucial one. We should also note that whilst we can have degrees of belief in the sense that nothing is certain, we also have what we might dub ‘opinion’ (Dennett 1978). Our opinion might be that, given the odds, one theory is correct and another false. When analysts plump for one theory over others, what they are doing is forming their opinion given the evidence available. Good academics, however, will always leave some room for doubt. Often when we see an explanatory theory falsified or disconfirmed, what we mean is, given the evidence, it is our opinion that it is false. When we see a theory corroborated or confirmed, it is our opinion that it is true. But our degree of belief in the falsity or the truth of theories might vary; our opinions can be weakly or strongly held. In reality, case studies never disconfirm theories. They might falsify them in Popper’s sense – reduce the odds that a given model is true relative to a rival one – but the odds are reduced far more when other studies are conducted; and particularly when we have accounts of why a model is likely to fail in a given case alongside rival predictions from that other account. In that regard, theoretical (and sometimes normative) considerations count as much as empirical ones (as in the Amnesty International case). A case study might change our opinion, but it cannot (unless we are considering models which produce point predictions, and even then we have to be careful) shift the odds that far on its own, no matter how carefully designed it is.

5.4 Hempel’s paradox and Popper’s falsifiability Popper’s account of science as conjecture and refutation was intended to do several jobs. First, it was supposed to demarcate science from non-science. Second, it was supposed to solve the problem of induction. As part of that latter aim it was meant to solve Hempel’s (or the confirmation) paradox. It is now generally agreed by philosophers that Popper’s project failed. Its failure was a technical one, based upon how he defined corroboration, the nature of the empirical content of theories, and his account of truth or verisimilitude. Nevertheless, there are lessons to be learnt from Popper. I shall draw them out by describing Hempel’s paradox, for that concerns the distinction between confirmation and mere consistency of evidence. I shall go on to offer pragmatic advice about theory testing given these lessons. The issue discussed in the previous section of changing the definitions of dependent variables following apparently disconfirming evidence is related to a technical problem with the standard empirical positivist model of explanation and potential solutions to that problem. It is often referred to as the ‘paradox of confirmation’ or ‘Hempel’s paradox’. As the DN model of

Hypotheses and Theory Testing

117

explanation is based on logical deducibility, any evidence that is consistent with a hypothesis drawn from a model logically implies or confirms that hypothesis. Thus the fact that an apple falls from a tree confirms both pluralist and elite theories of state power structures, since gravity is consistent with both pluralism and elitist accounts of the power structure of the modern state. Or, to give another example (discussed below), discovering that cars do not geographically move for fiscal reasons confirms that households do, since it follows from all households moving for fiscal reasons that anything not moving for fiscal reasons is not a household. Karl Popper’s very different account of testing hypotheses, his account of corroboration and falsifiability is, in part, designed to avoid both Hempel’s paradox and the problem of induction. According to Popper, one does not test explanatory theories against the evidence. Rather, one tests such theories against each other given the evidence. That is, one tests one model against another model that generates a rival hypothesis with regard to some empirical matter. It is only when one has a rival hypothesis that one can be sure that the models are rival with regard to the empirical claim at stake, and only then can we see whether, given the evidence, one model is superior to the other. One reason Popper thinks we cannot simply test explanatory theories against the evidence is that evidence is ‘theory-laden’. I touched on this important point in Section 5.3 above, where we saw how a dependent variable might change its definition when purported counterexamples arise, and shall return to it later. Once evidence comes in, we often reconsider the relevant description of a dependent variable, depending on the nature of the theory (in this case process or mechanism) of which that dependent variable is a part. Sometimes the theory drives our interpretation of our results as much as our results drive our interpretation of a theory. The important reason for collecting evidence for rival hypotheses is not to increase the probability of a hypothesis being true, but rather to affect the relative odds of the truth of one model in relationship to another. With confirmation, the idea is that each time we collect a piece of evidence that is consistent with a hypothesis, the probability of the hypothesis and (we might say) therefore the model from which it is drawn increases. Thus, if the hypothesis is that all swans are white, then every time we find a white swan, or a black crow, the probability that all swans are white increases. As soon as we find one black swan, the hypothesis is disconfirmed, and so we no longer hold that hypothesis. Or if the hypothesis is that democracies do not go to war with each other, then every case of non-warring democracies confirms the hypothesis but one case where two democracies do go to war will disconfirm it (Brown et al. 1996; Ish-Shalom 2013). For this model of explanation, once the deduced hypothesis is found to be false, it cannot be (rationally) held. Now we note that disconfirmation in the sense of demonstrating a claim to be false only applies to strictly invariant generalizations. Empirical

118 The Philosophy and Methods of Political Science positivists (and at times Popper too) thought that all scientific theories fundamentally were such invariant generalizations. (At base, for the empirical positivists, scientific theories are simply descriptions of the universe, and these general descriptions can then be used to deductively explain all singular descriptions.) The democratic-peace example might not be considered by empirical positivists to be a strict law-like generalization. Popper holds that no single instance of disconfirming evidence need ever lead us to give up a hypothesis, not even a strictly invariant generalization. Hence he does not think there is such a thing as a ‘crucial test case’. Finding a black swan falsifies the generalization that all swans are white (remember, for Popper, falsification is not the same as meaning the claim is false). There might be – and this is widely accepted – a ‘measurement error’. More importantly, since evidence is theory-laden, it might be more reasonable to change our definition of ‘swan’ than give up the generalization in the light of the evidence. For Popper, falsifying means that the odds favouring the theory (generalization in most of his work, but model in my terms) from which the hypothesis is drawn, relative to a rival theory (generalization, model), go down, whilst the odds that the rival theory is correct go up. If the evidence collected falsifies one hypothesis, it must also corroborate the rival hypothesis (in this case perhaps ‘all swans are either white or black’) and thus increase the odds that the rival model is correct. The rival generalization is still falsifiable. The claim all swans are either white or black will be falsified by finding a red swan, corroborating another claim: ‘All swans are white, black or red.’ Sometimes changing the model slightly because of the evidence is reasonable; to keep doing so (‘OK, chaps, let’s add another colour to the list!’) is not. When it comes to the democratic-peace generalization, querying whether the democracies that fought really were democracies or whether the conflict truly constitutes a war is less obviously unreasonable. How far such claims are reasonable should be judged in terms of the underlying mechanisms that are thought to generate the democratic-peace thesis. Generating rival hypotheses from developed models is Popper’s answer to the problem of reasonableness. And I think it is a good one, despite other flaws in his account. Falsifiability and corroboration are thus odds ratios (well, almost, as I explain below) and not probabilities. You might think that one could turn the odds ratios into probabilities of each theory (model) being true. But that would only be possible if you had the relative odds of all potential theories (models) of any given phenomenon. Standard or classical statistical tests of significance were developed with falsifiability in mind, though the nature of falsification here is slightly different. R. A. Fisher, like Popper, thought the purpose of testing hypotheses was refutation rather than confirmation. He thought the important aspect of evidence for a model is likelihood, which for him is the evidence for a hypothesis relative to the rejection of the null hypothesis. He thought failing

Hypotheses and Theory Testing

119

to reject the null does not give one a positive reason for adopting any particular model, as there are an almost infinite number of alternative models that could furnish such a positive result (Fisher 1935). This is why, when adopting the Fisher test of statistical significance, we talk of rejecting the null rather than confirming the model. The null here is identified with chance agreement with the data. For Popper, the evidence for a model is relative to a rival model and so the relevant likelihood is the odds of one model being true relative to another. The problem with Fisher tests of statistical significance is that the null hypothesis is identified with a particular chance distribution of the data (Howson 2000: ch. 2), so rejecting the null when statistical significance reaches some point is not only arbitrary, but can also be misleading. Popper’s account seems to have the advantage that the claim of corroboration of one model is only relative to another model, but still suffers the problem that the evidence corroborating the chosen model over the falsified one is arbitrary, since that evidence could equally corroborate myriad other models which share the same prediction as the chosen one. Popper recognizes this, hence his emphasis that the odds of a model being true are only relative to the rival model, and not a probability of the model being true. Of course, we might have other, theoretical, reasons, to choose this model to test as opposed to myriad others. Popper’s problem is justifying these theoretical reasons without specifying them in terms of prior probabilities, as in Bayesian thinking, and seeing tests as being designed to bring posterior probabilities as close to 1  as possible given our reasons (our prior probability) for testing any particular model. A further issue is related to the Duhem–Quine thesis. Even if model A is falsified relative to model B because the evidence supports a hypothesis from B and not from A, we might still hold on to model A rather than B, since there might be more support overall for A. We might need to modify model A in light of the evidence, but we do not necessarily switch to model B. We need to take into account all of the evidence. The assumption that evidence inconsistent with a model automatically renders the model false Popper calls ‘naive falsification’. He was continually frustrated by others assuming he was a naive falsificationist, despite carefully distinguishing himself from naive falsification (and the positivist account of confirmation) in The Logic of Scientific Discovery (first published in 1934). Responsibility for the continued misreading of Popper lies largely with Imre Lakatos, a one-time colleague of Popper; in reading Lakatos (1978), one must always keep in mind that what he means by falsifiability is not what Popper means. Popper claims that the difference between science and non-science is that scientific statements are falsifiable, but non-scientific ones are not. Any statement that is not falsifiable in principle – that is, we can imagine what evidence could falsify it (reduce the odds that the model from which

120 The Philosophy and Methods of Political Science the statement is drawn is true) – is not a scientific one. He called this the demarcation principle. Lakatos and many since claim that explanatory theories or models can never be falsified (they mean disconfirmed) because ‘auxiliary hypotheses’ can always save them. In fact, Popper discusses this issue in The Logic of Scientific Discovery, published 40 years before Lakatos. (Popper explicitly states that only ‘auxiliary hypotheses’ that increase falsifiability should be allowed (1972b: 82–3; see also 1989: 57–9; and 1974: 1004–9, which is a wonderful destruction of Lakatos). For Popper, the empirical content of a model is important to corroboration (and therefore falsifiability). Empirical content is defined in terms of what the model excludes (Popper 1989: 217–20, Addendum 1, 385–7). Thus the generalization ‘all swans are white’ excludes more than the generalization ‘all swans are either white or black’. We can consider an example of a non-formal model from urban politics and economics, the Tiebout Model (Tiebout 1956), which generates claims about whether efficient urban service delivery can rely on signals produced by the movement of households. According to (a reasonable version of) Charles Tiebout’s non-formal model (parts of which have been formally modelled (Zodrow 1983; Greenburg and Weber 1986)), in large urban areas with multiple competing jurisdictions, the geographical movement of households across jurisdictional boundaries can supply signals to local governments about the efficiency of the tax–service package they provide. In Tiebout’s famous phrase, ‘Citizen-consumers vote with their feet’ through their moving decisions. For the claim to hold, households must take fiscal factors into account when they decide to move. Thus one hypothesis that is generated from Tiebout’s account is that households take fiscal factors into account when moving. That is, part of the empirical content of the non-formal model is that households take into account fiscal factors when they move. According to Popper, though, what is important for empirical content is what it excludes. For example, the hypothesis that ‘people voluntarily move house for all sorts of reasons’ can be derived from a ‘theory’ about geographical mobility, but it does not exclude as much as the hypothesis that ‘people move house only for fiscal reasons’. It does not exclude as much, since the latter can be falsified by finding at least one household that has voluntarily moved for reasons that are not fiscal. Indeed, the more general hypothesis could only be falsified if we discover that households never relocate across jurisdictions or that they do so completely randomly. The claim that households only relocate for fiscal reasons is more falsifiable than the general claim or the claim that households take into account fiscal factors, because there is more evidence that could falsify it. Hence it has more empirical content, even though all three explanatory theories purport to explain precisely the same physical reality, the geographical mobility of households.

Hypotheses and Theory Testing

121

If the theory passes its tests each time – when we examine household mobility we do indeed find that people always move for fiscal reasons – then the explanatory theory is corroborated each time. Popper is adamant that this does not increase the probability that the theory is true. That is because he believes that induction cannot be justified; each discovery that a household moves for fiscal reasons cannot logically (deductively) demonstrate that all households do (or that all households in the future will move for fiscal reasons). Each test corroborates the theory; but, as explained above corroboration is not the same as confirmation. The important point is that this allows Popper to avoid Hempel’s paradox. Since any empirical evidence that is logically derivable from a model confirms it, discovering that cars do not geographically move for fiscal reasons confirms that households do, since it follows from all households moving for fiscal reasons that anything not moving for fiscal reasons is a not a household. However, examining why cars move will never falsify a model about fiscal household geographical mobility. The fact that cars do not fiscally move does not change the odds on any model about household fiscal mobility in relation to any rival model that claims neither cars nor households move for fiscal reasons. Hence such findings cannot corroborate the model either. In that sense, falsifiability and corroboration are converse concepts. However, there is a complication. I have been suggesting that corroboration and falsifiability constitute odds ratios for or against rival models. A careful look at Popper’s formula shows that the degree of corroboration is not straightforwardly increased each time a theory passes a falsifiability test. This is so for two reasons (Popper 1989: Addenda 1–3, 385–96; 1983: 217–55). First, some theories have more empirical content because they are more falsifiable, even though they purport to explain precisely the same physical events; but secondly because some purport to explain a greater number of physical events. More general theories have greater empirical content than less general ones. Theories purporting to explain all human behaviour are more general than ones purporting to explain only geographical household mobility; though all theories of the latter might be derived from the more general theory. Does a theory that is more general but less falsifiable have more or less content than a less general one which is more falsifiable? According to Popper, it has less. So an unfalsifiable theory about all human behaviour is less general than a falsifiable one purporting to explain only geographical mobility – in which case the latter could not be derived from the more general one. More general theories only have more empirical content if they are more falsifiable. In fact Popper only discusses more general theories in relation to less general ones when the former include the latter as special cases. He says we can only compare by subclass relations (for example, Popper 1972b: 269).

122 The Philosophy and Methods of Political Science However, Popper does say that passing tests corroborates explanatory theories and increasing corroboration does increase their similitude or ‘truthlikeness’. He says we can calculate this degree of corroboration or similitude using the probability calculus, but the degree is a ‘probability but not in the sense of the probability calculus’ (Popper 1972a: ch. 1; 1974: 1013; 1992: 100). Looking at his formula for degree of corroboration across two models, it seems that the relative degree of corroboration is a relationship between how much a model can be falsified and how far it has passed those tests. For two explanatory theories with the same empirical content, the corroboration of one theory and the falsification of another is a straightforward odds ratio. Where they have different empirical content, whilst the odds of one going up are related to the odds of the other going down, the relationship is not so straightforward. Popper repeatedly claims that the more falsifiable an explanatory theory, the more likely it is to be actually false and, as a consequence as our scientific knowledge grows, the more probable it is that it is false (Popper 1983: 223–7; 1989: 192–3). (The less we think we know the less likely, overall, we are going to be wrong.) The greater the corroboration of a model the greater its similitude: we get closer to the truth the more our (non-falsified) theories explain. Now, there are serious problems with Popper’s account of verisimilitude, as I discuss below. Nevertheless, they affect his account of corroboration only tangentially to the distinction between corroboration and confirmation. Having set up the Popperian account, we can now ask whether explanatory theories or models can always be saved by changing definitions or by ‘auxiliary hypotheses’. Lakatos’s (1978) favourite example is drawn from astronomy. In the Ptolemaic system of calculating planetary motion, all planets move in perfect circles around perfect circles. However, these calculations only work for a relatively short while and in order to predict the correct path of the planets relative to the earth, new concentric circles continually need to be added. Lakatos claims this is just like Newton’s astronomical theory, which also fails to predict accurately, leading astronomers to posit an extra planet. In both cases ‘auxiliary hypotheses’ can save the models. However, the two cases are importantly different. Adding extra circles saves Ptolemy but only metaphysically. Continually adding circles to the model does not provide a means that is open to empirical refutation, since the circles themselves are not empirically observable. For that reason the Ptolemaic theory is not scientific. Newton’s model, however, provides a prediction or hypothesis that is open to empirical testing: namely, we should expect to find another planet. We did, corroborating the theory. We might find that Newtonian physics does not quite predict the planetary orbits correctly, but whilst we can look for other planets or material orbiting the sun, it

Hypotheses and Theory Testing

123

remains a scientific theory. And we should stick with it, according to Popper, until a better theory comes along. One did. The principle that you can’t beat something with nothing is a good one. Global climate-change deniers are quick to point out that climate scientists continually get their predictions wrong. (They do, unfortunately most of the time in the wrong direction: things are worse than they earlier suggested.) But this does not challenge climate change as a model unless climate-change sceptics can come up with a rival model that predicts better. As far as I am aware, none has even tried to do so. We might say of political or social theory that, to the extent new assumptions do not produce new predictions, then social science resembles Ptolemaic ‘ad hocery’ and, to the extent it adds new predictions, it provides scientific explanation. I think we can probably choose examples from political science that fit one or the other. Duverger defined semi-presidentialism and his famous ‘laws’ about electoral and party systems (Duverger 1980, 1954). Literature on the first has had a tendency to follow Ptolemaic ad hocery, chasing new sub-types to explain divergences from predictions based on the categories of presidentialism, semi-presidentialism and parliamentarianism. One of the issues here is to what purpose the categories are being used, what are the dependent variables (Elgie 2004); another is variation in factors beyond that being measured (Cheibub et al. 2014); yet another is that some work suggests that other factors are more important than the categories for some of the dependent variables (Cheibub and Chernykh 2009). Literature on the second has led to refinements in predictions, following accounts of the effective number of parties, and concentrating attention upon district magnitude and competition within districts rather than at national levels (Taagepera and Shugart 1989; Taagepera 2007). In other words, Popper’s demarcation criterion cuts through different social scientific models. But what we learn from the Newtonian account is that we do not discard a theory simply because some predictions derived from it do not seem to fit the evidence. First we need to find an explanatory theory that does predict that evidence and then we need to examine the evidence and the theories more carefully. Adding new predictions is a reasonable thing to do. So is changing some of the concepts in the theory. Neither makes such an explanatory theory ‘unfalsifiable’ since both allow for further empirical falsification. Importantly, this means that we modify our models even as the evidence comes in. However, we can also see that redefining and worrying too much about categories leads to sterile debates and few new predictions. Some people might be sceptical that we can always and forever change things so a model is never falsified. However, rather than worrying about that logical possibility, we would be better to concentrate upon actual modifications and conceptual changes and see how reasonable they are, rather than condemning a practice on the grounds that it might be abused. We might also

124 The Philosophy and Methods of Political Science

Box 5.4

Popper and verisimilitude

Holding the capital-letter metaphysical notion of ‘Truth’, Popper at first tried to avoid the concept, believing that corroborated theories could still end up being false. He later introduced the idea of ‘truthlikeness’ or ‘verisimilitude’, suggesting that better theories had greater verisimilitude than lesser ones. The amount of verisimilitude depends on a model’s empirical content, but Popper now introduced ‘truth-content’ and ‘falsity-content’ (Popper 1972b, 1983 and 1989). The first is the class of true propositions that can be derived from a theory, the latter the false propositions – this being empty if the theory was completely true. He suggested two methods of comparing models in terms of verisimilitude: the qualitative and the quantitative. Of the former he said that if we assume the truth-content and the falsity-content of two theories t1 and t2 are comparable, t2 is closer to the truth than t1, if and only if either: (a) the truth-content but not the falsity-content of t2 exceeds that of t1, or (b) the falsity-content of t1, but not its truth-content, exceeds that of t2. In this sense verisimilitude is seen entirely in terms of subclass relationships: t2 has greater verisimilitude than t1 if and only if the truth- and falsity-contents of each are comparable through subclass relationships, and either (i) t2’s truthcontent includes t1’s and t2’s falsity-content, is the same as, or is a subset of t1’s, or (ii) t2’s truth-content includes, or is the same as t1’s, and t2’s falsity-content is a subset of t1’s. On Popper’s quantitative account, verisimilitude is defined by assigning quantities to contents. Here an index of content is created by a theory’s logical improbability since content and probability vary inversely. Formally then, Popper defines the quantitative verisimilitude which a statement ‘a’ possesses by: Vs(a) = CtT (a) − CtF (a), where Vs(a) is the verisimilitude of a; CtT(a) is the truth-content of a, and CtF(a) is its falsity-content. Popper claims that any theory t2 with a higher content than a rival theory t1, which is subsequently falsified relative to a third theory t3, can still legitimately be regarded as superior to t1 in the sense that it is closer to the truth than t1. Science progresses by establishing theories with ever-higher verisimilitude. However, a number of scholars in the 1970s showed that the superiority of theories by Popper’s qualitative and quantitative definitions only worked for theories that turned out to be true, since verisimilitude does not necessarily intersect positively with corroboration (Miller 1974a, 1974b; Tichý 1974, 1978). It was proved that a theory t2 which had greater content than a rival t1 would have both greater truth- and falsity-content than t1. How important is this failure? Some see verisimilitude as central to Popper’s work and hence its failure topples the entire edifice. A great deal of work has gone into trying to repair his account, but Popper himself suggested that numerically calculating verisimilitude is not the key issue (Popper 1983: xxxv– xxxix; Oddie 1986; Niiniluoto 1987; Brink 1989; Brink and Britz 1995). I largely agree. What is important is that we need rival hypotheses to test models against each other.

Hypotheses and Theory Testing

125

add that we should not discard any organizing perspective despite suspicious ad hocery in some of the models developed within that perspective. Not, at least, until we have a better organizing perspective – that is, one that produces models with greater corroboration. It also follows from these reflections that we can modify a model, either by some internal change in its formal structure or, keeping the original formal structure, by new empirical predictions. Furthermore, given that most statistical tests themselves require a host of specific assumptions and are modelled by those assumptions (so-called ‘empirical models’: see below), often whether or not a particular explanatory theory can be deemed to have been corroborated is not straightforward. Related to these claims is the argument of the Duhem–Quine thesis that you cannot disconfirm a single hypothesis, because hypotheses are self-supporting (Quine 1960, 1993). Quine’s point here is about the meaning of scientific concepts. He says that when we alter the meaning of one theoretical concept because of empirical evidence, other theoretical concepts also change. Changing one concept reverberates through the web of other concepts, affecting all hypotheses involving those concepts. I think we can allow Quine this logical point without worrying about it too much. Whilst it is true that conceptual change will mean a shifting, to some extent, of all other concepts, we will not notice any change in those that are ‘far away’. Quine suggests we can imagine our language as a balloon where everything is connected to everything else. Conceptual change in one place can affect everything else. We might rather think of it like the planet. Massive conceptual change can occur like an earthquake, rearranging other concepts nearby. However, on the other side of the planet, whilst very sensitive instruments might measure such tremors, most people will not notice them. Indeed, we can view the difference between confirmation and mere consistency as whether we would feel the need to shift our definitions of concepts in the light of contrary evidence. If the evidence we find is what we expect, but had it been different we would have needed to change our concepts to keep the model in line with the new evidence, then that evidence is confirmatory. If, however, the supporting evidence could have been different but not cause us to change our concepts, then it is merely consistent. If gravity did not apply, then surely our theories of the state would change, but would we need to change any of our concepts within the models of pluralism or elitism? If people always take into account fiscal factors when driving (say, all roads are tollways), do we need to change our concepts of ‘household’ or ‘fiscal factors’? In the first case I think not. Thus gravity is only consistent with elitism and pluralism. In the latter I would also think not; thus whether cars move for fiscal reasons is only consistent with the theory. It might be argued that we need to extend our understanding of ‘fiscal factors’,

126 The Philosophy and Methods of Political Science but I do not see that as a problem. People doing things for fiscal factors outside of household locational decisions might indeed be confirmatory of them doing so in household locational decisions, if only slightly. (In fact, car movements might be affected by fiscal factors, such as tolls and city congestion charging – and we can explain that by the economic theories underlying both household and car movements.) I have tried to draw some lessons out of Popper’s attempt to avoid Hempel’s paradox. We cannot rely upon Popper too much. His corroboration/falsifiability account itself fell into deep paradoxical problems over his attempt to formalize how we compare theories (models) with regard to the truth or what he called verisimilitude or ‘truthlikeness’ (see Box 5.4). The problem is that theories are constantly falsified. We keep on replacing one model with another, or modifying our models in the face of recalcitrant evidence. In that sense we might say that virtually everything anyone once believed (scientifically) was false. That is another reason why some writers prefer to think of models as more or less useful, rather than more or less true. Again, not much might really hang to this dispute, especially for the empirical researcher. Nevertheless, I do think that people look for truth and not just utility, though what we mean by truth might simply be that which is predictive or useful to us. As I argued in Chapter 3, models are true to the extent that their structural elements correspond to structural elements in the universe, and can be thought to be true to that extent. If those structural elements are all that one is using to generate predictions, or utility, in a model, then I think we can call the model true. It is true to its purpose. I have distinguished between confirmation/disconfirmation, corroboration/falsification and consistent with/disconfirmation. The failure of Popper’s attempt to bring the notion of verisimilitude to defend his falsifiability thesis means that these distinctions cannot really be maintained. We can see that to an extent in my examples of ‘consistent with’ above, which we can certainly press to see that evidence that is inconsistent with a hypothesis would in fact lead us to modify our concepts or perhaps not make sense of an explanatory theory at all. If there was no gravity, could we really leave pluralism or elite theories intact? But as a practical working political scientist, it seems pretty obvious to me that some evidence tests theories and some does not. There might not be a hard philosophically defensible distinction between the three concepts, but in most cases we can see a big enough gulf to warrant using separate terms. The test is to ask in each case: if the evidence were different, how would it make me modify my explanatory theory? If it is not obvious that it would, then the evidence is merely consistent. If it would lead to a change in the model then it is corroborating my model relative to a rival. We can think about evidence consistent with a hypothesis drawn from a model as confirming that

Hypotheses and Theory Testing

127

hypothesis, and we are interested in such confirmations when they corroborate our explanatory theories. That is because we are not interested in hypotheses as such, but rather in the general explanatory theories in which they are embedded and the mechanisms that narrate the explanation given by those theories. It is those theories that endow hypotheses with much of their meaning and suggest what kind of evidence might confirm or disconfirm them. The important point about corroboration and falsifiability is bringing to the fore the need to understand that we test models against each other through rival hypotheses that show the models are rival. Following those tests, we support one model rather than another, either because of the empirical demonstration or because we see the need to modify our model in important ways. That is the main message of the journey through the often misunderstood and misrepresented work of Karl Popper. All too often in political science non-formal models are pitted against each other as though they are rivals, when in fact the evidence used to defend one is also consistent with its purported rival. Being Popperian in the way we think about comparing different theories does not mean we have to buy into the full Popperian account of scientific method. If the only thing we take from Popper is that we need to contrast our hypotheses with others, then Popper’s work is still worthwhile. The inversion strategy (see Chapter 4) is a quick way of thinking about the contrasting theories against which we are testing and the prior probabilities we might assign to any hypothesis and associated theory to see how strong a test we think our evidence is.

Box 5.5

Hypotheses

The term ‘hypothesis’ has several meanings. It can be a subordinate particular of a more general thesis; a particular example of a more general proposition. In formal logic it can be supposition that forms the antecedent for a conditional proposition, and derivatively can thus be any proposition or assumption that is stated merely as a basis for reasoning without any empirical support. That leads to its pejorative sense for a claim that is insufficiently grounded (that is just a hypothesis). It might be a theoretical statement that is supposed to account for empirical findings. It is used to mean ‘conjecture’ that can then be empirically tested, and it is in that sense (derivatively from the first) that it sometimes is used to mean a prediction from a model (rather than an assumption in a model). I will use it to mean any prediction that has been derived from a formal model. Thus from a model we might draw the hypothesis that democracies will not go to war together (an empirical prediction), or the hypothesis that the only popular wars are ones conducted in self-defence.

128 The Philosophy and Methods of Political Science

5.5 Concepts are theory-laden I conclude this chapter with a developed account of how important concepts in rival models can be completely theory-laden and why theory is involved in empirical testing. The theory-laden nature of concepts is sometimes used to suggest that models cannot be tested. That claim needs to be demonstrated not assumed. In the following example we can see how one conceives of a theoretical concept – in this case the spatial location of political parties can affect our interpretation of rival hypotheses. In spatial models of voting and party behaviour both voters and parties are supposed to occupy a location in n-dimensional space. Often the number of dimensions is restricted to one, or two, for tractability and because the small number seems to be able to track voter and party behaviour. The idea of location in n-dimensional space is a theoretical concept that is useful, both because it has proved to be predictive but also because we can imagine it. We perceive objects in locations in two dimensions on maps or in three when we look at the world, so we can picture the theoretical location of people and parties. There are many ways of measuring the ideological location of parties – for example, analysis of election manifestos or speeches (Budge et al. 2001), automated processing (Laver et al. 2010) or expert assessment (Laver and Hunt 1992). Both the proximity and the directional modellers locate both parties and voters in the same way. They ask voters to locate themselves and parties with regard to a series of issues. So people are to say where a party sits on, say, a seven-point scale with regard to defence policy, with one end being hawkish and the other dovish; or attitudes to health care, and so on. From the sets of responses to these questions the position of the parties and the voter’s own position in ideological space are calculated. The two rival spatial models of party and voter behaviour, despite using similar measures, generate different predictions, at least in part because of different ways of conceptualizing how to measure the placement of parties in ideological space. Proximity models predict that people will vote for parties that are closest to them in ideological space. They also predict, at least when politics can be reduced to a single dimension, that major parties will tend to converge in ideological space. Directional models predict that people will see themselves as ‘left’ or ‘right’ wing, and vote for the party that, within certain bounds of acceptability, is the most left or right wing. They do not predict party convergence. Since the models have clearly different predictions, they look like rivals, both with regard to how voters should vote and party strategy. Testing one model against the other ought to be relatively easy. However, both sides in the debate claim that their model is corroborated and the other falsified. I leave aside some issues concerning the data and appropriate statistical models, as well as a specific criticism of the directional model (that

Hypotheses and Theory Testing

129

the region of acceptability is not well specified). I concentrate upon a specific conceptual debate: how do we empirically locate parties or candidates and voters within ideological space? Proximity researchers compare the self-assessed placement of the respondent with the respondent’s assessment of the location of the parties (the idiosyncratic placement). Most researchers using this technique find results that tend to corroborate the proximity model. Macdonald and others argue, however, that this means parties are located in several different places at once, depending on the idiosyncratic placement by all voters (Rabinowitz and Macdonald 1989; Macdonald et al. 1991; Macdonald and Rabinowitz 1993). Yet spatial theory assumes that parties each occupy a single position. Accordingly, researchers take the mean of the placements of all their respondents to locate a single ideological place for the parties (or candidates). When they do this, they find the directional model tends to be supported. When the placement of political parties is put this simply (and the issues are rather more complex), it is hardly surprising that the empirics produce different results. It is not surprising that voters are more likely to vote for parties they have idiosyncratically located in the same ideological space as themselves. On the other hand, it is not surprising that when we take the mean measure, more extreme parties are likely to appeal to voters. Why? Given that those who vote for parties are likely to place them close to where they place themselves, and we might assume about equally to the left or the right, then the relative placement of parties away from that point is likely to be largely determined by respondents who do not vote for them. Voters on the right of the spectrum are more likely to place left-wing parties further to the left than are left-leaning voters; voters on the left more likely to place right-wing parties further to the right than are right-leaning voters. The issue is a conceptual one. Surely Macdonald and colleagues are correct that in theory parties need to be placed on a single point on the ideological spectrum. On the other hand, since people vote for a party given how they perceive it, we should judge the proximity of voters to the parties they vote for given voters’ judgement of themselves and the parties. Some argue, against the mean placement, that a voter’s own location of the candidate or party is important, since that voter cannot know what the national mean placement is. Some argue that spatial utility theory only requires that a voter’s assessment of their spatial position be compared with the voter’s assessment of the candidate or party’s spatial position. However, theoretical spatial theory assumes that voters and parties each only take up one position, so why shift the theory when it comes to empirics? Without offering judgements on the particular issue (and the later literature transcends some of these issues – probabilistic proximity models do not always lead to convergence and models of mixed voter behaviour seem to provide clearer predictions (Merrill III and Grofman 1999; Adams et al.

130 The Philosophy and Methods of Political Science 2005)), we can see that a specific dispute over the meaning and measurement of ‘party location’ affected the empirical analysis. Conceptual analysis and consideration of measuring both party and voter position can overcome these issues. The example is meant to demonstrate that sometimes how we understand theoretical concepts affects our measurement and understanding of the empirics.

5.6 Conclusion We can test models that draw strict hypotheses by collecting evidence that might be consistent or inconsistent with that predicted by the hypotheses. One way in which we might think about this is that we have a prior probability that the model is true. If the model is rather trivial or obvious, our priors (what we consider the probability of its being true before we test) will be high, and any test will not much affect the probability. If the hypotheses are surprising, then our priors might be low and a test that is consistent with them will increase the probability that the model is true by a much higher amount. Evidence consistent with the obvious does not do much to change our confidence that the obvious is indeed true, evidence consistent with the surprising can. This can be the case whether the evidence is consistent or inconsistent with the model. However, evidence that is inconsistent with what seems secure might be trusted less. We might be more wary of the assumptions that went into the empirical model or more wary of accepting the data without careful checking. In that sense how ‘interesting’ the hypotheses are matters a great deal for how we assess the evidence for them. However, models can be consistent with a great deal of evidence, which is why we want to draw specific hypotheses from them. What makes hypotheses interesting is how surprising they are or if they are inconsistent with other models, maybe ones for which our priors are higher. Here we can in practice test models. A philosophical problem with this idea is that we can (so it is claimed) come up with variant models from which we can derive the same hypotheses and therefore we cannot really test models. The response to this claim, I have been suggesting, is that whilst we might, in theory, be able to come up with a rival model, for the critics’ point to be anything other than weasel philosophy, they need to actually come up with that rival model. Furthermore, if a model is truly rival, then it will produce some predictions different from the one tested, so, in theory, we can still choose between them. As we saw in the extended example in Section 5.5, even where predictions seem to be clearly different, what is doing the work might be the concepts within the theory and the assumptions made within the empirics to test the predictions in relationship to those concepts. This does not mean that we cannot test models, but

Hypotheses and Theory Testing

131

it does mean that we need to be careful. Furthermore, the real test of how far we make judgements depends upon the actual models and tests, and is not an issue that can be sorted out philosophically. To be sure, there are deep philosophical issues with regard to testing, but what matters to the empirical researcher is not the logical but the empirical issues they face. Often in political science we do not produce hypotheses that have been strictly derived from a model. Our non-formal models suggest that certain kinds of results should be found, and we can find evidence consistent with them. Here the danger from rival non-formal models predicting the same hypotheses is much greater. In these cases we cannot be sure that evidence does increase the probability that the non-formal model is true. This is because it is relatively easy to modify non-formal models without having much effect on the empirical claims that can be drawn from them (see Chapter 4, Section 4.4). Thus we need to be much more careful when making claims that our non-formal models have been tested. Often all we do is find evidence that is consistent with them. We really have to pit a non-formal model against a rival one, and show that the evidence is consistent with one and inconsistent with the other, or at least seems to support one rather than the other. Here we are claiming one non-formal model is corroborated and the other falsified relative to each other. And we might make that claim without asserting that we have increased the probability over our priors that the non-formal is true, since there is so much more that could be plausibly added. In that sense, non-formal models act more like ways of framing questions and the evidence we wish to collect, to inductively generate ideas about the mechanisms that we think operate to lead to certain types of outcomes. They ‘frame’ our research – they could be called ‘frameworks’ rather than ‘non-formal models’. When it comes to statistical work, Fisher tests of significance at best only suggest that the model is corroborated relative to the null. At times, especially when no rival models exist, this can be interesting. It is suggestive that the kinds of mechanisms that are associated with the model under test have plausibility. The more interesting a model generating the hypothesis – that is, the more unlikely or un-considered it was, prior to the test – the more likely we are to think that we have corroborated the model. Bayesians are sceptical of this (Bayesian) reading of such tests, and certainly we cannot put any probabilities on rival models that have not been pitted against each other, only against the null. To do that, empirical work needs to be given a formal Bayesian setting. Outside of that more formal and statistical setting, researchers need to set up rival models with clearly generated rival hypotheses, from which, after assessing their evidence, they can make claims about corroborating or falsifying the theories in regard only to those that have been thus explicitly assessed. We can, note, still be highly sceptical of the probability of a model corroborated in this sense actually being true.

132 The Philosophy and Methods of Political Science The philosophical issues that I have discussed in this chapter are potentially problematic for empirical political scientists, but I do not think we need to take the logical problems too seriously. By which I mean we should not be too frightened to make theoretical claims based on empirical evidence. Nevertheless, researchers are sometimes prone to too quickly making comparative inferences when comparative tests have not truly been made, and to assuming evidence consistent with theory is confirmatory or corroborative when it is neither.

Chapter 6

Narratives, Mechanisms and Causation

6.1 Introduction Some believe that the role of science is to find causes. Many political scientists also believe that the ‘gold standard’ of their discipline is to discover causes. Whilst not disavowing this general aim – after all, identifying what causes one event is a fundamental route to predicting others – it is not the only way to make predictions. Discovering constraints, or structures, enables prediction, and we can make many predictions without precisely pinning down causation. Indeed it can be difficult to define exactly what causation is. My previous sentence suggests that constraints and structures are not causes. We often make distinctions between background conditions and causes, but does full causal analysis have to include everything? Some think so, often in the name of ‘causal particularism’. There are more accounts of causation than there are causal accounts of many events. In this chapter I am not going to try to provide my account of the nature of causation. Rather, I will examine a few ways of looking at causation and suggest that personal background, training and perhaps inclination or some natural disposition might lead one person to view causation somewhat differently from another. I will also suggest that finding causes is not the only desideratum in social science and we should not denigrate description. So in this chapter I will: 1 2 3 4

Examine some ways in which causation has been analysed. Suggest that they are not as different from each other as often imagined. Suggest that they all involve narration and interpretation. Suggest that two ways of examining causation lead to different research strategies, one using what I will broadly refer to as ‘but for’ (BF) conditions; the other being probabilistic – and these two approaches broadly underlie qualitative and quantitative explanation respectively. 5 Draw links between these two and proximate and ultimate; type and token explanation. 6 Visit along the way different forms of prediction, causation and explanation.

133

134 The Philosophy and Methods of Political Science 7 Alert the researcher to the need to be careful about what notions she has about these matters and whether her techniques can address the precise questions she has raised. 8 Examine ‘process tracing’ and the notion of causation underlying it. 9 Discuss mechanisms once again.

6.2 Causation as narrative The world’s shortest book: Chapter 1: Coughing. Chapter 2: Coffin. When we read this we can’t help but think that the coughing of the first chapter is causally related to the death of the second. Of course, we haven’t even been explicitly told that there is a death in Chapter 2; we have inferred that from the word ‘coffin’. We have created a narrative out of two words, encouraged by being told that they form a book, a story. The mere fact that someone (me) thought it worthwhile to write down two words in this form leads the reader to construct (or reconstruct) a narrative, a causal story: that someone coughed and died as a result. Maybe it was TB, or just a bad cold that got out of hand. What we see as causation is always a narrative of this kind, though not normally one so pared down, and usually with a bit more evidence. Causation is always a narrative because that is how we describe, explain and understand the world. Establishing causation has become a major theme of political science and the debate is largely about the nature of the evidence for it. On the simple account here, causation is always a narrative of some kind; it is a story of one event following another, not by chance but through the second being caused by the first. When we say ‘not by chance’, we thereby imply there was some kind of necessity or, more carefully, non-contingency (non-randomness) between the two events. There is a correlation between them, but a correlation of a certain kind. What is that kind? It is a narrative that satisfies our demand for an explanation of why the first leads (is nonrandomly related) to the second. Events that are correlated are not causally correlated when they are narratively connected by some other event, and that event is then said to cause them both. For some people this way of thinking about causation implies that it is relativist in the sense that causation only exists in the mind, or rather is a psychological phenomenon that we impose on the universe. However, for the realism I have been suggesting, all patterns are ones we impose on the universe through our observations of it – but this makes them no less real. Their realism is constrained by how predictive they prove to be. (And we

Narratives, Mechanisms and Causation

135

can note the circularity or analyticity here. It is their non-randomness that makes them predictive.) Subjective impositions on the universe that were not so constrained might still pick patterns, but not ones that enable us to make predictions beyond those contained within description of the pattern itself. They could make no predictions beyond the data themselves. (Six heads in a row when tossing a coin is a pattern we can pick out, but we have to be careful what we make of that.) Causation is something we impose on the data, but then a datum is itself a pattern, something we have already imposed. What makes something a cause, by this account, is the nature of imposition. Its nature is a narrative that is predictive in the right sort of way. The barometer falling predicts rain. It only explains the rain by the narrative that quickly descending pressure causes the barometer to fall quickly; quickly descending pressure indicates that a low pressure front is approaching; low pressure fronts are associated with storm clouds. (Is rain caused by low or falling pressure? The way some social scientists use the barometer example to show that correlation is not causation suggests they think so. I think a bit more work needs to be done to explain rain. And that is not unimportant to the example – reconsider it after you have read the next section.) Historical narrative is designed to order our understandings of sets of events. This understanding does not have to be exclusively causal, but generally speaking the narrative the historian tells is supposed to be noncontingent; historical narrative is not supposed to be simply one damn event after another. No one can make this point better than Edward Gibbon, often credited as being the first modern historian: The age of the great Constantine and his sons is filled with important events; but the historian must be oppressed by their number and variety, unless he diligently separates from each other the scenes which are connected only by the order of time. He will describe the political institutions that gave strength and stability to the empire before he proceeds to relate the wars and revolutions which hastened its decline. He will adopt the division unknown to the ancients of civil and ecclesiastical affairs: the victory of the Christians, and their intestine discord, will supply copious and distinct material both for edification and for scandal. (Gibbon 1993/1781: 81) Was it the political institutions that caused the strength and stability of the empire, or did they structure the empire in a manner that enabled stability to be caused by other things? The scenes that are separated and the divisions Gibbon makes are categories created to determine the variables upon which the causal narrative is to be overlain, even though they might be unrecognizable to those who lived through them. Separating the events ensures that we

136 The Philosophy and Methods of Political Science ignore extraneous factors beyond those contained in the narrative for each causal story, and which are selected from all the potential stories running contemporaneously. One needs to separate independent variables from irrelevant ones. Moreover, we are interested in these stories in order to make our lives better (edification) and more pleasurable (scandal). The historian does more than tell causal stories, but she is interested in causation even if she cannot establish it. We cannot help but impose causal patterns or mechanisms on narratives, for that is, in part, how we make sense of them. In fact the psychological desire to find causes goes deeper, for we look for them everywhere. I have argued that the nature of explanation involves prediction, but also claimed that prediction does not entail being able to foretell the future. Prediction is simply the claim that if you think X explains Y under conditions w, then you are predicting Y given X under conditions w. And this holds even where, say, the relationship of X and Y given w is probabilistic. Such a condition might never obtain in the future. Nevertheless, predicting the future is the motivation for the explanations we seek. And finding causes of events enables us to predict future events when similar causal conditions obtain. Our predilection for seeking causes occurs even in situations that we know are not causal. The equation 5 + 2 = 7 is not a causal relationship. Five and two do not cause seven. The relationship defined by the equals sign is one of identity. Nevertheless, psychologically, such formulas are often interpreted as operations much like flour + eggs + mix + bake = cake (Hofstadter and Sander 2013: 409–14). We often think about identity relations as though they are causal ones. Equations representing operations are asymmetrical but identity relationships are symmetrical. Many of the fundamental equations of theoretical physics are symmetrical; they are identity relations rather than causal ones. Similarly most of the equations in formal (‘rational choice’) political science are symmetrical. In other words, they are descriptions. But often we do not understand that each side of the equals sign means the same thing. They have to be further interpreted for causal explanation. Philosophers of language distinguish sense from reference, or intension from extension, to enable us to understand that we do not always realize that something is identical with itself. A large part of science is the discovery of such identities. Of course, fundamental relationships, once discovered, are used as part of causal explanations. That is one of the reasons why we should not disparage careful description in political science, because careful description can help us think about identity as well as causal relationships. We should not allow our propensity to look for causes everywhere to overcome the need to look for identity relations, functional relationships and other structural features of the social world that configure and affect causal processes. We should recognize these for what they are. Historians do not have to pin down causation.

Narratives, Mechanisms and Causation

Box 6.1 Effects of causes Causes of effects

137

Causes of effects and effects of causes

We try to establish the degree to which some factors impact upon some phenomenon. We want to know what effects these factors cause. There is some phenomenon that occurs, and we want to establish why. We want to know the cause of some effect.

And one cannot demonstrate causation for any unique case. One can tell plausible stories that suggest causation, though even here their plausibility will rely upon the assumption of mechanisms that we accept because of evidence from other cases. A case study might be only a single data point for the claim that is being made, but we have to accept all sorts of other generalizations, mechanisms and assumptions in order to be able to narrate the case study in the first place. The distinction that is often made between looking for the causes of effects and examining the effects of causes might be invoked here. We cannot explain some unique outcome by only looking at what happened around that outcome (we cannot establish the cause of the effect), but we can examine what difference in that unique case some element had on the effect (the effects of causes). However, we should not make too much of this distinction which to some extent is merely a play on words. We can no more determine the effects of causes through one occurrence than we can determine the causes of effects (see the discussion of the ‘equivalence problem’ below). However, the unique case might lead us to pick and examine a particular element in a purported cause because, in that example, it seems so important. In recent years quantitative scholars have become more sceptical about the evidence for causation from standard linear regression models and have turned to new techniques. This scepticism has led to a renewed interest in conducting experiments in both the lab and the field, as well as so-called natural experiments where events allow a study under (close to) experimental conditions. Within qualitative research too, there has been a revival of interest in causation. In part this is due to reaction to King et al. (1994) and their unified logic of explanation. There is an attempt to resuscitate careful historical analysis on the grounds that only such analysis, under what has come to be known as ‘process tracing’, can pin down actual causation. Some have suggested that qualitative research is interested in necessary and/or sufficient conditions for causal processes that involve a type of logic (Ragin 2008; Goertz and Mahoney 2012) very different from that of quantitative research. I will consider those views in some detail in Section 6.4. First, however, I want to look at some elements of causation by considering a set

138 The Philosophy and Methods of Political Science

Box 6.2 Objective ‘But for’ (BF) Regularity Generalizations Distal-correlational Production Token-particularism Explanations

Binary causal oppositions — — — — — — — —

Subjective Probabilistic Counterfactual Mechanisms Proximal-serial Difference-making Type-generalization Inferences

of binary oppositions (see Box 6.2). These are not hard and fast distinctions and people take different lines on each of them. But they are elements that go into various considerations about causation.

6.3 Dichotomies in causal accounts Objective–subjective Do causal relations exist outside of our perception of them or are they something we impose on the data? Both ideas can be extracted from the work of David Hume. At times he seems to suggest that causal relations are things we impose on the world, at other times that they are processes existing within the world (Blackburn 2008). Is it a human psychological propensity to impose relations of causality on to the information we receive, or is there really some physical motive force outside of our perceptions? Perhaps whether ‘cause’ is something we impose on the data is not that important, at least for empirical political scientists. I have been suggesting that all patterns we see in the data are ones we perceive, but the possible patterns are constrained by how predictive they are. In this form of realism, causal patterns are simply patterns in the data, though ones that take a certain form or forms that we call causal. Below I will suggest that the elements in a pattern that we denote as a cause (as opposed to ‘background conditions’ or ‘structural conditions’) are often related to the question we are addressing. Thus what we privilege as a ‘cause’ is related to how we are looking at the world at a given moment. Nevertheless, that does not mean that what we are seeing as a cause and what we are seeing as background conditions to that cause are simply question-specific. Many philosophers have made the claim that explanation is reduction to causes. Certainly, even when that claim is not made explicitly, causal relationships are often given primacy over and above other relationships. In recent years political scientists have turned away from traditional quantitative

Narratives, Mechanisms and Causation

139

regression techniques in the quest for causation, arguing that these cannot establish causation, merely correlation for which causation is then assumed or claimed on theoretical grounds. One might try to derail the entire push for causal explanation by claiming that causation is simply a correlation of a certain character. But, even if this is true, that ‘certain character’ is vital to our understanding of the world and how we might intervene to change it. Understanding causation is a key to future predictions. Experimentalists push for causal inference rather than causal explanation, whilst qualitative scholars have a non-probabilistic understanding of causation that differs from standard quantitative understandings. So these drives for new causal understandings rely upon other dichotomies, discussed below. Whether we see causation as something outside of our perception of the world might make little difference to empirical research; but it might make a difference to how we interpret data. The desire for causal inferences as opposed to causal explanations seems to be the attempt to logically constrain the patterns we see outside of any theoretical imposition on the data. Underlying it seems to be the desire to directly observe causation in the data rather than theoretically impose it upon them. Those more inclined to see causation as patterns imposed on the data by us might be more inclined towards causal explanations than causal inference.

‘But for’ (BF)–probabilistic One of the standard ways of defining causality is in terms of necessary and sufficient conditions. An outcome is often defined in terms of sets of conditions that are either necessary or sufficient for the outcome. Various different ways of representing necessary and sufficient conditions exist in the literature. John Mackie (1974) introduced the influential idea of INUS conditions. An INUS condition is an insufficient but non-redundant part of an unnecessary but sufficient condition. So each INUS condition is neither individually necessary nor individually sufficient for the outcome. Each is, however, a non-redundant element of the sufficient set for the outcomes. Any set of conditions sufficient for some outcomes is composed of either necessary or INUS conditions. We note here that INUS conditions could be considered ‘background conditions’ rather than the ‘cause’ itself. So the spark from an electrical circuit might be given as the cause of a specific fire in a house, but the ready supply of air is seen as a background condition since we would expect air to be present in a building of that nature. However, statistical differences in the propensity of fires in buildings of different types might be due to differences in the ready supply of air in their design. Buildings that are relatively compartmentalized might suffer fewer fires on average than those with a greater airflow. The cause of the difference in propensity to fires is therefore the

140 The Philosophy and Methods of Political Science ready supply of air. (Some causal particularists deny the distinction between background conditions and causes, but it is a natural folk or psychological category we impose on our explanations of causes.) Where the sufficient set for some outcome is composed of INUS conditions, there could be several different ways in which the same outcome could arise. Circumstances where there are different causal pathways to the same outcome are often referred to as ‘equifinality’ in the literature. The idea of equifinality is very important to many qualitative researchers, since they are interested in identifying the particular causal path or chain that led to some outcome in a given case, rather than the set of such paths. (I say more about equifinality when I discuss process tracing in Section 6.4.) Another version of necessary and sufficient conditions used extensively in jurisprudence is NESS tests (a necessary element of a sufficient set) (Wright 1985, 1988). Wright (2001: 1102–3) says, a ‘condition contributed to some consequence if and only if it was necessary for the sufficiency of a set of existing antecedent conditions that was sufficient for the occurrence of the consequences’. In both INUS and NESS tests the focus of the analysis is upon each element of a cause. Thus when discussing what caused some E, we consider each element that is part of the causal story, in the sense that if that element was not there, E would not have occurred (or E would have been different in some regard). I will reduce all such analyses to a simpler idea: the ‘but for’ (BF) test. This is popular in legal circles where argument resides over the nature and degree of responsibility. The idea is that E would not have occurred ‘but for’ C. Again we can note that not all BF conditions are ordinarily thought of as causal. Some are ‘background conditions’. What constitutes background conditions and what constitutes causes to some extent depends on the (causal) question we are asking. I will refer to such accounts in terms of necessary and sufficient conditions as BF, on the grounds that BF this causal condition or these causal conditions the effect would not have occurred. One can still assign degrees of responsibility, as several different factors can all add up to the BF, but essentially the idea is that a set of conditions determines an outcome. Contrariwise is the idea that causation is probabilistic. Any given X will, with a given probability, cause some Y. Qualitative researchers tend to think about causation in terms of BF conditions, whilst quantitative researchers tend to think in terms of probabilities. In standard linear regressions we measure the amount of variance we predict in Y given by variables in X, by each of {x1, x2, x3}. The amount of variance can be thought of as how much each of these variables brings to the party that constitutes Y, so to speak. It is easy to think that the variation measured is the probability of Y occurring given each element of X, but that equation can only be justified given a specific interpretation of the constitution of Y. Without some variables in X, then Y would be constituted differently.

Narratives, Mechanisms and Causation

141

Nevertheless, the probabilistic interpretation of causality is implicated in the statistical analysis. In the example above, the amount of variance in the supply of air affects the variance in the number of fires. In this example it might be natural to think that what is ordinarily considered as ‘the cause’ is what is captured by the BF condition; and what is ordinarily thought of as a ‘background condition’ is better captured in the statistical analysis. I return to this idea below. However, not all ways of thinking about causation would allow statistical analysis to be the ‘background conditions’. Think of the following example. Say a competent snooker or pool player would sink the object ball in the corner pocket 9/10 times given the relative positions of it and the cue ball. One way of looking at the sinking of the object ball is that the snooker player caused it to be pocketed with probability 9/10. Another way of looking at it is to say he caused it to be pocketed. It is true the player would miss 1 in 10 goes, but each time he fails we can explore the reasons why. For example, some failures are caused by a ‘kick’ on the object ball when struck by the cue ball; some by the slight ‘side’ the player put on the ball by accident; occasionally analysis suggests he has sighted up the angle incorrectly; we might note a bad habit he sometimes falls into with his cue action, and so on. These all add up to a 1/10 failure rate on that particular shot. Those who favour BF explanations will want to fill in these details, since they need a BF to explain the failure when the normal causal conditions obtain. Those with a more probabilistic bent might be happier with the probabilistic relationship. In part it may depend on what one is interested in. If one is betting on a game, all one needs to know are the probabilities; if one is training the snooker player to try to improve his game, the reasons for the failure become more important. We can consider the binary distinction here in terms of ‘granularity’ demanded of the information in an explanation. One way of thinking about the narrative for any given set of events is that the events are the equivalence class of a set of finer-grained histories with respect to the narrative offered. For any given narrative of a specific set of events, we are claiming that all equivalence classes will have the specific features we have denoted in that narrative. This narrative is contained of those events and those events only. For the snooker example, the narrative of equivalence classes is that the snooker player behaves thus and so, and 9/10 times he pots the object ball. To discover why he fails 1/10 times requires us to go down to a finer-grained set of equivalence classes. Thus the set of equivalence classes for failure due to unintended side will all have a specific set of features that lead to the unintended side, as will the equivalence classes where the player gets a kick on the object ball. These equivalence classes will also have finer-grained versions. The kick will not be the same each time; the unintended side will be greater or less, and might derive from a slightly different cue action.

142 The Philosophy and Methods of Political Science We might think that we can go all the way down to an equivalence class containing one member: that is, the member that only has the details of the actual event. However, the finest graining will be descriptive of that even all the way down to the level of quanta. But we find that there are equivalence classes here too, since at the quantum level everything is probabilistic. Thus the same event has numerous quantum possibilities. In one sense, therefore, it turns out that all causes are probabilistic. That does not mean that we cannot analyse effects in terms of BF conditions, but we understand that any BF condition is given as a condition for an equivalence class of the specific event. This does have some consequences, as we see below when we consider the equivalence classes in terms of counterfactuals – under the title of ‘the specification problem’. It also makes us realize that whether we look at any causal event in terms of probabilities or BF conditions depends entirely upon the specific point at which we stop examining the set of equivalence classes.

Regularity–counterfactual David Hume’s analysis of causation is ambiguous on this divide, as he was with our first. He seems to suggest that causation can be simply seen as constant conjunction, and also that causation is a counterfactual notion (Lewis 1973; Goertz and Mahoney 2012). Constant conjunction is associated with necessity when two things are always associated. Correlation is a similar idea but allows that the conjunction is probabilistic. Correlation is a statistical measure of the association between two or more variables. Variables are associated if they co-vary or change together. In statistics such association is called dependence, so a correlation demonstrates a lack of independence. Correlation does not establish causation: two variables might be strongly correlated because both are caused by some other variable. Whilst correlational evidence does not establish causation, it can be a strong indicator, and theory often assumes that correlations demonstrate causal effects. Or rather, a theory posits a causal relationship between variables, and correlations are thought to help establish evidence that the theory is correct. Indeed, the most obvious examples of spurious causation occur when we cannot think of a theoretical reason (or a reasonable mechanism) by which one causes the other. (Often people talk about ‘spurious correlations’ when they mean spurious causation. A correlation is not spurious if two items are indeed correlated because they are caused by a third variable. Spurious correlations, or spurious regressions, can be generated when the purported relationship is created out of the model specification and is not robust across different specifications.) At its simplest, a correlation of 1 between X and Y means that they are perfectly positively associated, a correlation of –1 that they are perfectly negatively correlated, and a value of 0 that they are independent. There are

Narratives, Mechanisms and Causation

143

many different correlational statistical measures in linear and non-linear, parametric and non-parametric forms. Whilst political science is turning away from the mere demonstration of correlations to infer causal relationship, they remain foundational and a major starting point for much empirical work. The move away from correlations is towards demonstrating counterfactuals and that involves a turn towards a different account of causation. In part, this is led by a move against ‘mere theory’ providing the causal element found in a correlation. As a reason, that is a poor one, though this is not to suggest that interrogating the counterfactual in experimental work is not advantageous. More importantly, though, it involves the idea that the estimated effects can be assumed to be the causal effects: that is, it is one thing to find Y correlated with X and assume X causes Y; it is another to say that the estimated difference in Y is caused by the estimated difference in X. The issue that arises in finding correlations across many cases is that this does not involve a direct observation of a causal effect in a given case. The counterfactual account relies on what would have happened to Y if we vary some aspect of X. However, for a given X we cannot both vary some aspect of it and leave X the same. This is called the Fundamental Problem of Causal Inference, because one cannot simultaneously both leave X alone (control for X) and change it (treat it). However, one can have a control set of Xc and a treated set of Xt over a population of X and estimate the average causal effect of the treated variable (Xt) over the whole population (Holland 1986: 947). That average causal effect then provides the counterfactual estimate for the individual case. With the experimental turn in political science, the counterfactual approach is coming to dominate statistical analysis. Goertz and Mahoney (2012: 79) argue that in fact this approach trades upon the idea of both the counterfactual and constant conjunction, for it is the repeated effects of X on Y that give the average causal effect. They treat the pure counterfactual case as existing in qualitative research, saying that the counterfactual is treated as a necessary condition such that if X had not occurred, then Y had not occurred. This is then generalized to all counterfactual cases such that if not X then not Y, giving a general statement that X is a necessary condition for all Ys, and then using the X’s presence in a particular case of Y as being a cause of Y (Goertz and Mahoney 2012: 80–1). Of course, we can treat X in a particular case as being a cause of Y where both are present, but the fact of it does not demonstrate the counterfactual and we cannot make an inference from one to the other. If there are no extant cases of Y without X then we might infer that X is a necessary condition for Y, though this would only be plausible with some theory that suggests X is necessary for Y. The problem for the Goertz and Mahoney argument is the specification problem. Good counterfactual inferences can only come from or at least be justified by statistical analysis for which there is empirical or theoretical justification

144 The Philosophy and Methods of Political Science (Dawes 1996). One way in which we might think about this problem is that ‘if only’ implies ‘because’ in the sense that ‘if only X had occurred, Y would have been different’, and that is to claim that the nature of Y is partly dependent on X. This is not the same as giving a necessary condition. To say ‘if X had not occurred, Y would not have occurred’ is to claim X is a necessary condition for Y. But the latter statement can only be derived from general principles – that is, theories or statistical relationships – because we simply do not know enough about society to make deterministic causal claims. One of Goertz and Mahoney’s examples is the democratic-peace generalization discussed in Chapters 4 and 5. They suggest that the relationship between democracy and peace is a necessary relationship. If that is indeed the case, then the relationship between democracy and peace would be a law-like generalization. But it is not. It is an empirical generalization. We saw in Chapter 3 (Section 3.6) (and see Chapter 8, Section 8.4) that one way of thinking about the difference between empirical and law-like generalizations is whether or not we can imagine (in a scientific not a fantasy manner) the generalization failing. We might say that the more invariant a generalization the more law-like it is, because the more invariant the less credible its failure. The degree of invariance might be based upon empirical evidence. However, some empirical generalizations might seem highly invariant. A relationship between democratic countries and their not waging war against each other has few exceptions. However, is it really law-like? Not only can we imagine democracies going to war with each other but, more pertinently, we can imagine conditions under which democratic countries might go to war with one another. Examining those conditions will get us close to the mechanisms that keep democracies from warring, and will get us closer to the law-like generalizations that make the empirical generalization we started with. One might suggest that we can imagine law-like generalizations not being true. After all, many well-attested laws in physics were discovered to be false; prior to those discoveries not only were other relationships imagined, they too were thought to be true. However, when I say that we can imagine empirical generalizations being false, I mean without shifting the boundaries of our knowledge by very much. Can we imagine water not being H2O? In Hilary Putnam’s (1973, 1985) twin-earth thought experiment, a substance precisely like water is made up of the elements X, Y, Z. This particular thought experiment might be justifiable for the point Putnam is making about sense and reference, but we can’t really imagine such a substance without reimagining the entire periodic table. And reimagining it, moreover, in a way that allows us to continue to rely on the evidence we have that water is indeed composed of H2O. For reimagining the periodic table to admit a substance that acts like water in all circumstances except under chemical analysis, where it turns out to be composed of three

Narratives, Mechanisms and Causation

145

unknown elements, would almost certainly mean that the current way in which we analyse the chemical evidence that water is indeed H2O would no longer be reliable. In reality we cannot scientifically imagine water not being H2O. To do so would overthrow far too much else. Goertz and Mahoney (2012) suggest that we can do such counterfactual analysis through case studies to find necessary and sufficient conditions. But without theory driving the analysis we cannot. And good theory is corroborated by large-n analysis. Why can we not do it through hypothetical counterfactual analysis? We can, if that counterfactual analysis is based upon strong theory and strong empirical evidence (generalizations), but even then we face the specification problem (Dawes 2001: ch. 7). In Dawes’s example we explain why someone took great risks on the stock market in order to avoid bankruptcy. We might plausibly generate such an explanation by the finding that people are generally risk averse except when facing a sure loss. But this plausible explanation has rivals: proud people might sooner gamble than borrow money from friends; or people are risk takers in some situations. As Dawes summarizes the issue: Even when we can start to talk meaningfully about ‘this here’ situation in terms of a hypothetical alternative, such a counterfactual discussion makes sense only if we believe there is some generalization that leads it to make sense. Thus, we can’t avoid the problem of specification, of deciding which generalization to reference. (Dawes 2001: 113, my italics) In other words, counterfactual analysis always requires generalizations to reference the changes in the specified conditions. We have to turn to quantitative analysis to gain an idea of which generalizations are best to reference in such cases. In other words, the plausibility of any given counterfactual is based upon a (large-n) generalization. Often there are too many sets of counterfactuals that can justify the inference – another way of putting it is that there are too many mechanisms that could be operating in a given case. With the democracies-peace generalization, there are many possible explanations of the relationship (many potential mechanisms) and that means that the counterfactuals that would support the breakdown of the relationship are underspecified. This is the specification problem. Only by examining when peace does break down can we begin to get a handle on which counterfactuals are the pertinent ones. If there really is no war between democracies, then we need to turn to non-democracies to see when peace breaks down and then use some theory (some imagination in our counterfactuals) to assess which are the best counterfactuals to make judgements about the mechanisms that lead to peace between democracies. The fact that democracy comes in degrees and forms enables us to make such judgements about the mechanisms that underlie the empirical generalization.

146 The Philosophy and Methods of Political Science

Generalizations–mechanisms We considered the nature of law-like generalizations in Chapter 3 when we looked at the failure of the DN model of explanation. We considered the difference between law-like and empirical generalizations, and followed Hitchcock and Woodward (2003) in suggesting that a better way of thinking about generalizations is in terms of their invariance. We also saw that a generalization might not always appear as an explanation (because we have to specify the relevant knowledge conditions of the person asking the questions), though it can certainly enter into one. It was the perceived failure of law-like generalizations, especially in the social sciences, that led many to suggest that mechanisms are a better model (Elster 1989; Dessler 1991; Wendt and Shapiro 1992; Hedstrom and Swedberg 1998). The idea of causal mechanisms is often closely associated with realism – the view that there are structures which tend in certain circumstances to lead to certain types of outcomes, but which are not nomothetic in the sense of strict invariant generalizations. Indeed Elster defines mechanisms in terms of generalizations with variance, saying that ‘if C1, C2,…Cn obtain, then sometimes E’ (Elster 1998: 48). These then simply become a form of generalization. Others want to avoid this close a relationship between laws and mechanisms, suggesting that the latter should be understood in terms of the causal power of events, conditions or structures to bring about outcomes (Little 1998: 197–8; George and Bennett 2005: 137). However, it is not really clear that we can explain causation in terms of causal powers. Causation needs to be explained in a way that does not use the term ‘cause’ or any direct synonym in the explanans. George and Bennett also utilize the idea of causal pathways, whilst some, for example Gerring (2008), define a causal mechanism in terms of ‘intervening variables’. Whilst tracking pathways might help us to understand how some outcome transpires, it does not really explain causation. After all, for every intervening variable that we add we could ask what intervenes between the added variable and the previous ones. We might call this the ‘bumpbump’ account of causation, as it seems to imply that for any causal pathway we need a set of variables crashing into each other like dominoes knocking each other down. However, we’ve known at least since Isaac Newton that we do not require such bumps to specify causation. Statistical correlations do not provide causal explanation, as we know from the example of the falling barometer reading and the onset of rain. To explain the correlation, we need to see how the barometer reading and the rain are linked. And of course they are linked by something which (is part of) the cause of each. Causal explanations of phenomena require the right sort of linkage. In one sense explanation by mechanisms provides that linkage. For some, the linkage is simply the specification of causal pathways. Gerring

Narratives, Mechanisms and Causation

147

(2008: 163), for example, explains a mechanism as the linkage (D) between the event viewed as the (initial) cause (C) of the event that is seen as the effect (E). But such explanations of mechanisms might always be underspecified, for we might want to know the variables that intervene between C and D, and between D and E (Waldner 2012). For any such ‘bump-bump’ account of causation, we can always request the extra bumps. One way around this is to recognize that some intervening variables add more explanatory oomph than others. Theoretically specifying which, however, is difficult. Again, the issue is that what adds more depends at least in part upon the knowledge of the receiver of the information. Waldner (2012) suggests that we think in terms of the added value of different models of causation. In his example we think of a ball breaking a window. By a regularity theory, the prior event of the ball striking the window with sufficient force provides the explanation; by a counterfactual theory, were it not for the first event the second would not have occurred; by a manipulation model, the events are linked by a change in the probability distribution of the second event when preceded by the first. According to Waldner (2012), each later explanation provides more information than earlier ones. An explanation by mechanism adds even more value by tracing ‘causal influences in a way that ultimately generates the second event’. So the mechanism explanation involves the ball’s momentum, the transference of force to the window, the glass fracturing by stretching and breaking atomic bonds, and so on. However, we notice that, as more and more information is provided, we get more and more invariant generalizations. The momentum of the ball will explain why the glass breaks only sometimes: when it is sufficient the glass will break. But what is the process of the breaking – and when is the momentum sufficient? The process is then the transfer of force from ball to glass (a law-like generalization); whilst the strength of the atomic bonds and the level of force explain the sufficiency. Waldner’s explanation of mechanisms as explanations is the provision of links that boil down to more and more invariant law-like generalizations. Objects break when the atomic bonds between their molecules are stretched and broken; atomic bonds are broken when sufficient force is placed upon them. Mechanisms end up as a series of generalizations fitted together to form a narrative that is explanatory. When background knowledge is assumed, we do not need to drill all the way down. We stop when, as noted above, we are satisfied with a specific level of granularity. We can conclude here an issue left hanging in Chapter 3 (the end of Section 3.6). Part of the problem of explanation by law-like generalizations is that the generalizations given in the examples are all too granular. Explanatory generalizations occur at very fine levels of granularity, and these will be descriptions and identity statements. At the finest levels of granularity

148 The Philosophy and Methods of Political Science they will be probabilistic. However, political scientists do not operate at that level of granularity. They can operate with mechanisms that make narrative sense. In other words, mechanisms are indeed bound together by law-like generalizations, but political scientists do not operate with those generalizations; they operate with mechanisms. These mechanisms at the finest level of granularity will be probabilistic (they have to be, since at the quantum level so are the generalizations), but we can treat (some of) them as being (almost) deterministic since they are, or rather underlie, empirical generalizations that we can discover with our analyses. We treat them as almost deterministic because at the level of granularity at which we operate that is what they appear to be.

Distal-correlational–proximal-serial The idea that causation has to involve ‘bump-bump’ accounts is clearly false. If Y always follows X and there is no W that causes both, then we can say (given some theory/narrative/mechanism) that X causes Y. There might be intervening variables, and there might be more to say via some mechanism or sets of base generalizations about those intervening variables. But these facts, if they are so, do not mean that X does not cause Y. The fact that there are intervening variables does not, in any way, undermine the claim that X causes Y. I label such causal claims distal-correlational. Some may be theoretical, in the sense that X causing Y is part of what we mean by those theoretical terms. The number of veto players being correlated with the degree of policy stability follows from the meaning of veto players and the assumption that the probability of an agent wanting to veto is independent of their number. It might not be thought that it is analytic in this manner that democracies are less likely to go to war with each other than non-democracies, but some simple assumptions about the behaviour of governments given the need for re-election and the nature of the international trade system generate the obvious correlation. With a detailed process-tracing account of how a specific causal path led from X to Y, one is giving proximal-serial causal explanation. This might be interesting in specific cases. As we saw in Chapter 3, sometimes we desire more from our explanations than mere statistical relationships. People tend to be dissatisfied if they are told the reason they have cancer is that the probability of getting cancer conditioned on their specific personal characteristics and circumstances is 1/114; they would sooner be told some aspect of their behaviour or physiology is correlated with that form of cancer. Similarly, we tend to be more satisfied with a detailed narrative of how some political event came about than with being offered some probability distribution of that event occurring in a country with those characteristics.

Narratives, Mechanisms and Causation

149

Proximal-serial causal explanations are often more psychologically satisfying than distal-correlational ones, though they might be less helpful for making future predictions. Thus detailed analyses of why a plane crashed might be less useful for future air safety than more general statistical analysis of the causes of plane crashes, or the causes of plane crashes of this type (Dawes 2001: ch. 7); though the detailed analysis of this crash (in terms of BF conditions) is more satisfying for the victims’ families. Political narratives are more satisfying for those interested in the particular outcome; though less useful, except as a data point, for more general understandings in comparative politics.

Production–difference-making In many ways this opposition (for example, Hall 2004) cuts across some of the earlier oppositions. The production side can be seen as the narration of process or a mechanism, whilst ‘difference-making’ is a counterfactual or regularity understanding, so it opposes two ‘oppositions’ with a third account. ‘Difference-making’ is not the same as ‘generalization’, however; it tends to be a rather broader understanding, if only because it can allow for quite weak correlations to stand as causal relations. The difference-making approach is interested simply in robust (even if weak) correlations, so here standard regression analysis can be evidence of causation even if it does not demonstrate it. Then, within that differencemaking approach, experimental counterfactual approaches can be used to provide stronger evidence (or demonstrate or allow the observation) of causation. The production side, as explained in Hall (2004), is the paradigmatic ‘bump-bump’ account, and in the social sciences tends to mean the careful description of each event leading to another, with the narration providing the production (or mechanism) of the account. One way of looking at the distinction is in terms of the level of social reality that we are considering; another way is to look at it in terms of token and type, the binary opposition I will consider next.

Token-particularism–type-generalization In my standing example of veto-player theory, policy change occurs in a political system when relevant actors agree on a change. Veto players are defined as agents who can stop policy change; all else being equal, the more veto players in a system, the more likely it is that one will veto policy change. Hence the more veto players in a system, the greater policy stability we expect from that system. Change is possible, of course, when all players agree. This theoretical mechanism is then applied to political systems. In general, presidential systems have more veto players than parliamentary ones (though there are exceptions), so we can generate empirical hypotheses about the comparative policy

150 The Philosophy and Methods of Political Science stability of presidential versus parliamentary systems (Tsebelis 2002). But what does this tell us about a particular change in policy? Saying that all the veto players did not veto because they were in agreement is a trivial explanation. Veto-player theory is useful for answering specific questions at the type level. It is a generalization – the greater the number of veto players the greater the stability of policy over time. It does not answer the sorts of questions we are interested in at the token-particular level. Indeed, even using the phrase ‘veto player’ in a specific historical description of the process of some policy change being blocked in a given country would be entirely otiose. That is where careful historical analysis or process tracing comes into its own. Such token-particular explanations can tell us why the relevant agents were in agreement, providing some background and perhaps detailed explanation of how some were persuaded to back policy change.

Explanations–inferences I have largely been discussing causation in terms of causal explanations. There is a movement in modern political science to displace causal explanation in favour of causal inference. The latter term is preferred by statisticians who use techniques that distinguish between inferences that mark associations and those that mark causes. Causal inferences are invoked when the dependent variable is affected when an independent variable is altered. This occurs most obviously in experiments, but causal inferences are also invoked in structural equation models between the different ways in which the independent variables regress on each other. An underlying assumption of the invocation of causal inference, however, is that causal inference is essentially an observational and not a theoretical process. In the experimental case we see causation through the manipulation of some of the conditions. Causal explanation invokes more theoretical considerations with regard to the interpretation of associations. However, we note that the observation in experiments is still theorized, since it invokes an average causal effect to provide the estimate for the counterfactual of any individual case. We do not just observe the causal effect in any given case. Whilst experimental techniques can certainly give greater confidence in causal claims, at base we still need to be able to make sense of inferences. Theoretical assumptions underlie the statistical analysis of causes and effects in experiments such as the Neyman–Rubin causal model (Holland 1986). Causal inference is theorized. That theorization involves the narration of mechanisms that make sense of the inferences we draw. Causal inferences are also invoked by qualitative scholars using process tracing, but again, when we observe any process and see in that process a cause or set of causal links, we are theorizing. In any narrative or interpretation of a set of events we have to do that. We can observe causation in

Narratives, Mechanisms and Causation

151

particular cases only given our species-history: we have learned to interpret evidence causally. Babies only a few days old seem to observe causation (or, rather, show surprise when causal paths are unexpectedly interrupted). That does not make the observation any less theoretical. As argued in Chapter 3, datum is already a theorized pattern. There is no theoretical difference between claims of causal inference and claims of causal explanation.

What is the correct way of viewing causation? Is there a correct way of viewing causation? It seems that the way people are trained leads them to think about causation differently. Amsel et al. (1991) asked people to evaluate a statement about whether events in a series are causally related or coincidental. In one (pre-digital-age) test, the subjects were asked about kicking a television and clearing the picture. Three ideas were put forward: the mechanistic, where the cause was specified by narrating physical causal paths (kicking the TV reconnected some wires); covariance, where the cause was specified in terms of types of causes usually associated with events and which can therefore be assumed as causes (kicking the TV usually clears the picture); and counterfactual, where the explanation relies on the idea that if one event had not taken place nor would the subsequent event (if the TV had not been kicked, the picture would have remained fuzzy). For all groups except the psychologists, invoking a mechanism was the most convincing; psychologists found co-variation the most convincing. For all groups except the lawyers, the counterfactual explanation was the least preferred. Training leads people to see causation differently. In my view, causation is so closely related to explanation that the same sorts of considerations come into it. We are satisfied by certain explanations given our background knowledge and interests. This will include what we select as important for our needs and for the predictions we want to make. Causal explanations are narratives, and precisely what we expect from a narrative depends on the question, and what elements of causal path we need filling in. That is not to say that many inferences we make, and many explanations that satisfy us, are bad or misleading. As we have seen, some poor reasoning has been offered by academics well versed and well studied in these matters. (And that, of course, includes me.)

6.4 Case studies and causation: process tracing We can corroborate or confirm causal claims with single case studies in the sense of increasing the probability they are true; and one can falsify or disconfirm in the sense of increasing the probability they are false. However, individual cases might not change the probabilities very much, especially in

152 The Philosophy and Methods of Political Science the former type. Crucial case studies can be important for falsifying even if, as I have suggested, they rarely if ever demonstrate that a theory is false. But can one actually show causation with a single case study? If we can observe causation then perhaps we can; but, as I have argued, causation is a theoretical concept and so the observation is already based upon generalizing across many other (similar) cases. If one thinks one cannot see theoretical concepts, then one cannot see causation. In that case, we say what we see is a series of events; what makes them causal is that we envisage the counterfactuals that would mean that, had some of what we saw not occurred, nor would what we saw later. However, we might say we can see causation all the time. But what does this mean? What we see are expectations based upon past repetitions of identical or highly similar cases. Some of these expectations are pre-programmed, having been seen by our species for millennia. Causal inferences always involve theory, even if it is not overtly specified. We are seeing theoretical inferences, and if we can see such things, we have no warrant for claiming we cannot see other theoretical entities such as the centre of gravity or the spatial position of a political party. I want to look a little more closely at the claims of some qualitative scholars that they can demonstrate causation in case studies in a deeper or more thorough manner than can be accomplished in quantitative studies. This claim is made most strongly in the work of Alexander George and Andrew Bennett (2005). There is much to recommend in the qualitative methods they describe, which they named process tracing; it has attracted a vast literature (George 1979; King et al. 1994; Brady and Collier 2004; Gerring 2007; Collier 2011). To some extent, process tracing is careful historical analysis, but in George and Bennett’s hands such analysis is propelled towards a more theorized and scientific analysis that often involves both quantitative and qualitative work in what is now fashionably called ‘mixed methods’. Students can learn much about how to go about qualitative work from George and Bennett’s book and I heartily recommend it. Here, though, I criticize their underlying justification for process tracing and their claims that, at least in some contexts, it is superior to large-n studies for identifying causal mechanisms. Single case studies cannot establish causation because of the specification problem. However, once potential mechanisms have been established (or plausibly identified) by large-n studies, careful qualitative work can suggest which of several potential mechanisms might have been operative in the particular case. This is what I will term the diagnostic use of process tracing. Case studies can also suggest hypotheses that can be formalized in deductive work, or examined and tested with further case studies and statistically in large-n studies. This is the hypothetical or conjectural use of process tracing (see Box 6.3). Both are valuable, but neither can establish causal mechanisms. I will return to

Narratives, Mechanisms and Causation

Box 6.3 Diagnostic Hypothetical or Conjectural

153

Uses of process tracing

Examining a set of political events to see which of several potential mechanisms seems to operate in this particular case Telling a story from a case study that suggests a potential mechanism for that type of outcome

these two uses of process tracing, but shall first examine in more detail the claims of George and Bennett. King et al. (1994) suggest that establishing a causal effect (through experiment) is logically prior to establishing a causal mechanism. What they mean, I think, is that one can identify a change in an outcome dependent upon a change in a manipulated variable without having any idea what the mechanism might be. We often theorize about mechanisms after we have the evidence, and then try to establish the mechanism as theorized with further experiments or analysis. On the other hand, we might theorize a mechanism without having a specific datum, and conduct an experiment to see if we can establish it. So I am not sure that causal effects are ‘logically’ prior to mechanisms. Furthermore, I suspect that most work that examines causal processes is conducted because we have ideas about those processes simply induced from the evidence we come across. George and Bennett have a different complaint about the claims of King et  al. (1994). They say King et al. conflate the definition of causality with that of causal effect: ‘The definition of causal effect is an ontological one that invokes an unobservable counter-factual outcome: the causal effect is the expected value of the change in outcome if we could run a perfect experiment in which only one independent variable changes’ (George and Bennett 2005: 138). Whilst King et al. (1994: 81–2) define causal effect epistemologically, it is easy, for a realist, to give an ontological version of causal effect as the value of a change in an outcome dependent on a change in an independent variable (or any number of them) (ontological), and then to say that can best be estimated by a perfect experiment (epistemological). George and Bennett (2005: 138) say a mechanism ‘invokes’ a causal process and process tracing is a procedure for identifying the ‘observable implications’ of such mechanisms. Box 6.4 sets out two ontological claims, causal effect and causal mechanisms, and two ways of measuring or identifying them. George and Bennett (2005: 139) suggest causal effects are found by experiment and mechanisms by process tracing. I have argued that mechanisms are theoretical accounts (narratives or interpretations) and they can be invoked by large-n quantitative studies just as easily as by single-n case studies, and indeed that we cannot establish causation by the latter.

154 The Philosophy and Methods of Political Science

Box 6.4

Experimental methods and process tracing

Ontology

Methods

Causal effects

Controlled experiment; quasi-experiment; natural experiment; various regression and other statistical techniques Process tracing

Causal mechanism

George and Bennett make two claims for the superiority of mechanism over finding causal effects: a contiguity claim and a reductionist one. The contiguity claim is the idea that looking at the detail of events will fill each of the links between an A and a Z so the full causal story is told. This is not simply a reductionist claim. It goes further. Reductionism is the idea that at finer grains of description we can see deeper into the mechanisms that bring about outcomes. So we might note that heating the air in a balloon will cause it to expand. Now we can explain that expansion by the mechanism that heating excites the air molecules, causing them to move around more, and this has the effect of expanding the volume the air takes up at any given moment, putting pressure on the fabric of the balloon, which increases its size until the pressure on the inside and the outside of the skin equalizes. We could reduce further, but we stop here. However, this reductionist claim does not involve any particular description of the movement of molecules. It is not applied to a single case study, but to all balloons under the relevant conditions (such as constant pressure outside the balloon). In other words, we do not need to look at the actual (token) movement of air molecules to understand why this balloon expanded. The general type of explanation provides an explanation for this balloon’s expansion once we see that the air was heated. In George and Bennett’s defence of process tracing, they say we need to look at the actual causal paths between A and Z. They suggest there might be different causal paths. To be sure, the actual air molecules in any one balloon will not move around in identical fashion to those in another balloon. However, that is irrelevant to the reductionist explanation of why heating air causes it to expand. Thus the fact that actually there are different paths from A (heat) to Z (expansion) – due to variation in the movement of air molecules – is irrelevant. In other words, reduction does not involve reducing type explanations to token explanations. It involves reducing type explanations at one level to type explanations at a reduced level. In one of George and Bennett’s examples, we are told that there might be many different explanatory paths, combinations of factors or sequences of events that deter different democratic nations from going to war with

Narratives, Mechanisms and Causation

155

each other. The democratic-peace theory has different explanations. True. However, any such reductionist claim would only require the type democratic-peace claim to be reduced to different type claims. The different sets of possible combinations of factors that can lead to the same outcome George and Bennett label equifinality, which they define as ‘many alternative causal paths to the same outcome’ (George and Bennett 2005: 10). However, whilst there might indeed be many different causal paths to the same outcome, if it is the case that A always leads to Z, then we would expect, on reductionist grounds, that there will be some underlying mechanism (see next paragraph). We would expect that those different combinations of factors or sequences would include something that brought them all together. But, more to the point, if that were not the case, if indeed the democratic peace is underlain by different mechanisms once considered in more detail, those different mechanisms can only be established by large-n analysis. Case studies can identify hypotheses, but they cannot demonstrate causality because of the specification problem. In fact, equifinality is usually formally defined (in the original open-systems account of Bertalanffy (1968) and others taken from the biological sciences) in terms of system predictability: a system whose behaviour is predictable from more than one preceding system. (We can see this in terms of different starting points leading to the same type of outcome.) Moreover, that predictability depends on one mechanism. In biology, of course, that mechanism is natural selection. So in the democratic-peace example, the argument would proceed from seeing lots of different histories that lead to either war (between non-democracies or between non-democracies and democracies) or non-war (between democracies) and the reasons will be sought in the nature of democracy. There might well be different variables involved, but those variables will be tied together by the nature of democratic systems. Equifinality is also taken to mean that the same output can be achieved by different inputs through a different process; but here we would expect to identify different processes through theory and through large-n analysis. Case studies can only provide hypotheses to be examined more carefully. The logic of equifinality accounts is that of reducing complexity to simpler mechanisms. George and Bennett seem to reverse that logic to argue that we need to understand the complexity to provide grounds for the simple. They confuse this with the irrelevant bump-bump account of different causal narratives. It is to these I now turn. I stated in Chapter 3 that there is an important difference between type and ultimate explanations and between token and proximate explanations. At times we want – that is, we are psychologically satisfied by – token and proximate explanation. Sometimes we do not want to learn just that democracies do not go to war; we want to know precisely how tensions between Pakistan and India (which have sometimes led to war) were avoided on

156 The Philosophy and Methods of Political Science specific occasions. Being told they were ‘both in democratic phases’ at those times is not sufficient. We want to know the detailed story of who did what and what systems were employed to reduce border tensions. In their careful account of different ways of looking at reductionist claims, List and Spiekerman (2013) give the example of the Copenhagen climate summit of 2010. Narrating the breakdown in talks might include Wen Jiabao withdrawing to his hotel room, Nicolas Sarkozy losing his temper, Angela Merkel being sidelined, Barack Obama cutting side-deals. However, without establishing the relevant counterfactuals, all these events are of limited utility in causally explaining the summit’s failure. That is the specification problem and it cannot be solved easily. A more general explanation of the failure might involve insights such as the number of actors, the competing interests, the unhelpful legal framework, and so on; factors relating not solely to this summit, but generalized from many cases and then applied to this example. These structural (type) reasons are better candidates for providing the explanation. One response of process tracers to this dismissal of the detail is to say, ‘but the actual events tell us the actual process of the cause of the breakdown’. However, the specification problem arises. If all the events narrated had not occurred and the talks still broke down, then they have not pinned down the explanation. Indeed, some of the narrated events are only relevant because they are of the type that is part of the causal explanation. We infer which are relevant from our background understandings of similar occasions where episodes like these unfold. In other words, narrations gain plausibility because they are instantiated within broader understandings. Since a detailed historical account often satisfies our psychological desire to see what happened in an actual case, such a narration can suggest that one mechanism rather than another was at play. This is where process tracing and careful historical analysis add real value to explanations. Process tracing can be used diagnostically in a particular case to establish that some hypotheses regarding ‘democratic peace’ seem not to be satisfied, while others are. There are different paths to the same outcome and evidence is provided that, in this case, one path is more important. How is this diagnostic justification of process tracing different from George and Bennett? In one sense it is not: they discuss the utility of process tracing in these terms (without calling it diagnostic justification). It differs in making no claim that this is a superior method of identifying causal mechanisms. Its justification is that it suggests which mechanism is involved in this particular case. Moreover, it does not establish it on its own. It only plausibly suggests it, given (often tacit) evidence from other cases. I think that is justification enough for a careful historical case study. I do not think we need to try to provide grander philosophical justifications and start making claims for some superior method of identifying causation. Any such attempts will

Narratives, Mechanisms and Causation

157

fail because of the specification problem. In order to identify and pin down a causal mechanism, one needs more than one case. A case study can, however, inductively suggest mechanisms that have not previously been considered. This is the hypothetical or conjectural use of process tracing. Both diagnostic and conjectural claims could operate in the same case study. The case study might suggest that one particular (rather than another) mechanism was operative, but also that perhaps some third mechanism might have been at work. This can then lead to some further (qualitative and quantitative) work to try to establish that possibility. At base any case study is simply data for a large-n study. That does not mean it is not important. After all, both you and I are data points for all sorts of social scientific claims once we are entered into someone’s spreadsheet, but that does not make the narration of my life any less important to me and to my kith and kin. Often we require the narration in order to feel satisfied that we have understood some set of events, but establishing that those narrated events are causal will rely upon large-n analysis. We can only ‘see’ causation in any given case because our ‘seeing’ is based upon large-n previous perceptions, both from us as individuals and from us as a species, since it was all those previous events that led us to see those events in the way we causally narrate them. But, as all lawyers know, observational testimony is the weakest form of evidence (though often weighted most heavily by juries), which is why we have developed more sophisticated ways of seeing the social world in our developed social science. That does not besmirch qualitative evidence; it merely puts it into proper context and perspective.

6.5 Conclusion If one thinks that causation can be established with constant conjunction plus some model or mechanism that tells us which conjunct is the cause and which the effect, then one may believe that large-n regressions can pin down causation. They can demonstrate association between dependent and independent variables that, together with a theory, an account, narration or mechanism, plausibly or less plausibly gives us the causal account. In many ways such regressions are simply the collation of descriptive narratives shorn of those elements deemed to be irrelevant. Usually we demand stricter conditions for establishing the counterfactual conditions in order to make the claim that causation is demonstrated. Note that regularity and repeatability are required to establish counterfactuals. One case, even in controlled conditions, will not be accepted if it cannot be replicated. Neither way, however, distinguishes between ‘background’ and ‘causal’ conditions. That is, if we narrate a specific case, we might see some elements as ‘background’ and others as the ‘cause’ depending on the specific question being addressed.

158 The Philosophy and Methods of Political Science One reason we tend to like case studies is that we might observe the actual cause – the specific process in a given example. That observation can only be assumed if we have already established a mechanism in such cases and so can be confident in our counterfactual presumption. Often, however, especially in public policy, we want the background conditions. We want to know how to reduce the propensity of buildings to catch fire (given the conditional probability of dropped matches). Or we want to reduce the probability of dropped matches (or more plausibly, electric sparks). We want the probabilities associated with the starting of wars or with policy stability. In his blog Phil Arena says (thinking about Neyman–Rubin causality) that it is wrong to think the cause of death of someone who was shot is that person’s failure to wear a Kevlar vest (Arena 2011). He’s right, of course: if we want to prosecute the shooter. However, if we are trying to reduce death amongst security guards, the lack of a Kevlar vest might be the pertinent variable. It might also be the pertinent fact in an insurance claim by the victim’s family (if he was required to wear one, but on this occasion did not). The problem perhaps is the attempt to give a once-and-for-all definition of causality when, in fact, what we consider to be causal depends on the questions we ask. Given the contract agreed, the insurance company is interested in the lack of the Kevlar vest, not the identity, motives or specific actions of the shooter. In terms of the death, the vest is the causal variable they are interested in. The counterfactual they are trying to establish is that if the victim had worn the vest he would not have died; so the contract is void. I have tried to demonstrate that causation has many aspects. I introduced those aspects by contrasting sets of binary divides. Some of these divides are very similar, but one does not have to plump for the same side of each of the divides in order to defend a plausible account of causation. Causation has been defended in many different ways. Those different accounts are often plausible, especially in the context of examples described within the justification of the account. Because people have different accounts, they might be susceptible to different types of evidence in causal claims. However, under any account one cannot justify causation with a single case because the specification problem (or narrative fallacy) gets in the way. When we believe that we can observe causation in specific events, it is because we have been programmed to read them that way because of the large-n of cases that our species (and forebears) have collated over the millennia. We are seeing patterns in the data, some of which, given a specific perspective and question, we call causal. It is by those constant conjunctions in cases like these, and different outcomes in cases relevantly different, that we think we have established the counterfactuals. There are more formal ways of trying to do so, in multi-level statistical analysis and in experimental and quasi-experimental methods (using matching, discontinuity design, and so on). What we see is still theorized – causation is a theoretical

Narratives, Mechanisms and Causation

159

notion – but the justification for the theory is the prediction based upon that repetition or large-n. The specification problem for single cases also applies to simple regression. The critique that simple regression fails to establish causation by failing to establish counterfactuals is a version of the specification problem. At base the specification problem is that several different causal stories can be told of the same evidence. The fallacy is that this narration is the only one that is applied to the pattern identified when other interpretations are plausible. The problem with simple regression in establishing causation is the same. There are (or might be) other interpretations of the regression pattern. Usually when simple regression evidence is taken as causal, it is because it seems to fit with what we expect given lots of other evidence, and this is so with causal claims based on case studies. Either way, the plausibility of a causal claim is based on the plausibility of rival narrations; that is why we sometimes accept simple regression or case study evidence. Though equally, when we do, we plan further study to establish the case more plausibly still. But the plausibility of such evidence is not the best justification for qualitative work. That comes from interest in the case itself; either diagnostically or conjecturally. Token and proximate explanation can be justified for its own sake.

Chapter 7

Methods and Methodologies

7.1 Introduction Within the frame of organizing perspectives one might have many different methods at one’s command. In fact, whilst some ways of looking at the world suggest specific questions that require particular methods to address them, the results from any method might be utilized within any organizing perspective. In this chapter I will briefly describe some of the major methodologies of political science and the sorts of questions they can address. Some people claim that different methods and methodologies belong to separate accounts of social reality: they rely upon different epistemologies and ontologies (Marsh and Furlong 2002). To some extent this might be the case – if only because analysts think this. In terms of ‘what there is’, there are two issues. One concerns how we treat theoretical entities. In what sense, if any, do they exist? In political science the concern is motivated by whether simple descriptions of surface reality or less obvious descriptions such as correlations are enough. I have suggested that descriptions do answer some of our questions and are therefore explanatory. However, we look for deeper explanations that examine functional, structural and institutional relationships. Whilst we seek empirical generalizations, we also want to provide mechanisms to understand the underlying structure and causes of empirical generalizations. Furthermore, I have argued that these relationships, albeit theoretical, exist every bit as much as surface reality. Indeed, to the extent that such deeper relationships help us predict events, they can be thought of as more real. Much of the literature that discusses ontological and epistemological issues in political science does so in order to challenge ‘everyday’ understandings of reality. To some extent, of course, the claim that difficult-to-observe relationships, generalizations and mechanisms are ‘more real’ is also a challenge to everyday understandings of reality. However, for relativists (see Chapter 2) the challenge is posed the other way round. They challenge the very idea of an objective reality, and argue that all reality, or at least all social reality, is in the mind of the perceiver. At one end of the spectrum lie constructivists, who see social life as constructed by people. Constructivism is perfectly compatible with the realism

160

Methods and Methodologies

161

I have described. A convention accepted by all members of a group exists precisely so everyone can predict what others will do and react accordingly. When two groups, each ignorant of the other’s conventions, meet neither can predict how the other will react; but a third observer can analyse their interactions in light of their differing constructions. There is nothing anti-realist in such a picture. The constructions are real in the way they affect behaviour (again, see Chapter 2). At the other end of the spectrum are radical relativists, who see no reality at all. One cannot engage with them in conversation; and they should not engage in it, since for them conversation can have no meaning across people. Habermas (1987) criticizes such post-structuralists for their ‘performative contradiction’: that is, their hypocrisy in not living their lives or engaging in their work in ways in keeping with the theories they espouse (see also Matusik 1989). Another ontological issue is over the nature of causation (the subject of Chapter 6) – whether causal relations are what we impose on the data or whether they are, so to speak, within the data. Again, the type of realism I have been espousing suggests this is a false contrast. Causation is a theoretical concept, but given aeons of evolution where thinking about it has been to our advantage, we observe it around us. Going beneath the surface, we find other relationships that conceptually seem causal. This leads us to epistemological questions. Again, attitudes towards the nature of causation, what counts as evidence for causal inferences or explanations, and how to weigh different types of evidence are implicated in researchers’ epistemology. I prefer to see such ‘epistemological’ issues as related to the type of questions one is asking, and believe the different ‘epistemologies’ concern different interests. I am sceptical about the claims of different epistemologies because often their approaches are not contradictory. They are simply asking different questions, and can be critiqued in terms of their own evidence. What matters is what counts as justified belief and if something counts as justified belief it is part and parcel of our epistemology. We can argue about that, but political science should really concern itself with how well claims are substantiated and maintained by the evidence provided. Let the evidence and the arguments stand up to their inquisitors. If the issues matter, the truth is likely to out. If they do not matter, then it might not, but we need not worry too much about that. As I briefly discuss each method of analysis, I will indicate the kinds of issues with regard to causation and evidence they bring up. I have pressed for a more general account of conceptual realism that allows for different sorts of questions and different accounts of the same objects of study but which are non-contradictory. In that account the different methods and methodologies can sit side by side. To some extent, this fits with the recent surge in justification of ‘mixed methods’ (though many of us have always used them). That is, different methods can be employed to

162 The Philosophy and Methods of Political Science grapple with different aspects of the same political enquiry to give a more rounded and complete story. Usually mixed methods employed by the same person or team are designed to give a fuller causal story of some political process or institution. However, there might be different, though non-rival, accounts of some political process that involve very different ways of approaching it. They are non-rival in that they are not attempting to explain or causally describe that process; rather, they see different objects of interest in the phenomena. They use different methods, since their aims in the study they are attempting are very diverse. In other words, pluralism in research does not imply we all have to be pluralist, but simply to accept that other researchers approach the topic with different aims and use appropriately different methods to try to answer their questions. Moreover, there is no need in a pluralist discipline for everyone always to adopt several methods. A single method may best address some specific questions. The big divide in political science is between quantitative and qualitative methods. To some extent, as discussed earlier, the modern debate was set off by King et al. (1994) and their claim that there is one logic of inference. The underlying implication of the logic of inference is that only quantitative evidence can (a) pin down causation and (b) test hypotheses drawn from theory. One response has been that qualitative evidence can pin down causation by (c) filling in the gaps and helping to demonstrate actual mechanisms and (d) using a different model of causation. Interestingly, partly as a result of the debate, quantitative studies have come under increasing scrutiny of their causal claims, leading to new statistical techniques and the experimental turn. In Chapter 6, I attempted to do two things. First, to deflate the importance of pinning down causation, by arguing that whilst, of course, we are interested in causation, not all explanation is causal. In Chapter 3, I argued that we cannot have a single ‘logic’ or ‘model’ of explanation since explanation is, at least in part, a psychological demand. What counts as an explanation is what satisfies that demand, based on the context and precise nature of the question and the knowledge of the person posing it. Finding patterns in the data is explanatory; they can include descriptions of the functions of aspects of systems and interpretations of actions. None of these is straightforwardly causal. Indeed, many of our theories about the world, including evolutionary or structural or institutional accounts, are not straightforwardly causal. In Chapter 6, I suggested that what we see as ‘causes’ and what as ‘background conditions’ vary depending on the question. It is for that reason that qualitative studies, which tend to concentrate upon the token and the proximate, examine certain elements of the narrative, whereas quantitative studies, tending to concentrate upon type and ultimate accounts, often focus on other elements. This is sometimes represented as the difference between the causes of effects and the effects of causes, but it is better seen in terms

Methods and Methodologies

163

of type and token; ultimate and proximate. Quantitative methods might be seen as specifying structural conditions that condition outcomes given specific (proximate) processes that cause them. And of course, only quantitative studies can find empirical generalizations. When we think of probabilities conditioned on characteristics, we are thinking of what difference those characteristics make overall. The ‘difference’ can be regarded as causal, but if one is thinking in terms of the narrative of a specific case these conditions might be seen as the ‘background conditions’ affecting the particular causes. The ‘ontological and epistemological’ issues that vex some writers derive their force from the belief that some organizing perspectives and the methods associated with them are ideological in some manner. The Marxist and critical studies tradition suggests that mainstream political science exists to support the dominant ideology and dominant groups in society. I doubt that; though I am quite sure that dominant groups will use whatever means are available to maintain their dominance. It seems to me that the groups and organizations who have furthered their own aims most effectively by using lessons about the power of language delivered by critical discourse analysis are multinational firms, media organizations and governments. Arguments and evidence, once public, can be used by anyone, so no part of the profession is immune from helping dominant groups. What is undoubtedly the case is that academics tend to ask questions that motivate them. Motivation can come from many sources, some of which might indeed be ideological – and from any point along the ideological spectrum. Other questions proceed from fashion and career concerns, others still from sheer inquisitiveness or the belief that these questions are the ones of greatest importance to our current predicaments. It is perfectly legitimate to query why someone is asking these questions and not others, but here I am more concerned with the relevant methods of finding evidence to answer questions.

7.2 Qualitative and quantitative research Following Dennett, I have been suggesting that what we see through our research is patterns in the data. I have also defined data as any reasonably systematic information about the world. Data are already patterned, in the sense that what we see as a datum is already an interpretation. Both that datum and the patterns we see in data help predict in the sense I have defined. We might see patterns that ultimately are not predictive. The predictive ones are the real patterns. Reality is defined by the constraints it imposes on what we can do. This means that surface properties are not necessarily as real as underlying or deeper ones. Science is about finding patterns that we could not see without the techniques we adopt. This means that structures and

164 The Philosophy and Methods of Political Science

Box 7.1

Objective and subjective

Objective

Subjective

Refers to objects e.g.: The PM is a member of the Conservative Party Consumers respond to price changes

Refers to subjects e.g.: Mary thinks the PM is a buffoon Mary thought the chair was good value for money

theoretical concepts are not only as real as surface observations, but might have a stronger reality. One of the divides in political science methods is the perception of the relative worth of objective and subjective data. The distinction between objective and subjective data concerns whether a claim refers to an object or to a subject: that is, a person (see Chapter 2). A person’s responses to a survey question are subjective data; descriptions of how many people give particular answers are objective data. Explaining the behaviour of a public servant through their beliefs is a subjective explanation; explaining it in terms of the incentives that people tend to respond to is an objective explanation (‘subjective’ here should not be confused with relativism: see Chapter 2). In both cases, the evidence is the analyst’s interpretation. Subjective data are the subject’s perceptions as interpreted by the analyst (no matter how close and personal the analysis becomes). Similarly, objective data are characteristics interpreted by the analyst. For some people the key elements in explanations are the subjects’ interpretations. We need to be able to understand what people think they are doing in order to explain what they are doing. In one sense that claim has to be true. Actions are typically defined in terms of people’s reasons for action (Davidson 1980). However, if we can better predict what someone will do by giving an explanation that is not couched in terms of that person’s beliefs about what they are doing, we have found a better explanation. Usually, subjective evidence helps us to interpret objective evidence, but where they conflict we need to privilege the most predictive. One of the problems of political science is that because its subject matter includes such important items as justice, war, democracy, human rights, and so on, some people consider that our methods cannot be objective, but are ‘inherently subjective’. That is to misunderstand the objective/subjective distinction (see also Chapter 2). What they mean is that because people have different value systems, they make different moral judgements. As political scientists, we can measure those different moral judgements across people; we can be aware of our own, and know that they affect what we think it is

Methods and Methodologies

165

important to study. What we should not do is to allow our own values to infect the methods that we use or the inferences that we draw. If one believes that differences in value judgements do infect what counts as evidence, then one is relativist, not a subjectivist. Most of the methods I discuss below actually use both objective and subjective evidence. Even the most interpretative of scholars use some description of the objective characteristics of their subjects; and outside of the most macro-sociological or macro-economic analyses, social data are typically viewed through the lens of how they impinge on human action. Analysis with some micro-foundations will almost certainly have some subjective elements to it. The big divide in political science methods is usually considered to be between those who manipulate quantitative, large-n data and those who manipulate qualitative or small-n data. In fact, divisions within these camps are just as sharp. Some quantitative scholars are theory-driven, attempting to corroborate or falsify hypotheses drawn from models, whilst others eschew such model-driven techniques in favour of more inductive methods. Some qualitative scholars are highly theoretical, using high-flown theory to drive the search for and dictate the nature of evidence used, whilst others are more descriptive, again eschewing theory altogether. Theory-driven scholars often use both quantitative and qualitative evidence; and behaviouralists use both quantitative and descriptive evidence. Many of the big debates cross the quantitative–qualitative divide, and rightly so, for they involve issues that concern both type and token explanations. We are interested both in the large structural issues that affect the types of outcomes we see and in the particulars and the proximate causes of specific outcomes. I have argued that small-n qualitative data cannot pin down causation. It can diagnostically suggest that a particular causal process was operative in a specific case. Even here, though, because of the specification problem, it cannot demonstrate causation. We noted that the experimentalists’ critique of standard regression techniques’ inability to pin down causation is also due to the specification problem. The problem therefore is not simply whether or not one is utilizing qualitative or quantitative evidence, but what it is we can conclude from the evidence given the constraints on our interpretation of it. I also argued (in Chapter 5) that we are not able to confirm or disconfirm models based on a single case. Single cases provide confirmatory or disconfirmatory evidence, no more. An emerging method of synthetic control tries to bridge the gap between quantitative and qualitative research (Abadie et al. 2010, forthcoming) by attempting to synthetically create a counterfactual world to that of the case under study, in order to estimate the difference some major event has made. So a set of units that approximate the unit under study by structural characteristics is selected. The set is then assumed to consist of a panel that can

166 The Philosophy and Methods of Political Science be observed at a set of time points. The set is then used to create a synthetic counterfactual of the case under study to be compared to the case study after the major event has occurred, where the synthetic case is estimated by changes in the approximated units. So in Abadie et al. (forthcoming) the effects German unification had on the economy in comparison with a synthetic West Germany are estimated. This method of utilizing quantitative methods to study single cases is useful only for estimating the effects of major events. In other words, whilst it is an important new technique in our armoury, it cannot address many of the questions we are interested in as political scientists and policy analysts. Nevertheless, it does enable more thorough claims about the effects of singular events in specific cases. Its quasi-experimental form enables the statistical evidence to test particular claims about singular events (or events in general that have the form of that singular event). We can note that the Bayesian critique of Fisherian ways of interpreting statistical evidence concerns how far such interpretations can be taken to confirm or disconfirm hypotheses that have been drawn. To be sure, there is a distinction between qualitative and quantitative evidence, but when considering what we can do with our evidence, no matter what form it comes in, we would be better off considering the inferences we can legitimately draw from it given the interpretive possibilities open to us. However, the most important element of our evidence for rendering the study of politics scientific is making it available for others to manipulate and interpret. The most important move in science (not only political science) in the last decade or so is the emphasis on ensuring analysis can be replicated. This is just as important for interpretative qualitative analysis as it is for quantitative.

7.3 Data access and research transparency New guidelines for making data available and being more transparent about the process on which research conclusions are based were adopted by the American Political Science Association (APSA) in 2012 (see Lupia and Elman 2012). For quantitative scholars, this entails making available the coding frames, advice to coders, original data, original survey material, and original program files used for analysis. This enables others to replicate the results using the same statistical models to check there are no errors, to examine model specification problems, check robustness across similar models, and so on. The interpretation of the data must also be replicable, thus requiring access to the details of the coding frame, precise codes or procedures and the particular statistical models that were used (Dafoe 2014). This openness enables other researchers to see how good the data are, whether there are too many random or systematic errors that impugn the

Methods and Methodologies

167

results. It also means that they have bedrock on which to place new data, or similar data from other countries or systems, through which to check the causal claims and mechanisms that were inferred from the results (King 1995; Lupia and Alter 2014; Dafoe 2014). Experimentalists need to record carefully all stages of their experimental design, and make available all the original material recorded as part of the experiment, as well as all aspects of the analysis (McDermott 2014). Qualitative scholars also need to provide their materials for similar replication, making available the full transcripts of interviews, details on archival materials consulted, links to archives online, notes made during shadowing exercises, and so on (Elman and Kapiszewski 2014; Moravcsik 2014). This enables others to go back and examine how far qualitative scholars cherry-picked what to highlight in their narratives from the materials at hand. Just as the results of statistical manipulation can produce bias through systematic miscoding, poor model specification or missing data, so too can qualitative scholars ignore or play down evidence in their materials, highlight aspects that fit with the narrative, or not consult pertinent material. Indeed both quantitative and qualitative scholars ought to make clear what they discarded – what statistical models were tried on the data but did not make their way into the published articles, which archives were consulted but not quoted or referred to. The guidelines for replication are highly demanding, but young researchers should get into good habits immediately, and older ones need to conduct their research more carefully and professionally than perhaps they have been used to. Scientific practice might become more demanding still. The well-known problem of publication bias against null findings tends to mean that researchers may run data through different statistical models until they find positive results (King 1986). One way of overcoming this is to adopt a technique becoming established in medical research (Humphreys et al. 2013): to announce in advance precisely what hypotheses one is going to test, what data one is going to collect, and how that data will be analysed, thus precluding fishing for good or positive results. One of the objections to making data available is that those who have put a great deal of effort into collecting them will not be credited. They might be scooped. This issue is particularly pertinent to young scholars, doctoral students who do not have the time or personnel (younger colleagues, research assistants, PhDs under their supervision) to carry out further analysis, collect supplementary data, and so on. One solution is for data to be made available only on publication, allowing time for the original scholar to carry out further analysis (Lupia and Alter 2014). Even then, if your data are made available, others might copy them and pretend they are original. After all, many data sets are compiled from readily available sources. (How would you know if your data had been

168 The Philosophy and Methods of Political Science stolen after they are made available? One well-known political scientist (so he informed me) includes systematic errors in the data he makes publicly available, and only points them out to those who apply to him for use. Those who use the data without acknowledgement could easily be caught once the systematic errors are found in ‘their’ data.) However, the biggest problem with data collection is not theft, but spending time amassing in much the same form data that have already been collected by others. We need to share more, not less. At present many data archives insist that use of their data be acknowledged, but perhaps more is required for dedicated datasets compiled by individual researchers. Similar practices could be adopted for code (Lupia and Alter 2014: 58). My preferred solution is to follow practice in some areas of medical science: make data readily available, but include on any publications the name of the person who collected the data (if that person wants their name included). When one requests data of some medical researchers, they demand to know precisely how they are to be used, what the research questions are to be addressed, what sort of modelling will be done, and so on before they will release their data. They want to see if any research results are likely to be of good enough quality for their name to be on the piece. And why not? Data collection is time-consuming, and if such teamwork were to become the norm, then data might be made available more quickly. Collectors would be more willing to share their data with others, especially those with greater resources. It would also be more likely to lead to collaborative rather than destructive work. New modelling might throw doubts on previous results, but be less likely to lead to bad-tempered debates if that work included the original researchers or some subset of them, who in later publications recognize the limitations of their earlier analyses. Another problem is the growing market for data, where organizations collect and then sell their data with the proviso that it not be made public as this would undermine its proprietary value. Researchers need to be aware of the problems of buying such data, since it might restrict their ability to publish in the best journals – though it is also the case that many multinationals have data available of which social scientists can only dream. There is a danger that the companies might well learn more about human behaviour than academics ever will and keep that for themselves. There are problems for qualitative scholars too: issues such as confidentiality for interviewees, the non-precise replicability of anthropological observations, anonymity of survey respondents, and so on. There are ways to overcome some of these problems. Anthropologists keep dedicated notebooks (and these can be made available), and all qualitative researchers should keep daily logs of what archives they consulted and when; who they interviewed and for how long, what books they consulted. My understanding from anthropological colleagues is that such notebooks are less readily

Methods and Methodologies

169

available than was once the case, as the pressure to publish leads scholars to write up as much of their material as possible for publication. Again, we need to acknowledge at an institutional level the worth of the data rather than just recognizing the worth of the analysis therefrom. There will always be some reasonable limits upon complete replicability. Nevertheless, no matter what one’s preferred methods and analysis, the aim should be to make one’s research fully replicable.

7.4 Policy-oriented research Some academic research is deliberately policy-oriented, often targeted at specific journals read by practitioners as well as academics. There are important differences between policy-oriented and more directly academic research. First, policy-oriented work is often directed at specific problems and is not theorized, nor expected to explain institutions or outcomes. It is problemdriven. We should not make too much of that contrast. Much pure academic work is problem-driven too, but perhaps the problems are not so immediate. Second, policy-oriented work is often funded directly by government or other organizations. Hence it is not academics, but practitioners who drive it. This should not be a problem, except that the funder often wants results more quickly than is conducive to good academic work. The best academic work is iterative. Data are collected, written up, presented, criticized, more data are collected, reanalysed, re-presented, sent out to journals, critiqued, revised, and so on. It takes a long time. Policy-oriented work often finishes with the first stage. To be sure, it may build upon earlier academic work, but the specific work directed at the problem addressed only goes so far. Third, academic work is essentially critical. It examines processes and tries to explain what is happening. Problem-driven academic research is aimed at characterizing problems and seeing how they arose. It is not directed, usually, at providing solutions. Policy-oriented work, on the other hand, might also look at how problems arose, but is focused on providing solutions and suggesting better ways of carrying out policies. However, in order to see whether these suggestions really work, we need time for them to operate and for problems to emerge. The policy-oriented work finishes its job before pure academic research has any data to get started. In my experience too, we never get to see how academic policy-oriented solutions would work, because government selects those elements it likes, can afford, or that fit with its ideology: a pick ’n’ mix approach which often results in a worse policy than existed previously. Finally, academic work is essentially sceptical. Academics, when they are being honest, will always have doubts and see potential problems in any suggested policy reform. Politicians cannot survive on doubts. If you are

170 The Philosophy and Methods of Political Science going to influence government policy, you have to present work in a positive and conclusive manner. One way of putting this is that policy advice needs to be opinion – ‘(in my opinion) we should do X’ – whereas academic work is about (Bayesian) belief – how much credence we put in the potential success of different possible policies. The distinction between belief and opinion is Dennett’s (1978) that I mention in Chapter 5 (Section 5.3) and return to in Chapter 10. There are other differences too. I have found that, unless you are talking to finance or treasury officials, explanatory statistics are of less interest to policy makers than simple tables of descriptive statistics; whilst simple descriptive history works better than strong interpretative or theorized qualitative evidence. In part this is because many of the recipients of policyoriented research are not trained to understand either explanatory statistics or strongly theorized interpretative work; and in part this is because they want simple answers that they can explain simply to their political masters and the public. There is nothing wrong with academics taking money from government or organizations for policy-oriented research, just as there is nothing wrong with academics having opinions, though they ought at least to expose their beliefs to the rigours of science. Academics should not exclusively do policy-oriented research or they will forget what academic work is really all about. Opportunities are growing for academics to take a greater lead in policy-oriented research. Rather than responding to government by taking contracts offered, academics can produce policy-oriented research and then lobby government using both direct – through public servants and representatives – and indirect – through the media and lobbying activities – means, to influence government policy. With the advent of ‘big data’ through social networks and the greater collaboration of research teams across disciplinary divides, there is scope for much greater understanding than could once be achieved (King 2009). Such research might receive government or external funding, but needs to be directed at solving the problems that academics and the public see as important and not simply in the interests of political parties or governments.

7.5 Different methods: institutional-structural In the following sections I briefly describe some of the methods currently utilized in political science. I divide them into three types: institutionalstructural; behavioural; and interpretative. In many methods textbooks, the distinctions are qualitative, quantitative, experimental and comparative. I find this a strange breakdown. In some sense, any explanatory claim has to be comparative: it is a claim about how things are in comparison to how they

Methods and Methodologies

171

could be under different conditions. Of course, in politics comparative often means comparing across different countries in order to gain variance. But we can gain such variance within a country if we break it down by region, constituency or whatever, depending on the research question; and of course one can do comparative politics quantitatively or qualitatively. So my distinction is not a standard one, but I think it helps us to see certain methodological assumptions underlying the methods adopted. These are distinctions of convenience and should not be given too much emphasis in terms of breaking up the world into specific categories. The methods discussed under each can slide into the other categories, and much actual research involves two or three general types. Institutional-structural methods usually make important assumptions about behaviour; indeed, part of the point of such methods is to explain political behaviour or to explain how institutions affect such behaviour. Robert Dahl’s (1961a) original definition of behaviouralism (contrasting it with ‘old institutionalism’ or constitutional law) was the study of human behaviour in different institutional settings in order to discover continuities in behaviour. Institutional-structural methods tend to utilize the assumptions of revealed preference evidence. That is, they assume that the behaviour of agents reveals their preferences as they face different sets of incentives within different institutional or structural contexts. Such methods often start with some very basic assumptions about agent motivations that are then used together with empirical evidence about behaviour in different institutional settings, both to modify our understanding of motivations and to see how those different settings affect agent behaviour. Such work is usually highly theoretical: for example, veto-player theory examining the comparative stability of policy across different institutional systems; the vast array of work examining how electoral systems affect party systems, and how both affect voting, party coalition and government formation; principal–agent models of bureaucracies; social network analysis; policy network theory; critical theory; structuralfunctionalism; neo-Marxist and Marxist approaches; large parts of international political economy; realism in international relations.

Mathematical and formal modelling Rational choice models assume certain consistency conditions for agents to enable prediction of behaviour in novel situations based upon past evidence of behaviour (Austen-Smith and Banks 1999). These conditions of revealed preference are thus interpretative conditions. However, both rational choice and other mathematical models get their purchase (generally speaking) from modelling agents within roles – as voters, parties, public servants, consumers – and thus structurally situate agents within institutional parameters. The results of such modelling suggest outcomes within different institutional

172 The Philosophy and Methods of Political Science forums based on assumptions of such role–agent preferences. Used for explanatory purposes, the models produce hypotheses that can be tested empirically through quantitative or qualitative empirical methods. They are often also used normatively to look at the inefficiencies of different ways of organizing political, social and economic processes. Whilst formal modelling is deductive and tests hypotheses when quantitative data are used in type and ultimate explanation, more inductive usage of mathematical models is also common. When applied to token proximate cases in analytic narratives (Bates et al. 1998) or to theoretically analyse data as, for example, in the examination of party strategies (Schofield 2006), induction from the evidence to the models is common. That is not to say further hypotheses are not drawn, especially from the latter, for further empirical testing, but induction from evidence is usually the beginning point of any formal model. Within the formal tradition some very general theses about institutions’ effects on political behaviour have formed. From Arrow’s (1951/1963) original possibility theorem demonstrating that ‘institutions matter’ to work by Ostrom (1990, 2005), North (1990, 2005) and many others, they show how different institutions create different outcomes and also the inefficiencies that can become sustained over time. To some extent, such analyses are puzzle-solving as much as hypothesis-generating. They model and thereby explain why inefficiencies are generated despite individually rational, that is, individually explicable, behaviour. Whilst the models are deductive, often the applications are inductive and interpretative. Their strength lies in the institutional lessons that can be learned from them and the path-dependence of much of politics. Social choice theory emerging from Arrow’s original work is largely normative, looking at conditions on constitutional forms. Arrow’s original work can be interpreted as asking a normative question about the meaning of the public or general will (Riker 1982b; Mackie 2003; Dowding 2006a). So too is much work on preference-aggregation, examining how fair or manipulable voting systems are (Gibbard 1977; Satterthwaite 1975; Riker 1982b; Dowding and van Hees 2008). From this normative work, positive work emerges that tries to explain how events play out (Riker 1986; McLean 2001; Schofield 2006). The median voter theorem explains how, given conditions, outcomes emerge, and then from n-dimensional space how diverse outcomes can be, leading us to look for institutions or structure that tend to lead to certain types of outcomes (McKelvey 1976; Schofield 1978; Austen-Smith and Banks 1999). Game theory examines these processes in a more dynamic frame, and again often poses questions about what we expect to occur and investigates when it does not. Mathematical modelling is less interesting when it simply models what we would expect to find anyway or produces predictions that fit reality

Methods and Methodologies

173

but without discounting other potential mechanisms that can explain that outcome. A formal model essentially describes a mechanism, but just because its predictions are largely met does not prove the mechanism exists. That is when case-study analysis can help us observe the machinery of the model. Extant empirical generalizations might be underlain by the model’s machinery, but they act more as a constraint on actions (background conditions) than the proximate causal processes. I regard the predictions of coalition theory as much like this (Dowding 1995b). To some extent all formal models can be seen as normative in the sense that they make predictions based upon consistent behaviour (‘rationality’). Mathematical modelling provides a normative standard by which to judge actual behaviour. It can help us to explain departures from our expectations by providing the means by which we can observe those departures and ask why they occurred. This can be contrasted with agent-based modelling.

Agent-based modelling Agent-based models (ABM) simulate the actions of agents in structural settings (Johnson 1999). They can operate with different sets of assumptions, using classical game theory, evolutionary game theory, or other behavioural assumptions. Where rational agents in game theory optimize and the models produce equilibriums which are then checked for robustness, in ABM agents have limitations on information, search costs and often heuristic devices, and there is no particular search for equilibriums. ABMs are more interested in the dynamics revealed through the processes. In politics they have been applied to such things as group formation (Johnson 1996) and the behaviour of parties in parliament (Laver and Sergenti 2012). Often such computational models are essentially mathematical models of such complexity that the maths cannot be done by hand. Their best successes show that small changes in assumptions about behaviour can lead to dramatic changes in systems; that various competing equilibriums can be stable within a system; that some equilibriums are stable across systems; that processes can lead to cycling; or they examine the probabilities that can be assigned to diverse stable outcomes. The first of these, showing that assumptions can matter a lot, should give pause to mathematical or formal modellers. One must always ask when reading formal models how robust the conclusions are with regard to the assumptions. The more robust the conclusions, the less important specific assumptions; if a result rests upon a key assumption, how far that assumption is descriptively accurate (either motivationally for an agent or structurally for the system) will determine how useful the model is. The second of those results is important for qualitative and historical researchers. If different outcomes are possible across the same institutional

174 The Philosophy and Methods of Political Science or structural form, why one emerges in one society and another elsewhere is dependent upon particular historical processes. These might be entirely contingent or they might depend upon specific decisions or aspects of an historical process, perhaps the timing of events. However, whilst we might surmise that particular decisions or timings were important in a given situation, we can only test how important such specific determinations are by larger-n analyses. Whilst obviously cruder and more pared down than thickly described histories, agent-based models can provide multiple cases to show just how important determinations such as timing are. They provide a structural justification that supports the surmises based upon detailed historical analysis. The other results interest institutionalists, showing, once again, that institutions matter both for agent behaviour and for outcomes. Sometimes outcomes are almost entirely determined (given reasonable assumptions about agent motivation) by institutional or structural form. Sometimes, however, institutions seem to make little difference and the contingencies of specific decisions create the path dependencies that result. Agent-based models alongside other formal methods, large-n empirical analysis and detailed analysis of actual cases can help us to understand when institutions matter and when other factors play a larger role.

7.6 Different methods: behavioural Behavioural research is designed to make inferences about politics from the way people act. Of course, institutional research makes assumptions about how people behave, modifying those assumptions according to empirical evidence, but institutionalist and structural research depends more heavily upon theories of motivation. Behavioural research places less emphasis on theory and more on the empirics. It interprets what it finds after the empirical evidence has been collected. Thus behavioural studies tend to be more inductive. I include within the ambit of behavioural research survey research and experimental methods, as well as archival work. All of these also involve interpretation, and can encompass both revealed preference and stated preference evidence.

Survey research Since we need to interpret the results of surveys in terms of reasons or beliefs one might think surveys should enter into the next section, interpretative. However, survey research is traditionally considered behavioural since surveys are used to explain (‘interpret’) behaviour. Hence it is usually juxtaposed to institutional research and (narrow) interpretivists are generally critical of survey work. Titles do not matter that much. A survey is the collection of

Methods and Methodologies

175

information through a questionnaire imparted to a sample of individuals from a population relevant to the research question. It is a popular research method because it is relatively cheap and flexible, and provides a wide range of information. Its long heritage ensures standardized ways of conducting surveys and understandings of how to analyse the results (De Vaus 2002; Kalton 1983; Brady 2000; Lohr 2009). It is best used to sample generalizability: that is, it allows probability sampling from large populations. These can be thought to represent the views, attitudes or characteristics of populations. Essentially a survey is a description of a population, based upon generalizing from a sample of that population. Surveys are prone to observation errors through bad measurement techniques and non-observation errors through omitting important cases. Observation errors occur through poor questionnaire design, inaccurate answers or observer bias. Hence survey design is important. How surveys are administered might create biases, and the relative merits of face-toface, postal, telephone and web-based surveys are contested (Cooper 2000; Dillman 2007; Orr 2005; Sanders et al. 2007; Chang and Krosnick 2009). The key question is how far samples are randomized, and debate exists over appropriate weighting and re-weighting that might give greater validity to results. The issues involved are statistical – the degree of confidence we can have in the results and the validity of the marginal differences in the dependent variable that are thought to have been measured via differences in the independent variables. Where surveys measure attitudes, they measure stated preferences, which might not correspond to revealed preferences. Sometimes the two might be close – the way people say they will or have recently voted might correspond to how they will or have recently voted. However, how people say they might respond in novel situations, or what they say they take into account in social or economic choices, might be far from what we would interpret when we observe their actual behaviour in those situations. For that reason, whilst survey research is an important tool of social science, we need to be very careful about what we infer from it. In the main, survey research is a device to assist us to understand behaviour. Electoral surveys help us interpret what was happening in the actual ballot. Mobility surveys help us interpret what considerations people take into account when moving house or emigrating. Attitudinal surveys help us interpret what people think is important and thus predict how they might behave given different public policies. In the latter case the attitudes displayed in surveys might not reflect actual behaviour. Whilst surveys provide objective characteristics in the sense that respondents can be expected to truthfully report characteristics such as age, education, gender and, under suitably designed questions, socio-economic characteristics, stated preferences are an amorphous characteristic.

176 The Philosophy and Methods of Political Science One of the most standard criticisms of comparative survey work is that apparent synonyms do not always translate well across languages and cultures. Extensive use of surveys, such as barometer surveys now conducted across the world, reveal rather different attitudes to some social and political circumstances. How far responses to the questions demonstrate deeprooted and underlying attitudinal differences and how much they reflect different understandings of concepts, or the different institutional contexts in which the questions are posed, is a constant issue for analysts. This shows the deep interpretative element of survey research and how researchers need thorough institutional and cultural knowledge of the subjects’ backgrounds. Survey methods have been the backbone of political science since the behavioural revolution. However, findings from behavioural research are increasingly under scrutiny. This is not to disparage the careful claims that have been made in various fields, notably electoral research, over many years. We know far more about the nature of elections, the causes of electoral success and the pattern of support for parties than we once did. Surveys need careful design and careful interpretation. To conduct them thoroughly is often expensive in time and resources. These days small surveys can be conducted online using systems such as Amazon Mechanical Turk (Palolacci et al. 2010). Whilst these can prove good and useful tools for empirical topics, especially for undergraduate or doctoral theses, they have also been used extensively in quasi-experimental settings.

Experimental methods Experimental methods have taken off in political science in the past 20 years, both in the lab and in the field (Durckman and Green 2011; Gerber and Green 2012; Green and John 2010). They are popular for all sorts of reasons, not least because, as discussed in Chapter 6, they are thought to give a better handle on causation (Morton and Williams 2010). An experiment is a study that involves a randomized treatment: that is, there is a control and a treatment group that should be comparable, thus having no confounding elements. This gives experiments internal validity. We experiment by manipulating an aspect of the world, witness the effects and assume the manipulation was the cause of the change. The more precisely one can control the elements, the greater assurance one has that the intervention was directly causal of changes in the outcome. So in political science the researcher creates (or in natural experiments finds) random variation, intervenes in a sample of a subject pool, and sees what difference results in the treatment and the control group. The difference between the treatment group that had the intervention and the control group that did not will be due to the intervention plus some random variation. Laboratory experiments were once the preserve of psychologists or social psychologists, but first economists and now political scientists have

Methods and Methodologies

177

moved in. The behavioural revolution in economics was originally outside the mainstream and was designed to challenge the models of humans as rational economic maximizers (Guala 2005). Researchers adopted the methods of experimental psychologists, some of whom were conducting similar experiments which economists felt were flawed (Binmore 2007: ch. 19). Experimental economics developed a different set of protocols for running experiments. Over time the different sides have tended to move together and talk past each other less often (Camerer 2003). The major movement now, involving both neuroscientists and economists, is towards conducting experiments in tandem with functional Magnetic Resonance Imaging (fMRI) scans to explore how motivational forces can be explained alongside neural processes (Ross 1998, 2008; Glimcher 2004, 2011; Camerer et al. 2005). Political scientists became involved slightly later and, broadly speaking, tend to be more interested in how institutions affect behaviour than in the motivational bases of behaviour. To some extent, comparative politics is about the behavioural effects (on voters, politicians, parties, citizens, firms) of different political institutions. These can be examined in lab experiments and in natural experiments. Experiments are used extensively by public policy analysts and are of great interest to governments invested in the ‘nudge movement’ (Sunstein and Thaler 2008). This is the idea that citizens need not be directed or forced by law to change their behaviours, but can be nudged towards better ones (healthier lifestyles, participating in community life) through institutional design and informational acquisition. Experiments can suffer from experiment effects. The fact that subjects are being experimented upon can affect their behaviour, so both control and treatment groups might not behave as they would outside of the experiment. (Experiment effects do not only occur with human subjects. In the sciences we cannot always be sure that what we use to measure is not affecting the thing being measured. For example, the analysis of the uptake in nitrates in plants under different conditions is measured by the use of isotopes to track the passage of those nitrates through the plant, but we cannot be sure that under some conditions the isotopes are not themselves affecting the passage of nitrates.) One problem unique to human experiments is ethical guidance procedures: subjects need to consent to take part. The sample therefore is not truly random, since randomization usually takes place after self-selection. This is a problem of the external validity of the study. The effects noted only strictly apply to those people who take part in the study; when applied more widely, the fact that these people are self-selected might be a problem. Psychological experiments are bedevilled by the fact that their subjects are students. It has been shown, for example, that students in the laboratory do not display the same attitudes to risk as people who are in work, putting in doubt many previous experiments examining ‘rationality’. Today protocols in economics and political science dictate that students should not normally be subjects.

178 The Philosophy and Methods of Political Science Experimental studies can suffer other problems of external validity, since the conditions in labs differ from those outside and so results might not be generalizable to the social and political situations that led to the questions the experiments are designed to answer. Moving experiments into the field is one response, though it does not always overcome the self-selection problem. Field experiments in political science date back to the 1920s (Gosnell 1926, 1927; Eldersveld 1956), but, with larger samples and more sophisticated statistical analyses, have taken off in the twenty-first century (Gerber and Green 1999, 2000; see their 2008 for a review). There are particular problems faced in political science beyond those in other fields. One big problem is maintaining control over the participants. In medical trials, for example, it is relatively easy to ensure that the patients do nothing else beyond the intervention, and experiments can be clustered over large areas. In politics it is much harder to control what the participants are receiving by way of, for example, information on an election, and hence the effect size of the intervention is likely to be small. Clustering is problematic, since there are likely to be non-random cluster effects, which also affect the external validity of any finding. One way of dealing with external validity is to have different experiments in different areas, followed by meta-analysis, but the incentives to replicate studies in political science are low, if only because major journals are looking for original, not replication, studies. (Reviewers often ask, and graduate students are often taught to look, for the extra contribution made by a study. That is a professional problem for the science in political science.) Sometimes a feature of the social world enables studies that resemble an experiment (termed quasi-experimental), when a government or some natural event creates near-natural control and treatment groups. The results of such studies are examined through a statistical technique known as regressiondiscontinuity design (Shadish et al. 2002: ch. 7; Imbens and Lemieux 2008; Lee and Lemieux 2010). Lee at al. (2004), for example, try to examine whether electoral competition moves politicians towards the centre of the ideological spectrum or whether the competition for votes is simply a choice between ideologies. They look at elections in constituencies where there is a change in electoral strength (the discontinuity) to determine whether subsequently this has an effect on how the winning candidates vote in the legislature. They argue that politicians cannot credibly commit to moderate their policies, and find no effect of competition on the way politicians vote in the legislature.

Big data The other massive innovation in the past decade or so is the advent of big data enabled through the speed patterns of computer use. What is big data? Surprisingly, not everyone has the same understanding. It can be big in

Methods and Methodologies

179

several senses: number of observations, number of covariates, file size. Big data can mean computational innovation, such as data mining, or collection techniques such as tracking, web scraping, or other forms of extracting information from online sources – documents, Twitter feeds and anything in between. More generally, however, it refers to inferential techniques such as statistical or machine learning and so implies different statistical techniques from those traditionally used for analysing surveys. For some this means atheoretical work. In some uses, such as forecasting models, past predictions are used to forecast future ones. There is no role for theory. (Of course, this is subject to the problems of induction that I have argued (especially in Section 5.3), theory helps to address.) Using new statistical techniques is important, since simply increasing the size of the sample does not solve inferential problems. We can see that statistical algorithms can find multiple patterns in vast quantities of data. These inductive processes allow patterns to emerge; theory can then suggest whether we have found causal or structural features. This might make big data techniques seem to be rivals to experimental methods. However, big data are subject to the same methodological strictures as any simple regression technique. One should never throw the algorithms at the complete set of data. The data need to be randomly split into the inductive sample and the holdout sample. Any patterns seen and theorized about in the inductive sample can then be tested on the holdout sample. What a large number of observations allow, however, is to trade bias with variance. Hersch (2013) examined how people close to the victims of the 9/11 attacks behaved politically in comparison with similarly matched people. Those close became more politically involved and more conservative. Ansolabehere and Hersch (2012) use big data to examine differences between those who answer surveys and those who do not; whilst Ansolabehere et al. (2012) used large number of observations to test theories of voter registration (see Monroe 2015 for discussion of these and other examples). Large-n here facilitates hypothesis testing, since it enables specific suspected variances to be tested. Big data can exacerbate familiar inferential problems with multiple hypothesis testing and omitted variable bias. Bias can grow with sample size. Where there is a fixed bias, it will dominate the mean square of the estimate as the sample size grows. Big data can certainly help descriptions, explanation and hypothesis generation, but cannot substitute for theoretical analysis. The choice is not between deduction and induction, but rather between conscious theorizing and non-conscious theorizing. We can think of the techniques of big data as the sort of work we once did prior to analysis: they collect observations. Then we theorize and then we test. Using the holdout sample enables testing as in experimental methods. We have our hypothesis – this pattern exists and we think it shows X – and we test it against data – the

180 The Philosophy and Methods of Political Science holdout sample. Experimental methods can look equally atheoretical at the first stage. Experiments involve manipulating a variable to see what changes. We could manipulate each variable in turn, but usually, for reasons of efficiency (theory and/or practicability), we have reasons for choosing particular ones to manipulate. So we can see a close relationship between using big data and experimental methods. Are they rival to formal methods? Patty and Penn (2015) point out that when we find patterns in the data, we are describing it in certain ways. What machine-learning techniques do is find potential patterns from the mass of data and then concentrate attention upon those patterns. They start to look for them, much as a researcher might start to look for particular information in an archive, ignoring other information. To be sure, the algorithms do this more systematically than a human researcher in a physical archive, but they still concentrate attention upon the patterns that emerge. In doing so, information about other potential patterns is lost. Machine-learning techniques make choices about what to be responsive to and what to ignore. Patty and Penn point out that one of the major results of formal theory, Arrow’s theorem, teaches us that there are different patterns in the data depending on how we decide to describe them. Formal theory, in their account, can help us to make decisions over what choices we can make or can direct algorithms to make when manipulating big data. And for much of the data that we analyse, whether they are social networks, genomes, written texts, and so on, we have a conceptual structure that underlies our understanding (Patty and Penn 2015).

Evolutionary and genetic analysis In order to produce Darwinian evolution, one needs both a selection mechanism and a replicator. The replicator repeats itself and its interactions with the environment around creates the surface qualities – phenotypes in biology – that are often the subject of our analyses. The idea that social, economic and cultural forms ought to be explicable in evolutionary terms seems plausible and there are many attempts to provide explanations using evolutionary game theoretic or less formal models (Nelson and Winter 1982; Hodgson 1999; Johnson 1995; John 1998: ch. 8; Dunbar 1999; Alford and Hibbing 2004; Ben-Nur and Putterman 2000; Binmore 2005; Ross 2006). Dawkins’s (1976, 1999; Blackmore 1999) ‘meme’ – ideas that are repeated – once seemed promising, but the problem with memes is that they are too close to the replicated behaviours and institutions or conventions that we are trying to explain. Many ‘evolutionary’ accounts in the social sciences struggle to define both the selection mechanism and, more particularly, the replicator, and thus fall under a related but different mechanism, that of path dependence (Thelen 2004; Steinmo 2010). Furthermore, only very general and

Methods and Methodologies

181

wide-ranging predictions are possible in evolutionary models, since minor environmental changes can have dramatic effects. Thus effective applications to the general topics of political science are few and far between. Recently, genetic analysis of political behaviour has taken off (see Hatemi and McDermott 2011 for an introduction). Always controversial, such analyses tend to utilize big data or twin-matching studies. A general confusion that tends to surround gene-based research is that we know that both genes and the environment are always implicated in all human behaviour. Thus any claimed gene effect is always relative to some environmental factor and the total genome. Most specific gene effects for political behaviour are also rather marginal (Fowler et al. 2008). That does not mean that demonstrating them is not important, though recognizing this fact ought to put debate into proper perspective. Gene studies tend to be controversial because, for some reason, genetic influence on behaviour is considered somehow more provocative than environmental influence. In all statistical models, the measures of influence of genes or environment are determinative.

7.7 Different methods: interpretative In this sub-section I set out a very general understanding of interpretative research. A narrower understanding entails that in order to comprehend political or social interaction, one must understand it from the inside. That means understanding interactions as those engaged in them understand them; that understanding requires deep cultural, historical and generic understanding of context (Berger and Luckman 1971; Geertz 1973). Interpretative approaches in the narrow sense rely heavily upon anthropological and ethnographic techniques. In older forms interpretative understandings were juxtaposed to causal understandings. That is, interpretivists were not claiming to understand the causes of political processes, just how they should be interpreted in order to comprehend what those engaged in them are doing. In this section I include approaches much broader than this narrow interpretation – ones, that is, that delve to some extent into the reasons that agents have for acting as they do. I include approaches that are rather far from anthropological and ethnographic methods, though those too are included.

Archival research Traditionally, archives are conceived of as repositories of artefacts, usually documents, manuscripts and images, often the official records of organizations, but also personal diaries, letters, and so on (Burnham et al. 2009: ch. 7). These digital days, archives encompass those and anything existing

182 The Philosophy and Methods of Political Science in digital form. The diversity of traditional archives is narrated by Bradley (1999), whose work uses nineteenth-century parliamentary papers, various documents in Leicester’s local record office, a transcript of interviews she carried out herself with working people in northern England, and diaries, letters and clippings left by her mother. In digital form we might find any record that has been posted on an accessible website, whether designed as a repository of knowledge or as a personal record (for example, Facebook, Twitter, emails). Archives hold both records that the writers knew would be made accessible one day (or were public prior to or at the point of archiving) and others perhaps not intended for public consumption. We might discriminate between these differently intentioned documents, though we must take care interpreting all of them. In practice, political scientists are likely to visit online or physical archives to find official (usually government) documents to provide evidence on what organizations were doing or thinking they were doing. Such records can also reveal conflicts in organizations, for example the depth of disagreement within a cabinet that was carefully or largely hidden at the time decisions were publicly announced. Archives provide the opportunity for researchers to go where organizations keep their secrets – paradoxically, as organizations have learned the perils of Freedom of Information acts, they might prove less useful in the future. Traditional archival research can be done systematically but tends not to be. A systematic search of an archive would review either comprehensively or randomly all documents that might be relevant to a particular research question. Or one could weight a random search by likely relevance. One would systematically record everything pertinent to the question and analyse that. Usually, however, due to time and resource constraints, archival research is more like detective work, where hypotheses are formulated on the basis of evidence. Once a particular view is formulated, further evidence is sifted to try to prove the hypothesis. This has obvious problems of bias and the selective use of evidence. Its saving grace is that, professionally done, the evidence for a hypothesis is marshalled, recorded and referenced; later researchers can return to the archives and look to reselect and challenge bias (Moravcsik 2014). Of course, archives themselves are already potentially biased. This can be explicitly so, where the record is precisely what the writers want the historical record to document. Even private notes and letters tend to record what the writers want to show, to themselves or the recipients. Nevertheless, as those of us who delve in archives often find, the records can reveal more than their writers wanted us to see. People may demand that some things go ‘on the record’, but they informally record further information about others in an organization or send memos about third parties. Triangulating across sets of such messages can be revealing. Whilst official documents are often

Methods and Methodologies

183

written in the knowledge that they might be later surfaced, formerly totalitarian countries, for instance, are full of archives that no one expected to be made public (Shentalinsky 1995).

Interviewing Interviewing relevant actors is one of the major ways of analysing current events qualitatively. Once one knows the research question, choosing whom to interview is the next step (Wengraf 2001; Burnham et al. 2009: ch. 9; Brinkmann 2013). If one is interviewing affected individuals in a large group, one might use conventional random sampling techniques. However, most interviewing is of elites intimately involved in policy formation or decision making and interviewees are chosen in order to elicit particular viewpoints. Here statistical bias is not the issue. One is choosing to gain certain perspectives and also, often, to garner information that could not easily be surfaced any other way. How many interviews one conducts depends upon cost and access considerations. One might adopt triangulation or snowballing strategies to choose interviewees, asking each interviewee who else to talk to or who was important in some decision. Interviewing techniques should be carefully considered. Thoughtfully calculated degrees of leading, probing, challenging can prove productive (Rubin and Rubin 1995). One should be prepared to go off script if interesting information is offered, even if one not sure of its pertinence at the time. Interviews at early stages of research can be vital in developing an explanatory framework, rather than for ‘testing’ hypotheses about what is going on. Cultural issues are important, and interviewers must be sensitive to the culture of those whom they are interviewing. They must also set out the confidentiality criteria carefully prior to the interview. Ethics statements often require one to explain to the interviewee the purpose and use of the interview material. This should be done in the most general terms, lest interviewees, seeking to be helpful, slant their answers in favour of specific hypotheses or arguments. Wengraf (2001) is one of the most comprehensive introductions for the utility and limitations of interview techniques. The aims of replication hold here as elsewhere, so interview evidence in the form of recordings or transcripts should be made available for subsequent researchers. (This can be problematic, given confidentiality commitments.) Systematically analysing one’s interviews through the creation of simple coding schemes or dedicated software gives greater credibility. However, interview evidence is qualitative and not really open to a great deal of quantification. Often the best use of interview material is illustrative. Evidence is built up through many sources, but a sharp quote from an interviewee might be what remains in the mind of the reader. Interview evidence can also, for that same reason, be quite misleading.

184 The Philosophy and Methods of Political Science

Focus groups Focus group research, originally developed in the 1920s, returned to academia via market research and became very fashionable from the 1990s (Wilkinson 2004). It essentially consists of conversation between (usually) a small group of people helped by a facilitator, who might lead and structure the discussion or perhaps simply ensure that everyone is treated with respect. It is a method by which researchers can get to hear what people think about issues in a relaxed setting, going deeper than surveys and broader than interview evidence would allow (Morgan 1996; Barbour 2008; Krueger and Casey 2014). Market researchers felt surveys did not always elicit thoughtful comments. Wanting to probe attitudes more critically, they found group settings more congenial and conducive to the development of views. Focus groups have been used effectively to chart how attitudes can be changed, at least in the short term. They cannot be thought of as representative, however. Each focus group takes on a life of its own, and they can be led by the facilitator or by dominant members. They are better as tools for examining interaction than representativeness. That said, focus groups can be important for eliciting the views of marginalized groups of people – though not necessarily more effective than sensitive interviewing or ethnographic work. They have also been effectively used to see how the setting can affect group dynamics and the utility of deliberative settings (Farrar et al. 2010); but in that sense the focus group is itself the object of research, rather than a tool to examine other questions.

Ethnography and interpretivism I categorize ethnography and interpretivism as interpretive methods, but they obviously also employ detailed behavioural analysis. Ethnographic or participant-observer methods involve shadowing or living with people to try to fully understand their situation. Ethnography is essentially descriptive, aiming to describe the life or work of other societies, groups or organizations in a detailed and accurate manner, following long exposure and first-hand knowledge. Its insights can be used to help produce or examine more general theories of human nature or organizational culture. However, some researchers believe that the sole purpose of ethnographic work is to describe and interpret the specific practices of those who are studied. This nomothetic versus idiographic split can be seen in anthropology (Radcliffe-Brown 1952: Introduction; see also Ingold 2008, who makes the distinction in terms of ethnography and anthropology, with the latter being nomothetic). Idiographic ethnology is divided between those who see the role of the participant – observer as understanding and explaining the culture and practices of those they study, and those who hold that such ethnographic

Methods and Methodologies

185

understandings can only be parsed in the terms and meanings of those who are being studied. Thus the patterns that matter in a culture are the patterns seen by the participants. This has radical implications. For example, if a society has no concept of rape, what we as observers see as rape must be understood differently (Fricker 2007). Less dramatically, we might say that interpretivism suggests that to comprehend the motivations and understandings of a group we need to understand what they think they are doing; but we can challenge that understanding. It is perfectly possible for a practice to be conducted for one reason, but to have other effects unknown to the participants. One of the main purposes of social scientific study is to examine those other effects. In political science participant observation is usually less esoteric than in anthropology and is designed to understand the working and practices of institutions within our own cultures. Fenno (1978, 1990) shadowed members of Congress both within that institution and outside. Rhodes (2011) shadowed ministers and senior civil servants at the heart of British government to find how they go about their working lives. Each describes working practices and tries to understand the inner motivations. For Rhodes, the exercise is interpretivist, and he sees decisions in Whitehall as part of a culture inculcated over a long time (Bevir and Rhodes 2003), while Fenno takes a more pragmatic approach. Rhodes supplements his shadowing exercise with interviews, and to some extent it might be queried what advantage ethnographic methods have, at least when studying organizations within a literate culture, over simply examining the biographical and autobiographical records of such elites. There are several responses to such a criticism. The first is the one we encountered with weighing interview evidence. People might not be completely honest; even when they are trying to be so, they describe events from a particular view or with a particular slant. Second, with participant-observer studies the political scientist can enter into the organization with specific questions in mind and can search out the answers. He is not just shadowing and following, but also interpreting and questioning. However, this latter advantage negates interpretivist understandings of ethnographic methods. Interpretivists hold that we need to understand what subjects are doing exclusively from their own point of view. If the researcher goes in with his questions and views, then he is not fully immersing himself in the subject. Either you are a civil servant or you are an observer of civil servants. Sometimes ethnographic methods are required, since the subjects are unlikely to be writing memoirs, honest or not. Only by immersing oneself in their lives can one appreciate their reality, as, for example, we see from ScheperHughes’s (1992) study of the violent lives of people in Brazilian shanty towns. Ethnographic techniques can suffer from experiment effects. The very fact of having a shadow can change the behaviour of the subjects. This is a bigger

186 The Philosophy and Methods of Political Science problem when the shadow is there only briefly (for example, Rhodes 2011) than when immersion extends over a significant time (for example, ScheperHughes 1992). There are also external validity effects: what we learn about US Congressmen or British ministers might not apply outside those forums or might be very time-specific. However, whilst these are logical problems, some lessons can be readily generalized; and whilst behaviours change over time, they do not usually do so radically unless there is some major institutional change or social upheaval. Interpretivists or those doing only idiographic study do not acknowledge the external validity problem, since for them the whole point of the study is to understand from the subjects’ point of view and only describe that particular group or organization at that point in time. At the extreme, however, that would mean every story is unique and could only be told once by one person. All such idiographic study is just a collection of lives from which nothing can be concluded or built up. The real issue is how far one wants to generalize and to what end.

Discourse analysis The essential elements of discourse analysis relate to our ideas of precisely what social processes are, the normative baggage that accompanies social, political and economic policy, and how that changes over time. In part, it is about how social power relations are reflected and reinforced through the nature of language. Discourses often reproduce the underlying assumptions of a culture or a group and thus can frame the way people look at issues. Discourse analysts look at changing language over time to see how dominant ideologies grow, persist and shift. This connects up with the Cambridge school of conceptual analysis (see Chapter 9), whose time frames of analysis spread over centuries. Some analysts look at how different groups frame the same issue at the same moment. Others explore how governments deliberately try to frame issues, as discourse analysts help teach powerful groups how to use language for their own ends. Those who see discourse as a frame have stimulated much experimental work, and analysis of bias in media (Gerber et al. 2009). They have shown that framing can lead to people changing their attitudes, but also that people tend to pick out frames that already fit their worldview. Socialization begins at an early age. Some discourse analysis uses relatively casual techniques: analysts read through their material in search of particular terms and phrases to track the changing nature of language. However, the field is being transformed with the use of content- and text-analytic techniques. With the increase in computing power and the ready availability of many text-analytic programmes, discourse analysis can be conducted systematically. Content analysis is a technique whereby terms are coded into categories. These categories could be general terms such as policy fields, in the way the

Methods and Methodologies

187

policy agenda project codes legislation, committee work, media coverage (Baumgartner and Jones 1993; Jones and Baumgartner 2005). Terms might be coded into ideological categories such as left–right or hawk–dove. The manifesto project has coded party manifestos and major policy speeches at election time into such categories to see the changing ideological landscape over the world (Budge et al. 2001). Such coding was once done manually. Now we can code a set of documents and software programmes can learn the coding system and apply it at least as accurately as human coders (Hillard et al. 2007; Quinn et al. 2010). Such content analysis is usually theory-driven. Terms are chosen to be picked out by the programmes because they label a given area (topic codes) or for their ideological or framed content. Another way of analysing discourse is to examine text ‘theory free’. Here the software programmes are set free to find patterns in the data. The analyst then theorizes about what those patterns represent. This process is usually called ‘text analysis’. Discourse analysis has traditionally been heavily theory-driven. However, new computer techniques are leading to more inductive methods. To the extent that claims and hypotheses can be drawn, computer content and text analysis can demonstrate those claims or test hypotheses. Text and content analysis are distinguished by the techniques adopted. Text analysis usually refers to techniques where the software algorithms are allowed to surface the relationships between words or meta-sentences (chunks of text) as they appear. Once the software has found these relationships, the researcher induces what is being shown, perhaps derives claims or hypotheses, and then tests these against further text. The best technique is to take a set of texts and randomly assign them to the initial and holdout samples. The software is then used on the initial text to find a set of relationships. The strength of the relationships is measured and the theoretical reasons why they emerge posited. Then the theory is tested on the holdout sample. This technique mirrors experimental methods. The first stage is an inductive version of producing a deductive theory, or an induction from more causal theorizing. Examining it to see if the relationships found are sustained in the holdout sample constitutes the test. This experimental method is required because such techniques are almost bound to find some relationship in the text. We need to be sure that it is not simply a chance pattern over which we can plausibly theorize – we want to avoid committing the narrative fallacy. The test gives us some confidence in the pattern we induced, since it also holds in a second sample. There are many uses of text analysis. For example LIWC (Linguistic Inquiry and Word Count) calculates the degree to which a set of people use different types of words across the sample of texts in psychologically meaningful categories (Tausczik and Pennebaker 2010). This can give clues to the thought processes of individuals or to psychological profiles. In political

188 The Philosophy and Methods of Political Science science it has been used for predictive purposes. Dalvean (2012a, 2012b) uses LIWC on a sample of maiden speeches in the Australian parliament to produce highly predictive models of who will become ministers, based on experience, education and abstractness of the maiden speech. Other topic modelling allows issues to emerge. Catalinac (forthcoming) examines speeches in the Japanese parliament to show how the nature of the policy process shifted following change to the electoral system in 1996. Prior to 1996, government was bureau-led, with politicians just interested in constituency pork, but became more policy-led as political competition between parties grew. There are issues with causal inference from such approaches that need to be handled carefully. Those who use software programs to systematically analyse text do not necessarily have any other programmatic aims in mind. However, the long tradition of non-systematic discourse analysis is critical in having established the aim of contributing to human emancipation by exposing the way language can be used to manipulate people (Fairclough and Wodak 1997; Wodak and Meyer 2001). Of course, any academic analysis can be used in many different ways. The lessons about manipulation through language have almost certainly been learned most fruitfully by the powerful groups and organizations in society.

7.8 Conclusion In this chapter I have considered different ways of going about political science research. Some claim that these methods rely upon very different ways of looking at the world. Whilst there is some truth in that, often the methods look at different sorts of patterns, ask different sorts of questions, and are not contradictory. The most important elements for any political science approach are that the evidence is marshalled in support of the conclusions drawn, which are then replicable by others. If others do not draw the same conclusions from the same evidence, then that evidence is not conclusive but rather speculative. Some areas of research have produced cumulative results, with later work building on earlier work. Others tend to revolve in circles. Where work is non-cumulative, we can be assured that the conclusions drawn from the evidence are non-conclusive. That is the most effective way of judging the best methods.

Chapter 8

Concepts and Conceptual Analysis

8.1 Introduction The degree of controversy over any concept is dependent upon a complex interplay of three aspects of conceptual analysis. First, how close the defined concept comes to a natural kind; second, how close that concept is to our ‘folk understanding’ of some surface phenomenon; and third, how easy it is to measure the concept once defined. You might note that none of these three explicitly brings in normative dispute. That is obviously an issue of concern with conceptual analysis in political science, but I believe it intersects with all three of these aspects. I will return to specifically normative concepts within political theory in Chapter 9. Here I will simply say that the easier it is to define a term across logically or naturally possible worlds, the less controversial that definition will be. The greater the utility of a concept in prediction, the deeper it is conceptually; the more technical its definition, the closer it comes to a natural kind. Conceptual analysis should be conducted only when one has a specific research question in mind. Concepts might be constructed as theoretical entities within a model or framework that tries to demonstrate the mechanism through which predictions are generated. Such concepts might have little obvious relationship to surface phenomena. The realism I have been supposing, however, ensures that they are tied to those surface phenomena, in the sense that they are predictive, and our tests of those predictions will examine patterns in those surface phenomena. I will begin by suggesting some principles of conceptual analysis that I will apply to issues in, first, empirical political science and then normative political philosophy. When we define a term we are usually looking for necessary and sufficient conditions for the application of the concept. A great deal of critical analysis of concepts in social theory derives from the process of finding counterexamples that intuitively constitute an example of the concept, but which are not contained in the previously defined conditions. Or, conversely, finding an example that fits those conditions, but which does not seem to be an example of the concept as defined. We cannot but help conceptualizing concepts in terms of necessary and sufficient conditions, though even as we

189

190 The Philosophy and Methods of Political Science do, we might feel sceptical that we can ever find such conditions for many social concepts. This sort of scepticism is expressed by those following Wittgenstein’s analysis of family resemblances, where two examples that come under a concept do not share any features with each other but do share features with other examples that share features with each other. I am not sure that the family resemblance metaphor is a good one. After all, family resemblances in reality are underlain by the expressions of genes, and so we might expect to find subsurface qualities that do constitute conditions that exist for examples of similar concepts, even if their surface qualities differ. A better argument for suggesting that there might be very different examples of the same underlying concept comes from historical accounts of naming. We saw this in Chapter 3 with Kripke’s historical account of naming. So ‘water’ is the name given to a specific substance that turns out (epistemically contingently) to be metaphysically necessarily H2O. Such truly metaphysically necessary relationships for the identification of an entity can lead us to dub such entities natural kinds. However, we should expect to find few, if any, such natural kinds in political science. We can expect to find entities with different underlying characteristics but which can be traced back historically to an initial definition. Thus we have very different forms of democracy that can be traced back to the original use of the term. Obvious examples in politics include ‘government’, ‘democracy’, ‘liberty’, and ‘political party’. There really is no point in trying to provide necessary and sufficient conditions to characterize such entities for all times and places. Rather, we should only define them theoretically for a given purpose at a given time. If one’s thesis is that political parties are in decline, then one is making some claim about how their general form was and now is (Dalton et al. 2011; van Biezen and Poguntke 2014). If one is claiming they are undergoing transformation, then one is making a general claim about how they are changing. The two theses might be identical apart from interpretation. ‘Decline’ has different normative connotations from ‘transformation’. We can make normative claims that, say, deliberative democracy is superior to participatory democracy or republican freedom superior to negative freedom. That they concern the same general entity can be traced historically, but which is superior requires normative demonstration. Theoretical entities such as ‘veto players’ can be given a priori metaphysically necessary conditions. We then apply a posteriori analysis to see whether any political agents have those characteristics to some degree. Our type-level explanation provides a mechanism to predict outcomes and we see to what extent those mechanisms apply at the token level. To the extent they apply, the agents in actual systems resemble veto players in theory. But note that the forces identified in models might constrain agents, even though they do not have the necessary and sufficient conditions that define the theoretical

Concepts and Conceptual Analysis

191

agents they resemble. Structural relationships exist even if the characteristics of agents do not conform one-to-one (Ross 2014). We should not be trying to provide once-and-for all definitions when doing conceptual analysis. Rather, concepts are to be applied to specific cases for specific purposes. Generally speaking, when we are doing qualitative work we need not concern ourselves overly much with conceptual analysis, for we are unlikely to be able to provide a definition that is not subject to critique. When doing quantitative analysis, or justifying why we include certain examples in some qualitative comparison, we might need to define our terms. This is especially the case when providing rules for coding purposes. Where we use machine-learning techniques we are more likely to be looking for similarities after patterns have been discovered. Whilst looking for a good definition involves trying to specify necessary and sufficient conditions, we should not be too worried about succeeding (unless we are constructing a formal closed model). That is not to say that we cannot produce some principles for good conceptual analysis.

8.2 Principles of conceptual analysis Concepts should be as primitive as possible The first principles of conceptual analysis should be easy to state. We should classify like with like; distinguish objects and ideas that are importantly different; and allow for further subdivision if that should prove necessary. But the sense in which something is like something else, or different from something else, is relative to our interest, our research question. Some fruit are alike in colour, others in shape. Either might matter for dispersal of the seeds within them, and so how they are classified might depend on the specifics of the dispersal of the seeds. Are we interested in democracies in terms of their party systems, the liberality of their legal systems, their governance structures? These should define the characteristics for comparison between types of democracy. We might want to think that ideally concepts should be as much like primitives as possible. In philosophy a primitive is something that cannot be further analysed. Once we have defined what a minister is, there is nothing more to say about that object. Of course, some concepts are more complex than others, but the idea of developing political concepts to be as primitive as possible is to ensure that theoretical change in any given concept reverberates as little as possible over all concepts within the theory. We do not want any concept to depend upon the whole of a theory. If our concepts are completely primitive, they cannot be further analysed and will not change as theory changes (that is, as we jettison some models in favour of others).

192 The Philosophy and Methods of Political Science

Box 8.1

Concepts should be as primitive as possible

• Classify like with like • We need to distinguish objects and ideas that are importantly different • Ideally concepts should be as much like primitives as possible – that means further sub-division should not be necessary • But many concepts are complex, so we need to allow for subdivision where necessary • Any concept might not prove fundamental, but rather based upon something more primitive still • Coherence across concepts is necessary, but concepts should not have to be redefined based on changes in other concepts no matter how small those changes are • Concepts should not depend upon the entirety of the theory in which they are instantiated

This may, of course, be a desideratum that we cannot expect political concepts to meet outside of formal models. We also need to allow that any of our concepts might prove not to be fundamental, but itself based upon some more primitive notion still. Neither of these latter principles denies the coherence or holism of theories, but they do suggest that no concept should rely upon the whole of theory for all of its meaning. No concept should become incoherent given change in any other concept, no matter how small that change. Of course, to the extent that concepts are mutually supporting, then big changes elsewhere might well force a change in every other concept. Complex concepts such as ‘democracy’ ought to allow for further subdivision. There might be major normative debate over what is important within a democracy. But rather than trying to define democracy in the round and defending that conception against all comers, the principle of making concepts as primitive as possible entails subdividing the elements of democracy. We might think democracy has many non-exclusive elements, such as: Participation Voting – different systems of voting Freedom of speech – forums for free speech Access to government – means of contacting public officials Deliberation Forums for public deliberation

Concepts and Conceptual Analysis

193

Organization of government Levels of government Responsibilities of government Rights and freedoms Constitutional guarantees for rights and freedom Constitutional guarantees for non-domination And so on. At any level we might feel that democracy might be enhanced, or threatened, by some change. Gerrymandering (redrawing boundaries for political advantage), for example, might be thought to threaten democracy even if other elements remain unchanged. Thus the rules for drawing boundaries and how political actors operate within those rules might affect our judgements about how democratic a given system is. Again, however, how we do this should be determined by the research question, and we should not be concerned with producing a definition of ‘democracy’ that is the best definition. I will say a little more about such complex and normative concepts as democracy in Section 8.5. The only point I want to bring out here is that the real theoretical work in thinking about a complex concept goes on at the subdivisions. Since these subdivisions could be weighted quite differently in terms of importance in democratic forms, we might not expect agreement over ‘how democratic’ different democracies are. Nevertheless, we can still compare the workings of different democracies by examining particular features. For example, there is a great deal of work on different electoral systems and their effects on electoral participation (Franklin 2004; Eijk and Franklin 2009); increasing study of the relationships between number of forums, participation and the general happiness of the citizenry (Frey and Stutzer 2002; Inglehart 2009); and on the deliberativeness of the democratic institutions and the quality of government and quality of life (Pacek 2009; Barker and Martin 2011). Much simpler coding of democracy has also been used to study economic growth and quality-of-life indicators, though careful analysis is needed to pick out the mechanisms (Doucouliagos and Ali 2008). In some regards conceptual analysis is about the purpose of using a concept, and to some extent the proof of the recipe is in the tasting. Referring back to the discussion above, however, precisely what is being conceptualized through the measurement and through the independent variables and how these relate to what is being measured in the dependent variable is what needs to be narrated in the purported mechanism.

194 The Philosophy and Methods of Political Science

Concepts should be as non-normative as possible Complex concepts in political science such as ‘democracy’, ‘freedom’, ‘rights’, and so on are strongly normatively laden. Our views and theories about what the good life of people should be affect our very idea of these concepts. The idea of making concepts as primitive as possible is to reduce the amount of normativity. When we break down democracy into baser elements, as I suggested at the end of the previous section, we are attempting to reduce the scope of normative dispute. We cannot hope to reduce it to zero even so. What constitutes freedom of speech and what constitutes legitimate forums for such speech are likely to be controversial. We can technically define different electoral systems, but given the remarkable results of social choice theory, we know that which systems are preferable depends on how we weight the different desiderata that they encompass. Normative debate will not go away as we analyse democracy into component parts. All concepts are normative in so far as they are used for prediction; and we want prediction to enhance our lives. Thus all concepts are normative to a degree. And it is clear that some concepts are more normative than others. Nevertheless, concepts should be as non-normative as possible. A concept can be as non-normative as possible in two senses (that I take from a remarkable, though as yet unpublished paper by Ian Carter); they can be as value-free as possible, or as value-neutral as possible. Carter (2011) defines value-freeness for any concept if the concept is defined in such a way that its definiens contains no ethically evaluative terms. A concept is value-neutral if it is defined in a way that is acceptable from any substantive ethical point of view. Another way of thinking about these two notions on non-normativity of concepts is (1) concepts should not be gerrymandered; rather they should fit as naturally as possible into normative desiderata; and (2) moral concepts should help us to develop our normative theories, rather than our normative theories determine our concepts. Value-freeness Just as electoral districts should be formed on the basis of desirable features (such as equal size, physical geography, social homogeneity) and not to fix electoral outcomes, our concepts should be developed on the basis of natural desiderata rather than in order to promote specific conclusions that we would like to see. This principle leads to the view that concepts should be as value-free as possible. For example, to define coercion as ‘non-justified interference’, rather than as ‘interference of a particular character’, is to further moralize coercion from the normativity the term already enjoys. The very notion of ‘coercion’ suggests being made to do something against one’s will. As long as we maintain the idea that freedom of the will of the individual is morally desirable, any

Concepts and Conceptual Analysis

195

coercive act has a point against it. Nevertheless, describing an act as coercive is not to say that the act was not justified. Sometimes coercing people might, all things considered, be justified. Hence to define coercion as ‘non-justified interference’ rather than as ‘interference of a certain kind’ is to gerrymander the concept. It is to extract from it what you want – coercion is never justified – rather than defining it in terms of normative desiderata and then seeing what your broader theory of the good social life will allow with regard to coercive acts. Some practices might clearly be coercion under a given conceptualization; how far and under what conditions that coercion might be justified should be left open to further analysis. It is far better to accept the analysis of coercion as some form of interference, and then allow subdivision into justified and non-justified coercion, recognizing that justified coercion might be wrong or unfortunate in some regards even if it is justified ‘all things considered’. This allows for greater nuances in the analysis. One way of thinking about value-freeness is that the hypothesis one is considering should have no part in the conceptualization of the concept one is considering. Thus, if examining whether an agent (a person or the state) is justified in forcing another agent to do some act, then one’s views on the purported justification should not affect one’s definition of coercion. It can be accepted that coercion is (generally) unjustified, but one should establish what makes coercion unjustified by considering paradigm cases of coercion, not outlier examples, and not let one’s own intuitions infect the definition of coercion itself. (Sometimes outliers do affect normative reflection: see Chapter 9, Section 9.4.) Here is an important caveat to the non-gerrymandering (value-freeness) rule. Sometimes empirical evidence so disturbs our concepts that we have to go back to the drawing board altogether. Sometimes we have to completely rewrite our theories and that involves reconsidering our concepts. In this case, the hypothesis might appear to remain in place, but only superficially, since the concepts within it have changed so much. Value-neutrality Moral concepts should help us to develop our normative theories, rather than our normative theories determine our concepts. In other words, concepts should not define theories. Now, the two processes go together. As argued above, the precise research question will help determine how we are going to define a concept for our purposes, and the model or mechanism (theory) will be part of that question. So in that sense theories contribute to our definition of concepts. However, we should not let our theories fully determine our concepts. Whilst empirical work will lead to modifications of concepts as we chase down the important elements in mechanisms (see Chapter 4, Section 4.5), our concepts should not determine the model form itself.

196 The Philosophy and Methods of Political Science The same principle applies when doing normative theory. The two processes go together, but we should not let our moral theory determine what our concepts are; rather our moral concepts should help us decide what is important when developing our moral theory. If our theory runs counter to our intuitions about the value of some moral concept, we should not simply redefine the concept in order to avert criticism of the theory. A concept of freedom that relies upon our understanding of Kantian or Rawlsian accounts of social justice in order to be comprehensible is the wrong way round. We should be developing a justification for our accounts of social justice in terms of how they promote desirable features of our lives, not finding out what is desirable about our lives through understanding particular theories of justice. Of course, we should reflect upon our moral concepts given what we come to believe through moral and political theorizing. Analysis is not simply from primitives up; rather the construction of necessary primitives is dependent upon what we need to develop theory. (This interrelationship of concept and theory is one aspect of the idea of ‘reflective equilibrium’ (see Chapter 9).) This is how it is in explanatory theory too. Our theories of justice will inform how we conceptualize freedom, coercion, welfare and the like; without a theory (no matter how ill-formed) these concepts would never have been formed, but they should not be conceptualized independently of other sources of evidence – social intuition, past understandings, general usage, how they might be measured – as to what they are. The account here is partially at variance with that of Carter, who argues that concepts can be defined in a value-free manner or a value-neutral manner, but not both simultaneously. Certainly, my claim that we ought to move from concept to theory and not theory to concept (with the important caveat in the above paragraph) discounts some types of normative theory, at least as they have been developed, and is thus not value-neutral on Carter’s account. Some moral and political theories trade precisely on their basic concepts being defined in a specific way. However, we should be open to the possibility that those theories can be respecified to ensure that their basic terms can be defined in as value-free manner as possible, taking as our starting point the concept not the theory. Whilst we should want our concepts to be as primitive as possible, they can of course be defined in terms of each other, particularly in formal systems. Thus in Euclidean geometry a point can be defined in terms of line or a line defined in terms of a set of points: either points or lines can act as the primitive. The same might be true within normative theories. A given conception of social justice might use either welfare or freedom as the basic building-block. We could define individual welfare in terms of individual freedom, suggesting, for example, that the welfare of a person is defined in terms of what they would choose under certain kinds of constraints. Or we might define freedom in terms of welfare: what is a person’s choice set under

Concepts and Conceptual Analysis

197

certain kinds of constraints? In other words, the same set of institutions we think normatively desirable under a given theory of social justice might be defended in terms of either welfare or freedom. In both theories we recognize the importance of freedom and welfare, but in one freedom is considered basic (the primitive) and in the other it is welfare that is basic. This is not to deny that taking welfare or freedom as the base normative concept is usually associated with rival systems of social justice (with different preferred ways of structuring society), but it is not the case that that which is primitive necessarily defines the just social structure. Welfare and freedom can be equally primitive; perhaps much social debate occurs because neither can be defined in terms of the other and the moral weight attached to each helps determine our attitudes towards rival ways of structuring society. Generally speaking, dispute over major contenders in theories of justice concern what relevant trade-offs there should be between freedom and welfare. The two concepts are not thought to be definable in terms of each other. Those on the right tend to weight freedom more heavily than welfare, those on the left welfare rather than freedom. But there are notable exceptions. The Nussbaum–Sen concept of ‘capabilities’ might be represented as an attempt to combine the important elements of freedom and welfare in one concept (Sen 1993, 1999; Nussbaum 2011), though its success is doubtful (Dowding 2006b). However, when considering the two concepts of freedom and welfare, we should begin by thinking whether or not one could be defined in terms of the other. We should not decide that one concept is superior to the other and then simply generate a normative theory from that concept. When writers choose that course, they soon bump up against moral intuitions and find they need to conceptualize their basic concepts in ways that depart further from intuitions in order to avoid counterexamples. There are many ways of cutting up the social world into basic concepts, but when those basic concepts are already in use in natural language, and that language affects our behaviour – it affects what we see as right and wrong, justifiable and not justifiable – it should also constrain our conceptual analysis. That is why making concepts as value-free as possible impinges on the possibility of making them as value-neutral as possible. In terms of empirical theory we can defend one way of conceptualizing, say, party spatial location as the only plausible one, and then develop theory based upon that conceptualization regardless of what hypotheses might be produced. In normative theory such a process generates implausible theories. For example, Nozick (1974) defines a particular moral concept, a right, as inalienable and fundamental and then generates a theory of justice. His rights-based theory seems to suggest that no matter what the welfare consequences, it is never justifiable to ignore individual rights. Even to Nozick, this claim is so implausible that he footnotes a caveat that rights might be

198 The Philosophy and Methods of Political Science overridden in times of moral catastrophe. We should inspect our concepts given what they seem to entail, and reconsider their worth in the light of those considerations. Sometimes definitions seem plausible until they are challenged with evidence. Some empirical studies of democracy define democracy in ways that exclude many of the countries in many of the years in their datasets. Many writers on conceptual analysis, often following the lead of Sartori (1970), write about different levels of conceptual analysis. Sartori represented this notion as the ladder of abstraction: as one goes up the ladder, concepts become more abstract or, we might say, theoretical. Conceptual change on the lower rungs affects our concepts higher up. Gary Goertz (2006) writes about the Basic Level, the Secondary Level, the Indicator/Data Level and then the Method of Measuring or Aggregation. These metaphors suggest rather too rigid a structure of conceptual analysis. To be sure, there are more complex concepts that can be subdivided and discussed in terms of other elements, but the more complex terms do not have to be more abstract than any of the elements. Certainly, if concepts should always be as primitive as possible, we should define concepts in ways that exhaust further analysis as much as possible; but that does not imply more concreteness of a term. We might have a reasonable measure of unobservable entities based upon their theorized effects, and these unobservable concepts might be quite abstract. Sartori’s particular concern was conceptual stretching and using proxy indicators in measuring the effects of ‘democracy’ or ‘freedom’ in ways that really did not capture those concepts properly. That assumes that there is (historically) a correct application. Certainly, we might feel that if a concept departs too far from the historical understanding of a term, it has lost its initial referent. But by Kripke’s historical account that matters when there are features that rigidly designate a concept. To the extent that rigid designation is implausible for political concepts, such shifting might not matter. It is true that people might conceptually stretch for ideological reasons: take a term with particular normative implications and stretch it over cases in order to impart that normativity to those cases. But that is a separate issue. Here we need to concentrate upon the characteristics that give the positive or negative normative connotations and see how far they apply to each application. That is another reason why we should make our concepts as non-normative as possible. Such problems can be handled by creating more complex concepts by extending the elements that go into their analysis (though that brings its own measurement problems, since there might be different ways of weighting those elements). However, in this way of thinking the contrast is simplicity and complexity, rather than abstractness and concreteness. In part that is because what matters most is how easy it is to measure concepts. I turn to this issue in Section 8.5.

Concepts and Conceptual Analysis

199

Conceptual space should be as simple as possible Conceptual space should be kept as simple as possible. The important element of any explanation is that the concepts utilized provide the motor of the mechanism or narrative that generates the outcomes. When we pare our concepts down to their most basic elements, we are attempting to clip away all extraneous factors. We are making our concepts simple. Returning to an example we have considered in earlier chapters, the idea that the number of veto players within political systems can predict policy volatility is powerful because it is so simple. The concept of ‘veto player’ has a simple, uncontroversial definition. Identifying which agents in actual systems perform the function of veto players might be more controversial, especially when we move beyond constitutional veto players such as presidents who can refuse to sign legislation to social veto players such as professional groups who can effectively veto by virtue of their role in implementing policies. This is the stuff of careful empirical analysis; it is the application of the model to specific cases rather than the model itself that might be open to doubt. We also see that by defining a theoretical concept, such as veto players, that is part of the structure rather than the surface properties of a political system (such as presidents, legislators, and so on), we ensure conceptual space is kept simple. In the theory of veto players, only two agents occupy conceptual space – agenda setters and veto players – rather than the myriad of roles we see in thicker descriptions of political systems. Type and ultimate explanation utilize theoretical concepts to a greater degree than token and proximate explanation. Ultimate explanations rely upon the below-the-surface structural features that are denoted by the identification of theoretical concepts with mechanisms and models. Again the veto-player model is a good example. When it comes to proximate explanations of token, such as why health reform failed in country X at time t1, but succeeded at time t2, we look to narrate the actual events involving token actors. The fact that these actors are veto players and agenda setters is not irrelevant to the story – it is why we examine their behaviour and not someone else’s – but any particular outcome will be a result of the interplay of their preferences, attitudes and the circumstances of the time. The proximate story will not be inconsistent with the ultimate explanation of relative policy volatility, but will be directed at explaining something rather different: policy change at different times. We keep conceptual space simple for ultimate explanation. We expand the space of our explanation when the nature of what we are explaining changes. Our simple conceptual space allows specifics of individual roles in token and proximate explanation. If conceptual space is too complex or detailed, such expansion can become problematic. In other words, if a researcher tries to make specific outcomes rely entirely upon their theory, then their theory is likely to be highly complex. Any empirical evidence that appears to

200 The Philosophy and Methods of Political Science contradict what that complex theory (model, mechanism) allows will require substantial and detailed theoretical modification. That will impinge upon how we view the concepts within it.

Allow for development The way in which we carve up conceptual space should as be as simple as possible, but further divisions might be necessary. Despite keeping theory simple, we should always allow further development of concepts. This is especially so the more normative the conclusions to be derived from your argument. In developmental terms, as our moral and political philosophy develops, so do our understandings of our basic concepts. When the social world is suffused with dictatorships and oligarchies, it might be enough to define democracy in terms of representativeness underpinned by elections. Such an idea of democracy is sufficiently radical and sufficiently desirable to virtually exhaust the meaning of ‘democracy’. Later, however, when representative democracy becomes a more normal state of affairs, and shows both its desirable and less desirable aspects, we might want to further develop a more inclusive account of democracy. We might feel that representation through delegates does not satisfy the requirements of participation, as it seemed to do when the only alternative was dictatorship. Thus we develop an idea of democracy that delves into other areas: a polity’s openness to the media; what rights and duties exist; how participatory they are (who is included, how easy it is to participate, how deep is participation in different economic and social organizations) – and then considerations of more direct input into democratic processes might require further divisions into deliberative and less deliberative forms; and so on. In other words, social conditions and new moral desiderata will develop our conceptions of democracy, but should do so by building upon previous conceptions rather than denying that those conceptions were of democracy. We should not deny previous concepts because we have a more developed, nuanced or complex version created over time, partly built upon our understanding of the practice of the earlier conception. We should be careful in criticizing earlier writers for their conceptualizations – we should always keep in mind what they were defining their terms against: it might not be what we are now redefining those terms against (see Chapter 9, Section 9.2.) We should recognize different conceptions of democracy and argue the normative desirability of some over others, rather than labelling only some ‘democratic’ or ‘true democracy’. That would be to gerrymander the democracy/non-democracy boundary in favour of particular theories of democracy. Whilst each division I have mentioned marks some normatively important boundaries, each should have its place in our conceptual lexicon. Within its conceptual space, the term ‘democracy’ itself should be kept as primitive as possible.

Concepts and Conceptual Analysis

201

Concepts should be as independent as possible Concepts are related to each other in obvious ways, but they should be kept as independent of each other as possible. I have been using the terms concept and conception largely interchangeably. There is a tradition in political philosophy to mark a distinction between a concept (such as ‘democracy’) and conceptions of it (such as ‘representative democracy’, ‘participatory democracy’, ‘deliberative democracy’, and so on). This distinction might be useful at times, but there is likely to be little purchase in trying to get it to do too much work. Conceptual analysis should be a more fluid process, engaged for specific reasons in different contexts. It is more important to understand that our concepts are theory-laden and that they depend upon other concepts, assumptions and arguments within the theory. We can think of an analytic theory as an interlocking whole of ideas, concepts, arguments, and so on. If the theory is perfectly analytic, then the meaning of each of its terms will depend on the meaning of all the others. A change in one will reverberate throughout the whole theory (this is the Duhem–Quine thesis in another form). Good conceptual analysis, however, will try to stand apart from the details of any specific model as far as it can. Whilst the meanings of concepts will change with changes elsewhere in the theory, concepts should not have to be rethought and rewritten following such changes. That might be impossible in interpreting formal closed models (which is what motivates Quine’s argument), but when we are using less formal models (and indeed natural language) to construct purported mechanisms, the Quinean thesis has much less purchase. This desideratum of concepts standing on their own as much as possible follows from the idea that concepts should be as primitive and as non-normative as possible. Ideally the same concept should be interchangeable across different models or normative theories.

8.3 The existence of concepts Do concepts exist? This depends upon one’s stance on the realist–anti-realist division. For some, concepts are simply instrumental. We saw this argument when we considered whether models can be true or false. For some people, models cannot be true or false: they merely represent reality in some manner. For anti-realists, concepts do not exist either. Thus freedom, democracy, veto players, the spatial location of parties do not exist. They are merely representations of that which does exist. For realists, to the extent that concepts map real patterns in the data, then concepts do exist. All conceptual analysis does, as a research activity, is to formalize a little what we do all the time. We examine patterns in the data and describe them with our concepts. This means that

202 The Philosophy and Methods of Political Science whilst we might have rival ways of patterning, those rival ways are not necessarily contradictory. (For the realist the surface reality is one pattern, and the theoretical reality is another. The latter is just as real as the former; indeed to the extent it is more predictive, it is ‘more real’.) When our concepts are simply descriptive of surface phenomena, the patterns we find are those that are immediately observable. They are often those that are straightforwardly socially identifiable. They might be fuzzy at the edges in some sense, but that might not matter. If you are introduced to a government minister at a social gathering, that introduction identifies that person in a political role. It might not matter whether she is a full cabinet minister, a junior minister, nor perhaps whether she is an elected or appointed politician (unless that matters to your personal esteem or you want to use her as a contact for some purpose). Only when those characteristics of the class of objects are being used to identify some specific pattern in the data – junior ministers are less durable (that is, more likely to lose their jobs on any given day) than full cabinet ministers – does such information matter. Only when one is conducting research with specific questions in mind might one want to define ‘minister’ more carefully. What specific characteristics we think might be important will be generated by our proposed model or mechanism. In this case our theory will generate issues over the specific conceptual analysis we conduct. Our conception of a minister thus generated will be theoretical. But we can consider that the concept exists. It exists in the sense that those characteristics we are imputing to ‘ministers’ and to different types of ministers are attached to people. And it is doubtful whether there are any significant changes in empirical analysis if you prefer to take an anti-realist or instrumentalist stance and say the concept is a ‘useful fiction’ to make judgements about ‘real ministers’. We do not always need to conduct such conceptual analysis prior to doing research. Sometimes it is done following data collection. We saw in Chapter 7 that some inductive methods do not require conceptual analysis prior to investigation. In machine learning we let the computer algorithms surface patterns and then try to work out what they show. We might utilize concepts and models borrowed from earlier research to interpret our findings, or we might create some new concepts to try to explain the patterns found. Such techniques can be abused if not handled carefully, but done properly inductive methods are a powerful research weapon. In text analysis, no specific theory has been adopted to lead the software program to concentrate upon any specific features of words or quasi-sentences. That is not to say there are no conceptualized objects – the program recognizes chunks of data in the form of grammatical symbols (full stops, commas, and so on) and words (or it could be used to pick out letters), and will demarcate strings of words into quasi-sentences which correlate with variance in the dependent variable. However, the researcher will impute the

Concepts and Conceptual Analysis

203

significance of such quasi-sentences after the program has identified the patterns. We have allowed a computer program to look below the surface of data to find a pattern that otherwise we might not have seen. We have not used theory to hypothesize a pattern that we then look for; rather, a pattern is found for us and we then produce a theoretical construction to explain it. For example, if we find that ‘concreteness’ as opposed to ‘abstractness’ in parliamentarians’ first speeches is correlated with later being chosen as ministers, we can theorize about that pattern (Dalvean 2012a, 2012b). The program has picked out certain features in the words, and we then conceptualize those features as displaying ‘concreteness’. Conceptual analysis here is an explanatory process. The ‘concreteness’ is the reality underlying the finding. Turning something into a concept does involve decisions. If we put an object into one category rather than another, we are making a decision. We should recognize, however, that many such decisions are decisions of distinction where others might only perceive degrees of difference. We might say that virtually all concepts are threshold concepts: we make a decision on a threshold on one side of which we place one concept (or subdivision of a concept) and another on the other side. We make decisions about when blue shades to green, when day becomes night, when a person is over the age of consent, when stones become a pile, when a group is large, and so on. Some of these threshold concepts – the age of consent, for example – hold simply through convention, though even conventions should be philosophically defendable. Though the precise boundary of where we categorize day or night, adult or minor, is one of judgement, it does not follow that the two categories are not themselves distinct. Night and day are different, and creatures react differently to them. Children are importantly different from adults in many ways. Quite when a given day turns to night or a given person attains the maturity of adulthood might vary (Cowden 2012). When thresholds are crossed, there are qualitative alterations that constitute the change from one to another. For some concepts there is a distinct break between one and the other, but not for others. Chemical compositions or social tipping phenomena are two examples where threshold changes have predictable and sometimes dramatic consequences. Where there are distinct changes in phenomena we might see them in terms of the philosophical concept of natural kinds.

8.4 Natural kinds Traditionally terms such as ‘water’, ‘cat’, ‘heat’ and ‘blue’ are natural-kind terms. They are naturally occurring and once the thing is denoted by its moniker then those things that share the essential features with that original item are of the same kind. Social or political objects, such as ‘government’,

204 The Philosophy and Methods of Political Science ‘government ministers’, ‘pressure groups’ or ‘freedom’ are not. They might be considered non-natural, as they are artefacts of society. However, we might consider that the manner in which they arise is equally natural. Government in some form (‘governance’, as recent fashion dictates) might naturally arise in any human society. Freedom might naturally arise (to a greater or lesser extent) in any human society, just as heat or shades of blue might arise under different conditions. In fact we might sooner take a rather different approach and discover the metaphysical necessities and then apply those features to examples. We have already discussed how decisions are made theoretically on how to distinguish animal species. It is not obvious that there are metaphysically necessary features of what constitutes any animal, as features might naturally diverge; what we place inside and outside of the cat category is open to interpretation. Kripke and many philosophers of language treat species as unproblematically natural kinds. Philosophers of biology either see species as not being natural kinds or weaken the conditions for natural kinds in order to make them so (Ereshefsky 2009). What we can say is that if we believe there are metaphysically necessary features of some patterns to which we apply nouns, then those nouns designate natural kinds. It is not obvious to me that species, let alone any political concepts, have metaphysically necessary features. To change the essential features of an animal requires only minor genetic or environmental conditions. Terms in the natural sciences, such as water and heat, seem rather different. Water as a compound of molecules of H2O seems metaphysically necessary. Once we have identified ‘water’ and given it that name, then it turns out, once we can do the chemical analysis, that being H2O is a necessary feature. That is so, since to change water into another compound and retain those features by which it was originally identified would require massive changes in the laws that govern physical universe. Heat, once that term has been applied, has the essential feature of transfer of energy with a net increase in entropy. Colours, once named, are defined by surface qualities under specified lighting conditions. What is important to these metaphysically necessary qualities is that they are subject to invariant generalizations. Thus rather than distinguishing the natural from the artificial, it would be better to distinguish natural kinds in social categories in terms of the invariance of their features. Soames (2010: 89; see also Soames 2003b: ch. 17) represents this Kripkean way of defining natural kind terms with three prerequisites, which I reword thus: P1 Objects to which the term is applied are similar in certain respects that guide our ordinary application of the term, fallibly but reasonably accurately, to new cases.

Concepts and Conceptual Analysis

205

P2 These similarities have a single unifying explanation that involves counterfactual invariant generalizations. P3 We use the term in generalizations and explanations, and wish to do so without identifying the term with the cluster of observed similarities. So water refers to H2O, not ‘colourless odourless liquid’ or whatever qualities associated with it by which it was first identified. Heat becomes the transfer of energy, not the ‘feeling of warmth’ by which it was first identified, and so on. So for political concepts, once a term such as ‘government’, ‘minister’ or ‘freedom’ is originally identified, anything that is metaphysically necessary to that term would then constitute that term as its natural kind. Of course, the problem then is, is there anything that is invariant in that manner by which we want to identify the concept outside of its original identifying features? There might be some necessary feature of government ministers, but it is not obvious to me that this is so, since ‘government’ can take many different forms. Is ‘freedom’ like ‘heat’? I do not believe the answer is obvious. Certainly, when we do conceptual analysis, what we are doing, by and large, is suggesting necessary and sufficient conditions for the identification of the concept under discussion. And much conceptual battle involves demonstrating that suggested definitions either fail to capture or fail to exclude examples of the concept under discussion. However, these definitions are usually refinements to the original identifying descriptions, not to something discovered a posteriori. If we want to look for natural-kind terms in the social sciences, we need to look for items whose scientific description departs from original usage. Often these might be terms created from models that typically more or less describe invariant mechanisms. The idea of ‘public goods’ came about because economists noted that some types of goods and services seemed efficiently undersupplied in relation to other types of goods and services. Modelling why this came about led to a definition of ‘public good’ in terms of its jointness of supply (or ‘non-rivalness’) and ‘non-excludability’; in this manner ‘public goods’ and ‘private goods’ might be considered natural-kind terms. Another example would be ‘veto players’ and ‘agenda setters’. Informally, a veto player is one who must agree to a legislative change for the status quo to change. An agenda setter is one who can place an item on the political agenda for discussion and/or decide the order in which items will be considered under a voting rule. These terms can be given more precise definitions, enabling modelling to suggest invariant generalizations across specific settings. Of course, many argue that there are no pure public or private goods, that all such goods vary in degree of rivalness and excludability (and indeed these vary with changing technical and economic conditions). Or that there are no people who have to agree in order for legislative change to occur (since

206 The Philosophy and Methods of Political Science everyone is subject to informal pressures beyond formal legislative rules), so that these terms only apply to items to some degree. To that extent, none of our social science concepts are natural kinds. The definitions of terms in the social sciences (‘ideal types’ in some analyses) do not apply perfectly to any item in the actual social world, hence no item can be considered a natural kind. We can define social natural-kind terms, but they do not perfectly correspond to any actual item. Where we have actual items that can be roughly described by their qualities, such as ‘government minister’, we find that they might be best defined differently (at least for purposes of coding in some analysis), depending on the research question (see Fischer et al. 2012). Nevertheless the failure of correspondence of natural-kind terms and the surface social objects does not mean that the processes represented in the models that instantiate those terms do not correspond to processes in the actual world. The structural features of the models correspond to structural features in the actual world to the extent that the models provide predictions that (non-miraculously) are corroborated by empirical evidence. For example, perhaps no capitalist enterprise actually profit maximizes. But how far enterprises flourish within a capitalist economic system reflects the extent to which they profit maximize. Profit maximizing is the theoretical concept that is important to understand their flourishing within ultimate explanation; why one firm manages to maximize to a greater extent than another would be the proximate explanation of the relative success of each.

8.5 Measurement A key factor in conceptual analysis is the issue of measurement. What one cannot measure one does not have a warrant for thinking exists. Measurement does not have to be precise. If you say, ‘he loves her more than she loves him’, then you have some kind of measurement in mind, and could supply evidence for such a claim. Interpersonal comparisons of utility are difficult or controversial, but we continually make choices for people (buying presents, for example). Government has to make choices over public expenditure that implicitly or explicitly make such comparisons (Dowding 2009). The first stage of measurement is nominal. Merely placing items under a category heading is the beginning of measurement. The issue that taxed Giovanni Sartori (1970) in his critique of conceptual stretching is using a proxy measure for some concept as though it were the concept itself. What you measure is what you measure and it is important not to trade on a concept’s good name. Coding the legal rights of citizens codes their legal rights; it does not measure the amount of freedom they have under most definitions of liberty. The legal rights of citizens might correlate with the amount of liberty they have under some reasonable definition, but that would have to be

Concepts and Conceptual Analysis

207

argued or demonstrated. But if it can be demonstrated, then why not use the measures that exist to do that rather than the list of legal rights? One needs to be careful about what one is claiming. The more one goes up through the hierarchy of measurement – nominal, interval, cardinal – the more what is being measured is likely to be a proxy for the concept you are trying to measure, unless the concept is a primitive. The more complex the concept, the more there might be to measure, and the more difficult it will be to measure accurately. If one has many components in a measure of democracy, then not only does each need measuring independently, but they might also need to be weighted in some manner for importance. And then again they might not be all independent of each other; their interaction might be important for how we judge the democracy of different societies. Hence theory and conceptual analysis will determine your answers as much as your empirics – which is another reason for keeping your concepts as simple and as non-normative as possible. The saving grace, perhaps, is that a simple measure of what country is the ‘most democratic’ is of little interest in itself. Rather, we are interested in explaining something for which a measure of democracy is useful. To the extent that it proves useful, it is capturing something that we can theorize is affecting our dependent variable. After we have found the importance of a given set of dependent variables, then we might utilize our conceptual analysis to try to determine precisely what is the important aspects of the mechanism or relationship we have found. Conceptual analysis involves both inductive and deductive processes. We begin by classifying items together that seem to go together. We look for underlying similarities, but we also deduce important relationships and consider logical relationships that bring concepts together. Our computers can aid this process. For example, principal component analysis converts sets of observations into uncorrelated variables or principal components (or factors) (Jolliffe 2002). Each principal component is independent of the others (and thus conceptually separate) as long as the original data set is jointly normally distributed. Factor analysis is closely related, but has more specific assumptions (Child 2006). In the social sciences these techniques were first used in psychology to try to discover underlying unobservable factors of the human psyche which could then be theorized about to produce concepts explanatory of human behaviour. In political science the techniques can be useful for combining data. Poole and Rosenthal’s NOMINATE is a similar multidimensional scaling model for combining and analysing roll-call data (Poole and Rosenthal 1997). Once combined, we can theorize about why they operate together as variables that potentially have independent effects upon our dependent variable. Machinelearning techniques (Mohri et al. 2012) take massive sources of data and inductively find some that correlate with our dependent variable. Again, such techniques sometimes find variables that we would not have picked out a

208 The Philosophy and Methods of Political Science priori as explanatorily important, but which we then theorize about. Such techniques are computational aids that do precisely what our minds do when we are conceptualizing and should be treated as such.

8.6 Some principles of classification There are various formal or semi-formal principles of classification (McKinney 1966; Bailey 1994). Personally, I do not believe there is much traction in formalizing classification principles beyond sanctifying common sense, except where the classification is designed as part of a theoretical explanation. The first purpose of classification is to describe the world. We create tables, typologies or taxonomies in order to do this. We should bear in mind that they are simplifications and are based upon decisions that we make to place certain items together under given headings. The rules we use should be considered carefully and we need to be aware of the purpose of classifying. Sometimes we might order items in a table simply to ‘get our head around’ the issue. A chronology of events might help us do this. A temporal typology might also serve some purpose within an explanation. For example, specifying the three waves of democratization (Doorenspleet 2000) gives a classification that is thought to require explanation. Historically classificatory systems form the basis of theory construction (Lenski 1994). They can underlie very different theories and hence be highly controversial (Ereshefsky 2001). Theorized classifications or taxonomies use two types of principles: (1) sorting principles and (2) motivating principles. The former sort entities into taxonomic units; the latter justify the use of the former. So classifying public policies into issues such as ‘environment’, ‘defence’ and ‘law and order’ might be justified by a wish to compare the relative importance of issues as seen by a government at a given point in time; or to compare the relative importance of issues in different countries; or to see if issues are dealt with at different levels of government in different countries; and so on. Classifying public policy in terms of how it affects citizens, such as ‘reducing freedom’, ‘increasing rights’, ‘regulating behaviour’, and so on, would be justified if we wanted to compare the freedom and rights of people across countries or to see if the state was extending its influence on people’s lives. Some typologies in political science are supposed to be explanatory ones (Bailey 1994, Elman 2005): that is, they list sets of explanatory variables in tabular form. In this shape they can be no more than notes or an aidememoire to either the causal claims narratively made elsewhere or to help us find our way through a model specified more formally elsewhere. One of the problems is that whilst a tabularizing set of variables can be helpful, once the table expands its heuristic utility quickly contracts. Increasing the favourite

Concepts and Conceptual Analysis

209

two-by-two table to three-by-three increases the cells to nine; recoding four dichotomous variables trichotomously raises the number of cells from 16 to 81. Already we have too much information to comprehend without further algorithmic manipulation. There are articles and books written on reducing the cells from the logical space by reducing the levels of measurement of variables (rescaling), combining variables (indexing), getting rid of improbable combinations (logical compression), eliminating empty cells (empirical compression), or simply deleting some that do no work in the explanation (pragmatic compression). If an explanation requires all these variables, then they are best handled in an empirical (statistical) model or in a formal model. Tabularizing is useful only in so far as it helps us get our mind around an explanation, and therefore needs to be kept relatively simple. I suspect explanatory typologies, tables and classifications are often more useful to writers than readers, especially in the early stages of model formation. We should never allow the nature of a theory to be determined by the heuristic utility of keeping typologies simple. Elman (2005) nicely illustrates a problem with using typologies as explanations in his discussion of Schweller’s (1998) balance-of-interests typology of states’ capabilities and interests in the pre-Second World War period. Names given to the types of states stand in lieu of the explanation, and the problem becomes reification of the names or types into the explanation itself, rather than their standing as labels for what is being explained. The names become independent variables explaining state actions, rather than names for the types of states in need of explanation. This demonstrates that ‘explanatory typologies’ are only ever representations or summaries of the mechanisms or narrations of the explanation itself. We can classify different types of diplomacy, different security dilemmas, types of revolutions, and so on, but these classifications are only ever labels for details of the mechanisms that are supposed to explain the cases. Typological theories of the democratic peace might have refined their concepts to different types of democracies and degrees of conflict, but these classifications are not themselves explanations. They suggest elements that are important in the specific mechanisms that are deemed to operate within each type. Typological theories operate as frameworks from initial classifications that structure the questions, the hypotheses and the explanations (George and Bennett 2005: 237–9). Various principles of classification might determine what one is doing. Usually, of course, items are given the same label because they share features, so similarity is the general decision rule. But what features are important is determined by the purpose of the exercise. Where the items are dependent variables, the similarity is based on hypotheses associated with the mechanism that is purporting to explain variance across the set of dependent variables with each different label within the classification. These differences might be related to hypotheses about causes or functional relationships.

210 The Philosophy and Methods of Political Science One needs to be aware of the decision rule when using the classification. Some rules are quite arbitrary and can be misleading if their arbitrariness is forgotten. One of the most obvious arbitrary rules is to classify a set of items by year. We might graph the number of parliamentary bills under different policy areas and note that a particular area has a large number of bills in a given year, and think that needs explaining. Graphing them by month or by parliamentary session (which might not correspond to a regular year, or be longer or shorter than 12 months) might smooth out the apparent peak. There might be nothing to explain. Thus, focusing on a single country, we query why the choice has been to classify by year rather than parliamentary session. But comparing across countries that have differing start and finish times and lengths of parliamentary sessions, classifying by year might make sense. In other words, there are different possible principles of classification. How we choose to classify depends on what we are trying to do. So if we are interested in examining the role of public servants in a bureaucracy, how might we classify them? By their place in the hierarchy, by the role they perform, by their terms of employment. It all depends upon what we are trying to explain. However, typologies can never operate as explanation in political science, unless they take on cladistic form.

Typologies in biology and social science It is in the use of classifications that social science differs most from modern biology. Political scientists classify by similarity, which is how flora and fauna were classified in biology pre-cladism. In order to explain how similar types of plants and animals grew up in different places, structural explanations are utilized through the eyes of natural selection within environmental settings. It is the similarity of the environments that lead to similarity of the forms. I describe this as structural, since it is the environment that structures those forms through changes (causes) via sexual reproduction. Cladistic typologies classify through lineages. So life forms are related to each other through the causal processes and the structures that lead to their similarities. If we want to understand the causal detail of political phenomena, and want to use typologies as part of the explanation, then typologies through clades are required. For example, the Comparative Agendas Project (www.comparativeagendas.org) that grew out of Baumgartner and Jones’s Policy Agendas Project (www.policyagendas.org) (Baumgartner and Jones 1993; Jones and Baumgartner 2005; John 2006) has a complex set of 19 major and 250 minor policy codes. Whilst some of the sub-codes differ across countries, and each country has developed specific interpretative rules, essentially the same codes are used across all countries to label policy agendas in categories such as government intentions, actual legislation, political debate (parliamentary

Concepts and Conceptual Analysis

211

questions, committee work), media, public opinion and government spending. The purpose is to compare the nature of policy agendas across countries. The work has demonstrated similar descriptive forms, often described as punctuated equilibrium: relatively stable agendas punctuated by periods where certain policies become more important for short periods of time (Baumgartner et al. 2009; Jones et al. 2009). This is a typical coding exercise for political science. It enables us to see how policy agendas share features across different countries with varying institutional forms, and facing (to some extent) different environments. Thus we can observe that the general outline of policy agendas does not change despite individual differences. We can also see how some institutional differences do make a difference to policy agendas; for example, the stronger party systems of European parliamentary systems produce party effects absent in the weak party and presidential system of the USA (Walgrave et al. 2006; Walgrave and Varone 2008). Thus the classificatory system provides a set of dependent variables for which comparative analysis provides structural variation. However, the classificatory system cannot itself provide much help by way of causal analysis. The evolution or, more properly, the path dependencies that lead to the generation of specific policies and how these produce a certain pattern in the policy agenda of a country require narratives and analysis. If a typology is to play a direct role in that explanation, as opposed to helping label the dependent variables to be explained, then sorting by clade is needed. The government might take a given issue to be related to defence – for example, the development of a missile system. The policy adopted (develop, test and utilize the missile) will affect industry (who builds it), the environment and individual health (consequences of the manufacturing and testing processes), and so on. The taxonomy might include primary features – what the issue is primarily about – which might be coded in terms of the department which is taking the decisions or steering legislation through parliament (say defence); secondary features – who else in the policy community has been involved (say industry); and external features (environment), which are defined in terms of the positive and negative externalities which become involved in the policy. The last might not emerge for several years after the policy has been implemented. These are all characteristics of the policy that to some degree could be coded or measured. In other words, the history and impact of the policy would have to form part of the sorting principles. Bluntly, such a use for explanatory typologies in political science is unrealistic, since we have no (relatively) easy means of tracing lineages for policy in the manner of genetic markers. Until such tracing becomes possible, classification in political science will always have to play a much simpler role.

212 The Philosophy and Methods of Political Science

8.7 Contests and contestability of concepts Most philosophical discussions of conceptual analysis bring up the issue of ‘essential contestability’. This is a thesis in political philosophy that suggests that some concepts are so normatively laden that there can be no agreement over their meaning. In part, the argument I have made that we should define concepts as non-normatively as possible is an attempt to undercut the argument that there can be no agreement. Of course, concepts are contested. If they were not, there would be no conceptual analysis. However, is it necessarily the case that there can be no agreement over concepts? We might agree that people have different views about what the good or just society should look like. In that sense, there will always be normative disagreement about what we want, about what we would like to see. The fact of such normative disagreement does not mean that we cannot agree about our concepts. Two people could both agree over the meaning of ‘liberty’ or ‘freedom’ and about what constitutes human welfare. However, they could disagree about the moral weight to be attached to freedom versus welfare. One person might think we need to trade losses in personal freedom for overall social welfare, another that we need to trade losses in welfare for overall personal freedom. Nevertheless, normative disagreement over the good or just society often involves disagreement about what constitutes freedom or welfare. One can adopt the subscript strategy (Chalmers 2011, 2012) to analyse verbal disputes. Here differing understandings of the same term can be ‘subscripted’ so that different meanings attach to different versions of the same word. (I adopt the strategy for the terms ‘interest’ (Dowding 1991: ch. 8) and ‘power’ (Dowding 2003).) Sometimes the subscript strategy reveals that there is no substantive dispute at all. However, it can only go so far towards revealing underlying tensions. There might be substantial dispute underlying the use of a concept. The disagreement over what ‘freedom’ means might boil down to disagreement over the weight we attach to different elements that make up personal freedom. Good conceptual analysis provides a process whereby we can try to reduce disagreement over concepts as such to disagreement over the moral weight we attach to characteristics of those concepts. This process will not guarantee agreement over concepts, especially where those concepts are deeply normative. Still, trying to define our concepts in a manner that is as non-normative as possible will enable greater commensurability in debate across moral theories.

Chapter 9

Analytic Political Philosophy

9.1 Introduction Most of the chapters in this book are about empirical political science. Theory, or political theory, has mostly been concerned with what is sometimes called (oxymoronically) ‘empirical theory’. In this chapter I want to consider some methods for normative political theory, or what I will refer to as political philosophy. The relationship between political philosophy and political science has occasionally been an uneasy one. Political philosophers are located in both political science and philosophy departments, and one can often detect a slightly different approach from normative theorists inhabiting each discipline. Theorists in philosophy departments tend to be more philosophical, as you might expect, more inclined to work in both moral and political philosophy, including meta-ethics, or in other branches of philosophy altogether. They also tend to be less interested in how their work might apply to the real world (see Box 9.1). Theorists in political science departments are more inclined to think about political philosophy in (what political scientists would recognize as) institutional terms, more inclined towards conceptual analysis and theory that crosses normative and empirical boundaries, and less likely to publish on moral and more straightforward philosophical topics. They are more likely to think about and apply their ideas to real-world events. Formal scholars are more inclined to logic in philosophy departments, and to decision or game theory in political science departments. These broad generalizations are not surprising, since academics are likely to be influenced by the work of their colleagues. There is, too, a great deal of crossover, as political philosophers move from one type of department to another during their careers. The relationship between political philosophy and political science is becoming increasingly uneasy as positive political theory grows in strength. The tension might be seen in the area of deliberative democracy, as its more philosophical underpinnings come under sharper empirical attack (for being unrealistic and inapplicable) and as more empirically inclined political philosophers work with purely empirical colleagues to develop deliberative forums as a practical step towards a stronger democracy (Parkinson and Mansbridge

213

214 The Philosophy and Methods of Political Science

Box 9.1

Rawls and Northern Ireland

In 1987 John Rawls gave the first H. L. A. Hart lecture in Oxford, introduced by Ronnie Dworkin with the words, ‘Today we listen to one of the world’s greatest political philosophers give a lecture in honour of the other one.’ After the lecture, a version of ‘The Idea of Overlapping Consensus’ (Rawls 2001: ch. 21), someone in the audience asked Rawls how his theory might be applied to the conflict in Northern Ireland. As I remember, Rawls somewhat sheepishly replied, ‘It doesn’t. I really have nothing to offer about the conflict in Northern Ireland.’

2013). I think it would be unfortunate if political science departments were to abandon political philosophy to the philosophy departments, because I think moral theory without political nous tends towards the irrelevant, and political science should remain normatively and philosophically inclined even as it becomes more technical and empirical. I cannot cover all the methods of political philosophy in one chapter, so here I restrict my attention to analytical methods. I say a few words about what I see as the central issues in work in the history of ideas, and then turn to what I consider the central methods of analytic political philosophy. First, let me suggest a simple schema by which to understand the role of political philosophy. Figure 9.1 schematically lays out some relationships. Meta-ethical principles inform both moral and political philosophy Figure 9.1

Stylized relationship of political philosophy to philosophy and political science

Philosophy

Political Science

Meta Ethics Political Theory

Moral Philosophy

Political Philosophy

Institutional Analysis

Behavioural Analysis

History of Political Thought

Analytic Political Philosophy

215

directly, but moral philosophy sometimes bears directly on political issues. Sometimes an argument could be seen as a contribution to either moral or political philosophy, or – of course – to both. On the right-hand side I have laid out three elements of political science: political theory, institutional analysis and behavioural analysis. I have given the last two the only twoheaded arrow, since each closely informs the other; indeed, analysing one requires making some assumptions about the other. In all other cases I have put single arrows, as the subject is political philosophy and I am looking at how the other elements inform it. In fact, though, one could just as easily put two-headed arrows, since political philosophy ought to be reflected back upon more foundational aspects for some questions. The history of political thought works through political philosophy to inform our other analysis on the schema. Figure 9.1 is a rough sketch, but it reflects what I see as the role of political philosophy in the broader scheme of political science. Political philosophy brings together some of the concerns of meta-ethics and moral philosophy; it should also be informed by some of our knowledge in political science. Thence it can also usefully inform other branches of philosophy and, more importantly, political science. One of the major topics in political philosophy in recent years has been how much political philosophy should be concerned with empirical or institutional issues: that is, what is the role of ‘ideal theory’ (Stemplowska 2008; Valentini 2012). This debate concerns several issues (largely the big questions of social justice), including how far our moral obligations go given any theory of justice; how far feasibility needs to be taken into consideration at the point of developing a system of justice (rather than at the stage of trying to institute it); and how much philosophers should think about the ideal of social justice and how much they should consider more pertinent questions of how we improve our lot now. The first and third are relatively straightforward, since there is no deeply philosophical reason why they should be mutually exclusive. The issue is the importance of each activity, not its meaningfulness. The second aspect – the issue of feasibility constraints – gets to the very heart of philosophical method. I will say a little about that debate here, since it affects my methodological guidelines below. The data of political philosophy include our views about what is right or wrong. Our views, our ‘intuitions’, are facts about ourselves. We cannot generate principles of justice without utilizing some facts, since those facts constitute part of the data of the subject. But how and when we utilize facts is one of the issues in the ideal/non-ideal debate. Before I get to these issues I will say something about working in the history of ideas (Section 9.2) and about grand theory or examining moral problems (Section 9.3). In Section 9.4 I discuss the major methods of analytic theory, including intuition pumps and how they are used in political philosophy.

216 The Philosophy and Methods of Political Science

9.2 Interpreting the work of dead people The essential problem of the history of ideas concerns how we can rightly and thoughtfully interpret the work of dead people. To some extent, the work of interpreting others remains the same whether they are living or dead. The dead, obviously, are less easily interrogated. One can ask the living ‘Did you mean x or y when you said y?’ and be told, ‘Well, more like ý.’ And one can then work out with the writer exactly what ý entails. Even so, responses from the living can be as slippery of interpretation as their initial writing. The main problem is that one’s own views get in the way. Burton Dreben (1992: 294) writes of Putnam’s interpretation of Quine: Like Hume, Quine writes superb English, and (hence?) like Hume, Quine is easy to misread. Putnam has misread him far less than many, but his misreadings matter. They occur wherever he most differs with Quine. … the misreadings signal the depth of the differences. They are also the almost inevitable result of Quine’s fundamental philosophical stance. (Since 1946 I have been a student of Quine. Starting with ‘Speaking of Objects’ – in 1957 – I have read and discussed with Quine, before publication, nearly everything he has written. Yet I am continually surprised at my own misunderstandings.) Dreben goes on, ‘One cause of the misunderstanding, the misreading, of Quine is the frequent difficulty in discerning from whence he speaks.’ Misunderstanding from whence someone speaks can happen with the living or the dead, but the longer someone has been dead the greater the chance of misunderstanding. To interpret the work of anyone, one must understand the culture in which they wrote. I have suggested several times in this book that one must understand the purpose of a theory (in the sense of a model) – that is, the reason why it was created – in order to properly understand it and its application. Often criticisms of models derive from their being applied to issues or problems for which they were not designed. This issue becomes even more pertinent when interpreting the work of dead people. The first step is to put oneself into their frame of reference: both what was motivating at the time – the moral and political issues of the day – and the culture – the background beliefs and attitudes all around them. An important aspect is what they could have been expected to know. Sometimes the biggest problem is appreciating not the big frame of reference for people from the past, the major issues of the day, but the little things that we take for granted. We often develop our thinking about new ideas through analogies and metaphors (Hofstadter and Sander 2013), but what was appropriate in the past might not be helpful today; and our analogies, that is, the very way we think, would not make sense to long-dead people.

Analytic Political Philosophy

217

We cannot have conversations with the dead nor dialogue through the pages of academic journals. Conversation is important. Sometimes I have learned more about the beliefs of an academic from a short conversation than from reading their books and articles. People do not always write down what they take to be obvious, and their starting place can be important to understanding the nature of their conclusion. Debate through academic journals is ultimately still more rewarding, as one can see protagonists explain more clearly their views, shift their ground or reveal hidden assumptions. And, importantly, it has a permanence that enables later interrogation to uncover problems and inconsistencies. It is that permanence which is important when discussing and criticizing the work of dead people. Although authors change their views, one often finds that their earlier work is still cited and used by others as though its originator had not moved their position. It might be that the original position is more interesting or perhaps the critic finds it an easier target. However, writers sometimes make claims that seem to be entailed by the earlier, not the later, position they adopt. Our thought processes often take familiar lines, and sometimes we forget that we discarded a train of reasoning when we saw it entailed a problem elsewhere in our web of belief. And after all, one cannot expect people to be consistent throughout their careers: we all change our minds and we all hold some inconsistent beliefs or fail to express ourselves well. Furthermore, our words take on a life of their own, an important issue to which I shall return. The way we normally interpret others is to use the principle of charity and try to work out the most coherent version of what we think that person is claiming. We might point out, then pass over, minor inconsistencies, make charitable interpretations of sections we do not fully grasp, before putting the point that we want to make. This might be to extend their work somehow or use it as part of our own argument. Or we might want to criticize on the grounds of internal inconsistency, empirical irrelevance or moral absurdity the best version of their argument we can create. But have we really interpreted the person’s views correctly? ‘Save me from charitable interpretations of my work’ is an oft-heard cry. I have at times found myself on the horns of a rather different dilemma. In the seminar room when a member of the audience comes out with a really nice account and asks me if that is my position, I find I want to take the credit, but fear I am being set up. So I usually answer, ‘Essentially that is my position, though I wouldn’t normally put it quite like that.’ In that way I claim credit, but give myself two outs: ‘essentially’ implies my argument is a littler subtler than the speaker allowed, whilst ‘wouldn’t normally put it like that’ allows me to disavow anything in the account that turns out to be a hostage to fortune. These are not just academic weasel words, but also an acknowledgement that we might not always be aware of all of the implications and entailments of certain ways of making points. Any claim we make should

218 The Philosophy and Methods of Political Science be conditional upon reasonable interpretations of that way of making the point. Dead people cannot condition in this manner on later formulations of their claims. The fact is that writers are not always themselves sure of the details of their own positions, at least until they are forced to elucidate them under criticism. Indeed it is difficult not to elide parts of one’s argument, particularly when one is not sure what one thinks. It is a mistake to give precedence to some unstated thought or intention over the written word. Written words have a permanency that such thoughts or intentions do not. Of course, we can disavow our words once we realize what they entail or how they have been interpreted, but that is not the same as giving greater weight to unstated thoughts or intentions. Rather, it is to give greater weight to the later disavowal. Sometimes people robustly defend positions that you know they cannot really believe – irritation may be a signal that they know they are wrong, but it can also indicate that they think the objections trivial; others of a different temper might give up positions too soon. Given these facts about the living, trying to work out what a dead person ‘really thought’ about something they did not explicitly state, or even about what they did write, is almost impossible. Rather, we need to concentrate upon reasonable interpretations. Indeed, what we express of our own views are only reasonable interpretations and sometimes others’ reasonable interpretations are better expressions of our beliefs. It is impossible to consider all the implications and entailments of any position one adopts; when these are pointed out, one might want to reconsider at least the nuances of one’s position. And that is the second problem of interpreting the work of dead people. Their views are incomplete and we can no longer ask them to complete them. So interpreters often do this for them. We extend their beliefs. We can imagine what they would have said about an issue, given how a contemporary could have interrogated them during their life. Or we think about how they might have responded to our questions were they alive today. To a large extent these two routes belong to two very different traditions of political philosophy. The history of ideas is conducted, broadly, in two ways. One is essentially elucidatory. It describes and explains the views of dead people. Sometimes this is done through interpreting the oeuvre of a particular person, sometimes through interpreting work within a tradition, though the two often go together. The second way is more critical or problem-driven. Here the work of the past is used more to reflect on events and ideas today. We might almost treat the dead writer, even if they died a long time ago, as though they were living. The two approaches come together when we use the work of the past to critique the way we think today. We might use past writers to explain how we got to our current views, and why perhaps we need to get out of those

Analytic Political Philosophy

219

ways of thinking, or to suggest that current views are incoherent and we need to return to the greater coherency of a particular writer or tradition. The true historian of ideas will be careful not to extend the ideas of a dead person beyond what they could have thought in their own time. They might be inclined to stretch a tradition a little more and integrate its thoughts into modern times. This is usually done critically. For example, modern republicans have used republican tradition to critique modern liberalism (Pettit 1997; Skinner 1998), and whilst a great deal of effort goes into elucidating republican thought in the terms of the dead, modern republicans are prepared to cut out anything from that original republican tradition that is not acceptable today (to democratize, feminize and socialize it, for example). Others suggest that updating in this manner makes republicanism incoherent in some manner (see Phillips 2000 for some of the ins and outs of these possibilities with regard to feminism). Whole ways of thinking can be criticized for leading us into certain types of conclusion. Carole Pateman (1988) argues that contractual analysis leads to a male-centred way of looking at social and political problems. Charles Mills (1997) has similarly suggested that the contractual tradition has been blind to ethnic considerations and the history of thought dominated by white males that have led to specific types of conclusions. Mills and Pateman take different lines, however, on precisely what this means for using the contractual tradition. Pateman believes contractual theory is itself implicated in biased reasoning; Mills argues that as long as the contract is set up correctly with the right types of contractors, then it can specify conditions of justice (see their debate in Pateman and Mills 2007). Others use the writings of dead people anachronistically. Rather than asking how the writer himself could have extended his thought, they ask how we can do so, though within the general bounds of what that writer might have thought were he around today. Of course this route opens up many more possibilities; when taking it, students should not be too concerned about defending their interpretation of the writer as the correct one. ‘Correctness’ under this extension has almost no meaning in terms of the interpretation of the writer’s actual view, but rather extends to the plausibility, coherence and general wisdom of the extension. The student needs to stand on her own more. In the history of ideas, the Cambridge School is associated with trying to get to grips with tradition and to understand the culture of the time. The leader of this school, Quentin Skinner, argues that we need to understand what an author is doing in relation both to other contemporary texts and to the politics of a time (Skinner 2002; Tully 1988). We need to understand the general tenor of the times, the ideologies as they existed and how the key texts changed those ideologies. Skinner points out that the texts we know as classics became so (often) because they change the way we look at the world; minor texts might provide better insights into conventional elite wisdom of

220 The Philosophy and Methods of Political Science the time. Of course, cultures are in flux and are less coherent and consistent than the writing of any interesting author. This means that we cannot expect to pin down the precise meaning of any dead writer, any more than and for the same reason as his contemporaries could. The essence of Skinner’s own account is his use of the intensional philosophy of language associated with Austin (1962), Grice (1989) and Searle (1969) that distinguishes the propositional meaning of texts and the point or intentions of speech. Skinner (2002: 82) wants us to understand not merely what dead people are saying, but what they are doing when they write. He is saying that words take on their own (conventional) meaning, but the writer also has something he is intending to convey. This might be a useful distinction, since intentions might include trying to obfuscate or simply to persuade without making any analytic sense; however, writers can achieve both these without intending to. As with other traditions, it is important to understand what the Cambridge School was reacting to, which was what they saw as naive and anachronistic readings of past writers (see Plamenatz 1963: Introduction, for a flavour) or Marxist interpretations. The importance of the Cambridge School lies less in that aspect (which I think everyone has taken on board now) than in their attempt to lead us to understand more clearly what our current conceptions of, say, freedom, rights or justice are, given where they came from; that, and their attempt to lead us away from current trends of thought to new ones. The deep problem is in what we mean when we say what dead writers ‘really mean’. It comes about if we privilege the original writer with the ‘real’ meaning. If we claim that the writer himself knows the real meaning of what he is saying, then we are making a very strong claim, and one of which I am not confident, especially when attempting to answer the tough questions I am prepared to ask myself. It is easy to shift from intensional accounts of meaning to think that it is thoughts (intentions) that shape sentences, forgetting that our sentences collaborate with our thoughts to express our beliefs. Once sentences are written down, writers do not exclusively own them; they belong equally to the sympathetic reader. That is as true of dead writers as it is of the living. We must respect the context in which dead writers were composing, but the text belongs to us too, and what it says to us now in our context might be more important than what it said to the writer’s contemporaries. For the historian of ideas, what the sentence said to the writer’s contemporaries is more important; for the political philosopher, what it says to us takes precedence. The historian of ideas needs to understand what motivated the author, and what the political issues of the day were. Skinner believes that ideological motivations are important for understanding what scholars are doing when they are writing political philosophy. However, it is the strength of their arguments that determines their worth in terms of political

Analytic Political Philosophy

221

philosophy. And for the political philosopher, what is more important is how those words speak to the issues of our day. Here we are not simply trying to understand ideas from the past in context, but applying modern knowledge of the underlying structures and mechanisms of social life to our understanding of what past writers understood. In that sense, we can understand better what dead writers were saying than they could themselves, in so far as they and we are talking about the same underlying realities. We might not expect a dead writer to agree with our (charitable) understanding, because they could not interpret their claims with our knowledge. Chemists know things that alchemists could not know, and can understand what alchemists were doing with their compounds better than the alchemists could themselves. To the extent that political science has found some of the underlying mechanisms of political life (and how far it has is, of course, disputed), we can understand better what was happening in the politics of societies in the past better than those who lived in them. And so we can be critical of their claims just as we can of those of our contemporaries. It might be ridiculous to treat Plato’s polis as Hobbes’s state, just as it would be to treat the trireme as a steamship (Collingwood 1939: 64), but trireme and steamship must respect weather conditions just as polis and state must respect social realities. When we look beneath the surface to the mechanisms that give plausible solutions to issues discussed over the ages, we can see why writers propose different sorts of solutions given the contexts of the past. And we can also see how different normative implications emerge from those different considerations. Furthermore, a great deal of the justification of moral positions depends upon empirical understandings. What is reasonable at one point in time might not be reasonable at another (Dowding 2013a). Concepts do not need to change for their moral status to change; their moral status can remain the same while their extension (what they refer to) alters; or the intension and extension can vary together. What interests Skinner and the Cambridge School is how our interpretations of current concepts are infused by past interpretations. Terms carry normative baggage, and as their extension changes the normative baggage might remain. By seeing how terms change their meaning, we can shed light on the ideology that suffuses our thought and affects our clear thinking. The distinction I suggested between the interests of the historian of ideas and the analytic political philosopher then disappears. The historian’s investigation into how terms change their meaning, carrying their moral connotations with them, enhances the weapons of the analytic philosopher. Empirically, I think this is correct; however, we might be sceptical that logically one needs to run through the history of conceptual change in order to examine the justification for the moral connotations of concepts. I can argue that the justification made for free markets for goods cannot justify a free market

222 The Philosophy and Methods of Political Science

Box 9.2

Interpreting in the sea of verbiage

Some sentences do not really make sense or, rather, what they say is so obscure that many different senses can be made of them. When you are struggling with new ideas, it is sometimes hard to make them clear. If you lack the relevant vocabulary, if the concepts at your disposal do not do the job, then you are forced into analogy or metaphor. Sometimes you are not sure what you are trying to convey but you are sure there is some truth in your attempts to do so. When a person writes obscurely there might be several interpretations of what he wrote. We might have more or less sensible or plausible accounts of what an author was trying to say. We choose for plausibility: that is to say, for the correctness or truth of the claim, not plausibility or correctness of the meaning ‘intended’, except in so far as it was the intention of the author to make plausible or true claims. This is not always the case. People can be deliberately obscure: John Searle once told me about a conversation he had with the late Michel Foucault: ‘Michel, you’re so clear in conversation; why is your written work so obscure?’ To which Foucault replied, ‘That’s because, in order to be taken seriously by French philosophers, twenty-five percent of what you write has to be impenetrable nonsense.’ (Dennett 2006: 405 n. 12) The clearer your writing, the easier you make it for critics, and the less room there is for interpretation. If you want to be a cult figure, then you should write unintelligibly or, better still, spatter unintelligible garbage with some trivial truths, as this gives others lots of room for interpretation, debate and argument about what you ‘truly meant’. It is also easier to be correct if your writing is obscure, since you can then choose an interpretation later. Read Foucault quickly. The central ideas are straightforward, and life is too short to bother about trying to interpret impenetrable nonsense – unless, that is, you want to make a career out of being a Foucault interpreter.

in financial commodities without tracing the history of the concept of a ‘market’ – though pointing out that criticisms of financial markets in the past resonate today will not harm the case (Hirschman 1986). Again we can utilize Kripke’s historical account of meaning. The original usage is important in fixing a term to a claim; but as our social world changes the designation can change. The ‘state’ today does not have the same characteristics as the ‘state’ in Hobbes’s day, and that matters. But what really matters to understanding and applying Hobbes today is how far his claims about the state hold now. What matters to the historian is how far his claims about the state applied when he was writing. In another frame of reference, discourse theory supposes our language shapes the entire (social) world (Gee 2005). For some this means that we cannot truly understand the meaning of those who use a different discourse,

Analytic Political Philosophy

223

making critiques of long-dead writers impossible. For others it is not so much the impossibility of critique that drives interest in discourse, more that we can track misunderstandings across time or groups based upon the different ways in which they use concepts and argumentative structures. In Chapter 6 I reported evidence that lawyers and scientists view the concept of causation somewhat differently, which means their assessment of evidence (notably statistical evidence) tends to differ. Foucault’s empirical work examining how our understandings of sexuality and punishment, for example, have changed over time is an important contribution – though in his hands discourse analysis examines how power relationships are expressed through language (Foucault 1975, 1980 and 2006). Foucault also provides a good example of the problems of interpreting dead people, that is, of trying to read too much into their writing (see Box 9.2). This is why I recommend that we mostly let the texts talk to us for our problems of today, and let the context do more work when we want to understand what the dead thought was going on in the past.

9.3 Moral problems and grand theory Reflecting on political theory in the new preface to his 1965 Political Argument, Brian Barry (1990) suggests that political philosophy took on a new lease of life with the work of John Rawls. Post-war Anglo-American political philosophy was moribund; utilitarianism dominated and thus the only question to be faced in any moral or political issue was how interests were to be weighed for the greater good. Rawls suggested that there were still big moral issues to be addressed and questions to answer, and revitalized contract theory by which to frame answers. The journal Philosophy and Public Affairs was instituted in 1971 (the same year as Rawls’s Theory of Justice was published), originally to unleash moral and political philosophers on to major public policy issues such as abortion, just war and exploitation. We might think that modern analytic political philosophy has two aspects. First, grand theories of social justice, democracy or, as the fashion for the really big issues wanes, slightly less grand theories of other aspects of politics; second, consideration of particular moral or political problems. This split mirrors one of the divisions I mentioned with regard to political science: that between theory-driven and problem-driven research. Grand theorizing should not really be indulged in until one is a grand theorist, though younger writers can legitimately produce general theories on more restricted topics. Far more is written now on what we might see as moral or political problems, though much of it is at a high level of abstraction, where the problem itself is less discussed than the issues underlying it. A great deal of effort goes into attempting to work out principles by using paradigm

224 The Philosophy and Methods of Political Science examples as thought experiments or intuition pumps to generate broad principles to apply. A recent book by LePora and Goodin (2013) exemplifies the approach. LePora is Program Manager for Médecins Sans Frontières for the Middle East, and the book opens with a particular dilemma. As a doctor, LePora was asked by a young soldier whether he ought to wear condoms when engaging in rape. She answered in the affirmative, since condom wearing would be advantageous to the victim. But does that make her complicit in the crime? The book uses a series of intuition pumps, as well as dictionary delving, to catalogue types of compromise and complicity in terms of moral worth; the authors then apply their analysis to some more general moral issues. It is worthwhile elucidating some good and bad practices with these techniques, and also to reflect upon what these techniques can tell us about grand theory or foundational moral and political theories.

9.4 Conceptual analysis, thought experiments and intuition pumps Empirical political science has data on which to base its workings. One can have empirical data on moral and political issues, of course – one can survey people’s views, and track general moral and political trends in different societies. Political philosophy does utilize such data occasionally (Swift 1999) (and should do so more: see below). The main purpose of political philosophy, however, is not to analyse what people think, but to establish what they should think about major political issues. Its subject is what Barry (1990) calls ‘political argument’; that is, not the mere elucidation of political or moral views, but their justification. It can utilize empirical evidence – what we as a society think and notably what the political philosopher herself thinks. It is legitimate to criticize certain normative claims on the grounds that empirical evidence disproves or makes them unlikely. Where once it might have been reasonable to think that ethnic groups had different moral worth, based upon their level of development, that is now not so. We know that because of what we have found out about different cultures, and also from what we know genetically about human capabilities. Our moral ‘intuitions’ form the background data. However, on their own they are not enough. Our intuitions are, of course, subject to our own culture and upbringing, but we can still use them, subjecting them to critical analysis. Data, as I have described them, comprise a relatively organized set of evidence, and such background data need more organization. This occurs through conceptual analysis, along with new data being created through thought experiments and intuition pumps. Together, these processes feed into

Analytic Political Philosophy

225

creating the grander theories, and allow reflection on the moral and political problems. Theory informs intuitions and these then inform theory to create coherence in what Rawls (1971, 2001: ch. 1) calls ‘reflective equilibrium’ (see also Goodman 1955; Daniels 1996). How far this process is a methodologically legitimate exercise is discussed below; despite any flaws, we might find it difficult to think what we might replace it with. To some extent political philosophy lifts itself up by its own bootstraps by creating its subject matter through argument. In doing so, however, some rules need to be created and followed.

Conceptual analysis I elucidated principles of good conceptual analysis in Chapter 8 and will only add points specifically relevant to political philosophy here. I argued there that our concepts should be defined as non-normatively as possible. This is so for normative as well as positive theory. The idea is that the meaning of our concepts should be not determined by the aims of the grand normative theory. Thus if one is defending a libertarian theory of justice – the just state should maximize liberty as much as possible – one should not define liberty in such a way that any justified state interference in a person’s life would not count as a restriction upon liberty. Libertarians might accept minimal constraints on liberty, but we do not want to gerrymander the concept in such a way that the theorists’ preferred account of justice automatically provides those minimal constraints. Rather, we should define our account of libertarian justice, hold it up against liberty defined as non-normatively as possible, and then argue that constraints upon that liberty that the theory justifies are the minimally justified ones. Starting with a good dictionary definition is not a bad idea. The Oxford English Dictionary (OED) online is comprehensive not only for meaning, but also the history of a term. Seeing the etymological generation and development of a term can be rather revealing and helpful for conceptual analysis; one can see how it acquires some of its normative baggage. The fact that modern philosophical usage is often not included in the OED definition also brings home the fact that such usage is jargon; one ought to be wary of allowing the normative baggage to ride on the philosophical usage without careful reflection. We can think about three criteria for conceptual analysis within political philosophy: the semantic, the normative and the methodological (Dowding and van Hees 2007). The first suggests that our usage should accord with our everyday usage. In positive theory there might be advantages in creating technical terms that examine structural features underlying our usual perceptual gaze. These terms extend over structural and causal elements that might only be discovered theoretically and measured quantitatively. However, political

226 The Philosophy and Methods of Political Science philosophy is concerned with the intentional attitudes and behaviour of people. It is about our morality, and thus its foundational concepts should, as far as possible, mirror those in our usual discourse. Since conceptual analysis is about sharpening our discourse, concepts will take on meanings or be restricted from meanings used ordinarily. And in order to expand our moral universe we need to change the extension of concepts to take on new understandings or make up words. The concept of ‘exploitation’ has been expanded to cover behaviours well beyond its original scope, as our notions of what is socially unjust have enlarged. A language with no word for ‘rape’ would make it more difficult for those sexually abused to explain that abuse and affect moral attitudes to behaviours. Thus moral and political philosophy is not just about elucidating more carefully the moral discourse of ‘ordinary people’, but about expanding that discourse. Nevertheless, when we try to change language we should take care not to depart too far too quickly from everyday meanings. The normative criterion is that in sharpening our vocabulary we should not change the moral force of terms. Whilst I have insisted that our political concepts should be as non-normative as possible, we cannot, and should not, try to pare away that normative force too much. For example, many people have died in the name of ‘liberty’; to define the concept of liberty in a way that makes their willingness to fight for it eccentric would fail the normative criterion. There are sometimes political reasons for overturning the normative significance of a word. The methodological criterion is designed to keep in mind the question being posed. The topic of ‘freedom’ in a deterministic world is not quite the same topic as ‘freedom’ in the modern state. We might expect there to be some bridge between the two, and perhaps our musings in one field will affect our considerations in the other; but we should not necessarily expect the term ‘freedom’ to bear only one meaning. Indeed the methodological criterion can be kept in mind by simply perusing a dictionary to see how many words take on quite different meanings, though – importantly (see below) – often deriving from the same root. The main purpose of conceptual analysis is to bring coherence and consistency. We want our moral concepts to be as clear as possible and free from contradiction. Formal analysis in empirical work can help sharpen clear hypotheses to test with empirical data. Formal methods, whether mathematical, semi-mathematical, social choice, decision and game theory or logic, can perform a similar task in normative theorizing. They can clarify precisely the claims being made. Theorists in philosophy departments are more likely to use logic and those in political science decision or game-theoretic techniques, though they are increasingly being combined. Formal techniques and clear analytic methods provide clarity, and contradictions or incoherencies are less likely to be hidden than with more literary and expansive styles. Something

Analytic Political Philosophy

227

can be lost with such Spartan language, however. The language of moral concepts can, and perhaps should, contain the rhetorical flourishes that inspire people. Aristotle (1976: 65 [Book I, 1094a]) commented about politics: It is the mark of the trained mind never to expect more precision in the treatment of any subject than the nature of that subject permits; for demanding logical demonstrations from a teacher of rhetoric is clearly about as reasonable as accepting mere plausibility from a mathematician. It might be the case that our ideas about liberty cannot be made completely consistent. To put the same point Aristotle’s way, it might be that if you make the subject ‘liberty’ precise, then you will find that you have lost the subject of liberty altogether. In Chapter 8, Section 8.5, I suggested that if we cannot measure something, then we have no warrant for thinking that thing exists. We can measure freedom in the sense that it is plausible to assert that one society has more freedom than another; or to know that once one has left the constraints of a prison, or an overbearing family or community, one’s own individual freedom has increased. However, measuring freedom or freedom of choice in a manner that seems consistent with all of our intuitions about what that concept is has proved problematic (see Dowding and van Hees 2009 for a review). That might suggest that whilst we are warranted in claiming that freedom exists, we have no justification for thinking our conception of freedom is, or indeed can be, entirely consistent. Liberty might be a basket of desiderata that in the end proves impossible to shape into one complete and coherent entity. It is often suggested that moral and political concepts are essentially contested (Gallie 1956; Connolly 1983), largely on the grounds that they are evaluative, internally complex and often open-ended. Analysis cannot produce agreement over the meaning of such terms. But I believe that by making our concepts as non-normative as possible we can, by and large, avoid moralizing concepts to make them essentially contestable. And where that proves difficult we can follow the ‘subscript strategy’ (Chalmers 2011, 2012) of giving concepts subscripts so that we can see that the same word is being used with different entailments by different people. Another way of thinking about this is to distinguish a concept, such as fairness, from conceptions of it, such as egalitarian notions of fairness (see, for example, Dworkin 1977: 134–6). It is possible that both the concept and the conceptions thereof might remain contested. However, to the extent that moral and political concepts cannot be rendered fully coherent because they contain a basket of inconsistent desiderata, there is a residue of contestation. In any debate one person might give greater weight to one desideratum, the next to another. Indeed, at different moments one might do so oneself.

228 The Philosophy and Methods of Political Science

Thought experiments Anyone wanting to know about thought experiments and intuition pumps and how useful they are should read Dennett (2013). Indeed the term ‘intuition pump’ is Dennett’s (1980); it was originally used to suggest that Searle’s (1980) ‘Chinese Room’ example did not provide a proper argument, just an ‘intuition pump’. Despite that criticism, Dennett considers intuition pumps to be useful devices if deployed correctly. In this section I want to lay out some methodological considerations about good and bad ways of using intuition pumps in moral and political philosophy. I will refer to a set of well-known intuition pumps by the names I give them in Boxes 9.6 and 9.7. Readers unfamiliar with these examples will need to consult the boxes. Dennett suggests there is a distinction between thought experiments and intuition pumps: the first are rigorous, the second ‘little stories designed to provoke a heartfelt, table-thumping intuition – “Yes, of course, it has to be so!” – about whatever thesis is being defended’ (Dennett 2013: 5–6). I define thought experiments as requiring one to think through the logic in order to work out the result. They do not pump intuitions as much as challenge them. Intuition pumps, on the other hand, stimulate the immediate intuition, as Dennett suggests. But what is an intuition? Philosophers disagree: some consider them ‘simple opinions’ (Lewis 1983: x), others ‘simply our beliefs’ (van Inwagen 1997: 309) and others still ‘intellectual seemings’ (Bealer 1998: 207). Historically the scholastics saw ‘intuition’ as spiritual perception or immediate knowledge; but perhaps it is best seen as an immediate apprehension beyond any conscious reasoning process. Certainly intuition is usually considered to strike us immediately rather than be something that has been worked out. One might immediately intuit that the product of 3200 × 15 is higher than that of 4200 × 11, but one does not work out that it is indeed so by intuition, but by doing the maths. But maths examples also show us that intuitions can be wrong, especially when it comes to probabilities, as Zeckhauser’s paradox illustrates (Binmore 2007: 133–4). Sometimes when we immediately intuit that something must be the case, reflection can make us realize our intuition is in fact wrong. Thought experiments can help provide that reflection. Thought experiments are usually used for one of two purposes. First, they are used critically to produce some contradiction in a theory; second, for heuristic purposes, to provide some analogy or metaphor to aid understanding. They are widely used in science as well as philosophy. Galileo’s ‘Joining the Objects’ thought experiment is designed to demonstrate a contradiction in Aristotle’s account of velocity. Einstein (unsuccessfully)

Analytic Political Philosophy

Box 9.3

229

Joining the objects

Galileo Galilei’s often-referred-to thought experiment involves showing that two objects passing through a medium of the same resistance will fall at approximately the same speed (dependent on resistance), and hence at the same speed in a vacuum, contrary to received wisdom from Aristotle who suggested bodies of different weights had different natural velocities. Galileo demonstrates this through a dialogue that involves Salvador leading Simplicio into a contradiction. Salvador asks Simplicio whether he admits that falling bodies acquire a definite speed fixed by nature that cannot be increased except by force or resistance. Salvador then asks what happens when two bodies whose natural speeds are different are united: will the more rapid be retarded by the slower and the slower hastened by the more rapid? Simplicio agrees that is indeed what will happen. Salvador then points out that the two objects united weigh more than the heavy one alone, hence the new object should fall more rapidly than either of the initial two, a direct contradiction. Simplicio admits, ‘I am all at sea because it appears to me that the smaller stone when added to the larger increases its weight and by adding weight I do not see how it can fail to increase its speed or, at least, not to diminish it.’ Source: Galilei 2012/1638 (see First Dialogue)

attempted to knock down the Copenhagen interpretation of quantum mechanics through thought experiments. These are the destructive use of thought experiments. Einstein also produced thought experiments to help explain both special and general relativity (indeed they helped him to come up with both models). Virtually all of Einstein’s positive results emanated from thought experiments. Thought experiments often do not pump intuitions at all and can be hard work. Einstein’s thought experiment ‘Train’ is meant to explain why time is relative and not absolute; it requires some thought to work through the logic and it can puzzle. If the man on the embankment sees the man on the train simultaneously with his seeing the two bolts of lightning, one might think the man on the train must also see the bolts of lightning simultaneously. But he cannot. If the two men are moving relative to each other they cannot both see the bolts of light simultaneously. Our initial intuition is wrong. Most educated people today are not surprised to hear that simultaneity of events is relative to position, because most educated people, if only through osmosis, have a smattering of understanding of Einstein’s theory

230 The Philosophy and Methods of Political Science

Box 9.4 Train Suppose a very long train is travelling along a railway line with constant velocity v. Two events at points A and B on the embankment will appear to occur at particular points (A and B) along this very long train. So simultaneity can be referenced relative to the train in terms of the embankment. Are two events that are simultaneous with reference to the embankment also simultaneous with reference to the train? v Mj

A

Mi

Train

B

Embankment

If lightning bolts strike the embankment at A and B and observer Mi halfway between them sees them as simultaneous, the bolts will also be events at A and B on the train. Will person Mj on the train also see those flashes as occurring simultaneously? Whilst person Mj coincides with Mi just when, from Mi’s viewpoint the lightning strikes simultaneously, Mj is moving towards B and away from A with velocity v, and so he will see the flash from B before he sees the flash from A. From the point of view of the embankment, flashes A and B are simultaneous; from the point of view of someone on the train, they are not. So events that are simultaneous with respect to the embankment are not simultaneous with respect to the train. Relativity says there is no way to decree that the embankment is at rest and the train in motion, only that they are in motion relative to each other. Thus there is no means by which to say that two events are absolutely simultaneous, only that they are simultaneous relative to some point. Source: Einstein 1916: 25–7

of relativity. For Einstein and his generation, the Train thought experiment proved revelatory. In political science we can think of game-theoretic models, such as Prisoners’ Dilemma or the Stag Hunt, as thought experiments, where (occasionally) surprising conclusions are deduced from the premises. Sometimes we forget how surprising the conclusions once were. The conclusion of the simple Prisoners’ Dilemma is now generally accepted, but through the 1970s and beyond some very clever people tried to explain it away. The conclusions are now intuitive; back then they were counterintuitive. Thought experiments challenge intuitions until their lessons become intuitive.

Analytic Political Philosophy

231

Box 9.5 Two toy game thought experiments Prisoners’ Dilemma Confess

Not Confess

Confess

–5,–5

0,–8

Not Confess

–8,0

–2,–2

The numbers stand for years in jail, but can be considered with associated ordinal utility scores. Each prisoner is offered a deal: if you confess, ensuring your accomplice is convicted, you go free, and your accomplice will get the maximum tariff of eight years; if you both confess, you both get remission, so will serve five years; if neither confesses, you will both be framed on trumpedup charges and serve two years. Confessing is the best strategy, no matter what the accomplice does. The dominant strategy leads to a worse outcome for each player than the dominated strategy. Stag Hunt Hunt Deer

Chase Hare

Hunt Deer

3,3

0,1

Chase Hare

1,0

1,1

The numbers here can be considered the utility from eating venison or hare or going hungry. If it were a matter of hunting a deer, everyone well realized that he must remain faithfully at his post; but if a hare happened to pass within the reach of one of them, we cannot doubt that he would have gone off in pursuit of it without scruple and, having caught his own prey, he would have cared very little about having caused his companions to lose theirs. Rousseau 1754/1984: 111

The idea is that, hunting together, the players eat venison, but if one player should break the circle the deer will escape and everyone will go hungry apart from the one who caught the hare. There are two Nash equilibriums, (3,3) and (1,1), since once each player is choosing one strategy, there is no individual incentive to deviate.

Intuition pumps Intuition pumps do not work like these thought experiments, though they can be used destructively as in ‘Joining the Objects’ or positively as in ‘Train’. Searle’s original ‘Chinese Room’ intuition pump was designed to criticize an account of consciousness – but here he expects the apprehension to strike

232 The Philosophy and Methods of Political Science us immediately. However, we have seen that intuitions can be wrong. But in order for an intuition to be wrong, there needs to be some process by which we can judge it to be wrong. As we shall see, methodologically, this is something of a problem for the use of intuition pumps in moral and political philosophy. Intuition pumps work by producing an immediate reaction to the example: they stimulate the intuition. Intuition pumps are designed to create data (our intuitions) that help assess the plausibility of a theory or argument. Thus ‘Surgeon’ is sometimes used to critique utilitarianism and ‘Shoot the One?’ to critique the idea that maximizing utility can be the only moral imperative (Box 9.6). The idea is that even if we all agree that the right thing to do is to shoot the one person, we feel this is wrong. If maximizing utility was all there was to say, morally speaking, we should not feel that. ‘Surgeon’ can only work to refute utilitarianism if we do not accept that the utilitarian calculation, once carefully and soberly made, does not prove that the immediate intuition (that the surgeon should not operate) is false. A utilitarian might claim that the maths does in fact prove the surgeon should operate. ‘No,’ says the critic, ‘moral theories are not just maths and “Surgeon” shows us that utilitarianism does not represent our morality.’ The data prove the moral theory wrong. But can we trust the data? We note here an important difference between thought experiments and intuition pumps. The former are designed to make us work out the reality underlying surface phenomena. They provide insight into that which we do not immediately apprehend, much as I suggested good mechanisms do for understanding social or physical processes. Intuition pumps use surface

Box 9.6

Surgeons and shootings

Surgeon You are a surgeon with five patients who will soon die if you cannot provide transplants. Two need a lung each, two a kidney each, and one needs a heart. Luckily another patient has just arrived for a yearly check-up, who is a perfect genetic match for all your patients; you could whip out his lungs, kidneys and heart to save the five, though of course he would die as a result. Should you kill the healthy patient to save the five others? Source: Thomson 1985, suggested by comments in Foot 1968

Shoot the One? A madman is going to shoot 20 people, but says that he will spare 19 of them if you shoot one. Should you shoot the one? Source: Williams 1973

Analytic Political Philosophy

233

phenomena (data) to challenge claims about underlying (moral) reality, to contest moral theories. However, moral theories are designed to guide our morality, so how can moral intuitions challenge theories? John Rawls (2001: ch. 1; 1971: 20) suggests a process by which we use our intuitions to challenge theories but also allow theories to challenge our intuitions. We modify each to get them into what he called reflective equilibrium (see also Goodman 1955: 65–80). Part of Rawls’s idea here is to attack logical or foundational accounts of ethics (read utilitarianism) that seem to get the wrong answers. He explicitly invokes the idea of using observational data to test physical theories as an analogy in his account of reflective equilibrium to attack foundational ethical theories. The idea is that a moral theory, even a completely coherent and consistent one, which does not fit our intuitions must have something wrong with it. (Compare this to my remarks about normative conceptual analysis on pp. 225–6 above.) In the light of observations we modify our theories, though our theories might also lead us to reinterpret our observations. There is a general problem with using intuitions as data in this way, since the intuition is a psychological fact. However, psychological facts about my moral assessment of ‘Surgeon’ or ‘Trolley’ are not the same as evidence about the moral assessment of what is the right thing to do in these examples (Williamson 2007: ch. 7). Unless, that is, the psychological facts are what constitute the moral evidence. If we are seriously going to use people’s intuitions as observational data, we would have to conduct psychological analysis of moral beliefs in a community in a systematic manner. This is an empirical not a normative exercise and, as we have seen in Chapter 5, we cannot make generalizations from an n of one, nor can we use a single case study to confirm or disconfirm a thesis. We might also note that not only do we need to take account of attitudes from the intuition pump examples, but also to vary the order in which they are presented, as this might affect people’s intuitions. Otsuka (2008) provides an exhaustive analysis of a set of ‘Trolley’ and ‘Bridge’ cases and suggests that our intuitions in some of these cases spring from the order in which they are presented. That is, we have a strong intuition in one case T1, which leads us to a similar intuition in T2, and so on to T5. However, had we started with T5, we would have had a different intuition about it, leading to different intuitions about T4 and T3, even if our original intuition about T1 remains the same. Otsuka provides some casual evidence and references some slightly stronger empirical evidence to that effect (Hauser 2006). To the extent that reflective equilibrium relies upon our ‘moral intuitions’ as being data in much the same way as data from scientific experiments, then it changes the nature of the enquiry. Such systematic analysis provides a description of ethical attitudes within a community, and we might find that different communities have different intuitions. Such empirical analyses

234 The Philosophy and Methods of Political Science

Box 9.7 Trolleys and bridges Trolley I A trolley (or tram) is running out of control down a track. It will hit and kill five people on the track unless you pull a lever beside you that will send it down a spur. However, there is another person on the spur track and by diverting the trolley you will kill that person. What should you do? Source: initially Foot 1968; Thomson 1985

Trolley II A trolley (or tram) is running out of control down a track. It will hit and kill five workmen on the track unless you pull a lever beside you that will send it down a spur. However, there is a tourist on the spur track and by diverting the trolley you will kill that tourist. What should you do? Trolley III A trolley (or tram) is running out of control down a track. It will hit and kill five workmen on the track unless you pull a lever beside you that will send it down a spur. However, there is a child on the spur track and by diverting the trolley you will kill that child. What should you do? Trolley IV A trolley (or tram) is running out of control down a track. It will hit and kill five tourists on the track unless you pull a lever beside you that will send it down a spur. However, there is a workman on the spur track and by diverting the trolley you will kill that workman. What should you do? Trolley Loop A trolley (or tram) is running out of control down a looping track. It will hit and kill five people on the track unless you pull a lever beside you that will send the trolley down a loop where it will hit and kill a fat man. Without the fat man the trolley would continue along the loop and kill the five from the other direction.* Similarly, without the five there, the fat man would be killed by the trolley from its original direction. The fat man is large enough to stop the trolley on his own, but it will take five thin people to stop it hitting him. What should you do? Bridge A trolley (or tram) is running out of control down a track. It will hit and kill five people on the track unless you push the fat man just in front of you off the bridge to impede the trolley. Only someone that fat could stop it. What should you do? Bridge II A trolley (or tram) is running out of control down a track. It will hit and kill five people on the track unless you pull a lever that will make a fat man drop from the bridge to impede the trolley. What should you do? * Don’t ask why the five will not spot the trolley no matter which direction it comes from.

Analytic Political Philosophy

235

might be very interesting and useful, but they do not constitute what is usually considered to be moral or political philosophy as such. The point of ethical theory is not to describe and explain our moral intuitions but to guide and, perhaps, change them. As described by Rawls, reflective equilibrium is inherently conservative (Singer 1974, 2005). Sometimes we need to pursue out-of-equilibrium reasoning to reach a new moral equilibrium; that is, we need to fight against our intuitions of today in order to create other and (by our argument) superior intuitions for tomorrow. In fact, of course, academics are doing this all the time. Professional moral philosophers are not good guides to the moral intuitions of their culture. Indeed, a large part of philosophical training in ethics classes involves confusing students about their intuitions. Norman Daniels’s (1979, 1996) ‘wide’ interpretation of reflective equilibrium allows us to countenance rejecting all ordinary beliefs to overcome the conservatism inherent in the process. However, if we allow all of our intuitions to be thrown out after reflection of a consistent moral theory, then there is little difference between the method and foundationalism (Singer 2005: 347), which constructs a rational morality from first principles and provided Rawls’s target for the use of the reflective equilibrium strategy in the first place. If we take intuitions to be immediate apprehensions, then intuition pumps ought to be used to test the veracity of those immediate apprehensions, rather than trying to explain why we have the apprehensions in the first place. (That should be the preserve of the systematic empirical research.) Using ‘Trolley’ to explain the doctrine of double effect (that is, it is permissible to cause serious harm as a side-effect of some action, but not to aim to cause serious harm) or to produce complex arguments about what is ‘really driving’ our intuitions (see, for example, Naylor 1988) is the wrong way round. (Aquinas’s original account of why killing someone in self-defence is justifiable is a better illustration of the double-effect doctrine than ‘Trolley’.) Indeed, if one examines the history of the use of intuition pumps, one often finds that the initial use is critical. Foot (1968) used ‘Trolley’ specifically to analyse shortcomings in the doctrine of double effect, whilst Thomson (1976, 1985) and others use the example to make various different points about several issues. If we really want to find out what ‘our’ intuitions are, then we need dedicated surveys with all the usual controls and analysis that survey researchers are good at. But that would only show what our culture tends to think in these examples, and that is not really what ethics or moral philosophy is about. The work of Jonathan Haidt (2001, 2003), for example, has shown in these sorts of stories (and many others) that people make quick automatic judgements which they then, sometimes, have difficulty defending. This suggests that there are two sets of thought processes – and indeed neuroimaging techniques have shown precisely that. Experiments using functional Magnetic Resonance Imaging (fMRI) have shown that our emotional responses to

236 The Philosophy and Methods of Political Science ‘Trolley’ are different from those to ‘Bridge’. It is the personal aspect of pushing someone off a bridge as opposed to sending the trolley down the spur that seems to lead to different intuitions about the cases (Greene et al. 2001, 2004; Greene and Haidt 2002; Greene 2013). Furthermore, Greene suggests that those who would push the stranger would take longer to decide (unless they were psychopaths) than those who said no. Experimentally this proves to be the case. The brain activity of those people who make the decision to push the stranger off the bridge is greater in the parts of the brain associated with ‘cognitive’ rather than ‘emotional’ activity. They are overcoming their first emotional response. These results also suggest why we might be suspicious of or even horrified by those of us who would push the man off the bridge – such actions are suggestive of psychopathology, and of course we do not want to give off that signal. It might be the case that, all things considered, a particular course of action would be the right thing to do in a given intuition pump experiment (IPE). However, carrying it out would also give a signal to others that we are prepared to do such things in those ‘sorts’ of circumstances. That might be a dangerous or unfortunate signal to send. Faced with the ‘Trolley’ or ‘Bridge’ problem in reality, of course, you would have to make a quick decision based on intuition, or what Kahneman (2011) calls fast thinking. Later you would try to justify what you did (or apologize for getting it wrong) using slow thinking. We can regard fast thinking as what we directly apprehend to be the right thing to do in these situations; slow thinking then tries to justify our consequent actions. Sometimes you might realize that you did the wrong thing. If, as seems to be the case, different parts of the brain are implicated in fast and slow thinking, there might be no ‘correct’ answer unless you privilege one part of the brain (one way of making decisions) over the other. But then fast thinking has developed for a reason – and so has slow thinking. Maybe we can train our intuitions, train our reactions, so that fast thinking does what, all things considered somehow, we want. Indeed, training for soldiers, pilots and the like seems to deliver something of that which is relevant to their roles. If that is the case, then perhaps ethical training (slow thinking) can help guide our intuitions when fast thinking is needed. Of course, though, that process privileges slow thinking. Indeed, discussion of IPEs in the seminar room is rife with this problem. Asked about a given IPE, a member of the audience gives an answer – fast thinking – that is then challenged by someone else, and must be responded to by slow thinking. What has happened to the intuition? What we learn from Greene and Haidt is that moral judgements in many IPEs are typically the outcomes of fast, almost automatic responses and the deliberation or conscious justification tends to come later. It tries to rationalize the fast response. This is relevant to the ‘Trolley’ and ‘Surgeon’ examples, since pulling the lever in the ‘Trolley’ example would have to be a fast response (level 1 thinking);

Analytic Political Philosophy

237

the surgeon could not operate that quickly, so his decisions would have to rely on level 2 thinking. The fact that the two IPEs would demand different levels of thinking is not irrelevant to our intuition (level 1 fast response) regarding them or our considered judgement (level 2 slow thinking). The outlier problem The way in which intuition pumps are sometimes used in the form of a datum point to test foundational theories differs from how empirical political scientists tend to use data. Intuition pumps are often unusual or strange cases, ‘outliers’ to the norm. In empirical political science, an outlier to some general finding (a case off the regression line, say: see Figure 9.2) is not usually thought to falsify the general finding of the regression line. It is not thought, at least straightforwardly, to disprove the general finding. Generally speaking, outliers are examined to find why they do not seem to conform to the other cases (Bäck and Dumont 2009). It might be that the mechanism that works in most cases does not work for the outlier because of some other feature. That is, at least at first, we attempt to explain how the outlier fits with the broad mechanism that is supposed to explain general findings. Outliers only become falsifiers or disconfirmers of general theories (mechanisms) when their underlying features fit the mechanism perfectly and thus other differences between the outlier and the general trend explain the disparities. In moral philosophy, examples like ‘Surgeon’ tend to be used to disconfirm the theory. So ‘Surgeon’ is used (especially in the classroom) as a knockdown argument against utilitarianism. Can it fulfil that role? Perhaps, but can it show that utilitarianism is false, or just that other considerations need to

Figure 9.2

‘Outliers’

PR

PF

238 The Philosophy and Methods of Political Science be taken into account? We can certainly think that ‘Surgeon’ and ‘Shoot the One?’ demonstrate that maximizing welfare is not the only consideration. By showing that simplistic maximizing does not fit with our moral intuitions, they prove that simplistic utilitarianism cannot be the basis of all morality. But on the outlier argument, they do not show that we should not make utilitarian calculations when taking moral decisions, just that we need to be aware that there might be other considerations. For example, we might say that ‘Surgeon’ demonstrates that government needs to be aware of certain basic rights (respect for bodily integrity) when it produces policy that tries to maximize welfare. Can it tell us anything more than that government should respect bodily integrity? I am not sure it can. In order to demonstrate that government needs to respect other rights, one needs other types of examples. Whilst intuition pumps can suggest that grand (foundational) moral theories have limitations, they cannot, on their own, show that the grand theories are not generally applicable; nor can they show that the grand theories are not generally applicable beyond the specifics of the case in the intuition pump. In other words, if intuitions are to be used as data to test theories, they need to be treated in the same way as data are treated in empirical political science. And that is a lot more carefully. Any given IPE might simply be an outlier to a straightforward principle of human (or state) behaviour that teaches us to look for similar cases that share the specific feature of the outlier. The outlier problem might be particularly acute in political philosophy since here we really need to demonstrate the importance of counterexamples. For example, it is no critique of comparative measures of human well-being across two societies based on mortality or morbidity rates, even if one can convince that keeping patients alive in a vegetative state is against human well-being. One also needs to show that this fact bears on the specific comparison between the countries (Dowding 2009). Whatever we learn from ‘Trolley’ or any other example has to be shown to be relevant to any comparative questions of social justice, not simply assumed. How should moral philosophers use intuition pumps? IPEs are good at pumping intuitions and perhaps at little else. And given that we need to be suspicious of intuitions pumped in this manner, if they are then used to test ‘grand’ moral theory (foundational theories) as in the method of reflective equilibrium, how should we use them? Perhaps the superior method for the moral and political philosopher is to use a version of my inversion strategy. Faced with a specific intuition pump, authors tend to respond with a different example to see what new intuitions are pumped by it. So first ‘Trolley’ is trotted out, then ‘Trolley Loop’ or ‘Bridge’ or ‘Surgeon’, and we are asked to compare them to the original. However, each new

Analytic Political Philosophy

239

example brings its own baggage whose psychological effects need to be factored in; and some might rely on slow thinking, others on fast, making intuition comparison even more problematic. A better way is to ask what has to change in the example to bring about the opposite intuition. So we invert ‘Surgeon’ to ethically examine our intuitions. Ask yourself: ‘Under what conditions would it be right for the surgeon to operate to kill one to save five?’ Or if you find that too difficult to swallow, ask yourself what mitigating circumstances you would take into account to reduce the sentence of a surgeon found guilty of murder to save five lives, if you were the judge at his trial. In that way you interrogate your intuitions rather than letting them rule the roost. Dennett (2013) offers similar advice, suggesting that one should take every line in a story and see what difference changing it would make to one’s intuitions. I have done that for ‘Trolley’ (Box 9.7 above). Without trying to lead your intuitions, what difference does it make if the victims are not ‘people’ but workmen or tourists? In Trolley II you have five workmen who ought to be aware of the trolley schedule and be looking out, but the tourist is more innocent (and not on the dangerous track): does that make a difference? Or what about Trolley IV, where the initial victims are tourists (what they hell are they doing on the track!) and on the spur is a workman who perhaps knows he is on a ‘safe’ track, so is not watching so carefully. One can also modify each of these examples by changing the numbers. In Trolley IV, for example, turn the five tourists into three and the one workman into two: does that make any difference? Or think about the relative ages and social roles of the people on the track. The point of this strategy is not to let the intuitions lead the way, but to make you think about the moral issues in a focused manner. When applied to government activity or to the sorts of principles we want in a constitution, these are the aspects that matter. And remember the type/token distinction. Intuition pumps involve token people – ‘the fat man’ – but political philosophy is really about types of people – ‘the obese people’, ‘men’, ‘women’, and so on. What difference does that make? I began by contrasting thought experiments and intuition pumps, suggesting the first work by challenging our intuitions and the second by utilizing those immediate apprehensions. To the extent that intuitions and reasoned consideration proceed from different ways of thinking, indeed utilize different parts of the brain, there might be no definitive answer as to which is correct. (Unlike our immediate apprehension to a maths puzzle where the maths supplies the correct answer.) If we privilege the moral intuition, then we are using it as data, but there are problems in doing that. If we are going to use intuitions as data, then we ought to do so systematically – but that is not what moral and political philosophers do. So we do need to interrogate IPEs, and may thus see that they are only heuristic and not decisive.

240 The Philosophy and Methods of Political Science When discussing normative conceptual analysis, I suggested that normative concepts cannot be completely unambiguously and coherently defined, since we might not have coherent intuitions about them. The method of reflective equilibrium cannot be analogous to observational data-testing theory, since moral theory is supposed to directly affect the data that are being used to test it. Moral theories are supposed to change our beliefs, our intuitions, and not simply to predict them. If we make the subject of ethics too precise, we will lose it from our discourse. At the end of the day, reflective equilibrium might do no more than scrutinize our intuitions to see how ambiguous and at times contradictory they are. Since I take it we do not, psychologically, like contradictions, then these problems might lead us (or perhaps more likely our children or grandchildren) to develop different, less contradictory, intuitions; but we should not expect complete ethical closure for any of our descendants. Reflective equilibrium might simply be a description of what we do when we morally theorize, and perhaps it cannot be rationally justified, much as we cannot rationally justify induction even though it is part and parcel of our (scientific) reasoning process. Perhaps we need to remember that, just as induction is not deduction, ethics is not science. That does not mean that there are not better and worse ways of thinking about ethics or doing political philosophy, nor that there are not more or less useful ways of using intuition pumps.

9.5 Conclusion The method of reflective equilibrium is supposed to reflect the theory-testing methods of science. There are problems with understanding the method that way. However, its appeal is understandable, if only because it is hard to see how else one could theorize without taking into account contemporary moral and cultural understandings. Skinner’s approach to the history of ideas also appeals to the fact that when judging what people are saying – their theories and the concepts and ideas that compose them – we need to understand the moral and cultural contexts of their day. Classic texts often become classic texts because they change the way people think. Those changes in thought might be ‘ideologically biased’ in the sense that the authors wish to defend some special interests in society – those of landowners or the proletariat, for example – and understanding those ideological underpinnings might give us pause when considering the rational basis and philosophical worth of those writings. These values are better considered in their own terms, however. Whatever dastardly motivations someone has for a viewpoint, the arguments for that viewpoint are what count in judging its justification.

Analytic Political Philosophy

241

Concepts are part and parcel of theories and as arguments develop, concepts change their meaning. As we learn empirical facts about society the extension of our concepts changes. As our moral understandings alter and develop, the extension of concepts can change, as indeed can their moral worth. Conducting conceptual analysis in as non-normative a fashion as possible helps us to see more clearly what role values are playing in theories, and makes discourse across different worldviews more compatible. It reduces the sleight of hand that is the trademark of the ideologue and ensures a more analytic approach. Intuition pumps make our theoretical ideas stand up against our moral intuitions, but they should not be considered simple tests. Moral and political argument is supposed to challenge and change our values, not simply reflect them. The inversion strategy is designed to shape those intuitions from both sides, so to speak, and careful reanalysis of each element of an intuition pump can force us to see more clearly where our intuitions come from. Once we know that, we can more easily decide whether to follow or disown intuition. At the beginning of this chapter I made a distinction between moral and political philosophy. The chapter has been more about moral theorizing than political theorizing. That is because it is moral theorizing that is more distanced from the methods of political science. What is distinctive about normative political philosophy, as opposed to political theory or theoretical politics, is the moral element. Political philosophy needs to take into account the lessons of political science. Some of those lessons seem to hold true generally about human societies; others are more culturally specific. Political philosophers need to be aware of the lessons that seem more culturally determined, for those are the ones that might be more effectively challenged or need more careful defence. Just as political scientists need to think carefully about the methods they adopt, so too do political philosophers. Cohen (2008: ch. 6) argues that there is a clear distinction between facts and normative principles. A normative principle is a general directive that tells agents what they ought to do. A fact is any truth other than a principle, of a sort that might reasonably be thought to support a principle. Cohen argues that whilst some moral principles might rely upon the facts of the matter, any such principles must rest upon other principles that do not rely upon any facts. He gives the example of ‘we should keep our promises’ because ‘only when promises are kept can promisees successfully pursue their projects’. The first looks like a principle based on a factual claim in the justification. However, the second seems to rely on another principle namely ‘that promisees should be able to successfully pursue their projects’. Whenever we chase principles, there will always be one principle that is ungrounded in facts. So for every fact-sensitive principle there is a fact-insensitive principle. Underlying this argument is the claim that there is always an explanation of why any ground that bases a claim does so. In other words, there

242 The Philosophy and Methods of Political Science is always a justification for such claims. That is surely so, at least if we can engage in rational justification of our moral claims at all. If one believes that such grounds for principles are generated simply by pure reasoning, then the fact of them relies upon nothing empirical. If, however, one believes that any reasoning for moral principles is based upon some intuitions, then one sees an empirical basis for any principle. If intuitions can vary given the circumstances of the people who generate them, then we can see how principles can vary. Furthermore, our intuitions often carry feasibility constraints: what we can reasonably expect people to be able to do. When considering the narratives of IPEs, we do not imagine going outside of what we can achieve. One normative principle might be: ‘One can’t expect people to do what they are not able to bring themselves to do.’ We might hold a principle of human equality, but how we think of that equality – what we consider equal treatment within any set of specifications – will be bound by what is intuitively feasible. However, such a principle is a rather strong version of ‘ought implies can’, since one might believe that someone ought to have been able to bring themselves to do something that in fact they could not. The strong version might countenance letting people off the moral hook. Again, morality is supposed to lead people, not simply follow them. A second, hard issue of the ideal/non-ideal debate is whether principles do rely upon empirical matters. If one utilizes intuition pumps, then one is bound by some feasibility constraints and relying upon some non-ideal data, namely our direct apprehensions about some issues. Unless we follow the scholastics in believing such intuitions are the perceptions of angels, or pure knowledge, then we must accept that something empirical enters in to moral and political theory. Ideal theorists in that pure sense cannot use reflective equilibrium; their morality must be generated from first principles. How far one wants to build in other feasibility constraints (of the first and third variety – see p. 241 above) will then determine how far one wants one’s political theory to affect the world. To worry too much about what is feasible will not move the world very far; to worry too little is not to move it at all or, worse, to move it in the wrong direction. There are various methods of political philosophy. Perhaps we need to reflect upon them more often. Philosophers go about their task using reason, logic and argument; they look for contradictions and incoherencies; and they attack or defend assumptions. Whence their underlying views, their assumptions derive and how these are to be treated, given that they are both descriptive data and the object of prescription, is the problematic of the subject for which there is no clear-cut response.

Chapter 10

Political Science as a Vocation

I drafted this chapter before coming across Anne Norton’s (2004) and Robert Keohane’s (2009) similarly named essays, also taking their titles from Weber’s (1918, 1919) two famous lectures. As should be apparent from the chapter, Weber’s essays provided the inspiration, but Norton and especially Keohane also influenced the final version.

10.1 Political science What do we understand by the term ‘political science’? As in the nature of the analysis of this book, I do not think there is much worth in defining it at length. It depends on what you are interested in examining. Politics can be understood narrowly or broadly – as concerned only with matters of the state or polity or as expressed in virtually any set of human interactions. Wherever there is some kind of power relationship, we can see any attempt to bargain or barter, to shift opinion through argument or persuasion, to change the interests of partners or opponents through altering their situation (the institutions and structure around them) as part and parcel of politics. Politics has been defined as the authoritative allocation of values (Easton 1965); it has also been defined in terms of peaceful ways of overcoming conflicts of interests (Blattberg 2009: ch. 1; Keohane 2009); in terms of power, or ‘who gets what, when and how’ (Lasswell 1950). These all seem to me reasonable shots at defining politics, even though they are quite different. I have no real concern if we want to go with a narrow definition concerning only political communities, leaving the rest to sociology or economics. Political science can be what is studied in political science or political studies or government – and international relations departments, though that will barely narrow the field, as such departments include academics doing what some will claim is history, sociology or economics or even psychology, social psychology or genetics. (Or, for that matter, philosophy, literature, statistics, econometrics or mathematics.) However, I do want to distinguish political science from politics itself, from political journalism, and from public policy, just as we can distinguish belief from opinion, science from action. I want to define political science as science, broadly understood, following the sorts of methods recommended in this book.

243

244 The Philosophy and Methods of Political Science

10.2 Belief and opinion Dennett (1978), following on from ideas from Annette Baier and Ronald de Sousa, makes an important distinction between ‘belief’ and ‘opinion’. In many accounts belief operates in a Bayesian-like fashion that admits of degrees, so that one might believe 0.8 that p whilst believing 0.2 that not-p. People’s beliefs, including their degree, can be judged by their behaviour using revealed preference methods (Dowding 2002; Ross 2014). There are various difficulties in revealed preference methods (Sen 1982a, 1982b, 1982c, 2002a, 2002b, 2002c), but even critics would agree that we can gain some idea of beliefs through behaviour. Humans are self-reflective and can assent to their beliefs verbally; the state of assenting to a belief is what Dennett calls an opinion. Dennett then suggests that coming to have an opinion is to make up one’s mind about something; changing one’s opinion is to change one’s mind. Dennett suggests there are many ways of collecting opinions – one can inherit them, fall into possession of them without noticing, fail to discard them despite deciding to do so, borrow them and then forget they were merely borrowed. To these we might add misremembering a story as the truth and creating an opinion therefrom. Habits can create opinions; one might suspend disbelief and then forget this and turn an assumption for the sake of an argument into an opinion. One’s reason for believing something might be long gone, but the opinion remains. We change our mind the same way we change someone else’s – by the colloquy or soliloquy of persuasion. Sometimes we cannot withstand an argument, but still find it difficult to believe. It might be our opinion is too ingrained, though I find I can come to believe an argument over time as I get used to it. Opinions are a bit like old jumpers: you don’t want to discard them even when they get a bit tatty, until you get used to the new jumper (and even then the old jumper still holds some appeal). Though a really old jumper that you find at the bottom of the closet many years after discarding it can seem quite dreadful if not positively disgusting. (‘I used to wear that, how embarrassing!’) Sometimes an old opinion, perhaps discovered written in a diary or a letter, or an old academic article, especially (or perhaps exclusively) if you had forgotten you had ever held it, can seem equally disgusting and embarrassing. Some academics really don’t care about the truth. They espouse opinions in their writing that their actions clearly reveal them not to believe. They can get away with that because talk is cheap. You can endorse opinions easily if you do not let them infect your other behaviour. Moreover, the outcome these academics care about is not enhancing human knowledge, but rather influencing others and the rewards that such influence brings (status, income, prestige) relative to the costs. Peddling theory with few empirical consequences is cheap, but the academic rewards, if you write well enough in

Political Science as a Vocation

245

obscurity with tidbits of insights, can be enormous. That is one sort of vocation. It is not one I admire. Political scientists ought to care about the truth and ought to care about trying to demonstrate that truth. To be sure, there are some truths that people might not want to find out: their partner’s infidelity perhaps, or what their children were up to last night. We might not like to have deeply held opinions challenged, especially our moral or political opinions. It is not difficult to allow these dispositions to infect our research strategies and interpretations of results, but we should at least try to overcome those prejudices. It is also easy to shield them from facts and results. In fact, inconvenient truths are often not as opinion-destroying as they might first appear (we simply need to shift the grounds of why we think such and such is the morally or politically right thing to do); and when they do change our beliefs and opinions, we can grow equally attached to the new ones. My claim here, for which I have no scientific support, is based on the assumption that, generally speaking, people want to believe what is true and avoid believing what is false (Boudon 1981). We care about the truth. However, not all truths are equal, in the sense that some are more valuable than others. Some questions are intrinsically more interesting. I am tempted to claim that the truths that are the most valuable are the ones we need to use most in prediction, but many of our beliefs which we never think about are never thought about because we use them all the time in living our lives. As such, they are some of the most predictive beliefs we have, but they are not the ones we usually think are the most interesting or valuable. I do not know how to go about providing an account of what questions are the most interesting. The only such account I know is Carballo (2012), who suggests that the value of questions can be judged in terms of the value of the beliefs that answers to such questions are likely to yield, and couches that thought in terms of explanatory closure. One way of thinking about this evaluation is that the type of answers that lead to cumulative research are better than ones that will not. How one phrases questions is important to how we go about answering them.

10.3 Surface and structure I have argued that there is nothing wrong with detailed historical description and we should contest the modern view that seems to suggest that only work involving some explicit theory, complete with hypotheses to be tested or theorized explanation concerning causes, mechanisms or functional relationships or within broad theorized interpretive schemes, constitutes good political science. It is true that detailed description might only be the first stage in some explanation of political life; it might only constitute the data

246 The Philosophy and Methods of Political Science upon which further work can be built. But collecting high-quality data is an exercise in itself, and collecting data from primary-source materials, from interviews, from historical records bolstered with a thorough understanding of the accounts of others of the case at hand, in order to produce a rich and detailed description is worth the applause of the academy, and the award of doctoral honours. Others might erect more from such descriptions, but all science has to be constructed from this bedrock. The history of biology can provide a lesson. For several hundred years much of what we now call biology was simply describing the natural world, both the familiar from biologists’ own countries and the unfamiliar observed in forays to far-off lands. What passed as theory was often speculation that now has little credence. It was on the back of detailed description, on the production of classifications and typologies, that modern biology was built. Political science has gone beyond mere description, but there is so much that has not yet been described, and the very nature of human society with its constant turmoil and development, even perhaps evolution (though we must be careful using that term), deserves constant detailed description. And some of what goes on in the name of political science is no more than speculation that one day will be given little credence. Description is not opinion. To be sure, description must involve interpretation and interpretation is affected by our views, our perspective and our interests, but interpretation and description can still be scientific. Scientific description must still be based upon evidence, be systematic and be replicable as far as possible. The references to original sources, transcripts of interviews and all other evidential bases need to be collected and be available for inspection. But at the end of the day, good detailed description, wrapped around a narrative (for otherwise the description would be uninteresting to the reader), makes for good data. The narrative might be considered to be a hypothesis, as I describe in Chapter 6 (Section 6.4), but the data themselves cannot do more than provide a framework that makes the hypothesis plausible; they cannot themselves corroborate or falsify it. We should not denigrate such narrative, however, for it is the bedrock of all else that exists in political science. Political journalism does not provide such narrative. The accounts of journalists might form some of the original sources for political science, but journalists cannot be expected to make their sources replicable nor provide the detailed references and careful evidence of the political scientist. Journalists, no matter how careful they are, need copy and they often need it fast. They also need a story that will capture the imagination of their readers. Often they are, explicitly or implicitly, constrained by the ideology or the line that their news organization takes; their words, too, are constrained by the needs of the day, and may be changed without consultation by subeditors. In these ways their copy represents opinion rather than belief. It evidences certainty and clarity, not probability and nuance. I am not referring only to opinion pieces.

Political Science as a Vocation

247

Even the most descriptive reportage of political events has a narrative – spin, in modern parlance – that fits the line the journalist or paper is taking, often from cues signalled by other media. For journalists, the scandal is always a danger for the government; the administrative error is always a headache for the minister; there is always the search to pin blame. Journalists cannot afford nuance and doubt. They must report in the black and white associated with traditional newsprint. Political scientists are not immune to opinion. They will give their opinion, and on occasions should give their opinion. However, in their work they need to operate with belief backed by evidence. Even in descriptive narrative, they must not ignore evidence that tends to suggest other possible narratives. At times that might mean they have to trade style for accuracy, but that is their job as scientists. They must give credit to others whose narratives differ. Still more importantly, political scientists must try to delve beneath the surface to mechanisms and underlying structures that shape the course of events. The description thus becomes theorized, with theories that make claims about mechanisms and generalizations and hence provide predictions for another day. Pure theory will be involved in such shaping and can sometimes show that previous narratives must surely be wrong. In Chapter 4, I gave the example of McCloskey (1970), and several times in the book have alluded to vetoplayer theory, which suggests an underlying reality that explains more about policy stability across different political systems than the generalities offered in surface perceptions and models in terms of presidential, semi-presidential and parliamentary descriptions. Unlike political journalists, political scientists need to take a step back. Where journalists want to uncover crimes and misdemeanours in government circles, to pin blame on ministers or departments, political scientists can take a more aggregated and deeper perspective. We can understand that in any human process there will be a distribution of error. People make mistakes. The error today was made in the home affairs and not defence portfolio; that does not mean that home affairs is inefficient and defence is not. To be sure, we might be able to point to the human(s) or process(es) that created the problem. But it does not follow that these people are more error-prone than others, or that this process is less efficient than that in another ministry. They might be, but a single error does not demonstrate it; rather, it is simply a piece of evidence to be considered with others. The political scientist, even when providing proximate and token explanations, needs to delve deeper than do journalists. To that end political science needs to be comparative. Not in the overt sense that one must not study the politics of a particular nation, or state, or local government; but the political scientist needs to ask comparative, counterfactual questions, and these can be best answered with some knowledge of other countries. Country specialists need to ask questions of specific countries or states that are international ones – ones that could be asked

248 The Philosophy and Methods of Political Science elsewhere – and not just ones that are unique to their own country. Political science needs a process of discovery that delves beneath the surface of political phenomena to the structures and mechanisms that shape them. The same global forces might operate across different countries, but they play out with very divergent consequences in different institutional settings. The primary point of comparison is usually those institutional differences, as these lead to similar processes having diverse and divergent outcomes. Whilst economic growth rates might rise and fall broadly in unison across the world, there will still be individual differences or discontinuities. Growth rates might converge because of the structural convergence of separate national economies, but demonstrating such structural convergence requires separate (institutional) evidence and cannot be shown simply by evidence of growth rate convergence. Here both large-n quantitative evidence and evidence from process-tracing techniques need to be used in conjunction. Journalists and country specialists are sometimes too quick to point to specific features of their countries to explain current political events or processes, when the same sorts of events and processes are occurring simultaneously in many countries. Conversely, certain features of political landscapes, such as party systems, might be converging, yet the forces contained there are very different in different countries. For example, a broad thesis taken up by some parliamentarians and many journalists and political scientists suggests that prime ministers in parliamentary systems are becoming more presidential (Foley 1993, 2000 and 2004; Fabrini 1994; Allen 2001; Helms 2005; Poguntke and Webb 2005; Bäck et al. 2009; Webb et al. 2011; Kefford 2013). I think this thesis provides a good example of putting too much emphasis on superficial or surface reality and not enough on analysis of the mechanisms that operate in different institutional settings (Hart 1991, 1992; Norton 2003; Jones 2006). To be sure the media might focus on prime ministers more than in the past; prime ministers might take on the air of presidents. They might also, like presidents, have increased in importance relative to their cabinet ministers. But parliamentary systems are deeply different from presidential ones. Legislation in prime ministerial systems emanates from the executive; in presidential systems from the legislature. Oversight procedures are different; governmental coalitions are often created and run very differently; and, importantly, elections where candidates for parliament run alongside the party leaders generate a very different dynamic from systems where the legislative and presidential elections are separate in form and timing. These are the institutional differences that mean that environmental changes impact differentially on presidential and prime ministerial systems (Dowding 2013b, 2013c). Institutional differences create behavioural differences, and we need to be aware of how they both lead actors within systems to interpret what they are doing and change their actions accordingly.

Political Science as a Vocation

249

This is why I do not think that the institutional behavioural and interpretive methods of Chapter 7 belong to separate accounts of social reality. Nor do they rely upon different epistemologies and ontologies, as some claim. It is true that people might weight evidence differently and be interested in different sorts of evidence, but that does not show they hold different theories about the nature of belief acquisition. Humans cannot help but take account of the constraints that lead them to see patterns in the universe. However, those who see ‘ontological commitments’ as important to social science are correct in one respect. One of the ways, perhaps the chief way, in which social science exists separately from natural science is that it studies many things that exist only because we think they exist. In the natural sciences what we, as scientists and as humans, see as patterns in the data are predictively constrained by what exists; what exists in that sense is not constrained by us. Those patterns cannot be changed by what we would like to find. Conversely, what we, as social scientists, see as patterns in the data are predictively constrained by what exists, but here to some extent what exists is what we as human beings think exists. Some of those patterns exist because we think they exist. That does not mean the patterns we see and respect do not constrain our actions, but they can much more easily be changed (Wendt 1999). Those patterns can be changed by what we, as humans, would like to find – though what we can change in those patterns is constrained by the world and our expectations about the world. Social science is largely about relationships between people, and those relationships are formed by mutual expectations. Conventions, moral rules, behavioural and verbal signals exist in the forms they do because we think they take those forms. If we were all to believe they take another form, then they would take another form. If you offer your right hand as a polite greeting, it is because you think that extending your right hand to shake another’s right hand is a polite greeting. If the person you greet also believes that to be the case, then it is indeed the case. If a society holds that private property rights in some objects mean that only the property holder can use that object without permission, the property holder’s right in that object (materially) exists to the extent that others respect that right (Dowding and van Hees 2003). Shepsle and Bonchek (1997) in their introductory book on rational choice have a section entitled ‘It’s Not Rocket Science But …’. They are right. Social science is not rocket science: it is much harder than that. We can predict more far more easily in the natural sciences than in the social sciences, but that is not to say we cannot predict at all in the social sciences. Indeed, if we could not reasonably predict human behaviour most of the time, we could not live together. One of the problems in political science is that politics is a strategic game, where the winners often emerge because their opponents failed to predict their behaviour; in non-cooperative situations, such as electoral contests or war, the name of the game is surprise. Nevertheless, whilst token contests

250 The Philosophy and Methods of Political Science are difficult to predict, we can predict the emergence of types of strategies within certain institutional and structural contexts. Political science should concentrate on those sorts of predictions, for it is unlikely to do much better than acute political commentators when it comes to the actual token contests. (We can take the analogy with rocket science a bit further. Rocket science uses Newtonian equations, since they are good enough to get rockets roughly where we want them in the solar system; once there we can nudge them as necessary. In other words, rocket science uses principles known to be false, but good enough for purpose. We might think we can use models in political science that are known to be false in detail but good enough for purpose, and then we can nudge a bit to get everything just right. However, one of the problems with this analogy is that it is much harder to discover how close political science models are to reality than it is for the spatial location of rockets. There is much debate about just how wrong our predictions really are.) Conventions, moral rules, institutions, language, and so on, exist as the conventions, moral rules, institutions and the language they are to the extent that we each recognize them. Thus social science studies objects that exist only because people think they exist. That does not make their existence any less real. It does not make examining and measuring these objects any less real than examining and measuring the objects of the physical sciences. Nevertheless, it does give social science a rather different character. It must not be thought that, merely because the main object of study of social science are those things that exist only because we think they exist, natural science is constrained by the ‘external world’ but social science is not. Conventions, moral rules, institutions, and so on, are ‘external’ to each person just as trees and rocks are. And just as we interpret data about conventions and institutions, so we interpret data about trees and rocks. We might live by conventions and institutions and we might work on them to change them, but we can also work on and change some aspects of trees and rocks. We might find that, just as we cannot change some aspects of trees and rocks, we cannot change some aspects of conventions and moral behaviour. Political science is devoted to examining these deep underlying structures. Political philosophy has the aim of deciding which we want to change and how we might go about changing them. The two are, to some extent, mutually supporting, but they should not be confused.

10.4 The profession Pursuing the profession of political science is not just doing research. Teaching is an integral part of it. Most of the best researchers are also teachers, and often the best teachers. Inspiration for research often comes from one’s students, who ask the most fundamental questions. Teaching in

Political Science as a Vocation

251

the classroom is stressful and leads to heightened awareness. Occasionally I have been making a point in a lecture I have given several times when I suddenly think, ‘this cannot be possibly be right’. The heightened awareness stimulates the thought processes and engages the intellect. The experience of teaching also proves that if one cannot explain something, then one does not fully understand it. Sometimes in the classroom my inability to get a point across, or a question from a student, leads me to realize that I have not truly grasped something I thought was relatively straightforward. (The shift to online or recorded lectures is a major threat to university education. One of the points of lecturing live is for the lecturer to be able to judge how well their audience is following the explanation. It also enables the interaction that is lacking with recordings. Seminars perform a different function and cannot fully compensate for this loss of face-toface lecturing.) Lessons about one’s subject of research can come from anywhere. The most important element of originality is taking an idea from one context and walking it over to a new one. I am bemused by colleagues who do not attend a particular research presentation because they say it is not in their subject area. If they are not interested in the study of politics, why study it? It is also my opinion that anyone who specializes too long in the same area of political science is someone who has nothing more to say about it: saying someone is ‘an authority on a subject’ is really to say ‘they have nothing new to say on that subject’. I get many ideas from people who work in areas far from my own. Moreover, good ideas can be stimulated by bad presentations. (I sometimes think: ‘I would not try to answer that question using those methods – how would I go about it?’) The point of teaching political science is to make students more expert in it. Just telling them some facts, narrating stories or even getting them to recount complex theories will not make them more expert. In disciplines where expertise exists, those who have been trained should come to converge on similar analyses of the same situation, or come to similar solutions to problems, or at the very least they should come to recognize the outline of the situation in the same way. Some disciplines seem not to involve such convergence. Those at the beginning and end of their training in psychoanalysis, for example, do not converge in opinion, suggesting that opinion and not expertise is all that exists there (Kahnemann 2011: ch. 21). Studies of expertise suggest that novices and experts categorize problems differently, the former concentrating upon surface features (such as keywords in problem statements), the latter upon deeper structures (Chi et al. 1981). (The converse can happen if only the structure is taught. Students taught only the mathematics of game theory sometimes struggle to apply the techniques they have learned to situations casually described but essentially having the underlying structure of games they have been taught.)

252 The Philosophy and Methods of Political Science On the other side of the coin is learning from others. To publish topquality research means publishing in top-quality peer-reviewed journals. To some extent publishing means following a series of norms that includes developing a particular style of writing. Different journals and different areas of the discipline have different styles. These are not always fertile conventions and many areas use technical jargon. One may have to follow the style, conventions and jargon of the discipline areas, but one should also be self-critical of them. To satisfy reviewers one must sometimes do things one considers unnecessary: perhaps provide a literature review when these are ten-a-penny (‘author needs to put the paper into context’); refer to the reviewer’s work (‘I was surprised the author had not given enough account of previous findings’); or defend assumptions that have been defended elsewhere (Frey 2003). Often, in fact, reviewers provide sage and important advice. Even if one finds precisely the same results when running a different statistical model specification, using slightly different data, adding further controls, or developing and defending one’s interpretations of the findings, the research is likely to be enhanced. All academics have some bête noire over refereeing practices, and these should be remembered when reviewing oneself (Schneiderhan 2014). In the end, however, the research findings are greatly enhanced by the review process, and if one of the costs of that process is learning how to satisfy reviewers with a few tricks of the trade, then that is the nature of the profession. Early in your career, you might have to follow the current dictates to get published and become established; later you may be better placed to publicly challenge them. The aim of such methodological challenges, however, should be same as the research directive itself: to learn more about the social and political world. Of course, all political scientists not only want to discover new and important truths; they also want to convince others that the truths they have found are indeed the important ones. This can be a frustrating exercise, especially for young academics who often find their older colleagues unwilling to take on new findings. One needs to remember Max Planck’s (1949: 32–3) caution: ‘A new scientific truth does not triumph by convincing its opponents and making them see the light, but rather because its opponents eventually die, and a new generation grows up that is familiar with it’; or, as Brian Barry more pithily (and combatively) put it to me, ‘if the graduate students quote you, you win. If they quote your opponents, your opponents win’. That is one reason why writing well can be as important as writing truly. And it is also why academics are fixated on their citation rates. Still, when all is said and done, citation rates are as good a guide to what we think important, interesting and true in political science as anything else.

References

Abadie, Alberto, Diamond, Alexis, and Hainmuller, Jens (2010) ‘Synthetic Control Methods for Comparative Case Studies: Estimating the Effects of California’s Tobacco Control Program’, Journal of the American Statistical Association, 105 (490), 493–505. Abadie, Alberto, Diamond, Alexis, and Hainmuller, Jens (forthcoming) ‘Comparative Politics and the Synthetic Control Method’, American Journal of Political Science, DOI: 10.1111/ajps.12116. Acemoglu, Daron and Robinson, James A. (2012) Why Nations Fail: The Origins of Power, Prosperity and Poverty (London: Profile Books). Achinstein, Peter (1981) ‘Can There Be a Model of Explanation?’, Theory and Decision, 13, 201–27. Achinstein, Peter (1983) The Nature of Explanation (Oxford: Oxford University Press). Adams, James F., Merrill III, Samuel, and Grofman, Bernard (2005) A Unified Theory of Party Competition (Cambridge: Cambridge University Press). Aldrich, John H., Montgomery, Jacob M., and Wood, Wendy (2011) ‘Turnout as a Habit’, Political Behavior, 33 (4), 535–63. Alford, John R. and Hibbing, John R. (2004) ‘The Origin of Politics: An Evolutionary Theory of Political Behavior’, Perspectives on Politics, 2 (4), 707–24. Allen, Graham (2001) The Last Prime Minister (London: Graham Allen). Almond, Gabriel A. (1988) ‘Separate Tables: Schools and Sects in Political Science’, PS: Political Science and Politics, 21 (4), 828–42. Amsel, E., Langer, R., and Loutzenhiser, M. (1991) ‘Do Lawyers Reason Differently from Psychologists? A Comparative Design for Studying Expertise’, in R. J. Sterberg and P. A. Frensch (eds), Problem Solving: Principles and Mechanisms (Hilldate, NJ: Lawrence Erlbaum Associates). Ansolabehere, Stephen and Hersch, Eitan (2012) ‘Validation: What Big Data Reveal about Survey Misreporting and the Real Electorate’, Political Analysis, 20 (4), 437–59. Ansolabehere, Stephen, Hersch, Eitan, and Shepsle, Kenneth A. (2012) ‘Movers, Stayers, and Registration: Why Age Is Correlated with Registration in the US’, Quarterly Journal of Political Science, 7 (4), 333–63. Arena, Phil (2011) ‘What Does That Mean?’ Available at http://fparena.blogspot .com.au/2011/01/what-does-that-mean_30.html. Accessed 28 May 2014. Ariew, André (2003) ‘Ernst Mayr’s “Ultimate/Proximate” Distinction Reconsidered and Reconstructed’, Biology and Philosophy, 18, 553–65. Aristotle (1976) Ethics [Nichomachean Ethics], trans. J. A. K. Thomson and Hugh Tredennick (Harmondsworth: Penguin). Arrow, Kenneth J. (1951/1963) Social Choice and Individual Values (1st/2nd edn; New Haven: Yale University Press).

253

254

References

Austen-Smith, David and Banks, Jeffrey S. (1998) ‘Social Choice Theory, Game Theory, and Positive Political Theory’, Annual Reviews of Political Science, 1, 259–87. Austen-Smith, David and Banks, Jeffrey S. (1999) Positive Political Theory I (Ann Arbor: University of Michigan Press). Austin, John L. (1962) How to Do Things with Words: The William James Lectures Delivered at Harvard University in 1955, ed. J. O. Urmson (Oxford: Clarendon Press). Bäck, Hanna and Dumont, Patrick (2009) ‘Combining Large-n and Small-n Strategies: The Way Forward in Coalition Research’, West European Politics, 30 (3), 467–501. Bäck, Hanna, Dumont, Patrick, Meier, Henk Erik, Persson, Thomas, Vernby, Kare (2009) ‘Does European Integration Lead to a “Presidentialism” of Executive Politics?’, European Union Politics, 10 (2), 226–52. Bailey, Kenneth D. (1994) Typologies and Taxonomies: An Introduction to Classification Techniques (Thousand Oaks, CA: Sage). Barbour, Rosaline (2008) Doing Focus Groups (London: Sage). Barker, Chris and Martin, Brian (2011) ‘Participation: The Happiness Connection’, Journal of Public Deliberation, 7 (1), 1–16. Barry, Brian (1990) Political Argument: A Reissue with a New Introduction (New York: Harvester Wheatsheaf). Bates, Robert H., Greif, Avner, Levi, Margaret, Rosenthal, Jean-Laurant, and Weingast, Barry R. (1998) Analytic Narratives (Princeton: Princeton University Press). Baumgartner, Frank R. and Jones, Bryan D. (1993) Agendas and Instability in American Politics (Chicago: Chicago University Press). Baumgartner, Frank R., Breunig, Christian, Green-Pedersen, Christoffer, Jones, Bryan D., Mortensen, Peter B., Neytenmans, Michael, and Walgrave, Stefaan (2009) ‘Punctuated Equilibrium in Comparative Perspective’, American Journal of Political Science, 53 (3), 602–19. Bealer, George (1998) ‘Intuition and the Autonomy of Philosophy’, in M. Depaul and W. Ramsey (eds), Rethinking Intuition: The Psychology of Intuition and its Role in Philosophical Inquiry (Lanham, MD: Rowman and Littlefield). Ben-Nur, Averner and Putterman, Louis (2000) ‘On Some Implications of Evolutionary Psychology for the Study of Preferences and Institutions’, Journal of Economic Behavior and Organization, 43, 91–99. Berger, P. and Luckman, T. (1971) The Social Construction of Reality: A Treatise in the Sociology of Knowledge (Harmondsworth: Penguin). Berlinski, Samuel, Dewan, Torun, and Dowding, Keith (2007) ‘The Length of Ministerial Tenure in the UK, 1945–1997’, British Journal of Political Science, 37 (2), 245–62. Berlinski, Samuel, Dewan, Torun, and Dowding, Keith (2010) ‘The Impact of Individual and Collective Performance on Ministerial Tenure’, Journal of Politics, 72 (1), 1–13. Berlinski, Samuel, Dewan, Torun, and Dowding, Keith (2012) Accounting for Ministers: Scandal and Survival in British Government 1945–2007 (Cambridge: Cambridge University Press).

References

255

Bertalanffy, Ludwig von (1968) General Systems Theory: Foundations, Development, Applications (New York: George Braziller). Bevir, Mark and Rhodes, R. A. W. (2003) Interpreting British Governance (London: Routledge). Bhaskar, Roy (1979) The Possibility of Naturalism: A Philosophical Critique of the Contemporary Human Sciences (London: Routledge). Binmore, Ken (1994) Game Theory and the Social Contract I: Playing Fair (Cambridge, MA: MIT Press). Binmore, Ken (1998) Game Theory and the Social Contract II: Just Playing (Cambridge, MA: MIT Press). Binmore, Ken (2005) Natural Justice (Oxford: Oxford University Press). Binmore, Ken (2007) Playing for Real: A Text on Game Theory (Oxford: Oxford University Press). Blackburn, Simon (2008) How to Read Hume (London: Granta). Blackmore, Susan (1999) The Meme Machine (Oxford: Oxford University Press). Blais, André (2000) To Vote or Not to Vote: The Merits and Limits of Rational Choice Theory (Pittsburgh, PA: University of Pittsburgh Press). Blattberg, Charles (2009) Patriotic Elaborations: Essays in Practical Philosophy (Montreal: McGill-Queen’s University Press). Boudon, Raymond (1981) The Analysis of Ideology (Oxford: Polity). Bradley, Harriet (1999) ‘The Seductions of the Archive: Voices Lost and Found’, History of the Human Sciences, 12 (2), 107–22. Brady, Henry (2000) ‘Contributions of Survey Research to Political Science’, PS: Political Science and Politics, 33, 47–57. Brady, Henry E. and Collier, David C. (eds) (2004) Rethinking Social Inquiry: Diverse Tools, Shared Standards (Lanham, MD: Rowman and Littlefield). Brink, Chris (1989) ‘Verisimilitude: Views and Reviews’, History and Philosophy of Logic, 10 (1), 181–201. Brink, Chris and Britz, Katarina (1995) ‘Computing Verisimilitude’, Notre Dame Journal of Formal Logic, 36 (1), 31–43. Brinkmann, Svend (2013) Qualitative Interviewing: Understanding Qualitative Research (Oxford: Oxford University Press). Brown, Michael E., Lynn-Jones, Sean M., and Miller, Steven E. (1996) Debating the Democratic Peace (Cambridge, MA: MIT Press). Budge, Ian, Klingemann, Hans-Dieter, Volkens, Andrew, Bara, Judith, and Tanenbaum, Eric (2001) Mapping Policy Preferences: Estimates for Parties, Electors, and Governments 1945–1998 (Oxford: Oxford University Press). Burnham, Peter, Gilland, Karin, Grant, Wyn, and Layton-Henry, Zig (2009) Research Methods in Political Science (Houndmills: Palgrave Macmillan). Camerer, Colin F. (2003) Behavioral Game Theory: Experiments in Strategic Interaction (New York: Russell Sage). Camerer, Colin F., Loewenstein, George, and Prelec, Drazen (2005) ‘Neuroeconomics: How Neuroscience Can Inform Economics’, Journal of Economic Literature, 43, 9–64. Carballo, Alejandro Perez (2012) ‘Good Questions’, http://perezcarballo.org/work .html. Accessed 20 July 2014.

256

References

Carter, Ian (2011) ‘Value-freeness and Value-neutrality in the Analysis of Political Concepts’, MS. Catalinac, Amy (forthcoming) Electoral Reform and National Security in Japan: From Pork to Foreign Policy (Cambridge: Cambridge University Press). Chalmers, David J. (2011) ‘Verbal Disputes’, Philosophical Review, 120 (4), 515–66. Chalmers, David J. (2012) Constructing the World (Oxford: Oxford University Press). Chang, Linchiat and Krosnick, Jon A. (2009) ‘National Surveys Via RDD Telephone Interviewing Versus the Internet Comparing Sample Representativeness and Response Quality’, Public Opinion Quarterly, 73 (6), 641–78. Cheibub, José Antonio and Chernykh, Svitlana (2009) ‘Are Semi-Presidential Constitutions Bad for Democratic Performance?’, Constitutional Political Economy, 20 (2), 202–29. Cheibub, José Antonio, Elkins, Zachary, and Ginsburg, Tom (2014) ‘Beyond Presidentialism and Parliamentarism’, British Journal of Political Science, 44 (3), 515–44. Chi, M. T., Feltovich, P. J., and Glaser, R. (1981) ‘Categorization and Representation of Physics Problems by Experts and Novices’, Cognitive Science, 5 (2), 121–52. Child, Dennis (2006) The Essentials of Factor Analysis (3rd edn; New York: Continuum). Clarke, Kevin A. and Primo, David M. (2012) A Model Discipline: Political Science and the Logic of Representations (Oxford: Oxford University Press). Cohen, G. A. (1976) Karl Marx’s Theory of History: A Defence (Oxford: Clarendon Press). Cohen, G. A. (1982) ‘Functional Explanation, Consequence Explanation, and Marxism’, Inquiry, 25, 27–56. Cohen, G. A. (2008) Rescuing Justice and Equality (Cambridge, MA: Harvard University Press). Collier, David (2011) ‘Understanding Process Tracing’, PS: Political Science and Politics, 44 (4), 823–30. Collingwood, R. G. (1939) An Autobiography (Oxford: Oxford University Press.). Colomer, Josep M. (2005) ‘It’s Parties that Choose Electoral Systems (or, Duverger’s Laws Upside Down)’, Political Studies, 53 (1), 1–21. Connolly, William E. (1983) The Terms of Political Discourse, 2nd edn (Oxford: Martin Robertson). Cooper, Mick P. (2000) ‘Web Surveys: A Review of the Issues and Approaches’, Public Opinion Quarterly, 64, 464–94. Cowden, Mhairi (2012) ‘Capacity, Claims and Children’s Rights’, Contemporary Political Theory, 11 (4), 362–80. Dafoe, Allan (2014) ‘Science Deserves Better: The Imperative to Share Complete Replication Files’, PS: Political Science and Politics, 47 (1), 60–6. Dahl, Robert A. (1961a) Who Governs? Democracy and Power in an American City (New Haven: Yale University Press). Dahl, Robert A. (1961b) ‘The Behavioral Approach in Political Science: Epitaph for a Monument to a Successful Protest’, American Political Science Review, 55, 763–72. Dahl, Robert A. (1982) Dilemmas of Pluralist Democracy: Autonomy vs. Control (New Haven: Yale University Press).

References

257

Dalton, Russell J., Farrell, David M., and McAllister, Ian (2011) Political Parties and Democratic Linkage: How Parties Organize Democracy (Oxford: Oxford University Press). Dalvean, Michael (2012a) ‘The Selection of Cabinet Ministers in the Australian Federal Parliament’ (Australian National University, PhD thesis). Dalvean, Michael (2012b) ‘Predicting Cabinet Ministers: A Psychological Approach’, in Keith Dowding and Chris Lewis (eds), Ministerial Careers and Accountability in the Australian Commonwealth Government (Canberra: ANU E Press), 35–65. Damer, T. Edward (2005) Attacking Faulty Reasoning: A Practical Guide to Fallacyfree Arguments (Boston, MA: Wadsworth). Daniels, Norman (1979) ‘Wide Reflective Equilibrium and Theory Acceptance in Ethics’, Journal of Philosophy, 76 (5), 256–82. Daniels, Norman (1996) Justice and Justification: Reflective Equilibrium Theory in Practice (New York: Cambridge University Press). Davidson, Donald (1980) ‘Actions, Reasons and Causes’, in his Essays on Actions and Events (Oxford: Oxford University Press). Davidson, Donald (2004) ‘Paradoxes of Irrationality’, in his Problems of Rationality (Oxford: Oxford University Press), 169–87. Dawes, Robyn M. (1996) ‘Comment: Counterfactual Inferences as Instances of Statistical Inferences’, in Paul E. Tetlock and A. Belkin (eds), Counterfactual Thought Experiments in World Politics: Logical, Methodological and Pyschological Issues (Princeton: Princeton University Press), 300–8. Dawes, Robyn M. (2001) Everyday Irrationality: How Pseudo-Scientists, Lunatics, and the Rest of Us Systematically Fail to Think Rationally (Boulder, CO: Westview). Dawkins, Richard (1976) The Selfish Gene (Oxford: Oxford University Press). Dawkins, Richard (1999) The Extended Phenotype: The Long Reach of the Gene (Oxford: Oxford University Press). Dennett, Daniel C. (1978) ‘How to Change Your Mind’, in his Brainstorms: Philosophical Essays on Mind and Psychology (London: Penguin). Dennett, Daniel C. (1980) ‘The Milk of Human Intentionality’, Behavioral and Brain Sciences, 3 (4), 428–30. Dennett, Daniel C. (1998) ‘Real Patterns’, Brainchildren: Essays on Designing Minds (Harmondsworth: Penguin). Dennett, Daniel C. (2006) Breaking the Spell: Religion as a Natural Phenomenon (London: Allen Lane). Dennett, Daniel C. (2013) Intuition Pumps and Other Tools for Thinking (London: Allen Lane). Dessler, David (1991) ‘Beyond Correlations: Toward a Causal Theory of War’, International Studies Quarterly, 35 (4), 337–55. De Vaus, David (2002) Surveys in Social Research (Sydney: Allen and Unwin). Diamond, Jared (1998) Guns, Germs and Steel: A Short History of Everybody for the Last 13,000 Years (London: Vintage). Dillman, Don A. (2007) Mail and Internet Surveys (New York: John Wiley). Diwakar, Rekha (2007) ‘Duverger’s Law and the Size of the Indian Party System’, Party Politics, 13 (5), 539-62.

258

References

Doorenspleet, Renske (2000) ‘Reassessing the Three Waves of Democratization’, World Politics, 52 (3), 384–406. Doucouliagos, Hristos and Ali, Mehmet (2008) ‘Democracy and Economic Growth: A Meta Analysis’, American Journal of Political Science, 52 (1), 61–83. Dowding, Keith (1991) Rational Choice and Political Power (Aldershot: Edward Elgar). Dowding, Keith (1995a) ‘Model or Metaphor? A Critical Review of the Policy Network Approach’, Political Studies, 45 (1), 136–58. Dowding, Keith (1995b) ‘Interpreting Formal Coalition Theory’, in Keith Dowding and Desmond S. King (eds), Preferences, Institutions and Rational Choice (Oxford: Clarendon Press), 43–59. Dowding, Keith (2002) ‘Revealed Preference and External Reference’, Rationality and Society, 14 (3), 259–84. Dowding, Keith (2003) ‘Resources, Power and Systematic Luck: Reply to Barry’, Politics, Philosophy and Economics, 3 (2), 305–22. Dowding, Keith (2005) ‘Is It Rational to Vote? Five Types of Answer and a Suggestion’, British Journal of Politics and International Relations, 7, 1–18. Dowding, Keith (2006a) ‘Can Populism Be Defended? William Riker, Gerry Mackie and the Interpretation of Democracy’, Government and Opposition, 41 (3), 327–46. Dowding, Keith (2006b) ‘Can Capabilities Reconcile Freedom and Equality?’, Journal of Political Philosophy, 14 (3), 323–36. Dowding, Keith (2008) ‘Agency and Structure: Interpreting Power Relationships’, Journal of Power, 1 (1), 21–36. Dowding, Keith (2009) ‘What Is Welfare and How Can We Measure It?’, in Harold Kincaid and Don Ross (eds), The Oxford Handbook of Philosophy of Economics (Oxford: Oxford University Press), 511–39. Dowding, Keith (2013a) ‘The Role of Political Argument in Justice as Impartiality’, Political Studies, 61 (1), 67–81. Dowding, Keith (2013b) ‘The Prime Ministerialization of the British Prime Minister’, Parliamentary Affairs, 66 (3), 617–35. Dowding, Keith (2013c), ‘Presidentialisation Again: A Comment on Kefford’, American Journal of Political Science, 48 (2), 147–9. Dowding, Keith and Christiansen, Lars (1994) ‘Pluralism or State Autonomy? The Case of Amnesty International (British Section): The Insider/Outsider Group’, Political Studies, 42 (1), 15–24. Dowding, Keith and John, Peter (2012) Exits, Voices and Social Investment: Citizens’ Reactions to Public Services (Cambridge: Cambridge University Press). Dowding, Keith and Kang, Won-Taek (1998) ‘Ministerial Resignations 1945–97’, Public Administration, 76 (3), 411–29. Dowding, Keith and Lewis, Chris (2012) ‘Culture, Newspaper Reporting and Changing Perceptions of Ministerial Accountability in Australia’, Australian Journal of Politics and History, 58, (2), 226–50. Dowding, Keith and van Hees, Martin (2003) ‘The Construction of Rights’, American Political Science Review, 97 (2), 281–93. Dowding, Keith and van Hees, Martin (2007) ‘Counterfactual Success and Negative Freedom’, Economics and Philosophy, 23, 141–62.

References

259

Dowding, Keith and van Hees, Martin (2008) ‘In Praise of Manipulation’, British Journal of Political Science, 38 (1), 1–16. Dowding, Keith and van Hees, Martin (2009) ‘Freedom of Choice’, in Paul Anand, Prasanta K. Pattanaik, and Clemens Puppe (eds), Oxford Handbook of Rational and Social Choice (Oxford: Oxford University Press), 374–92. Dowding, Keith, Lewis, Chris, and Packer, Adam (2012) ‘The Pattern of Forced Exits from the Ministry’, in Keith Dowding and Chris Lewis (eds), Ministerial Careers and Accountability in the Australian Commonwealth Government (Canberra: ANU E Press). Dowding, Keith, John, Peter, Mergoupis, Thanos, and Vugt, Mark van (2000) ‘Exit, Voice and Loyalty: Analytic and Empirical Developments’, European Journal of Political Research, 37 (4), 469–95. Downs, Anthony (1957) An Economic Theory of Democracy (New York: Harper and Row). Dreben, Burton (1992) ‘Putnam, Quine – and the Facts’, Philosophical Topics, 20 (1), 293–315. Druckman, James N. and Green, Donald P. (2011) Cambridge Handbook of Experimental Political Science (Cambridge: Cambridge University Press). Dryzek, John and Dunleavy, Patrick (2009) Theories of the Democratic State (Houndmills: Palgrave Macmillan). Duhem, Pierre (1954) The Aim and Structure of Physical Theory (Princeton: Princeton University Press). Dunbar, Robin (ed.) (1999) The Evolution of Culture (New Brunswick, NJ: Rutgers University Press). Dunleavy, Patrick and Boucek, Françoise (2003) ‘Constructing the Number of Parties’, Party Politics, 9 (3), 291–315. Duverger, Maurice (1954) Political Parties: Their Organization and Activity in the Modern State (London: Methuen). Duverger, Maurice (1980) ‘A New Political System Model: Semi-Presidential Government’, European Journal of Political Research, 8 (2), 165–87. Dworkin, Ronald (1977) Taking Rights Seriously (Cambridge, MA: Harvard University Press). Easton, David (1965) A Framework for Political Analysis (Englewood Cliffs, NJ: Prentice-Hall). Eijk, Cees van der and Franklin, Mark N. (2009) Elections and Voters (Houndmills: Palgrave Macmillan). Einstein, Albert (1923/1911) ‘On the Influence of Gravitation on the Propagation of Light’, in H. A. Lorentz, A. Einstein, H. Minkowski and H. Weyl (eds), with notes by A Sommerfield, trans. W. Perrett and G. B. Perry, The Principle of Relativity: A Collection of Original Memoirs on the Special and General Theory of Relativity (London: Dover), 97–108. Einstein, Albert (1961/1916) Relativity: The Special and General Theory, trans. Robert W. Lawson (New York: Bonanza Books). Eldersveld, Samuel T. (1956) ‘Experimental Propaganda Techniques and Voting Behaviour’, American Political Science Review, 50 (1), 154–65. Elgie, Robert (2004) ‘Semi-Presidentialism: Concepts, Consquences and Contesting Explanations’, Working Paper in International Studies, Centre for International Studies (Dublin University).

260

References

Elman, Colin (2005) ‘Explanatory Typologies in Qualitative Studies of International Politics’, International Organization, 59 (2), 293–326. Elman, Colin and Kapiszewski, Diana (2014) ‘Data Access and Research Transparency in the Qualitative Tradition’, PS: Political Science and Politics, 47 (1), 43–47. Elster, Jon (1989) Nuts and Bolts for the Social Sciences (Cambridge: Cambridge University Press). Elster, Jon (1998) ‘A Plea for Mechanisms’, in Peter Hedstrom and Richard Swedberg (eds), Social Mechanisms: An Analytical Approach to Social Theory (Cambridge: Cambridge University Press), 45–73. Enc, Berent (1979) ‘Function Attributions and Functional Explanations’, Philosophy of Science, 46 (3), 343–65. Ereshefsky, Marc (2001) The Poverty of the Linnaean Hierarchy (Cambridge: Cambridge University Press). Ereshefsky, Marc (2009) ‘Natural Kinds in Biology’, in Edward Craig (ed.), Routledge Encyclopedia of Philosophy (1; London: Routledge), https://http://www.rep.rout ledge.com/articles/natural-kinds-in-biology, Accessed 15 May 2015. Eriksson, Lina and Hajek, Alan (2007) ‘What Are Degrees of Belief?’, Studia Logica, 86 (2), 183–213. Fabrini, S. (1994) ‘Presidentialism as Americanization? The Rise and Fall of LeaderDominated Strategies in Western Europe in the “Eighties”’, American Studies International, 32 (1), 51–65. Fairclough, N. and Wodak, R. (1997) ‘Critical Discourse Analysis’, in T. A. van Dijk (ed.), Discourse Studies: A Multidisciplinary Introduction, vol. 2 (London: Sage). Farrar, Cynthia, Fishkin, James S., Green, Donald P., List, Christian, Luskin, Robert C., and Paluck, Elizabeth Levy (2010) ‘Disaggregating Deliberation’s Effects: An Experiment within a Deliberative Poll’, British Journal of Political Science, 40 (4), 333–47. Fenno, Richard F. Jnr (1978) Home Style: House Members in their Districts (Boston, MA: Little Brown). Fenno, Richard F. Jnr (1990) Watching Politicians: Essays on Participant Observation (Berkeley, CA: IGS Press). Feyerabend, Paul (1975) Against Method (London: Verso). Feyerabend, Paul (1978) Science in a Free Society (London: Verso). Fischer, Jörn, Dowding, Keith, and Dumont, Patrick (2012) ‘The Duration and Durability of Cabinet Ministers’, International Political Science Review, 33 (4), 1–15. Fisher, R. A. (1935) The Design of Experiments (London: Oliver and Boyd). Foley, Michael (1993) The Rise of the British Presidency (Manchester: Manchester University Press). Foley, Michael (2000) The British Presidency: Tony Blair and the Politics of British Leadership (Manchester: Manchester University Press). Foley, Michael (2004) ‘Presidential Attribution as an Agency of Prime Ministerial Critique in a Parliamentary Democracy: The Case of Tony Blair’, British Journal of Politics and International Relations, 6 (3), 292–311. Foot, Philippa (1968) ‘The Problem of Abortion and the Doctrine of Double Effect’, in her Virtues and Vices and Other Essays in Moral Philosophy (Oxford: Basil Blackwell), 19–32.

References

261

Foucault, Michel (1975) Discipline and Punish, trans. Allan Sheridan (New York: Pantheon). Foucault, Michel (1980) Power/Knowledge: Selected Interviews and Other Writings 1972–1977 (Brighton, Sussex: Harvester). Foucault, Michel (2006) History of Madness, trans. Jonathan Murphy and Jean Khalfa (New York: Routledge). Fowler, James, Baker, Laura A., and Dawes, Christopher T. (2008) ‘Genetic Variation in Political Participation’, American Political Science Review, 102 (2), 233–48. Franklin, Mark N. (2004) Voter Turnout and the Dynamics of Electoral Competition in Established Democracies since 1945 (Cambridge: Cambridge University Press). Frey, Bruno S. (2003) ‘Publishing as Prostitution? Choosing between One’s Own Ideas and Academic Success’, Public Choice, 116, 205–23. Frey, Bruno S. and Stutzer, Alois (2002) Happiness and Economics: How the Economy and Institutions Affect Human Well-Being (Princeton: Princeton University Press). Fricker, Miranda (2007) Epistemic Injustice: Power and the Ethics of Knowing (Oxford: Oxford University Press). Friedman, Milton (1953) ‘The Methodology of Positive Economics’, in his Essays in Positive Economics (Chicago: Chicago University Press). Gaines, Brian J. (1999) ‘Duverger’s Law and the Meaning of Canadian Exceptionalism’, Comparative Political Studies, 32 (7), 835–61. Galilei, Galileo (2012/1638) Dialogues Concerning Two New Sciences, trans. Henry Crew and Alfonso de Salvio, http://ebooks.adelaide.edu.au/g/galileo/dialogues/ complete.html; Adelaide: University of Adelaide ebooks. Gallie, W. B. (1956) ‘Essentially Contested Concepts’, Proceedings of the Aristotelian Society, 56 (2), 167–98. Gamble, Andrew (1990) ‘Theories of British Politics’, Political Studies, 38 (3), 404–20. Gawel, R. and Godden, P. W. (2008) ‘Evaluation of Consistency of Wine Quality Assessments from Expert Wine Tasters’, Australian Journal of Wine Research, 14 (1), 1–8. Gee, Paul (2005) An Introduction to Discourse Analysis: Theory and Method (London: Routledge). Geertz, Clifford (1973) The Interpretation of Cultures: Selected Essays (New York: Basic Books). Gelman, S. A. (2009) ‘Learning from Others: Children’s Construction of Concepts’, Annual Review of Psychology, 60, 115–40. George, Alexander L. (1979) ‘Case Studies and Theory Development: The Method of Structured, Focused Comparison’, in Paul Gordon Lauren (ed.), Diplomacy: New Approaches in History, Theory and Policy (New York: Free Press), 43–68. George, Alexander L. and Bennett, Andrew (2005) Case Studies and Theory Development in the Social Sciences (Cambridge, MA: MIT Press). Gerber, Alan S. and Green, Donald P. (1999) ‘Does Canvassing Increase Voter Turnout? A Field Experiment’, Proceedings of the National Academy of Sciences of the United States of America, 96 (19), 10939–42. Gerber, Alan S. and Green, Donald P. (2000) ‘The Effects of Canvassing, Telephone Calls, and Direct Mail on Voter Turnout: A Field Experiment’, American Political Science Review, 94 (3), 653–63.

262

References

Gerber, Alan S. and Green, Donald P. (2008) Get Out the Vote: How to Increase Voter Turnout (Washington, DC: Brookings Institution Press). Gerber, Alan S. and Green, Donald P. (2012) Field Experiments: Design, Analysis and Interpretation (New York, NY: W.W. Norton and Co). Gerber, Alan S., Karlan, S. D., and Bergan, D. (2009) ‘Does the Media Matter? A Field Experiment Measuring the Effects of Newspapers on Voting Behavior and Political Opinions’, American Economic Journal: Applied Economics, 1 (2), 35–52. Gerring, John (2007) Case Study Research: Principles and Practices (Cambridge: Cambridge University Press.). Gerring, John (2008) ‘The Mechanistic Worldview: Thinking Inside the Box’, British Journal of Political Science, 38 (1), 161–79. Gerring, John (2012) Social Science Methodology: A Unified Framework, 2nd edn (Cambridge: Cambridge University Press). Gibbard, Allan (1977) ‘Manipulation of Voting Schemes that Mix Voting with Chance’, Econometrica, 45, 665–79. Gibbon, Edward (1993/1781) The Decline and Fall of the Roman Empire, vol. 2 (New York: Alfred A. Knopf). Glimcher, Paul W. (2004) Decisions, Uncertainty, and the Brain: The Science of Neuroeconomics (Cambridge, MA: MIT Press). Glimcher, Paul W. (2011) Foundations of Neuroeconomic Analysis (Oxford: Oxford University Press). Goertz, Gary (2006) Social Science Concepts: A User’s Guide (Princeton: Princeton University Press). Goertz, Gary and Mahoney, James (2012) A Tale of Two Cultures: Qualitative and Quantitative Research in the Social Sciences (Princeton: Princeton University Press). Goodman, Nelson (1955) Fact, Fiction and Forecast (Cambridge, MA: Harvard University Press). Gosnell, Harold F. (1926) ‘An Experiment in the Stimulation of Voting’, American Political Science Review, 20 (4), 869–74. Gosnell, Harold F. (1927) Getting-Out-The-Vote: An Experiment in the Stimulation of Voting (Chicago: Chicago University Press.). Green, Donald P. and Shapiro, Ian (1994) Pathologies of Rational Choice Theory (New Haven: Yale University Press). Green, Donald P. and John, Peter (eds) (2010) Field Experiments in Comparative Politics and Policy (London: Sage). Greenburg, Joseph and Weber, Shlomo (1986) ‘Strong Tiebout Equilibrium under Restricted Preferences Domain’, Journal of Economic Theory, 38, 101–17. Greene, Joshua D. (2013) Moral Tribes: Emotion, Reason, and the Gap between Us and Them (New York: Penguin). Greene, Joshua D. and Haidt, Jonathan (2002) ‘How (and Where) Does Moral Judgement Work?’, Trends in Cognitive Sciences, 6 (12), 517–23. Greene, Joshua D., Sommerville, R. Brian, Nystrom, Leigh E., Darley, John M., and Cohen, Jonathan D. (2001) ‘An fMRI Investigation of Emotional Engagement in Moral Judgement’, Science, 293 (5537), 2105–8.

References

263

Greene, Joshua D., Nystrom, Leigh E., Engell, Andrew D., Darley, John M., and Cohen, Jonathan D. (2004) ‘The Neural Bases of Cognitive Conflict and Control in Moral Judgement’, Neuron, 44 (2), 389. Greenleaf, W. H. (1983) The British Political Tradition (London: Routledge). Grice, Paul (1989) Studies in the Way of Words (Cambridge, MA: Harvard University Press). Grimmer, Justin (2015) ‘We Are All Social Scientists Now: How Big Data, Machine Learning, and Causal Inference Work Together’, PS: Political Science and Politics, 48 (1), 80–3. Grofman, Bernard (2006) ‘The Impact of Electoral Laws on Political Parties’, in Barry R. Weingast and Donald A. Wittman (eds), The Oxford Handbook of Political Institutions (Oxford: Oxford University Press), 102–18. Grofman, Bernard and Kline, Reuben (2011) ‘How Many Political Parties Are There, Really? A New Measure of the Ideologically Cognizable Number of Parties/Party Groupings’, Party Politics, 18 (4), 523–44. Grofman, Bernard and Lijphart, Arend (eds) (1994) Electoral Laws and their Political Consequences (New York: Agathon Press). Grofman, Bernard, Blais, André, and Bowler, Shaun (eds) (2009) Duverger’s Law of Plurality Voting: The Logic of Party Competition in Canada, India, the United Kingdom and the United States (New York: Springer). Guala, Francesco (2005) The Methodology of Experimental Economics (Cambridge: Cambridge University Press). Habermas, Jürgen (1984) The Theory of Communicative Action, vol. 1: Reason and the Rationalization of Society, trans. Thomas McCarthy (Cambridge: Policy). Habermas, Jürgen (1987) The Philosophical Discourse of Modernity: Twelve Lectures (Cambridge, MA: MIT Press). Habermas, Jürgen (1998) ‘Social Action, Purposive Activity, and Communication’, in Maeve Cooke (ed.), On the Pragmatics of Communication (Cambridge, MA: MIT Press), 105–82. Haidt, Jonathan (2001) ‘The Emotional Dog and Its Rational Tail: A Social Intuitionist Approach to Moral Judgement’, Psychological Review, 108 (4), 814–34. Haidt, Jonathan (2003) ‘The Moral Emotions’, in R. J. Davidson, K. R. Scherer, and H. H. Goldsmith (eds), Handbook of Affective Sciences (Oxford: Oxford University Press), 852–70. Hall, N. (2004) ‘Two Concepts of Causation’, in J. Collins, N. Hall, and L. A. Paul (eds), Causation and Counterfactuals (Cambridge, MA: MIT Press), 225–76. Hart, John (1991) ‘President and Prime Minister: Convergence or Divergence?’, Parliamentary Affairs, 44 (2), 208–25. Hart, John (1992) ‘An Australian President? A Comparative Perspective’, in Patrick Weller (ed.), Menzies to Keating: The Development of the Australian Prime Ministership (Carlton, Vic.: Melbourne University Press). Hatemi, Peter K. and McDermott, Rose (eds) (2011) Man Is by Nature a Political Animal: Evolution, Biology and Politics (Chicago: University of Chicago Press). Hauser, Mark (2006) Moral Minds (New York: Ecco). Hedstrom, Peter and Swedberg, Richard (eds) (1998) Social Mechanisms: An Analytical Approach to Social Theory (Cambridge: Cambridge University Press).

264

References

Helms, Ludger (2005) ‘The Presidentialisation of Political Leadership: British Notions and German Observations’, Political Quarterly, 76 (3), 430–8. Hempel, Carl G. (1965) Aspects of Scientific Explanation and Other Essays (New York: Free Press). Hempel, Carl G. (1966) Philosophy of Natural Science (Englewood Cliffs, NJ: Prentice-Hall). Hersch, Eitan (2013) ‘Long-Term Effects of September 11 on the Political Behavior of Victims’ Families and Neighbors’, Proceedings of the National Academy of Sciences, 110 (52), 20959–63. Hillard, Dustin, Purpura, Stephen, and Wilkerson, John (2007) ‘Computer-Assisted Topic Classification for Mixed-Methods Social Science Research’, Journal of Information Technology and Politics, 4 (1), 31–46. Hirschman, Albert O. (1970) Exit, Voice and Loyalty: Responses to Decline in Firms, Organizations and States (Cambridge, MA: Harvard University Press). Hirschman, Albert O. (1986) Rival Views of Market Society and Other Recent Essays (New York: Viking). Hitchcock, Christopher and Woodward, James (2003) ‘Explanatory Generalizations, Part II: Plumbing Explanatory Depth’, Noûs, 37 (2), 181–99. Hodgson, Geoffrey M. (1999) Evolution and Institutions: On Evolutionary Economics and the Evolution of Economics (Cheltenhan: Edward Elgar). Hodgson, Robert T. (2008) ‘An Examination of Judge Reliability at a Major U.S. Wine Competition’, Journal of Wine Economics, 3 (2), 105–13. Hodgson, Robert T. (2009) ‘How Expert Are “Expert” Wine Judges?’, Journal of Wine Economics, 4 (2), 233–41. Hofstadter, Douglas and Sander, Emmanuel (2013) Surfaces and Essences: Analogy as the Fuel and Fire of Thinking (New York: Basic Books). Holland, Paul W. (1986) ‘Statistics and Causal Inference’, Journal of the American Statistical Association, 81 (396), 945–60. Howson, Colin (2000) Hume’s Problem: Induction and the Justification of Belief (Oxford: Oxford University Press). Hume, David (1742/1978) A Treatise of Human Nature (Oxford: Oxford University Press). Humphreys, Macartan, Sanchex de la Sierra, Raul, and van der Windt, Peter (2013) ‘Fishing, Commitment, and Communication: A Proposal for Comprehensive Nonbinding Research Registration’, Political Analysis, 21 (1), 1–20. Imbens, Guido W. and Lemieux, Thomas (2008) ‘Regression Discontinuity Designs: A Guide to Practice’, Journal of Econometrics, 142 (2), 615–35. Inglehart, Ronald (2009) ‘Democracy and Happiness: What Causes What?’, in A. K. Dutt and Ben Radcliff (eds), Happiness, Economics and Politics: Towards a MultiDisciplinary Approach (Cheltenham: Edward Elgar), 256–70. Ingold, Tim (2008) ‘Anthropology Is Not Ethnography’, Proceedings of the British Academy, 154 (1), 69–92. Isaacson, Walter (2007) Einstein: His Life and Universe (New York: Simon and Schuster). Ish-Shalom, Piki (2013) Democratic Peace: A Political Biography (Michigan, IL: University of Michigan Press).

References

265

John, Peter (1998) Analysing Public Policy (London: Pinter). John, Peter (2006) ‘The Policy Agendas Project: A Review’, Journal of European Public Policy, 13 (7), 975–86. Johnson, Gary R. (1995) ‘The Evolutionary Origins of Government and Politics’, in Albert Somit and Joseph Losco (eds), Human Nature and Politics (Greenwich, CT: JAI Press). Johnson, Paul E. (1996) ‘Unraveling in a Variety of Institutional Settings’, Journal of Theoretical Politics, 8 (4), 299–331. Johnson, Paul E. (1999) ‘Simulation Modeling in Political Science’, American Behavioral Scientist, 42 (8), 1509–30. Jolliffe, I. T. (2002) Principal Component Analysis, 2nd edn (Berlin: Springer). Jones, Bryan D. and Baumgartner, Frank R. (2005) The Politics of Attention: How Government Prioritizes Problems (Chicago: Chicago University Press). Jones, Bryan D., Baumgartner, Frank R. Breunig, Christian, Wlezien, Christopher, Soroka, Stuart, Foucault, Martial, François, Abel, Green-Pedersen, Christoffer, Koski, Chris, John, Peter, Mortensen, Peter B. Varone, Frédéric, and Walgrave, Stefaan (2009) ‘A General Empirical Law of Public Budgets: A Comparative Analysis’, American Journal of Political Science, 53 (4), 855–73 Jones, G. W. (2006) ‘The British Presidency: Tony Blair and the Politics of Public Leadership’, American Political Science Review, 95 (4), 1017–18. Jones, M. (1995) Electoral Laws and the Survival of Presidential Democracies (Notre Dame, IN: University of Notre Dame Press). Kahneman, Daniel (2011) Thinking Fast and Slow (New York: Farrar, Straus and Giroux). Kalton, Grahan (1983) Introduction to Survey Sampling (Los Angeles, CA: Sage). Kefford, Glenn (2013) ‘The Presidentialisation of Australian Politics? Kevin Rudd’s Leadership of the Australian Labor Party’, Australian Journal of Political Science, 48 (2), 135–46. Keohane, Robert (2009) ‘Political Science as a Vocation’, PS: Political Science and Politics, 42 (2), 359–63. Kim, Hyung Min and Rousseau, David L. (2005) ‘The Classical Liberals Were Half Right (or Half Wrong): New Tests of the Democratic Peace’, Journal of Peace Research, 42 (5), 523–43. King, Gary (1986) ‘How Not to Lie with Statistics: Avoiding Common Mistakes in Quantitative Political Science’, American Journal of Political Science, 30 (4), 666–87. King, Gary (1995) ‘Replication, Replication’, PS: Political Science and Politics, 28 (3), 443–99. King, Gary (2009) ‘The Changing Evidence Base of Social Science Research’, in Gary King, Kay Schlozman, and Norman Nie (eds), The Future of Political Science: 100 Perspectives (New York: Routledge), 91–3. King, Gary, Keohane, Robert O., and Verba, Sidney (1994) Designing Social Inquiry: Scientific Inference in Qualitative Research (Princeton: Princeton University Press). Krehbiel, Keith (1991) Information and Legislative Organization (Ann Arbor: University of Michigan Press).

266

References

Kripke, Saul (1972) ‘Naming and Necessity’, in Donald Davidson and Gilbert Harman (eds), Semantics of Natural Language (Dordrecht: Reidel), 252–355. Kripke, Saul (1980) Naming and Necessity (Oxford: Blackwell). Krueger, Richard A. and Casey, Mary Anne (2014) Focus Groups: A Practical Guide for Applied Research (London: Sage). Kuhn, Thomas (1962) The Structure of Scientific Revolutions (Chicago: University of Chicago Press). Kuhn, Thomas (1970) ‘Reflection on My Critics’, in Imre Lakatos and Alan Musgrave (eds), Criticism and the Growth of Knowledge (Cambridge: Cambridge University Press). Kuhn, Thomas (1977) ‘Second Thoughts on Paradigms’, in Frederick Suppe (ed.), The Structure of Scientific Revolutions, 2nd edn (Urbana, IL: University of Illinois Press), 459–82. Ladyman, James and Ross, Don, with Spurrett, David and Collier, John (2007) Every Thing Must Go (Oxford: Oxford University Press). Lakatos, Imre (1978) The Methodology of Scientific Research Programmes (Cambridge: Cambridge University Press). Lasswell, Harold D. (1950) Politics: Who Gets What, When, How (New York: McGraw-Hill). Laver, Michael and Hunt, W. Ben (1992) Policy and Party Competition (London: Routledge). Laver, Michael and Sergenti, Ernest (2012) Party Competition: An Agent-Based Model (Princeton: Princeton University Press). Laver, Michael, Benoit, Kenneth, and Mikhavlov, Slava (2010) ‘Treating Words and Data with Error: Uncertainty in Text Statements of Policy Positions’, American Journal of Political Science, 53 (2), 495–513. Lee, David S. and Lemieux, Thomas (2010) ‘Regression Discontinuity Designs in Economics’, Journal of Economic Literature, 48 (2), 281–355. Lee, David S., Moretti, Enrico, and Butler, Matthew J. (2004) ‘Do Voters Affect or Elect Policies? Evidence from the U.S. House’, Quarterly Journal of Economics, 119 (3), 807–59. Lenski, Gerhard (1994) ‘Societal Taxonomies: Mapping the Social Universe’, Annual Review of Sociology, 20, 1–26. LePora, Chiara and Goodin, Robert E. (2013) On Complicity and Compromise (Oxford: Oxford University Press). Lewis, David K. (1973) ‘Causation’, Journal of Philosophy, 70 (4), 556–67. Lewis, David K. (1983) Philosophical Papers, vol. 1 (Oxford: Oxford University Press). Lewis-Beck, Michael S. (2005) ‘Election Forecasting: Principles and Practice’, British Journal of Politics and International Relations, 7 (2), 145–64. Lijphart, Arend (1968) The Politics of Accommodation: Pluralism and Democracy in the Netherlands (Berkeley, CA: University of California Press). List, Christian and Spiekermann, Kai (2013) ‘Methodological Individualism and Holism in Political Science: A Reconciliation’, American Political Science Review, 107 (4), 629–43. Little, Daniel (1998) Microfoundations, Method and Causation (New Brunswick, NJ: Transaction).

References

267

Lohr, Sharon L. (2009) Sampling: Design and Analysis (Boston, MA: Brooks/Cole). Lupia, Arthur and Elman, Colin (2012) ‘Openness in Political Science: Data Access and Research Transparency’, PS: Political Science and Politics, 47 (1), 19–42. Lupia, Arthur and Alter, George (2014) ‘Data Analysis and Research Transparency in the Quantitative Tradition’, PS: Political Science and Politics, 47 (1), 54–9. Macdonald, Stuart Elaine and Rabinowitz, George (1993) ‘Direction and Uncertainty in a Model of Issue Voting’, Journal of Theoretical Politics, 5 (1), 61–87. Macdonald, Stuart Elaine, Listhaug, Ola, and Rabinowitz, George (1991) ‘Issues and Party Support in Multiparty Systems’, American Political Science Review, 85 (4), 1107–31. Mackie, Gerry (2003) Democracy Defended (Cambridge: Cambridge University Press). Mackie, John. L (1974) The Cement of the Universe: A Study in Causation (Oxford: Clarendon Press). Marsh, David and Furlong, Paul (2002) ‘A Skin, not a Sweater: Ontology and Epistemology in Political Science’, in David Marsh and Gerry Stoker (eds), Theory and Methods in Political Science (London: Palgrave). Masterman, Margaret (1970) ‘The Nature of a Paradigm’, in Imre Lakatos and Alan Musgrave (eds), Criticism and the Growth of Knowledge (Cambridge: Cambridge University Press), 59–89. Matthen, Mohan and Ariew, André (2009) ‘Selection and Causation’, Philosophy of Science, 6 (7), 201–24. Matusik, Martin J. (1989) ‘Habermas on Communicative Reason and Performative Contradiction’, New German Critique, 47 (1), 143–72. Mayr, Ernst (1961) ‘Cause and Effect in Biology’, Science, 131, 1501–6. McCloskey, Donald (1970) ‘Did Victorian Britain Fail?’, Economic History Review, 23 (3), 446–59. McDermott, Rose (2014) ‘Research Transparency and Data Archiving for Experiments’, PS: Political Science and Politics, 47 (1), 67–71. McKelvey, Richard D. (1976) ‘Intransitivities in Multi-dimensional Voting Models and Some Implications for Agenda Control’, Journal of Economic Theory, 12, 472–82. McKinney, John C. (1966) Constructive Typlogy and Social Theory (New York: Appleton-Century-Crofts). McLean, Iain (2001) Rational Choice and British Politics: An Analysis of Rhetoric and Manipulation from Peel to Blair (Oxford: Oxford University Press). Merrill III, Samuel and Grofman, Bernard (1999) A Unified Theory of Voting: Directional and Proximity Spatial Models (Cambridge: Cambridge University Press). Miller, David (1974a) ‘On the Comparison of False Theories by their Bases’, British Journal for the Philosophy of Science, 25 (1), 178–88. Miller, David (1974b) ‘Popper’s Qualitative Theory of Verisimilitude’, British Journal for the Philosophy of Science, 25 (1), 166–77. Mills, Charles W. (1997) The Racial Contract (Ithaca: Cornell University Press). Mohri, Mehryar, Rostamizadeh, Afshin, and Talwalkar, Ameet (2012) Foundations of Machine Learning (Cambridge, MA: MIT Press). Molotch, Harvey (1976) ‘The City as a Growth Machine’, American Journal of Sociology, 82, 309–30.

268

References

Monroe, Burt, Pan, Jennifer, Roberts, Margaret E., Sen, Maya, and Sinclair, Betsy (2015) ‘No! Formal Theory, Causal Inference, and Big Data Are Not Contradictory Trends in Political Science’, PS: Political Science and Politics, 48 (1), 71–4. Moore, G. E. (1903) Principia Ethica (Cambridge: Cambridge University Press). Moravcsik, Andrew (2014) ‘Transparency: The Revolution in Qualitative Research’, PS: Political Science and Politics, 47 (1), 48–53. Morgan, David L. (1996) Focus Groups as Qualitative Research (London: Sage). Morris, Michael (2007) An Introduction to the Philosophy of Language (Cambridge: Cambridge University Press). Morton, Rebecca B. (1999) Methods and Models: A Guide to the Empirical Analysis of Formal Models in Political Science (Cambridge: Cambridge University Press). Morton, Rebecca B. and Williams, Kenneth C. (2010) Experimental Political Science and the Study of Causality: From Nature to the Lab (Cambridge: Cambridge University Press). Murray, J. D. (1989) Mathematical Biology (Düsseldorf: Springer-Verlag). Naylor, Margery Bedford (1988) ‘The Moral of the Trolley Problem’, Philosophy and Phenomenological Research, 48 (4), 711–22. Nelson, Richard R. and Winter, Sidney G. (1982) An Evolutionary Theory of Economic Change (Cambridge, MA: Belknap). Niiniluoto, Ilkka (1987) Truthlikeness (Amsterdam: Reidel). Nordlinger, Eric (1981) On the Autonomy of the Democratic State (Cambridge, MA: Harvard University Press). North, Douglass C. (1990) Institutions, Institutional Change and Economic Performance (Cambridge: Cambridge University Press). North, Douglass C. (2005) Understanding the Process of Economic Change (Princeton: Princeton University Press). Norton, Anne (2004) ‘Political Science as a Vocation’, in Ian Shapiro, Rogers M. Smith, and Tarek E. Masoud (eds), Problems and Methods in the Study of Politics (Cambridge: Cambridge University Press), 67–82. Norton, Philip (2003) ‘The Presidentialization of British Politics’, Government and Opposition, 38 (2), 274–8. Nozick, Robert (1974) Anarchy, State and Utopia (Oxford: Basil Blackwell). Nussbaum, Martha C. (2011) Creating Capabilities: The Human Development Approach (London: Harvard University Press). Oddie, Graham (1986) Likeness to Truth (Amsterdam: Reidel). Orr, Shannon K. (2005) ‘New Technology and Research: An Analysis of Internet Survey Methodology in Political Science’, PS Political Science and Politics, 33 (2), 263–7. Ostrom, Elinor (1990) Governing the Commons: The Evolution of Institutions for Collective Action (Cambridge: Cambridge University Press). Ostrom, Elinor (2001) ‘Decentralization and Development: The New Panacea’, in Keith Dowding, James Hughes, and Helen Margetts (eds), Challenges to Democracy (Houndmills: Palgrave). Ostrom, Elinor (2005) Understanding Institutional Diversity (Princeton: Princeton University Press). Otsuka, Michael (2008) ‘Double Effect, Triple Effect and the Trolley Problem: Squaring the Circle in Looping Cases’, Utilitas, 20 (1), 92–110.

References

269

Pacek, A. C. (2009) ‘Politics and Happiness: An Empirical Ledger’, in A. K. Dutt and Ben Radcliff (eds), Happiness, Economics and Politics: Towards a MultiDisciplinary Approach (Cheltenham: Edward Elgar), 231–55. Palolacci, Garielle, Chandler, Jesse, and Ipeirotis, Panagiotis (2010) ‘Running Experiments on Amazon Mechanical Turk’, Judgement and Decision Making, 5 (5), 411–19. Parkinson, John and Mansbridge, Jane (eds) (2013) The Deliberative Society (Cambridge: Cambridge University Press). Pateman, Carole (1988) The Sexual Contract (Oxford: Polity Press). Pateman, Carole and Mills, Charles W. (2007) Contract and Domination (Oxford: Polity Press). Patty, John and Penn, Elizabeth Maggie (2015) ‘Analyzing Big Data: Social Choice and Measurement’, PS: Political Science and Politics, 48 (1), 95–101. Pennings, Paul and Keman, Hans (2008) ‘The Changing Landscape of Dutch Politics since the 1970s: A Comparative Exploration’, Act Politica, 43 (2), 154–79. Persson, Torsten and Tabellini, Guido (2005) The Economic Effects of Constitutions (Cambridge, MA: MIT Press). Pettit, Philip (1997) Republicanism: A Theory of Freedom and Government (Oxford: Oxford University Press). Phillips, Anne (2000) ‘Survey Article: Feminism and Republicanism: Is This a Plausible Alliance?’, Journal of Political Philosophy, 8 (2), 279–93. Phillips, Nelson and Hardy, Cynthia (2002) Discourse Analysis: Investigating Processes of Social Construction (London: Sage). Plamenatz, John (1963) Man and Society: A Critical Examination of Some Important Social and Political Theories from Machiavelli to Marx, vol. 1 (London: Longman). Planck, Max (1949) Scientific Autobiography and Other Papers (New York: Philosophical Library). Poguntke, Thomas and Webb, Paul (2005) ‘The Presidentialization of Politics in Democratic Societies’, in Thomas Poguntke and Paul Webb (eds), The Presidentialization of Politics: A Comparative Study of Modern Democracies (Oxford: Oxford University Press). Poole, Keith and Rosenthal, Howard (1997) Congress: A Political-Economic History of Roll-Call Voting (Oxford: Oxford University Press). Popper, Karl R. (1972a) Objective Knowledge: An Evolutionary Approach (Oxford: Clarendon Press). Popper, Karl R. (1972b) The Logic of Scientific Discovery (6th impression (revised edn); London: Hutchinson). Popper, Karl R. (1974) ‘Replies to My Critics’, in Paul Arthur Schilpp (ed.), The Philosophy of Karl Popper: Library of Living Philosophers (La Salle, IL: Open Court), 961–1197. Popper, Karl R. (1983) Realism and the Aim of Science (London: Routledge). Popper, Karl R. (1989) Conjectures and Refutations: The Growth of Scientific Knowledge (revised and corrected edition edn; London: Routledge). Popper, Karl R. (1992) Unended Quest: An Intellectual Autobiography (London: Routledge). Putnam, Hilary (1973) ‘Meaning and Reference’, Journal of Philosophy, 70 (4), 699–711.

270

References

Putnam, Hilary (1985) ‘The Meaning of “Meaning”’, in his (ed.), Mind, Language and Reality: Philosophical Papers, vol. 2 (Cambridge: Cambridge University Press), 215–71. Quine, Willard van Orman (1953) From a Logical Point of View (New York: Harper and Row). Quine, Willard van Orman (1960) Word and Object (Cambridge, MA: MIT Press). Quine, Willard van Orman (1993) ‘In Praise of Observation Sentences’, Journal of Philosophy, 90 (3), 107–16. Quinn, K. M., Monroe, Burt L., Colarsi, M., Crespin, M., and Radev, D. (2010) ‘How to Analyse Political Attention with Minimal Assumptions and Costs’, American Journal of Political Science, 54 (1), 209–28. Rabinowitz, George and Macdonald, Stuart Elaine (1989) ‘A Directional Model of Issue Voting’, American Political Science Review, 83 (1), 93–121. Radcliffe-Brown, Alfred R. (1952) Structure and Function in Primitive Society: Essays and Addresses (Glencoe, IL: Free Press). Ragin, Charles (2008) Redesigning Social Inquiry: Fuzzy Sets and Beyond (Chicago: Chicago University Press). Rawls, John (1971) A Theory of Justice (Oxford: Oxford University Press). Rawls, John (2001) Collected Papers (Cambridge, MA: Harvard University Press). Rhodes, R. A. W. (2011) Everyday Life in British Government (Oxford: Oxford University Press). Richardson, J. J. and Jordan, A. G. (1979) Governing under Pressure: The Policy Process in a Post-Parliamentary Democracy (Oxford: Martin Robertson). Riker, William H. (1976) ‘The Number of Political Parties’, Comparative Politics, 9, 93–106. Riker, William H. (1982a) ‘The Two Party System and Duverger’s Law: An Essay in the History of Political Science’, American Political Science Review, 76 (4), 753–66. Riker, William H. (1982b) Liberalism against Populism: A Confrontation between the Theory of Democracy and the Theory of Social Choice (San Francisco: W. H. Freeman and Co). Riker, William H. (1986) The Art of Political Manipulation (New Haven: Yale University Press). Rorty, Richard (1980) Philosophy and the Mirror of Nature (Princeton: Princeton University Press). Rorty, Richard (2007) Philosophy as Cultural Politics: Philosophical Papers, vol. 4 (Cambridge: Cambridge University Press). Ross, Don (1998) ‘Neuroeconomics and Economic Methodology’, in John B. Davids, D. W. Hands, and Uskali Maki (eds), The Handbook of Economic Methodology (Aldershot: Edward Elgar). Ross, Don (2005) Economic Theory and Cognitive Science: Microexplanation (Cambridge, MA: MIT Press). Ross, Don (2006) ‘Evolutionary Game Theory and the Normative Theory of Institutional Design: Binmore and Behavioral Economics’, Politics, Philosophy and Economics, 5 (1), 51–79. Ross, Don (2008) ‘Two Styles of Neuroeconomics’, Economics and Philosophy, 24 (3), 473–83.

References

271

Ross, Don (2014) Philosophy of Economics (Palgrave: Macmillan). Rousseau, Jean-Jacques (1754/1984) A Discourse on Inequality, trans. Maurice Cranston (Harmondsworth: Penguin). Rubin, H. J. and Rubin, I. S. (1995) Qualitative Interviewing: The Art of Hearing Data (London: Sage). Russett, Bruce (1993) Grasping the Democratic Peace (Princeton, NJ: Princeton University Press). Sabatier, Paul A. and Jenkins-Smith, Hank C. (1999) ‘The Advocacy Coalition Framework: An Assessment’, in Paul A. Sabatier (ed.), Theories of the Policy Process (Oxford: Westview), 117–66. Sanders, David, Clarke, Harold D., Stewart, Marianne C., and Whiteley, Paul (2007) ‘Does Mode Matter for Modelling Choice? Evidence from the 2005 British Election Study’, Political Analysis, 15 (3), 257–85. Sartori, Giovanni (1970) ‘Concept Misformation in Comparative Politics’, American Political Science Review, 64 (4), 1033–53. Satterthwaite, Mark (1975) ‘Strategy Proofness and Arrow’s Conditions: Existence and Correspondence Theorems for Voting Procedures and Social Welfare Functions’, Journal of Economic Theory, 10 (2), 187–217. Sayer, Anthony (2000) Realism and Social Science (London: Sage). Scheper-Hughes, Nancy (1992) Death without Weeping: The Violence of Everyday Life in Brazil (Berkeley, CA: University of California Press). Schneiderhan, Eric (2014) ‘Why You Gotta Be So Mean’, http://chronicle.com/article/ Why-You-Gotta-Be-So-Mean/140469. Accessed 20 July 2014. Schofield, Norman (1978) ‘Instability of Simple Dynamic Games’, Review of Economic Studies, 45, 575–94. Schofield, Norman (2006) Architects of Political Change: Constitutional Quandaries and Social Choice Theory (Cambridge: Cambridge University Press). Schweller, Randall L. (1998) Deadly Imbalances: Tripolarity and Hitler’s Strategy of World Conquest (New York: Columbia University Press). Searle, John (1969) Speech Acts: An Essay on the Philosophy of Language (Cambridge: Cambridge University Press). Searle, John (1980) ‘Minds, Brains and Programs’, Behavioral and Brain Sciences, 3 (4), 417–58. Sen, Amartya (1982a) ‘Behaviour and the Concept of Preference’, in his Choice, Welfare and Measurement (Oxford: Blackwell), 54–73. Sen, Amartya (1982b) ‘Choice Functions and Revealed Preference’, in his Choice, Welfare and Measurement (Oxford: Blackwell), 41–53. Sen, Amartya (1982c) ‘Rational Fools: A Critique of the Behavioural Foundations of Economic Theory’, in his Choice, Welfare and Measurement (Oxford: Blackwell). Sen, Amartya (1993) ‘Capability and Well-Being’, in Martha Nussbaum and Amartya Sen (eds), The Quality of Life (Oxford: Oxford University Press). Sen, Amartya (1999) Development as Freedom (Oxford: Oxford University Press). Sen, Amartya (2002a) ‘Goals, Commitment, and Identity’, in his Rationality and Freedom (Cambridge, MA: Harvard University Press), 317–44. Sen, Amartya (2002b) ‘Internal Consistency of Choice’, in his Rationality and Freedom (Cambridge, MA: Harvard University Press).

272

References

Sen, Amartya (2002c) ‘Maximization and the Act of Choice’, in his Rationality and Freedom (Cambridge, MA: Harvard University Press). Shadish, William R., Cook, Thomas D., and Campbell, Donald T. (2002) Experimental and Quasi-Experimental Designs for Generalized Causal Inference (Boston, MA: Houghton Mifflin). Shapere, Dudley (1964) ‘The Structure of Scientific Revolutions’, Philosophical Review, 73 (3), 383–94. Shentalinsky, V. (1995) The KGB’s Literary Archive (London: Harvill). Shepsle, Kenneth A. and Weingast, Barry R. (1981) ‘Structure-Induced Equilibrium and Legislative Choice’, Public Choice, 37 (3), 503–19. Shepsle, Kenneth A. and Bonchek, Mark S. (1997) Analyzing Politics: Rationality, Behavior and Institutions (New York: W. W. Norton and Co.). Sil, Rudra and Katzenstein, Peter J. (2010) Beyond Paradigms: Analytic Eclecticism in the Study of World Politics (Houndmills, Basingstoke: Palgrave Macmillan). Singer, Peter (1974) ‘Sidgwick and Reflective Equilibrium’, Monist, 58 (4), 490–517. Singer, Peter (2005) ‘Ethics and Intuitions’, Journal of Ethics, 9 (4), 331–52. Skinner, Quentin (1998) Liberty before Liberalism (Cambridge: Cambridge University Press). Skinner, Quentin (2002) Visions of Politics, vol. 1: Regarding Method (Cambridge: Cambridge University Press). Soames, Scott (2003a) Philosophical Analysis in the Twentieth Century, vol. 1: The Dawn of Analysis (Princeton: Princeton University Press). Soames, Scott (2003b) Philosophical Analysis in the Twentieth Century, vol. 2: The Age of Meaning (Princeton: Princeton University Press). Soames, Scott (2010) Philosophy of Language (Princeton: Princeton University Press). Sober, Elliott (1984) The Nature of Selection: Evolutionary Theory in Philosophical Focus (Cambridge, MA: MIT Press). Sober, Elliott (1993) Philosophy of Biology (Oxford: Oxford University Press). Sperber, Dan and Wilson, Deirdre (1986) Relevance: Communication and Cognition (Oxford: Blackwell). Steinmo, Sven (2010) The Evolution of Modern States: Sweden, Japan and the Unitied States (Cambridge: Cambridge University Press). Stemplowska, Zofia (2008) ‘What Is Ideal about Ideal Theory?’, Social Theory and Practice, 34 (3), 319–40. Strawson, Peter F. and Grice, H. P. (1956) ‘In Defence of a Dogma’, Philosophical Review, 65 (2), 141–58. Sunstein, Cass R. and Thaler, Richard R. (2008) Nudge: Improving Decisions about Health, Wealth and Happiness (New Haven: Yale University Press). Swift, Adam (1999) ‘Public Opinion and Political Philosophy: The Relation between Social-Scientific and Philosophical Analyses of Distributive Justice’, Ethical Theory and Moral Practice, 2, 337–63. Taagepera, Rein (2007) Predicting Party Sizes: The Logic of Simple Electoral Systems (Oxford: Oxford University Press). Taagepera, Rein (2008) Making Social Sciences More Scientific: The Need for Predictive Models (Oxford: Oxford University Press). Taagepera, Rein and Shugart, Matthew Soberg (1989) Seats and Votes: The Effect and Determinants of Electoral Systems (New Haven: Yale).

References

273

Taleb, Nassim Nicholas (2007) The Black Swan: The Impact of the Highly Improbable (New York: Random House). Tausczik, Yia R. and Pennebaker, James W. (2010) ‘The Psychological Meaning of Words: LIWC and Computerized Text Analysis Methods’, Journal of Language and Social Psychology, 29 (1), 24–54. Tetlock, Philip E. (2005) Expert Political Judgment: How Good Is It? How Can We Know? (Princeton: Princeton University Press). Thelen, Kathleen (2004) How Institutions Evolve: The Political Economy of Skills in Germany, Britain, the United States, and Japan (Cambridge: Cambridge University Press). Thomson, Judith Jarvis (1976) ‘Killing, Letting Die, and the Trolley Problem’, Monist, 59 (2). Thomson, Judith Jarvis (1985) ‘The Trolley Problem’, Yale Law Journal, 94 (6), 1395–1415. Tichý, Paul (1974) ‘On Popper’s Definitions of Verisimilitude’, British Journal for the Philosophy of Science, 25 (1), 155–60. Tichý, Paul (1978) ‘Verisimilitude Revisited’, Synthèse, 38 (1), 175–96. Tiebout, Charles M. (1956) ‘A Pure Theory of Local Expenditures’, Journal of Political Economy, 64, 416–24. Tsebelis, George (2002) Veto Players: How Political Institutions Work (Princeton: Princeton University Press). Tully, James (ed.) (1988) Meaning and Context: Quentin Skinner and His Critics (Oxford: Polity). Udehn, Lars (1996) The Limits of Public Choice: A Sociological Critique of the Economic Theory of Politics (London: Routledge). Valentini, Laura (2012) ‘Ideal vs. Non-Ideal Theory: A Conceptual Map’, Philosophy Compass, 7 (9), 654–64. Van Biezen, Ingrid and Poguntke, Thomas (2014) ‘The Decline of Membership-Based Politics’, Party Politics, 20 (2), 205–16. Van Evera, Stephen (1997) Guide to Methods for Students of Political Science (Ithaca, NY: Cornell University Press). Van Inwagen, Peter (1997) ‘Materialism and the Psychological-Continuity Account of Personal Identity’, in J. Tomberlin (ed.), Philosophical Perspectives, 11: Mind, Causation and World (Oxford: Blackwell). Von Wright, Georg Henrik (1971) Explanation and Understanding (Ithaca, NY: Cornell University Press). Waldner, David (2012) ‘Process Tracing and Causal Mechanism’, in Harold Kincaid (ed.), Oxford Handbook of the Philosophy of Social Science (Oxford: Oxford University Press). Walgrave, Stefaan and Varone, Frédéric (2008) ‘Punctuated Equilibrium and AgendaSetting: Bringing Parties Back in: Policy Change after the Dutroux Crisis in Belgium’, Governance, 21 (3), 365–95. Walgrave, Stefaan, Varone, Frédéric, and Dumont, Patrick (2006) ‘Policy with or without Parties? A Comparative Analysis of Policy Priorities and Policy Change in Belgium, 1991 to 2000’, Journal of European Public Policy, 13 (7), 1021–38. Webb, Paul, Poguntke, Thomas, and Kolodny, Robin (2011) ‘The Presidentialization of Party Leadership? Evaluating Party Leadership and Party Government in the

274

References

World’, SSRN eLibrary, http://ssrn.com/abstract=1915657. Accessed November 2012. Weber, Max (1918) ‘Science as a Vocation’, http://www.wisdom.weizmann.ac.il/~oded/ X/WeberScienceVocation.pdf, Accessed 30 August 2015. Weber, Max (1919) ‘Politics as a Vocation’, http://www.sscnet.ucla.edu/polisci/ethos/ Weber-vocation.pdf, Accessed 30 August 2015. Weingast, Barry and Marshall, William (1988) ‘The Industrial Organization of Congress’, Journal of Political Economy, 96, 132–63. Wendt, Alexander (1999) Social Theory and International Politics (Cambridge: Cambridge University Press). Wendt, Alexander and Shapiro, Ian (1992) ‘The Difference Realism Makes: Social Science and the Politics of Consent’, Politics and Society, 20 (2), 197–223. Wengraf, Tom (2001) Qualitative Research Interviewing: Biographic Narrative and Semi-Structured Methods (London: Sage). Wildavsky, Aaron (1963) ‘A Methodological Critique of Duverger’s Political Parties’, in Harry Eckstein and David Apter (eds), Comparative Politics: A Reader (New York: Free Press). Wilkinson, S. (2004) ‘Focus Groups: A Feminist Method’, in S. N. Hesse-Biber and M. L. Yaiser (eds), Feminist Perspectives on Social Research (New York: Oxford University Press), 271–95. Wilks, Steven and Wright, Maurice (1987) Comparative Government–Industry Relations (Oxford: Clarendon Press). Williams, Bernard (1973) ‘A Critique of Utilitarianism’, in J. C. C. Smart and Bernard Williams (eds), Utilitarianism: For and Against (Cambridge: Cambridge University Press), 77–150. Williamson, Timothy (2007) The Philosophy of Philosophy (Oxford: Blackwell). Wodak, R. and Meyer, M. (eds) (2001) Methods of Critical Discourse Analysis (London: Sage). Woodward, James and Hitchcock, Christopher (2003) ‘Explanatory Generalizations, Part I: A Counterfactual Account’, Noûs, 37 (1), 1–24. Wright, Richard (1985) ‘Causation in Tort Law’, California Law Review, 73 (6), 1735–828. Wright, Richard (1988) ‘Causation, Responsibility, Risk, Probability, Naked Statistics, and Proof: Pruning the Bramble Bush by Clarifying the Concepts’, Iowa Law Review, 73 (5), 1001–77. Wright, Richard (2001) ‘Once More into the Bramble Bush: Duty, Causal Responsibility, and the Extent of Legal Responsibility’, Vanderbilt Law Review, 54 (3), 1071–132. Zodrow, G. R. (ed.) (1983) Modelling Local Political Economy: The Tiebout Model after Twenty-Five Years (New York: Academic Press).

Index

Achinstein, Peter 36–7, 45–50 ad hominem arguments 60 affirming the consequent 112–13 Alex the lion 52–3, 62 Amnesty International 115–16 analytic–synthetic distinction 38–9 anti-realism 10–14 a posteriori 39–43, 190, 205 a priori 18, 34, 39–44, 190 Barry, Brian 223, 224, 252 Bayesian 104, 119, 131–2, 166, 170, 244–50 behaviouralism 23, 69, 78, 171 belief 116, 170, 244–50 big data 68, 102–3, 170, 178–81 Binmore, Ken 27, 32 BMR 60–1 ‘boo–hurrah theory’ 18, 21, 34 Bush, George 50 cabinet ministers 16, 38, 53–5, 185–6, 188, 202–6, 247 Cambridge school 219–23 Cameron, David 50–1 Carter, Ian 194–6 case studies 5, 51–2, 59, 66–7, 80, 95–9, 105, 112–16, 143–6, 151–9, 166 crucial 113–16 cause, causation 6, 42, 57–60, 94–5, 133–59 as narrative 10, 23, 40, 71, 75, 80, 111, 134–8 background conditions 98–9, 133, 138–41, 158, 162–3, 173 binary oppositions 138 bump-bump 67, 147–9, 155 but for (BF) 6, 63–4, 133, 138–42, 149

constant conjunction 142–3, 157–8 correlations 146–8 counterfactual 142–5 difference-making 149 distal-correlational 148–9 effects of 137–8, 162–3 fundamental problem of 143 inferences 40, 68, 84–6, 88, 139, 143, 150–2, 161, 188 INUS 139–40 NESS 140–1 Neyman–Rubin 150, 158 of effects 137–8, 162–3 probabilistic 6, 43–9, 63–4, 133, 138–42, 147–8 production 149 proximal-serial 148–9 proximate 50, 52–5, 133, 163, 105 psychological 136–7 reasons as 57 regularity 142–5, 157–8 specification, see specification problem centre of gravity 29, 152 Clarke and Primo 46, 81–4, 88, 112–13 classification, principles of 208–11 motivating 208 sorting 208 coherentism 34–5 Comte, Auguste 12, 17 conceivability 41 scientific 41, 144–6 concepts 37–43, 189–212 change 124, 192–3, 200 contestability 212, 227 development 200 existence 203–6 gerrymandering 193

275

276

Index

concepts (Continued) independence 201 non-normative 194–9 primitive 191–3 simplicity 199–200 subdivision of 192–3 subscript strategy 212, 227 theoretical 38 theory-laden 17, 128–30, 191 threshold 202–3 value-freeness 194–6 value-neutrality 195–8 conceptual analysis 42, 189-212, 224–8 deductive 189–90, 207 inductive 207–8 methodological criteria 225–6 normative criteria 225–6 semantic criteria 225–6 stretching 198 confirmation 99, 103–4, 106–27 paradox 116–27 conjecture 70 constructivism 3, 11–12, 27–30, 160–1 in international relations 4, 27 contingency 38–42 contractual tradition 219 conventionalism 10, 12, 18–19 corroboration 103–27 critical realism 10, 24–6 critical theory 10, 21–2, 78 cumulative research 98–100, 123 Dawes, Robin 95–7, 144–6, 149 deductive inference, see inferences democratic peace 42, 68, 72, 117–18, 144, 155–6, 209 Dennett, Daniel 29, 43, 116, 163, 170, 223, 228, 239, 244–5 description, see explanation descriptive inference 6, 40, 58–60, 90, 99 Diamond, Jared 43 dictators 6 directional party models 128–30 discourse theory 70, 76–7, 186–8, 222–3

Dreben, Burton 216 Duhem–Quine, see under Quine Duverger, Maurice 99–100, 123 Eddington, Arthur 113–14 Einstein, Albert 113–14, 229–30 empirical content 86, 116, 120–2, 126 empiricism 10, 15–18 epistemology 5, 26, 40, 161 epistemological commitments 2, 69, 160–3, 249 equifinality 140, 154–7 equivalence sets 64–5 essentialism 10, 12, 19–21 evidence, theory-laden 117 evolution 5, 27, 50, 180–1 experiment 6, 68, 96–7, 109, 137–9, 143–4, 149–54, 158, 162, 165–7, 176–80, 185–7 experiment effects 177, 185 external validity of 177 experts 65, 98–9, 236–7, 251 explanation 36–67 barometer 47 black box 64 causal 42 description as 36, 37–43, 55–60, 85, 245–50 DN model 36–7, 45–50, 107, 116–27, 146 functional 57 IS model 45–50 leverage 86–7 mechanisms as, see mechanisms model of 45–50 narrative 94–6, 111 NES requirement 45–6 non-rival 57, 89–92 prediction 43–4, 55–6, 70–1, 79–88, 102–32 proximate, see proximate explanation psychological phenomenon 48, 95–6 relevance of 60 structural 64–5 token, see token type, see type

Index typologies as 208–11 ultimate, see ultimate explanation fallacious reasoning 59–60 falsifiability 104–7, 115–28 falsification (falsified) 73–9 family resemblances 190 feminist 68, 69 Feyerabend, Paul 73 Fisher, R. A. 118–19, 131–2, 166 Foucault, Michel 2, 222–3 foundationalism 34–5 framework, see model, non-formal Friedman, Milton 81–3

277

hypocrite 2, 23, 96 hypotheses 14–17, 66–70, 76–7, 79–88, 92–3, 102–32, 149, 152–62, 167–72, 182, 183, 187, 197, 209, 226, 245

70

game theory 74, 78 generalizations 17, 35, 41, 48, 58–67, 107–27 empirical 45–50, 68, 70–1, 99–100, 102, 107–27 invariant 17, 41–2, 57–8, 61–5, 100, 102, 106, 114, 117–18, 144– 7, 204–5 law-like 17, 45, 57, 114, 117–18, 144–6 George and Bennett 146–8, 152–7, 209 Gibbon, Edward 52, 135–6 Gillard, Julia 50 good argument 59–60 Gould, Stephen Jay 96–7 granularity 102, 141–8 Habermas, Jürgen 21, 76 Harry Potter 56–7 health reform 50–1 Hempel, Carl 2, 36, 45–50, 116–27 Hirschman, Albert O. 88–9, 92 history/historians of ideas 7, 52, 80–2, 95, 97–8, 216–23 Hitchcock, Christopher 61, 146 Hitler, Adolf 7, 39 Hoggart, Simon 93 holism 35, 38–43 Hume, David 26–7, 41, 103–7, 138, 142, 216

idealism 10, 14–15 ideal theory 215, 241–2 ideal type 206 identity statement 6, 39–43, 57, 84, 95, 136–7, 145–6, induction, problem of 103–5, 107–16 inferences causal 68 deductive 58–60 inductive 68 individualism 35 instrumentalism 10, 18–19 interpretivism 10, 23–4, 184–6, 216–23 interesting questions 245 intuition pumps 215, 231–40 Bridge 233–7, 238–9 outlier problem 237–8 Shoot the One? 232 Surgeon 232, 237–9 Trolley 233–7, 239 intuitions 215, 224–6, 228, 232–7, 242 fast thinking 236–7 invariant generalizations, see generalizations inversion strategy 61, 92–3, 128, 238–41 isms 3–6, 9–35 Jiabo, Wen 156 Kant, Immanuel 15, 18, 28–32, 196 Kevlar vest 158 King, Gary (King et al.) 59, 85–6, 137–8, 153–5, 162 Kripke , Saul 38–45, 62–3, 190, 198, 204–6, 222–3 Kuhn, Thomas 72–5

278

Index

Lakatos, Imre 72, 119–23 law-like generalizations, see generalizations; laws laws 17, 41–3, 55–8, 60–7, 103–6 Duverger’s 99, 123 of excluded middle 12–14 of nature 17, 41–2, 105 of non-contradiction 12–14 Lijphart, Arend 114–15 LIWC 187–8 Marty the Zebra 52–3, 62–4 Masterman, Margaret 73–5 Marxism 21, 68, 69 McCloskey, Deirdre 97–8, 113–14, 247 meaning extension 37–43 historical (causal) theory 38–45, 190, 198, 204–6, 222 reference 37–43 rigid designators 39–42 verification theory 34 mechanisms 49, 58, 60–7, 70–1, 94–8, 111–12, 117, 133–59 memes 180 Merkel, Angela 156 metaphysical necessity, see necessity methodological individualism, see individualism methods 160–88 agent-based modelling 173–4 archival 181–3 behavioural 174–81 content analysis 187–8 discourse analysis 186–8 ethnography 184–6 evolutionary 180–1 experimental 176–8, 179–80 focus groups 184 genetic 181 institutional-structural 170–4 interpretative 181–8 interviewing 183 mathematical-formal 171–7 participant observation 185–6

shadowing 185–6 survey research 174–6 text analysis 187–8, 202–3 mirages 29 moa 60–2 models 70–1, 79–101, 102–2 false assumptions 81–3 forecasting 82 formal 70–1, 72, 73–6, 79–88, 94–6, 100–1, 102–32 growth machine 90–2 non-formal 70–1, 84–7, 88–100, 131 rival 76, 86–8 Rhodes non-formal 89–92 spatial 128–30 Tiebout 120 true or false 81–4 Moore, G. E. 27 Morton, Rebecca 84–5 narrative fallacy, see specification problem naturalism 10, 26–7 natural kinds 38–40, 203–6 natural selection 6 necessary and sufficient conditions 189–90 necessity a posteriori 40 logical 42 metaphysical 39, 190 natural 41 nominalism 10, 14–15 normative commitments theory 16–18 nouns 37 Obama, Barack 50–2, 156 objective 164–5 objectivism 10, 19–21 obscurity 23 ontological commitments 2, 69, 160–3, 249 opinion 21, 116, 170, 244–50 organizing perspective 70–1, 70–9 Ostrom, Elinor 92–3

Index

279

paradigms 72–5 Planck, Max 252 pluralism 68–9 ismistic 5 policy agendas 210–11 policy-oriented research 169–70 political argument 224 political journalism 246–9 political philosophy 41, 213–42 political science 243–52 profession of 250–2 Popper, Karl 2, 3, 5, 6, 17, 60–2, 69–70, 76, 86–5, 104, 106, 115–28 positivism 3, 15–18 postmodernism 10, 23–4, 78 power elite theory 69 pragmatism 10, 18–19 prediction, see explanation; see also hypothesis presidentialization 248 presidents 50, 65–7, 123, 248 prime ministers 20, 50, 66, 248 principle of bivalence, see law of excluded middle principle of charity 217 private goods 205–6 process tracing 6, 64, 148–9, 151–7 diagnostic 152–3, 155–7, 159 hypothetical or conjectural 152–3, 159 proximate explanation 50, 52–5, 133, 163 proximity party models 128–30 Ptolemy 122–3 public goods 205–6

rational choice theory 68, 77–9 rational turnout 87 Rawls, John 72, 196, 214, 223, 225, 233–5 real patterns 43, 56–7 realism scientific 2, 10–14, 160–1 in ethics 27, 30–2 in International Relations 4 reasons 56–7 reductionism 33, 153–4 reflective equilibrium 196, 225, 233–5, 238–40, 242 relativism 2, 10–14, 72–5, 160 in ethics 27, 30–2 replication 166–9, 182–3 revealed preference 58 Rhodes model, see models rigid designators, see meaning rocket science 249–50

qualitative research 59–60, 96–7, 105, 151–7, 163–6 quantitative research 59–60, 96–7, 163–6 Quine, Willard van Orman 3, 6, 18, 19, 34, 38–9, 43, 124, 201, 216 Duhem–Quine 88, 103–4, 112, 119–20, 124, 128–9, 135, 140, 201, 217

teaching 250–1 theory (general) 68–101 complexity 68 explanatory 105 see also models normative 72 perspectival 70–9, 72, 102 thought experiments 228–31, 239 Joining the Objects 229 Stag Hunt 230–1

Sarkozy, Nicolas 156 Sartori, Giovanni 198, 206 scientific conceivability, see conceivability semi-presidentialism 65–7, 123 Severus Snape 56–7 Skinner, Quentin 219–22 Soames, Scott 204–5 specification problem 80, 95–7, 112–13, 143–6, 158, 166 subjective 164–5 subjectivism 20–1 surface phenomenon 189, 202, 245–50 synthetic control 165–6

280

Index

thought experiments (Continued) Prisoners’ Dilemma 230–1 Train 229–30 Tiebout, Charles 120–1, 124–5 token 50–5, 149–50, 155–6, 163 transcendental realism 15 trinity of God 14 trireme 221 truth correspondence theory of 34 necessary 38–40 truthlikeness 125–7 verisimilitude 124–7 Tsebelis, George 65–6, 150 TV kicking 151

type

50–5, 149–50, 155–7, 163, 199–200 typologies 65, 208–11 ultimate explanation 52–5, 133, 163, 199–200 verbal disputes 90, 101 verisimilitude, see truth veto-player model 65–7, 82, 94–5, 149–50, 190–1, 199 water 29, 40–1, 144–5, 190, 204–5 Wittgenstein, Ludwig 190 Woodward, James 61