Voter Turnout and the Dynamics of Electoral Competition in Established Democracies since 1945

This page intentionally left blank Voter Turnout and the Dynamics of Electoral Competition in Established Democracies...

Author: Mark N. Franklin

11 downloads 528 Views 1MB Size Report

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
Report copyright / DMCA form

DOWNLOAD PDF

This page intentionally left blank

Voter Turnout and the Dynamics of Electoral Competition in Established Democracies since 1945 Voting is a habit. People learn the habit of voting, or not, based on experience in their first few elections. Elections that do not stimulate high turnout among young adults leave a “footprint” of low turnout in the age structure of the electorate as many individuals who were new at those elections fail to vote at subsequent elections. Elections that stimulate high turnout leave a high turnout footprint. So a country’s turnout history provides a baseline for current turnout that is largely set, except for young adults. This baseline shifts as older generations leave the electorate and as changes in political and institutional circumstances affect the turnout of new generations. Among the changes that have affected turnout in recent years, the lowering of the voting age in most established democracies has been particularly important in creating a low turnout footprint that has grown with each election. Mark N. Franklin is the John R. Reitemeyer Professor of International Politics at Trinity College Connecticut and past Chair of the Political Science Department there. He was previously a John and Rebecca Moores Professor at the University of Houston, Texas, where he was also Director of Graduate Studies in the Department of Political Science. Before that he taught for twenty years at the University of Strathclyde, Scotland. Dr. Franklin has been a Principal Investigator of the European Election Studies Project since 1987; is the author or co-author of seven books, including Choosing Europe? (1996), Electoral Change (1992), and The Decline of Class Voting in Britain (1985); and is the editor or coeditor of five more, including The Future of Election Studies (2002) and Parliamentary Questions (1993).

Voter Turnout and the Dynamics of Electoral Competition in Established Democracies since 1945

MARK N. FRANKLIN Trinity College Connecticut

with assistance from Cees van der Eijk, Diana Evans, Michael Fotos, Wolfgang Hirczy de Mino, Michael Marsh, and Bernard Wessels

cambridge university press Cambridge, New York, Melbourne, Madrid, Cape Town, Singapore, São Paulo Cambridge University Press The Edinburgh Building, Cambridge cb2 2ru, UK Published in the United States of America by Cambridge University Press, New York www.cambridge.org Information on this title: www.cambridge.org/9780521833646 © Mark N. Franklin 2004 This publication is in copyright. Subject to statutory exception and to the provision of relevant collective licensing agreements, no reproduction of any part may take place without the written permission of Cambridge University Press. First published in print format 2004 isbn-13 isbn-10

978-0-511-21040-2 eBook (EBL) 0-511-21217-8 eBook (EBL)

isbn-13 isbn-10

978-0-521-83364-6 hardback 0-521-83364-7 hardback

isbn-13 isbn-10

978-0-521-54147-3 paperback 0-521-54147-6 paperback

Cambridge University Press has no responsibility for the persistence or accuracy of urls for external or third-party internet websites referred to in this publication, and does not guarantee that any content on such websites is, or will remain, accurate or appropriate.

This book is for Diane

Contents

Figures Tables

page viii ix

Preface The Authors

xi xv

Introduction 1. Confronting the Puzzles of Voter Turnout 2. A New Approach to the Calculus of Voting

1 9 37

3. The Role of Generational Replacement in Turnout Change 4. Rational Responses to Electoral Competition 5. Explaining Turnout Change in Twenty-Two Countries 6. The Character of Elections and the Individual Citizen

59 91 119 151

7. Understanding Turnout Decline 8. The Turnout Puzzles Revisited

171 201

appendices A. The Surveys Employed in This Book B. Aggregate Data for Established Democracies, 1945–1999 C. Supplementary Findings

225 231 237

Bibliography Author Index Subject Index

251 263 267 vii

Figures

1.1 Average Turnout in Decades for Thirty-Nine Countries since 1945 3.1 Turnout at Legislative Elections in Six Countries, 1960–1999 3.2 Expected Long-Term Evolution of Turnout Following an Extension of the Franchise to Younger Voters 3.3 Expected Long-Term Evolution of Turnout Following Extension of the Franchise to a Previously Disenfranchised Group 3.4 Typical Expected Long-Term Evolution of Turnout Following a Change in Election Law 4.1 The Evolution of Turnout in Malta and Switzerland, 1947–1999 4.2 Changing Turnout in Presidential Year Congressional Elections, 1840–1988, Compared with Changes in Margin of Victory 5.1 Average Turnout in Five-Year Periods for Twenty-Two Countries, 1945–1999 7.1 Predicted and Actual Average Turnout in Five-Year Periods for Twenty-Two Countries, 1950s–1990s C.1 Expected and Actual Turnouts for Austria to Germany C.2 Expected and Actual Turnouts for Iceland to The Netherlands C.3 Expected and Actual Turnouts for New Zealand to the United States

viii

page 10 67 82

83 84 93

110 120 172 238 240 242

Tables

1.1 Mean, Standard Deviation, Trend, and Standard Error of Estimate for Turnout in Twenty-Two Countries, 1945–1999 2.1 Three Levels of Political Context and the Types of Information They Provide 3.1 Average Turnout by Cohort and Year for British Elections, 1964–1997 (percentages) 3.2 Average Turnout by Cohort for Six Countries, Earliest and Most Recent Cohorts 3.3 Mean Absolute Cohort Turnout Differences in Six Countries 3.4 Regression Findings (Panel Corrected Standard Errors in Parentheses) 4.1 Effects of Divided Government and Distance Between Parties on Turnout (Standard Errors in Parentheses) 4.2 Correlations with Turnout at U.S. Presidential Elections, 1840–1988 4.3 Explaining House and Presidential Turnout at U.S. Midterm and Presidential Elections, 1840–1988 5.1 Three Models Explaining Turnout in Twenty-Two Countries, 1945–1999, Using AR1 Correction for Autocorrelation (Panel Corrected Standard Errors in Parentheses) 5.2 Three Models Explaining Turnout in Twenty-Two Countries, 1945–1999, Using Fixed-Effects Regression (Standard Errors in Parentheses)

page 11 44 69 72 73 77 102 106 109

133

135

ix

x

Tables

5.3 Effects of Further Aspects of the Character of Elections When Added, One at a Time, to Model E (Standard Errors in Parentheses) 6.1 Individual Level Variables for Explaining Electoral Participation in Germany 6.2 Differences in Proportions Voting that Result from Shifts from Minimum to Maximum in Independent Variables When Other Such Variables Are Held at Their Mean Values 6.3 Differences in Proportions Voting that Result from Shifts from Minimum to Maximum in Independent Variables with Other Variables Held at Their Mean Values, Separated Interactions 6.4 Effects on Campaign Variables of Other Independent Variables in Tables 6.2 and 6.3 from OLS Regression Analysis (Standard Errors in Parentheses) 6.5 Direct, Indirect (via Campaign Variables), and Total Effects of Different Categories of Variables for New Cohorts 7.1 Values of Dependent and Independent Variables, with Actual and Estimated Turnout Changes 7.2 Tracking of Turnout by Country Using Predictions from Chapter 5’s Model E 7.3 Number of Countries for Which Turnout Moved Up or Down Between the 1960s and 1990s due to Changes in Particular Independent Variables 7.4 Comparing Findings Without Cohort Effects with Findings that Take Account of Cohort Effects Using Fixed-Effects Regression (Standard Errors in Parentheses) C.1 Cohort Effects in Six Countries (for Table 3.3 in Chapter 3) C.2 Logistic Regression Results for Table 6.2, Model C C.3 Logistic Regression Results for Table 6.3, Model E C.4 Predicated and Actual Turnout Change, 1960s to 1990s, Australia to Israel (for Table 7.3) C.5 Predicated and Actual Turnout Change, 1960s to 1990s, Italy to United States (for Table 7.3)

143 154

156

159

163

165 174 181

183

187 244 246 247 249 250

Preface

Human beings, it has been suggested, have a “puzzle instinct” (Danesi 2002) – a fascination with puzzles and an aptitude for solving them. Academic research in most disciplines is all about puzzle-solving, but political science is perhaps unusual in being home to a great many puzzles that are of interest beyond the walls of academe. Bernard Grofman recently edited a book with the title Political Science as Puzzle-Solving (2001) whose premise was that interesting puzzles lead us to topics where the tools and skills of political science can be brought to bear, teaching us useful things about the world. If the vexing questions of political science can be regarded as puzzles, the particular topic of voter turnout could be called the “grand enchilada” of puzzles in political science. As we will see in Chapter 1, almost everything about voter turnout is puzzling, from the question of why anyone bothers to vote at all to the question of why certain variables appear to explain voter turnout in some circumstances but not in others. I became interested in these puzzles in the early 1990s as a byproduct of my interest in elections to the European Parliament. Turnout in these elections is very low despite the fact that they occur in countries (of the European Union) where turnout levels in national elections are generally high. The attempt to explain low turnout in European Parliament elections in terms that would be generalizable – an explanation that would say more than that these elections are different or exceptional – led me to an interest in turnout in general, reinforced by xi

xii

Preface

the fact that I found myself supervising a dissertation at the University of Houston on the topic of voter turnout. These two experiences, which coincided in time, brought home to me the fact that this is a vitally important area of research that is very poorly understood. As my interest in voter turnout moved beyond European Parliament elections to elections in general, I became aware that the United States (to which I had moved in 1989) was a country in which voter turnout presented a particular puzzle. Why, in the world’s oldest democracy (or perhaps second-oldest, depending on how you count), was turnout at national elections so abysmally low? When I became aware that the other country, Switzerland, which vies for the title of “world’s oldest democracy,” was also one in which turnout was abysmally low, I decided that this phenomenon was definitely worthy of extended study. These three explanations for my interest in voter turnout also provide the start of an explanation for why this book has so many coauthors. Some were my co-investigators on the European Elections Study project. Another is the scholar who was the student writing that dissertation. Still others became involved when my interest in voter turnout moved beyond my areas of expertise (either in terms of countries I know well or of political science subfields in which I am well versed). I have always believed in making use of colleagues when I find myself out of my depth, and the number of co-authors who share credit for this book attests to the frequency with which I was confronted by my own ignorance when addressing the voter turnout puzzles. The contributions made by those with whom I share this book’s authorship vary greatly, from helping me to sort out the vagaries of German and other survey data (Bernard Wessels) through the coauthorship of a conference paper that is the basis of a chapter that I could not have written alone (Mike Fotos), to long discussions and years of collaboration on multiple projects central and peripheral to this one (Cees van der Eijk, Diana Evans, Wolfgang Hirczy, and Michael Marsh). Where these contributors share authorship of particular passages, sections, or chapters, this is acknowledged in footnotes. I am grateful for their willingness to share the fruit of our joint labors. The effects of these individuals on the book more generally is acknowledged here, along with my debts to other scholars who have labored in the vineyard of voter turnout studies. These are too numerous to list exhaustively, and I am sure they know who they are, but I must mention

Preface

xiii

Vanessa Baird, André Blais, Bernie Grofman, Peter Hall, Hans-Dieter Klingemann, Arend Lijphart, Patrick Lyons, Johan Martinsson, Frank Niles, Pippa Norris, Thomas Reilly, Peter Selb, James Tilley, and Jack Vowles. These are colleagues who helped me with analysis or interpretation, provided me with data, and/or commented on all or portions of this book. I also owe a debt of gratitude to Heather McMullen and John A. Collins, of Harvard College Libraries, who helped me to track down union membership figures and other hard-to-find social statistics. The research was supported by a generous Fellowship from the John Simon Guggenheim Memorial Foundation (topped up by a faculty research grant from Trinity College Connecticut), by research funds from the Reitemeyer endowment at Trinity College, and by the hospitality of both the Harvard Center for European Studies during the academic year 2001–2002 and the Wissenschaftszentrum, Berlin, during the spring of 2002. For suggesting the title, checking the references, listening patiently as I worked through my ideas, and helping in countless other ways (not least by tolerating months of preoccupation verging on obsession), the debt I owe my wife, Diane, cannot possibly be adequately expressed. Though all of these people and institutions deserve credit and thanks, none of them bears responsibility for remaining errors and infelicities, which no doubt are numerous despite all my efforts and theirs. Mark N. Franklin Cambridge, Massachusetts

The Authors

Mark Franklin is the Reitemeyer Professor of International Politics at Trinity College Connecticut and Associated Professor at the Amsterdam School of Communications Research, University of Amsterdam. Cees van der Eijk is Professor of Political Science at the Amsterdam School of Communications Research, University of Amsterdam. Diana Evans is Professor of Political Science at Trinity College Connecticut. Michael Fotos is Adjunct Professor of Political Science at Trinity College Connecticut. Wolfgang Hirczy de Mino is Adjunct Professor of Political Science at the University of Houston. Michael Marsh is Associate Professor of Political Science at Trinity College, Dublin. Bernard Wessels is Director of Research at the Wissenschaftzentrum, Berlin.

xv

Introduction

Voter turnout regularly makes news. Seemingly, whenever an election is held, the question comes up: How many people voted? Sometimes the turnout is unexpectedly high. Commentators were amazed at how many people stood for hours in the hot African sun waiting to vote in South Africa’s first truly free and universal election. But it is rare to see stories about higher than expected turnout. More often we see stories that express concern at the fact that turnout is lower than expected – so much more often, indeed, that one might be forgiven for supposing that low or declining turnout was ubiquitous in contemporary democratic elections. One prosaic reason for this is the newsworthiness of turnout decline. Stable turnout is not news. Moderately increased turnout is not news. Low or declining turnout is newsworthy. So commentators draw attention to the level of turnout mainly when it is down. How many people are aware that turnout was higher at the American presidential election of November 2000 than at the previous presidential election, in 1996? Of course, the level of turnout in the more recent of those elections was overshadowed by its other, more newsworthy, features – butterfly ballots and such. But the lack of press attention given to increased turnout, when it occurs, is one reason why we have this general perception that turnout everywhere is in decline. What is true is that, whenever turnout is down, the decline makes news. The reason why declining turnout makes news seems to be because it allows commentators to pontificate about the dire state of democracy in the country concerned. Low electoral turnout is often considered 1

2

Voter Turnout and Dynamics of Electoral Competition

to be bad for democracy, whether inherently or because it calls legitimacy into question or by suggesting a lack of representation of certain groups and inegalitarian policies (Piven and Cloward 1988, 2000; Teixeira 1992; Patterson 2002; Wattenberg 2002). Above all, low turnout appears to be seen by commentators as calling into question the civic-mindedness of a country’s citizens and their commitment to democratic norms and duties. Indeed, falling turnout is often seen as a mark of disengagement, if not of actual disaffection (Teixeira 1992; Dalton 1999; Norris 1999). That turnout should be a mark of civic virtue is not self-evident. In the early days of empirical social enquiry, those who studied turnout (Merriam and Gosnell 1924; Gosnell 1927; Boechel 1928; Tingsten 1937) took it for granted that turnout would be higher when an election’s outcome hung in the balance and when “issues of vital concern are presented” (Boechel 1928:517). Seen in this light, low voter turnout would be blamed on parties and politicians for failing to present issues of vital concern – or for failing to present such issues in an election where the outcome was seen as likely to determine the course of public policy. Thus low voter turnout would have been blamed on the character of the election, not on the characters of those who failed to vote. It was the rational choice approach to explaining political behavior that changed our ideas about why people vote. Writing in 1968, Riker and Ordeshook, elaborating on the ideas of Downs (1957), pointed out that the chances of any one vote affecting the outcome of an election for nationwide public office were virtually zero – even in a close race. For this reason, they went on to argue, people (unless they had quite unreasonable expectations about the importance of their vote) could not be voting with the purpose of benefiting from the outcome (Riker and Ordeshook 1968:28). Whatever the benefits any individual might receive as a consequence of policies adopted or blocked by an election’s outcome, those benefits would be enjoyed whether the individual voted or not. So the only rational reason for an individual to vote would be to gain nonmaterial benefits, such as the satisfaction of pulling one’s weight and other aspects of civic virtue. In the years that followed the publication of this argument, those who studied electoral participation seldom paid attention to benefits that might accrue to voters from the outcome of any specific electoral contest. Instead they focused on voting as a habit that people

Introduction

3

learned during their formative years – a learning experience dominated by education and social status.1 In their seminal work Participation in America, Verba and Nie (1972) built their explanation of electoral participation on what they called a “baseline model” consisting of income, occupation, and education. This baseline model (later renamed the “resource model”) dominated explanations of individual turnout decisions in the United States and elsewhere until the present time (Wolfinger and Rosenstone 1980; Parry, Moiser, and Day 1992; Rosenstone and Hansen 1993; Verba, Schlozman and Brady 1995, Wernli 2001). In recent research, the resource model has been joined by a “mobilization model” that takes into account the fact that people also vote because they are mobilized to do so by parties, interest groups, and candidates (Rosenstone and Hansen 1993; Verba, Schlozman, and Brady 1995). But in all of this work the focus for explaining why people vote is centered on the individual and things that happen to the individual rather than (as in earlier research) on the election and things about the election.2 In the light of this focus, it is not surprising that commentators should take low or declining turnout to be a reflection on the capacity and motivation of individual citizens. Yet the idea that declining turnout is due largely to “something about citizens” runs counter to some very obvious facts. In the first place, turnout varies from election to election both up and down; and while it is possible to imagine secular trends in civic virtue, it is hard to imagine what would cause it to fluctuate both up and down from election to election. Moreover, if civic virtue drives turnout, why does virtue have more effect in some elections (U.S. presidential elections, for example) than in others (U.S. midterm elections, for example)? The same citizens vote in a presidential election who fail to vote in the following midterm election. Presumably it is not something about those citizens that makes them more likely to vote in some elections than in others. In the second place, turnout varies enormously between countries. There are countries (like Australia, Belgium, and Malta) where virtually

1 2

The dominance of education and social status has not been found ubiquitously outside the United States (see Chapter 1). The mobilization model, of course, serves to some extent to bridge the gap between the two (see Chapter 1).

4


everyone votes. If high turnout is due to “something about citizens,” then how come the citizens of these three countries are so different from the citizens of the United States and Switzerland (two countries where turnout in national elections is particularly low)? It is true that in Australia and Belgium voting is compulsory, but the law that makes abstention illegal does not affect the character of those countries’ citizens. On the contrary, what it affects is the character of the elections in those countries; and, if compulsory voting can affect voter turnout, then perhaps other things about the character of elections can also affect voter turnout. Voting is not compulsory in Malta, so (unless we want to assume that Maltese citizens are uniquely civic-minded) it seems clear that there must be at least one other feature of elections’ character that can bring about universal turnout – or perhaps a combination of several features. The purpose of this book is to take a closer look at these and other puzzles that bedevil the study of voter turnout. Its argument is that Riker and Ordeshook’s view of elections and electoral behavior was incomplete (cf. Whiteley 1995). When we allow ourselves to suppose that some citizens might be motivated to vote primarily by a self-interested desire to gain the political benefits that victory by one party or candidate promises to bestow on them (or to prevent those benefits from going to someone else), then both of the just-mentioned puzzles resolve themselves. Indeed, it is by demonstrating that these (and other) puzzles do resolve themselves that we prove (insofar as such things can be proven) that the approach to understanding voter turnout set out in this book improves on the conventional wisdom. The importance of the book lies in its ability to establish how voter turnout serves as an indicator of the health of a democracy and to enumerate the conditions that can result in low voter turnout. Most of the commentators who decry declining turnout seem to assume that the development is consequential – either that falling turnout is bad in itself or that it is an indicator of bad things happening to the society. However, falling turnout might be incidental to deliberate changes made in a country’s electoral system or other political arrangements. Thus, the abolition of compulsory voting would result in lower turnout, but this would have been an anticipated consequence of a reform that was enacted nevertheless. Lower turnout might also be accidental, if an election occurred on a day of particularly bad weather (for example).

Introduction

5

Accidental developments can be expected to reverse themselves in due course. The weather is a clear factor of this kind, but other developments may be hard to place in terms of whether they are accidental, incidental, or consequential. A major objective of this book is to distinguish accidental and incidental developments from consequential ones and to make suggestions for reforms that might alleviate the consequential problems that have led to low or declining turnout in certain countries.3 In Chapter 2 we will revisit the assumptions made by Riker and Ordeshook and show how they can be elaborated in the light of more recent political science research. This will enable us to build a model of voter turnout that rests on a more elaborate set of assumptions – a model whose assumptions and implications are the real subject of this book. But first, in Chapter 1, we describe the various puzzles in greater detail and set out our strategy for solving them. In brief, this book seeks to revive and develop a long-neglected approach to the study of voter turnout that focuses on individual motivations.4 The strategy is to pull together a number of building blocks that come from different branches of political science – research into the rational underpinnings for the decision to vote, research into the socialization and immunization of new voters, research into voter turnout at the aggregate and individual levels, and research into the generational basis of political change – and use them to develop a theoretical edifice that is both more elaborate and more comprehensive than those previously constructed to examine voter turnout. This framework generates hypotheses at the level of the country, election, electoral cohort, and individual voter. These hypotheses in turn are tested by using, on the one hand, datasets constructed from survey data for all of the elections ever studied, back to the 1960s, in every country (all six of them) for which there is a continuous series of such studies for every

3

4

We do not investigate the deleterious effects that low voter turnout may bring by virtue of the unrepresentative nature of those who vote when turnout is low. These effects are in need of more careful research than they have previously received (see Chapter 8), but that is not the purpose of this book. Research on aggregate-level turnout often assumes such motivations while evidently failing to test for their existence (such a test cannot be implimented at the aggregate level). The mismatch in findings between individual-level and aggregate-level studies gives rise to one of the puzzles that this book seeks to resolve (see Chapter 1).

6


national election held since then, and, on the other hand, a specially collected set of aggregate data spanning fifty-five years from the end of World War II, covering all of the national elections conducted during that period in all of the twenty-two countries that have an accessible and continuous statistical record of free and fair national elections held since within one electoral cycle of the end of that singular global convulsion – 356 elections in all. The hypotheses that we derive from our theorizing fly in the face of a number of common presumptions about turnout and its decline. In particular, the findings of this book are not expected to support the notion that turnout declines (where it does) because of a shortfall in civic virtue and dutiful behavior or because of political alienation or disaffection. Electoral turnout, we will contend, is not (generally speaking) about how people approach elections; rather, it is mainly about how elections appear to people. The theoretical viewpoint developed here expects turnout to vary either when elections change their character or when demographic shifts change the sizes of groups that pay attention to the character that elections have. Features of the character of elections that affect turnout are hypothesized to include some of the rules under which elections are conducted (including the electoral system that is employed), features of the party system at the time of each election (particularly its fractionalization and cohesion), and features of particular elections (including how much time has elapsed since the previous election and whether its outcome was too close to call). Most of these features are encapsulated in the concept of electoral competition, and our model could be typified as the “electoral competition” model of voter turnout. Notably absent from our hypotheses are expectations that voter turnout in established democracies will be affected by changes in the resources of individuals or of the social structures in which they are embedded (though the age structure of the electorate is expected to be important). These common presumptions are, however, investigated to see whether they add anything to the electoral competition model that is built to test this book’s hypotheses. It is important to stress at the outset that our hypotheses are intended to apply to established democracies – expectations for transitional and consolidating democracies would be different (as explained in Chapter 1), but those are not studied in this book for reasons that

Introduction

7

will become clear as our theoretical framework unfolds. Our objective is to arrive at general conclusions that would apply to any established democracy. Democracies that are not yet established should in due course display the same dynamics as are shown in the countries that we study. In Chapter 7 we will discuss the ways in which democracies that are not yet established can be expected to display different dynamics from those that we find in established democracies. The book is organized as follows. Chapter 1, as already stated, describes the key puzzles that will be addressed in the remainder of the book. Chapters 2 and 3 set out the theoretical framework that will guide our investigations – and that will give rise to the hypotheses put forward in later chapters. Chapter 3 also investigates fundamental aspects of this framework at the level of electoral cohorts. Chapter 4 presents a number of case studies and statistical analyses that support the operationalization of variables to be employed in subsequent chapters. Chapters 5 and 6 use those variables to test hypotheses (derived from the theories developed in earlier chapters) at the level of the country, election, and individual. Chapters 7 and 8 elaborate on the findings, establishing to what extent they account for actual changes in turnout that have occurred in the twenty-two countries that this book studies and discussing the implications of those findings in terms of the likely future evolution of turnout and measures that could or should be taken to avoid, or even reverse, future turnout decline.

1 Confronting the Puzzles of Voter Turnout

The puzzles alluded to in the Preface and Introduction to this volume come in two major varieties. The first are puzzles about voter turnout itself, among them, why do people bother to vote at all? Why is turnout so relatively stable over time (compared to the enormous differences we see between countries)? Why does it decline when it does? And why (in some countries) does it not decline at all – or even rise? The second variety of puzzles have to do with how to study voter turnout. For example, is turnout an aggregate-level or an individual-level phenomenon? Can we understand it best by studying turnout change over time? By studying differences in turnout among countries or by studying why some individuals vote while others do not? What can we learn from the fact that turnout varies more among younger cohorts and less among established cohorts? In this chapter we take these and other puzzles one at a time and use the process of going through them as an opportunity to set out not only the puzzles themselves but also the major approaches that have been employed in past research for solving them. Why So Much Turnout Change – or Why So Little? It has already been pointed out that turnout decline is not ubiquitous. Turnout did rise in the 2000 U.S. presidential election, for example. Turnout also rose markedly in the German election of 1998, in the Norwegian election of 1997, in the Swedish election of 1994, 9


10 Turnout %

80

70

Established democracies 60

50

40

All other countries 30

20

10

0 1945-1949

1950-1959

1960-1969

1970-1979

1980-1989

1990-1997

figure 1.1. Average turnout in decades for thirty-nine countries since 1945. Source: International IDEA (1997). The graph is not contained in the second edition of this work, which is otherwise updated to 2000.

and in the British election of 1992 (just to take another four recent cases of turnout increase that received little publicity). In transitional and consolidating democracies, turnout increases have been more common than declines in recent years (IDEA 1997; Norris, 2002). Even in established democracies, the overall decline in turnout has not been exactly remarkable. As shown in Figure 1.1, the Institute for Democracy and Electoral Assistance (IDEA) voter turnout project estimates that turnout in established democracies had declined by just 4 percent, on average, between the decade of the 1970s and the decade of the 1990s (though the final dip for established democracies would have been 1 percent greater had the chart taken us to 1999). This is not a big decline over a twenty-year period. More importantly, the associated

Confronting the Puzzles of Voter Turnout

11

table 1.1 Mean, Standard Deviation, Trend, and Standard Error of Estimate for Turnout in Twenty-Two Countries, 1945–1999 Country

Mean(turnout)

SD(turnout)∗

Trend (b)∗

SEE∗

Malta Sweden Australia Denmark Norway Belgium Iceland Germany Israel Ireland United Kingdom Luxembourg United States Canada Italy New Zealand Japan France Finland Austria Netherlands Switzerland Overall

88.2 86.1 94.6 85.6 80.6 92.6 89.5 85.6 80.4 73.2 76.4 89.8 55.8 74.6 90.6 88.6 71.3 76.7 75.9 92.1 87.6 56.6 81.4

8.7 4.7 1.8 2.5 3.0 1.7 2.0 4.4 3.1 3.2 3.6 2.0 4.0 4.2 3.3 4.1 4.6 5.9 5.8 3.9 7.5 11.0 4.4

+23.5 +7.8 +2.0 +1.7 +1.3 +0.1 −3.2 −4.1 −4.8 −5.5 −5.6 −5.7 −6.2 −6.3 −7.0 −9.3 −9.8 −10.9 −11.0 −11.3 −21.1 −34.7 −5.5

3.2 3.8 1.3 1.9 2.8 1.4 1.5 4.3 2.5 2.6 3.0 0.8 3.7 3.4 2.3 2.2 2.8 3.9 3.9 1.8 3.5 2.3 2.7

∗

SD is the standard deviation (the span, + or −, within which turnout is found two-third of the time). The trend is from a regression of turnout on time, where time is a continuous variable running from 0 in 1945 to 1 in 1999, and the trend of the best-fitting line is given by its slope (b). SEE is the standard error of the estimated trend (the standard deviation of turnout around the trend). Source: International IDEA (1997, 2002).

election to election drop of less than 0.5 percent is almost imperceptible. It would be hard to establish by statistical methods the reasons for such a small decline. Perhaps, in the light of Figure 1.1, the question that we should be asking is not why is turnout declining, but why is turnout so stable? This question is reinforced if we look at the extent to which turnout has varied since 1945 in the twenty-two countries that have held elections continuously over this period (see Table 1.1). Most of them have seen turnout varying by no more than some 5 percent from the mean for the period. Indeed, only in five countries (Finland,

12


France, Malta, The Netherlands, and Switzerland) does turnout show a standard deviation as great as this; and these are all countries that saw marked trends – trends involving more than a 10 percent change in turnout over the period – upward in Malta, downward elsewhere. Even these five countries see turnout that seldom strays by as much as 4 percent from a linear trend describing their overall change in turnout during the period.1 The most plausible answer to the question of why is turnout so stable ironically gives new weight to the small amount of turnout decline that is occurring. Turnout appears to be stable because, for most people, the habit of voting is established relatively early in their adult lives. Those who find reason to vote in one of the first elections at which they are eligible generally continue to vote in subsequent elections, even less important ones. On the other hand, those who find no reason to vote in their first few elections generally continue not to vote in subsequent elections, even more important ones.2 This insight has been documented using data for the United States (Miller and Shanks, 1996; Putnam 2000; Plutzer 2002) and Canada (Blais et al. 2001), but because it ties in so well with our understanding of other aspects of voting behavior (and because turnout stability is ubiquitous, as we have seen) we can confidently expect that the same thing will be found elsewhere (though establishing that this is indeed the case is one objective of this book). We will elaborate on the theory underlying this insight in Chapter 3. The decision to vote, according to the logic explained there, is like the decision to support a particular party. Just as support for political parties is established for most people early in adult life, so with turnout.

1

2

The trend line in Table 1.1 is only intended for descriptive purposes (the analysis from which it derives does not claim to employ a properly specified model). The standard error of estimate in the final column is the range, plus or minus (from turnout expected on the basis of time elapsed since 1945), within which actual turnout falls two-thirds of the time. The story is somewhat more complex than this because citizens often undergo development during their first few elections, and many of those who did not vote in their initial election do learn the habit of voting shortly thereafter – what most researchers have called an age effect but which Plutzer (2002) refers to as a transition. Nevertheless, those who first vote in an election with low turnout do retain a distinctive pattern of lower turnout throughout their adult lives, as we shall see in Chapter 3.


13

Only in the earliest elections at which individuals are eligible to vote would there be any real question as to whether they will vote or not. This insight has a number of implications for our understanding of turnout variations, some of which go beyond anything that has been suggested in past research. For one thing, if most variation occurs within new cohorts, then the small turnout variations that we have documented within electorates taken as a whole must correspond to much larger variations in the turnout of those who are recently eligible (this is the subject of Chapter 3). More importantly, any sustained change in turnout (due to new voters voting at different rates than their predecessor cohorts) will have a cumulative effect as those new voters progressively come to make up a larger and larger portion of the electorate. If new cohorts are voting at lower rates than established cohorts, then turnout will decline; and, even with no further drop in turnout among new voters, that decline will continue until the new rate of voting is reflected throughout the electorate – a development that could take as much as fifty years to run its course. On this basis, a drop of 5 percent in twenty years may yield a drop of 10 percent over forty years and bottom out after a further 2.5 percent drop during the final ten years that it takes to replace an entire electorate: That is a 12.5 percent drop overall. When we consider that the mean drop in turnout is an average deriving from countries in which turnout is rising as well as of countries in which turnout is falling, we have to concede that turnout might be undergoing a quite substantial long-term decline in many established democracies. Table 1.1 tells us that, in the fifteen countries with a negative trend in turnout, only one has seen a turnout decline of less than 4 percent, which implies that all but one of these countries could see at least a ten-point drop in turnout or over a fifty-year period. So the small amount of turnout decline that we have seen in recent years could presage much larger turnout falls. For this reason we do need to take seriously the two questions that so often appear to excite the commentators. These are 1. Why is turnout declining in so many established democracies? 2. Does it matter?

14


To address these questions we need to confront a number of additional puzzles. Many of these are puzzles that bedevil all of political science research to a greater or lesser extent, but few other research topics confront so many puzzles all at the same time in the way that this one does.

Change Compared to What? Other Countries or Other Times? What should we compare when we compare turnout? Most commentators who decry declining turnout compare current with past turnout in the same country, and Bernard Grofman (1993) once argued in his aptly titled “Is Turnout the Paradox That Ate Rational Choice Theory?” that political scientists should emulate economists in focusing not on why the level of a variable is what it is, but rather on why it varies over time. This is an approach that is common in studies of turnout within individual countries but not in the comparative turnout literature, which has focused primarily on turnout differences between countries. Those who have studied multiple elections in each of a number of countries generally used the additional time points as replications of each country (what might be called quasi-countries) with the object of increasing their study’s N;3 but their focus was still on making comparisons between countries.4 Are comparisons between countries equivalent to comparisons over time? Most scholars studying aggregate turnout levels (Powell 1980, 1982, 1986; Jackman 1987; Crepaz 1990; Jackman and Miller 1995; Franklin 1996; Blais 2000; Norris 2002) have implicitely assumed that if they can identify factors that are associated with differences between countries, these same factors will account for why turnout changes over time. A country with low turnout is assumed to have the characteristics that a country with declining turnout is in the process of acquiring. But is this assumption reasonable? But does what we learn from comparing countries translate into an understanding of why turnout changes over time within one country (cf. Rose 2002)? 3 4

Important exceptions are found in Gray and Caul (2000) and in Franklin (2002). We defer until Chapter 3 a discussion about whether the comparisons should employ turnout as a percentage of the registered electorate or as a percentage of the voting age population.


15

One important determinant of turnout across countries is weekend voting. Countries that go to the polls on Saturday or Sunday see higher turnout, to the extent of some 6 percent, than countries that go to the polls on weekdays (Franklin 1996, 2002). But when countries adopt weekend voting their turnout does not appear to increase, and nor does turnout appear to drop when they move to weekday voting (Franklin 2002). The same lack of consistency applies to the proportionality of electoral systems (how closely the proportion of votes cast for a party is translated into the proportion of seats that party wins). Turnout appears to respond to differences in average proportionality between countries with a more proportional translation of votes into seats leading to higher turnout because there will be fewer individuals who feel their vote would be wasted. But changes in proportionality over time in the same country have no equivalent effect on turnout (Franklin 2002). A similar inconsistency has been found with compulsory voting when established democracies are compared with other countries (Rose 2002). Such inconsistencies are puzzling, and any new basis for understanding the mainsprings of electoral participation should be able to explain them. These observations make it clear that understanding the reasons why some people vote while others do not, or why some countries see higher turnout than others, will not necessarily tell us why turnout is declining or how to make it increase. Comparisons across space and individuals may be the only way certain research questions can be answered, but the answers we get are not necessarily the answers we want; and in turnout studies we are not restricted to this approach. We need to compare people and countries when we lack information about change, but when studying turnout we have no such lack. If, for example, we want to know whether introducing weekend voting will cause turnout to increase, we arguably get more appropriate evidence from examining countries that have changed the day upon which they hold elections than from examining a set of countries that vote on different days.5 If such changes are not associated with change in turnout, then any cross-country association between rest-day voting 5

If countries were to prove themselves responsive to different variables, or to the same variables in different ways, then of course we would be interested in these country differences.

16


and turnout must be spurious or contingent.6 The same applies to other potential independent variables.7

Something About Societies or Something About People? While voting is a matter of individual decisions, turnout is an aggregatelevel phenomenon. It is a feature of an electorate not a voter. And, while it is true that electorates are made up of aggregates of voters, the process of aggregation is not simply one of adding up relevant features of the individuals who form part of it. An electorate is not just a voter writ large any more than an economy is a consumer writ large.8 Yet most of the work on what makes people vote has been conducted at the individual level. We know from survey research that the factors most strongly associated with the likelihood that someone will vote are their age (young people are less likely to turn out), their education (those with a college education are more likely to vote, at least in the United States, and literacy has been found to be important in transitional democracies), and the extent to which they are embedded in social structures (people who are members of churches, unions, and other organizations are more likely to vote; loners are less likely to do so). The three variables are connected in the theory that underpins past research because membership in organizations makes people easier to mobilize (Rosenstone and Hanson 1993), education serves to inculcate civic virtue and various skills that can make electoral participation easier (Lipset 1960), and age governs the opportunities that people have had to receive an education and to become embedded in social structures (Putnam 1995, 2000; Verba, Schlozman, and Brady 1995).

6

7

8

A contingent relationship might occur when certain cultural prerequisites are needed for variables to have the expected effect (some call this path-dependency). One benefit from focusing on established democracies is to reduce the likelihood of path-dependent effects; but, in any case, such effects cannot be responsible for turnout change in one country. Of course, looking only at change over time could equally give rise to misleading findings if certain important variables did not change their values during the time span under investigation. In this book we will look both at change across units (people, cohorts, countries) and change over time. Dirkheim’s famous study of suicide rates confronted the same problem of distinguishing between the causes of suicide and the prevalence of high suicide rates.


17

When we take the aggregate view, however, we need to concern ourselves with the precise manner in which individual-level effects are aggregated, a concern that has not been prominent in past theorizing. The idea that resources and mobilization affect turnout leads to the idea that turnout depends on the social makeup of a society – the proportion of the population falling into different age groups, having different levels of education, or having strong links to social groups – with a more highly educated population, for example, being one that should demonstrate higher turnout. But the social makeup of society does not pertain directly to the voting act, and attempting to derive aggregate-level implications for turnout is not straightforward. For one thing, there is a mechanical problem arising from the fact that quite large changes in the structure of the population do not necessarily have important effects on turnout in established democracies.9 Again taking education as our example, if we were to see a doubling of the proportion of the college-age population engaged in higher education in some country (from, say, 10 percent to 20 percent) over the course of twenty years, this would represent an enormous investment in higher education, even for a rich country. But what would be the effect on turnout? The most generous estimate of the effect of higher education in established democracies is that college-educated individuals in the United States are some 16 percent more likely to vote than those with no college education (Rosenstone and Hansen 1993). On average, over twenty-two countries, the effect found by Franklin (1996, 2002) was more like 12 percent. For ease of computation let us take it to be 15 percent. A fifteen-point increase in turnout among 10 percent of the population is a 1.5 percent increase overall. But note that this increase takes twenty years to accomplish. More importantly, the full effect of the reform will not be felt until the new proportion of college educated is reflected throughout the population, and that will take a further fifty years or more, until the last of those educated before 9

We would have different expectations if we were dealing with transitional and consolidating democracies, since social differences there can be more stark. Being able to read, for example, makes a difference to turnout in a way in which university education does not. Norris (2002) finds ceiling effects in the links between social characteristics and turnout, such that many relationships tail off in established democracies. So it is important to look separately at the two groups of countries. In this book we focus on established democracies.

18


the reform leaves the electorate. A 1.5 percent rise in turnout over a seventy-year period would certainly be swamped by other changes and would likely be impossible to detect, even in a properly specified model of turnout change. Certain demographic changes (for example, in the age structure of the population) might take effect more immediately, as we shall see. However, problems similar to those regarding education arise when we attempt to aggregate many other measures of individual-level resources, or measures of people’s susceptibility, to mobilizing efforts.10 Proponents of the resource and mobilization models of turnout change have not addressed these problems. Instead they argue by analogy with the individual level – an analogy that may not hold.11 But the role of education and other resource variables in explaining the decision to vote is put into question in other ways as well.

How Do We Know When an Association Is Causal? The contrast between low turnout in Switzerland and the United States and high turnout in Malta is curious for a reason that goes beyond putting into question the primacy of civic virtue in explaining why people vote. It also puts into question the role of education in inculcating that civic virtue. This is because Switzerland and the United States are countries with among the most highly educated populations on earth. This question about the role of education is reinforced when

10

11

For example, changes in union membership, such as the 50 percent reduction that occurred in the United States in the forty years after 1960, would also have had rather small effects on turnout. Even in the 1960s, union members made up less than 20 percent of the voting age population and were only some 4 percent more likely to vote (according to a recent analysis by Gray and Caul, 2000). This implies less than a 1 percent change in turnout for the voting population as a whole over a forty-year period. Outside the United States, union density has declined less – by under 2 percent on average across eighteen advanced democracies, according to Gray and Caul, so this is not likely to be a variable that could account for widespread turnout decline. It is as though an economist were to argue that, since few of those with college degrees are unemployed, we can solve a country’s unemployment problem by educating everyone to the college level. The experiences of countries like India and Egypt, with high proportions of unemployed graduates, tells us that there is no aggregate-level counterpart to the individual-level relationship. Why should it be any different with turnout?


19

we consider the fact that most countries have seen a huge expansion of secondary and tertiary education since the 1960s – precisely the period during which turnout has declined in many of these countries. Over the twenty-two countries that have held elections continuously since the end of World War II, the correlation between the proportion of the population with a university education and turnout is negative and highly significant at −0.36. These anomalies call into question the causal primacy of education in determining turnout (cf. Brody 1978) – a primacy established firmly at the individual level, at least in the United States. This primacy is also called into question by the failure to find a similar role for education in many other countries (Verba, Nie, and Kim 1978; Topf 1995; Heath and Taylor 1999). There is no doubt about the individual-level link between education and electoral participation in the United States, but perhaps we make a faulty inference when we assume that, in established democracies, more education leads to higher turnout.12 Just because some factor is associated with electoral participation does not mean that the association is causal. Suppose education is not itself a cause of voter turnout. Suppose that education is simply an indicator of an enquiring mind. Perhaps more people vote who take an interest in the world around them, and perhaps such people are also more likely to pursue an extended education. On this view, the number of people with higher education is a function of higher education provision, and, the more opportunities there are for people to become educated, the more educated people there will be. Interest in politics, however, could be quite unaffected by whether there are more or fewer university places.13 Yet education will correlate with turnout because politically motivated people will be the ones who pursue a higher education when the opportunity is presented, and these are also the people who vote. Indeed, the generally assumed causal connection between education and interest in politics might at times actually be reversed, if some people have been motivated to acquire a higher education because

12

13

In transitional and consolidating democracies, education and income have sometimes served as indicators of economic and social development that have been shown to be linked to turnout (Blais 2000; Norris 2002). Verba, Schlozman, and Brady (1995) make a similar point when they emphasize the importance of relative resources.

20


of increased political consciousness. A similar argument could be made with regard to other resources. If the association between education and turnout is not causal, this would explain why many countries can have high turnout despite having much less well-educated populations than the United States or Switzerland. It would also explain why the United States and Switzerland themselves have had higher turnout in past years when they had less well-educated populations.14

Two Faces of Socialization Those who have tried to explain why people vote on the basis of learned patterns of behavior stress the socializing influences of family and school. But socialization is a tricky concept. The idea that people are socialized into certain patterns of behavior at a young age is generally thought of as having to do with early training. Parents and teachers are seen to be trying to get children to behave in certain ways, and the fact that those children grow up to be decent citizens is often thought to reflect the success of that training. Certainly, when new behavior patterns that are seen as undesirable arise among young adults, the blame is often placed on lax parental and school training. Yet the origins of the notion of socialization, in the work of the first Yale professor of sociology, William Graham Sumner, lie not in training but in the desire to conform (Sumner 1906). Indeed, the desire of children to conform their behavior to that of other children is often easier to observe than the effects of parental training. What probably happens in fact is that a child’s first reference group is his or her parents, and she starts by conforming her behavior to theirs. Later the child’s reference group becomes the playmates he or she meets outside the home, and her behavior changes with the identity of those to whose behavior she is trying to conform.15 14 15

The correlation between university education and turnout over the period since 1945 is −0.68 in the United States and −0.96 in Switzerland. Seen in this way, the declining ability of the family to govern the behavior of children has nothing to do with the effectiveness or otherwise of parental efforts at training them and everything to do with the decline in the sizes of families and the increasing amount of leisure time for children who thus find both the need and the means to gain companionship elsewhere.


21

Political behavior evidently has much in common with other types of behavior, and individuals are held to learn their political orientations from the orientations of those around them. The decline of class voting (Franklin 1985) or, more generally, the decline of cleavage politics (Franklin 1992) had much to do with individual mobility and the declining importance of face-to-face communications in the modern world. These developments have given individuals increasing freedom to find and change the reference groups to which they conform their voting behavior (van der Eijk et al. 1992). But there is another component to socialization, one that it is important to bear in mind when trying to understand the failure of certain individuals to conform to changing patterns: This is the fact that people get set in their ways. The longer some pattern of behavior has been pursued, the harder it is to change. The reason why children are so susceptible to new influences is that they have not been alive long enough to get set in their ways.16 But the repeated reiteration of any particular pattern of behavior lays down tracks in a person’s personality, making it harder and harder for that person to change the behavior concerned. This is why adults retain so many of the behavior patterns of their childhood, which in turn is why socialization is such an important concept. But note two critically important implications: (1) that tracks can be laid down at any age, not just in childhood, and (2) that the failure to repeatedly follow any particular behavior pattern implies the failure to lay down tracks. People who do not repeat the same behavior do not (in the language of Butler and Stokes, 1974) become “immunized” against change. Butler and Stokes found that the magic number was three: Anyone who had voted the same way three times had become essentially immune to the appeals of any other party (in this book we will generally refer to this as “inertia”). People who had switched their party allegiance more recently than three elections ago were much more likely to switch back again. Indeed, these authors found quite elderly respondents who were still not immunized against change, having repeatedly switched their party allegiances all through their lives. There were few such, however, because the mere happenstances of political

16

Indeed, the adaptability of children was probably an important survival trait for our species, allowing new generations to adapt to new circumstances.

22


fortune generally resulted in someone voting the same way three times at some point in their adult lives. The way in which we have presented socialization theory in the previous paragraphs is in terms of rational behavior. People adopt a pattern of behavior as a rational response to circumstances (conforming to peer group pressure or responding to political arguments are both rational responses to circumstances). Over time, the responses may cease to be so rational, if people become immunized against further argument,17 but the fact that we see so much rote behavior by older individuals does not preclude a rational basis for that behavior originally. When we consider the decision to vote we need a similar starting point. Those who tried to find a basis in socialization theory for the decision to vote – a basis that would explain why people vote when they supposedly gain no political benefits from doing so – ignored the fact that even socialized behavior rests on a basis of rational decision-making. How were people supposed to get stuck in the habit of voting if they had no basis for choosing whether to vote in the first place? One possibility is that parents would inculcate in their children the value of the voting act without reference to its instrumental benefits. But why should parents do this unless the voting act had value to them? Socialization theory merely moves the point at which people decide whether to vote back in time from the current election to those elections at which they learned the habits of voting and not voting. It does not provide an alternative basis for explaining the origins of those habits. This brings us back to the contrast between individual-level behavior and the behavior of aggregate entities such as electorates. At the individual-level, most of the individuals we study are socialized into their behavior patterns. If we want to understand those patterns, we have to study the nature of the socialization process that they underwent. At the aggregate level, the importance of the socialization process is reduced to this: It ensures that change in the behavior of the aggregate entity will be slow rather than rapid. Turnout changes only gradually

17

Unless we consider inertia to be a rational response to the pressures of decisionmaking – economizing scarce decision-making resources by changing decision-based into rule-based behavior whenever possible and adopting “standing decisions” to handle situations that arise repeatedly (see Chapter 2).


23

because most people have adopted a “standing decision” to vote or not to vote, based on their early experiences of elections in their country. So turnout change depends on the behavior of new members of each electorate, and the things that determine turnout change will be the things that impinge on these new electoral cohorts. We will return to these matters later. The Role of Mobilization Meanwhile, we still need to consider the role of mobilization. This is another concept that seems to explain turnout without the need to assume that voters are motivated by concerns about an election’s outcome. After all, if voters are mobilized by party and group leaders, this means they are just responding to suggestions from those they respect or who have some hold over them. Or does it? Actually, in the modern world it is hard to imagine anyone being able to command the blind allegiance of followers to the point where they would vote just because they were encouraged to do so (cf. Whiteley 1995). Some reason would have to be given, and it seems likely that the reason given would relate to group advantage. Those who wanted to mobilize a group of potential voters would have to explain to them that, if a particular candidate wins, she will favor the group whereas, if she loses, her opponent will favor someone else. At the very least, group members have to be persuaded that loyalty to the group demands that they vote, which requires recognition at least that other members of the group believe that a favorable outcome matters. But in this case the mobilization effort is helping to explain to people that their votes have value (political or social) – that the calculus of voting espoused by Riker and Ordeshook is incomplete. In Chapter 2 we will build on this insight when we come to develop a calculus of voting that does give each vote a value considerably greater than the “vanishingly small” amount it was found to have through the application of the logic developed in that classic article. It is hard to see how the mobilization mechanism could in fact be independent of instrumental rationality. If mobilization is successful – as Rosenstone and Hansen (1993) tell us that it is – then this must be because voters are persuaded to see their votes as potentially bringing them the benefits that might flow from a favorable electoral outcome.

24


But the mobilization of electorates, if it occurs, also provides us with a second reason to suppose that votes have instrumental value. Those who take the trouble to mobilize their supporters must want to win the election. If the outcome did not matter to them, then they would not bother. But the outcome is likely to matter to those who do the mobilizing for exactly the same reasons that would make it matter to someone who was mobilized: An election that promises the possibility of policy change, especially if the contest is close-fought, will motivate the mobilization efforts of all of those who have the possibility of influencing others. Indeed, low-turnout elections are often referred to as “low-mobilization” elections in recognition of precisely this mechanism. The declining importance of labor in the United States since the 1960s was not just due to the falling number of union members. It was also due to the fact that union leaders ceased to attempt to mobilize their members. In recent U.S. elections unions have been a more important force, despite their smaller number of members, because union leaders have begun once again to attempt to mobilize those members. The question of why union leaders should have stopped trying to mobilize their supporters thirty years ago and started doing so again today goes far beyond the remit of this book. What can be said is that it makes a big difference whether the attempt is made. Those who argue that the sizes of social groups (such as unions) will determine the level of turnout ignore the critical role of the leaders of those groups. In the United States, the decline of unionization indeed went along for many years with a decline in the attempt to mobilize union members, but the recent change in this mobilizing behavior is independent of any increase in the size of the unionized U.S. workforce. Finally, we should ask, if the attempt is made to mobilize the members of some group, who will respond? On the basis of the socialization theory put forward earlier in this chapter, it seems clear that those who will respond to a renewed call for union solidarity are union members who are also new voters – individuals too young to have learned the habit of voting before the long years of leadership neglect and also too young to have learned the habit of nonvoting during the period when unions made no attempts to mobilize new members. So mobilization is not a separate route to electoral participation; it is a mechanism that works by way of both rationality and socialization.


25

In the end there is no escaping the fact that both rationality and socialization must play a role in turnout. In particular, the resource theory of turnout, while helping to distinguish between voters and nonvoters in most countries, does little to help us understand turnout differences between countries or over time – at least in established democracies. The mobilization theory of turnout is more promising, but only because it is intertwined so closely with rationality and socialization – the two basic forces that govern all of voting behavior. Insofar as it is intertwined with those forces, mobilization becomes a route by which those forces have their effects, not a separate force in its own right, and certainly not a substitute for rational action.

Individual-Level Change or Change in the Structure of Society? Although changes in social structure do need to be extensive to have measurable effects, major changes in political behavior have often resulted from changes in the composition of society. Though there is argument over whether the New Deal realignment in the United States was due to the mobilization of previously nonvoting individuals or to changes in party allegiance by established voters (Anderson 1979; Campbell 1985; Erikson and Tedin 1981), there is no similar question about the generational basis of the British realignment of 1945 (Franklin and Ladner 1995). And the decline of cleavage politics in most advanced industrial democracies appears to have been due primarily to the same phenomenon (Franklin 1992). Changes in social composition can lead to changes in aggregate behavior even when no individual citizen changes her behavior in any way.18 A number of changes have occurred in the composition of the electorates of established democracies that need to be taken into account in any analysis of turnout decline. In the first place, the electorates of most countries have been swollen, starting in the late 1960s, by the entry of the so-called baby boom generation born after the Second World War.19 Because new voters turn 18 19

Changes in the composition of the European Union appear to have been primarily responsible for turnout decline at European Parliament elections (Franklin 2001). In several countries electorates were also swollen by the enfranchisement of women (in France, Italy, Japan, and Malta immediately after the Second World War; in Belgium in 1948; and in Switzerland in 1972).

26


out at lower rates than established voters, anything that increases the proportion of new voters will cause turnout to fall. At approximately the same time, the voting age itself was lowered in many countries – generally from twenty-one to eighteen. These two events in conjunction had the effect of producing a newly enfranchised cohort that was double the size of previous cohorts in most countries, in turn doubling the contribution of new members of the electorate to turnout. Since (as was pointed out earlier) newly enfranchised voters are those largely responsible for turnout change, this provided the opportunity in many countries for an unusually large drop in turnout during the early 1970s. To the extent that (because of their younger ages) new cohorts lacked educational and membership characteristics of their predecessor cohorts (and to the extent that these features would be acquired with the passage of time), the drop would have been a temporary one. The large group of new voters would have eventually acquired the characteristics of their elders and turnout would have risen again to previous levels (cf. Topf 1995). But this is a process that could take fifty years to complete, and in the meantime turnout could drop progressively for a number of years before starting to recover. However, we are talking about relatively small changes in turnout. Calculations that we will spell out in Chapter 3 suggest that, at the point of inflection where turnout would cease to fall on account of the baby boomers and start to rise again, the amount by which turnout might have declined would be no more than 0.7 percent.20 Still, we will have reason to refer to this quantity in later chapters. More importantly, the lowering of the voting age may have had additional, larger, and more permanent (if slower acting) effects, as will be suggested in Chapter 3. The increase in the number of young voters will of course also increase the overall size of the electorates, so it should come as no surprise that increases in the sizes of electorates have been credited (Blais 2000; Gray and Caul 2000) with affecting turnout by bringing in new people who are harder to mobilize – whether because they are young or because they are newcomers to the society. Though these authors see electorate size as a variable bearing on electoral mobilization, their argument does not suffer from the same problems as other mobilization 20

McDonald and Popkin (2001) estimate this effect for the United States at 1 percent.


27

arguments because electorate size has a direct link to our measure of turnout: It constitutes the denominator in the calculation of that variable. Changes in electorate size must have immediate effects on turnout, unless new voters vote at the same rate as existing voters.21 But any increase in the proportion of new voters contained in the electorate may well have an additional consequence if new voters, as has already been proposed, are so largely responsible for turnout change. An electorate with more new voters will be an electorate that is more responsive to any factors that alter voter motivations.22 So changes in the character of elections may have become more potent in effecting turnout change just because contemporary electorates contain a larger proportion of those who are responsive to such changes. Of course, the classic rational choice basis for understanding the motivation to vote, as put forward by Riker and Ordeshook (1968) and underpinning most research at the individual level as already explained, would deny that there are any factors that can influence voter motivations, whether in interaction with the proportion of new voters or directly. But we have already indicated that we have strong reservations about the classic rational choice basis for understanding voter motivations, and it is to those reservations that we now turn.

Something About People’s Characteristics or Something About Their Motivations? Just as important as the fact that some individual-level effects on turnout, such as those of education, appear not to have counterparts at the aggregate level is the fact that aggregate-level research has found important effects on turnout that would not be expected on the basis of Riker and Ordeshook’s theorizing and have not been identified in past individual-level studies – effects that arise from variations in the competitiveness of elections. It appears as though people react to these differences because they alter the instrumental benefits of voting, 21

22

The variable will have a particularly marked effect when turnout is calculated on the basis of the voting age population, since immigration raises the denominator without having any possibility of raising the numerator commensurately (see Mcdonald and Popkin, 2001). We discuss this problem in Chapter 3. Peter Mair (2002) has remarked on the increasing willingness of voters to transfer their preferences between parties – a hallmark of an increasingly youthful electorate.

28


creating an apparent contradiction, since (according to rational choice theory) people should not take account of their vote’s instrumental benefits.23 This apparent contradiction has become known as the “turnout paradox.” But the very recognition of a paradox (Fiorina 1990; Groffman 1992) underlines the fact that people vote in large numbers, defying a clear implication of rational choice theory, while other implications that flow from assuming voter rationality do hold. No doubt theorists will one day resolve the paradox – indeed, we hope to make a contribution toward its resolution in this book. Meanwhile, it is enough to note that turnout varies as though in response to rational behavior – perhaps a form of aggregate rationality that is not readily amenable to Downs’ (1956) individualistic rational choice calculus that underpins Riker and Ordeshook’s argument (cf. Green and Shapiro 1994).24 Research in this older tradition has repeatedly demonstrated the importance of institutional and political context. Citizens in some institutional settings vote at much higher rates than citizens in other settings. Indeed, Franklin (1996) showed how country differences had effects on turnout that were four times as great, on average, as individual-level differences. Country differences have two obvious components: costs (generally established by institutional features of each country that are largely set in the short term – cf. Blais 2000) and benefits (generally arising from the political situation and encapsulated in electoral competitiveness that varies from election to election – cf. Franklin 2002). It is the second of these components that puts into question the notion that people vote primarily because of benefits that are independent of who wins or loses. In some countries (notably the United States and Switzerland) institutional provisions make it impossible for legislative elections to directly affect the complexion (and hence the policies) of the executive; and turnout in U.S. and Swiss legislative elections in recent years has 23

24

Such research builds on a tradition – associated with works by Merriam and Gosnell (1924) and Tingsten (1937) – that predates the survey-based individual-level research that we have been discussing, though that older literature has been updated by contemporary concerns (Powell 1980, 1982; 1986; Jackman 1987; Crepaz 1990; Jackman and Miller 1995; Franklin 1996, 2002; Blais 2000; Norris 2002; Rose 2002). We will give this question more attention in Chapter 2.


29

always been low.25 Elsewhere, much depends on the current state of party competition, which can vary greatly from election to election. Of particular importance are the closeness of the race and the decisiveness of the election – the likelihood that it could result in a (different) single-party majority government (Blais 2000; Franklin 2002). Close races that are liable to result in changes in government policy are highly competitive and will motivate both supporters and opponents of the government to vote, while also stimulating efforts to mobilize the undecided. The degree of electoral competition could only affect the behavior of individuals who cared which party or candidate won (or who were mobilized by other individuals who cared which party or candidate won). So, even while the importance of the resource and mobilization approaches were being confirmed at the individual level, the theoretical basis for trying to find noninstrumental explanations for voting and nonvoting were being undermined at the aggregate level. The aggregate-level findings have, moreover, been reinforced by research into strategic voting (Cain 1978; Abramson, Aldrich, Paolino, and Rohde 1992; Niemi, Whitten, and Franklin 1992; Blais and Nadeau 1996). Such research has shown repeatedly that people behave as though their vote had instrumental value, not wishing to waste it on a candidate or party that had no hope of winning (even if that was their preferred party or candidate) when they could use it to further the chances of a party or candidate that did have a chance of winning (cf. Blais 2000). Does all of this mean that the individual-level findings for resources and mobilization are spurious? Not necessarily. Resources and mobilization may just be playing a different role than has been assumed in past research. If socialization plays the important role that we outlined earlier in this chapter, then people’s resources may be acting as indicators of the ways in which they have been socialized. And if electoral competition plays an equally important but neglected role, then mobilization variables may be providing one route by which changes in the character of elections are felt. If these variables serve as indicators of other processes, or as mechanisms that transmit the effects of other processes, this could explain why, in the analysis of turnout differences 25

We will develop these ideas in Chapter 4.

30


among established democracies, variables like education, occupation, and income have not shown themselves able to play a significant role.26 So there is no reason why individual-level effects of resources and mobilization should not co-exist with instrumental motivations. The problem is rather that there is no theoretical basis for assuming the existence of such motivations. This is why research at the individual level has not attempted to measure the impact of electoral competition on individual voters. In Chapter 2 we will tackle this problem head-on, attempting to provide an alternative theoretical basis for the decision to vote to that established by Riker and Ordeshook (1968) that does make it rational to take account of the instrumental benefits of voting. A model built on these ideas would include variables currently employed only in aggregate-level studies. If the aggregate-level findings are to be believed, such factors should add to the individual-level variance that is currently explained by effects attributed to resources and mobilization, and they might even replace some of those effects. The Relevant Contest: Local or National? So far we have talked as though all elections were national events, but in some electoral systems the actual election that is participated in (or not) is a local event: an election that sends to the nation’s legislature one representative (sometimes several) from a particular district or constituency. It is an open question to what extent in such elections the stimulus that voters respond to is national or local. If we assert (as we did previously) that turnout variations can be due to “something about elections,” then which election is the one whose character will motivate voters – the national or the local one? For example, the closeness of the race nationally has been found to motivate turnout (or the lack of it); but what about turnout in the local race? It is clear that people do pay attention to the local race, since much work on tactical voting has shown that people are aware of the character of the race in their local constituencies, as already pointed out. Research on American congressional elections has shown that the number of uncontested seats and the average margin of victory have had strong effects on turnout in congressional elections since 1830 (Franklin and Evans 2000), and an analysis of British constituency-level turnout at the 2001 26

We will consider these matters at length in Chapter 6.


31

election (Clarke, Stewart, and Whiteley 2002) is compatible with the same interpretation. Still, many people may not be aware of the specific circumstances in their local constituency, so the closeness of the race nationally may still be a source of turnout motivation even in countries with the sort of electoral system (a majoritarian or “first past the post” electoral system) that makes the local race a relevant matter.27 In countries with proportional representation of any type, the character of the local race should not be relevant, because a major objective of such systems is to ensure that every vote counts, irrespective of where it is cast. Solving the Voter Turnout Puzzles In this chapter we have set ourselves a daunting task. Any answer to the question “Why is turnout declining in established democracies?” must be such as to also provide answers to the various puzzles enumerated in this chapter. Indeed, the list of puzzles to be solved can be reformulated in terms of a list of constraints that any solution to the voter turnout puzzle must conform to. The list of constraints, moreover, is longer than the list of puzzles because there are constraints to be met that are not puzzling. The additional constraints that any solution must conform to are as follows: 1. Our answer should provide estimates of turnout in different countries that approximate the historical pattern of turnout evolution in each country (remembering that some countries have seen turnout increases as well as declines). 2. Our answer should not only explain the evolution of turnout in each country, but also the turnout evolution for different electoral cohorts in each country (distinguishing new from established cohorts) and differences (at the individual level) between individuals who vote and those who do not. 3. Our answer should show how differences in electoral context (distinguishing elections according to their closeness, decisiveness, and other characteristics) influence not only overall turnout levels but also individual decisions to vote. 27

A close race nationally (one in which the dominance of legislative forces hangs in the balance) may also stimulate awareness of close races – races that might tip the balance nationally – in specific districts or constituencies.

32


In the remainder of this book we will attempt to produce an answer to the question “Why is turnout declining in established democracies?” that does meet all of these constraints and those set out earlier in this chapter and which, in the process, also tells us whether it matters that turnout is declining. Because of the three constraints just enumerated, our research design is a complex one involving datasets constructed at the individual level, at the level of the electoral cohort, and at the country level. All of these datasets contain multiple time points as well as multiple countries and/or individuals, so as to permit both crosssectional and over-time analysis. Chapter 2 takes the bull by the horns and addresses the theoretical basis for expecting people to vote. As already mentioned, this book is founded on the premise that Riker and Ordeshook’s (1968) view of the calculus of voting needs extending; and Chapter 2 proposes that, in certain quite common circumstances, it would be perfectly rational for everyone to vote who cared about an election’s outcome. The chapter sets out the assumptions and logic on which this proposal is built and attempts to show that these are just as reasonable as the assumptions and logic embodied in Riker and Ordeshook’s argument. Chapter 3 addresses the other fundamental building block on which this book’s research strategy is based: that citizens are socialized into the habit of voting or not voting during their first few elections, and that the habit (once solidified) governs the behavior of most members of established cohorts. Only voters in their first few elections are fully responsive to changes in the character of elections that might lead to greater or lesser abstention. This chapter tests the importance of the socialization process by using it as the basis for a hypothesis not previously proposed in the turnout literature: that the lowering of the voting age resulted in a different socializing experience for those facing their first few elections, a socializing experience that would guarantee lower turnout for each cohort in turn that entered the electorate at a younger age than previously. This hypothesis implies progressively declining turnout as cohorts entering at a younger age come to make up larger and larger proportions of the electorates of many countries. Using a cohort-level dataset culled from election studies conducted since the 1960s in six countries – a dataset that includes countries that lowered the voting age at a variety of different points in time – the chapter proceeds to confirm this hypothesis, affirming the theoretical


33

basis on which the remainder of the book is constructed and providing a first reason for expecting gradually falling turnout in elections that followed the lowering of the voting age. Chapter 4 builds on the findings of Chapter 3 by determining which features of an election’s character can be expected to have cumulative effects on voter turnout by socializing new electoral cohorts into different behaviors than preceding cohorts had displayed. The argument of this chapter is that only certain variables engage the generational basis of cumulative turnout change: variables such as those relating to the introduction of compulsory voting or reform of the electoral system that, after changing their values, retain their new values without further change. Only a very few variables have this character (in our data only six). Other variables have primarily short-term influences, affecting the behavior of new cohorts in specific elections but not providing a basis for cumulative effects. The chapter starts with three case studies of countries – Malta, Switzerland, and the United States – whose historic record of turnout evolution suggests the importance of certain types of variables in generating turnout change and uses the logic uncovered in these case studies to develop an inventory of variables likely to help determine the level of turnout. Chapter 5 builds on the previous two chapters by putting together into a single model the cumulative and short-term effects that we expect to see on turnout and by investigating their power to explain turnout variations in all twenty-two countries that have held elections continuously since World War II. The model explains some 82 percent of the variance in turnout in the 350 elections conducted in these countries during this time period and confirms the conjectures made in Chapter 4 about the differential impact of short-term and cumulative aspects of the character of elections. Chapter 6 changes the focus of the investigation by asking, what are the implications of our cohort and national-level findings for models of electoral participation at the individual level? If changes in the character of elections have short-term and cumulative effects on turnout, this has to be because individual citizens are aware of variations in electoral competitiveness, and other features of the character of elections, and are affected by these variations in terms of their propensities to vote or not in particular elections. But if this is the case, then most conventional models of electoral participation at the individual level have been

34


misspecified by failing to include measures of the character of elections as independent variables (exceptions are to be found in Franklin, van der Eijk, and Oppenhuis 1996; Norris 2002; Perea 2002). Indeed, it may be that individual-level models are inherently misspecified unless they contain data for multiple elections, since variables pertaining to the character of elections are constant for data collected at the time of any single election in one country. Individual-level models will also have been misspecified in past research by failing to take account of the fact that some individuals (those facing their first few elections) are more strongly affected than others by short-term influences on electoral behavior. In this chapter we employ survey data, collected in Germany at every election since the 1960s, to develop and test a model of participation at the individual level that takes account of both of these phenomena. The model finds significant and strong effects from electoral competitiveness on individual-level decisions to vote and yields different estimates of the effects of variables that traditionally have been included in models of electoral participation. More importantly, it puts a new complexion on the interpretation of certain individuallevel effects, showing that certain variables used in past studies gain at least part of their importance by transmitting indirect effects from the changing character of elections. Chapter 7 estimates the extent to which turnout change in twentytwo countries can be accounted for by our models. The chapter focuses particularly on turnout decline and evaluates the contributions of different variables to such decline as occurred in different countries, distinguishing between the direct and indirect effects of short-term and cumulative forces. Finally, Chapter 8 sums up the findings of the study and discusses the implications for the health of electoral politics in established democracies. It concludes with an assessment of the extent to which turnout decline in different countries can be regarded as incidental to reforms (such as the abolition of compulsory voting in certain countries) or as an accidental by-product of other developments (fractionalizing party systems, for example), and it discusses the possibility that politicians may themselves have played a part in bringing about consequential turnout decline through their success in insulating themselves from electoral verdicts in certain countries.


35

A Note on Statistical Tests In this book, our tests of hypotheses report levels of significance (onetailed, since we know the direction of effects that are expected), but our interest is primarily in the magnitude of the effects. We are trying to explain the changes in turnout that have occurred historically in twenty-two countries over fifty-five years, so the effects of independent variables are assessed primarily in terms of the extent to which they help us to understand those changes. Recent turnout change has not been extensive – most countries have seen less than a 10 percent change in turnout over the period (see Table 1.1). Small changes in turnout imply small effects – often effects that are too small to achieve statistical significance when the election is the unit of analysis. However, those effects might be exactly what our theories lead us to expect. If we expect a 1 percent effect on turnout and find exactly that effect, then this is a success in prediction. Rejecting the finding as not statistically significant would be to make a Type II error (rejecting a hypothesis that is in fact true) in terms of statistical inference.28 Because we are more interested in the magnitude of effects than in their statistical significance, we do not employ multilevel models in this book. The benefits of such models are mainly in yielding tests of significance that are deemed more accurate when some variables are measured at a higher level of aggregation than other variables.29 However, there are costs in using multilevel modeling techniques in terms of ease of data manipulation and flexibility in testing alternative models. With 150,000 cases in our survey data files that had to be aggregated in various ways for analysis, we placed a premium on ease of data management and model respecification. Since the accuracy of parameter estimates would not be affected, and parameter estimates are what primarily interest us in this book, a decision was made to employ more conventional analysis methods. 28 29

We will not be asking the reader to swallow many such findings – only one, and the finding in question is by no means critical to the argument of this book. Though it is not clear why the level at which the data are collected should be critical, see van der Brug, van der Eijk, and Franklin (2003).

2 A New Approach to the Calculus of Voting

In Chapter 1 we saw that the attempt to explain voter turnout without reference to the instrumental motivations of voters gives rise to a number of puzzles, and we asserted that those puzzles would be easier to resolve if we could derive a calculus of voting that made it rational for individuals to take account of the political benefits they might enjoy as a consequence of the electoral victory of one party or candidate rather than another. It is perhaps somewhat brave of us to think that we can, in this book, resolve a paradox that has baffled some of the best minds in the profession over several decades. However, it is clear that contemporary political scientists know a great deal about the voting act that was not known to Riker and Ordeshook when they published their seminal article in 1968. With the benefit of this knowledge (and building on several advances in rational choice theorizing about individual decisionmaking in the context of collective action) we are able to put forward a rationale for the voting act that makes it quite understandable that voters would turn out in greater numbers in elections that aggregate-level findings tell us should indeed be high-turnout elections.1 We acknowledge, as have many following Downs (1957) and Olson (1965), that voters face two collective action problems when it comes 1

Much of this chapter was presented at the annual meeting of the Midwest Political Science Association (Fotos and Franklin 2002). We are grateful to panel participants for helpful comments.

37

38


to voting: gathering enough information to form a preference among the candidates or issues, and expending the effort required to cast a ballot. The outcomes of elections are quintessential public goods that all citizens enjoy (or not) regardless of whether they cast a ballot. Among a large and presumably anonymous mass electorate, the temptation to free-ride on the electoral efforts of others appears enormous. The so-called impossibility result – a positive turnout in the face of theory that predicts zero – has sparked a great deal of productive inquiry. We begin our own inquiry with the individual decision-making model proposed by Riker and Ordeshook over thirty years ago. We retain the individualistic form of the model but proceed to “layer on” a number of relevant contextual factors. These factors effectively transform the model from a strictly individualistic calculus to one that is infused with considerations of strategic interaction. The proposed model is very much under the influence of the rational choice approach to political theory. However, our aim is not to achieve theoretical purity. Like John Aldrich’s (1995) aim in developing a theory of political parties, the aim of this chapter is to develop a theoretical account of voter turnout that can help us to make sense of the widest possible array of empirical findings. In what follows we will first review the assumptions made by Riker and Ordeshook. We then survey what contemporary political science knows about turnout that was not known thirty years ago, sifting from among the vast array of findings a small number that have potential relevance for the individual decision to vote. On the basis of the resulting (larger) set of considerations, we build a more extensive model than Riker and Ordeshook’s and discuss its implications.

The Riker and Ordeshook Model Riker and Ordeshook (1968:25) state that the net benefit of voting to any individual is equal to the benefits of the preferred election outcome to that individual times the probability that his or her vote will affect that outcome less the cost of voting, or R = BP − C,

(1)

where R = the return, or net benefit of voting; B = the benefits of the

A New Approach to the Calculus of Voting

39

preferred outcome; P = the probability that one’s vote will affect the outcome; and C = the costs of voting. The authors conclude, as have others, that the probability that a single vote will affect the outcome of a national election is infinitesimally small and that, absent heroic assumptions about the size of expected benefits, rational voters must therefore gain other “satisfactions” (1968:28) from the act of voting or they would not vote at all. Recently, Ordeshook (1996) has restated the basic model in a twocandidate race. “[A]ccording to rational choice assumptions, the person in question should vote, if and only if |u j − u j | pi j + A > 0

(2)

[where A represents] the net costs and benefits from voting that are independent of the influence a vote has on who wins and loses” (1996:182). In the equation, | | denotes absolute value, ui and uj denote the expected utility of candidates i and j, and pij denotes the probability that the person’s vote is decisive (Ordeshook 1996:182). Because pij is presumed to be virtually zero, Equation 2 emphasizes the central importance of the conclusion that it is rational to vote only if the net utility of voting is positive regardless of who wins. A point-by-point discussion of each term in equation (2) will serve to illuminate this theory of voting utility and provide a basis for our later elaboration of it. The first term (|ui – uj |pij ) implies that the only policy-oriented utility of voting derives from who wins and who loses. Although this assumption makes intuitive good sense, it may not hold up empirically, as we shall see. The A term in Equation 2 combines a number of theoretically distinct costs and benefits of voting. We sort them as follows. A includes selective material incentives that are applied by various outside actors to reward the voter or punish the nonvoter (cf. Powell 1986) and psychic rewards that exist largely independent of the immediate social environment (Riker and Ordeshook 1968).2 2

In Olson’s (1965, 1971) classic formulation, the electorate, due to its large size and undifferentiated output (the cast ballot), is a latent group – one that Olson predicts will be unable to overcome the free-rider problem without the imposition of selective incentives. We note that several nations impose fines or other sanctions on nonvoters. Not coincidentally, the rates of voter turnout are higher in those countries (Powell 1986; Franklin 1996; Lijphart 1997).

40


Externally applied incentives are observable and measurable; psychic rewards fail on both counts (cf. Green and Shapiro 1994:59). Clairvoyance aside, few research scenarios offer a way to distinguish psychic rewards from the unexplained variance in any instrumental model of voter turnout. The model that we propose omits psychic rewards. A number of unspoken assumptions also underlie the Riker and Ordeshook model. The most important is that the only benefits that count are benefits to the individual who is considering whether or not to vote. Benefits to peer group members or to society as a whole are not valued. Equally, in assuming that one’s vote will not affect the outcome, no account is taken of the role that one’s participation might have in helping to induce the participation of others. In other words, Riker and Ordeshook’s individual is seen as entirely divorced from any social context or peer group that might give her vote value beyond its numerical contribution to the election outcome. Finally, the instrumental benefits of voting are seen purely in terms of the outcome. The idea that voting sends a message as well as contributing to an election outcome is entirely absent from the Riker and Ordeshook calculus.3

What We Have Learned that was not Known to Riker and Ordeshook As already explained in Chapter 1, in the immediate aftermath of Riker and Ordeshook’s article, those who studied electoral participation paid little attention to benefits that might accrue to voters from the outcome of any specific contest. Instead, following Riker and Ordeshook’s lead, they focused on voting as a habit that was inculcated into people during their formative years by their educational experiences, so that more highly educated individuals were more likely to vote. But even in the early work on this topic by Verba and Nie (1971) there are clear hints of ways in which Riker and Ordeshook misconceived 3

This is despite the fact, well known even in the 1960s, that a landslide victory sends a message that gives the victor special legitimacy and freedom of action. Of course, if the contribution of one individual vote toward a candidate’s victory is too small to motivate an individual voter, the same will be true of its contribution to a candidate’s mandate. Nevertheless, this is an additional component in the calculus of voting that will need to be taken into account in our more elaborate formulation of this calculus.


41

the voting act. Verba and Nie develop a “baseline model” of participation that predicted whether an individual will vote from their education and social status. However, they found that two of the strongest forces causing deviations from the baseline model are organizational memberships and group consciousness. Those who belonged to more organizations, and those who were more active in those organizations, were more likely to vote than would have been expected from their (generally high) social status. Equally, those who were aware of their membership in an underprivileged minority were more likely to vote than would have been expected from their (generally low) social status. Later in the 1970s these findings were elaborated to encompass data gathered beyond America’s shores (Verba, Nie, and Kim 1976), but the fundamental findings were reinforced in this work, as in the later work of Wolfinger and Rosenstone (1980). These findings should have signaled to theorists that the calculus of voting needed to take account of the social context in which individuals are embedded. Instead, researchers in due course developed a theory of mobilization (Rosenstone and Hansen 1992; Verba, Schlozman, and Brady 1995) that appeared to account for the importance of social context without the need to update notions of voter rationality derived from Riker and Ordeshook, as already explained in Chapter 1. Even stronger evidence of the importance of social context was to come from the work of Huckfeldt and Sprague (1987, 1991; see also Huckfeldt, Plutzer, and Sprague 1993), who spent much of the 1980s immersed in campaigns conducted in South Bend, Indiana – to the extent of counting the number of yard signs supporting one candidate or another. These works were mainly concerned with documenting the effects of context on individual attitudes, but they also demonstrate conclusively that individuals are immersed in social contexts that matter politically. What we learn from these findings is that the decision to vote is not just a private decision, separately made by each individual. On the contrary, even though each actual vote is separate and secret,4 the decision to vote, for many people, has some fundamental characteristics of a collective act, with implications for the calculus of voting that we will explore later. Indeed, this fact may even impinge on the cost component of Riker and Ordeshook’s equation, giving rise 4

Though whether someone has voted cannot easily be kept secret, a point of indirect relevance to our model.

42


to a possible “cost of not voting” to help balance the costs of voting if associates would be appalled by lack of group solidarity.5 A rather different line of research has implications for the benefits of voting. We already mentioned in passing (see footnote 3) that the number of those voting in support of a winning candidate can send a message about the fervor with which that candidate was supported, with possible implications for legitimacy and freedom of action. This has been known since the dawn of the democratic age and implies that there was at least one benefit of voting known to their contemporaries that Riker and Ordeshook ignored: the benefit of putting your preferred candidate into office with a clear mandate. If we presume that one motivation for voting is to effect policy – which Riker and Ordeshook, following Downs (1957), certainly do assume – this benefit could motivate many to vote even when their vote is not needed to secure the candidate’s victory.6 But research conducted mainly during the 1990s has shown that “sending a message” can go far beyond the desire to give a winning candidate a clear mandate. Another possible message that voters can send is one about the extent to which a losing candidate has support in the population. Indeed, voters whose preferred candidate is sure to win (or sure of not winning) have been found to switch their votes to other candidates with no hope of winning (Franklin, Niemi, and Whitten 1994) apparently to signal the fact that the other candidate’s policies are attractive to them – perhaps in the hopes that their preferred party or candidate will take those policies on board. In elections to the European Parliament, which have no function in bringing about policy change, voters appear to use these elections to 5

6

Note that this is to be distinguished from a generalized psychic need to vote, such as mentioned by Riker and Ordeshook, in being oriented toward a particular group with particular interests rather than relating to a more general civic duty. We already mentioned in Chapter 1 that research into strategic voting (Cain 1978; Abramson, Aldrich, Paolino, and Rohde 1992; Niemi, Whitten, and Franklin 1992; Blais and Nadeau 1996) has shown that people behave as though their vote had instrumental value (cf. Blais 2000). Fowler and Smirnov (2002) go further and, consistent with Riker (1962), assume that a large margin of victory gives a party freedom to radicalize its platform and move away from the middle ground, whereas a close race will encourage it to protect its majority by adopting a more moderate platform. This logic provides an incentive to every potential voter who gains by pushing future party platforms further in the direction of his or her vote. This idea is quite consistent with ours.


43

“send a message” regarding support for policies espoused by parties whose actual participation in government might be anathema to them (van der Eijk and Franklin 1996). Some variant of this motivation appears to account for some of the support received by Ross Perot in the American presidential election of 1992.7 Finally, we need to mention a line of research that gives rise to a whole new conception of the habit of voting. Evidently, to the extent that voting is a habit, people will vote (or not vote) simply because they have done so before (or not done so before), focusing the question of why turnout varies onto those who have not yet acquired the habit of voting or abstention. Recent research in Canada (Blais 2001) and the United States (Miller and Shanks 1996; Lyons and Alexander 2000) demonstrates that the most important group of those who have not acquired the habit of voting are new members of the electorate. There is good reason to expect that these findings hold in other countries as well, as already argued in Chapter 1 of this book. On this basis we might expect that, whether turnout is rising, falling, or oscillating about a more or less stationary mean, the major source of election-to-election change will be found among those voting for the first time. Perhaps more importantly, those whose first opportunity to vote comes at a low-turnout election will retain a profile of lower turnout in subsequent elections, even elections in which yet newer cohorts vote at a higher rate; so the past leaves a “footprint” in subsequent elections that reflects the low turnout of an earlier period. The same sort of footprint will be left by abnormally high-turnout elections.

Voters in Context: Obtaining Information, Eliminating Uncertainty How does this new knowledge about the foundations of the voting act affect Riker and Ordeshook’s model of rational voting? The most 7

In 1992, a disproportionate number of Perot voters rated the federal budget deficit as one of the three most important issues (Alvarez and Nagler 1995). Although Perot almost certainly lost votes because of the “wasted vote” argument, he received 5 percent of his vote from people who reported that they would not have voted for him if they thought he could win (Abramson et al. 1992:359). For these individuals, the motivation was presumably to show support for deficit reduction or other Perot policies that major parties were felt to be ignoring.

44


table 2.1 Three Levels of Political Context and the Types of Information They Provide 1. Institutional Context a) Policy Consequences of Election b) Benefits and Costs of Voting 2. Temporal (or Campaign) Context a) Electoral competitiveness b) Size and Nature of the Stakes c) Reinforcement of (or Appeals to) Group Awareness 3. Social Context a) Group Awareness b) Identification of Group and Self Interest c) Voting Intentions of Political Discussants d) Social Consequences of Voting and Abstention

important implication, we believe, is to require any contemporary model that strives to explain turnout variations to include a role for context. The evidence that at least some voters behave as though rationally would rule out any model that did not predict turnout variations based on differences in the context (institutional, political, and social) in which individuals are embedded at the time of an election. As suggested in Table 2.1, the institutional context reflects the legal and constitutional arrangements that determine the costs of voting (Blais 2000) and the likely benefits in policy consequences (or net return) of election outcomes (Powell 1986; Jackman 1987; Franklin 1996; Franklin and ˜ 1998). Hirczy de Mino Comparisons across time and political boundaries indicate that institutional factors influence the salience of elections and, by inference, the rates at which voters turn out. It seems that citizens generally understand the “stakes” of any given election, either through their political discussants, media coverage, or the efforts of the respective campaigns (Berelson, Lazarsfeld, and McPhee 1954; Huckfeldt and Sprague 1987; Lodge and Steenbergen 1995; Bowler and Donovan 1998).8 Turning to the temporal context, campaigns and their news coverage are prominent features. Parties and candidates advertise the 8

Riker and Ordeshook do of course take account of the policy consequences of an election (though they heavily discount its importance). Here we view this factor as a source of information rather than as a motivation to vote.


45

benefits of their policy positions. News organizations report on the closeness of the race and the disagreements among the contenders. It seems incontestable that voters who watch the evening news or listen to talk radio (even occasionally) can gather this information with little difficulty (cf. Downs 1957). The third source of political information is the voter’s social context. People work in certain occupations, live in particular neighborhoods, belong to churches or civic organizations, and engage in recreational activities. These almost unavoidably involve social contacts that reinforce group awareness, encourage individual voters to align their voting intentions with group interests, and inform voters of the likely voting behavior of other group members (Berelson, Lazarsfeld, and McPhee 1954; Finifter 1974; Huckfeldt and Sprague 1987, 1988). Huckfeldt and Sprague detail one particular effect of social networks on the flow of political information that is especially relevant to our analysis. The choice of political discussants is socially constrained (1988:470), meaning that the information available to individual voters, and the gloss that this information receives from discussants, tends to reinforce preexisting political preferences rather than diluting them (cf. Berelson, Lazarsfeld, and McPhee 1954). The likely voting behavior of other group members is also of particular relevance, since the mere intention of others to vote can suggest a widespread belief that, with conscientious behavior by all group members, the group can prevail.9 These contextual influences have important implications. As Green and Shapiro (1994:56f) artfully point out, few rational choice models of voting predict voter turnout that is both rational and greater than zero. One promising exception is a multiperson game that formally accounts for the strategic interdependence of voters. The model has one problem: Its capacity to predict nonzero turnout “disintegrates in the presence of a ‘relatively small degree of uncertainty’” (1994:57, quoting Palfrey and Rosenthal 1985:73). Palfrey and Rosenthal (1985:62) identify three 9

Of course the act of voting generally comes too late to motivate others, but the expressed intention to vote might well have been timely, and the temptation to free-ride on others motivated by one’s expressed intention will be tempered by the knowledge that failure to vote could reduce the credibility of one’s stated intention on some later occasion. Elections are not one-shot games. They are repeated games, and actions taken at one election have implications for future plays.

46


types of uncertainty that affect predicted turnout: uncertainty over the alternative policy outcomes of the election, uncertainty over whether other citizens will vote or abstain, and uncertainty over the preferences of other voters. We believe that, by taking account of the three levels of context just described, our model of voter rationality can include assumptions that effectively dispel these three forms of uncertainty for a large number of voters.10 As a consequence, we also believe that Palfrey and Rosenthal may have been closer to resolving the turnout paradox than either they or their critics acknowledge. But, before we show how this works in practice, we need to add one more way in which a contemporary theory of voter rationality should take account of new knowledge.

Schelling’s “Sufficiency” Model: Rational Voting in the Context of a Coalition Riker and Ordeshook, following Downs (1957) and followed by many in the rational choice tradition, assumed that the relevant pool of votes to which the individual would consider adding her own vote was the pool consisting of all votes cast. Our survey of the contexts within which voters operate suggests that the relevant pool may in reality be the pool of like-minded voters with whom the individual identifies. Like-minded voters face “a uniform multiperson prisoner’s dilemma” (Schelling 1978:218 – italics in the original). All are likely to be better off if a sufficient number vote. If not enough vote, none are better off and those who do vote are worse off than the nonvoters to the extent of the costs of voting. According to Schelling, a uniform multiperson prisoner’s dilemma (or MPD) meets the following conditions: 1. There are n people, each with the same binary choice and the same payoff. 2. Each has a preferred choice whatever the others do; and the same choice is preferred by everybody. 3. Whichever choice a person makes, he or she is better off, the more there are among the others who choose [the same] alternative. 10

We will show later that there is another form of uncertainty that may not be dispelled in the same way but whose presence actually enhances the probability of the voting act.


47

4. There is some number k, greater than 1, such that, if individuals numbering k or more choose their preferred alternative and the rest do not, those who do are better off than if [none of them had] chosen their preferred alternatives, but if the number is less than k this is not true (1978:218).

The important parameter is the term k. It represents the minimum coalition size necessary for any group of voters to achieve its collective aim, that is, the election result that assures a favorable policy outcome. Schelling proposes a form of the MPD, called “sufficiency,” that models the strategic interdependence of a group of similarly situated voters (1978:239–40). His example of a hypothetical committee meeting illustrates the relationships among turnout, individual incentives, and collective benefits. A more familiar example is the committee meeting. Everybody suffers if nobody goes; it is not worth going unless there is likely to be a quorum; over some numerical range, one’s presence makes enough difference to make attendance worthwhile; and if the meeting is large enough, there is no need to give up the afternoon just to attend (1978:240).

Applied to the logic of voter turnout, attending the meeting is analogous to going the polls. The likelihood of a quorum is analogous to the probability that the respective bloc of voters will constitute a winning coalition. The phrase “if the meeting is large enough” captures an individual’s temptation to free-ride on the efforts of other similarly situated but more conscientious voters. This brings us once again to the fundamental theoretical question: Can any model generate a reasonable expectation that a sufficiently large number of putatively anonymous and rational voters will individually overcome the temptation to free-ride on the votes of others in their respective voting coalitions, especially if the coalition contains thousands or millions of voters? Schelling defines the term “coalition” in a way that helps answer the question: [A coalition] is a subset of the population that has enough structure to arrive at a collective decision for its members, or for some among them, or for all of them with some probability . . . (1978:224).

To illustrate how such coalitions work in a practical setting, Sheplse and Bonchek (1997:226–33) offer the quaintly named “Possum Hollow

48


Rod & Gun Club” analogy. Possum Hollow members, contribute annual dues and volunteer to work on the club grounds one weekend a year. The club has n members, of whom k are required to honor their work commitment if the club grounds are to be properly maintained. According to the authors, the club is more likely to overcome the free-rider problem when the number of workers needed for annual maintenance is large in relation to the club’s membership base (when the gap between k and n is small). Under these circumstances, each member is more inclined to believe that his or her contribution will be decisive in ensuring the club’s welfare, and the likelihood will be high that enough members will show up on the appointed day (1997:229). As the number of members increases beyond k, the temptation of each of them to free-ride grows until the probability that less than k will show up becomes a near certainty. As long as the collective benefits of the annual cleanup exceed the total costs to individual members, the gap between k and n “is a crucial determinant of whether or not this group is able to get its act together” (1997:230). The example of Possum Hollow is directly applicable to a contextual analysis of the voting decision. The institutional context informs the voter of the costs of voting and the policy effects of the election (Powell 1986; Franklin 1996). The parties or candidates emphasize the magnitude of the stakes (|ui – uj |pij ) and make appeals that reinforce feelings of group membership (Berelson, Lazarsfeld, and McPhee 1954; Uhlaner 1989; Franklin 1992). Media coverage focuses on the closeness of the “horse race” and, by inference, reveals the size of k. The voter’s social network and perhaps local party or other formal group activities inform her of the likely effect of the election on similarly situated voters and of the voting intentions of fellow group members (Huckfeldt and Sprague 1987, 1988). In other words, one need not make heroic assumptions about the average voter’s intelligence or civic virtues to suppose that he or she can make reasonably accurate estimates of the benefits and costs of voting and whether the size of the group of like-minded individuals (k) is large enough to potentially constitute a winning coalition. This is the great contribution of contextual analysis to instrumental theories of voting. It maps a flow of political information and generates a logically complete and observable path to rational collective action.


49

The considerations we have set out can be formalized in terms of a number of assumptions that add to (or modify) Riker and Ordeshook’s explicit or implicit assumptions, as follows: A1. There is a natural limit on the extent to which turnout can vary from election to election, based on the number of voters who are already set in their (voting or nonvoting) ways. A2. Among those who have not (yet) become set in their ways, the benefits of voting center on the likelihood that a given vote will encourage and contribute to full turnout by all the members of a potentially winning coalition of voters. A3. The importance of the policy stakes will condition this likelihood, as will the degree of electoral competition and the extent of uncertainty about the true likelihood that all members of the potential coalition will vote. A4. For some of those who are not motivated by the opportunity to determine the outcome of the election, the opportunity to vote can still be used to “send a message” to party leaders or the media about concerns and considerations that the election provides an opportunity to voice. Such opportunities are more frequent in multiparty (or multicandidate) races than in twoparty (or two-candidate) races.11 A Contextualized Model of the Decision to Vote Based on the assumptions outlined in the previous section, we can construct a turnout model for the average voter. To simplify the exposition, we assume a two-candidate race of Dot versus Ruby, denoted as D and R.12 Our voter is a member of a representative group distinguished by the fact that most members of the group prefer Dot over Ruby. The preference of the ith member is expressed as the utility (U) 11

12

Even in a two-horse race, voters may want to contribute to a landslide victory, as already indicated. Alternatively, in such elections voters who anticipate that their preferred party will gain an overwhelming victory might occasionally vote for the (anticipated) loser so as to help balance an anticipated lopsided result, warning their preferred party not to take victory for granted (and not to adopt extreme policies). Ordeshook (1996:183) demonstrates how to generalize the model to encompass races with three or more candidates and thus to introduce the possibility of strategic voting.

50


of a Dot victory minus the utility (U) of a Ruby victory, or Ui (D) – Ui (R). If Ui (D) – Ui (R) is positive, the voter is aligned with the majority of group members; if it is negative, he or she is in the minority. The absolute value represents the intensity of the voter’s preference. If Dot is to win, k of her n supporters must vote for her. The probability that k supporters will turn out and vote for Dot is denoted as π k . Since few election outcomes are sure things, the model includes a second probability term. It is the likelihood that Dot wins given that k group members vote for her, or π D|k . So the expected utility of the election for the ith group member is Ui (election) = πk ∗ π D|k(Ui (D) − Ui (R)).

(3)

Equation 3 reveals the following about the voting behavior of the subject group. If the election is likely to be close, then π D|k is maximized when k = n. In a cliffhanger, a Dot victory requires the unanimous support of her voting coalition. Unanimity is a “focal point” wherein each voter believes that his or her vote is decisive and acts accordingly (Shepsle and Bonchek 1997:230).13 This proposition generates the term for the probability of decisiveness that will be included in Equation 4. For the present analysis of Equation 3, the requirement that k = n generates conditions that maximize the probability that k group members will turn out to vote for Dot, given that all members of the group prefer her to Ruby.14 Let us now address the probability term that has proven most problematic to individual models of voter turnout. It is the probability that the ith voter will cast the decisive ballot, denoted as π d . The probability of decisiveness is a function of two other interacting probabilities, 13

14

We believe that this approach to understanding why people vote is an improvement on the “unity principle” (Muller and Opp 1986; Finkel, Muller, and Opp 1989), which arguably is not compatible with a rational choice approach (Whiteley 1995). The probability that k = n votes will ensure a victory for Dot (π D|k ) is maximized when the election is close. Assume, for example, that the subject group is unanimous in its support for Dot and that the number of voters in the group (n) is exactly two more than one-half of the total number of all voters (denoted as N), or n = N/2 + 2. Thus, the winning coalition, given that a simple majority rules, is k = n − 1 votes. Furthermore, π D|k = 1 when either n or n − 1 voters in group A turn out. The same applies when k = n − 2, and so on. The same logic can be used to account for cases when k > n. Dot’s chances of picking up enough votes to win a close election are maximized (even if they remain less than 1) when all of her supporters vote.


51

the probability that the preferred candidate wins, given that her coalition turns out to vote, and the probability that a sufficient number of members of the coalition do in fact vote. These relationships accurately reflect the strategic electoral environment. If Dot appeals to a hopelessly small minority, then her chances of winning are zero and the expected utility of voting for her is zero. On the other hand, if there is a good chance that k votes will put Dot over the top, then the voter must estimate the likelihood that k of her supporters will show up. Finally, the voter must decide the likely effect of his or her known intention to vote on other voters in the coalition, and hence her contribution to ensuring that k voters do in fact show up.15 Uncertainty plays a different role in this calculation than in the Riker–Ordeshook calculation. If individuals are confident that their preferred coalition can win but are unsure about its actual margin of victory, this maximizes the probability that their own vote will be critical – not just in terms of its own singular addition to the total number of votes cast by coalition members, but also as an inducement to other members of the coalition to vote.16 So the calculus of voting is quite different for socially connected people. If each vote has a motivating impact on other members of a group, then each vote effectively counts more than once. People in social networks would also incur costs of nonvoting because other members of their group care whether they vote or not. By contrast, for those not involved in social networks, their vote is only worth one single vote, and, furthermore, they incur no social costs from freeloading. So, the benefits of voting and the costs of nonvoting are higher for socially connected people.

15

16

It is possible that many voters see themselves not as part of a nationwide potential majority but as part of a potential majority among a much smaller group – the group of their own acquaintances – and unconsciously confuse the national coalition relevant to the election outcome with the smaller group that is salient to them. This smaller group is the one containing people whose good opinion they value. The confusion could only be partial, however, because it is the closeness of the race nationally that governs behavior. We do not develop our model on this basis. Every individual who makes their intention to vote known thereby increases the odds that other members of the same coalition will see their votes as potentially pivotal. One might argue that intentions do not actually have to be carried out, but in practice most of the costs of voting have already been incurred by the time a coalition member makes their intention known (see also footnotes 4 and 9).

52


This is the key departure of our analysis from the Riker and Ordeshook model. In the Riker and Ordeshook model, individuals calculate the probability of decisiveness in relation to the entire electorate. In that context, only the single marginal vote required for victory is considered decisive. The rational voter in our model calculates the probability of decisiveness in relation to his or her social reference group. The important thing here is not that the social reference group is smaller than the entire electorate but that, unless the election outcome is a foregone conclusion, each of the group members considers their vote to be potentially decisive. The critical difference is that uncertainty leaves the atomized voter with no reason to vote while uncertainty leaves the member of a potentially winning coalition with no excuse for not voting.17 In equation 4 below, we combine the probability terms (π d ∗ π k ∗ π D|k ) and denote the combined probability of factors that affect the election outcome as o . Given the expected value of the election as presented in formula (3) and including the terms (CVi and CNVi ) that account for the ith voter’s costs of voting and not voting, voter i’s expected utility of voting is Ui (voting) = o (Ui (D) − Ui (R)) − (CVi − CNVi ).

(4)

In some contexts, the way voters vote will influence public policy regardless of the outcome of an election. This is most readily apparent following elections that themselves have no immediate policy implications, such as elections to the European Parliament. Examples of policy effects from extremist or otherwise unexpected votes in such elections are to be found in van der Eijk and Franklin (1996). Such effects may be intentional. Voters sending messages may be members of organized groups that seek to demonstrate their power at the polls to gain policy concessions (cf. Uhlaner 1989; Franklin, Niemi, and Whitten 1994). Or they may be attracted spontaneously to alternative parties as a way of sending an “angry letter” to incumbents, given assurance by the polls or other indicators of public opinion that a sufficient number of other voters intend to do the same. If only one voter “sends a message,” he or she is unlikely to be heard. However, the number of voters who 17

Our substitution of the word “excuse” for the word “reason” in the second half of this sentence is intended to emphasize the fact that the member of a potentially winning coalition will be motivated in part by the desire to maintain good relations with other members of her group.


53

want to send a message need not comprise a winning majority as long as they are sufficiently numerous to command the attention of the media and political elites. The purpose of this discussion is not to specify exactly how numerous they must be, rather, it is to suggest that a fully specified model of voter turnout should include a term for the value of political messages.18 A voter may also intend by her vote to send a message to other group members. As indicated earlier, the intention to vote can motivate others to do the same, and the act of voting may be conceived as the consequence of sending such a message (see footnote 9). Such an act can be seen as a demonstration of confidence in the potential of the voter’s coalition to win – not confidence that the coalition will win (that would make other votes unnecessary) but confidence that, if all members of the coalition pull their weight, their cause can prevail.19 The model in Equation 5 includes a term that fulfills both of these requirements, denoted as Mi . We model it in the same way as the previously mentioned election outcome component. To reflect the strategic interdependence of the various communication-minded voters, we include a combined probability term, denoted as M . M is the product of two probabilities, that a sufficient number of voters turn out and vote the desired way, and that the target of the message (whether a politician or group member) recognizes the message and responds in the desired way. The value of voting as a form of political communication for the ith voter is thus equal to M times Mi . Equation 5 presents a comprehensive utility function regarding the vote decision facing those who have not yet acquired the habit of voting or nonvoting.20 All else being equal, the ith member of this group will turn out when the following equation has a value greater than zero: Ui (voting) = o | Ui (D) − Ui (R) | + M (Mi ) − (Ci − CNVi ).

18

19 20

(5)

Though we believe that any complete picture of the calculus of voting should take account of political messages, we will not attempt in this book to measure this aspect of the message component in the utility of voting. This second aspect of the message component can be assumed to magnify the effects of whatever forces might enhance the utility of voting in Equation 4. Equation 5 incorporates the utility calculus of all voters, not just the “pro-Dot” coalition discussed previously. As a consequence, we express the expected utility term [Ui(D) − Ui(R)] as an absolute value so as to ensure that it is positive for all voters who have a preference for either of the two candidates.

54


Equation 5 may take a range of positive and negative values. It may also take a value exactly equal to zero, in which case we presume that the voter is indifferent about whether to vote. As an analytical convenience for what follows, and to avoid the nasty problems that accompany interpersonal comparisons of utility, we will stipulate that the equation may take any value less than or equal to +1 or greater than or equal to −1.21 For ease of reference, henceforth we refer to the entire equation 5 by its net value, Ui (voting). Our model of voting requires one final term to take account of the probability that certain individuals will vote even in an election that yields them low utility (or fail to vote even in an election where voting would have yielded them high utility) on the basis of what is effectively a “standing decision” reflecting the individual’s previous experience with the political system, as explained earlier. The voter’s formative political experiences can be thought of as instilling a “Bayesianesque” set of prior beliefs about the benefits of voting. Absent strong disconfirming evidence, the “cognitively economizing” rational voter (or nonvoter) will have little inclination or reason to change his or her standing decision about the net benefit of voting.22 This standing decision for the ith individual is denoted in the following equation as Hi . Pi (vote) = Hi + {(1 − Hi ) ( f ( Ui (voting) )}.

(6)

Equations 1 through 5 were individual-level utility functions expressed as statements of benefits and costs. Equation 6 is formulated instead in terms that enable one to test the likelihood of voting via a multinomial regression equation. In Equation 6, the H term designates the 21

22

The reader will recognize this form as a simple adaptation of the von NeumannMorgenstern expression of cardinal utilities as “equivalent lotteries” with a value of 0 to +1 (Morrow 1994:23, 32). In this instance, we extend the range to −1 to reflect the inclusion of the cost term in Equation 5. Note that we remain true to our original aim of dispensing with psychic rewards such as civic virtue in the effort to develop a model of voting as an instrumental act. Group solidarity at the polls, even (perhaps especially) when their votes do not serve to put a party or candidate “over the top” are well known to be a valuable commodity for group leaders, as seen whenever such leaders claim the power to “deliver” their voters to some candidate or party – generally in return for policy or other promises. Political leaders have every incentive to engage in activities that reinforce the standing decisions of dependable loyalists, including appeals to civic virtue or other material and nonmaterial considerations.


55

probability of voting based on habitual considerations (along with other semi-fixed influences on the likelihood of voting) for the ith individual; and the probability of not voting (1 – Hi ) is used to condition, for the same individual, the utility of voting U (standing in for the whole of Equation 5) because the measure of utility only applies to the ith individual insofar as that individual’s vote (or failure to vote) is not already determined by habitual factors. The utility of voting will also be conditioned by a function that we leave unspecified in Equation 6, whose effect will be to turn those utilities into probabilities through some form of logit transformation.

Subjective Probabilities and the Strategic Calculus of Multiperson Games Palfrey and Rosenthal (1985:62) model the strategic interdependence of voters as a game in which “Nature” moves first by assigning costs and preferences to all voters and the voters move simultaneously with incomplete information – that is, they cannot observe the moves of the other voters or of Nature. As a consequence, game equilibria take on the characteristics of the “one-shot Prisoners’ dilemma,” where voters infer the worst about what the other voters and the candidates will do. Our combined probability terms embody a similar strategic dynamic, but the contextual approach allows the analyst (and the voter) more opportunities to evaluate the likely behavior of the other players and hence to eliminate a substantial portion of the strategic uncertainty.23 The voter’s frame of reference when deciding whether and how to vote is fundamentally richer in the contextual model than it is in other more individualistic approaches. The model represented by Equations 1 and 2 presumed that voters were atomistic decision-makers with minimal knowledge of their strategic environment. Palfrey and Rosenthal’s (1985) game-theoretic adaptation of that original model offered a fuller treatment of the strategic environment but restricted the information 23

Ostrom, Gardner, and Walker (1994) demonstrate that, in experimental settings, opportunities to communicate (so-called cheap talk) can support Pareto superior (but not Pareto optimal) cooperative outcomes in multiperson games that replicate self-governing, common property situations.

56


available to the voters. Assumptions made there about limited knowledge and simultaneous play left voters in ignorance of the voting intentions of others in their reference group(s) and of the electoral stakes for these groups. Our analysis, by contrast, suggests that voters are neither misinformed nor irrationally optimistic when they behave as though the expected value of their vote is substantially greater than zero. This analysis builds on Schelling’s (1978) model of the conditions under which coalitions achieve sufficient turnout. Contextual analysis of the flow of political information suggests that voters can know quite a lot about their strategic environment and the likely voting behavior of like-minded voters. Moreover, pervasive media coverage of the “horse race” aspects of campaigns (even in the absence of the almost daily polling that takes place today in advanced democracies) imply that voters can infer the likely voting behavior of very large groups. The human capacity for some forms of collective action (and the maximum size of voter groups that can overcome the free-rider problem) may be quite a bit greater than has generally been supposed.24

Discussion The logic informing this model is inductive. It is built on the recognition that Riker and Ordeshook’s model fails to accord with specific research findings, as suggested in Chapter 1 and set out in detail in this chapter. We proceed from these findings to general propositions about voting behavior. It is thus entirely possible that our model reflects an unintended tailoring of the variables to preexisting data. So it is important that its implications be tested on data other than the data on which it was built. The implications that are most distinct, in terms of the testable propositions that they give rise to, are those regarding uncertainty and inertia. We have already mentioned that uncertainty plays a role in our model that is different from its role in Riker and Ordeshook’s model. When the character of an election is such that there is no uncertainty 24

Olson’s (1965) typology of groups does not formally estimate or predict the respective sizes of privileged, intermediate, and latent groups, although he does admit the possibility that the degree of anonymity (or its absence) can influence the potential for collective action.


57

as to the numerical outcome of that election, the election is a foregone conclusion and there is little reason to vote. Only in elections where there is uncertainty as to the numerical outcome does each voter who cares about that outcome have reason to believe that their votes might make a difference. But uncertainty also plays another role, one closer to the role it plays in Riker and Ordeshook’s model. This is the strategic uncertainty referred to earlier. Only when this other form of uncertainty is low (for example, because the policy consequences of an election are clear) will people care about the numerical outcome. So electoral competition is maximized when strategic uncertainty is low while uncertainty about the outcome is high. In later chapters we will need to distinguish between these two different sorts of uncertainty. The combined effect is expected to lead to turnout being higher in competitive elections and lower in elections where competition between parties (whether in terms of the closeness of the race or the likelihood of the election leading to substantively important policy change) is less. The other critical feature of our model that makes it different from previous models is the role of inertia. Were it not for inertia, turnout would vary radically from election to election as all cohorts reacted to the very different character that successive elections can have. Arguably, it is only because of inertia that it has taken us so long to understand the driving forces behind turnout change. In subsequent chapters we will focus our attention on those members of the electorate least affected by inertia – members of new electoral cohorts – in the hopes that the relatively greater variability of turnout among these cohorts will permit us to identify the independent variables responsible for that variability. Our model predicts that turnout will be highest when electoral competition is greatest – and when electoral competition has been uniformly high throughout the lifetimes of all members of the electorate. Our model is also consistent with the idea that a networks of formal and informal group memberships are required if members of the population are to care about election outcome and become aware of the closeness of the race and other relevant factors. But notice that the model does not require any particular level of social capital (as social networks are now generally referred to – cf. Putnam 1995, 2000, 2002); and nor does it lead us to expect turnout to vary with the level

58


of social capital.25 Indeed, our model does not contain any terms for social capital. Rather, the existence of sufficient social capital is a prerequisite for the model to function – an assumption upon which it is built, not a variable that it contains. This assumption will be tested in due course, when we look to see whether the model can be elaborated through the addition of variables measuring social capital along with other factors suggested in past research to be linked with voter turnout. As it stands, our model implies that turnout will be predictable, in the case of any specific election, on the basis of the character of that election, not the character of the individuals voting in that election. In Chapter 5 we will test this implication using data from all of the elections conducted in twenty-two countries since World War II. But before we do that we need to establish in what specific ways we can expect the character of elections to affect the behavior of individuals. This topic is addressed in the next two chapters. 25

Because social capital has declined in many countries, it is common to blame declining voter turnout on waning social capital. Recent research on changes in social capital in Britain and Sweden (Hall 2002; Rothstein 2002) puts the assumed link with turnout into question, however, by documenting a society (Britain) in which social capital remains vibrant even while turnout has declined, in contrast to the Swedish case, where social capital has declined even while turnout has risen. Putnam himself does not regard turnout decline as a consequence of eroding social capital, however, seeing electoral mobilization as being just as likely to stimulate social capital formation (Putnam 2000).

3 The Role of Generational Replacement in Turnout Change

In Chapter 2 we established a rationale for the voting act that would make it plausible to assume that people are influenced by electoral context when deciding whether to vote, thereby laying the groundwork for resolving several of the puzzles outlined in Chapter 1. This context changes from election to election, depending on something we have called the character of elections – primarily their competitiveness. But we asserted in Chapter 1 that the context of the election has more impact on some cohorts (newer cohorts) than others (older cohorts). Before we can proceed to an analysis of the effects of the changing character of elections on turnout, therefore, we need to address the question of why and to what extent cohorts differ in their responsiveness to the changing character of elections. Until we have done so we will not be able to properly specify any model of turnout change. The idea that generational replacement plays a role in turnout change is one that has not been extensively explored in the literature.1 Newly enfranchised individuals are known to be particularly open to recruitment by new parties and to be largely responsible for such changes as occur in the support for existing parties (Campbell et al. 1960; Butler and Stokes 1974; Nie, Verba, and Petrocik 1978; Inglehart 1977, 1990, 1997; Rose and McAllister 1990; Franklin et al. 1992; 1

Much of this chapter was presented at the annual meeting of the American Political Science Association (Franklin and Wessels 2002). We are grateful to panel participants for helpful comments.

59

60


Miller and Shanks 1996). The same importance of newly enfranchised individuals has been found in regard to turnout (Miller and Shanks 1996): Change in turnout most often comes from a new cohort of voters turning out at a rate that is different from the turnout rate among the previous cohorts when they were new.2 Newly enfranchised individuals are also known to rapidly become immunized against changing their minds if they support the same party at even a quite small number of consecutive elections (Butler and Stokes 1974). The implications of this last insight for turnout change have not been explored, though it may well supply an explanation for the fact that it took fifty years after female suffrage for the gender gap in U.S. turnout to be eliminated (Christy 1987; Norris 2001). Such an explanation would focus on the fact that only women who had not yet learned the habit of nonvoting would have taken full advantage of a franchise extension when it occurred. New female voters, as they entered the electorate after the extension of the franchise, would learn the habit of voting at the same rate as men, but it might take as much as fifty years for the last of those socialized before the franchise extension to leave the electorate.3 Similar processes will presumably also have been at work following other franchise extensions in the United States and elsewhere. In this chapter we establish the role of generational change in voter turnout across six countries that have conducted election studies continuously since the 1960s. We show that incoming cohorts are particularly responsive to changes in the character of elections, being at the forefront of increases in turnout as well as of declines in all countries. Further, we document the tendency of new cohorts to retain the habit of higher or lower turnout gained in their first elections, leaving a “footprint” in the electorate (like the ring of an aging tree) that both records and transmits forward in time the formative experience of the

2 3

In a ground-breaking article, Plutzer (2002) looks at persistence and inertia in the propensity to turn out, a different take on the same concerns that we address here. The idea that increasing female turnout was generational in nature has been well understood (see, for example, Norris 2001), but there appears to have been no understanding that this generational shift might have to do with the socializing power of the electoral experience itself. The diagnosis was rather that “Long-term secular trends in social norms and in structural lifestyles seem to have contributed towards removing many factors that inhibited women’s voting participation” (Norris 2001).

The Role of Generational Replacement in Turnout Change

61

cohort.4 We also confirm the unsurprising fact, already documented in past research, that newer cohorts turn out at lower rates than more established cohorts. On the basis of these findings we set out a rather prosaic proposal that contributes toward an explanation of declining turnout in countries that lowered the voting age (generally from twenty-one to eighteen) in elections after the mid-1960s. In brief, we propose that, by lowering the voting age, politicians in these countries exposed young citizens to the experience of voting who were not yet in a position to benefit from the opportunity. These were individuals many of whom had not yet completed their education, not yet had the opportunity to become established in an occupation, and not yet had the opportunity to establish the links to groups and communities that would give an election relevance (see Chapter 2). The result will have been a lower turnout rate than the same cohort would have registered had their first electoral experience come at a slightly older age, and the learning experience will have had long-term consequences for the future turnout of the cohort. The point goes beyond the conventional wisdom regarding the extension of the franchise to eighteen-year-olds. It has been well understood that the lowering of the voting age would bring into the electorate a group of individuals with lower than average turnout (see, for example, Topf 1995). Indeed, the drop in overall turnout to be expected from bringing this group into the electorate should have been particularly large because the change generally coincided with the entry into the electorate of the post-1945 “baby boom” generation, yielding a cohort twice as large, in most countries, as any prior cohort. If voters aged eighteen to twenty-one are 4 percent less likely to vote (as we shall see) than those aged twenty-one to twenty-four,5 and if the typical electorate in established democracies consisted of thirteen equally 4 5

Though the turnout rate for any given cohort does tend to rise with the passage of time as its members learn the habit of voting (see footnote 7). Most countries hold parliamentary elections every four years, so an entering cohort encompasses those coming of age in the four years prior to each election. The voting age was lowered by three years in most countries, so there will have been an overlap of one year between those voting under the new rules and those who would have voted under the old rules. This is why the twenty-one-year-olds occur in each of the two groups being compared. We will determine the actual difference between these two groups later in the chapter.

62


sized four-year electoral cohorts,6 then the decline in turnout to have been expected from lowering the voting age to eighteen would have been 0.3 percent. If the cohort concerned was twice the size of other cohorts, then the effect would have been to cause a turnout decline of 0.6 percent – still a barely measurable blip in the turnout of most countries. Two concomitants of this blip, however, will have conspired to generate a more noteworthy effect from lowering the voting age. The first is the fact, alluded to in Chapter 1, that the increase in the proportion of young voters was not a one-shot deal. From the end of the 1960s until the end of the 1980s, every cohort that entered the electorates of most countries was larger than usual. Over those twenty years, on average across the twenty-two established democracies in our aggregate-level dataset, electorates increased in size by 47 percent overall – 9.4 percent at each election, on average. Since each election that maintains a constant-sized electorate must on average consist of 8.3 percent new voters to replace those lost through mortality, during this twenty-year period new cohorts were more than twice the size required to maintain such a steady state. Doubling the size of the incoming cohort over a twenty-year period implies a measurable decline in turnout over that period. In the countries for which we have survey data going back to before this franchise reform, members of the electorate aged forty and over turned out at a rate 4 percent higher than those under forty. Doubling the size of the group under forty would cause the turnout to decline by 0.7 percent. Much of that decline would, however, be made good during the final twenty years that it would take for the baby boom generation to totally replace the preexisting electorates in these countries – a period during which the rising turnout associated with the aging process should finally overweigh the falling turnout due to increasing numbers of young voters – a process that should now be starting in many countries. There is, however, the possibility of a second process with more permanent effects on turnout in countries that lowered the voting age, and that process is the subject of the present chapter. 6

Thirteen four-year cohorts encompasses a fifty-year span, about the remaining life expectancy of an eighteen-year-old in 1970. Of course the assumption of equal-sized cohorts is only employed to make the arithmetic easier and will be dropped when dealing with real rather than hypothetical examples.


63

The Socializing Effect of Younger Enfranchisement Learning to vote is costly. Plutzer (2002) has a wonderful description of the costs incurred by would-be first-time voters who not only have to figure out how to register and where to find the polling place, but also have to learn about the contestants and their policy proposals. And they have to do this at a time when their lives are filled with the problems of establishing themselves in adult relationships and circumstances. Our argument in this chapter is that the costs of learning to vote are raised considerably if a person’s first election falls during the period immediately after leaving high school. In most countries people leave high school at about eighteen years of age. The four years that follow are fraught with the problems of early adulthood already referred to. More importantly (in the light of our discussion in Chapter 2), those four years are years in which young adults are only starting to establish the social networks that will ultimately serve to guide their political choice and motivate their vote. For young people in college, the four years may well be spent in a location that will not be their permanent abode, so they have little reason (as well as little opportunity) to establish such networks. Entering the electorate at twenty-one years of age would have been bad enough. On average, such a person would face their first election two years after that, at a point at which those who had attended college would be just starting to establish themselves in a community, while those who had not attended college would presumably be even better situated. But entering the electorate at eighteen means that the average person faces their first election at twenty, while still in college (if they attend college) or while still establishing themselves in their first job otherwise. The costs of learning to vote at the younger age will clearly be higher than at the older age, even if those costs were themselves pretty high. So the experience of facing one’s first election will, for individuals who entered the electorate at eighteen, be less rewarding and more likely to result in a failure to vote. If the socialization process explored in Chapter 1 applies in the realm of turnout, then the fact that those who gain the vote at eighteen have a worse first-time experience than they would have had at an older age will leave an imprint on their later behavior. One of their three opportunities to transition to the habit of voting (to use Plutzer’s terminology) will be more likely to have been wasted.

64


If this is a result of lowering the voting age, then the fifty-year fall and recovery of turnout due to the evolving age structure of electorates is only the beginning of the story. The fact that incoming cohorts are younger than before not only affects the electorate of which they form part, but also the formative experiences of those cohorts. Because their first opportunity to vote yields an experience less likely to be positive for the members of a new cohort than a later initial experience would have been, the socializing effect of this experience will have created a footprint in the electorate of lower turnout than would have been generated for an older entering cohort. In other words, the socializing experience of the act of voting (or of nonvoting) will have tended to lock in the lower turnout of the newly enfranchised eighteen-year-olds, making them permanently less likely to vote than an older entering cohort would have been.7 Cohorts following this initial group into the electorate at four-year intervals will also vote at the same lower rate. Older cohorts, who entered the electorate when the voting age was higher, and who received a more positive formative electoral experience, will at the same time be dying off. So the lower rate of turnout that was generated by the lowering of the voting age will have slowly spread through the electorate as older cohorts were replaced by new ones with a generally lower rate of turnout. A drop in turnout of only 0.7 percent or so when those first eligible to vote at eighteen made up only one of thirteen cohorts (though a doublesized one, as already mentioned) could rise to as much as 5 percent when everyone socialized in an earlier era has left the electorate and all thirteen electoral cohorts have come to consist of those whose first electoral experience was relatively less meaningful than it would have been during that earlier era. These numbers are imaginary. One of the purposes of this chapter is to estimate the actual decline in turnout that was to be expected as a direct consequence of the lowering of the voting age; another is to describe the nature of the footprint resulting from the socializing force of 7

We shall see that individuals do become more likely to vote as they age – an insight that provides the rationale for Rosenstone and Hanson’s (1993) life-experience approach to explaining individual-level electoral participation. However, the initial level of turnout in any group serves as a constraint on the level to which turnout in that group can rise (Plutzer 2002). In this research we take that insight and apply it to groups defined in terms of the age at which they enter the electorate.


65

these electoral experiences. Butler and Stokes (1974) estimated that it took three consecutive elections to immunize a voter against change in partisanship. A single unsatisfactory electoral experience will presumably not immunize the entire cohort against later electoral participation, so, even if the initial impact was 0.7 percent, the enduring impact might be less than 0.7 percent per cohort. We cannot be sure about these numbers until we have done the analysis. What we can be confident in hypothesizing is that there will have been an initial effect, and that this initial effect should have had enduring consequences for each and every cohort entering the electorate at a younger age than previously. These enduring consequences should have resulted in declining turnout in every country that lowered its voting age, and that decline will continue until the last of those socialized in an earlier era has left the electorate. In other words, the decline due to this change in electoral laws has only halfway run its course. Thirty-some years after the reforms were enacted that lowered the voting age in most established democracies, there are still some five electoral cohorts left to go before the decline in turnout set in motion by those reforms will come to an end. Until we have estimated the extent of this decline, there is no point in talking about any other effects on turnout change. In Chapter 2 we suggested a number of features of the character of elections that we believe should affect the level of turnout in a country, but the extent of these effects cannot be estimated until we can include in any properly specified model the effects of an electoral reform that occurred thirty years ago but whose long-term consequences will continue to be felt for twenty more years to come. One of the most important implications of the ideas just set out is that changes in turnout are primarily due to the ways in which new voters (and those voting for the second and third times) react to the character of each election. More established voters are more set in their ways and so have less influence on turnout. Interestingly, in Cambridge, Massachusetts, the voting age in local elections has recently been lowered to seventeen (Bloom 2002). By giving children the vote while they are still in high school, it is possible that the proportion of newly enfranchised voters who will be socialized by a good first-time voting experience will increase. The election could be turned into a meaningful class project for new would-be voters still in high school, with students being graded on their ability to find


66

campaign literature and assess campaign promises. Voting would be a natural part of such a class project, and the fact that everyone in the class was participating would help to ensure its relevance. If this opportunity is taken, in due course the reform will provide us with a natural experiment that may help to confirm the theory presented in this chapter. Meanwhile, we can only employ a quasi-experimental method to confirm our reasoning, taking advantage of natural variation in the character of elections (from country to country and over time) to confirm these ideas.

Hypotheses The hypotheses that follow from our line of reasoning are as follows: H1. The most recently enfranchised cohorts voting in any election are the ones most likely to register changes in turnout, increases as well as decreases, by voting at a rate that is different from the turnout rate for the previous cohort when that cohort was new. H2. Inertia (the tendency to vote at the same rate as previously) will be strongest among established cohorts. H3. Features of an election that drive turnout change (party structure, closeness of the race, voting age, etc.) will have the greatest effect on new cohorts. H4. Cohorts entering the electorate after the lowering of the voting age will vote at a lower rate than cohorts entering the electorate before that reform. H5. Once post-reform cohorts become established, the gap between their turnout and that of pre-reform cohorts will not be eliminated with the passage of time. These hypotheses are logically connected. The first two take the findings of past research regarding party choice and apply them to turnout. They concern the preconditions that, if present, should ensure that H5 will follow from H4. H3 is an implication of H1 and implies the need to take account of other determinants of turnout change in any investigation of the effects of voting age if the latter effects are to be estimated properly.


67

Turnout, percent

95

H

H S g

g S S G

turnout at legislative election

S

g

N

S h

s

g h

n

n

h

g s

h

b

g b

B

h g

b

n

n

b

b

B

g

s

s

s

h n

n N

H B

g

h

N

s

s

b

b

h

B Britain G Germany H Holland (Netherlands) N Norway S Sweden U USA (lower case after lowering of voting age)

U U u

u

u u

u u

47

u

1963

1998 year of election

figure 3.1. Turnout at legislative elections in six countries, 1960–1999.

Data and Methods Our study is made possible by the existence of survey data going back to the time of the reform in six established democracies. These are Britain, Germany, The Netherlands, Norway, Sweden, and the United States.8 Figure 3.1 shows the evolution of turnout in these countries over the period. The countries fall into two clear groups in the figure, with the lower group of points belonging to the United States, while other countries provide a cloud of points in the upper portion of the graph. As can be seen, both groups of points show a gentle decline in turnout over the period since immediately before the lowering of the voting age.9

8

9

The data were collected by different (and changing) groups of researchers in each country and made available by the various national data archives of the countries concerned (see Appendix A). For details of turnout evolution in individual countries, see the figures in Appendix C. The series ends in 1998 because none of these countries held elections in 1999 and elections from 2000 are not included in the dataset.

68


In the pages that follow we will first present tables that give evidence that the preconditions outlined in H1 and H2 indeed are in place – tables that also provide the raw material for estimating the effects of cohort replacement. Our analysis of the consequence of lowering the voting age (H3 to H5) is done at the level of the electoral cohort in each election, where we can pick out the precise groups of individuals in each country who entered the electorates following the reform of the voting age and employ multivariate analysis to compare these groups with those who entered earlier and later, controlling for other effects on voter turnout.10 The Generational Basis of Turnout Variations Table 3.1 shows the full cohort table for one exemplary country, Britain. In the upper part of the table, electoral cohorts are tabulated across election years, and each cell contains the mean turnout recorded for that cohort at that election.11 These figures should not be taken entirely at face value, because, even though British election studies are among the largest fielded anywhere, with sample sizes often in excess of 4,000 cases (with an average of 3,000 cases), the number of cases in each cell is generally under 300 and often under 100 (the minimum size of any cell is 73). Thus it follows that much of the movement that we see there will be due to sampling variation. We will be more confident of the reliability of our summary table based on all six countries taken together (see below), but, since that overall pattern largely repeats the pattern seen for the United Kingdom, it is appropriate to use Table 3.1 as a basis for explaining how the summary table will be derived. The summarizing process is illustrated in the lower half of Table 3.1, where we see five rows of percentages. The first row 10

11

Though we derive these cohorts from individual-level data – specifically the age of each respondent and whether or not they voted – age and vote are almost the only variables available in identical form across countries, so we would gain nothing from analyzing the data at the individual level. Because our theory applies to the behavior of electoral cohorts, the cohort-year (or, more properly, the cohort-election) is the appropriate level of analysis in this investigation. For these analyses (unlike the multivariate analyses presented later) the data have not been weighted to the historical turnout recorded at each election. It should be borne in mind that misreporting of turnout tends to be greater with lower turnout, so intercohort differences in turnout will be underestimated in the data that we employ.

69

b

a

∗

8.4 9.9 1.1 1.4

−19.7 −11.0 1.5 3.2

72.2

−3.8

76.0

77.2

85.5 77.6 75.3 75.4 73.9

1970

−1.2

87.0 78.8 74.2 65.5

90.1 77.4 85.1

1966

1.9 4.5 13.2 8.5

3.7

75.9

89.8 89.7 84.3 88.6 78.4 75.8

1974

−5.5 2.5 1.9 2.1

0.4

76.3

90.6 89.8 83.9 80.8 80.3 78.3 70.3

1979

3.4 2.2 −1.2 3.3

−3.5

72.8

86.4 89.8 86.6 91.9 82.2 77.1 72.5 73.7

1983

−0.6 6.6 10.1 5.5

2.6

75.4

88.5 92.5 91.8 82.0 85.3 87.1 82.7 80.3 73.2

1987

2.1 6.5 7.1 3.6

2.8

78.2

88.9 86.5 91.3 90.8 89.6 87.8 87.4 87.4 79.7 75.3

1992

−15.9 −12.1 −9.6 7.3

−6.2

86.0 87.6 88.6 80.7 84.8 80.5 74.5 70.6 70.0 63.2 59.4 72.0

1997

−3.2 1.1 3.0

−0.7

87.5 87.6 84.0 82.1 82.4 82.6 77.6 78.0 74.3 69.2 59.4 82.6

Mean

7.2 6.9 5.7 4.4

Mean absolute change 3.0

Two elections in 1974 taken together. Actual turnout is not the mean of the figures above, which are not weighted to the election outcome. Mean turnout change is not the average change among cohorts. Mean turnout change is less than change within any cohort because the latter are all subject to more random error.

Summaries Mean turnout changeb # of election for cohort 1. New voters difference 2. Second cohort change 3. Third cohort change 4+ All prior cohorts change

Cohorts Pre-1955 cohorts 1959 cohort 1964 cohort 1966 cohort 1970 cohort 1974 cohort 1979 cohort 1983 cohort 1987 cohort 1992 cohort 1997 cohort Actual turnouta

1964

Year of election(s)∗

table 3.1 Average Turnout by Cohort and Year for British Elections, 1964–1997 (percentages)

70


gives the change in turnout for each election compared with the previous election (differences across the actual turnout row). The second row (labeled Cohort #1) gives the change in turnout for the incoming cohort compared to the previous cohort when it entered the electorate (differences down the diagonal). The Cohort #2 row shows the change in turnout for that previous cohort compared to its turnout at its first election, and the next row provides the same comparison for the cohort now facing its third election. The final row provides an average of all of the differences in turnout higher up the same column. At the right of the summary we see two columns of means: The first is simply the average of all of the changes registered in the row concerned; the second is the average of the absolute changes, where reductions in turnout are taken as positive and added to increases in turnout. It is that final column that we will focus on, but an explication of the penultimate column will be helpful in understanding what it is that is being summarized. First of all, when we look at that column in the upper half of the table (it is there the rightmost column) we see confirmation of the generally understood phenomenon of declining turnout in successive electoral cohorts. Even in Britain, where overall turnout only dropped some five points between 1964 and 1997 (and where the pattern of turnout change included quite a large rise in 1987 and 1992), we see a steady decline from cohort to cohort, amounting to some twenty-eight points in all,12 with only the smallest of occasional upticks in average turnout as we move down the column. Second, when we look at this cell in the “Mean turnout change” row (this cell is the penultimate one in that row), we see that (on average) turnout declined in Britain by 0.7 percent per election. (Over eight interelection periods this totals 5.6 percent, which would be the same as the difference between the 1997 and 1964 actual turnout cells, except for rounding errors.) Looking lower down the same column, we see that this average decline is entirely the work of new voters. More established cohorts – especially (in the British case) those facing their third election – see a net increase in turnout with the

12

Estimated by subtracting 59.4 (the mean for the 1997 cohort) from 87.5 (the mean for the pre-1955 cohorts). For most practical purposes we would need to bear in mind that the cohorts are of very different sizes.


71

passage of time (presumably the generally expected effect of aging, as we will discuss later). Finally, let us turn to the “Mean absolute change” column. There we see, first, that the average decline of 0.7 percent per election does not give a remotely adequate accounting of the real amount of change from election to election. As already noted, British elections saw rises in turnout as well as declines. The average gross change, treating declines as positive and averaging them with the increases, is 3.0 percent – more than four times the average net change of 0.7 percent. Looking further down the column we again see that the greatest contribution to this overall change comes from new cohorts, with the cohorts that face their second and third elections contributing almost as much. The big distinction in this column is between the most recent three cohorts and the rest. Of course, much of the movement that we record could be sampling error, so what must be focused on is the difference between cohorts with different amounts of experience. Newer cohorts see more volatility than do more established cohorts.13 Tables 3.2 and 3.3 summarize the findings for Britain and five other countries – all the countries for which election studies have been conducted continuously at least since the lowering of the voting age. Table 3.2 contains the mean turnout for each cohort in each country (for Britain this is the final column in the top half of Table 3.1). For the sake of comparability, we focus only on the ten most recent cohorts in each country, with prior cohorts being taken together in the first row of the table.14 The table shows that all countries except Germany saw a steady decline in turnout by cohort since the lowering of the voting age (the cohorts that had the chance to vote at eighteen are indicated in bold). We will have more to say later about the German case. 13

14

It might be argued that the entire table was an artifactual result of the smaller cell sizes lower in the table. The top row, after all, averages a large number of cohorts and constitutes a large fraction of the “all prior cohorts” row. But if the differences between the rows of the lower part of Table 3.1 were an artifact of this method of constructing the table, then row 5 would show a steady increase in the amount of change from election to election as later cohorts contribute a larger share of this change, to the right of the table. This does not happen. Moreover, a summary based on an entire cohort table, with thirteen rows (not shown), yields the same results as Table 3.1. Note that the number of years encompassed by each cohort is different in different countries and (except in the United States) even for different cohorts in the same country.

72


table 3.2 Average Turnout by Cohort for Six Countries, Earliest and Most Recent Cohorts

Earliest cohorts Cohort 10 Cohort 9 Cohort 8 Cohort 7 Cohort 6 Cohort 5 Cohort 4 Cohort 3 Cohort 2 Cohort 1 Change since lowering voting age

Britain

Germany

Netherlands

Norway

Sweden

United States

87.5 87.6 84.0 82.1 82.4 82.6 77.6 78.0 74.3 69.3 59.4 −22.7

88.3 86.2 86.6 88.4 85.3 79.8 77.7 72.8 80.2 70.3 89.9 3.2

84.7 82.4 83.5 83.4 82.6 76.3 69.1 74.4 67.9 73.7 69.8 −13.7

87.2 85.8 84.5 83.3 83.5 79.6 76.5 74.7 75.3 65.3 61.5 −21.9

89.5 89.0 88.1 87.1 85.4 83.8 83.0 82.0 78.0 85.5 78.0 −11.5

59.6 57.0 50.8 52.1 47.5 40.5 40.3 35.3 31.1 25.4 23.9 −16.4

Note: Turnout for cohorts entering after the lowering of the voting age appears in bold.

Table 3.3 reports the final column of the bottom half of Table 3.1 or its equivalent for each of the six countries (additional detail is provided in Appendix C). The final column of this table gives an overall summary, which shows about twice the volatility among newest cohorts as among established cohorts. Second-election and third-election cohorts fall in between, with little to distinguish them. There are differences, however, between the columns for particular countries. The Dutch and U.S. cases show less distinctiveness between new and established cohorts than does the British case, while the Norwegian, Swedish, and German cases show greater distinctiveness. However, the overall pattern is essentially repeated for each country. The only real anomaly is to be found in the Dutch column. The Netherlands is the only country in which the most recently entering cohorts do not on average show the most volatility. The newest three cohorts do show more volatility than established cohorts do, but there is no real progression among the first three cohorts, and the distinction between this group and the group of established cohorts is the least shown by any country. This is not the first time that lack of structure has been noted among Dutch electoral cohorts (van der Eijk and


73

table 3.3 Mean Absolute Cohort Turnout Differences in Six Countries British Dutch German Norwegian Swedish U.S. Overall 1. New voters difference 2. Second cohort change 3. Previous (third) cohort change 4+ All prior cohort changes

7.2

9.0

12.4

9.0

5.0

8.1

8.5

6.9

9.8

7.5

7.6

3.7

7.2

7.1

5.7

8.6

9.4

5.6

5.4

7.8

7.2

4.4

6.3

5.0

3.8

2.3

5.0

4.5

Neimoller 1983; Franklin et al. 1992),15 but the general pattern is still evident. These findings conform to hypotheses H1 and H2 listed earlier. The greatest change is evident among newly enfranchised cohorts and the greatest evidence of inertia is to be seen among established cohorts.16 Although the remaining hypotheses can only be formally tested in a multivariate perspective, Table 3.2 does also seem to confirm H4 by showing turnout decline among electoral cohorts that entered the electorate after the lowering of the voting age. The reason why we need to address this hypothesis in multivariate perspective is also clear in Table 3.2. Though all countries saw a decline in turnout among new cohorts entering the electorate after the lowering of the voting age (boldface entries), all of the countries also see later increases in the turnout of entering cohorts in one election or another – most notably in the most recent German cohort and the second to most recent Swedish cohort. So other forces can override the process of generational replacement, if it exists, and those other forces must be taken into account 15

16

Dutch turnout saw a massive drop in the first election for which we have a survey, 1971, which was the first election after the abolition of compulsory voting. Turnout recovered in the following election, primarily among newly enfranchised cohorts, even though this second election was the one in which the voting age was lowered to eighteen. This unfortunate starting point to our series of Dutch elections could be responsible for the unusual pattern shown in that country. To validate our assumption that cohorts become established after their third election, the average change in turnout from election to election for the cohort facing its fourth election was calculated for each country. In the overall column the average absolute change for this fourth cohort is 5.5, a clear step down from the overall third cohort average of 7.2 and considerably closer to the mean for all prior cohorts (4.5).

74


if we are going to detect the impact of early socialization on turnout change. This is the purpose of the next section. Multivariate Analysis If newly enfranchised cohorts show the greatest volatility from election to election, is this just random movement, or does it reflect rational responses to differences in the competitiveness and other features of successive elections, as required by our second hypothesis? And does it also reflect the different socializing experience of those who first voted after the lowering of the voting age, as required by our fourth hypothesis? These two questions can only be addressed by use of multivariate analysis. They have to be addressed together because, if addressed in different models, each model would be misspecified through the omission of the other critical influence on turnout. Apart from whether each cohort entered the electorate after the lowering of the voting age (what we will call “young initiation”), what variables need we include in order to properly specify the considerations that would lead turnout to be higher or lower? Recent research on wider sets of countries (Franklin 2002; Norris 2002) has shown that turnout responds to a variety of phenomena that together make up an election’s character: the size of the largest party (how close to commanding a majority), the margin of victory,17 compulsory voting, absentee voting, and electoral salience (a dummy variable picking out Switzerland and the United States). However, the countries for which we have survey data extending back to the time when the voting age was lowered do not differ in terms of absentee voting, or of compulsory voting during the period of our concern (the Netherlands abolished compulsory voting with effect from the election prior to the one at which they lowered the voting age), and do not include Switzerland. 17

For countries with plurality (first past the post) elections (Britain and the United States) we employ the average of the lead in each congressional district (U.S.) or in each constituency (Britain), on the basis of research into turnout at U.S. congressional elections, which has shown average margin of victory to be a powerful predictor of turnout (Franklin and Evans 2000). The variable essentially speaks to the “wasted vote” syndrome (Blais 2000), which does not exist in proportional representation (PR) countries, so those countries are coded 0 on this variable (matching the code that would be given in Britain and the United States if the average margin of victory across all electoral units were zero).


75

For the countries that are included in our data, the variables needed to characterize each election are thus size of largest party (what we will call its “majority status”),18 margin of victory,19 time since last election, and a dummy variable picking out the United States as a country where elections are particularly lacking in salience (Franklin 2002:156–7). In addition, we include average district magnitude – a variable found significant in several other past studies (Powell 1986; Jackman 1987; Radcliff 1992; Jackman and Miller 1995) and about which Blais (2000) argues that more contestants in each district stimulate turnout by offering more choice. Other variables employed in past research were evaluated for inclusion in our models, but, as was found by Franklin (2002), none of them proved significant. The way we model the role of new voters (and test our second hypothesis, H2) is by interacting each of the previously listed components of electoral character with a variable (“new”) that picks out the cohorts facing any one of their first, second, or third elections. In order to test H5, we would also like to know whether those who enter the electorate after the lowering of the voting age see an increase in turnout, relative to older cohorts, as they gain experience of elections. So we define a variable, referred to as “experience,” that starts at 0 in the year a cohort enters the electorate and rises continuously to 1 over the following fifty years (the maximum length of time that any cohort did remain in any of our electorates). This variable is employed in interaction with young initiation to see if there is evidence of a narrowing gap between those who entered the electorate at eighteen and earlier cohorts. We would have liked to also employ experience on its own, to control for the general aging effect that is evident in our cohort tables (older cohorts generally see higher turnout in Table 3.1). However, the regression equation has multicolinearity problems when it contains the interaction together with both of its components,20 so instead we

18 19

20

Measured as the absolute percentage distance of the party from receiving 50 percent of the vote (the smaller this gap, the higher the expected turnout). For countries with majoritarian electoral systems (Britain and the United States, in this dataset) the margin of victory is taken on average over all districts or constituencies as well as overall. The multiple correlation between the three variables is above 0.8, suggesting estimation problems. The variables included in the models have no such problems and no tolerances above 0.7.

76


include a variable (“established cohort”) that picks out cohorts in their fourth or subsequent elections. We expect a positive value for this variable’s coefficient, since older cohorts do vote at a higher rate than new cohorts, as we have seen. We have strong expectations for the direction of the effects of all of these variables. Established, time since previous election, average district magnitude, and the interaction of experience with young initiation should all see positive coefficients. Young initiation itself, together with majority status and margin of victory, should have negative coefficients. Interactions with new cohort membership should retain the same signs as each of the relevant variables without interaction. For all our independent variables we thus employ one-tailed tests. The unit of analysis is the cohort-year (more properly the “cohortelection,” but the word “election” lacks a temporal implication). Each nonblank cell in the upper half of Table 3.1 represented a cohort-year, as did each equivalent cell in each of the other cohort tables summarized in Tables 3.2 and 3.3. We have 10–13 cohorts in each country, and 9– 15 elections, yielding 564 cohort-years altogether. Before aggregation, the data were weighted first so as to reproduce the actual turnout observed historically at each election and, second so as to give equal weight to each survey; but after aggregation the N for our analyses becomes the number of cohort-years. Our coefficients are estimated using ordinary least squares (OLS) regression with panel-corrected standard errors to take account of the fact that turnout variations will to some extent be country-specific (Beck and Katz 1995).21 21

A problem arises in the use of this method because of the nature of our dependent variable. Though turnout is measured as a percentage and is thus an interval-level variable, it is subject to ceiling effects. It cannot exceed 100 percent, and actually reaching this figure is difficult or impossible. For this reason it has the same limitations in regression analysis as does a binary dependent variable, and ordinary or least squares (OLS, GLS) analyses may yield the same sort of biased coefficients as they would with a dummy dependent variable. To see whether this was occurring in practice, we subjected our dependent variable to a logistic transformation and repeated our analyses with the resulting transformed variable. The transformation is achieved by taking LOG ((turnout / MAX(turnout) / (1 – turnout / MAX(turnout)))). Analyses employing this variable explained less variance than the corresponding analyses with an untransformed dependent variable, indicating that effects on turnout are better estimated if they are assumed to be linear than if ceiling effects are assumed. Nor is the possibility that estimates of turnout could exceed 100 percent realized when predictions are made from any of the models used here. Thus we chose to stick with the untransformed turnout measure.


77

table 3.4 Regression Findings (Panel Corrected Standard Errors in Parentheses) Unstandardized coefficients Independent variables (values) (Constant) Young initiationa cohort (0,1) Establishedb cohort (0,1) Experiencec ∗ young initiation (0–1) Majority status,d percent (0–14) Mean margin of victory, percent (0–45) Newe ∗ majority status (0–14) New ∗ mean margin of victory (0–45) New ∗ average district magnitude (0–328) United States (0,1) Turnout of cohort at previous election Adjusted R2 Number of cohort-years

Model A

Model B ∗∗∗

86.032(1.443) −1.867(0.865)∗ 7.229(1.760)∗∗∗ −0.259(0.122)∗ −0.589(0.081)∗∗∗ −0.434(0.041)∗∗∗ −0.160(0.166) −0.296(0.054)∗∗∗ 0.047(0.017)∗∗ −16.539(1.325)∗∗∗ 0.741 546

79.446(3.074)∗∗∗ −3.396(0.676)∗∗∗ 1.206(1.216) −0.130(0.107) −0.402(0.066)∗∗∗ −0.351(0.039)∗∗∗ −0.363(0.134)∗ −0.288(0.048)∗∗∗ 0.036(0.017)∗ −15.421(1.445)∗∗∗ 0.127(0.034)∗∗∗ 0.842 464

Significant at the ∗ 0.05, ∗∗ 0.01, ∗∗∗ 0.001 level, one-tailed. a Voters facing their first election after turning eighteen. b Voters who have already experienced three elections. c Number of years since reaching voting age, as a proportion of 50. d Absolute distance in percent from having received 50 percent of the votes. e Members of one of the youngest three cohorts.

The findings in Table 3.4 are presented in two related models. Model A shows the effects of all of the variables that proved significant at the 0.05 level. In Model B we introduce the effect of inertia by including turnout for each cohort at the previous election. The major surprise in the table is the failure of margin of victory to prove significant, either in its simple or its interacted forms (only a mean margin of victory proves significant in the analysis).22 Otherwise, the findings are very much as hypothesized. In the lower part of Model B we see that the variables relating to electoral competition have significant interactions with new cohorts (confirming H3). At the top of the table we see that cohorts that experienced young initiation (as a consequence of entering the electorate at eighteen) are some 2 percent less likely to vote (3.4 percent when past turnout is included in Model B) than cohorts that entered the electorate at a later age (confirming H4), 22

Jack-knife tests show that, if The Netherlands is excluded from the dataset, the variable works as hypothesized. The effect on all cohorts remains insignificant, but the effect on new cohorts attains significance (and variance explained rises to 0.867).

78


and this gap is not reduced with the passage of time. Though cohorts do vote at higher rates as they age (the coefficient for established cohorts shows these voting at a rate of 7.2 percent higher than new cohorts on average) – the interaction of experience with young initiation is not positive (confirming H5) and is not even significant in Model B. The introduction of inertia into Model B greatly increases the effect of young initiation (presumably because Model B is better specified).23 It also has the effect of greatly reducing the coefficient for established cohorts. Evidently, the way that established cohorts maintain their higher turnout is through inertia – they vote at higher rates because that is what they have always done – so inertia accounts for the difference between new and established cohorts when it is included in the model, as suggested in Chapter 2 (the H and 1 − H terms in Equation 2.6). Inertia also affects new cohorts. However, this effect does not replace (or even much attenuate) the effects on new cohorts of other variables. The negative effect on turnout of being a cohort from the United States is somewhat smaller than the gap shown in Figure 3.1 between U.S. turnout and turnout in other countries. This implies that some part of the difference is accounted for by one or more of the substantive variables included in the model, most likely the mean margin of victory, which, in the United States, is very high (see Chapter 4). Assessing the Findings An effect of 3.4 percent on cohorts that experienced an initiation into the world of electoral politics at a younger age than previously corresponds to a 2.5 percent effect, to date, from lowering the voting 23

Including vote at previous election generally reduces the effects of the substantive variables included in Model C (in the case of the interaction between new and margin of victory, this reduction is enough to cost the variable statistical significance), as is generally expected in time-series models that employ a lagged version of the dependent variable (see Achen 2000, and further discussion of this point in Chapter 5). The fact that young initiation has a more powerful effect in Model B suggests that the absence of past turnout in model A was artificially depressing the effect of young initiation. Making use of turnout at the previous election as an independent variable highlights the fact that this dataset contains a sort of time series in which the same cohort is interviewed repeatedly at successive elections. However, there does not appear to be a time-serial dependency worth taking into account since, if we attempt to view the series as embodying an autoregressive (AR1) process, Rho is an insignificant –.05. We will discuss the problem of time-serial dependency in Chapter 5.


79

age.24 Such an effect is not large. It constitutes a small part of the 8 percent average decline in voter turnout experienced by the six countries investigated in this chapter over the period of our study. Even taken with the 0.7 percent decline anticipated earlier in this chapter due to the temporary enlargement of new cohorts, it accounts for less than one-half of the total. However, this effect of young initiation is an effect that has greater importance than its numerical consequence for voter turnout. Its presence confirms the importance of learned behavior, as well as the long-term consequences of early learning, with implications that go far beyond the subject of this chapter. Indeed, this insight will be instrumental in helping us to understand the remaining 5 percent drop that has occurred in these countries since the 1960s (together with specific variations, both up and down, that have occurred in one country or another).25 Our study of the consequences for voter turnout of a lowering of the voting age in six countries has made it clear that there are two components in turnout change – one from features of a particular election that differ from features of the previous election and one from the delayed effects of past changes. This implies that virtually all of the models previously used to study voter turnout (both at the aggregate and individual levels) have been misspecified. Most of the factors that impact turnout have significantly more impact on individuals facing one of their first three elections. Older voters are less strongly affected (indeed, our findings suggest that some factors impact established voters not at all, as is the case for the average district magnitude in

24

25

Three-fifths of the cohorts in each electorate (except Norway, where the proportion is only one-half) had by 1999 enjoyed (if that is the word) the experience of facing an election at eighteen to twenty-one years of age. However, as already noted, the earliest of these cohorts will have been enlarged by the baby boom generation that came of age at about the same time. Thus the proportion of our electorates exposed to young initiation will be more like three-fourths than three-fifths, yielding this estimate of the effect to date of lowering the voting age (0.75 × 3.4). The number of cohorts available in each country are too few for country-by-country analyses to be viable. However, jack-knife tests (in which each country in turn is eliminated from the analysis) demonstrate that the major findings are not dominated by any one country and produce quite consistent estimates of the parameters of interest. We have already mentioned the major anomaly that is caused by the inclusion of The Netherlands in the dataset (note 22). Boosting the N by doing the analysis at the level of individual voters would only give spurious significance to results that would still be subject to large variations from country to country.

80


Table 3.4).26 This in turn implies that many voters establish a “standing decision” to vote or not to vote and that models that do not take this into account will yield biased and unstable results. There are two ways in which conventional models are misspecified: The first is in attempting to measure effects on all voters that apply particularly to new voters. This will result in such effects being underestimated. The more important way in which conventional models are misspecified is in failing to take account of the cumulative effects of factors that played a role in the socializing of established voters. In individual-level models, effects ascribed to social status or education may in fact be the effects of early experiences. In aggregate models, we need to include the indirect consequences of the character of elections long past. In cross-national studies, the need is to take account of how long particular electoral arrangements have been in place. It is not enough, for example, to determine whether citizens are entitled to vote at eighteen years of age; it also matters how long this provision has existed. Only that will tell us how many electoral cohorts will have been affected by the provision and, thus, how much lower we can expect turnout to be than in a country with a voting age of twenty-one. The cumulative changes ascribable to a lower voting age that we found in this chapter are likely also to apply to other changes in election laws that primarily affect new cohorts – cohorts that, we have found, are most largely affected by the character of an election. Thus the abolition of compulsory voting in Italy in 1993 and the move to proportional representation in New Zealand in 1994 have so far had only limited effects; but perhaps these effects will cumulate over the period that it takes for all Italian and all New Zealand voters to have had their first electoral experiences under the new rules. This is in contrast to our expectations for variables such as the closeness of the race and the size of the largest party, which can move in both directions in the experience of individual citizens. Butler and Stokes, it will be recalled, suggested that it took three mutually reinforcing experiences to inculcate a habit of voting for one party. Our cohort tables suggest that this process applies more generally. Changes in institutional arrangements such as the voting age have the chance to partake of the phenomenon. 26

Though the fact that the effect is barely significant at the .05 level is more likely to be the reason why its even lesser effects on established cohorts are not significant.


81

Variables that do not generally take on the same value for that many elections would not be expected to have such strong cumulative effects. One by-product of this insight is a possible explanation for the apparent contradiction that arises from the fact that weekend voting appears potent in a cross-national perspective but not in an over-time perspective (see Chapter 1). Countries that change the day of the week on which they hold elections might do so repeatedly, thus eliminating the possibility of a learning experience among new cohorts. Countries that do not change the day on which they hold elections have (by definition) kept to that day over the course of many elections. Citizens of such countries will have thus had the chance to become socialized by the experience of easier or more difficult voting arrangements, affecting the overall turnout in such countries. The mechanics of voter turnout have been shown in this chapter to be quite tricky, depending as they do on the learning experiences of cohorts during their formative years and thus on the history of electoral arrangements as well as on their current state. So the study of voter turnout calls for inventive research strategies that can tease out from the data the gradually operating as well as the immediate causes of turnout change. Only by taking both cumulative and short-term effects into account can we hope to come to a proper understanding of the mainsprings of electoral participation. But we must do more than just distinguish cumulative from short-term effects; it is also important to distinguish different types of cumulative effects. Types of Cumulative Effects Not all variables have cumulative effects. As already pointed out, variables such as the closeness of the race and the size (majority status) of the largest party do not generally retain the same values long enough to engage the socializing processes that yield cumulative effects. Indeed, we will discover in later chapters that most of the variables that cause turnout change from election to election are short-term in nature. Still, short-term effects involve quite straightforward interactions, as employed in Table 3.4. Cumulative effects can involve complications that go beyond anything that we needed to take into account in Table 3.4. In the final pages of this chapter we will focus on these complications.


82

The long-term effect of lowering the voting age has been established in this chapter by showing that, as new cohorts enter the electorate, each in turn is found to have a level of turnout lower than it was for cohorts entering the electorate when slightly older (whether in other countries that have not yet lowered the voting age or in the same country in earlier years). This disparity between cohorts entering the electorate at different ages is not eliminated as cohorts grow older with the passage of time. It seems obvious that this development will affect turnout at later elections, lowering the overall turnout as young initiation voters progressively replace those that were socialized in an earlier era. We have referred to this effect as a “footprint” in the electorate, like the rings of an aging tree, but what will this footprint look like? To answer this question we need to transform the insight explored in this chapter into one expressed in terms of turnout change over time. Figure 3.2 shows an idealized and simplified picture of turnout evolution over time in response to the lowering of the voting age in some country. What is shown there is the effect on overall turnout of the entry into the electorate of successive cohorts that had received an initiation into electoral politics at a younger age than previously. This effect is based on the coefficient (3.396) estimated in this chapter for the difference between young initiation cohorts and others. At the first election after the lowering of the voting age, that difference has little effect on overall turnout because only the newest cohort is affected. Only when the entire electorate consists of cohorts that were subjected

Turnout Initial level

-1 -2 -3 -4 -5 -6 -7 -8 -9

Reform election

-2

-1

+1 +2 +3 +4 +5 +6 +7 +8 +9 +10 +11 +12 +13 Elections

figure 3.2. Expected long-term evolution of turnout following an extension of the franchise to younger voters.


83

to young initiation do we see a difference (comparing the turnout in such an election with the turnout before the lowering of the voting age) equivalent to the difference we measured in Table 3.4. The total effect thus takes thirteen elections to accomplish, starting with the reform election and ending with the reform election +12. At each election one cohort is added to the electorate, which turns out at a lower rate than previous cohorts, and one cohort is removed from the electorate that was voting at a higher rate. After thirteen elections the entire electorate has been replaced and everyone is voting at the lower rate. This is evidently an idealized picture that assumes successive cohorts of constant size maintaining an electorate of constant size. In practice we will relax these assumptions, which are only made for heuristic reasons to more easily describe the nature of the footprint created by a lowering of the voting age. This footprint is not the only possible footprint that could be left by an electoral reform. To clarify the logic, we will present two other footprints that would be left by reforms of different types. The second of these footprints is the one that is left by an extension of the suffrage to a group that was previously disenfranchised (women, for example, or citizens without property qualifications). In Figure 3.3 what we see is an initial fall in turnout as the previously disenfranchised group enters the electorate having not yet acquired the habit of voting. Three cohorts adapt immediately, those that have not yet become set in the habit of nonvoting, reducing the extent of the fall that would otherwise have occurred. The extent of the fall in overall turnout is also proportional

Turnout Initial level

-1 -2 -3 -4 -5 -6 -7 -8

Reform election

-2 -1 +1 +2 +3 +4 +5 +6 +7 +8 +9 +10 +11 +12 +13 Elections

figure 3.3. Expected long-term evolution of turnout following extension of the franchise to a previously disenfranchised group.


84 Turnout Initial level

-1 -2 -3 -4 -5 -6 -7 -8 -9

Reform election

-2 -1

+1 +2 +3 +4 +5 +6 +7 +8 +9 +10 +11 +12 +13 Elections

figure 3.4. Typical expected long-term evolution of turnout following a change in election law.

to the relative size of the group to which the franchise was extended. In the case of extensions of the franchise to women, the fall in overall turnout is one-half the difference between the initial female turnout and the male turnout, since the men’s turnout is not affected by the reform. In ensuing elections, turnout progressively rises as successive cohorts of the previously disenfranchised group learn the habit of voting at rates equivalent to the rates for other voters.27 Finally, let us turn to a more typical electoral reform: one that affects all voters and in which the initial change in turnout is amplified by the ensuing socialization process. This footprint will occur with a change to or from compulsory voting, with the introduction or elimination of absentee voting, or with any other permanent change in electoral arrangements with effects on turnout, other than a change in the franchise. Figure 3.4 illustrates the expected effect on turnout when this is in a downward direction. The introduction of absentee voting or the 27

Recent research (Schlozman, Burns, and Verba 1994, 1999) has attempted to explain lower female electoral participation on the basis of resource differences, but our theoretical approach suggests that resources would be intervening variables in the generational replacement of less participatory with more participatory women. The same would be true of the role of resources when previously nonvoting groups such as blacks and latinos learn the habit of voting (Verba et al. 1993). Our model would explain the rising turnout in Southern states following the Voting Rights Act of 1965 (cf. McDonald and Popkin 2001). Norris (2001) explains rising female turnout on the grounds of generational replacement, though the reasons she gives for the evolution of female turnout are quite different from ours.


85

abolition of compulsory voting would be expected to have an effect of this kind, where the first three cohorts react immediately but the bulk of the change takes effect only over the course of a further ten elections. Evidently, the reverse reform (the introduction of compulsory voting or the elimination of absentee voting, for example) would cause an initial rise followed by a slower amplification of that rise. Figures 3.3 and 3.4 were drawn assuming effects of similar magnitude. Thus, Figure 3.4 shows an overall drop in turnout of 5 percent. However, an effect of similar magnitude resulting from female enfranchisement (in Figure 3.3) only causes an initial fall in overall turnout of some 2 percent, since the extension of the franchise affects only about one-half the individuals in the electorate (and some of these adapt immediately to their change in circumstances). The drop is then eroded over the following ten elections. There is of course no good reason to suppose that the two effects would be of similar magnitude, and the actual magnitudes of these effects do not have to be assumed; they will be discovered empirically.28 However, the three different cumulation functions are determined theoretically. The only confirmation we will get that the evolution of turnout follows the three different patterns shown in Figures 3.2 to 3.4, depending on the type of reform concerned, will be that the resulting variables yield plausible findings and that the pattern of predictions that we make with the help of these functions matches the pattern of turnout observed empirically in each country. The figures are also constructed on the assumption of uniform cohort size, but this assumption is made only for the sake of providing an idealized picture of the expected consequences for voter turnout of different kinds of electoral reforms. In practice, we will be able to take account of changes in the sizes of incoming cohorts provided we are able to keep track of these changes. Indeed, to evaluate the effects anticipated in Figures 3.2 to 3.4 we need to be able, for each electorate, to keep track not only of the sizes of incoming cohorts but also of the evolving size of the group that constitutes the newest three cohorts 28

Tingsten (1937) found that the enfranchisement of women in Scandinavian countries gave rise to female turnout that was initially 10–12 percent less than male turnout, equivalent to a 5–6 percent drop in the overall turnout.

86


taken together. And of course we need to know the total size of the electorate as this evolves over time, since the important quantities are the sizes of incoming and impressionable cohorts as proportions of the size of the entire electorate in each country. None of these quantities are straightforward to determine, for reasons to which we now turn. Measuring the Size of the Electorate at Each Election To measure the proportionate share of new cohorts in the electorate of each country at the time of each election, we need to know the size of the electorate at each election. Indeed, we also need to have this figure to work out the turnout rate. The total number of votes cast is always reported, and this number is as near to being cross-nationally comparable as any statistic in comparative political research. So the numerator in the calculation of turnout is not problematic. However, the denominator is another matter, and the figure that constitutes the denominator in the calculation of turnout is also the denominator in calculating the proportion of new voters. For most countries this figure is the number of registered voters; but for the United States it is conventional to report turnout on the basis of the voting age population. This is because the United States is unique in having voluntary voter registration with a very large proportion of the population failing to register (France also has voluntary voter registration, but in that country registration rates are almost as high as in countries where the state takes responsibility for ensuring voter registration). In the United States, at least until recent motor voter legislation, those who did not vote were generally not registered, and those who took the trouble to register generally voted (Erikson 1981) – a situation that makes it quite impossible to imagine using U.S. registration figures as a basis for measuring turnout there. The mismatch between the basis for reporting turnout in the United States and elsewhere creates a problem for comparative electoral research. One solution is simply to omit the United States from comparative analyses, as was done by Blais (2000). Another solution (mainly employed by U.S. scholars) is to recalculate other countries’ turnout on the basis of their voting age populations (VAP – see especially Lijphart 1997). This solution, which has a plausible ring, actually has the effect of injecting a great deal of error into comparative measures of


87

turnout. This is because, except in some European countries that keep continuous track of their population size, the voting age population is not measured at the time of an election but is estimated from the most recent census, which could be more than ten years out of date. In such countries, the accuracy of the projected size of the current voting age population will vary greatly from election to election in the same country (depending on how recently a census was conducted) and even more so from country to country (depending on how accurate the census was and how much side information exists on which to base projections). The United States, which has much need for accurate population statistics, may be relatively well equipped to employ the voting age population as a basis for calculating turnout. For most other countries the estimates would not be official estimates but estimates made by individual researchers using assumptions that could give quite different results from those that would be obtained by other researchers using other assumptions. So the most accurate figures for the denominator that we need both for calculating turnout and for calculating the proportionate size of new cohorts are in fact not comparable: VAP for the United States; registered voters elsewhere. For a small number of countries, as already mentioned, accurate population figures are available at the time of each election. However, using these figures would result in measurement incompatibilities between the countries concerned and countries that do not have such figures, while not eliminating differences in measurement for the United States which, however much effort it puts into making them, still only has estimates of population size, not actual counts. By using the registered electorate as the denominator everywhere except the United States, we limit our problems of comparability, needing to adjust the figures for only one country. Happily, a great deal of effort, mainly by Walter Dean Burnham (1955 and personal communications with the authors of the International Almanac of Electoral History), has been put into correcting U.S. estimates of the voting age population to remove those not in fact eligible to vote. These corrections have been incorporated into the source employed here, the already mentioned International Almanac of Electoral History (Mackie and Rose 1991, with updates in the annual data issues of the European Journal of Political Research). There are still reasons to be cautious about interpreting U.S. turnout (and electorate)

88


figures as fully comparable with those published for other countries, but this source certainly incorporates less error than other sources.29

Measuring the Proportion of the Electorate that Is New at Each Election Ideally, one would like to be able to find out from census data the exact size of the cohort that is new at each election. On the basis of this figure, the size of the newest three cohorts could be established by simple addition. However, it is not in fact possible to get these figures in comparable terms from census data for twenty-two countries because the categories used for reporting census figures differ from country to country and change, even in particular countries, with the passage of time. To make matters worse, census data are not reported on the same basis as voting data. Most countries report turnout statistics in terms of the registered electorate; only the United States reports turnout in terms of the voting age population, as already explained. But census data relate to the voting age population, not the electorate. For the analyses to be conducted in later chapters, the proportion of the population that is new in each election is computed from the figures that we have for the sizes of electorates (Mackie and Rose 1991, and updates). The procedure is straightforward. Survey data from the six countries investigated in this chapter tell us that the average length of time that an individual remains in the electorate of an advanced democracy is fifty years and that this has not changed since the 1960s. It follows that each year 1/50th of the electorate will have had to be replaced if an electorate of constant size was to be maintained (1/47th prior to the lowering of the voting age). At the time of each election it is a matter of public record how many years (and parts of years) have elapsed since the previous national election. Time elapsed 29

We checked the denominator in our calculation of U.S. House turnout against the figures used by McDonald and Popkin (2001) for correcting their measure of presidential turnout and recalculated house turnout on the basis of their corrected figure whenever the Mackie–Rose figure differed from theirs (which only happened in the 1990s). There is no reason to suppose that remaining errors are anything other than random, so, to the extent that U.S. turnout is not adequately corrected, this will make it more difficult for us to obtain findings that prove statistically significant – a conservative measurement strategy.


89

in years multiplied by 1/50th of the electorate at the previous election yields the number of new members required to maintain a constant-size electorate. To this figure needs to be added (or subtracted, if negative) any difference in size between the electorate at the previous election and the electorate at the current election. The sum is an estimate of the total number of individuals newly enfranchised at any given election. Dividing the sum into the figure for the total electorate at the election concerned yields the proportion of the electorate that is contained in the incoming cohort. Adding up the figures for three successive elections before dividing the sum into the electorate size for the last of the three elections yields the proportion of that electorate facing one of its first three elections.30 Using the building blocks established in this chapter, we are now ready to inventory the variables that influence voter turnout levels. 30

This procedure would yield missing data for the first three elections in each country. In order not to lose such a large fraction of the available data, the size of the first three cohorts in each country’s second election is estimated by taking three times the estimated number of new potential voters since the first election; the size of the first three cohorts in each country’s third election is estimated by adding twice the number of potential voters new in the third election to the number of those new in the second election.

4 Rational Responses to Electoral Competition

In Chapter 2 we put forward a theoretical model that changes our approach to understanding voter turnout from one that focuses on why anyone would vote to one that focuses on why anyone would not vote. In Chapter 3 we established that one reason for not voting would be because of the failure to learn the habit of voting. In Chapter 3 we also established that there is a strong tendency for turnout variations to be led by the newest members of the electorates of each country, and especially by the entering cohort. Variables relating to the character of elections have much greater effects on individuals who have not yet become set in their ways – effects that can cause those individuals to vote at either a higher or a lower rate than their older counterparts. Indeed, we also see traces of the same effects acting on earlier cohorts: traces evident in the different levels that we see in the turnout of particular older cohorts. Though turnout in general has been declining in recent years, we saw clearly that the susceptibility of the youngest cohorts to the character of elections is not just a susceptibility to lower turnout; they can also be induced to vote at higher rates than preceding cohorts (most notably in Germany in 1998, in Sweden in 1994, and in The Netherlands in 1994 – see Table 3.2). What are these forces that induce more people to vote in some elections than in others? This chapter addresses the question first by setting out three case studies (of Malta, Switzerland, and the United States), which illustrate the operations of some of the most powerful influences concerned. On the basis of these illustrations, the chapter will 91

92


then move on to inventory additional influences that can be expected to prove important by analogy with those whose importance was shown in the three case studies. In previous research (Franklin 1996, 1999, 2002; Franklin and Hirczy de Mino 1998; Franklin, van der Eijk and Oppenhuis 1996) it has been argued that the variables with the greatest power to cause turnout variations are those that govern the likelihood that their vote will affect an election’s outcome in ways that are meaningful to them. In those works it was found that competitive elections exert a variety of pressures, presumably both directly on voters and indirectly via the efforts of those who try to mobilize them to vote, as anticipated by Schattschneider (1960) and Schneider (1980). In this chapter we want to extend those arguments by distinguishing between a number of ways in which elections differ in competitiveness – ways that, because of the possibility of generational effects on turnout levels, have to be treated differently in models that try to explain turnout change. Much of the chapter is taken up with a single variable – executive responsiveness – because that variable has not been used previously in turnout studies and proves to be of major importance in explaining both differences in turnout between countries and changes in turnout over time.1 To lay the groundwork for understanding this variable we start with two stories, one about Malta and one about Switzerland: two small countries whose very different electoral histories over the past fifty years are suggestive of the sorts of things that lead to high and low turnout, respectively. Building on these stories and a complementary analysis of turnout in the United States, we will then develop a list of the different ways in which the character of elections can affect the level of turnout. A Tale of Two Outliers We start by focusing on Malta and Switzerland not because these are the two countries that have today the highest and lowest rates of turnout, respectively, of all established democracies, but because each of these countries exhibits a progressive and dramatic change in turnout over 1

The variable is very close in concept to what Powell (2000) calls “responsiveness in choosing policymakers.”

Rational Responses to Electoral Competition

93

Turnout, percent 98

Maltese turnout

84 1962

70

56

Swiss turnout 1963

42 1947

1959

1971

1987

1999

election year

figure 4.1. The evolution of turnout in Malta and Switzerland, 1947–1999.

the course of the past forty years. Both countries started the period after World War II with turnout in the normal range for Parliamentary democracies without compulsory voting (70–80 percent). But we see in Figure 4.1 that, starting in 1963, Swiss turnout began a progressive fall that continued until the present day (the particularly large drop in 1971 follows a reform that gave women the vote in federal elections, reduced the voting age to eighteen, and also abolished compulsory voting in two of twenty-six cantons)2 and which ultimately brought Swiss turnout to a level below even that of the United States (long the established democracy with the lowest turnout). By contrast, Maltese turnout in 1962 began a progressive rise that ended only when it hit the ceiling established by the fact that turnout cannot actually reach 100 percent in any country, due to inevitable inaccuracies in the lists of eligible voters (which include people recently dead; people recently moved to other jurisdictions; and people temporarily out of the country on vacation, on business, or to study). 2

Prior to the reform, three cantons had compulsory voting. We will have more to say later in this chapter about extensions of the franchise to women.

94


Bearing in mind what we learned in the previous chapter about the incoming and most recent cohorts being largely responsible for turnout change, it is plausible to assume that something happened in Switzerland before 1963 and in Malta before 1962 that resulted in a change in the turnout of incoming cohorts – something that resulted in cumulative turnout change as the new cohorts progressively replaced older cohorts through natural demographic processes. In the case of Switzerland, we would expect the decline to have about ten years still to run, since the youngest of those who were socialized into habits of relatively high turnout – those who entered the electorate of that country before 1963 (that is to say, at the previous election, which was held in 1959) would by 1999 be only sixty-three to sixty-seven years old and be expected to have a good ten years of electoral participation still ahead of them. In Malta we would also have expected the effect to be continuing, except that the ceiling on turnout was apparently reached there some elections ago. Before we attempt to pin the blame for these radical changes in turnout on any one culprit (or culprits) we should first survey what is currently believed to be the explanation for low turnout in Switzerland and high turnout in Malta. Switzerland is a country with a long history of continuous elections, being one of the few to have continued holding elections during World War II as well as throughout World War I. It has proportional representation, which is generally believed to boost turnout,3 and no obvious peculiarities that would distinguish it from other countries with moderately high turnout. The Swiss themselves appear to have no very clear idea why turnout in their country has declined so much. A recent study of electoral participation there (Wernli 2001) suggests a number of factors that make turnout in Switzerland low, several of which are of recent vintage, but he does not attempt to apportion the decline that has occurred as between these different factors (we will enumerate them later). Turning to Malta, this is a country where high turnout has been studied specifically to gain clues as to what forces lead to high turnout. 3

The introduction of proportional representation to Swiss elections in 1919 had the effect of raising turnout there from nineteenth-century levels to the levels seen at the start of our period.


95

Wolfgang Hirczy (1995) credits Malta’s high turnout to a conjunction of characteristics that would seldom be found anywhere else: proportional representation with only two parties (a possibly unique combination) in which the two parties are polar opposites in terms of their policies (one being a virtually unreconstructed Marxist socialist party and the other a Thatcherite free-market conservative party). This is a country where the identity in the party in power really makes a difference (a country that applies for membership in the EU or withdraws that application, depending on which party is in power). And the national government is the only game in town. Malta is too small to have any local governance (no state or regional governments, not even any mayors) – all policy is ultimately determined by the unicameral parliament in Valetta. Moreover, the two parties are very closely matched in terms of support. In a country with only some 200,000 voters, the margin of victory in parliamentary elections is often only a few hundred votes. In elections as competitive as these, every vote counts, and on election day people are wheeled to the polling booth in their hospital beds if that is what it takes to get their votes (Hirczy 1995). The argument is persuasive and would go far toward explaining Malta’s unique situation at the top of the league table of established democracies in terms of turnout (despite the absence of compulsory voting there) were it not for one inconvenient fact: Turnout in Malta was not always so high, despite the fact that the characteristics described by Hirczy have been features of Maltese elections at least since World War II. As already mentioned (and shown graphically in Figure 4.1), until the early 1960s, turnout in Malta was in fact no higher than in other established democracies that lacked compulsory voting (no more than 10 percent above turnout in Switzerland during the same period). Evidently there must have been some other precondition for high turnout that was not met until the early 1960s – a precondition that, when met, made it possible for all of the features described by Hirczy to come into their own and play the roles they play in Maltese elections today. So what is it that happened in Malta and Switzerland in the early 1960s? What are the chances that one particular independent variable changed its value in both countries, though in opposite directions? Anyone faced with this question without having been primed would certainly suppose that the chances of the same variable proving crucial

96


in both countries was very small, yet we are going to argue that it was the same variable (changing its value in opposite directions) that played the critical role in both of these cases, as it plays a critical role in every democratic election. This is the variable that Arthur Banks (1997) calls “parliamentary responsibility” but that we (following Powell 2000) will call “executive responsiveness” – the extent to which the political complexion of the executive is responsive to the choices made at the time of an election. The World Handbook dataset compiled by Banks codes this variable uniquely in terms of the extent to which legislatures control the complexion and policies of the executive. The variable gives low scores to elections held in the United States, because of the separation of powers there, and to elections held in Switzerland for reasons that we will come to later. It also gives reduced scores to elections for constitutional assemblies in several countries and to elections for the French Assembly in 1958 and 1960, when the peculiar position of General Charles De Gaul essentially isolated the executive from popular control. We consider this variable to indicate an aspect of electoral competition, since, even in a close election, competitiveness is lacking when the outcome of an election has little or no effect on policy. The variable as included in the World Handbook dataset does not differentiate late-period Malta from early-period Malta because the government of Malta was appointed by the parliament and responsible to the parliament, just as in other parliamentary systems, during the entire period. What Banks (or his coders) neglected to notice was that, until 1964, that executive would not necessarily be free to implement the policies decided by the Maltese electorate because, until 1964, Malta was a British colony and the decisions of its elected government were subject to ratification by the British-appointed governor.4 We believe that the increase in turnout seen in Malta, starting at the election of 1962, when independence was decided upon by Maltese voters, was due to the fact that this variable changed its value for Maltese voters when they were first presented with choices that would not be subject to ratification by an unelected power. The fact that turnout did not immediately rise to its 1990s level would have prevented most observers 4

This was not just a hypothetical possibility. The fact that Britain could override the wishes of the Maltese government had been demonstrated as recently as 1958, when they suspended Malta’s constitution for two years.


97

from linking the Maltese rise in turnout to this change in status. Our own understanding of the generational basis of turnout change, confirmed by our findings in Chapter 3, lets us see the development in a different light: as an instance of a change that only affects new cohorts gradually coming to affect the whole electorate as a result of generational replacement. What about Switzerland? The World Handbook data give this country a low score on parliamentary responsibility because the government of Switzerland is a cartel in which the same parties form part of the governing coalition year after year (with the prime ministership rotating every year), no matter what the outcome of legislative elections. This arrangement (which the Swiss refer to as the “golden rule” and to which they attribute much of their vaunted stability) came into being in 1947 but became effective only after the election of 1959, when the Socialist Party was brought into the cartel and given its due share of ministries (Wernli 2001). The fact that this cartel became effective following the election before the one in which Swiss turnout began its long-term fall is certainly suggestive of a link between the two – though the link would not be apparent without an understanding of the generational basis of turnout change. Until 1963, Swiss legislative elections had some point to them. Supporters of the Socialist Party wanted to see their party given its due share of ministerial portfolios, and opponents of that party wanted to keep its representation as low as possible so as to defuse demands that it be given a share in governing. We can imagine, however, that those who entered the Swiss electorate in 1963 and after will have seen little point in voting in elections that would have no discernable policy implications (Wernli 2001:214–215). Elections that offered no real choices will not have been viewed as competitive. As these new cohorts progressively replaced the cohorts that entered the Swiss electorate at an earlier date, when elections were more competitive, turnout would progressively fall – exactly what happened in practice.5 5

Several other developments during the 1960s or shortly afterwards are also relevant. Extensions of the franchise to women and to eighteen-year-olds in 1971 accentuated the decline that started in 1963. We will return to these factors later. The 1960s were also a period when the number of referendums in which Swiss citizens were invited to vote increased dramatically (Wernli 2001:214). With all these additional opportunities to vote, one might suppose that the falling turnout in parliamentary elections

98


There is a neat symmetry to the fact that the variable whose change in value appears to have set Switzerland off on its long downward slide in turnout is the same variable whose change in value (in the other direction) appears to have set Malta off on its less extensive rise (less extensive only because turnout there very soon reached the ceiling above which turnout cannot rise any further – see Figure 4.1). Putting these two cases together, and looking at them in the context of other changes in the value of this variable (such as occurred in France at the end of the 4th Republic), it seems indisputable that the variable plays a crucial role not only in turnout change but in the level of turnout seen in different countries. Understanding Low Turnout in the United States6 We turn now to the United States – the third main outlier, along with Malta and Switzerland, in terms of turnout among advanced democracies. The United States is given a low score on the Banks measure of executive responsiveness because of the separation of powers there. This coding is not self-evident. Low voter turnout in the United States has in the past been blamed on a number of factors, but the separation of powers has not been prominent among them. One possible explanation for low U.S. voter turnout suggested in past research (Boyd 1981:145) is the frequency of elections. Among advanced democracies, only Switzerland and the United States call their voters to the polls more than once a year on average,7 and it has often been noted that these are also the two countries with the lowest turnout.

6

7

might be due to electoral fatigue, except that the level of turnout in referendums hardly changed. What happened was a convergence in turnout between parliamentary elections and referendums (Kriesi 1995:156–157). It is quite possible that the devaluing of parliamentary elections following the institution of the Swiss “Golden Rule” was exacerbated by the increasing use of referendums, but we will not be able to disentangle these two factors because they occurred more or less simultaneously. When we refer to the institution of the golden rule we thus refer to both phenomena. Much of this section was published in my 1998 article with Wolfgang Hirczy de Mino entitled “Divided Government and Turnout in US Presidential Elections Since 1840,” c 1998 Midwest Political Science American Journal of Political Science (42:1) 312–326. ( Association). I am grateful to the journal for being able to reuse this material. More than twice a year on average in the United States, according to Boyde (1981:145); about five times a year on average (mainly to render referendum verdicts) in Switzerland (LeDuc, Niemi, and Norris 1996:Table 1.2).


99

However, the fact (see footnote 5) that the increasing frequency of referendums in Switzerland did not cause turnout in referendums to fall raises a question as to the relevance of this variable. Thus the decline in Swiss turnout to U.S. levels probably has to be traced to some other development than frequent elections, and, that being the case, frequent elections lose plausibility as explanations for low voter turnout in the United States as well.8 Low U.S. turnout is often attributed to voluntary voter registration (Powell 1986; Teixeira 1992), but research suggests that automatic voter registration would only raise U.S. turnout by 8 percent at most (Mitchell and Wlezien 1995), leaving much of the difference between the United States and other countries unaccounted for.9 In the light of what happened to Switzerland after its main parties adopted their cartel agreement, it does appear that the strict separation of powers in the United States should be evaluated as a possible cause of low turnout. After all, separated powers remove the executive from legislative control. Even if the U.S. Congress is controlled by the same party as controls the presidency, this gives it little power over the president’s agenda (Burns 1963), and the decision as to which particular policies will in fact be enacted is made later, with a large input from interest groups and interested individuals. So the outcome of legislative elections in the United States does not determine policy,10 as it does in

8

9

10

This argument is not definitive. The increasing frequency of occasions to vote in Switzerland after 1959 might have played a part in declining Swiss turnout and been offset in the case of referendum votes by the relative increase in the importance of referendums in that country that was the mirror image of the declining importance of parliamentary elections. So frequency of elections may be part of the reason for low voter turnout in these two countries. Unfortunately it is not possible to tease out the separate effect of this variable when high values occur only in two countries whose low turnout can be attributed to other reasons as well, but we will see in Chapter 7 that there are additional reasons to suppose that the frequency of elections might have a bearing on turnout. Recent research into the effects of easing the registration process (through so-called motor voter laws) suggests that the effect is even less – no more than 5–6 percent (Franklin and Grier 1997; Knack 1999; Hanmer 2000). The countries of Western Europe typically have 30–40 percent higher turnout than the United States. Neither do the outcomes of presidential elections. A presidential candidate may promise that, if elected, he will try to enact health care reform (for example), but whether this reform will be enacted in practice is not within the president’s power to determine. Voters know this and do not normally even hold the president accountable for failure to deliver on such promises.

100


most parliamentary democracies, reducing the salience of this outcome to voters. A common denominator in theories of voting is that people are more likely to participate when they care about an election’s outcome (see in particular Downs 1957; Campbell 1960). The fact that turnout is higher when “issues of vital concern are presented” was noted in one of the very first statistical studies of turnout (Boecel 1928:517) and has been reiterated many times (e.g., Key 1964:578). Turnout increases when races offer clear contrasts between the policy stances of different candidates (cf. Grofman 1993). Anything that stands in the way of such contrasts also stands in the way of clear accountability for past government performance. Just so do separated powers – by dividing responsibilities between different branches of government – reduce the likelihood of enacting campaign promises, raise the information costs of assessing government performance, and dilute clear accountability. The idea that low U.S. turnout may be explained in part by the tenuous link between voting and the enactment of legislative programs is not new (Schattschneider 1960:100–101; Polsby and Wildavsky 1991:331; Asher 1992:56).11 Establishing the connection is more difficult, however, because of the uniqueness of the American case. Of the twenty-five countries with a readily available historical record of electoral statistics (Mackie and Rose 1991), the United States is the only one with strict separation of powers between legislature and executive, the only one with localizing electoral rules (Schugart 1995), and the only one except for France with strictly voluntary voter registration. So comparative multivariate analysis cannot tell us which of these variables matter. One strategy that overcomes this problem is to measure changes over time in the linkage between voting and its results. The idea is that if separated powers are bad for turnout, then anything that accentuates this separation should reduce turnout still further. So we are led to look closely at any variable that might affect this linkage. One such variable is divided government. When the same party controls the presidency and both houses of Congress, this oils the wheels of politics and makes it easier for that party to implement its program. More importantly, in such periods it is difficult for candidates of the 11

This is the “notorious gap” between promises and performance (Campbell et al. 1960:544).


101

ruling party to avoid blame for any failures. A divided government, by contrast, mutes clear policy directions (Mayhew 1991; cf. Wlezien 1996) and makes it easier to pass the buck for policy failures. Falling turnout has been blamed on the reduced sense that anything is at stake (Rosenstone and Hansen 1993). A divided government, by interfering with accountability, should certainly reduce the sense of anything being at stake and increase the information costs associated with attributing responsibility for good/bad times. The result will be less competitive elections with lower incentives to turn out.12 Longer periods of divided government should reinforce the sense that elections serve no function (which we assume to be a feature of truly separated powers), further lowering the incentives for electoral participation. By contrast, a return to a one-party government (especially if that government implements its program) should have the opposite effect – restoring some degree of clarity to electoral choices, reducing information costs, and perhaps even returning turnout to previous levels.13 Divided government thus serves as a litmus test for the importance of separated powers. By increasing or reducing the degree to which powers are separated, it can demonstrate that separated powers affect electoral competitiveness with consequential effects on turnout. Results of such a test have been reported in Franklin and Hirczy de Mino (1998). Table 4.1 shows their principal findings: the effects of divided government and distance between leading parties in three models of voter turnout. The first includes only the two variables of theoretical interest. This model shows the anticipated effects, significant at the .05 level, and explains a modest 34 percent of the variance in turnout. Taking account of the 1896–1916 turnout decline14 in Model B raises 12

13

14

A more elaborate version of this argument would state that only when the stakes are high will an election receive the sort of media attention that motivates marginal voters to go to the polls (cf. Mitchell and Wlezien 1995). During the period 1840–1992 there were thirty-nine presidential elections, of which twenty-one (53.8 percent) occurred during periods of divided government (ten of these occurred before 1900 and eleven of them after 1908). During the twenty years following the realignment of 1896, U.S. turnout both at presidential and at congressional elections dropped by some 16 percent on average – a drop that needs to be taken into account in order to properly specify the model. This method of taking account of the post–1896 turnout decline overfits the data, causing significant heteroskedasticity problems, as shown by the autoregressive conditional heteroskedasticity (ARCH) test for Model C in Table 4.1. Franklin and Hirczy de Mino find that the heteroskedasticity problem can be eliminated by replacing the

102


table 4.1 Effects of Divided Government and Distance Between Parties on Turnout (standard errors in parentheses) Independent variable

Model A

Model B

Model C

(Constant) Divided government (years) Margin of victory (%) Post-1896 turnout declinea Turnoutt−1

77.08 (2.93)∗ −1.1 (0.53)∗ −0.84 (0.20)∗

80.47 (1.35)∗ −0.49 (0.24)∗ −0.27 (0.10)∗ 18.4 (1.54)∗

73.94 (9.58)∗ −0.48 (0.25)∗ −0.26 (0.10)∗ 16.86 (2.73)∗ 0.08 (0.12)

Adjusted R2 Breusch–Godfrey test (5 df) ARCH test (5 df) N

0.31 22.38∗ 5.9 39

0.86 8.63 13.6∗ 39

0.86 8.59 12.93∗ 38

∗

Significant at p < .05 (one-tailed). A variable that takes on a value of 0 until 1896, 0.2 in 1900, 0.4 in 1904, 0.6 in 1908, 0.8 in 1912, and 1 thereafter to match the decline in turnout that occurred following this realignment (see text). a

variance explained to 86 percent and halves the coefficient for divided government while leaving its significance level unchanged (we discuss the substantive reasons for this decline in the next section). Including lagged turnout as a predictor in Model C – a standard test for the presence of time-serial dependencies – adds almost nothing to variance explained (the effect of Turnoutt-1 is less than its standard error) and hardly affects the value of any other coefficients. Based on this analysis, therefore, divided government matters. Moreover, its effects are cumulative. Let us now examine these effects in more detail. On average, four years of divided government precede any presidential election held under such conditions,15 so a coefficient for this variable must be multiplied by 4 to assess its average impact. On this basis, and using the coefficient from Model B, each successive election held under conditions of divided government reduces turnout by 1.96 percent, on average. After twelve years of divided government (the maximum period during which government remained divided in

15

post-1896 variable with a dummy variable that switches from 0 to 1 in 1906, but such a variable is less satisfactory theoretically and yields substantively identical findings. The question of what caused the post-1896 turnout decline will be discussed later. The onset of a divided government commonly occurs at a midterm congressional election (cf. Erikson 1988; Shugart 1995), so there are numerous cases where the number of prior years of divided government is two, accounting for the average mentioned here.


103

our data) turnout would be depressed by three times that amount, 5.88 percent. Separate analyses of nineteenth- and twentieth-century subsets of the data show much the same relationships in both centuries. These effects are not large, but they do not have to be. The object of this section was not to show that divided government is bad for turnout but that separated powers are bad for turnout. Analysing a divided government was just a strategy to get at the effects of separated powers in the absence of data about what happens when this institutional arrangement is introduced or abandoned. A divided government makes the separation of powers more marked, slightly reducing the already poor linkage between vote and public policy. Small differences in this linkage over time, brought about by the presence or absence of a divided government, have significant effects. This finding implies that the much larger differences over space, in terms of the presence or absence of separated powers, have even greater effects. In the chapters that follow we will use this finding to justify our use of the Banks measure of executive responsiveness as an indicator of the tightness of this linkage in different countries and different times. The fact that this measure renders redundant the dummy variable used in past research to pick out the United States (Powell 1986; Jackman 1987; Franklin 1996) will serve as further validation of this procedure. There is one objection that might be raised to the argument of this section. In the case of Malta, the prevailing account for that country’s high turnout was objected to, earlier in this chapter, on the grounds that turnout there had previously been much lower. In the case of the United States, the argument presented here could be objected to on the grounds that turnout there was previously much higher. During much of the nineteenth century in the United States, turnout ranged from 70 to 82 percent, in contrast to the range of 50 to 60 percent seen in the twentieth century; yet, in the nineteenth century, the separation of powers was no different from from what it is today. In answer to this potential objection it needs to be pointed out that even an average turnout of 76 percent is not that high in comparative perspective. The drop in turnout that occurred at the turn of the twentieth century can be accounted for by other things – perhaps the tightening of election laws, as assumed by some earlier scholars (Converse 1972, 1974; Rusk 1974), or perhaps the huge increase in the average margin of victory in congressional districts that followed the realignment of the 1890s,

104


as will be suggested later in this chapter.16 We saw in Chapter 3 that most of the 40-point difference that now separates U.S. turnout from average turnout elsewhere was taken up by this and other variables included in Table 3.4, leaving only a 16-point difference between U.S. turnout and turnout elsewhere to be accounted for by the U.S. dummy variable – a difference that may well have also existed in the nineteenth century. It is this 16-point difference that we attribute to the separation of powers. We will return to the effects of margin of victory in the next section of this chapter. Before we leave the topic of divided government it should be mentioned that the cumulative effects that we noted earlier, when a divided government extends over multiple presidential elections, are exactly what would be expected of an influence on turnout that was having its greatest effect on new members of the electorate. If the response to a divided government occurs primarily among new cohorts (as the findings of Chapter 3 would lead us to expect), then those effects will be slight if they last only for one election. If they continue to be felt at a second election, a second new cohort of voters falls under their influence. Since the number of years of divided government was never greater than twelve, the two occasions when a third election was held under conditions of divided government both resulted in a single-party government being restored. So there are no cases of a third election reinforcing the effects of two previous elections for any cohort. That is presumably why there is no indiction in Table 4.1 of residual effects from a divided government once party government has been restored. The effects apparently end as soon as the state of divided government itself ends, so that the first election conducted in conditions of undivided government sees a restoration of turnout to the level that would have been found had a divided government never occurred (controling for other factors), no matter how extensive its effects had been previously. In the light of the findings from Chapter 3, we can imagine that if a divided government had been reinstated by this third election, the members of one cohort (the one that was new at the first election held under conditions of divided government) would have been socialized by the experience of seeing a divided government reinstated on three 16

The mean margin of victory correlates 0.945 with the election laws variable used in the models in Table 4.1.


105

successive occasions, leaving a footprint in the electorate that testified to this experience and preventing turnout from completely returning to its previous level.17 Divided government is thus a potentially long-term force whose effects in practice are only short-term ones. In the next section of this chapter we will deal with a quintessentially short-term factor, but we will see that there are circumstances in which it could have long-term effects. Observing cumulative effects at work with a variable unconnected with voting age is important in demonstrating that the findings from Chapter 3 do apply generally to effects on turnout. The fact that the same variable (executive responsiveness) has been found to have had long-term effects in Switzerland and Malta but only short-term effects in the United States sets the scene for our discussion, later in this chapter, of the different ways in which independent variables need to be operationalized, depending on whether they can be expected to have cumulative effects. Turnout and Insulation from Election Outcomes The 16-point drop in turnout that followed the realigning election of 1896 raises a critical question that needs to be explored if we are to understand turnout change in the United States and elsewhere.18 It has often been observed (as already noted) that the closeness of the race affects turnout. What is seldom mentioned is that, in a country where elections are fought district by district (of which the United States is one of only a handful), the closeness of the race not only matters in terms of the overall election outcome but also at the district level. The more 17

18

Of course, since there were only two occasions when a third presidential election was conducted under conditions of a divided government, any socializing effect would not have produced a significant deviation from predictions in our data, so it is not impossible that on two occasions a slight footprint was left in the electorate as a result of twelve years of divided government. Our focus in this book on three elections as the magic number that divides expected short-term effects from expected cumulative effects does not preclude some long-term effects among voters in new cohorts any more than it precludes some short-term effects among voters in more established cohorts. Much of this section of the chapter was presented at a meeting of the Midwest Political Science Association (Evans and Franklin 2001). We are grateful to panel participants for helpful comments.

106


table 4.2 Correlations with Turnout at U.S. Presidential Elections, 1840–1988 House turnoutt−1 House turnoutt−2 Presidential turnoutt−2 House turnoutt 0.535 N 76 Presidential turnoutt 0.943 N 38

0.923 76 0.841 38

0.844 38 0.811 38

Note: t − 1 is the previous midterm election, and t − 2 is the previous presidential election.

districts with close races, the higher we can expect the overall turnout to be (see Franklin and Evans 2000 for further discussion of this point). At the turn of the nineteenth to twentieth centuries, not only were election laws tightened throughout the United States (Converse 1974; Rusk 1974), but there also was a realignment that greatly increased the number of safe seats in the congress (cf. Kleppner 1982). Of course, our analysis of the effects of a divided government in the United States was framed in terms of presidential turnout, whereas the safeness of the district relates more obviously to turnout in congressional elections. Since turnout in presidential races has virtually always (since the 1840s) been higher than turnout in congressional races, it is generally assumed that turnout in congressional races is determined by the importance of the presidential election rather than the other way around. This general assumption is false, however. Turnout in presidential elections is indeed generally higher than turnout in congressional elections, but it is the congressional election that appears more responsible for setting the level of turnout, with a presidential race apparently bringing to the polls an additional group of individuals who would not vote in an election that was uniquely a congressional election. This can be demonstrated in a number of ways. Table 4.2 shows the correlation of House and presidential turnouts at presidential elections (the t subscript in the table refers to presidential elections) with previous House and presidential turnouts (the t − 1 subscript refers to the previous midterm election and the t − 2 subscript refers to the previous presidential election).19 It shows clearly 19

The data for the analyses conducted in this section are taken from ICPSR study #0002 as reissued and updated in ICPSR study #7577 (1995). This release contains data only up to the election of 1990. Because the data would be hard to extend, and because we are primarily interested here in the evolution of turnout at the turn of the nineteenth


107

that turnout at presidential elections is strongly correlated (r = 0.94) with turnout at the previous midterm House election – far more strongly than it is correlated with presidential turnout at the previous presidential election (r = 0.81). Indeed, turnout in the House election at the time of the previous presidential election does a better job (r = 0.84) at predicting current presidential turnout than does presidential turnout at the previous presidential election. Another way of demonstrating the same thing is by means of a regression equation predicting presidential turnout from House turnout at the previous midterm election plus an adjustment for features of the national situation that come into play at presidential elections. Such an equation explains more than 88 percent of the variance in presidential turnout since 1840 (as we shall see in Table 4.3) – well above the 86 percent explained in Table 4.1. Both these demonstrations show the primacy of turnout in House elections.20 This book is primarily concerned with legislative elections, which means that most of our data for the United States refer to House turnout. Nevertheless, if we can understand House turnout, this will apparently contribute to our understanding of presidential turnout as well. But what explains turnout at House elections? In particular, what explains the huge decline in turnout at the start of the twentieth century, followed by a partial recovery at the time of the New Deal and more recent renewed decline after the 1960s (as we shall see in Figure 4.2). In the late nineteenth and early twentieth centuries, requirements for voter registration were instituted in most states in an effort to reduce or eliminate the “graveyard vote” and to make it more difficult to “vote early, vote often.” This tightening of election laws will have made it more difficult to vote and was a development that was widely understood, at least until recently, as being largely responsible for the post-1896 decline in U.S. turnout.21 That explanation, however,

20

21

to twentieth centuries, the final presidential election included in this analysis is that of 1988. The primacy of House election turnout is probably quite simply due to the larger variations in the determinants of House turnout. If the determinants of presidential turnout were to vary greatly over time while the mean margin of victory in House districts remained relatively constant, then presidential turnout would presumably dominate the interdependent system that these elections constitute. Mitchell and Wlezien (1995) argued that earlier studies had greatly overestimated the effects of voter registration. The recent apparent failure of motor voter legislation to

108


does not account for the ensuing evolution of U.S. turnout, since election laws were not changed further after 1920 until the 1960s. Moreover, the explanation makes the United States an anomolous case in comparative perspective (Franklin 1996, 2001). An alternative explanation was put forward by Burnham (1965, 1986), who blamed the early twentieth-century decline on a sectional realignment that followed the election of 1896. The resulting domination of the South and West by the Democrats and the Northeast and upper Midwest by the Republicans was so strong in many states that, within regions, the minority party’s capacity to launch competitive races was eliminated. In turn, the decline in intraregional competition may well have produced a sense that elections were no longer meaningful contests because the identity of the victorious party was predetermined. We have already seen (Table 4.1) that higher margins of victory bring lower turnout. Overall, the impact on the eligible electorate of the decline in electoral competitiveness was, according to Burnham, tantamount to political “desocialization” (Burnham 1974:1013) leading to declining participation (Burnham 1965). In other words, following the 1894– 1896 realignment and lasting nearly until the New Deal realignment, elections apparently lost their significance to a large proportion of the citizenry, who were thereby discouraged from voting. As seen in Figure 4.2, the drop in voter participation began immediately after those two realigning elections and continued until the mid-1920s. Table 4.3 gives the effects of mean margin of victory across districts together with overall margin of victory (in terms of votes cast for the Congress as a whole) on House turnout in midterm and presidential election years. It also shows the best prediction for presidential turnout that builds on our ability to explain midterm election turnout. Evidently our ability to predict turnout is best for midterm elections (Model A), where all that matters, apparently, is the marginality of individual races.22 When these are averaged across all races they explain nearly 92 percent of the variance in midterm election turnout, giving new meaning to Tip O’Niel’s dictum that “all politics

22

reverse the effects of registration reforms also call into question earlier assumptions in this regard (see, for example, Knack 1999; Evans and Franklin 2001). The models are well behaved in showing no evidence of autocorrelation or heteroskedicity (the tests for these conditions yield nonsignificant results). The reasons for concern about these conditions will be explained in the next chapter.


109

table 4.3 Explaining House and Presidential Turnout at U.S. Midterm and Presidential Elections, 1840–1988 Model A

Model B

House turnout at midterm election

House turnout at presidential election

Model C Presidential turnout

Mean margin across districts Overall vote margin (House) House turnout at previous midterm Overall vote margin (presidency) Constant

−1.279 (0.066)∗∗∗ −1.165 (0.081)∗∗∗

89.852 (2.150)∗∗∗

99.386 (2.343)∗∗∗

29.603 (3.451)∗∗∗

Number of elections (N) Adjusted R2 Breuch–Godfrey autocorrelation test (3 df) ARCH test for heteroskedasticity (3 df)

38

38

38

−0.312 (0.127)∗ 0.757 (0.061)∗∗∗ −0.222 (0.073)∗∗

0.919 1.00 ns

0.892 0.45 ns

0.884 1.27 ns

0.39 ns

0.41 ns

1.21 ns

Note: Prais–Winston regression estimates significant at ∗ 0.05, ∗∗ 0.01, ∗∗∗ 0.001. Standard errors in parenthases. ns = not significant (in tests for autocorrelation and heteroskedicity).

are local.” In a presidential year this is no longer true, and features of the national situation come into play (though only marginally) – the closeness of the race for the House as a whole in the case of House elections (Model B) and the closeness of the race for the presidency in the case of elections for the presidency (Model C). Even with the addition of the appropriate measure of overall margin of victory, we still do not explain as much variance in presidential year elections as we do in midterm elections, suggesting that in presidential years there are additional national factors that come into play over and above the closeness of the race. The importance of these findings in the context of our present concerns is to demonstrate the overwhelming influence of marginality in House races. As shown in Figure 4.2, the average margin of victory across the districts sending representatives to the U.S. Congress (taken

110


82

House turnout in presidential election years House margin of victory in presidential election years

r = −0.950 16 1840

1880

1914 year

1948

1988

figure 4.2. Changing turnout in presidential year congressional elections, 1840–1988, compared with changes in margin of victory. Note: House margin of victory is an index built from mean margin and overall margin of victory, weighted in proportion to their contribution to explaining turnout in Table 4.2, Model B (see text).

in an index together with changes in the overall margin of victory nationwide, where the two components are weighted in proportion to their regression effects in Model B, Table 4.3) moves more or less inversely with changes in turnout in House elections observed since 1840. The fit is not exact, but the correlation between the two series is exceptionally strong (r = −0.95), and it is possible to see the marginality of House races as providing a baseline from which actual turnout at House elections can drift for a time, but to which it tends to return. Long-term changes in House turnout apparently only occur with changes in that baseline. Importantly, the evolution of House margins does mirror the rise in House turnout during the 1930s and its subsequent fall after the 1960s, as well as the more dramatic turnout shift that occurred at the turn of the century. Since there were no further changes in election laws after 1920 until the 1960s, these findings show margin of victory to provide a better explanation than changes in election laws for the


111

overall evolution of House turnout since 1840.23 Increasingly competitive elections during the New Deal era have been explained by Kristi Andersen (1979) in terms of a challenge to the status quo in the context of economic crisis, and the more recent phenomenon of the “vanishing marginals” has been explained by Morris Fiorina (1989) and by David Mayhew (1974) in terms of incumbent legislators” manoevering to secure reelection. So an explanation for turnout change that focuses on electoral competition (district by district and over the country as a whole) is one that illuminates broad patterns in the evolution of U.S. turnout over the course of the two centuries. What we seem to observe is that competitive elections increase turnout, whereas uncompetitive races (which may result from natural causes, such as realignments, but which in the United States have more recently resulted from the efforts of politicians to insulate themselves from election verdicts) have adverse effects on turnout. We will return to this theme later in the book. The effect of changing margins of victory that we observe in the American case underlines the importance to voters of being able to affect by their votes the policy consequences of an election. If an electoral outcome is felt to be a foregone conclusion, it makes sense that fewer people will vote, because their vote cannot have any policy consequences in such elections. Our findings in the American case reinforce our findings in the Swiss and Maltese cases, leading us to expect influences on turnout from a number of other variables that affect the likely policy consequences of an election outcome. These findings also illustrate the workings of a quintessentially short-term factor that changes its value at every election.24 23

24

It might be argued that what really matters is not the margin in House races but the statewide margin (which also determines electoral college outcomes). Indeed, party dominance statewide could explain presidential turnout variations, raising the possibility that House turnout is driven by presidential turnout after all. However, state margins of victory (weighted by the number of congressional districts in each state before aggregating to the level of election years) correlate only −0.420 with presidential turnout and only −0.402 with House turnout in elections since 1840. Though for ease of exposition (and to simplify data analysis) we make a clean distinction between short-term and long-term forces, it is clear from Figure 4.2 that even a short-term factor like margin of victory can have a cumulative component if its level changes markedly (as it did in the United States at the end of the nineteenth century). Such a large fall in turnout could not occur purely on the basis of short-term forces. At each step in the seven-election fall in turnout, people who had already seen three

112


An Inventory of Variables Affecting Electoral Competition The effects on turnout of executive responsiveness and margin of victory provide us with reasons to expect certain other variables to also prove important in determining the level of electoral turnout. Both these variables are fundamentally concerned with the linkage between the voting act and policy change. In countries and at elections where executive responsiveness scores low, it does so because something (most notably separated powers, an appointed governor, or an anti-democratic cartel) stands in the way of an election having clear and immediate policy consequences. High margins of victory similarly stand in the way of the individual vote having clear and immediate consequences. Other things that could mitigate the policy consequences of an election should have similarly important effects. Among them are the question of whether an election’s outcome will be translated into policy without the intervention of coalition bargaining: The importance of an election will be less if the party that gains the most seats will have to bargain with other parties to determine the policies of an eventual coalition. But these variables are not binary. An election’s outcome is not either a foregone conclusion or not, and a party forced to enter into coalition bargaining can have a stronger or weaker position in that bargaining. Thus the precise degree of closeness is likely to be the important thing when considering the likelihood that votes can affect an election’s outcome (and we have already seen that turnout responds to the precise degree of closeness),25 and the precise size of the largest party is likely to be the important thing when considering the likelihood that a party will get its policies included among those of an eventual coalition government. Because it is not feasible to determine which parties are candidates for inclusion in each possible coalition government, we will measure this disparity simply in terms of the size of the largest party,

25

elections with falling turnout will have learned the habit of voting at a lower rate than they would have done three elections earlier, so the footprints of successive cohorts will have been of lower and lower turnout. Happily, we have no examples of this sort of hybrid short-term/long-term process during the period and for the countries that we study using multivariate analysis elsewhere in this book. Of course potential voters respond not to the actual closeness of a race but to its expected closeness. In this research (as has been ubiquitous in past research) we use actual closeness as an indicator of expected closeness.


113

what we will call its majority status (how far it diverges from receiving 50 percent of the votes). A party with 50 percent of the votes can be expected to rule alone, and hence to get its policies enacted (in the absence of impediments to executive responsiveness, which of course will need to be controlled for). As a party diverges from 50 percent in the downward direction, it has less chance of having its policies included unchanged in the government program, reducing the clarity of the expected election outcome.26 Other variables that should impinge on the link between voting and policy-making are the proportionality of the electoral system and the polarization and cohesiveness of the party system. Elections that convert votes into parliamentary seats with a high degree of accuracy avoid the “wasted vote” syndrome, where individuals may be discouraged from voting because their own vote does not seem likely to count for much. This could be either because they support a party that is not going to gain parliamentary representation or because they live in a constituency or district where one party enjoys such an overwhelming margin that voting either for it or for an opponent seems quixotic.27 Party polarization might prove important because parties that present policy choices that are more stark should excite more enthusiasm both among their supporters and among their opponents.28 Party cohesiveness should also prove important, because this variable determines whether a party that wins a nominal majority will be able in practice to marshal that majority behind its legislative proposals. A party system that is not very cohesive will be one in which election outcomes will not be very definitive.

26

27 28

As it diverges in the upward direction it should get less support because of a different mechanism: the fact that its victory is a foregone conclusion. It is, of course, not necessary that these two mechanisms be accounted for by a single variable, but it turns out empirically that one variable works better than two at indicating these two reasons for lower turnout (Franklin 2002). However, this effect (since it can only occur in plurality electoral systems) may well be subsumed by the average margin of victory in those systems. The stark policy differences between the two Maltese parties was proposed by Hirczy (1995) as one of the reasons for the high turnout in that country, but the idea has never before been tested empirically and it is possible that all elections seem important in this sense because people orient themselves to the alternatives on offer. Certainly turnout has not risen in the United States as its parties have become more polarized in recent years.

114


Two final variables are likely to have effects that could be considered part of the syndrome we are pointing to, even though they are not generally considered aspects of competitiveness: time since the previous election and size of the electorate. The length of time since the most recent election is also the length of time since voters most recently had the opportunity to make their choice between parties. If the most recent election was held not long ago, voters may not feel the same inclination to make their voices heard as they would if the most recent election had been held some time in the past (and especially if free elections had not previously been held). This variable can also be regarded as a measure of “electoral fatigue” (cf. van Egmond, de Graaf, and van der Eijk 1998; Norris 2002; Rallings, Thrasher, and Borisyk 2003). The size of the electorate has in the past been linked to the arrival of new voters with lower than average turnout (a mechanism we plan to operationalize more specifically through our measure of young initiation), but it could theoretically also affect turnout by devaluing the contribution to the overall electoral decision of individual votes in large electorates (Blais 2000; Franklin 2002; Norris 2002; but see Dahl and Tufte 1973 for contrary bivariate findings). A focus on this syndrome of variables should not be taken to imply that no other variables are important in determining the level of turnout in a country. Other variables can certainly prove important by reflecting administrative arrangements that encourage the voting act (compulsory voting, absentee voting), by facilitating the efforts of those who might work harder to get out the vote in an election that was close and had clear policy implications (union membership or close ties between parties and social groups) or by interacting with variables included in the syndrome, making their effects more or less pronounced (the relative size of the newest cohorts). Though we have a priori doubts about the potency of variables such as union membership or party– group linkages, for reasons given in Chapter 1, the importance of all of these variables can only be established empirically. Which Independent Variables Have Cumulative Effects? Before we can establish anything about the effects of these variables, we have to determine how their effects are likely to be felt. All of them should have a greater impact on new cohorts (citizens facing their


115

first, second, or third elections), but some will also have cumulative effects as these new cohorts replace older cohorts with different voting patterns. Why do all independent variables not have cumulative effects? The reason was already touched on in our discussion of the lack of longterm effects of divided government in the United States. It takes several consecutive electoral experiences to lock in a pattern of behavior, as explained in Chapter 3. Any variable that does not retain the same value for several consecutive elections does not generate effects that cumulate. In particular, variables like the size of the largest party and the closeness of the race are liable to vary considerably from election to election and may never acquire the opportunity to generate cumulative effects. Indeed, among the variables listed earlier in this chapter, the only ones likely to have cumulative effects are those that change their values and retain their new values over a considerable period. Such variables will, in general, be the institutional variables that relate to the type of electoral system employed in a country, the extent of the franchise, whether voting is compulsory, whether absentee ballots are available, and the variable that we believe explains turnout change in Switzerland and Malta: executive responsiveness. These are the variables hypothesized to have effects on turnout that, among the twenty-two countries that have conducted elections continuously since World War II, have changed their values in any country without that change being followed by further changes within the period of time (three elections) generally understood to be needed for the generation of cumulative effects. Notably absent from this list are the two variables that have been found to have effects in cross-national but not in over-time perspective: weekend voting and proportionality of the electoral system. Indeed, the absence of these variables from the list of those that have cumulative effects underlines the fact that we do have two different mechanisms at work in the effects that we measure on turnout variations. We mentioned in Chapter 1 that these two variables were found to have significant effects in cross-national perspective but not in over-time perspective. An understanding of the difference between short-term and long-term effects might explain why this is the case. Countries that change the day of the week on which they hold elections generally make such changes repeatedly, choosing the day as a

116


matter of political convenience. With some notable exceptions, much the same can be said about countries that change the proportionality of their electoral systems. In such cases we could only see short-term effects at work, and short-term effects of these variables might well be too small to be significant. No more than three cohorts are affected, or less than a quarter of each electorate, the remainder being already socialized into habits of voting or nonvoting. Countries that do not change the day of the week on which they hold elections, or the proportionality of their election systems, have (by definition) kept the same institutional arrangements in these two regards for long enough to have socialized each incoming cohort in turn into the turnout habits associated with that electoral arrangement to the point where the entire electorate has been influenced. Past research on country differences has found these two effects to be modest – some 6 percent for weekend voting and a 6 percent difference between the most proportional and least proportional countries – about half the difference in turnout associated with the presence or absence of compulsory voting (Franklin 1996:2001). Since less than a quarter of the electorate are susceptible to short-term forces (only the first three of thirteen cohorts), the overall effect of changing from least proportional to most proportional (or from weekday to Sunday voting) would be only about 1.5 percent – less than the effect required to achieve significance even with 350 cases. There are some countries that have changed their electoral systems in such a way as to permanently change the proportionality of those electoral systems: France in 1958, Austria in 1971, Italy in 1994, and Japan and New Zealand in 1996. In those countries, the changes might have given rise to cumulative effects. So this variable may well have given rise to different sorts of effects at different times as well as in different countries. In general it is thus critical to keep track of which variables can be expected to have cumulative effects and which variables can be expected to have only short-term effects, and to operationalize such variables appropriately in any models of turnout change. In the next chapter we will be using these insights as we design a research strategy to investigate the determinants of turnout change in twenty-two established democracies. But first we should take stock of the manner in which the distinction between short-term and cumulative effects fits into the theory that was developed in Chapter 2.


117

Short-Term Effects, Cumulative Effects, and the Habit of Voting Variables operationalized as short-term or cumulative effects are variables measured in an interaction that reflects the distinction made in Chapter 2’s Equation 6, which we repeat here: Pi (vote) = H i + {(1 − H i )( f (Ui (voting))}. As explained in Chapter 2, this equation asserts that, for each individual voter (i), the probability of voting is equal to the sum of two components: (1) Hi – the chances that the individual will vote out of habit, which we will operationalize in terms of whether that individual voted at previous elections – and (2) a function of the utility of voting multiplied by (1 − H) – the chances that the individual is not voting out of habit. In the light of the findings made in Chapters 3 and 4, it turns out that this formulation (complicated though it looks) is actually an oversimplification of an even more complex reality. Reality is more complicated than allowed for in equation 6 in Chapter 2 because each of the components of U (each of the variables listed in the previous section of this chapter) has the possibility of being subject to a different function. So {(1 − Hi )(f (Ui (voting))} might better have been written ((fj (Ui j (voting))), where the functions operationalizing each variable can differ from one another. Such functions would evidently incorporate the (1 − H) term as well as whatever further manipulations are required to handle the specific features of each variable.29 Whatever the function (whether it generates a short-term interaction or one of several different cumulative interactions described at the end of Chapter 3), it serves to limit the calculation of utility (and the evaluation of the character of elections) to just those voters who are not now (or were not at appropriate past elections) habitual voters or nonvoters – focusing variables that measure the utility of voting on

29

Even this simplifies a still more complex reality because “voting” was not a variable in Equation 6 but a term that stood in for the whole of another equation (Equation 5) in which different types of utility were distinguished. It is really each of those types of utility that we have broken into variables and those variables that have been subjected to different transformations that operationalize the differences between short-term and various types of cumulative factors. We will discuss this further in Chapter 8.

118


those voters, in fact, who pay attention to the character that elections have. Using the building blocks established in this and previous chapters, we are now ready to address the primary research questions with which this book is concerned. What causes turnout change? How can we understand the decline in turnout that has occurred so widely in recent years? In the next chapter we will address this question, employing data not for individuals or electoral cohorts but for elections. This requires that Chapter 2’s Equation 6 be viewed in aggregate terms and summed over an entire electorate, so that the i subscript vanishes and H refers to the level of past turnout, while (1 − H) refers to the proportion of impressionable cohorts, as will be explained.

5 Explaining Turnout Change in Twenty-Two Countries

The time has come to put what we have learned into a model that attempts to explain turnout change over the longest possible period for the largest possible number of countries.1 Twenty-two countries have a record of elections held continuously since within one electoral cycle (generally four years) of the end of World War II, amounting to 356 elections in all. We need to study elections held continuously because, if any elections are missing, we can investigate neither cumulative effects nor influences like the habit of voting that are coded with reference to past elections. Many countries that started holding elections shortly after World War II do not meet the criterion of continuous elections.2 If we attempt to increase the number of countries beyond twenty-two, we dramatically shorten the total period over which we can make the comparisons. At the same time, the period cannot be extended to include years before World War II because so many countries had at least two electoral cycles without elections before and during that singular cataclysm (Austria, Britain, France, Germany, Italy, Japan, Luxembourg, The Netherlands, and Norway) or did not exist (Israel). The countries 1

2

Earlier drafts of this chapter, co-authored by Michael Marsh and Patrick Lyons, were presented at conferences in Britain and the United States (Franklin, Marsh, and Lyons 2000, 2001). We are grateful to panel participants for helpful comments. We include Malta despite the suspension of its constitution in 1958 because this only created a gap two years longer than that country’s normal five-year gap between elections and because that country played a critical role in suggesting one of the primary theories to be tested in this chapter. No other country could have been added that did so little violence to the requirements for inclusion.

119


120 84

84

83

74

64

82

54 81 44 80 34 Turnout on 0-84 scale

79

Turnout on 77-84 scale

24

78 14 77

04 1945-49

1950-54

1955-59 1960-64

1965-69

1970-74

1975-79

1980-84

1985-89

1990-94

1995-99

figure 5.1. Average turnout in five-year periods for twenty-two countries, 1945–1999.

we can study from 1945 onward include those ten and also Australia, Belgium, Canada, Denmark, Finland, Iceland, Ireland, Malta, New Zealand, Sweden, Switzerland, and the United States. The period of our study ends in 1999 because data for some variables are hard to find for elections more recent than that. We study elections to the lower House of the national legislature in each country. We do not study presidential elections (where they occur) or midterm elections in the United States, because doing so would increase the weight of certain countries and might introduce anomalies if such elections do not respond to the same forces as elections to the lower Houses of other national legislatures. Some features of legislative elections in these countries were given in Table 1.1. Their overall evolution in terms of turnout is displayed in Figure 5.1, which shows a rise from the early 1950s through to the late 1960s, followed by a decline that appears to be continuing (though too much should not be made of a decline of only 6 or 7 percentage points, as mentioned in Chapter 1 and demonstrated graphically by the almost flat upper line in Figure 5.1). Research Strategy It should be evident from the chapters leading up to this one that our primary interest is in exploring the influences on turnout of various

Explaining Turnout Change in Twenty-Two Countries

121

features of the character of elections – features that relate to the utility and costs of voting that (perhaps enhanced by a message component) we believe to be the primary motivating forces that drive voter turnout (as explained in Chapter 2 and elaborated upon in Chapter 4). In addition to features that are strictly electoral (the type and proportionality of the electoral system, the extent of the franchise, whether absentee ballots are permitted, whether voting is compulsory, and district magnitude) we will consider aspects of an election’s character to include how much time has elapsed since the previous election, whether an election is held on a working day or over the weekend, the closeness of the race (both overall and on average across the districts or constituencies of countries with majoritarian electoral systems), the size of the electorate, the size of the largest party (seen in terms of how far it is from having the power to carry its policy proposals into law), the cohesiveness of parties in the legislature (the extent to which party discipline is maintained), their polarization (the extent to which their election manifestos make different appeals), and the responsiveness of the executive to changes in the balance of legislative forces. Because such a large proportion of the electorate is assumed to have made a standing decision to vote or not to vote, we also need to include a measure of inertia – a surrogate for the character of elections past (which we will indicate by the level of turnout at earlier elections).3 These variables have been introduced in the course of previous chapters, most extensively in Chapter 4. We have well-established expectations for the direction of effects to be expected from all of them, so we can test for significant effects using one-tailed tests. All of the variables are expected to operate by way of one of the major components in the calculus of voting developed in Chapter 2 and summarized in Equation 6 of that chapter. Either they are concerned with the proportion of the electorate that will be most largely affected by the character of the election (the proportion consisting of those whose habit of voting or non-voting is not yet established, component 1 − H in Equation 2.6) 3

This variable is a weighted average of turnout at the previous election with turnout at the election before that, giving twice as much weight to the more recent election. Very similar results are obtained if turnoutt-1 and turnoutt-2 are used together as separate measures of past turnout. In order not to lose the second election in each series (in addition to the first), past turnout for the second election in each country is set equal to turnoutt-1 .

122


or they are concerned with the costs of voting or nonvoting (compulsory voting, absentee voting, weekend voting – one part of component U in the same equation), or they are concerned with the motivation to affect the outcome, which will be highest in competitive elections (all other variables – the other part of component U). Operationally, the important distinction to be made is between variables that have their effects mainly on new cohorts – which we refer to as short-term factors – and variables whose effects on new cohorts are amplified by being repeated for cohort after cohort, eventually affecting the entire electorate – which we refer to as cumulative factors. Among the variables listed earlier, only absentee ballots, compulsory voting, extent of the franchise, size of the electorate, average district magnitude, and responsiveness of the executive are cumulative in nature,4 though the extent of the franchise provides us with two variables: one concerned with the lowering of the voting age and one concerned with (generally earlier) extensions of the franchise to women. The remaining independent variables change their values frequently enough (not in general retaining the same value for four elections or more) to yield only short-term effects.5 Since most short-term effects involve electoral competition, as does executive responsiveness, the model can be viewed as one that focuses on electoral competition as the primary driving force determining turnout levels. Both short-term and cumulative effects are quite easy to operationalize provided we have a measure of the proportion of the electorate that falls within the scope defined as “new cohorts” in Chapter 3 (those facing one of their first three elections). That proportion can be used to weight the changing values of the variables concerned, creating an interaction between each of the various measures of short-term forces and the proportion of the electorate upon which those forces are expected to operate most strongly. Changes in turnout can be expected to reflect both components of this interaction, occurring either when the substantive variable changes its value or when there is a change in 4

5

Electorate size and average district magnitude are particularly hard to code as cumulative factors, since increases, where they occur, are often followed by further increases. So the cumulative effect of these variables has to be a sum weighted not only by the size of the group affected by each change but also by the extent of each change. Cohesion is classified as a short-term factor even though it changed its values less frequently than this (see footnote 23 ).


123

the proportion of the electorate affected by the influence concerned – and especially when a large change in some factor coincides with a period when a large proportion of the electorate is responsive to such changes. Thus, for example, the closeness of the race at a particular election will matter more when more potential voters than usual are members of new cohorts. To measure cumulative effects we also need a measure of the proportion of the electorate affected by short-term factors. But, for the long-term factors, we are interested in the cumulative proportion of the electorate that was exposed to the changed influence during its formative years. Indeed, the only difference between short-term and cumulative factors is that short-term factors change their values before having had the chance to generate cumulative effects. Both have their effects primarily on new cohorts, so an accurate measure of the size of new cohorts is critical to our research strategy. This measure was described in Chapter 3. Since the size of the group consisting of new cohorts is a proportion bounded by 0 and 1, it follows that, when this proportion is multiplied by a substantive variable to generate an interaction term, the interpretation of coefficients for that interaction term is the same as whatever interpretation would have been made of coefficients for the substantive variable without interaction. The effect measured by such an interaction can be viewed as the effect on members of the group concerned (the smaller the group, the more the coefficient for the interaction is boosted; when the group constitutes the whole of the electorate, the interaction involves a multiplication by 1 and the resulting coefficient is unchanged). The effect measured for such an interaction can also be viewed as the effect that the variable would have if everyone in the electorate had been exposed to its influence during their formative years.6 To ease interpretation and comparison between coefficients, with two exceptions all variables have been scaled either as percentages or as proportions. The dependent variable, turnout, is measured in percentages, so all other figures related to votes are also measured in percentage 6

When considering an institutional change, this is the information a politician would need: the ultimate effect of the change once generational replacement has removed those unaffected by the reform.

124


terms. Measures relating to features of elections are generally dummy variables or proportions that vary between 0 and 1 (0 and –1, for convenience, in some cases). So variables with other scales have been reduced to a scale with a one-point difference (often –0.5 to +0.5).7 The exceptions are executive responsiveness and electorate size. Responsiveness is measured on a four-point scale, but most countries never move by more than one point on this scale. When considering effects of changing executive responsiveness in Switzerland, the coefficient must be multiplied by 4, the number of points by which executive responsiveness was reduced in Switzerland after 1959. Electorate size is measured in millions. In the case of extensions of the suffrage to women, our interest is not so much in the amount that turnout was depressed when the franchise was extended, since for most of our countries this happened long before the start of our period, but in the extent to which the presumed causes of lower turnout among women have been mitigated through the replacement of older women with women socialized after receiving the right to vote. For this reason we refer to the variable concerned as “female empowerment,” and the interaction of female empowerment with the proportion of women who had the opportunity to vote while still of an impressionable age is the effect that will pertain when all women have had that opportunity.8 This coding has two consequences. First, the effect should be positive, precisely because extensions of the suffrage to women will have caused a fall in turnout that, over time, will have been eroded as successive cohorts of women learned the habit of voting (Figure 3.3). Second, any extension of the franchise to women that actually occurs after the first election in our data for the country concerned will generate a large negative value for this variable. Female empowerment is low in the first election after the extension of the franchise to women, dragging the overall 7 8

The reason some variables are scaled with negative values is because values of 0 cannot be weighted by the proportion of the electorate affected by a change to that value. We halve the proportion of new voters to derive an estimate of this proportion. The resulting interaction is analageous to a count of the number of years since female emancipation (though scaled very differently), the measure used by Norris (2001) in research that has similar hypotheses regarding female emancipation to those put forward here. Note that this hypothesis runs counter to common assumptions about female turnout originating with Tingsten (1937), but recent research has not supported these assumptions. For an overview of such research see Norris (2001).


125

level of turnout down in consequence. The interaction becomes progressively less negative as female empowerment grows over the years that follow. By contrast, in the case of extensions of the franchise to eighteenyear-olds, we expect the interaction to become gradually more negative. This is because the process of giving to all of those who first vote at eighteen a worse experience of their early elections than they would have had at a later age (see Chapter 3) should generate a fall in turnout – a fall that should grow larger as the size of this group grows to encompass the entire electorate (Figure 3.2). It is the experience of being initiated into the political world at a younger age that is responsible for the effect, and for this reason we refer to the variable (as we have in previous chapters) as “young initiation.” It is logically possible for a variable to have both cumulative and short-term effects, if it changes its value repeatedly in some countries but not in others, or if variations occur around means that are very different at different times.9 So it is logically possible for measures of both short-term and long-term effects for the same variable to prove significant. It is also possible for the original substantive variables to prove significant, which would imply effects that are not differentiated according to whether voters are established – effects quite counter to the expectations we derive from the theory developed in past chapters. If such variables were to prove significant, this could mean one of two things: If the variable were constant over time but differed as between countries, it would indicate effects (“super-long-term” effects) due to causes long past; significant effects of any other kind would put the entire theoretical basis of this research into question – unless there are obvious reasons applying to specific variables that would account for the anomalies. Because all of these variables derive in one way or another from a measure of the size of new cohorts, they are going to be quite strongly related. So multicollinearity could be a concern in any analyses that employ them. But, more importantly, the dependence of so large a number of variables on the accuracy of a common component gives 9

We saw an example of dramatic change in a short-term variable’s mean in Figure 4.2. However, there are no examples of this sort of hybrid variable in the period (1945– 1999) that we investigate in this chapter.

126


us a large stake in how that component is measured, an issue that was considered extensively in Chapter 3.

Estimation Issues Estimating the effects of these independent variables on the level of turnout in established democracies is complicated by the nature of the data. Not only do these data contain variation over time, with between thirteen and twenty-two elections in each country, but also variation over space, with twenty-two different countries. To deal with interlocking sources of variation such as these, different estimation procedures focus on different aspects of this variation in different ways. In this section we will discuss the estimation problem in general terms and introduce the procedures that will be used later in attempting to do justice to the hypotheses investigated in this chapter. To employ a sequence of national elections in each of a set of countries we need to address two methodological issues. The first arises from the fact (shown in Table 1.1) that the amount of variance in turnout differs greatly as between countries. This suggests the likelihood of problems arising from estimation errors. When poorly estimated cases are not randomly distributed (a condition known technically as heteroskedasticity) this can result in inefficient estimates (Greene 1990: 416–418). The second issue is how to allow for the fact that voter turnout tends to retain much the same level over time in the same country, so that turnout at any given election can be quite well predicted from the turnout at the previous election (a condition known technically as time-serial dependency). Such dependencies could result in correlated error terms and biased regression coefficients (Maddala 1988:206–207). Starting with the problem of possible heteroskedasticity in the data, in the first instance we attempt to solve this by choosing a model that manages to fit country differences in turnout variations, and one of our concerns will be to see how successful we have been at this.10 10

Because a large role in our model is played by inertia, any large nonrandom variations in turnout at previous elections will be reflected in our estimate of the turnout at the current election. So we will tend to predict more variance in countries that have more


127

Two classes of approach exist for dealing with time-serial dependencies. The first approach attempts to adjust the coefficients so as to remove what is regarded as contamination due to what is referred to as an autoregressive (or ARl) process and so “clean up” the error term. When this has been satisfactorily achieved, the result is a set of so-called panels of cases, one panel for each country, consisting of replications of that country whose variables are measured repeatedly with randomly distributed error. This approach considers the time-serial dependencies as giving rise to nonrandom errors and tries to remove those errors. Even after removing the time-serial dependencies, however, the set of cases that constitute each panel are probably more like each other than they are like cases from other panels, giving rise additionally to the need for “panel corrected” standard errors (Beck and Katz 1995). The problem of possible time-serial dependencies can also be addressed by modeling them explicitly – including among the predictor variables one or more that measure the source of the dependency (Beck 2001). This approach specifies the expected contamination in the data instead of letting it fall into the error term. We have already made it clear that we expect turnout at earlier elections to have an effect on turnout at the current election by way of the habit of voting or nonvoting, so this approach is attractive. But using as an independent variable a lagged version of the dependent variable can result in inflated variance explained.11 It can also depress estimates of the effects of other independent variables (Achen 2000). Because we expect a time-serial dependency in our data (due to habits of voting and nonvoting), any technique that treats such dependencies as errors to be purged will not suit the research question addressed in this chapter. Moreover, omitting a lagged dependent variable that is called for theoretically may inflate the effects of other independent variables, due to exactly the same logic as leads one to eschew the use of a lagged dependent variable that is not called for theoretically. Specifically, omitting this variable could result in too much weight

11

variance, which is exactly what a good model should do. Whether this feature of our models succeeds in overcoming the heteroskedasticity problem is an empirical question that will be addressed with appropriate diagnostic tests. Because the regression procedures that we employ explain variance on different bases, for the sake of comparability the variance explained in our tables is always the square of the correlation between actual turnout and turnout predicted by each model.

128


being given to other variables that also predict stability (those that change little over time). However, incorporating past turnout brings with it a strong presumption of bias in the resulting estimates (Kennedy 1992:142) – a bias that comes from using a version of the dependent variable as though it were independent. This bias can be purged by employing a two-stage estimation procedure that replaces lagged turnout by a socalled instrumental variable that derives its values from the values of other independent variables that have also been lagged and are thus measured contemporaneously (Kennedy 1992:143). It will be seen that estimated effects of the character of elections are indeed quite sensitive to whether we include a lagged measure of turnout, but whether or not that lagged variable is instrumented will be seen to matter much less. It also does not seem to matter whether we consider our data as constituting a panel of independent cases for each of twenty-two countries or as a set of twenty-two separate time series.12 Nor does it much matter in practice how we enforce a focus on the over-time variability in the data at the expense of cross-country variability. A classic method of achieving this end (Cohen and Cohen 1983), still widely employed by political scientists, is to include among the predictors a dummy variable for each country (except for a base country whose dummy is omitted). The same possible objections can be raised to this method as have been raised regarding the use of a lagged dependent variable: Dummy variables to control for country differences might raise variance explained while depressing the effects of substantive variables. The alternative is statistically cleaner. This is to transform the data so that each variable is expressed in terms of its deviation from the mean for that variable in each country. This entirely removes country differences from the data, leaving (in our study) twenty-two separate time series rendered comparable by being transformed onto the same scales (though it might still be necessary, additionally, to remove country differences in the variance of certain variables should heteroskedasticity prove problematic in the manner discussed earlier). The trouble with this approach is that variables that had already produced their long-term consequences before the start of the time period covered in our data would have no 12

Whether the cases are considered separate or part of a time series is simply a matter of how they are analysed. We deal with the same 356 cases either way.


129

opportunity to demonstrate their importance. Myopically focusing on variance occurring during a specific period (even a period that extends for half a century) could arbitrarily exclude influences that were felt before the start of that period. In what follows we will employ both methods in tandem. The first method will be used to tell us whether there are significant country differences in turnout attributable to settled features of the character of elections that differ between countries, and the second method will be used to make the best possible estimates of the effects of over-time variations in independent variables. We will also demonstrate, however, that it makes little difference to parameter estimates which method is chosen, or whether country differences are eliminated through the inclusion of country dummies or by taking our variables in terms of deviations from country means.13

Hypotheses The hypotheses we can address with these data derive from the rational basis for choosing to vote or not that was put forward in Chapter 2. However, that rational calculus was concerned with the decision to vote or not in one particular election. To elaborate that calculus to handle multiple elections, we need to take account of the difference between short-term and cumulative factors, as already explained. Based on the notion that changes in the value of an independent variable had to be at least semi-permanent (last for at least four elections) in order to have cumulative effects, such effects were possible on the basis of only a small number of developments. These were the reduction of the voting age (in most countries), the removal of compulsory voting (in The Netherlands in 1971 and in Italy in 1994), the introduction of absentee voting (in Canada in 1972, Finland in 1970, and Switzerland in 1995) or its removal (in France in 1978), the loss of a cohesive party system (in Belgium in 1961, followed by its recovery sixteen years later), executive responsiveness, average district magnitude and the sizes of 13

All of our models are tested for robustness in various ways, and our preferred model is extensively tested for robustness in Chapter 7. A large number of different model specifications were cycled through and, where choice of specification appeared to make a difference, those differences are reported.

130


electorates (both of which were prone to successive increases over the course of the period under study), and the introduction of female suffrage (in Switzerland in 1971; Italy in 1948; Japan, Belgium, Malta, and France in 1946; and other countries at various dates prior to the first election in our dataset).14 Other variables were presumed to have only short-term effects, as already explained. Where it was possible to do so (in the case of absentee voting, party cohesion, electorate size, and average district magnitude) both shortterm and cumulative effects were measured, and for every variable the original version was tried in addition to short-term and/or cumulative versions. Effects of such variables would be interpreted as effects on all voters – “super-long-term” effects that only manifest themselves as country differences. We do not expect electoral characteristics that vary over time to affect all voters (see H4 below).15 Our resulting hypotheses are as follows: H1. We expect significant effects from the following variables operationalized as short-term factors: time since the previous election, weekend voting, majority status of the largest party (how close to receiving 50 percent of the votes), margin of victory of the largest party, mean margin of victory across the districts in majoritarian systems (coded 0 for nonmajoritarian systems), polarization and cohesion of the party system, decisiveness of the election, and disproportionality of the electoral system. H2. We expect significant effects from the following variables operationalized as cumulative factors: compulsory voting, absentee voting, female empowerment, extension of the franchise to eighteen-year-olds, size of electorate, average district magnitude, and responsiveness of the executive to the political complexion of the legislature. H3. We expect a significant effect of past turnout. 14

15

As an experiment, compulsory voting countries (Belgium, Luxembourg, and Italy) were excluded from the list of those who extended the franchise, on the grounds that compulsory voting would have forced women to vote as soon as they were enfranchised; however, a variable that coded these countries as countries in which the franchise was extended performed better. Operationalizations of variables employed in this chapter may be found in Appendix B.


131

H4. We do not expect any features of the character of elections that can vary over time (other than past turnout) to prove significant except when operationalized as a short-term or cumulative factor. H1 mainly involves electoral competitiveness, H2 mainly involves processes of generational replacement, and H3 involves inertia.

Preliminary Analyses Each independent variable can exist in as many as three versions if it made sense to calculate both cumulative and short-term versions in addition to the original (uninteracted) version. These different versions were tested against each other in a large number of preliminary analyses (not shown) in order to determine whether our expectations would be confirmed regarding which version was appropriate. Although the results of these preliminary tests are not shown (they were too numerous to make this feasible), once a preferred model has been established, later in this chapter, analyses will be reported demonstrating that no other (versions of) variables have significant effects when added to the preferred model. On the basis of preliminary analyses we found only one variable that performed better in its original form than in cumulative or short-term interactions. This was time since the previous election. This variable appears to act about equally on all citizens regardless of whether they are members of established cohorts or of new cohorts, in defiance of H4. That hypothesis was evidently overstated when it was taken to apply to all variables. When two legislative elections are held in close proximity (generally because a coalition government could not be formed on the basis of the outcome of the first election but sometimes because a minority government or a government with a bare majority hoped to improve its position) this appears to reduce the turnout of all cohorts, established as well as new. We will address the theoretical implications of this anomaly later in the chapter. Of considerable significance is the failure of any other simple variable to work better than the same variable employed in appropriate interactions. This means that variables such as weekend voting that might have obtained their cross-country significance from long-term

132


effects that occurred prior to the start of our period do not appear to have done so. We will return to this point when we consider in more detail the variables that do not prove to have significant effects in the analyses conducted in this chapter.

Findings Table 5.1 contains three models showing the effects on turnout of all variables found significant in any of these models (variables not found significant in any model will be listed later). The three models show the effects that derive from estimation procedures that retain crosscountry effects: GLS regression models with panel corrected standard errors that are further corrected for time-serial dependencies. Model A has no lagged variable or country dummies. Model B operationalizes the role of inertia (the H term in Chapter 2’s Equation 6) by bringing a lagged variable into play: turnout at previous elections. Model C further adds dummy variables for all of the twenty-two countries included in the data, except for Finland (the country that deviates least from the turnout predicted by the model).16 Of these models, Model B’s effects appear to vindicate Achen’s (2000) warning that incorporating a lagged dependent variable could reduce the effects ascribed to other variables. Model B’s effects are often considerably lower than Model A’s – sometimes implausibly low. Interestingly, however, these underestimates are not echoed in Model C, where country dummies are included. That model is the one that employs the method customary in political science for handling data of the kind analyzed in this chapter (time-series panel data). It also has the most plausible coefficients of the three models, with statistically significant effects of all of the independent variables except for absentee voting. Moreover, those effects are generally closer to those shown in Model A, the model without a lagged variable, than to those in Model B, where the use of a lagged dependent variable appears to have had biasing effects. Finally, Model C is well behaved in that the 16

The dummy variables have the effect of adjusting the intercept to suit individual countries, so that differences in mean turnout by country do not bias the parameter estimates for named independent variables. These adjustments are of no substantive interest and so are not reported.

133

334 0.720 12.29∗∗∗ 8.87∗∗∗

Note: Significant at ∗ 0.05, ∗∗ 0.01, ∗∗∗ 0.001 (one-tailed). a PCSE = panel corrected standard errors. b Includes dummy variables for all countries except Finland (effects not shown).

N of cases R2 Breusch–Godfrey autocorrelation test (3 df) ARCH test for heteroskedasticity (3 df)

334 0.915 4.96∗∗∗ 2.37∗

0.693(0.184)∗∗∗ −0.433(0.079)∗∗∗ −0.241(0.114)∗ −0.143(0.184) 3.760(1.689)∗ 1.776(0.774)∗ 1.130(0.563)∗ 0.759(0.546) −1.268(2.544) −0.014(0.013) −2.352(0.958)∗∗ 0.857(0.047)∗∗∗ 11.801(4.056)∗∗

0.547(0.162)∗∗∗ −0.523(0.146)∗∗∗ −0.484(0.149)∗∗∗ −0.282(0.317) 9.974(2.800)∗∗∗ 12.304(0.895)∗∗∗ 3.301(1.059)∗∗∗ 7.865(0.999)∗∗∗ 6.632(5.056) −0.070(0.023)∗∗ −4.664(1.904)∗∗

Time since previous election in years (0 to 7) Short term majority status (0 to 17) Short-term margin of victory (0 to 10) Short-term mean margin (0 to 10) Short-term cohesiveness (0 to 0.6) Cumulative compulsory voting (−1 to 0) Cumulative absentee voting (−0.5 to +0.5) Cumulative executive responsiveness (−1 to +2) Cumulative female empowerment (0 to 1) Cumulative electorate size (millions) Cumulative proportion of young initiation (0 to 0.82) Previous turnout (41 to 98) Constant 81.556(2.346)∗∗∗

PCSEa with lag

PCSEa basic

Model B

Variable (range of values)

Model A

334 0.940 0.42 1.51

0.643(0.178)∗∗∗ −0.402(0.089)∗∗∗ −0.356(0.128)∗∗ −0.675(0.265)∗∗ 7.736(2.157)∗∗∗ 12.897(3.656)∗∗∗ −1.523(1.708) 4.603(0.818)∗∗∗ 5.186(2.690)∗ −0.047(0.014)∗∗∗ −3.944(1.053)∗∗∗ 0.337(0.082)∗∗∗ 55.365(7.470)∗∗∗

PCSEa with dummiesb

Model C

table 5.1 Three Models Explaining Turnout in Twenty-Two Countries, 1945–1999, Using ARl Correction for Autocorrelation (panel corrected standard errors in parentheses)

134


tests for autocorrelation and heteroskedasticity, whose results are given in the table, prove to be not significant. One final feature of these models should be especially noted. The only independent variable that proves significant in its raw form is time since the previous election (as already mentioned). All of the other independent variables with significant effects are variables operationalized as either short-term or cumulative factors, largely confirming our fourth hypothesis. Though Model C tries to deal with country effects by including them in the model, we argued earlier that a better method for achieving this is to recast the analysis in terms of deviations from country means. Table 5.2 contains three models whose variables (both dependent and independent) have been transformed in this way (technically these are known as “fixed effects” models). These contain no cross-country variation, so dummy variables are not needed to account for such variation. Included are, once again, only variables that prove significant in one or more of the models. Again, the first model (Model D) includes no lagged version of the dependent variable, and the other models (Models E and F) do include such variables. The final model (Model F) attempts to cut any link between the lagged version of the dependent variable and each current time point by estimating that lagged version from the values of other lagged variables, as explained earlier. In these models we once again see how including a lagged dependent variable reduces the magnitudes of other effects. But in Table 5.2 this reduces the sizes of coefficients that are implausibly large in Model D to coefficients of more reasonable magnitude in Models E and F. Nevertheless, the effects of all the independent variables are significant in each of the three models. Models E and F are to be preferred, however. Not only do they include a variable (past turnout) that is called for theoretically, but also their coefficients are of a much more plausible magnitude, according closely with effects for the same variables found in past research. Even more importantly, both models are well behaved, having no significant autocorrelation or heteroskedasticity problems (as shown by the coefficients at the foot of the table).17 17

Multicollinearity also turns out to be unproblematic. The regression of all of the other independent variables on each one of them in turn yields a set of multiple correlations of which none come close to the multiple correlation between all independent variables and the dependent variable (Kennedy 1992:181).

135

334 0.660 10.40∗∗∗ 12.86∗∗∗

334 0.813 0.19 1.52

0.606(0.167)∗∗∗ −0.381(0.116)∗∗∗ −0.420(0.135)∗∗∗ −0.458(0.240)∗ 6.219(2.868)∗ 12.479(3.180)∗∗∗ 4.853(1.169)∗∗∗ 5.992(3.118)∗ −0.051(0.015)∗∗∗ −3.737(1.004)∗∗∗ 0.323(0.090)∗∗∗ 58.258(8.089)∗∗∗

Instrumented laga

Model F

312 0.799 0.30 1.29

Note: Significant at ∗ 0.05, ∗∗ 0.01, ∗∗∗ 0.001 (one-tailed). a Previous turnout is instrumented in Model F, using all other independent variables measured at the previous election.

N of cases R2 Breusch–Godfrey autocorrelation test (3 df) ARCH test for heteroskedasticity (3 df)

0.636(0.163)∗∗∗ −0.409(0.113)∗∗∗ −0.353(0.132)∗∗ −0.571(0.237)∗∗ 7.042(2.691)∗∗ 11.252(2.907)∗∗∗ 4.054(0.973)∗∗∗ 5.004(2.842)∗ −0.044(0.014)∗∗∗ −3.808(0.975)∗∗∗ 0.410(0.057)∗∗∗ 50.949(5.390)∗∗∗

0.526(0.175)∗∗∗ −0.498(0.121)∗∗∗ −0.432(0.142)∗∗∗ −0.701(0.254)∗∗ 9.424(2.791)∗∗∗ 17.956(2.967)∗∗∗ 7.678(0.901)∗∗∗ 8.772(3.008)∗∗ −0.064(0.014)∗∗∗ −4.774(1.039)∗∗∗

Time since previous election (0 to 7) Short-term majority status (0 to 17) Short-term margin of victory (0 to 10) Short-term mean margin (0 to 10) Short-term cohesiveness (0 to 0.6) Cumulative compulsory voting (−1 to 0) Cumulative executive responsiveness (−1 to +2) Cumulative female empowerment (0 to 0.5) Cumulative electorate size (millions) Cumulative proportion of young initiation (0 to 0.82) Previous turnout (41 to 98)a Constant 85.779(2.587)∗∗∗

Basic with lag

Basic

Model E

Variable (range of values)

Model D

table 5.2 Three Models Explaining Turnout in Twenty-Two Countries, 1945–1999, Using Fixed-Effects Regression (standard errors in parentheses)

136


Assessing the Findings It is satisfying to find such consistent estimates from six models that, while each of them is theoretically and methodologically defensible, employ quite different methods for dealing with over-time dependencies and country differences.18 Importantly, the deficiencies of each procedure are seen to be unproblematic in this research, since other procedures with different deficiencies give roughly the same results. In particular, removing country differences by including dummy variables for each country in the analysis is found to give almost identical results to those obtained when country differences are removed by taking each variable as a deviation from its country mean. Both techniques produce well-behaved models, free of significant heteroskedasticity or multicollinearity problems. But the inclusion of country dummies clearly inflates the variance explained and would make it problematic for us to use the equation to estimate turnout in any given country (as we will be doing in Chapter 7). Within any given country, of course, the country dummies are constants and would be dropped from the analysis. For this and other reasons the model we prefer (and will use in subsequent analyses) is Model E from Table 5.2. This model is based on more cases than Model F (which loses one additional case from each country in instrumenting past turnout) but nevertheless explains more variance. It also yields coefficients that (when they differ from those in Model F) are generally closer to the coefficients in Model C. Note that absentee voting was found significant in models A and B but not in Model C. It is omitted from all the models in Table 5.2 because, though its coefficient was sometimes more than 1.6 times its standard error (the threshold of significance for one-tailed tests), it always had the wrong sign (as it does in Model C). Absentee voting, though it appears to distinguish between countries with higher and lower turnout (as has also been found in past research), does not appear to cause turnout to rise when it is introduced into a country that previously did not provide for absentee ballots. To the contrary, the introduction of postal 18

Additional models were evaluated that included interaction terms for different sorts of electoral systems, on the assumption that variables might have different effects in proportional than in majoritarian systems. None of these models worked as well as the models presented here, though there were strong indications that the effects of female empowerment were greater in majoritarian systems.


137

ballots appears to be associated with falling turnout, quite contrary to expectations. We will return to this finding later. With this one exception, models that estimate their effects uniquely on the basis of over-time variation (Models C, E, and F) produce coefficients broadly comparable to those of a model that includes country differences in the variance to be explained (Model A). Omitting absentee voting from Model A hardly changes the other coefficients (not shown), which tells us that a fixed-effects model can adequately explain country differences, though the reverse is not true.19 Model E tells us that changes in the character of elections cause turnout to vary markedly. An election that occurs three years early will see turnout 2 percent lower, on average (0.64 × 3), than an election that follows a Parliament that runs to term.20 An election in which the largest party is 10 percent smaller or in which the margin of victory is 10 percent greater will see turnout drop by about 4 percent (0.41 × 10 or 0.35 × 10).21 In a plurality or “first-past-the-post” (FPTP) election, if the average margin of victory increases by 10 percent, turnout will drop by 5.7 percent.22 A country with compulsory voting will see turnout more than 11 percent higher, and one with a cohesive party system will see turnout 7 percent higher than a country without compulsory voting or with a party system in which party discipline is poor.23 Finally, a 19

20

21

22

23

Panel corrected standard errors are not calculated in a fixed-effects analysis, but the fact that there is no significant heteroskedasticity in Models E and F implies that the separate country panels are not sufficiently different as to require panel corrected standard errors, provided past turnout is included in the model. This is also implied by the fact that the standard errors calculated in Model C, where they are panel corrected, are very similar to those calculated in Model E, where they are not. If anything, standard errors are slightly larger (and thus more conservative) in Model E. Terms in different countries are not the same, of course. Our findings suggest that countries with five-year parliamentary terms will see turnout two-thirds of a percent higher than countries with four-year parliamentary terms, other things being equal. Remember that majority status (size of the largest party) is coded as the gap between the party’s size and 50 percent. The larger the gap, the less influence the party will have on policy; so the effect is negative. The interaction between margin and mean margin is even more powerful, indicating that people are even more likely to vote in a cliff-hanging district if the result is likely to affect the balance of forces in the legislature, and the interaction raises the variance explained in Model E to 0.83. However, because the coefficient is hard to interpret, we do not include this higher order interaction in the findings. This variable is a rather crude measure derived from the Banks data. Most countries are coded as cohesive, but the United States, Switzerland, Japan, Italy, and Belgium

138


country with an executive that is fully responsive to shifting majorities in the legislature will see turnout that is a whopping 16 percent higher (4 × 4.1)24 than one in which the executive is totally unresponsive – a finding that goes far toward confirming the intuition that was the basis for our anecdotes about Malta and Switzerland in Chapter 4 and also matches the estimate made in the same chapter for the effect of separated powers in the United States.25 As already mentioned, the model does contain an anomalous finding. Contrary to H4, time since the previous election shows significant effects for the uninteracted original variable rather than for a short-term or long-term version. This finding, though unanticipated, does make substantive sense. There are a number of circumstances that could cause people not to vote, which we generally see as random and unmeasurable. These include illness and obligations of various kinds that might take up all of a particular day. Such obligations can often be pushed to one side for an overriding reason such as the need to vote on that same day (just as a sick person might be willing to leave the hospital if the occasion seemed pressing enough). However, the more frequently this occurs, the less overriding the need to vote may well

24 25

between 1961 and 1977 are coded as not cohesive. The coding only changes over the course of the half-century for Belgium, but for that country it changes twice: from cohesive to uncohesive and back again. Because of that double change, the variable is classified as having short-term effects (the period during which Belgium is rated uncohesive is longer than three elections but too short to have permanent effects on more than a small fraction of the electorate). An alternate coding that treated cohesion as a cumulative factor did not have significant effects. However, the fact that more than the normal 3/13ths of the electorate will have been affected by these changes will have produced a larger coefficient than would have been produced had the two Belgian changes occurred closer together. The real effect of party cohesion is probably more like 5 percent. This variable is coded on a four-point scale. These examples simplify the findings by ignoring the fact (allowed for by Model E) that cohorts differ in size. The consequences of changing cohort sizes will be discussed in Chapter 7. They also ignore the fact that each interaction’s effect depends on both components of the interaction (another topic for Chapter 7). Evidently the standard errors are also conditional on the values of the component variables. In general we focus on the special case where the value of the cohort size components is 1 and the standard error for the interaction is as shown. This is a conservative strategy because the conditional values of the standard errors will be less for lower values of cohort size. We should note a methodological controversy as to whether interaction terms are interpretable without the presence of both of their components. Due to problems of multicolinearity, the presence of all three terms generally results in none of them reaching statistical significance in our data.


139

appear, compared with other obligations, and the likelihood that any particular person will find themselves in such a situation will not relate to whether they are new or established voters.26 Let us now turn to the question of the lowering of the voting age and the corresponding enlargement of the electorate. Chapters 1 and 3 presented two speculations about these developments: One was that the simple act of enlarging the electorate with individuals less likely to vote would cause turnout to fall. Our analysis provides no strong support for this hypothesis. The proportion of the electorate contained in its youngest three cohorts did not prove significant in any of the models presented in Tables 5.1 and 5.2, which is why it, along with several other variables, does not appear in those tables. Elaborating Model E in a separate analysis (as we shall see in Table 5.3) yields an effect for this variable of 2.5 percent – not large enough to reach statistical significance given its high standard error (which is why it is not included in Table 5.2). This coefficient appears larger than the estimate given in Chapter 3, but in fact the effect applies only to new voters.27 When it is averaged out across the entire electorate it comes to 0.7 percent – exactly the fall that the enlargement of the electorate was estimated to have had on the basis of reasoning from first principles in Chapter 3. Evidently this is a quantity too small to achieve significance in this analysis, but this does not mean that the effect is not real (see the note on statistical tests at the end of Chapter 1). The specific effect of lowering the voting age is more definitively established. All six of our models find a significant effect of young initiation, and Model E estimates that effect at 3.8 percent – a figure highly consistent with the 3.4 percent reported in Chapter 3. This number, of course, represents a drop in turnout that will not occur in full until every voter has entered the electorate at the younger age. With only a quarter of the total time needed to replace most electorates still to go, the effect tells us that close to a 3 percent fall in turnout (taking the two effects together) can already be attributed to the lowering of 26 27

This idea will be developed in Chapter 7, where we consider the possibility that our measure of time since the last election is wrongly conceptualized. The large standard error appears to be due to the fact that turnout does not invariably fall when there are large numbers of individuals in the first three cohorts. There have been several elections in which these cohorts were responsible for significantly increased turnout, as we saw in Chapter 3.

140


the voting age, where this occurred, and that a further 1 percent fall is still to be expected as a result of that reform. A skeptic might suggest that any division of the dataset into an earlier and a later component would produce a significant effect, and that the findings for young initiation merely reflect the fact of declining turnout that is so ubiquitous in contemporary established democracies. In fact, the best tests of the young initiation thesis do not come from the analysis conducted in this chapter, but from the cohort-level analysis conducted in Chapter 3 and the individual-level analysis to be conducted in Chapter 6. Nevertheless, it should be pointed out that the young initiation variable does not split the dataset into two at the same point in time for each country. The split occurs at the point at which the voting age was lowered – as early as 1966 in Germany and as late as 1995 in Austria. Because the young initiation variable is country-specific, it should (if the theory on which it is based is correct) do a better job of explaining turnout change than any split of the dataset into earlier and later parts. As a test we split the data into pre1970 and post-1970 parts, a split that should give maximum power to the resulting variable to index the fall in turnout that occurred so ubiquitously after 1970. When this variable is used in lieu of young initiation in Model E, it fails to achieve significance, and variance explained drops from 81.3 percent to 80.4 percent. So young initiation gains power from being tailored to the specific dates at which the voting age was lowered in different countries, making it hard to imagine that its effects are spurious, even leaving aside evidence from Chapters 3 and 6. The effects of female empowerment are significant in Models C through F. In Model E, the effect of 5 percent suggests that female turnout had risen this much on average by the time all women had enjoyed the opportunity to vote during their first three elections. This effect is probably underestimated because it is based mainly on the residual gender differences that long postdate the enfranchisement of women in most countries. There were indications from a separate analysis (footnote 18) of countries with majoritarian electoral systems that the effect of female empowerment was greater in those countries. The equivalent interaction term in the model that attempted to combine the separate effects for majoritarian and PR systems (see footnote 18) was 11 percent. Had this been statistically significant (which it was not),


141

it would have implied that, when women first entered the electorates of countries with majoritarian electoral systems, they were voting at a rate more than 10 percent below the rate for men in those countries, a gap that was then eroded by the cumulative replacement of older female cohorts. It makes some sense to assume that pressures on women to vote would have been greater in PR countries, since all of their votes would have immediately counted and immediately been courted, whereas in majoritarian countries a large number of women might (by virtue of living in districts where the outcome was a foregone conclusion) have been seen as irrelevant to the election’s outcome and consequently received little encouragement to vote. Still, we should not make too much of this on the basis of evidence from a small number of elections in countries that had granted the vote to women so long ago that the estimate is based on vestigial effects.28 Our lack of confidence in the analysis concerned is reflected in our preference for an unmodified Model E as a basis for further analysis. Finally, we should consider how to interpret the difference in variance explained between the models in Table 5.1 and those in Table 5.2. Our preferred model (in Table 5.2) explains considerably less variance (81 percent) than the models in Table 5.1 that included previous turnout (91.5 and 94 percent). The difference matters because the implication that flows from explaining over 90 percent of variance is that there can be little room for any other effect that would contribute substantively to our understanding of the phenomenon under study. With an additional 10 percent of variance unexplained, it might seem possible to improve the model. Much of the superiority of Models B and C over Models E and F is illusory. It comes from the fact that country differences in Models B and C are included in the variance to be explained, and country differences are very well explained by turnout at the previous election. Ironically, it can be harder to explain smaller amounts of variance than larger amounts. When variables are viewed as deviations from country means, we remove the variance that is easiest to explain. So the fact that we explain even 81 percent of the over-time variation in turnout within 28

No FPTP country granted women the vote on the same terms as men later than 1927. Nevertheless, this is a question that deserves investigation with more appropriate data than those collected for this study.

142


countries is impressive. But there is certainly room for improvement, as we shall see in Chapter 7. The high variance explained in Models B and C (Table 5.1) is partly illusory.

Effects of Other Variables Only ten substantive variables prove significant in Model E. Yet the literature on voter turnout has proposed more than a score of variables as being responsible for turnout levels and, by implication at least, for turnout change (see Appendix B). What about these other variables? Do they really play no role in an explanation of turnout change? Variables proposed in past research that impinge on the character of elections have already been listed in this and the previous chapter. All of them were tested for inclusion in our models, though some failed to prove significant. Those that did not survive preliminary tests comprise average district magnitude, disproportionality, weekend voting, decisiveness of the political system, polarization of the party system, the proportion of the electorate made up of new voters, and whether the government of the day was a coalition government. Table 5.3 shows the effects of each of these variables in turn when added to (a sometimes modified) Model E.29 Additionally, electoral system type is tested in this table because it has been proposed in past research, even though it was not hypothesized to add significantly to the more elaborate specification of the character of elections employed in this chapter. These variables were each operationalized in at least two ways, and Table 5.3 displays the effect (generally the short-term effect) that proved the more powerful. In the cases of absentee voting, weekend voting, and disproportionality, however, cumulative versions proved more potent even though not significant. As can be seen, none of these variables proved significant when added to Model E. However, they often fail to prove significant not because they play no role in voter turnout, but rather because their role is subsumed in the operation of other variables. As one example, electoral system type is subsumed in mean margin of victory, which only comes into play for countries with majoritarian electoral systems. Countries 29

The proportion of the electorate made up of new voters replaces electorate size when it is included in Model E (or any other model).


143

table 5.3 Effects of Further Aspects of the Character of Elections When Added, One at a Time, to Model E (standard errors in parentheses) Effect Independent variable as a cumulative factor Absentee voting Disproportionality Weekend voting

−1.518(1.189) −1.215(1.326) −1.690(1.207)

Independent variable as a short-term factor Average district magnitude (1–328) Coalition government Decisiveness of elections Majoritarian electoral system New cohorts proportiona Polarization of the party system

−0.032(0.020) 1.947(1.474) 1.549(4.026) −7.502(5.193) −2.499(4.491) 0.040(0.073)

Note: None of the coefficients are significant at the 0.05 level (one-tailed). a New cohorts proportion displaced size of registered electorate when included in Model E.

with proportional representation are coded 0 on this variable, matching the code that would be given to a majoritarian country in which the race in virtually every district was too close to call. The fact that this variable displaces electoral system emphasizes the reason why electoral system type is important. PR systems attempt to make every vote count everywhere.30 In majoritarian systems, every vote counts only in districts that are close fought. If all districts are close fought, then majoritarian systems perform like PR systems in terms of the wasted vote syndrome; but if the race is a foregone conclusion in even quite a small number of districts, then the number of votes wasted overall could be quite large.31 In districts where the race is a foregone conclusion, many potential voters will not see the point of voting and turnout will be lower. The extent to which this happens is indexed by the mean margin of victory district by district. This provides a better measure of the wasted vote syndrome than a simple dichotomy between majoritarian 30 31

The extent to which they in fact count everywhere depends on threshold requirements. Fewer votes are wasted as thresholds are lower. In a country such as the United States, where over 90 percent of the districts are considered safe, the proportion of wasted votes overall is very high indeed.

144


and other electoral systems. The same variable appears to displace disproportionality, which perhaps gained its power in previous research from serving as a second indicator of the wasted vote syndrome – again not nearly as effective an indicator as mean margin of victory.32 Another example is to be found in coalition government. Countries where coalition governments are the norm add a step to the process of translating votes into policy. Once the votes have been counted, the parties have to bargain over which policies will be pursued by any government that forms, blurring the connection between election outcome and policy. But this consideration is subsumed by the majority status variable – a measure of the distance of the largest party from receiving half the votes. The smaller this gap, the greater the power of the party to get its way. With slightly more than 50 percent the party can govern alone. With slightly less than 50 percent it can dominate any coalition of which it is part. But as the largest party’s size drops from 50 percent, so it loses its ability to dominate a government. This is a much more subtle indicator than a simple dichotomy between coalition governments and single-party governments, and it is not surprising that such a dichotomy is displaced by the majority status variable. A third example is decisiveness of elections. This variable is clearly subsumed by executive responsiveness, which also subsumes bicameralism (a variable we do not explicitly test since it is so clearly subsumed in one or the other or both of the above).33 We have already addressed the effects of the sheer size of new cohorts. Remaining variables that fail to explain turnout, despite their role in defining the character of elections, fail for no apparent reason. They just do not add significantly to the explanatory power of an already powerful model, whether in their raw form, in interaction with new cohorts, or as cumulative effects (where appropriate).34 The most interesting failure is weekend voting (which has repeatedly been found 32

33 34

The United States, to take one anomalous example, has a very low index of disproportionality (the proportion of seats won by each party is generally very close to the proportion of votes cast for that party), yet the United States is a country in which huge numbers of votes are wasted in districts that are safe seats for one party or the other, as we saw in Chapter 4. Bicameralism is dropped from the equation when we try to explicitly test it because of multicolinearity with other included variables. Average district magnitude proved barely significant in the six countries investigated in Chapter 3 but not in this wider universe.


145

significant in past research into country differences in turnout). Weekend voting was also found in Franklin (2002) not to have effects in over-time perspective even though its cross-sectional effects were confirmed in the same study – apparently a spurious finding emblematic of those that can so easily occur in small N studies. Absentee voting is another variable found to have positive effects in past research that loses those effects when investigated in an over-time perspective. Its effects in a cross-national perspective were not spurious (like those of weekend voting) because we replicate them in this study when country differences are taken into account, but the introduction of absentee voting does not have the same effect as its long-term presence. We will consider this apparent paradox in more detail later. Another interesting failure is party polarization. It will be recalled that the distinctiveness of Maltese parties was one of the reasons given (Hirczy 1995) for the high voter turnout in Malta. Yet the polarization of party systems, as measured by the parties’ manifestos (see Appendix B), does not appear to be responsible for turnout differences. One possible reason for this unexpected finding would be that voters orient themselves toward the choices on offer in their own political system and take seriously the difference between choices on offer, however small these differences may be. Looking beyond variables having to do with the character of elections, variables expected for other reasons to play a role in determining turnout levels have to do with features of social structure that either reflect the resources of a society or contribute to the ease with which its members can be mobilized. Among the latter we include disaffection and alienation on the grounds that alienated individuals might be hard to mobilize.35 We do not expect such variables to prove significant in a properly specified model, as explained in Chapter 1. But there is a problem in implementing tests for these variables. Should they be taken in the form proposed by previous authors, or should they too be converted by appropriate manipulation into indicators of short-term and/or long-term forces? Since our theory does not call for

35

These variables are often thought of as aspects of political culture, but we do not use this as a category for distinguishing the influences on voter turnout in this book, regarding it as a residual category that would only have to be wheeled into play if more specific forces provided insufficient explanations.

146


these variables to have any effects, it does not tell us whether such effects would be short-term or cumulative in nature. And, for most of them, defining short-term and long-term versions is logically problematic. What would a long-term version of gross domestic product (gdp) per head, for example, look like? What would a short-term version look like? Why should such a thing affect only new cohorts, or have cumulative effects? There are also problems associated with collecting the relevant variables for all countries and years. Several of them have not been measured for years as recent as 1999, and some of them (union membership is the prime example) are unobtainable for certain countries. Nevertheless, simple (i.e., neither cumulative nor short-term) versions of all of the variables believed to have been used in past research were collected and coded, and all of them were correlated with the residual from Model E – used as a measure of what was still to be explained after the variables included in that model had done all that they could.36 None of the correlations proved significant even at the .10 level.

Implications The findings of this chapter go far toward demonstrating the aggregatelevel implications of the cohort-level findings reported in Chapter 3. By taking into account the size of new cohorts, we can measure the short-term effects of variables whose values change too frequently to have long-term consequences. By keeping track of the cumulative proportion of the electorate that has had the opportunity to be affected by permanent changes in the character of elections, we can additionally measure those variables’ cumulative effects. Four variables having to do with electoral competition have shortterm effects: size of the largest party (implemented as majority status), 36

The variables were literacy percent of population, university educated percent, gdp per head, union membership percent of workforce, population density, links to social groups, party attachment, and trust in government (as a measure of alienation). Variables available for only part of the total dataset were correlated with the residual over those countries and years for which they were available. All the variables tested in this chapter are listed in Appendix B, together with references to past studies that have employed them and indications of the sources from which they were derived.


147

margin of victory, mean margin of victory (for majoritarian countries), and party cohesiveness. Four variables having to do with institutional arrangements have cumulative effects: compulsory voting, female empowerment, young initiation, and executive responsiveness (the last of these is also an aspect of electoral competition). A further variable with cumulative effects is the size of the electorate. Finally, time since the last election has effects both on new and on established cohorts – the only variable to perform better in its untransformed state than when implemented as a short-term or cumulative factor. Using variables implemented in short-term and cumulative versions, we explain turnout change in established democracies better than in past research that took no account of the short-term/cumulative distinction.37 More importantly, our strategy for determining the differential impact of different sorts of change on different groups of voters produces a model that unambiguously establishes the primacy of electoral competition in influencing turnout levels. Turnout change is not brought about by changes in the character of society or of its members. When we try to add variables that measure the proportion of university educated or of union members, or other features of a society that indicate its citizens’ resources or their susceptibility to mobilization efforts, these variables do not add significantly to variance explained. So commentators appear to be mistaken who see in declining turnout, where it occurs, an indication of unfortunate trends in the composition of society or the attitudes of its members (cf. McDonald and Popkin 2001). Turnout declines, if it does, because elections change their character. It is true that changes in the age structure of a society do increase or reduce the responsiveness of its citizens to the character of elections

37

Without taking account of the short-term/cumulative distinction, Gray and Caul (2000) explained 72 percent of the variance in a dataset consisting of eighteen of the countries studied here over a slightly shorter period. When our universe of countries and elections is restricted to match theirs, the variance we explain falls fractionally to 81.1 percent but still remains well above theirs. Franklin (2002) explained 71 percent of the variance in a dataset consisting of rather more countries (thirty-one) but over varying periods. The analysis conducted in this chapter cannot readily be made comparable with that one because gaps in the series of elections for many of the additional countries make some of our variables impossible to operationalize. In Chapter 7 we will estimate a model that contains no interaction terms, to see what difference is made by taking account of the short-term versus cumulative distinction that has figured so prominently in this chapter.

148


(by changing the proportion of the electorate likely to react to an election’s character), and in interaction this plays an important role in turnout change, as established in this chapter (this point will be elaborated in Chapter 7), but it is clear that the primary driving forces in turnout variations are changes in the character of elections. Of considerable interest is the fact that a focus on turnout change somewhat reduces the apparent importance of institutional differences on turnout levels, compared to the implications of previous research. In particular, institutions whose role has been seen as simplifying the voting act and reducing the costs of voting (weekend voting and absentee voting) do not have the significant effects in this research that have been found in previous research. Weekend voting loses significance apparently because its effects in past findings were spurious – the result of analyzing data that compared average turnout across a small number of countries. With an N of only 20 or so cases, spurious findings are easily come by, and weekend voting appears to have been one such. The same is not true of absentee voting. Its cross-country effect appears to be real and is replicated in this research. However, the provision of absentee ballots in a country where they were previously unavailable (or less readily available) does not cause turnout to rise. Perhaps because absentee voting is sometimes introduced as an attempted cure for poor turnout (as in Switzerland in 1995), the circumstances that surround its introduction may vitiate the purpose for which it was introduced. Or its effects may simply be overwhelmed by whatever circumstances were already causing turnout to fall in the country concerned. In such a case the causal link would be the reverse of what is generally assumed: Falling turnout would be leading to the introduction of absentee voting, and the negative effect we measure would be picking up this causal link. If this is so, then a model that includes absentee voting as an independent variable is badly misspecified. It is for fear that this is indeed the case that we omit the variable from the models in Table 5.2. Finally, the findings of this chapter vindicate the model proposed in Chapter 2 by showing clearly the importance of distinguishing between voters who are susceptible to influence from the character of elections (the 1-H term in Chapter 2’s equation 6) and those who are not (the H term in the same equation). Equation 6 did not distinguish, among habitual factors, those that apply to all potential voters and those that have not been in place long enough to do so. At the individual level at a


149

particular election (the context for which Equation 6 was designed) the question of which influences on the utility of voting are ephemeral and which become rooted in habitual behavior is not relevant. The transfer of influences from one category to the other is something that happens with the passage of time, and time was not a component of any of the equations in Chapter 2. Another feature of Equation 6 that is vindicated in this chapter is the role of uncertainty. As pointed out at the end of Chapter 2, uncertainty plays two different roles in the model developed there. All of the short-term factors, together with executive responsiveness, are variables whose theoretical rational has to do with uncertainty in one or the other of these two forms. If the outcome of the election hangs in the balance, then uncertainty about the outcome renders the election highly competitive and turnout will be high (cumulative executive responsiveness also operates in this fashion). By contrast, if majority status is such as to make the policy stakes of the election unclear, this reduces turnout because this form of uncertainty makes elections less competitive and yields less reason for voting. The findings reinforce some of the points made in Chapter 1 about the need to focus on turnout in context. For a specific population at a specific point in time, the context of the election is set. It is a constant. Multivariate analysis cannot take it into account. Multivariate analysis can only take account of context if there are multiple contexts to take account of. In the next chapter we will discover what difference it makes to an individual-level analysis of turnout when electoral context is taken into account.

6 The Character of Elections and the Individual Citizen

So far in this book we have focused our analyses on electoral cohorts (in Chapter 3) or on electorates (Chapters 4 and 5) as units of analysis. Yet, in the end, individuals are the ones who make decisions to vote or not vote, so we need to complete the picture with analyses conducted at the individual level using survey data. To be sure, survey data have been used before, in Chapter 3, as a source from which to derive the electoral cohorts that were the subject of that chapter. For that chapter, however, only two variables were taken from the individual-level data: date of birth and whether the individual voted or not. Already given the country in which the data were collected and the date of the election about which respondents had been questioned, these two additional pieces of information were all that was needed to construct the electoral cohorts that were studied in that chapter. The present chapter will be employing a larger number of individual-level variables to paint a more detailed picture of the individual citizen faced with an election. The primary research question that we set out to address in this chapter is, “How do changes in the character of elections manifest themselves at the individual level?” We now know that more competitive elections (those that are likely to result in policy change) bring more citizens to the polls. But what is it that brings these citizens to the polls? If the election’s character acts on them directly, then previous models of voting behavior have been badly misspecified in omitting measures of the character of elections as independent variables. However, it is 151

152


quite possible that an election’s character is manifested at the individual level by way of variables already included in most election studies. Important elections could increase the political interest of individual citizens or give them greater sympathy for one of the parties, or give them a greater sense of party identification – all variables included in past individual-level studies. Even if an election’s character manifests itself by way of variables that are normally included in models of electoral behavior, there is still a question as to how this happens. Are changes in political interest, for example, felt by all citizens, or are they felt mainly by citizens in their formative years? Past research at the individual level has generally assumed that everyone is affected in the same way by the influences that determine whether they vote. Research that has distinguished among different types of citizens has most often distinguished between them on the basis of their political orientations (Marsh 1991; Blondel, Sinnott, and Svensson 1998). Our research in past chapters has suggested that both approaches miss an important feature of individuallevel behavior: the tendency of citizens to become set in their ways. The models that were confirmed in Chapters 3 and 5 suggest that those who will be most affected by the character of elections are those who are not yet set in their ways – members of new cohorts. If this suggestion is correct, then models employed in previous research will have been misspecified in omitting interactions with respondents’ cohorts, even if it turns out that no similar problem arises from the omission of variables defining each election’s character. If neither of these expectations is fulfilled – if no significant effects are found either from interactions with new cohorts or from the addition of variables measuring the character of elections – then we will be led to question the findings of earlier chapters. Establishing the expected effects, on the other hand, will provide important confirmation of those findings. This chapter is more exploratory than earlier chapters of this book. It proceeds by building a model of voter turnout as close as possible to those used in past research and then elaborating that model by incorporating the insights reached in earlier chapters. As we compare successive models in which variables are added that are suggested by the findings of past chapters, we will be looking for changes in the coefficients of variables that appear in both models, and

The Character of Elections and the Individual Citizen

153

gains in variance explained if the revised models do a better job of accounting for whether people vote.1 In order to conduct this exploration we need to pick a country to study. Analyzing two or more countries separately is impractical. Either we would need to devote more than a chapter to the task or we would not do the countries justice. More than a single chapter would be overkill in terms of our objectives.2 Combining the data for two or more countries is impractical because of the small number of variables available in identical terms across more than one country (see Table 6.1). The data for this chapter come from academic post-election studies conducted in Germany since 1965. Three earlier studies conducted in that country could not be employed because respondents had been asked neither their exact age nor their date of birth, and so could not be placed in age cohorts.3 Using German election studies brings several advantages. German election studies have fielded a reasonable number of questions in identical or virtually identical terms across most of the elections studied. More importantly, Germany is a country in which turnout has risen as well as fallen in recent years, so findings will not be contaminated by possibly correlated trends. It would have been desirable to add data from additional countries, as we did in Chapter 3. However, date of birth (or age) and whether the respondent had voted (the two variables that were all that were needed from survey data in that chapter) come close to exhausting the set of variables asked in identical terms across multiple countries. Some

1

2

3

This is not the first time that attempts have been made to investigate possible interactions between aggregate- and individual-level effects on voting. However, past attempts have focused on institutional factors at the aggregate level (Franklin et al. 1996; Perea 2002), the generally fixed portion of the character of elections. In this chapter we focus on the effects of electoral competition. There is no reason to suppose that the findings would be substantively different for another country. We are not applying a theory tailored to any particular country. Our theory is expressed in general terms and should work in any country. The main differences in practice should come from having to use a different set of variables to indicate the concepts of interest. The 1965 study had the same problem of asking respondents only to place themselves in five-year age groups, but these groups overlapped sufficiently with electoral cohorts for us to be able to use them as surrogates for actual ages. We could not use the same procedure for elections prior to 1965 because the quasi-cohorts created for any earlier election would not match those created for 1965 and later.

154

Note: Significant at ∗ 0.05, ∗∗ 0.01, ∗∗∗ 0.001 (one-tailed).

0.002 −0.002 −0.068 −0.363∗∗∗

Linkage variables Member of new cohort (0,1 = yes) Established voter (0,1 = yes) Young initiations (0,1 = yes) Voted at previous election (0,1 = yes)

0.194∗∗∗ 0.187∗∗∗ 0.252∗∗∗

Campaign variables Political interest (0–1 = high) Party ID (0–1 = strong) Party sympathy (0–1 = strong) −0.085∗∗∗ 0.005∗∗∗ 0.012

0.025∗∗∗ 0.075∗∗∗ 0.032∗∗∗ 0.042∗∗∗

Mobilization variables Union member (0,1 = yes) Religion (0,1 = Christian) Church attendance (0–1 = weekly) Urban (0,1 = most urban)

Character of elections variables Majority status, percent (0 = 4 to 1 = 14) Margin of victory, percent (0–1 = 13) Time since last election, years (3–4)

0.005 0.017∗ 0.048∗∗∗ 0.065∗∗∗ 0.071∗∗∗ 0.014

Correlation with voting

Resource variables Age (years) Gender (0,1 = male) Education (0–1 = higher) Married (0,1 = yes) Respondent’s occupation (0–1 = white collar) Household occupation (0–1 = white collar)

Variable

(Needed to evaluate the character of elections) Entered electorate less than four elections ago Entered electorate four elections ago or earlier Entered electorate after lowering of voting age in 1968

(From aggregate data) Percent differences of largest party from 50% of votes

Rescaled Rescaled Rescaled

Non-Christians include other religions and nonreligious Rescaled Less than 50,000 versus more than 50,000 population

Rescaled Occupation of head of household

Rescaled

Rescaled

Notes

table 6.1 Individual Level Variables for Explaining Electoral Participation in Germany


155

variables can be added after judicious recoding, but, even sticking to a single country, there are only thirteen variables relevant to studying turnout that are sufficiently similar across elections to be employed in this chapter. Adding just one more country would cut the number of suitable variables to eight. Though not perfect, the German data allow us to do justice to the alternative approaches outlined in Chapter 1 of this book: the resource and the mobilization approaches to explaining voter turnout. Its findings should be regarded as illustrative of the sorts of effects that will be found in other countries as well. Table 6.1 lists the available variables, divided into categories according to which approach they reflect. Each variable is accompanied by its correlation with voting and other information. Also listed are aggregate-level variables measuring the character of elections that we use in conjunction with the survey data, as well as the linkage variables that we need to test at the individual level the young initiation hypothesis already tested in Chapters 3 and 5. Campaign variables are considered separately from mobilization variables in Table 6.1 to draw attention to our separate expectations for the behavior of these variables when measures relating to the character of elections are introduced into our analysis. Campaign variables may well carry some influences from the character of elections if more competitive elections serve to stimulate political interest or make people feel closer to or more in sympathy with their preferred party. Table 6.2 shows the results of an analysis that attempts to classify respondents into those who voted and those who did not on the basis of a logistic regression analysis. The reason we cannot use OLS regression in this table is because the dependent variable is a dichotomy (voted or not), and regression analysis can yield biased estimates with such a dependent variable, especially if (as in this case) the proportion voting sometimes approaches 1.0. As was done in Chapter 3, the data are weighted to reproduce the actual historic turnout levels at each election. But in this chapter we go further than we have done before to assure the comparability between variables. All variables have been rescaled (if necessary) to give them a minimum value of 0 and a maximum value of 1. Except for the character of election variables (for which we want to retain comparability with past chapters) the variables are also recoded so that a value of 1 corresponds with an influence that would tend to make turnout higher on the basis of conventional wisdom.

156


table 6.2 Differences in Proportions Voting that Result from Shifts from Minimum to Maximum in Independent Variables When Other Such Variables Are Held at Their Mean Values D Model A Constant Uninteracted variables Age, proportion of 99 (0 = 18 to 1 = 99) Education (0–1 = higher) Urban (0–1 = most urban) Respondent’s occupation (0–1 = white collar) Household occupation (0–1 = white collar) Religion (0,1 = Christian) Church attendence (0–1 = weekly) Political interest (0–1 = high) Party sympathy (0–1 = strong) Party ID (0–1 = strong)

Model B

−0.096∗∗∗

0.065∗∗∗ 0.037∗∗∗ 0.025∗∗∗ 0.111∗∗∗

0.052∗∗∗ 0.030∗∗∗ 0.024∗∗∗ 0.106∗∗∗

−0.056∗ 0.037∗∗∗ 0.027∗∗∗ 0.092∗∗∗

0.081∗∗∗ 0.176∗∗∗ 0.043∗∗∗ 0.156∗∗∗ 0.697∗∗∗ 0.035∗∗∗

0.063∗∗∗ 0.133∗∗∗ 0.039∗∗∗ 0.140∗∗∗ 0.650∗∗∗ 0.037∗∗∗

Variables interacted with new cohorts New ∗ gender (0,1 = male) New ∗ education (0 –1 = Univ) New ∗ urban (0–1 = most urban) New ∗ respondent’s occupation New ∗ household occup (0–1 = white collar) New ∗ religion (0,1 = Christian) New ∗ church attendence (0–1 = weekly) New ∗ party id strength (0–1 = strong)

−0.116

0.042∗∗∗ 0.030 0.032∗∗∗ 0.088>∗∗∗ 0.567∗∗∗ 0.038∗∗∗

−0.059∗∗∗ 0.024∗∗∗ −0.044∗∗ −0.078∗∗ 0.046∗

−0.041∗∗∗ 0.038∗∗∗ −0.044∗∗ −0.006 0.065∗∗∗

0.032∗∗∗ −0.014∗∗ 0.007

0.103∗∗∗ −0.012∗∗ 0.014∗

Effects of character of elections Majority status (0 = 4 to 21 = 1) New ∗ majority status (0 = 4 to 21 = 1) New ∗ margin (0–1 = 13) Young initiation (0,1 = yes) Voted at previous election (0,1 = yes) Nagelkerke R2 N

Model C

∗∗∗

−0.122

∗∗∗

−0.045∗∗∗ −0.232∗∗∗ −0.502∗∗∗ −0.054∗∗∗ 0.139∗∗∗ 0.205 8410

0.216 8410

0.277 8410

Note: Significant at ∗ .05, ∗∗ .01, ∗∗∗ .001 levels (one-tailed).

Because the coefficients that are produced by a logistic regression analysis are uninterpretable by mere mortals, they have been converted in this table to differences in proportions. Each effect (D) is the difference in the value of the dependent variable resulting from a change of one


157

unit in the corresponding independent variable when all other independent variables are held at their mean values.4 So the coefficients are interpreted in roughly the same way as b coefficients in regression analysis. Because each independent variable has been modified where necessary so that its values range from 0 to 1, the coefficients for that variable can be interpreted as the proportion change in the likelihood of voting that would result from a shift in the value of that variable from its minimum to its maximum value. Some of the implied shifts are physically or logically impossible (thus members of new cohorts cannot logically have ages that range from eighteen to ninety-nine). This is why the constant term is negative. In practice, there are no cases with a negative likelihood of voting because there are no cases for which every independent variable takes on its minimum possible value. The table contains three models. The first includes only variables of the type that have been common in past research at the individual level: variables relating to resources (age, education, and occupation), mobilization (urban, religion, and church attendence), and campaign effects (political interest, party identification, and party sympathy). Variables listed in Table 6.1 as being available in comparable terms across the German surveys are included in Table 6.2 if they were found to have significant effects in preliminary versions of any of the three models.5 Model A explains over 20 percent of the variance in turnout, a respectable figure for individual-level investigations into why people vote. The strongest effect is that of party sympathy. Only a very small number of people gave low sympathy scores to all the German parties, but those people were very unlikely to vote. By contrast, individuals who have strong feelings of sympathy for at least one party were 65 percent more likely to vote (according to Model A). Another fairly strong effect comes from religion. This was recoded to contrast those who identified themselves as Protestant or Catholic with other individuals, and the strong effect of Christian identification probably relates to the fact that such individuals are part of the mainstream of 4

5

The logistic regression coefficients from which these differences were calculated, along with their standard errors, are presented in Appendix C, together with the procedure employed to translate logistic regression coefficients into their more interpretable counterparts. The coefficients presented in Table 6.2 derive, however, from analyses that contain only the variables shown there.

158


German society. Model B adds interactions with membership in new cohorts, defined as in past chapters to include respondents facing their first, second, or third elections since reaching voting age. These are the same variables as included in Model A but operationalized so as to show whether new cohorts react differently than all cohorts in terms of these variables. Model C adds variables having to do with the character of elections that prove significant with German data. The main differences in findings between Models A and B lie in the fact that new ∗ age, new ∗ interest, and new ∗ sympathy do not appear in Model B despite the presence of age, interest, and sympathy in Model A, as well as in the presence of new ∗ gender in Model B, even though gender did not prove significant in Model A. The case of gender is particularly interesting. Among established cohorts of Germans, men vote at a slightly higher rate than women (though the difference is not significant). Among new cohorts, by contrast, women vote at a significantly higher rate than men, a development that has been remarked upon worldwide (Norris 2001). The more interesting omissions from Model B are interactions with political interest and party sympathy. Newer cohorts appear to be no different from established cohorts when it comes to the effects of these variables. Interpreting the interactions themselves is more difficult.6 In order to make it easier to understand their import, the analyses reported in Table 6.2 are repeated in Table 6.3 with all variables that appear in interacted form being made components of one interaction or another. Uninteracted variables corresponding to interactions with new cohorts in Table 6.2 are thus additionally interacted with established cohorts in Table 6.3. The two tables really need to be viewed in conjunction. Table 6.2 tells us whether members of newer cohorts are significantly different from members of all cohorts in terms of the effects of specific variables on turnout. Table 6.3 tells us what are the effects of different variables on established as opposed to newer cohorts.7 6

7

Estimating effects for specific groups is complicated by the fact that the table consists of estimates made when other variables stand at their mean values. Those means are different for new cohorts than for all cohorts or established cohorts. So the coefficients cannot be treated as OLS regression coefficients in this respect. Established cohorts are defined as all those who are not new, so the two categories of respondents are complementary and between them exhaust the pool of those for whom we have turnout information.


159

table 6.3 Differences in Proportions Voting that Result from Shifts from Minimum to Maximum in Independent Variables with Other Variables Held at Their Mean Values, Separated Interactions D Model D

Model E

∗∗∗

−0.096∗∗∗

Effects on established voters Established ∗ education (0–1 = university) Established ∗ urban (0–1 = most urban) Established ∗ respondent’s occupation (0–1 = white collar) Established ∗ household occupation (0–1 = white collar) Established ∗ religion (0,1 = Christian) Established ∗ church attendence (0–1 = weekly) Established ∗ party ID (0–1 = strong)

0.032∗∗∗ 0.024∗∗∗ 0.105∗∗∗ 0.062∗∗ 0.131 0.040∗∗∗ 0.038∗∗∗

0.038∗∗∗ 0.025∗∗∗ 0.081∗∗∗ 0.038∗∗ 0.026 0.031∗∗∗ 0.038∗∗∗

Effects on new and established voters Age, proportion of 99 (0 = 18 to 1 = 99) Political interest (0–1 = high) Party sympathy (0–1 = strong)

0.052∗ 0.140∗∗∗ 0.651∗∗∗

−0.056∗ 0.087∗∗∗ 0.570∗∗∗

−0.058∗∗∗ 0.042∗∗∗ −0.014∗ 0.033 0.073∗∗∗ 0.074∗∗∗ −0.016∗∗ 0.008∗

−0.041∗∗∗ 0.059∗∗∗ −0.012 0.053∗∗∗ 0.081∗∗∗ 0.108∗∗∗ −0.014∗∗∗ 0.014∗

Constant

Effects on new voters New ∗ gender New ∗ education New ∗ urban New ∗ respondent’s occupation New ∗ household occupation New ∗ religion New ∗ church attendence New ∗ party id strength

−0.115

Effects of character of elections Established ∗ majority status (0–1 = 21) New ∗ majority status (0–1 = 21) New ∗ margin (0–1 = 13) Young initiation (0,1) Previous election voter (0,1) Nagelkerke R2 N

−0.050∗∗∗ −0.379∗∗∗ −0.505∗∗∗ −0.054∗∗∗ 0.140∗∗∗ 0.216 8410

0.278 8410

Note: Significant at ∗ .05, ∗∗ .01, ∗∗∗ .001 levels (one-tailed).

In Table 6.2 we saw that education, household occupation, and religion have significantly more effect among members of newer than established cohorts, while urban residence, respondent’s occupation, and church attendance have significantly less effect among members of newer cohorts. Indeed, Table 6.3 tells us that these last two variables

160


change their signs for members of newer cohorts, having negative instead of positive effects on turnout. The reverse is true of the effect of religion, which proves to be positive only among newer cohorts. Established cohorts are seen in Table 6.3 to be unaffected by religion. These quite different effects of particular variables on newer than on established cohorts highlight the deficiencies of conventional wisdom regarding individual-level effects on turnout, based on models that do not take account of the role of new cohorts. Indeed, our findings in Table 6.3 would be impossible to interpret other than in the context of a theory of turnout based on the socialization process, whereby newer cohorts with low turnout become established cohorts with higher turnout in the manner described by Plutzer (2002). What we see is that certain factors conventionally associated with turnout (respondent’s own occupation, urban residence, and church attendance) do not become associated with higher turnout until voters have become established. Most likely these are factors that play a part in the way in which new cohorts learn the habit of voting. Indeed, until new cohorts become established, urban residence and church attendance actually appear to discourage turnout for reasons that we will consider later. It is noteworthy that measures of political interest and party sympathy are just as important for new as for established cohorts. This signals the possibility that the character of elections could have equal effects on both groups if the effects of an election with different character were to be felt indirectly via these campaign variables. But when we introduce measures of the character of elections, in Models C and E, we find that this is not the case. Not only does variance explained go up considerably in these models as compared with the other models, but variables measuring the character of elections are seen to have effects of their own – and very strong ones – primarily on new cohorts. We will return to this point later in the chapter. Other variables appearing in Models B and D are seen to have sometimes very different effects in Models C and E. Most importantly, the effects of age are reversed in Models C and E, suggesting that the positive effects of age in conventional models of electoral participation are standing in for effects that are specifically measured in Model E – notably the effect of inertia among established voters, which is measured by the final variable in Tables 6.2 and 6.3. Older cohorts vote at higher rates because they have learned the habit of voting. When the


161

habit of voting is explicitly included in the model, we see that there is actually some tendency for turnout levels to decay with age, consistent with a nonlinear age effect found in many studies (e.g., Rosenstone and Hansen 1993). What does Model E (and the parallel Model C) indicate about the ways in which the character of elections have their impact? Some of that impact is clearly indirect. Political interest and party sympathy both have notably less effect in Models C and E than in Models A, B, or D, indicating that at least some of the predictive capacity of these variables in conventional models comes from their ability to pick up on the importance or otherwise of elections.8 There is a multiple correlation of 0.425 between the character of elections and political interest, and of 0.212 between the character of elections and party sympathy. We will consider indirect effects on turnout in more detail in the next section. At the same time it is clear that most of the impact of the character of elections is not mediated by any of the individual-level variables included in Table 6.3. Rather, differences in the character of elections appear to influence citizens directly. The size of the largest party (its majority status) is quite important in Germany, contributing 5 percent to the turnout of established cohorts and fully 37.9 percent to the turnout of new cohorts, if elections in which the largest party receives 50 percent of the vote are compared with elections where the largest party receives a 21 percent smaller share of the vote. And the margin of victory is worth a whopping 50 percent to the turnout of new cohorts, if elections in which the leading parties are neck and neck are compared with elections in which there is a 13 percent gap between the main contenders (margin of victory has no significant effect on established cohorts). These variables rank with party sympathy in the importance of their effects on turnout, an order of magnitude greater than the effects of any of the resource or mobilization variables. Substantively, what this means is that citizens – especially those facing their first three elections – take note of the competitiveness of an

8

In conventional models that are estimated for only a single election, the strength of the effects of these variables (and also of party ID) will depend on the salience of the election – one reason why past studies have found inconsistent effects of these variables.

162


election, consistent with the findings of previous chapters. An election in which respondents’ votes could result in a new government taking office will motivate more of them to participate than an election that is unlikely to see a shift in party control. Comparing Effects of Elections’ Character with Effects of Individual Characteristics Though the effects of particular variables measuring the character of elections are very much stronger than the effects of particular resource or mobilization variables, still there are many more such variables than there are character of election variables. Any overall accounting of the extent to which individual characteristics are important, relative to the character of elections, needs to add up the effects of different variables to reach a summation for variables in each group. Such accounting must also consider the campaign variables included in our models and take account of any indirect effects that these variables may transmit, as suggested earlier. The status of the campaign variables is important. These are variables included in most models that attempt to explain why people vote, but they are not obviously resource or mobilization variables. Verba, Schlozman, and Brady (1995:436) argue that political interest is largely a function of education, making it a surrogate resource variable. However, in our data the correlation between education and political interest is only 0.168, and all of the resource and mobilization variables taken together only produce a multiple correlation of 0.254 with political interest, in contrast to the multiple correlation of 0.425 reported earlier between the character of elections and political interest. Party sympathy, though less strongly linked to the character of elections (R = 0.212), is nevertheless more closely connected to that concept than it is to resources and mobilization (R = 0.164). It should come as no surprise that the character of elections has indirect effects by way of the campaign variables. It makes sense that, at the time of a more competitive election, interest in politics goes up and so does the extent to which people feel sympathy toward their most preferred party. These indirect effects are shown in Table 6.4, which also sums the effects of independent variables so as to indicate the total effect of shifting all the independent variables in each group simultaneously from their


163

table 6.4 Effects on Campaign Variables of Other Independent Variables in Tables 6.2 and 6.3 from OLS Regression Analysis (standard errors in parentheses) B Effects on party sympathy Effects of resources Age (0 = 18–1 = 99) Education (0–1 = univ) Respondent’s occupation (0–1 = white collar) Household occupation (0–1 = white collar) Total effects of resource variables Effects of mobilization Urban (0–1 = most urban) Religion (0, 1 = Christian) Church attendence (0–1 = weekly) Total effects of mobilization variables

Effects on political interest

Effects on party ID

0.031 (0.006)∗∗∗ −0.013 (0.009) 0.075 (0.022)∗∗∗ ∗∗∗ ∗∗∗ −0.033 (0.003) 0.129 (0.005) 0.151 (0.012)∗∗∗ 0.004 (0.005) 0.017 (0.007)∗∗ 0.108 (0.016)∗∗∗ 0.000 (0.006)

−0.037 (0.009)∗∗∗ 0.031 (0.022)

0.002

0.098

0.365

0.014 (0.003)∗∗∗ 0.013 (0.004) 0.042 (0.011)∗∗∗ 0.044 (0.009)∗∗∗ 0.069 (0.012)∗∗∗ 0.243 (0.032)∗∗∗ 0.018 (0.004)∗∗∗ −0.026 (0.005)∗∗∗ 0.015 (0.014) 0.076

0.056

Effects of character of elections Majority status −0.118 (0.006)∗∗∗ −0.322 (0.009)∗∗∗ Margin of victory −0.102 (0.006)∗∗∗ −0.040 (0.009)∗∗∗ Total effects of −0.220 −0.362 character variables r2 0.057 0.212 N 8410 8410

0.290

0.150 (0.022)∗∗∗ 0.350 (0.021)∗∗∗ 0.500 0.058 8410

Note: Significant at ∗ .05; ∗∗ .01; ∗∗∗ .001 levels (one-tailed).

minimum to maximum values.9 Note that majority status and margin of victory have an inverse relationship with turnout. As races get tighter and the largest party gets closer to receiving 50 percent of the vote, reduced values of these variables go along with rising turnout. So the role of political interest and party sympathy as intervening variables 9

Indirect effects for all variables could be computed concurrently with their direct effects in an analysis using Lisrel or EQS or some other simultaneous equation modeling package, but we prefer to employ a more informal method that builds on the findings already presented in Tables 6.2 and 6.3, rather than having to introduce a whole new method of analysis.

164


is indicated by the strength of the negative effects of the character of elections on them. Critically, however, effects of the character of elections on party ID are positive and very strong. This means that the effects of elections’ character on turnout is to some extent suppressed by means of a mechanism involving party ID. When an election’s character is such as to generate low turnout, the effect is somewhat mitigated by the power of party ID to generate higher turnout. In elections whose character would tend to generate high turnout, party ID does not contribute to that turnout.10 So in individual-level studies that take no account of the character of elections, the function of party ID is misunderstood. It is generally seen as one of the factors that generates high turnout, but this is not so. It is a factor that prevents turnout from falling as far as it might otherwise fall in a low-turnout election, but it most assuredly does not help to boost turnout in a high-turnout election.11 This is the most important finding to derive from our use of measures of electoral competitiveness in conjunction with variables derived from survey data. Indeed, it reinforces one of the primary arguments of this book, which is that people will vote in important elections regardless of the presence or absence of other factors that are sometimes seen as enhancing turnout. Party ID also joins “voted at previous election” as a manifestation at the individual level of the force of inertia that we have found to be so important in aggregate-level analyses, with considerably stronger effects among established than new cohorts (see Table 6.3). The other important implication of this table is that effects on campaign variables of resources and mobilization, even taken together, are much less than the effects of electoral competition. Even the strongest 10

11

The effect on turnout of moving from weakest to strongest party ID is not very great (0.014 in Table 6.2), so this indirect effect does not mitigate by much the effects of the character of elections. The effect is not as perverse as it might seem at first glance. What we see here is one of the mechanisms by which intertia in voter turnout is maintained. Party ID invokes the loyalty that causes people to vote even at elections where there is no other good reason for voting. The lower the turnout would otherwise have been, the greater the apparent effects of party ID. In elections where there are good reasons for voting, by contrast, it is understandable that party ID plays no role. In high-turnout elections people vote regardless of the strength of their attachment to parties. This insight can readily be checked through reanalysis of existing election studies. In studies of uncompetitive elections we should find a stronger effect of party ID than in studies of competitive elections, other things being equal.


165

table 6.5 Direct, Indirect (via Campaign Variables), and Total Effects of Different Categories of Variables for New Cohorts

Resourcesa Mobilizationb Campaignc Character of electionsd a b c d

Direct effects

Indirect effects

Total/net effects

0.152 0.082 0.570 −0.884

0.015 0.052 −0.217 −0.150

0.167 0.134 0.353 −1.034

Age, education, occupation, and household occupation. Urban residence, religion, and church attendance. Political interest, party sympathy, and party ID. Majority status and margin of victory.

effects of resources and mobilization – their effects on party ID – are only a quarter of the effects of the character of elections on party ID. Campaign variables evidently reflect the character of elections far more strongly than they do the characteristics of individuals. However, to complete the picture we need to account for the direct effects of each of these sets of variables on turnout, together with their indirect effects via the campaign variables. Table 6.5 sums the direct and indirect effects of variables in each category. This balance sheet operation takes a portion of the effects that regression analysis attributes to intervening variables and reassigns that portion to the variables really responsible for the effects concerned.12 Direct effects are taken from Table 6.3, Model E. Indirect effects are computed by multiplying the effects shown in Table 6.4 with corresponding effects in Model E and summing the results across categories of variables as indicated in the footnotes to Table 6.5. The resulting indirect effects are taken into account in two places in this table. First, each indirect effect is subtracted from the effects that they were contributing to (campaign effects were being boosted, so 0.217 is subtracted). Second, the indirect effects are summed and added to the total effects of the categories where they really belong. The outcome is seen in the columns headed “Total/net effects” (net effects for campaign variables, total effects for other categories of variables).

12

The indirect effects that flow via party ID are positive, as already mentioned, which reduces the total effect of the character of elections.

166


The table shows clearly how campaign effects had been inflated by the indirect effects they were transmitting while the effects of other categories of variables had been understated when indirect effects were ignored (only a little in the case of resources and mobilization variables but rather more in the case of character of elections variables). Indeed, indirect effects of the character of elections, when added to their already large direct effects, yield a spectacular total effect of 1.03. Were all the character of elections variables to move from their minimum to their maximum values simultaneously, the result would be to shift the chances that any individual respondent would vote from certainty to no chance at all – more than the combined effect of all other variables moving in concert (which would shift the probability of voting by 0.65, the sum of the net effects of all other groups of variables). Evidently this does not happen in practice because the sum of the real-world effects cannot exceed 1.0. This reassignment of effects makes it clear that turnout even at the individual level is dominated by the character of elections, with campaign effects playing a decidedly secondary role. Resource and mobilization effects, by contrast, play a relatively minor role, with less than a third of the effects that the character of elections have on turnout.13 Moreover, there are strong hints in our findings that the role of resources has been misconstrued in past research. The suppressive effects of party ID, and the negative effects of urban dwelling and church attendance among new voters (in contrast to their positive effects among established voters), suggest that at least some of the social structures that have been presumed in past research to uphold voter turnout (Rosenstone and Hansen 1993; Verba, Schlozman, and Brady 1995) may really work only to mitigate the effects of low voter turnout. Above all, these factors may not so much help to inculcate the habit of voting as, instead, to maintain a low level of turnout in elections where there is no other good reason for voting. This would explain the negative effects among those who are paying attention to the importance of the election, according to the logic outlined previously. 13

Moreover, this accounting is, if anything, generous to the resource total, since it counts the suppressive effects of party ID against the character of elections variables (taking the negative indirect effects via party ID away from the total effects of the character of elections). An argument could be made that these negative effects should rather be taken away from the resource variables, or at least ignored, since party ID does not in practice benefit turnout.


167

These hints, if they are substantiated in future research, suggest that it is a pity that so much effort was expended on studying individuallevel determinants of turnout in a country where those determinants were doing little more than supporting a habit of voting in less competitive elections that had been acquired in earlier, more competitive, elections (see Figure 4.2). More importantly, those scholars who express concern about declining social capital (Dalton 1999) and partisanship (Wattenberg 2000, 2002) may be addressing the wrong concerns. The extensiveness of social networks and sizes of party organizations may be important more in terms of cushioning a decline in turnout than in supporting high turnout.14 These suggestions reinforce the importance of uncovering the real reasons for high turnout (as we tried to do in Chapter 5) and of determining which features of the character of elections are mainly responsible for turnout decline – a primary topic for Chapter 7.

Discussion Turnout is an aggregate-level concept, and we focused on the aggregate level in Chapter 5 when we looked at turnout variations in twenty-two countries across fifty-five years. Still, the basis for understanding cumulative and short-term effects on turnout requires an understanding of cohort-level forces, and we focused on the cohort level in Chapter 3 when we employed a dataset constructed from survey data collected in six countries since the 1960s. The two analyses were not designed with the purpose of confirming each other’s findings, but rather Chapter 3 provided the building blocks on which Chapter 5 could be constructed. In the present chapter we have completed an inventory of levels at which turnout can be investigated by focusing on the individual level. Again the purpose was not confirmatory; rather, it was to illustrate some of the implications of this book’s findings for individual-level voting behavior. Although none of these chapters were written with the intention of confirming the findings of other chapters, in practice the findings reported in this chapter are remarkably consistent with those reported 14

It will be recalled that the model set out in Chapter 2, while it required a certain level of social networking as a prerequisite, does not suggest that the extensiveness of social networks is relevant to turnout. We will return to this topic in Chapter 8.

168


in Chapters 3 and 5. Each of these chapters rediscovers the effects of certain features of the character of elections that were measured in all three datasets. All three find strong negative effects on turnout of margin of victory and majority status (how far the largest party is from gaining 50 percent of the votes), which is especially strong for cohorts not yet set in their ways. All three find that distinguishing between established and new cohorts considerably enhances explanations of turnout change.15 Our theory suggests that the primary effect on established cohorts comes from their early voting experiences, when they were especially open to influence from the character of elections. However, such early experiences only have a lasting impact if changes in the character of elections that gave rise to those experiences remain in place. So the distinction between new and established cohorts made in Chapters 3 and 6 corresponds with a distinction between short-term and cumulative effects on turnout made in Chapter 5. Insofar as the three datasets employed in this study allow comparable analyses to be performed, they show the same influences on turnout that primarily affect new cohorts having primarily short-term influences and the same influences on turnout that become established with the passage of time having primarily cumulative influences. We must be cautious about comparing the coefficients estimated in this chapter with coefficients estimated elsewhere in this book. In this chapter, variables were recoded where necessary so as to ensure that all would vary along an identical 0–1 scale that makes the effects of different variables comparable. Only variables that did not need to be recoded have effects that are immediately comparable with those estimated in previous chapters. One of the few such variables is young initiation, which is found to reduce the probability of voting by 0.054 in Tables 6.2 and 6.4 – which is rather more than the 3.8 percent found in Model E, Table 5.2, or than the 3.4 percent found in Model C, Table 3.4, but still quite close.16 15

16

Not so much in terms of variance explained as in terms of a more nuanced understanding of how different groups within the electorate contribute to turnout change, which in turn gives us an enhanced ability to understand the dynamics of long-term turnout evolution. German cohorts showed greater than average turnout differences in Chapter 3, so it makes sense that the effects of young initiation would be greater in Germany than in the average country.


169

In this chapter we have used a multiple-election case study to test the proposition that variations in the character of elections have effects on individual-level electoral participation. For reasons of convenience we used Germany as our case study. The findings, however, are very much as were to be expected on the basis of theory and aggregate-level tests. There is no reason why substantively different findings should come from studying a sequence of elections in any other country (but see Chapter 8 for some caveats on this point). Except for the desire to replicate findings regarding the lowering of the voting age, there would be no need to use election studies going back as far as those in Germany, so data from a great many countries could be used to replicate and check these findings. Our ability to replicate at the individual level findings regarding young initiation already made at the level of the country and electoral cohort is particularly important for the story unfolding in this book. The role of young initiation in the logic of this study is more than that of just another variable bearing on the character of elections. Because it results from an institutional change in virtually every country, this variable provides us with our principal test of the theory that socialization and generational replacement play primary roles in causing turnout variation. The fact that we get essentially the same story in this regard from three very different datasets is satisfying and will be revisited in more detail in Chapter 8. The fact that this critical finding has been closely replicated at three different levels of analysis gives us the confidence to build a whole new layer of reasoning on coefficients calculated in Chapter 5 from country-level analyses of this and other variables that employed the same theory of socialization and generational replacement in their operationalization. We are now in a position to use those coefficients as a means for tallying up the sources of turnout change and using that tally to explain turnout decline among established democracies in the contemporary era. This is the purpose of the next chapter.

7 Understanding Turnout Decline

The findings of past chapters – and especially those of Chapter 5 – have provided us with the raw material we need to understand turnout change of all kinds, but in this chapter we will focus particularly on turnout decline. This is a topic that has much exercised commentators, as reported in Chapter 1. In Chapter 5 it was established that the causes of turnout change primarily have to do with the character of elections, not the character of society, so commentators who see in falling turnout a reflection on the civic-mindedness of citizens, or on their commitment to democracy, appear to be mistaken. Variables such as party attachment (Wattenberg 2000, 2002), trust in government (Dalton 1999), or union membership (Gray and Caul 2000) contribute nothing to the explanation of turnout developed in this book and tested in Chapter 5 (see Chapter 5, note 36). So why has turnout declined? Evidently, the preferred model established in Chapter 5 plays out differently in different countries.1 We would certainly expect that changes in the responsiveness of the executive play a primary role in explaining Swiss turnout decline, while the abolition of compulsory voting must play a primary role in Dutch turnout decline. Nevertheless, the overall picture can still be painted in terms of all countries taken together provided that we bear in mind that 1

In Appendix C we present graphs for each country showing the evolution of both expected and actual turnout, together with indications of salient electoral reforms.

171

172


84

83

82

81

80

79

Predicted turnout Actual turnout

78

77 1950-55 1955-59 1960-64 1965-69 1970-74 1975-79 1980-84

1985-89 1990-94

1995-99

figure 7.1. Predicted and actual average turnout in five-year periods for twenty-two countries, 1950s–1990s.

certain developments take place (certain variables change their values) in some countries but not in others. Our overall success in tracking variations in average turnout over time, on the basis of predictions derived from the preferred model (Model E, Table 5.2) established in Chapter 5 is shown in Figure 7.1. There we see that predicted and actual turnout move together very well over the period for which we have values for all our independent variables.2 Indeed, tracking appears to improve over time, which makes sense because several countries had seen a long period without elections prior to and during World War II, creating irregularities in the cohort structure of turnout in those countries that will only gradually have been eliminated as voters socialized during a period without elections left the electorates of the countries we study.3 Tallying the Sources of Turnout Decline What changes in the character of elections are primarily responsible for the pattern of predictions that we see in Figure 7.1? We can answer 2 3

Note that this figure provides a particularly stringent test of our ability to track turnout change, since the scale is so exaggerated (cf. Figure 5.1). This finding has implications for the study of turnout in countries that are not established democracies – implications that we will explore later.

Understanding Turnout Decline

173

this question by using our regression findings to estimate the extent of turnout change attributable to each variable. Table 7.1 is divided into three parts. The first part occupies a single column and provides the regression coefficients (b’s) estimated in Model E of Table 5.2. The second part occupies three columns and provides the mean values of independent variables employed in that equation for three points in time spanning the period that was investigated in Chapter 5. The first time point is 1950, or the first election after 1950. This is the earliest point at which we have nonmissing values for all variables, including those computed on the basis of information from the previous election. The second time point is 1965, or the nearest election to 1965. This was chosen as representing the high point to which turnout rose before starting its long decline (see Figure 7.1). The final time point is 1999, or the last election prior to 1999 (the last case for each country in the time-series dataset employed in Chapter 5). The third part of the table consists of a further three columns, each of which contains the result of multiplying changes in the observed mean values of a variable by the regression coefficient for the same variable given in the first column of the table. These are the changes in turnout predicted by changes in the values of independent variables contained in Model E. The first of these last three columns relates to change during the early period (1950–1965), the second relates to change during the later period (1965–1999), and the final column relates to change during the entire period 1950–1999. At the foot of the table are a number of summaries. The first row contains average levels of turnout observed in the twenty-two countries at each of the three points in time. The second row contains the changes in turnout between the three time points, first observed historically and then estimated by summing the predicted effects higher up the table. The final row contains the same information for the entire time period, showing first the total change in turnout that occurred between 1950 and 1999 and then the estimate we make of that total change (again obtained by summing the rows higher up the same column). Look first at the bottom row. There we see that the overall decline in turnout across the twenty-two countries is predicted to within 0.75 by the coefficients of Model E (Table 5.2 in Chapter 5), which occupy the first column of Table 7.1 and are used in all the ensuing calculations. These would have led us to expect a decline of 5.84 percent, compared to an actual decline of 6.59 percent – a disparity that matches the

174

Dependent variable Turnout Incremental change in turnout Overall change in turnout 1950–1999

Time since previous election Short-term majority status Short-term margin of victory Short-term mean margin Short-term cohesiveness Cumulative compulsory voting Cumulative executive responsiveness Cumulative female empowerment Cumulative electorate size Cumulative young initiation Turnoutt−1

Independent variables 0.636 −0.409 −0.353 −0.571 7.042 11.252 4.054 5.004 −0.044 −3.808 0.410

b

Model E coefficients

Y1950 82.5

3.273 1.170 1.820 0.497 0.123 −0.727 0.818 0.374 1.385 0.000 81.289

X1950

Y1965 84.2 +1.6

3.682 2.239 1.793 0.608 0.163 −0.727 0.803 0.455 5.964 0.000 83.239

X1965

Y1999 75.91 −8.19 −6.59

3.409 4.876 2.535 0.814 0.253 −0.775 0.724 0.495 28.214 0.572 78.279

X1999

Mean observed values

b(X99 − X50 ) 0.086 −1.514 −0.252 −0.181 0.915 −0.540 −0.381 0.608 −1.170 −2.178 −1.233

b(X99 − X65 ) −0.174 −1.078 −0.262 −0.118 0.634 −0.540 −0.320 0.200 −0.971 −2.178 −2.032

+1.044

−6.838

−5.841

b(X65 − X50 ) b(X99 − X65 ) b(X99 − X50 )

0.260 −0.437 0.010 −0.063 0.282 0.000 −0.061 0.405 −0.201 0.000 0.800

b(X65 − X50 )

Estimated consequence of change in independent variables

table 7.1 Values of Dependent and Independent Variables, with Actual and Estimated Turnout Changes


175

difference between the actual and expected turnout for the final time point in Figure 7.1. The model does even better in tracking the earlier rise in turnout. As seen in the penultimate row, the real rise in average turnout of 1.6 percent that occurred between 1950 and 1965 is predicted to the extent of 1.04 – a shortfall of only 0.56 percent. However, the consequence of failing to accurately explain the rise in turnout in the first half of the period inevitably enhances our failure to completely explain its fall during the latter half. The real decline of 8.19 from the high point in the 1960s to the low point at the end of our period is only predicted to the extent of 6.84 percent, an underestimate of some 1.4 percent.4 The failures of prediction are important because they provide clues as to ways in which the model might be improved – clues as to ways in which it is misspecified – and we will return to consider these clues later in this chapter. Yet, in overall perspective, the model performs quite well, both in terms of tracking the fall in turnout that occurred in our twenty-two countries during the last third of the twentieth century and also the rise in turnout that occurred during the earlier period following the Second World War. To what extent can these changes be explained by the changing character of elections? We can answer this question by looking at the estimated effects of changes in the independent variables, which are summarized in the right-hand three columns of Table 7.1. These show to what extent change in the values of each specific variable listed in the table contributes to the overall turnout change. The final column looks at the entire period 1950 to 1999. It does so by multiplying the b coefficient for each variable in the first column of the table by the total amount of change that occurred in the observed values of that variable – the b(X99 = X50 ) column. A large chunk of the total change (1.23 percent out of 5.84 percent) is attributed to the lagged version of the dependent 4

According to Figure 7.4, the peak in turnout was expected to occur five years later, on average, than happened historically. Had we used this high point in expectations as our benchmark for picking the middle election in Table 7.1, rather than the high point in actual turnout, we would have overestimated the decline from that high point rather than underestimating it. This underlines the difficulty of illustrating the operation of our model in terms of specific elections: those exemplary cases may involve elections that are unrepresentative in being particularly poorly predicted – outliers, in fact, from a set of predictions that generally perform better than those specific cases would give us to believe.

176


variable. The values of this variable were of course predicted by the values of other independent variables in the previous time period, so those other independent variables have indirect effects via past turnout and should each be boosted by about a fifth of their value in order to take account of this. Such corrections will be mentioned as we proceed. Taking direct and indirect effects together, the largest contribution to declining turnout was the cumulating effect of young initiation in countries that lowered their voting age. This is seen in the table to have made a contribution of 2.2 percent which, when indirect effects via turnoutt−1 are taken into account, rises to 2.6 percent – not far from the 3 percent mentioned in Chapter 5 but less because so many countries lowered the voting age rather later than was assumed there, as indicated on the graphs for each country displayed in Appendix C. The other big contributor to overall turnout decline was changes in majority status, shown in Table 7.1 to have been worth 1.5 percent, which, with indirect effects via turnoutt−1 , amounts to 1.8 percent. This contribution to declining turnout was due to the manifestation in the data employed here of Wattenberg’s (2000, 2002) fractionalizing party systems. As party systems break into more and smaller parties (a development that appears ubiquitous outside the United States) the share of the vote received by the largest party drops increasingly below 50 percent. The largest party finds it more and more difficult to get its policies enacted in unadulterated form, and voters see less and less obvious connection between their votes and government policy. A number of smaller effects (from margins of victory, compulsory voting, and executive responsiveness) took a further 1.7 percent from overall turnout once their indirect effects were taken into account, and the increasing sizes of electorates takes 1.4 percent, bringing the total negative consequences of changes in the character of elections to some 7.5 percent. Offsetting these negative consequences, three positive effects (time between elections, party cohesiveness, and female empowerment) sum to 1.7 percent when their indirect effects are taken into account, leaving the balance of 5.84, which is the total amount by which turnout was expected to fall on account of changes in the character of elections. Most of these effects are very much as would have been expected on the basis of our theorizing in Chapters 3 and 4. The big surprises are


177

in the positive effect of female empowerment, given that Switzerland admitted women to its electorate during the second part of our period (which should have depressed turnout in that country) and the rather significant positive effect of short-term party cohesion, which offset the decline in turnout that would otherwise have occurred by fully 1.1 percent. We will start with cohesiveness. Average party cohesiveness will have been lower at the end of the first half and the start of the second half of our period because that was when Belgium’s party system lost cohesiveness. But cohesiveness in general remains largely unchanged over the entire period, so the increasing values of the interaction we called short-term cohesiveness in Table 7.1 must be due, not to the cohesiveness component of the interaction, but to the changing sizes of impressionable cohorts. As the second half of the twentieth century wore on, the changing proportions of new cohorts – cohorts facing one of their first three elections – evidently matched up with turnout changes that distinguished between countries with and without cohesive parties, demonstrating that party cohesiveness matters to impressionable cohorts. We will return to this point later. Turning to female empowerment, though the much-publicized enfranchisement of women did happen in Switzerland during the second half of the period, doubtlessly lowering turnout there at least for a time, the start of the first half of the period also saw the enfranchisement of women in France, Belgium, Italy, Japan, and Malta. The depression of turnout due to these enfranchisements was in the process of wearing off as progressively larger proportions of each of these electorates came to be made up of women who grew up after female enfranchisement (gaining the benefits of what we refer to as “female empowerment”), and the positive effects in these countries of generational replacement quite overwhelmed the negative effect on Swiss turnout of giving women the vote there. What Caused the Rise in Turnout Up to 1965? The long-term effect of female empowerment joins time since the previous election as one of only two variables to have greater consequences during the (rather shorter) early period than during the later period, contributing to the reversal in the direction of change after 1965, as

178


seen by comparing the b(X65 −X50 ) column with the b(X99 −X65 ) column. Indeed, female empowerment and time since the previous election, together with the increasing influence of party cohesiveness (and the relatively large indirect effects of all three of these variables via past turnout) are the only factors responsible for turnout rise in the first part of the period (helped by generally small countervailing forces – zero in the case of absentee voting, compulsory voting, and young initiation). Elsewhere I have argued (Franklin 2002) that the peak in turnout during the mid-1960s might well have been due to a period of close-fought high-profile elections, but I appear to have been wrong. Neither majority status nor margin of victory contributed to higher turnout in this period, and the increasing effects of party cohesion were merely part of a longer-term trend. So the rise in turnout up to the mid-1960s appears not very interesting when decomposed in this fashion: The high point was a happenstance in the evolution of long-term developments.

What Caused the Additional Fall in Turnout During the Later Period? A question that turns out to be rather more interesting is why do we underestimate the drop in turnout that occurred between 1965 and 1999? In one sense, of course, we hardly underestimate this fall. Overall, we track turnout change to within less than 1 percent, based primarily on young initiation (following the lowering of the voting age in most countries) together with a decline in electoral competitiveness and an increase in the sizes of electorates.5 But this apparent success 5

Increases in electorate size, in interaction with the sizes of new cohorts, accounts for about 1.4 percent of the overall decline in turnout after 1965, taking indirect effects via past turnout into account. This is a big chunk of turnout change to be attributed to a variable whose operations are not well understood. What the effect suggests is that impressionable cohorts are increasingly affected by the size of their electorates, being increasingly less likely to vote as electorates become larger and the opportunity that members of these cohorts see of being able to affect the outcome of an election becomes less. The finding certainly makes sense in terms of the theory that underpins this book’s hypotheses, but the variable’s importance as the second or third most powerful influence is worrisome. The variable appears to be essential to properly specifying the forces at work in determining turnout change. It adds little to variance explained, but it produces a model that is robust to changes in analysis strategies. Without it, the findings of the different models presented in Tables 5.1 and 5.2 differ quite markedly from one another, and in ways that appear arbitrary. With


179

is partly due to the fact that we do not accurately track the rise in turnout in the early part of the period. If we correct our estimate of average turnout in 1965, then we have a 1.4 percent discrepancy (8.19 − 6.84 in the penultimate row of Table 7.1) in our ability to track turnout decline thereafter. The disparity is well within the margin of error associated with our estimates of turnout, so in an important sense it can be regarded as zero. Nevertheless, there is heuristic value in speculating as to what might account for the additional decline in turnout over and above a crass projection of what could have been expected on the basis of Model E. We have already ruled out all published suggestions that have been tested in past empirical research. But there are many speculations that have never been tested empirically. One is the idea that globalization lowers turnout. This speculation, if it could be confirmed, would fit right into the picture we have been painting of voters responding to the character of elections in terms of the extent to which they can see their vote as possibly having policy implications. Globalization would operate much like declining executive responsiveness, reducing the power of governments to make policies independently of international obligations and the realities of the international market place.6 A more globalized world in which governments surrender powers to bodies such as the World Trade Organization is also a world in which elections are less meaningful as vehicles for achieving or blocking policy change. On this reasoning, the globalization of the world economy should go along with a lessening in the importance of elections, with consequently lower turnout. The fact that there is an unpredicted downturn in turnout in so many countries at the end of the 1990s fits this conjecture, because the downturn coincides with the point at which the concept of globalization first gained public prominence. The very fact that globalization became salient so late in our series makes it hard to test the conjecture. Nevertheless, the idea that ceding

6

the inclusion of this variable our models are well behaved. Though electorates have been growing with the passage of time, note that the variable is not simply a surrogate for the passage of time. A simple measure of elapsed time adds nothing to our preferred model, whether or not cumulative electorate size is included. Indeed, it is the exact counterpart to Malta’s acquisition of independence from British sovereignty (see Chapter 4), to which we have attributed Malta’s increasing turnout after that time.

180


power to a higher authority affects turnout can be tested in a slightly different way. A higher authority that has been in existence through most of the period under study is the European Union (EU). This is an authority to which different sets of countries have ceded sovereignty at different times, some nearly a half-century ago and some quite recently. This makes for a natural experiment if, as would accord with the findings of this book, a change such as this would first affect new cohorts. On this basis, the longest-established EU countries should show most evidence in their electorates of the awareness of what we will call “Europeanization,” and the most recent adherents should show least evidence, because longstanding members have had more time in which generational replacement can have affected their turnout levels due to “Europeanization.” To test this idea, a long-term Europeanization variable was created from the proportion of new voters in the electorate of each member country at the first election after accession, together with increments to this proportion in succeeding elections (nonmembers and future members being coded 0), and this variable was correlated with the residual from the analysis used to estimate Model E (Table 5.2). The correlation was –0.16, which is small but significant at the 0.01 level and suggests that the ceding of authority to a higher level of government does have a small negative effect on turnout. This test of course does not confirm an effect from globalization, but it does make it plausible to propose that the missing 1.4 percent decline in turnout since the 1960s could be explained, at least in part, by globalization forces. An Inventory of Turnout Change in Twenty-Two Countries The overall decline in turnout averaging 6.6 percent for all of the twenty-two countries whose recent histories we have investigated in this book masks considerable variations in practice, as we saw in Chapter 1’s Table 1.1, with Malta having seen a notable rise in turnout; Australia, Belgium, Denmark, Israel, Norway, and Sweden having seen slight rises; and other countries having seen declines of various magnitudes. How far does our preferred model go toward explaining turnout change in individual countries? To establish this, the predicted value of turnout from the analysis in Model E was used as the independent variable in a prediction of actual turnout, country by country. The


181

table 7.2 Tracking of Turnout by Country Using Predictions from Chapter 5’s Model E

Country

Standard deviation

SEEa

rb

Switzerland Malta Italy United States Netherlands Finland Ireland Luxembourg Austria Germany Iceland France Sweden New Zealand Japan Canada Denmark Norway Belgium United Kingdom Australia Israel Average

11.050 8.719 3.345 4.941 7.545 5.779 3.235 1.992 3.907 4.450 1.988 5.919 4.742 4.059 4.591 4.199 2.520 3.028 1.712 3.602 1.842 3.135 4.377

0.924 2.058 1.078 2.264 3.847 3.271 1.987 1.263 2.534 2.747 1.323 4.399 3.596 2.790 3.856 3.866 2.321 2.656 1.536 3.429 1.925 2.752 2.565

0.996∗∗∗ 0.970∗∗∗ 0.954∗∗∗ 0.897∗∗∗ 0.875∗∗∗ 0.850∗∗∗ 0.820∗∗∗ 0.806∗∗ 0.793∗∗∗ 0.779∗∗∗ 0.763∗∗∗ 0.714∗∗ 0.694∗∗ 0.622∗∗ 0.588∗∗ 0.503∗∗ 0.481∗ 0.438∗ 0.422∗ 0.408 0.087 0.022 0.658

Note: Significant at ∗ 0.05, ∗∗ 0.01, ∗∗∗ 0.001 (one-tailed). a Standard error of estimate. b Pearson’s product moment correlation.

results are shown in Table 7.2.7 There we see that the average correlation between predicted and actual turnout, country by country, is an impressive 0.66, and the average of the standard error of estimate is 2.56. This latter figure implies that, on average in each country, predicted turnout comes to within 2.56 percent of the actual turnout two-thirds of the time. Doubling the standard error of estimate, we discover that predicted turnout in the average country comes to within 5.1 percent of the actual turnout 95 percent of the time. 7

They are also spelled out in thumbnail graphs, one for each country, presented in Appendix C.

182


The table also includes the standard deviation for turnout in each country (taken from Table 1.1) – a measure of the extent to which turnout varies from election to election, and thus an indication of how much change there is to be explained. Interestingly, the countries in which we best predict turnout variations (those with the highest correlations, at the top of Table 7.2) are also those in which there is the greatest amount of change in turnout needing to be explained (those with the largest standard deviations). More importantly, the countries in which our model is least successful at explaining turnout change (those where the correlation is not significant at the 0.01 level) are all countries with relatively slight turnout variation (they all have standard deviations less than 4). Pleasingly, our model explains turnout change where turnout change is greatest and fails to explain turnout change where that change is least problematic. The correlation between the standard deviation and correlation columns of Table 7.2 is 0.57. As emphasized by the graphs in Appendix C, countries where turnout change is poorly explained are all countries where turnout hardly changes its mean level. For this reason it is not surprising that the countries in which the model predicts poorly are generally countries in which expected turnout is quite close to actual turnout (standard errors of estimate are generally better than average among countries at the bottom of Table 7.2). The major exception is the United Kingdom, which shows a moderately high standard error of estimate together with a correlation that does not quite reach significance at the 0.05 level.8 Interestingly, our ability to track the major changes in turnout that occurred between the 1960s and the 1990s in each country is hardly related to the general performance of the model in each country summarized in Table 7.2. The effects for each country of changes in independent variables between the mid-1960s and the end of the 1990s (the equivalent for each country of the penultimate column of Table 7.1) 8

Inspecting the thumbnail graphs in Appendix C yields the insight that expected turnout can be quite close to actual turnout without there being a significant correlation, supposing that there are no large shifts in turnout to be tracked. If we include in the British series the election of 2001, which saw a large drop in turnout, this gives us a large shift to be tracked and our model responds well to the challenge. The correlation between expected and actual turnout in Britain rises to 0.616, significant at the 0.01 level.


183

are given in Appendix C, Table C.4. These yield the best predictions for Austria, Belgium, Canada, Denmark, Italy, Malta, New Zealand, the United Kingdom, and the United States (all countries where the changes in turnout over this thirty-year period are tracked to within 1.3 percent on average) while the worst predictions are made for Finland, France, Germany, and Ireland (all countries where the average disparity between actual turnout change and expected turnout change is greater than 4 percent). In Britain, the change in turnout from the 1960s to the 1990s is tracked to within 0.8 percent, which is far better than the average of 3 percent over all twenty-two countries. These findings emphasize the fact that the model can explain salient turnout changes even in countries where it appears to perform poorly in general terms. The reverse is also true. Even in countries where the model does really well in general terms, it can still do a bad job of explaining particular pairs of elections if each election produced an outlier (as happened in the case of Finland – see Table C.4 in Appendix C). The main utility of recalculating Table 7.1 for individual countries, however, lies in the help that such a table can give us in understanding the ubiquity of particular influences. Table 7.3 summarizes the information in Appendix C by telling us, for each variable, how many countries saw their turnout increased or reduced by changes in that table 7.3 Number of Countries for Which Turnout Moved Up or Down Between the 1960s and 1990s due to Changes in Particular Independent Variables Effect in country moved turnout

Time since previous election Short-term majority status Short-term margin of victory Short-term mean margin Short-term party cohesiveness Cumulative compulsory voting Cumulative executive responsiveness Cumulative female empowerment Cumulative electorate size Cumulative proportion of young initiation Average per country

Unchanged

Lower

Higher

Average change %

11 1 1 18 4 20 20 16 3 2

7 19 14 3 2 2 1 1 19 20

4 2 7 1 16 0 1 5 0 0

1.180 1.188 0.833 1.057 1.082 6.376 6.254 0.978 1.056 2.273

4.3

4.2

1.4

184


variable between the 1960s and the 1990s, and what was the average change in turnout (up or down) induced by that variable in those countries where the variable had any effect. On average, each country had 4.3 effects that caused no change in turnout, 4.2 effects that caused turnout to decline, and 1.5 effects that caused turnout to rise. Only two variables had effects on turnout that were positive in more countries than they were negative: party cohesiveness and female empowerment. All other variables had effects that were predominately negative, as we already know. Note, however, that no effect was consequential in every country, and that four of the ten variables had no effects on turnout in by far the majority of countries.The variables with the most widespread effects (effects that are felt in the largest number of countries) are those having to do with the expansion of the electorate, the lowering of the voting age, and the degree and nature of party competition. Indeed, majority status and margin of victory each had effects in twenty-one out of twenty-two countries (and not the same twenty-one in each case). Young initiation had effects in twenty countries, electorate size in nineteen countries, and party cohesiveness in eighteen countries. No other variable had effects in more than half the countries. Of particular interest is the fact that the variables with the most widespread effect are generally short-term factors. Two cumulative factors (compulsory voting and executive responsiveness) each have effects in only two countries, leaving only two cumulative factors with widespread effects. So most changes in turnout are due to short-term factors: interactions of the proportion of the electorate that respond to the character of elections with a very small number of variables specifying that character. If one is concerned about the sizes of effects over the full thirty-year period, institutional changes can be seen in Table 7.3 to have been responsible for by far the greatest changes in turnout. Changes in compulsory voting laws, executive responsiveness, and the franchise were each responsible for effects that averaged between 3.39 and 6.4 percent in each country where such a change took place. No other variable was responsible for changes that averaged more than 1.9 percent. Nevertheless, in any given election, the dominant influences on voter turnout are short-term, having to do primarily with the nature and extent of 9

The sum of the effects of young initiation and female empowerment.


185

party competition at that election. In conjunction with changes in the proportion of the electorate responsive to short-term factors, the five short-term factors can each of them easily cause a turnout change of 1 percent or more between any election and the next. Working together, they could account for shifts of up to 6 percent from election to election. Only over a series of elections extending for decades do cumulative forces exceed those of short-term factors. Between any given election and the next, the incremental change due to cumulative factors is quite puny, which might be one reason why previous research has not detected the operation of long-term effects. The Mainsprings of Turnout Change The fundamental building blocks for the findings in this and previous chapters come from the analysis of voter motivations put forward in Chapter 2, along with the mechanisms of generational replacement and inertia (together with its corollary, lack of inertia) put forward in Chapter 3. Once we understand that certain voters are more open than others to influence from the character of elections, and that the consequences of those influences can be divided into cumulative and short-term components, it is easy enough to discover what it is about the character of elections that influences voters. In Chapter 5 we established that a small number of factors explained more than 80 percent of the variance in turnout across twenty-two nations over the course of half a century (more than 90 percent when country differences were taken into account). In this chapter we have shown how those same influences can largely explain both the early rise and later decline in turnout over the same period. Now that we have established the nature and power of these influences to affect turnout during a fifty-year period, it is time to take stock of what exactly we have learned. What are these factors, and how do they relate to the theoretical model describing the motivation to vote that was developed in Chapter 2? The factors responsible for turnout change over this fifty-year period can be classified in two ways. First, there is the division (explicit in our tables) between cumulative and short-term factors. It might be wondered how short-term factors cause change over a fifty-year period, but the “short-term” designation only means that the variables

186


concerned do not have cumulative effects – not that they cannot contribute to long-term turnout change. A variable such as majority status mainly affects the turnout of newer cohorts, but this does not prevent it from being associated with long-term change if its effect on new cohorts moves progressively in a given direction.10 Nevertheless, while short-term factors can indeed have long-term effects, their effects are not intrinsically long-term in nature. The effects we designate as cumulative in our tables are distinctive in having inherently long-term effects. Second, we can classify factors responsible for turnout change into those due to changes in the character of elections and those due to what we will call “cohort effects” – changes in the proportion of the electorate that is sensitive to the character of elections (for cumulative effects, changes in the proportion having been exposed during their formative years to a new characteristic). This classification is implicit in the tables presented so far and is built into the interactions employed to measure short-term and cumulative factors. Because each interaction has two components, a measure of an election’s character can remain unchanged and still contribute to turnout change if there is change in the proportion of the electorate paying attention to that aspect of an election’s character.11 We have already referred to the fact that some of the changes estimated in Table 7.1 are due primarily to changes in the sizes of attentive cohorts rather than to changes in the character of elections per se. In Table 7.4 we focus on the decline that occurred after the middle 1960s and apportion the effects of different variables into those due to cohort effects and those due to changes in the electoral characteristic that constitutes the other half of the interaction concerned. In order to do this

10

11

The strong interrelationship between short-term and cumulative measures prevents our finding both versions of any given variable to be significant in the analyses conducted in Chapter 5. In complementary analyses conducted in Chapters 3 and 6, we do see the character of elections having significant effects on established cohorts, even though effects on newer cohorts are stronger (very much stronger in the analyses conducted in Chapter 6). This is where the use of interaction terms involving the size of a particular group in the electorate really scores. In individual-level or cohort analyses (such as those conducted in Chapters 3 and 6) we cannot readily take account of changing sizes of electoral cohorts. Their contribution to turnout change can only be estimated using interaction effects in aggregate-level analyses such as those conducted in Chapter 5.

187

3.682 10.632 9.700 3.666 0.773 0.273 2.636 0.955 16.744 0.000 83.239

0.397(0.171)∗ −0.089(0.043)∗ −0.071(0.038)∗ −0.150(0.078)∗ 1.813(1.558) 5.127(1.416)∗∗∗ 5.065(1.486)∗∗∗ −9.334(1.952)∗∗∗ −0.101(0.022)∗∗∗ −1.117(0.427)∗∗ 0.513(0.052)∗∗∗ 35.925(6.164)∗∗∗ 334 0.771 0.120 3.360∗

Time since previous election Majority status Margin of victory Mean margin of victory Cohesiveness of party system Compulsory voting Executive responsiveness Female suffrage Electorate size Whether voting at eighteen years old Previous turnout Constant Total of estimated changes N Variance explained Breusch–Godfrey test (3 df) Arch test (3 df) Note: Significant at ∗ 0.05, ∗∗ 0.01, ∗∗∗ 0.001 (one-tailed).

X1965

b 3.409 16.103 8.361 2.864 0.818 0.182 2.636 1.000 26.164 0.909 78.279

X1999

−0.174 −1.078 −0.262 −0.118 0.634 −0.540 −0.320 0.200 −0.971 −2.178 −2.032 −6.838

−5.699

Table 7.1

−0.108 −0.484 0.095 0.120 0.082 −0.468 0.000 −0.420 −0.956 −1.016 −2.545

This table

Turnout change estimated in

−1.139

−0.065 −0.593 −0.357 −0.238 +0.552 −0.073 −0.320 +0.620 −0.015 −1.163 +0.513

Consequence of cohort effects

table 7.4 Comparing Findings Without Cohort Effects with Findings that Take Account of Cohort Effects Using Fixed-Effects Regression (standard errors in parentheses)

188


we need to estimate an equation that predicts turnout in our data without employing interactions. Comparing the expected change in turnout from this equation with the expected change in turnout from the equation used in Table 7.1 tells us what difference it makes whether we take responsiveness (and cumulative effects) into account. As a by-product, the table also enables us to gain an appreciation for the extent to which we better understand turnout change when we estimate a model that takes account of changing cohort sizes. The consequence of taking account of cohort effects for our ability to correctly estimate turnout decline is seen in the final three columns of Table 7.4. Comparing the penultimate column (headed “Turnout change estimated in Table 7.1”) with the previous column, we see that taking cohort effects into account generally amplifies the estimates that would have been made without the interactions, though in three cases (margin of victory, mean margin of victory, and female suffrage) the sign of the estimate is reversed and in one case (executive responsiveness) all of the estimated change is due to cohort effects. The final column of the table gives the difference between these two estimates, the consequences of taking cohort effects into account. These differences are generally large, even when they do not involve a change of sign. The reversal of sign for female suffrage is easily explained. It is a matter of whether we take account uniquely of the enfranchisement of women in Switzerland (which Table 7.4 tells us lowered turnout in that country by 9.33 percent – a quantity that averages out to 0.42 percent over all twenty-two countries, as shown in the fourth column of Table 7.4) or whether we also take account of the increases in turnout still occurring in other countries that gave the vote to women in the immediate post-war period. Similarly, there was no change in executive responsiveness in this period – only in the proportion of the Swiss electorate who had been exposed to a lower level of executive responsiveness when they were young. The effects that we estimate for longterm factors depend completely on how we conceptualize the manner in which turnout change takes place. The consequence of taking account of cohort size in the case of shortterm factors has to be thought of rather differently. If impressionable cohorts increase in size relative to the electorate as a whole, then the impact of a variable with a given effect can grow. This is easily seen


189

in the case of cohesiveness. The variable itself hardly changes in value, but changes in the proportion of impressionable cohorts evidently coincided well with the distribution of cohesiveness across countries, so in interaction the two yield an influence on turnout that would have pushed turnout up in the absence of countervailing forces. Interactions with margin of victory worked in the opposite direction, helping to explain falls in turnout. The best way to think of this is to see margin and mean margin as largely unchanging (like cohesiveness). The much greater changes in the sizes of young cohorts thus had the opportunity to help explain the direction of turnout change. Large margins of victory are bad for turnout, and they become worse when there is an increase in the relative sizes of cohorts responsive to such factors.12 The actual changes in the two measures of margin of victory, small though they were, served to reduce the extent of turnout change that would have taken place had these variables remained truly constant over time. Before we leave this table, we should take advantage of the opportunity it provides to evaluate the benefits of taking cohort effects into account. The regression equation summarized in Table 7.4 performs nearly as well as our preferred Model E from Table 5.2, explaining 77 percent of variance instead of 81 percent and yielding statistical significance for all variables except party cohesiveness.13 However, the variables that in Model E were taken in short-term interaction have much weaker effects in this table – barely significant, indeed. Additionally, the test for heteroskadesticity proves significant at the 0.05 level, suggesting that errors of prediction are not randomly distributed.14 The consequence of these deficiencies is that the equation makes a 12

13

14

It is clear that 9.1 percent (the average over all countries for the period) is a high margin of victory. Three percent seems to signal much closer races. However, the 3.4 percent average for mean margin of victory is an average in each majoritarian country over all its constituencies/districts and indicates the presence of a lot of very safe districts. These are the districts in which turnout will have been low. The equation cannot be improved by adding variables that failed to prove significant when added to Model E. None of these variables prove significant when we attempt to add them to this equation either. The new cohorts proportion is virtually unrelated to turnout or to any of the other variables in the equation, so multiplying independent variables by this term produces interactions whose distributions are more uniform across countries than are the distributions of uninteracted independent variables, evidently reducing the prevalence of nonrandom error.

190


notably inferior job of explaining turnout decline. As shown in the final coefficients of the last three columns, taking account of cohort sizes yields the estimate for turnout decline shown in Table 7.1: a decline of 6.84 percent. This, as we saw when discussing Table 7.1, from which it is taken, underestimates actual turnout decline by some 1.4 percent. Failing to take account of changing cohort sizes results in an estimated decline of 5.7 percent – more than 1 percent further from an accurate estimate of turnout decline. The ability of the equation summarized in this table to track turnout variations in individual countries (not shown) is similarly degraded.15 Discussion It remains for us to evaluate the actual effects on turnout that we have found. We take them roughly in order of the sizes of historic effects on turnout seen in Table 7.1 to have resulted from these forces during the last half of the twentieth century. By far the most powerful effect is that of young initiation. This is the variable that provided the focus for the analyses conducted in Chapter 3 of this book – the chapter that investigated the importance of socialization and inertia. In the present chapter we see that it bears more responsibility for turnout decline than any other. Though its contribution (at some 2.6 percent, including indirect effects) should not be exaggerated, we should add to it the 0.7 percent impact of the relative enlargement of younger cohorts whose contribution, while not statistically significant, seems indisputable, if only because it tallies with separate estimates made from first principles by McDonald and Popkin (2002) and by ourselves. Taking these effects together, the changes at the lower end of the age structure account for close to half the average turnout decline that took place over the half-century.16 Though 15

16

Taking account of interactions with cohort sizes might seem to smack of capitalizing on chance, except that the precise same effects prove significant when we employ individual-level data (Chapter 6). Failure to operationalize at the aggregate level insights developed in Chapter 3 and confirmed in Chapter 6 would result in a model that was theoretically misspecified. The fact that the misspecified model also explains less variance and has other deficiencies confirms the need to take account of relevant cohort effects. If we remove Italy, Malta, The Netherlands, and Switzerland from consideration – countries where major changes in turnout are incidental to changes in a single variable


191

responsible for far less overall turnout change, female empowerment has analogous effects, reinforcing the support we find for the socialization thesis. Cohorts with a distinctive experience of elections during their members’ formative years show a different pattern of behavior throughout the ensuing lifetimes of cohort members. This is a critical insight, which underlines the importance of separating cumulative from short-term forces. The findings also lend circumstantial support to some of the ideas expressed in Chapter 2 of this book. The decision to vote is hypothesized to be partly governed by social linkages. The acquisition of these linkages lowers the costs of voting for new voters, and the need to drum up support for one’s potentially winning coalition motivates established voters (and especially their interest groups) to seek out potential voters and provide them with the social linkages they need (cf. Plutzer 2002). Young initiation interferes with these processes by bringing people into the electorate who are not readily reached and who are not yet motivated to develop the linkages that would lower their costs of voting. The very fact that this variable has the effect it is seen to have helps to validate the assumption that social integration (or something very like it) plays a critical role in the decision to vote.17 The second largest contribution to turnout decline comes from majority status. This is a variable whose value has grown over time, as the

17

(compulsory voting in the case of Italy and The Netherlands and executive responsiveness in the case of Malta and Switzerland) and correct the remaining countries’ turnout for young initiation by adding to turnout at each election the product of the young initiation coefficient from Model E times the value of the young initiation variable at that election, the trend of the resulting variable is reduced from –5.46 (seen in Table 1.1) to –2.35. This figure is an average from changes both up and down in turnout (many of them quite large) that it is important to understand, but it is doubtful whether such a small drop in average turnout would have raised the degree of concern among commentators that has been lavished on the turnout decline that has actually occurred. However, we find no evidence to support the idea that the extensiveness of existing networks helps determine the level of voter turnout. Our theory, it will be recalled, requires such networks to exist but does not propose a link between the extensiveness of such networks and turnout. The important question seems to be not how extensive such networks are, but who do they involve? When we identify groups of individuals who are less likely to be involved in relevant social networks, we find the extensiveness of these groups to correlate with turnout. There is, however, no reason for the extensiveness of these groups to correlate with the density of social networks. Social networks can become more or less dense simply by involving the same number of people in more or fewer groups.

192


fractionalization of party systems has resulted in large parties falling further and further below the 50 percent support threshold that would deliver total control of the legislature into their hands. But the variable works much better in interaction with newer cohort size, indicating that the increasing proportion of the electorate responsive to the character of elections also plays a large part. The same is even more true of margin of victory in both its forms. For these two variables the effects on turnout are due entirely to the increasing size of the cohorts responsive to these factors, as just discussed. This increasing responsiveness has resulted in a negative effect on turnout despite the fact that races actually became closer, on average, over the course of our time period. These three variables together give a sophisticated picture of the state of party competition, and voters’ assessments of it, at any given election; and between them they account for more than 2.3 percent of turnout decline – almost as important a set of influences as the lowering of the voting age. We see consequences almost as strong as those of majority status (but in the opposite direction) from party cohesiveness. We have already reported some unease with the way this variable is coded (see Chapter 5, footnote 23), and, as already noted in this chapter, its effects in our analyses hardly come from changes in party cohesiveness itself but mainly from changes in the responsiveness of young cohorts to party cohesiveness (the same story as for margin of victory). The fact that so much work in our model is done by the changing sizes of impressionable cohorts is fundamental to understanding turnout change. Change in responsiveness constitutes the lynch-pin of the theory put forward in this book, and its operation in interaction with party cohesiveness provides an excellent example of how this works out in practice. As newer cohorts grow in size relative to older cohorts, so the proportion of the electorate for which this factor matters also grows. This is exactly as it should be. Indeed, the importance of party cohesiveness in interaction with newer cohort size makes great intuitive sense. It indicates the presence of a sophisticated electorate whose newest members do not see much point in voting for a party that cannot deliver its legislators in support of party policy proposals. The result is lower turnout for countries (notably the United States and Switzerland) for which party cohesion was already low. The overall effect of this variable is, however, positive because most


193

countries have parties with good cohesion. As the number of voters has increased for whom cohesion is important, this variable has helped to offset the decline in turnout – a decline that would otherwise have been even greater than the decline we observe historically. The fact that the overall effect has been to offset turnout decline makes it clear that this interaction is not just capitalizing on chance and validates the similar importance of increasing responsiveness in the case of margin of victory, where the interaction has, by contrast, helped explain turnout decline. The five just-mentioned variables (party cohesiveness, executive responsiveness, majority status, and the two versions of margin of victory) dominate our findings regarding turnout change. All of them are closely connected to the notion that voting bestows benefits that depend on the responsiveness of the political system to changes in voter preferences. If the largest party in a system is not large enough or cohesive enough to be able to enact its program, or if the outcome of the election is a foregone conclusion, or if institutional or party arrangements are such as to insulate executive power from the outcome of legislative elections, then elections serve much less purpose and the point of voting is less. As political systems move in the direction of less accountability to their electorates (or electorates become more sensitive to the lack of accountability that their systems show) so turnout declines, both immediately and over ensuing decades. One last variable remains to be mentioned. Time since the previous election has generally risen on average (though with a slight decline after the mid-1960s). This could just be happenstance, but it could also be linked to increasing margins of victory and party cohesiveness, which have made ruling parties generally safer from legislative revolts than they were in the immediate post-war era. This variable is seen to have effects on all voters, not just those in their formative years. So this variable is anomalous among significant influences on voter turnout that otherwise operate primarily on new cohorts. Most of these forces gain long-term effects (when they do) by having retained their values long enough for new voters to become established as they age. There is nothing in our theory that precludes established voters from reacting to changes in the character of elections; and in Chapters 3 and 6 we found them doing so, though at a lower rate than the rate at which newer voters reacted. So the real question when we confront this finding is not

194


why is this variable significant, but why do none of the other variables show evidence of affecting established cohorts? The answer in general terms is certainly bound up with the difficulty of achieving significant findings with small datasets. The dataset employed in Chapter 6 relates to 8,410 individual respondents to German survey questions. The dataset employed in Chapter 3 relates to 546 cohorts-years (the unit of analysis in those data). The dataset employed in Chapter 5 relates to only 346 elections – 334 once the first election in each country had been employed in computing variables with an over-time component. Problems may also arise from the fact that we cannot absolutely distinguish established from newer cohorts in a country-level dataset. The role played by different groups within each country in the dynamics of turnout change has to be teased out through the use of interaction terms. This worked very well in general in the analysis conducted in Chapter 5, but we should not be surprised if there are situations where the procedure is not adequate to model the complexities of the real world. One possible complexity that might apply to time since the previous election was hinted at earlier. Perhaps this variable is misconceptualized. Perhaps it is acting as an indicator for the frequency of elections. As time since the last election goes down, so the frequency of elections rises. If impediments to voting (such as illness) are randomly distributed, then more frequent elections mean a greater chance for each individual that particular elections will coincide with some impediment to voting. A greater chance of each individual failing to vote at any given election translates into lower turnout by individuals in general at elections in general (another way to think of this is in terms of a set number of voting opportunities being spread more thinly over a larger number of elections). In this way, time since the previous election would be operating as a surrogate for random events that get in the way or otherwise raise the cost of voting. If the distribution of such events is indeed random over time, then they will more frequently interfere with the voting act as elections are held closer together. Such factors would not distinguish between established and newer voters.18 18

This interpretation suggests that the frequency of elections might indeed be a factor influencing turnout, a possibility that was not included in our research design because of difficulties in distinguishing it from executive responsiveness. It will be recalled that the countries with the most frequent elections are also those with the lowest executive responsiveness. The possibility is reinforced by the fact that the institution of elections


195

Leaving this anomaly aside, our findings have broad implications for the manner in which people learn the habits of voting and nonvoting – implications that we will explore in Chapter 8, the final chapter in the book. Before we turn to that, however, we should enumerate some of the implications of our findings for future political science research.

Implications for Rational Choice Theorizing Our findings are consistent with the model presented in Chapter 2 and confirm some important components of that model. In particular, we have clearly confirmed the structure of Equation 6 in Chapter 2, which is repeated here: Pi (vote) = Hi + {(1 − Hi )( f (Ui (voting))}.

(6)

In all of our prediction equations, the H term, indicating the habit of voting or nonvoting, proves to be necessary (implemented either in terms of lagged turnout or in terms of whether each individual voted at the previous election). Equally, the (1 − H) term, which conditions the rest of the equation, proves to be necessary (implemented in terms of short-term and cumulative effects or of membership in new cohorts). The Ui (voting) component is also confirmed by our findings. People do vote at higher or lower rates depending on the utility of voting (implemented in terms of the character of elections).19 When we move back to earlier equations presented in Chapter 2 our success is less evident. The limitations of our findings can be illustrated in terms of Equation 5 in Chapter 2, which is repeated here: Ui (voting) =

|Ui (D) − Ui (R)| + ◦

M

(Mi ) − (Ci − CNVi ).

(5)

Once the U term from Equation 6 is “unpacked” into its component parts, which are laid bare in Equation 5, it becomes clear that our

19

to the European Parliament had depressive effects on turnout at national elections in member countries, but this appears to have been due not to the resulting increased frequency of elections in those countries so much as to the socializing effects on young voters of the experience of low turnout elections (Franklin 2003). We mentioned at the end of Chapter 4 that the complications involved in implementing short-term and cumulative interactions properly require a slight modification to this equation.

196


findings have not established the parameter values associated with all the components. To be sure, the C term (relating to the costs of voting and not voting) is validated and can be given an actual value by summing the appropriate effects from Chapter 5’s Model E, but we cannot do the same for the remainder of the equation because the independent variables relating to the benefits of voting do not distinguish between the intrinsic utility of each benefit and the impetus that it provides for sending a message to other members of a potentially winning electoral coalition (see Chapter 2). These two components could in principle be separated, but only with the benefit of innovative survey questions that have not so far been asked. Because we could not operationalize either of these components in such a way as to exclude the other, our measures of the character of elections incorporate both utility and message-sending in an undifferentiated fashion. It is true that the relative weight of each of the components of Equation 5 is only of somewhat esoteric interest. It would be satisfying to be able to confirm that both components played a role and interesting to know how large a role was played by each, but the important question is which factors serve to stimulate these motivations, and that question has largely been answered – largely, but not completely. The main respect in which the question has not been answered is in terms of messages of a different type: messages that send an “angry letter” to politicians by breaking party ranks and voting for an upstart party or candidate. We mentioned the importance of this motivation in the 1992 U.S. election where Ross Perot provided a focus for voters who could not make their point by voting for established parties. And in Appendix C we do see a large spike in U.S. turnout for the election of 1992, the recent election most likely to have been the focus of such motivations. That spike reduces the variance our analysis explains even though we believe we understand why the spike is there. The deficiency is one of operationalization, not of understanding. None of the variables suggested in past research (which are primarily the variables employed in this book) deal with message-sending, and it is hard to see how to operationalize a suitable variable without considerable contextual knowledge about each and every election. This must be a task for future research. Even leaving this aside, our overall success in deriving a model that fits the facts in terms of actual turnout variations is not a success for rational choice theorizing. Our model builds on such theorizing and


197

has the form of a rational choice model, yet it was not arrived at in the traditional manner for rational choice theorizing that generally proceeds from very basic first principles, as did Riker and Ordeshook’s original model. The lesson of this research is that productive theorizing may require considerable knowledge about how the world works, and this knowledge cannot itself be gained from rational choice theorizing (cf. Greene and Shapiro 1994). Implications for Research into Voting Choice The two most innovative components of the models tested in this book (inertia and young initiation) were suggested by an analysis of British party choice conducted thirty-five years ago (Butler and Stokes 1969, 1974). Yet the implications of Butler and Stokes’ findings for the analysis of party choice have in fact been largely ignored. Our findings in this book suggest that this may have been unfortunate. Butler and Stokes established that only certain voters are available at any point in time to establish or change their party allegiance. Most voters are set in their ways and will continue to support the same party under virtually any circumstances. Investigations into why voters choose to support particular parties or candidates routinely ignore the implications of Butler and Stokes’ findings. Just as we have shown traditional models of voter turnout to have been mispecified through the omission of an explicit role for inertia, so traditional models of party and candidate choice may well be equally defective. It is to be hoped that the findings of this book will stimulate future research into party and candidate choice that will take account of inertia as a factor that influences voting choice. The influence of electoral context may also prove important in properly specifying models of voter choice – perhaps not the same contextual influences that have engaged our attention in this book, but other contextual influences that do bear on party and candidate choice. For example, it is taken for granted by many commentators that economic conditions affect the vote (“It’s the economy, stupid!”), but models of party choice have seldom taken explicit account of variables measuring economic conditions – only voters’ perceptions of such conditions, which are almost certainly contaminated by voters’ partisan preferences (Wlezien, Franklin, and Twiggs 1997; van der Eijk 2002). This book has made it clear that voters take account of electoral context

198


when deciding whether to vote. Until careful research proves otherwise, it should therefore be assumed that, when voters choose between parties or candidates, much the same is true. Implications for the Study of Turnout in Other Countries In this book we have focused on turnout change in established democracies. We defined established democracies more narrowly than is customary, excluding countries that missed one or more elections in the normal sequence, such as Greece, and countries for which any of the variables employed in this study could not be coded, such as Costa Rica. The theory presented and tested in this book applies to countries for which we have a complete record of turnout and associated variables such as registration over a fifty-year period – the length of time the average person remains in an electorate after reaching adulthood. Countries for which we do not have a complete record could not be included without muddying our findings. The most obvious omission in Costa Rica, for which we lack critical registration data. There is no theoretical reason why voter turnout in Costa Rica should not respond to precisely the same forces as does voter turnout in the countries we studied. Moreover, we expect turnout in additional countries to become predictable on the same basis as the countries we have studied as their history of free elections grows to encompass a fifty-year span. We can think of no reason why our model should not eventually apply to all democracies, once all democracies have conducted free elections continuously for fifty years or more. The reasons why we require a complete lifetime of elections, without gaps, are both theoretical and practical. From a practical perspective, many of the variables we employ are interactions, and these cannot be accurately computed without a series of three elections from which to derive the sizes of new and impressionable cohorts. It is bad enough that we had to compromise in the calculation of these variables for the earliest cases in our dataset (one reason for the poorer predictions made for early elections – see Figure 7.1). Every gap not only creates missing data for the missing election, but missing or error-prone data for the following two elections as well. This rules out the inclusion of the country concerned.


199

There is also a theoretical reason for omitting countries that have short sequences of elections or missing elections. A critical component in our model comes from the record of past elections. This record is what tells us the nature of the footprints in the electorate that reflect the electoral socialization of established cohorts. There is no telling what sort of cohort structure a new democracy might have, so turnout in new democracies during their first fifty years of democratic rule will contain a component that will at best be random, but which in practice will reflect that country’s idiosyncratic history, which could be far from random. More importantly, the role of inertia may be quite different in new democracies. In the first few free elections there is no saying to what extent older cohorts will behave as though they were new cohorts. We seem to have discovered in this book that, when the suffrage is extended to women, at least some of the older women behave as though socialized into the habit of nonvoting, and this may also be true of older people in new democracies. Yet it cannot be completely true because new democracies often see very high turnout at their first elections. A lot may depend on whether elections were held in the predemocratic regime – even if those elections were not free and democratic. Clearly much research is needed before we will know to what extent our findings apply to new and transitional democracies. For now our findings can be generalized in full only to democracies established for long enough to have a cohort structure that can be mapped from the record of past elections. For other countries, the cumulative forces that we have found to be so critical will simply be unknown. That is the down side. The up side is that the short-term factors that do most to determine change in turnout from one election to the next should operate in new and transitional democracies in just the same way as in established democracies. Once we have learned how large a portion of the electorate initially behaves as though they were new cohorts (and this can surely be discovered from existing data), turnout change may well be predictable for any country that has held as few as three elections. So the effects of electoral competition, which make up the largest part of the forces shaping turnout evolution, can likely be estimated even for quite new democracies.

8 The Turnout Puzzles Revisited

This book opened by setting out a number of puzzles that face anyone who tries to investigate the reasons why voter turnout levels vary from one election to another. Some of these puzzles arose because relationships expected on theoretical grounds were not observed in practice (or because relationships were observed in practice that were not expected on theoretical grounds). Rather more of them were puzzles about how properly to investigate what turns out to be an elusive phenomenon. The book has addressed these puzzles in different ways. Puzzles about how to proceed were addressed mainly on the basis of inductive and deductive argument, and the success of the book in arriving at coherent and mutually supportive findings vindicates those arguments. In some cases, multiple approaches were adopted in parallel because it was not clear that one approach or another could be established as correct. Thus it was decided to address turnout at three different levels of analysis because turnout is a phenomenon that is inherently multilevel in nature. Similarly, multiple estimation strategies were employed in Chapter 5 because, with time-series data, it seemed important to validate the findings by arriving at them in several ways. Puzzles about relationships were addressed by developing two bodies of theory involving (1) why people vote and (2) how the mainsprings of voting behavior translate into turnout variations. These theories were then tested with data at the three different levels of analysis and largely confirmed.

201

202


This chapter will inventory the various puzzles that were set out in Chapter 1 and discuss the extent to which they have been solved. It will do so in the context of a recapitulation of the theories that were proposed, the hypotheses that were derived, the tests that were conducted, and the conclusions that were reached in the course of Chapters 2 to 7. It then puts forward our prognosis for future turnout change, drawing attention to some political implications that flow from our findings and suggesting reforms that might end or even reverse future turnout decline.

Why Do People Vote – Or Why Do They Not Vote? Chapter 2 set out to argue that the so-called turnout paradox (which comes about because people vote even though it is hard to see how they receive any policy benefits from doing so) arises from misconceptualizing the potential voter as an individual divorced from any social context that would give her vote meaning other than its unitary contribution to the pool of votes needed to elect a party or candidate. In this traditional conceptualization, uncertainty about the true balance of political forces leaves the potential voter with no reason to vote, since the chances that any one vote would decide the outcome are vanishingly small. In Chapter 2 we argued that potential voters are not atomized but find themselves members of social networks of various kinds and, most importantly, members of potentially winning electoral coalitions. Seen in that context, the argument that it is supremely unlikely for any one vote to make the difference between defeat and victory is turned on its head. For the member of a potentially winning coalition who believes that her coalition could win if every member voted, uncertainty about the true balance of political forces leaves her with no excuse for not voting.1 1

This argument has interesting implications for the use of opinion polling at election time. More and better polls more carefully followed by more people means less uncertainty as to the election outcome. This development might have served to magnify in recent years the effect of margins of victory over and above the magnification we assume to have occurred because of the increasing sizes of responsive cohorts. Indeed, our failure to anticipate such increases in the effect of margin of victory might amount to a slight misspecification of our models, which might perform even better with an additional interaction between margin of victory and time.

The Turnout Puzzles Revisited

203

Of course this argument only works to the extent that members of an electorate see themselves as members of potentially winning electoral coalitions. One reason why turnout is low among electoral cohorts who have recently reached voting age is precisely because they do not see themselves as members of such groups. Many of them are not in a position to have yet acquired the necessary social linkages, nor have they been adult long enough to have yet been mobilized by those who will attempt to enmesh them in such networks (Pultzer 2002). Even as people age, some of them will fail to become engaged in networks with political connections and will never learn the habit of voting. In due course they will become habitual nonvoters. Most people, however, will become engaged in activities that bring them into contact with the political world and most will learn the habit of voting. Social Networks and Social Capital This line of reasoning seems to suggest the importance of social capital as an intervening factor in the transition of young nonvoters to established voters. It seems indisputable that it has become more difficult in recent years to acquire multiple memberships of large groups, because of the decline of cleavage politics (Franklin 1992) and the loss of various aspects of social capital (Putnam 1995, 2000, 2002). Yet in Chapter 5 we did test for the effects of declining cleavage politics, together with at least one form of social capital (union membership), and found no significant effects. Perhaps we need more extensive measures of social capital, and future research should certainly address this problem. Nevertheless, there is another possibility, already suggested in Chapter 7, which is that what matters is how extensively young adults become engaged, not how widespread are the networks they could engage themselves in. After all, networks can be widespread without involving young adults; and it may be that, once the habit of voting has been learned, the number of social networks the habitual voter belongs to is irrelevant. So future research should perhaps focus not on networks in general but specifically on networks that involve young people. A further possibility, and one that is more consistent with the model we put forward in Chapter 2, is that the density of social networks is not relevant at all. The model does require that there be social networks, but it says nothing about how widespread and extensive they

204


need to be. Indeed, the simple interactions that are part of work and play may provide all the contacts needed to trigger the perceptions of group membership and obligation that provide the lynch-pin of our model. Evidently future research should be directed at this possibility as well, but the fact that we find female members of new cohorts voting at significantly higher rates than men – a finding that echoes recent work by Norris (2001) – suggests that turnout responds to skills and/or attributes that women have in greater abundance than men. This is consistent with the idea developed in Chapter 2 that people’s decision to vote constitutes a cooperative game involving a social process. It is well known that women develop social skills earlier than men and more readily adopt cooperative problem-solving strategies. So a social process such as that described in Chapter 2 naturally benefits women once they have overcome the disadvantages that they inherit from past voting restrictions. The Habit of Voting The transition between unengaged and established appears to happen during the first three elections that people are exposed to as votingage adults. Butler and Stokes (1974) found that this was the length of time that it took people to become habitual supporters of one political party or another, and Plutzer (2002), in his study of the Jennings and Niemi (1981, 1991) data, also finds people making a transition from habitual nonvoting to habitual voting during the span of three elections. To be sure, habitual nonvoters may sometimes vote in later elections; and our data, unlike Plutzer’s (which does not extend beyond three elections) appear to find this happening. But there is no evidence that habitual nonvoters become habitual voters in any large numbers after the third election to which they are exposed – and the cohort tables in Chapter 3 suggest relative immobility in terms of turnout after the third election.2 Such variability as does occur between voting and nonvoting in established cohorts seems logically more likely to 2

Still, the magic number could perhaps have been four elections rather than three (see Chapter 3, footnote 16 for evidence validating our choice of a three-election threshold). Or it could have been a span of ten years, or twelve, no matter how many elections were contained in that span. We did not try out alternate specifications of this variable because of the difficulty of operationalizing all our variables in additional versions. The data do not show our operationalization to have been wrong, but they do not rule out the possibility of a better operationalization.


205

occur among habitual voters, who may find themselves with less reason to vote in some elections than in others. Habitual nonvoters are also habitually inattentive citizens whose failure to vote is part and parcel of a lack of engagement with the political realm that would generally preclude them from knowing whether there were good reasons to vote in some particular election. The data employed in this book are not able to discover which group – the habitual voters or the habitual nonvoters – shows most variability in turnout, but it does verify that these two groups exist and that transitions between them occur much more seldom after the first three elections. On the basis of these findings, the answer to the question posed in Chapter 1 is that for habitual voters it makes more sense to ask why they would fail to vote, while for habitual nonvoters the reverse question is the one that should be asked. During the learning period that seems to follow after reaching adulthood for most people, the important question is neither of the above. The important question is, how do young voters become habituated into voting or nonvoting? The lowering of the voting age in many countries during the late twentieth century had the effect of making it harder to engage the (now-younger) young adults, and this effect of lowering the voting age has not yet run its course. Knowing about the effect of this reform on turnout gives us information about the learning process that we did not have before. In this way it not only provides a large part of the answer to one of the questions posed in Chapter 1 (why has turnout declined?) but also a stepping stone toward a better understanding of the processes of political learning, giving rise to the suggestions for future research listed in the previous section. Our findings also resolve the puzzle of why there is so little turnout variation in overall perspective. Turnout change is constrained by the fact that, between any pair of elections, most people are set in their ways and will not respond to changes in the character of those elections. Only over a period long enough for generational replacement to occur can turnout vary substantially. Something About People, Something About Society, or Something About Elections? The fundamental question that motivates this book is, what causes turnout variations? But this question is poorly specified. What sort of

206


causes should we be looking at? Should we look for causes having to do with the character of individuals, or of societies, or of elections? This puzzle is entangled with another puzzle (that we will address in the next section of this chapter) that concerns the nature of the variations we should be trying to explain. Previous researchers have argued that, because individual-level turnout appears to respond to the level of resources that individuals possess, it follows that aggregate-level turnout will respond to the level of resources that societies possess. This argument is logically flawed even on its own terms, since a link between turnout and (say) education at the individual level does not mean that people are voting because they are educated. It could just as easily mean that, at any given level of turnout, levels of education simply distinguish those who vote from those who do not. Our findings in Chapter 6 provide a number of hints that individual-level relationships reflect aggregate-level processes rather than causing those processes. Turnout appears to vary because of variations in the character of elections, not because of variations in the character of society (except for its age structure), and individual-level relationships between voting and social characteristics are conditioned by the level of turnout rather than the other way around. How Turnout Conditions Individual-Level Relationships These findings, detailed in Chapters 5 and 6, imply that individual-level relationships may well be different depending on the level of turnout and whether it is rising, falling, or remaining fairly constant. One obvious constraint placed on individual-level findings by aggregate turnout levels is that when virtually everyone votes there will be no significant relationships between individual-level turnout and social characteristics, for the simple reason (rediscovered several times in past chapters) that something that does not vary has no relationships with anything. When everyone votes there is no variance in voting and therefore no relationship between voting and any other individual-level characteristic. As the level of turnout falls, the opportunity arises for some individual-level characteristics to show a relationship with turnout, but the relationships we see will not necessarily account for the fall in turnout. More importantly, the relationships we see with high but not perfect turnout may well be different from the relationships we will see with still lower turnout. And they will be different again depending on


207

whether turnout is falling or rising. In particular, falling turnout will produce a relationship between age and voting such that older people, who learned the habit of voting when turnout was higher, will vote at higher rates than younger people, who are learning the habit of voting in less competitive elections (precisely what our data show). But note the critical fact that, if turnout were rising, quite the opposite might be found: Older voters might vote at lower rates because they would have been socialized in a low-turnout era while younger voters (once they had learned the habit of voting) might vote at higher rates because they were responding to the greater importance of elections to them than to their elders when their elders were young.3 Relationships between voting and all age-linked variables would be similarly affected, and most resources are age-linked because people acquire more resources as they age. So the relationships discovered in individual-level studies of why people vote are likely to be strongly affected by the recent history of aggregate-level turnout variations. This in turn implies that these relationships may well be different in different countries, depending on the historic pattern of turnout variation in those countries.4 The Primacy of Electoral Competition At all events, the effects of the resource-linked variables are small in comparison with those of the campaign variables (political interest and party sympathy) or the variables that serve to indicate in those analyses the character of elections (majority status and margin of victory). Moreover, some of these effects (those of party sympathy and political 3

4

Indeed, the tailing off of turnout among older age groups seen in so many individuallevel studies, starting with Verba and Nie (1972), might not reflect what those authors referred to as a “slow-down” so much as the fact that turnout was lower when voters over sixty were themselves acquiring the habits of voting and nonvoting. Putnam (2000) has documented the presence of a highly participatory group of cohorts who came of age from the 1930s to the 1960s. Prior cohorts voted at lower rates, just as later cohorts do, producing the apparent slow-down among older voters. If turnout were to start to rise again, we might see the curvilinear relationship of the late 1960s duplicated but with a much younger peak in turnout, as new cohorts learned the habit of voting at rates above those of older cohorts. For this reason the assertion made in Chapter 6 that our choice of country to investigate in that chapter was simply a matter of convenience turns out to have been somewhat simplistic. We can expect the effects of the character of elections to be much the same in every country, but other individual-level relationships should vary in a predictable fashion with the level and history of turnout in the country concerned.

208


interest in Table 6.2) are partly the consequence of the character of elections. As mentioned in Chapter 6, it should come as no surprise that the character of elections have indirect effects by way of these variables: At the time of a more competitive election, interest in politics goes up, and so does the extent to which people feel sympathy toward their most preferred party. Interestingly, we also found that competitive elections depress the effects of party ID. This apparently happens because in particularly competitive elections (when the race is close and the largest party appears poised to gain control of the machinery of government) people vote regardless of the strength of their party ID. In unimportant elections, by contrast, party ID takes people to the polls even when they have no other good reason for going there. So the unexpected role played by party ID is linked to its role in habitual voting – a variation on the theme of turnout history discussed earlier. The causal analysis that we performed in Chapter 6, whose results were presented in Tables 6.4 and 6.5, made it clear that turnout is dominated by campaign effects and the character of elections, both for established and for newer cohorts (though much more so for newer cohorts). This in turn confirms our supposition that turnout has much less to do with the character of society, in terms of people’s characteristics and the ease with which they can be mobilized, and much more to do with the character of elections, at least insofar as these affect members of newer cohorts – the ones who are likely to move from voting to nonvoting, and vice versa. Variations Relative to What? The question, “What causes turnout variations?” is poorly specified not only because it is not clear what the subject of the sentence refers to, but also because it is not clear what the object refers to. What do we mean by turnout variations? Commentators and politicians who decry declining turnout generally compare current turnout with turnout at the previous election, but political scientists who have investigated turnout have tended to focus either on what makes some people vote when others do not or on what makes some countries vote at higher rates than others. These bases for comparison are not very satisfactory because they do not address the real object of concern. In this book we have focused on turnout change over time, in one country using


209

individual-level data (Chapter 6), in a small set of countries using cohort data (Chapter 3), and in all of the countries for which long-term data are available using the country-election as the unit of analysis (Chapters 5 and 7). Estimating Effects on Turnout In the process we have discovered two things. First, individual-level analyses are inherently misspecified unless they incorporate data for more than one election, because only with multiple elections can one take account of the character of those elections. Variations in the character of elections not only add considerably to our ability to estimate correctly the probability that specific individuals will vote; they also affect in nontrivial ways the estimates made of other influences on turnout. In particular, some variables that appear to play an important role when the nature of electoral competition is left out of account play a different role, a lesser role, or no role at all when elections’ competitiveness is included in the analysis. Second, intercountry differences in turnout are better estimated from models developed to explain turnout change over time than the other way around. Knowing what makes turnout change over time within countries also enables us to explain why turnout differs between countries. Models estimated from country differences, on the other hand, cannot readily explain turnout variations over time because they ignore the fact that many differences in the character of elections only have effects on turnout after a delay occasioned by the need to socialize new voters into new electoral or political arrangements. Country-level investigations are also more liable to spurious findings due to the small N available in such studies. It is true that we did gain information from the over-time analyses in Chapter 7 that included country differences – information that was helpful in interpreting the findings from the fixed-effects models that were the focus of Chapter 5. So the investigation of country differences is not irrelevant to the study of turnout. But such investigations should play a supporting, not a starring, role. Change in Behavior or Change in Social Structure? It is quite common for social researchers to be confounded by the problem of distinguishing changes in the behavior of individuals from

210


changes in the structure of society. Since most people become set in their ways, relatively little social change is possible on the basis of changes in behavior. Most long-term change in most countries most of the time comes about by means of generational replacement.5

Responsiveness of New Cohorts In this book we have given a starring role to generational replacement, but changes in the structure of society have played a role as well. We posited that changes in the relative sizes of impressionable cohorts would affect responsiveness to changes in the character of elections. This idea proved to be a fundamental breakthrough to an understanding of the dynamics of turnout change. Members of newer cohorts are more responsive to changes in the character of elections, and the relative size of this group strongly affects turnout by changing the proportion of the electorate that responds to changes in an election’s character. In some cases, changes in the size of new cohorts can affect turnout even without any change in the particular variables that newer cohorts are reacting to. A characteristic that would boost turnout (strong party cohesion or a close election, for example) will have greater consequences for turnout when the proportion of newer cohorts is large than when the proportion is small. We ascribe an additional role to the size of younger cohorts, even though the effect of the relevant coefficient was not significant in the analyses where we tested this idea (in Chapter 5). We assumed that, because newer cohorts vote at a lower rate than more established cohorts, the size of this group should affect overall turnout. The fact that the effect is not significant does not discourage us. It is not significant because the effect is very small and the N in our analysis is not large enough to yield significance with such a small effect. But since the effect expected theoretically was itself not large enough to prove statistically significant in a dataset of the size available to us, we accept the effect as real despite this.

5

There are, however, clear exceptions to this general rule. See van der Eijk and Neimoller (1983) for an analysis that shows quite the opposite balance of forces over a fiveelection period in The Netherlands. For evidence of the fact that Dutch voters are unusual in this regard, see Franklin et al. (1992).


211

How Do We Know When a Relationship Is Causal? We don’t. The research conducted for this book was no more able to find definitive evidence of causation than any other scientific endeavor. Science cannot confirm theories; it can only fail to prove them wrong. We have done rather more than that in this book. We have confronted our theories from multiple directions using multiple datasets and multiple analysis techniques. The results have reinforced each other insofar as it was possible that they should do so. Where we investigate the same relationships we find the same effects, to within about one percentage point, whether we use aggregate cohort-level data, aggregate countrylevel data, or individual-level data. The fact that we fail to disconfirm the major findings in three different ways does not prove them to be real, but it provides a degree of confidence that is unusual in social research.

Major Findings The major findings that have survived separate tests in each of Chapters 3, 5, and 6 are that (1) the character of elections matters for turnout; (2) among the variables defining an election’s character, majority status and margin of victory are definitely included; (3) newer cohorts react more strongly (much more strongly, according to estimates made in Chapters 5 and 6) to changes in the character of elections than established cohorts do; and (4) extensions of the franchise to eighteen-year-olds in most countries during the late twentieth century has cost a bit less than 3 percent in turnout so far and will cost at least 4 percent in turnout by the time all prior cohorts have left the electorates of these countries.6 The major finding that has survived separate tests in Chapters 3 and 5 is that (5) turnout changes over time can be quite well tracked (with over 80 percent of the variance explained) using independent variables designed to measure the changing character of elections, alone and/or in interaction with cohort effects. Findings that arose from only a single analysis, but which were 6

Taking together, the direct and indirect effects that our model tells us can be expected when generational replacement has run its course. Evidently the effect can differ between countries since we know from our investigations in Chapter 6 that the effect in Germany is somewhat greater than elsewhere.

212


statistically significant, include an indication that the extension of the franchise to women had effects analogous to those that occurred with the extension of the franchise to eighteen-year-olds (Chapter 5), an enumeration of the aspects of an election’s character that have effects on turnout (Chapter 5), an estimate of the direct and indirect effects of those among these variables with significant effects on German turnout (Chapter 6), a determination of the extent to which turnout in specific countries can be predicted on the basis of a general model designed to fit the pattern of turnout variations for all countries taken together (Chapter 7), and an accounting of the extent to which we can explain turnout decline in each of those countries and over all countries taken together (Chapter 7). What We Have Learned The findings summarized in the previous section make a long list, but the list is not one of isolated and unconnected results. Remarkably, we have pretty much succeeded in the task that we set for ourselves in Chapter 1, of finding answers to the two primary research questions that we posed there (why is turnout declining, and does it matter?) And in doing so we have met all the constraints that were enumerated in that chapter – the constraints involved in providing satisfactory solutions to the various puzzles and also the constraints involved in providing consistent answers that made sense at each of three different levels of analysis. So much for our achievement, but what do our findings amount to? What have we really learned? Turnout Decline The first thing we have learned is that there is nothing inevitable about declining turnout. Turnout has declined (where it has declined) for a mixture of demographic and political reasons. The fact that compulsory voting increases turnout was well known when the legal requirement to vote was abolished in The Netherlands and Italy, so the lower turnout that resulted in those countries was a consequence that politicians were willing to see when they enacted these reforms. The fact that an electoral cartel would serve as a discouragement to voting was probably not known to Swiss politicians when they inaugurated their “golden rule,” but, nevertheless, falling turnout in Switzerland (like


213

rising turnout in Malta) was an incidental consequence of political change. Along the same lines, low voter turnout in the United States can be regarded in part as an incidental cost of separated powers – a system put in place for political reasons. The Voting Age Lowering the voting age to eighteen in most countries (the only franchise extension in the history of many of them that was not demanded by those to whom the franchise was extended) brought a similar cost in return for no very obvious gains. Indeed, looked at with the benefit of hindsight, this reform was a really bad idea. It turns out that the well-intentioned decision to enfranchise young adults one election earlier than previously had the unanticipated consequence of giving rise to a lifetime of disenfranchisement for many of the intended beneficiaries. Ironically, almost any other age from fifteen to twenty-five would be a better age for individuals to first be confronted with the need to acquire the skills and knowledge necessary for casting a vote; and since it would be politically difficult or impossible to now reestablish an older voting age, the most promising reform that might restore higher turnout would be to lower the voting age still further, perhaps to fifteen. Given the vote at fifteen, most children would face their first election while still in high school. They could then learn to vote in the context of a civics class project where they were graded on their ability to discover relevant information (including how to register and find the polling booth, where relevant) and assess party and candidate promises in the light of that information. The real-world experiment now being conducted in Cambridge, Massachusetts (see Chapter 3) may in due course provide at least suggestive evidence on this point. Various organizations and perhaps some political parties would oppose such a change on the grounds that it would give teachers a role in forming the political opinions of the next generation. However, teachers already have such a role, and the opportunity to guide young people through their first electoral experience would not necessarily lead to an increase in influence. In fact, the opportunity to teach young adults that they have the civic duty to think for themselves in the political realm might even diminish the specifically political influence of teachers. At all events, this appears to be the only feasible reform that might undo

214


some of the damage that was done when the voting age was lowered in most countries during the late twentieth century. Even should such a reform be enacted, some countries have the prospect of many years of turnout decline ahead of them. Those that lowered the voting age in the late 1970s or later are less than halfway through the replacement of their electorates with newer cohorts, who will each in turn be socialized into a lower rate of voting than their predecessor cohorts, and in those countries there will continue to be a downward push on turnout that it will be hard for other forces to overcome. The good news is that countries that lowered their voting age in the late 1960s and early 1970s are nearing the end of that process; and in the final ten years, we can expect the last of the decline due to changes in socialization to be counteracted by a shift in age profile for most countries as the bulge created by the baby boom moves into the oldest segment of the population – the one with the highest turnout. In countries that were late in lowering their voting age, the same demographic shift may offset some of the decline to be expected in coming years, but not all of it.7 Other Incidental Factors Politically motivated changes in the character of elections are not the only changes that have led to lower turnout in recent years. As we saw in Chapter 7 (Table 7.2), the other major contributor with general applicability has been the fractionalization of party systems in many countries (indicated in our data by the variable we call “majority status”) that has reduced the prospects for definitive electoral outcomes in those countries. The average difference between votes cast for the largest party and 50 percent has increased by more than half, as seen by comparing the mean for 1965 (10.6) with the mean for 1999 (16.1) – see Table 7.4.8 The sizes of electorates have also increased markedly 7

8

In some countries (notably the United States) the lowering of the voting age is almost the only thing that caused turnout decline since the 1960s. Indeed, there might even have been a rise in turnout at U.S. presidential elections had the voting age not been lowered (cf. McDonald and Popkin 2001). However, turnout at congressional elections (the elections studied in this book) would still have fallen even without this change to the franchise (see Table C.5). As stated in Chapter 4, the importance of this variable comes not only from indicating the likelihood that a party will win a majority in its own right, but (at smaller sizes) also the likelihood that it will dominate coalition bargaining and thus get more of its program accepted as government policy than other parties.


215

(from 16.7 million to 26.1 million, on average) during the period under study, as shown in the same table. Again this is a process that is not likely to continue outside the United States (in the absence of increased immigration), so this source of turnout decline also has limited future potential. Others among these independent variables have changed much less but nevertheless have together taken a considerable toll on turnout – especially in interaction with the increasing sizes of new cohorts. This final influence – the most critical variable of all because of its interactions with other variables – cannot go on rising indefinitely, and the demographics of especially European societies assure us that the proportion of new cohorts must very soon start falling, if it is not already doing so.9 So the effects of all the variables that gain leverage on turnout through their interactions with this variable will also start to fall. It is true that the negative effects of changing majority status on turnout could continue for the foreseeable future, but it is hard to see how this one influence could overwhelm all the positive influences that are already starting to be felt and whose momentum should increase in coming years.10 Civic Virtue and Disaffection The second thing we have learned is that turnout decline, to the extent that it occurs, is in no way due to any decline in civic virtue or increase in political disaffection.11 Turnout is not something about the way people approach elections but something about how elections appear to people. Turnout changes either when elections change their character or when demographic shifts place more (or fewer) people in the groups that pay attention to the character that elections have. Changes in the character of elections, in interaction with changes in the proportion of 9

10

11

Our data show the proportion still rising (though very slowly) even when the final five-year period is compared with the previous five-year period, but this is only true over all countries averaged together. In many individual countries the proportion is already falling. Ironically, the reduced sizes of new cohorts, which should slow any future turnout falls, will also make it hard for turnout to increase. Positive influences on turnout will mainly affect new cohorts, but overall turnout will be held down by the inertia of cohorts established in the habit of low turnout, and these cohorts are mainly large ones. Variables measuring these factors did not correlate significantly with the residual from the analysis presented in Model E, Table 5.2, as explained in Chapter 5 (see footnote 36 in that chapter, and surrounding text).

216


people paying attention to those changes, account for 81 percent of the variance in turnout using a relatively conservative estimation strategy (other strategies explained 91 and 94 percent of variance, respectively) and come within 0.75 percent, on average, of tracking the actual record of turnout change in twenty-two countries over half a century. Generational Replacement The third thing we have learned is something that we keep learning and keep forgetting: The future lies in the hands of young people. Young people hold the key to the future because they are the ones who react to new conditions. Older people are, on the whole, too set in their ways to be responsible for social or political change, so most longterm change comes about by way of generational replacement. Because young people hold the key to the future, any reform that primarily affects young people can have large effects on voting behavior. Changes in the voting age have had dramatic adverse consequences. But the possibility of further lowering the voting age should be carefully studied precisely because such a reform shows promise of dramatic beneficial consequences. Mobilizing Agencies Because of the pivotal role of young people, one reform that could have dramatic consequences would be a reform in the behavior of mobilizing agencies. Especially in the United States, political parties make little or no attempt to mobilize young voters, preferring instead to focus on voters whose concerns are familiar to them. A serious attempt by parties to engage young voters could have a dramatic effect on turnout. More importantly, such an attempt could have a dramatic effect on the political fortunes of any party that was to focus on young voters. Our findings suggest strongly that targeting established voters is a waste of money. They will either vote or not, depending on their past history, and (though our research has not shown this) they will probably vote, if they vote, for the party they have always voted for. Past research shows clearly and repeatedly what politicians myopically ignore, that the real payoff from electioneering is to be found in voters not yet immunized against changing their past behavior because they have no long track record of past behavior to change. If this fact were to be taken on board by parties and other mobilizing agencies, the result


217

could be dramatically more competitive elections. A surge in voter turnout among incoming cohorts would be an incidental consequence. The Future What is our prognosis for the future of turnout change? While the total sum of negative effects on turnout looks likely to be reduced in coming years, much depends on the evolution of party systems. Enough additional fractionalization could maintain the existing trend. Moreover, the character of elections that we have measured in this book by no means accounts for all the variations in turnout that we observe. No development among our independent variables accounted for the rise in turnout in the United States in 1992, or in Germany in 1998, or the fall in New Zealand in 1999 (see Appendix C). So, while we can be fairly confident about general trends, nothing said here should be taken to imply that we can make accurate predictions of turnout in specific countries at specific elections. Still, except perhaps in Switzerland and the United States (as we shall see in the next section), there is also no reason why turnout decline need continue indefinitely; and that is a major finding. Some Political Implications This book has been largely about turnout change. Yet countries differ enormously from each other in terms of turnout level. The supposition supported by our findings is that these differences arise from past effects on turnout that have run their course. Even during the time period covered by our data, we do see countries moving from medium to low turnout (Switzerland), from medium to high turnout (Malta), or from high to medium turnout (The Netherlands). Looked at in a data series that started in 2000, these three countries would show little evidence of how they attained the turnout that they now have. This implies that other countries that have occupied fairly fixed positions in terms of turnout since 1945 may also have changed their positions radically at some point in their history prior to 1945. The American Case The most interesting case in this respect is the United States. We saw in Chapter 4 that turnout in that country declined markedly at the end of

218


the nineteenth century (from a medium level of turnout to a low level of turnout). Analyses of U.S. district-level data ascribe this decline to radical changes in the mean margin of victory (see Chapter 4). At the time, those changes occurred because of a realignment of the party system, but the impressive margins of victory acquired at the start of the twentieth century were restored and even enhanced during the last half of the twentieth century as individual members of the House of Representatives learned to use their positions to secure themselves against future electoral defeat (Mayhew 1974; Fiorina 1989). The Swiss Case In some ways, this American story is a parallel to the Swiss story. In both cases politicians learned to guard themselves against the possibility that future elections would threaten the security of their political careers. The stories are different because, in America, individual congressmen secured themselves in large numbers against individual electoral defeat, while the Swiss only secured the positions of party leaders. But that difference is built into the two countries’ different electoral systems. A country in which each legislator is elected by winning a local election is one in which each legislator has the possibility of affecting his or her electoral fate. By ensuring that the huge margins created at the start of the twentieth century did not dissipate, American congressmen also made sure that it remained virtually pointless for voters to turn out in large numbers in congressional elections.12 On the other hand, a country in which representatives are elected from a party list on the basis of votes cast for that list already provides security of legislative tenure for those whom seniority has placed near the top of the list. The only way in which the Swiss electorate could use their votes in a legislative election to affect the policies of government was by changing the number of supporters that party leaders could count 12

By securing their electoral prospects against the vagaries of popular discontent in their districts, American congressmen also removed from ready popular control the overall political complexion of their legislature, but this is not central to the current argument. What is central is the fact that, in the 2002 midterm elections, of the 435 districts in which House elections were conducted, no more than about 35 were considered truly competitive. This means that, for more than 90 percent of the potential voters, their electoral coalition was either certain to win or had no chance of winning. For all these potential voters the uncertainty that would, according to the model developed in Chapter 2, have brought them to the polls was absent.


219

on in the Swiss Parliament. In other parliamentary countries (and in Switzerland until after the election of 1959) such shifts in support do lead to changes in the complexion of governments, and that is what voters go to the polls to try to achieve or prevent. In Switzerland, after the election of 1959, politicians cut the link between election results and government complexion, making voting just as pointless in a Swiss legislative election as it already was in most of the individual races making up an American legislative election. Rigging the System The ability of politicians to “rig the system” so as to secure the future of their political careers is troubling. It is especially troubling that such rigging of the system should have been most fully perfected in the world’s two oldest democracies. Evidently it takes time for politicians in any particular country to learn how to use the features of their political system so as to secure themselves against the slings and arrows of outraged voters. But the fact that this happened in these two countries reminds us of two things: (1) Elected politicians have no strong reason to love competitive elections, and (2) every political system probably has vulnerabilities (or may acquire vulnerabilities) that can be exploited by elected politicians anxious about job security. From this perspective, low voter turnout can act as a signal that something is wrong – not with the voters who fail to turn out or with the society of which they are part, but with a political system in which politicians have contrived to make elections less relevant. The resulting insulation of politicians from the need to be accountable to their voters is only one consequence. Low voter turnout may also result in certain classes of citizens being underrepresented among voters (Schattschneider 1960; Piven and Cloward 1971, 2000). In general, low voter turnout signals that elections may not be performing the functions that they are supposed to perform (representing voters’ interests, holding representatives accountable, and legitimizing their exercise of power) and, to the extent that low-turnout elections do retain any function, that function is restricted to certain classes and conditions of citizen, in defiance of voter eligibility rules. It was in recognition of this last point that Arendt Lijphart proposed in his Presidential Address to the American Political Science Association that the United States adopt compulsory voting, hoping to ensure

220


thereby that all classes and conditions of citizens would vote (Lijphart 1997). It was in recognition of the previous point – that, in low turnout countries, elections are not serving the function of holding politicians accountable – that I argued in my response to Lijphart (Franklin 1999) that to institute compulsory voting in a low-turnout country would be to “shoot the messenger.” Removing the signal that something was wrong would not make that thing right. Electoral Reform in Switzerland and the United States Of course it would be desirable to get a more representative selection of Americans to vote, but it would be even more desirable to restore competitive elections to the House of Representatives. The term-limits movement in the United States is motivated by the idea that politicians are not sufficiently vulnerable, electorally, in the system as it stands; but the solution proposed is the wrong one. Term limits do not restore accountability. They punish all legislators equally and would make it even harder than it is today find good candidates. It would be far preferable to reduce the number of terms that the average congressman (or woman) serves by restoring competitiveness to House elections. That reform (unlike term limits) would ensure that representatives with short careers on Capitol Hill would have short careers because their constituents wanted them replaced. Representatives who did a good job would be more likely to be retained, and natural selection would give rise in due course to a more responsible (and more responsive) legislative body.13 More competitive elections for the House of Representatives would also hold out the prospect of control of the House moving in step with control of the presidency, increasing the 13

It is beyond the remit of this book to spell out how competitive elections could be restored to the United States, but here are a few proposals. Strict limits on campaign spending together with free television time for candidates (as in other democratic countries) would clearly be a step in the right direction. Relaxing the locality rule, so that experienced politicians could challenge each other for the same seats (as they do in other countries and as now happens occasionally after redistricting) would also help. Redistricting itself could be taken out of the hands of state legislatures and/or performed on the basis of simple rules that eliminate collusion to benefit incumbent politicians (as is normal in other countries and as happened in Iowa following the 2000 census). A change of electoral system to one embodying some form of proportional representation would help even more. It is not hard to think of reforms that would have the desired effect. The problem is in generating the will to get one or more of those reforms enacted.


221

importance of elections by making their outcomes more likely to lead to policy change (cf. Franklin and Hirczy de Mino 1998). What about Switzerland? The cartel arrangement in that country has only been tolerated for so long because its effects have gone largely unrecognized. The effects have gone unrecognized partly because, in Switzerland, party politics are less important than elsewhere. Parties do not decide policy. Voters decide policy, since any important policy question is submitted to the voters in a referendum. Nevertheless, there is already a debate under way in Switzerland about the desirability of abandoning their “golden rule” because of its undemocratic nature. When the Swiss come to realize that this rule is also responsible in large part for their low and declining turnout, this will bring home to them the true price they pay for what in any case may be no more than an illusion of stability. I am somewhat more confident that the Swiss will eventually abolish their “golden rule” than that the Americans will restore competitive elections to the House of Representatives, though in optimistic moments I take comfort from Churchill’s remark that “you can always count on the Americans to do the right thing – after they have tried everything else first.” My focus on these two countries in the concluding pages of this book does not mean that low turnout in other countries is not a problem. All cases of low turnout are worrying to the extent that low turnout can cause biases in the system of representation, with certain classes of individuals being less likely to be heard. Research is needed on the exact point at which bias starts to be a problem. Such research should not only look to see how nonvoters would have divided themselves among the options on offer – a relatively straightforward exercise – but also to see what other options might have been offered had there been a wider electorate to assess them – a more difficult question to address (cf. Schattschneider 1960). I am not as concerned about falls in turnout that appear accidental or incidental as I am about falls in turnout that result from deliberate political maneuvering. Accidental turnout declines (by-products of the fractionalization of party systems, for example) will probably reverse themselves in due course. Incidental turnout decline (incidental to the removal of compulsory voting or the enfranchisement of eighteenyear-olds, for example) will not continue beyond the number of years it takes to replace the electorate of the countries concerned. But political

222


maneuvering to ensure continued electoral success affects voter turnout in a way that is neither accidental nor incidental. Such maneuvering is consequential and will not reverse itself unless politicians are stopped from finding new ways of defending their electoral prospects from the normal operations of the democratic process. If they are not stopped, then there is no obvious limit to the extent to which turnout might fall. Total job security by elected politicians might lead in the fullness of time to zero turnout at elections that would have become meaningless charades. Electoral Reform in the European Union The low turnout that can be expected in elections that have no policy consequences is most dramatically seen in elections to the European Parliament. Our study of those elections (van der Eijk and Franklin 1996) concluded that low turnout was inevitable in elections that did not provide voters with real choices. In this book we have reached the same conclusion by a different route. But our findings in this book have implications for European elections that could not be reached from a study of European elections alone. In Choosing Europe? (van der Eijk and Franklin 1996) it was suggested that low turnout in European Parliament elections was due in part to the lack of accountability of European political leaders (the Council of Ministers and Commission) to European voters. That was an assertion that could not be tested in the context of European elections because there was no variance in accountability to be explained. The investigation conducted in this book of accountability variations in the wider world makes it seem that the diagnosis suggested in Choosing Europe? was correct. Elections that fail to hold politicians accountable are elections that will not excite the interest of voters. The findings of this book also suggest that turnout at European Parliament elections might continue to decline, with no floor to that decline above 0 percent, unless those elections are provided with a more compelling rationale.14 14

There is, however, no evidence of declining turnout at European Parliament elections beyond the decline occasioned by the changing composition of the European Union, as established in Franklin (2001) in a puzzle mentioned at the end of that article. Perhaps the reason is that European Parliament elections are not independent of national elections in the member states (Mattila 2003). So long as there is reason to vote in national elections, there may well be spillover to other elections, even elections


223

Solving the Low Voter Turnout Problem In the two countries among established democracies where turnout in legislative elections is lowest, it seems clear that the solution to the low-turnout problem is to find a way to return competitiveness to national legislative elections. In Switzerland we need elections that have a realistic prospect of changing the political complexion of the executive; in the United States we need elections that provide a realistic opportunity to unseat incumbent politicians in a majority of congressional districts.15 Those two reforms should serve to reverse turnout decline in the two countries where turnout is lowest.16 Turnout decline in other countries seems likely to stabilize or even reverse itself without the need for heroic reforms – though a further lowering of the voting age in most countries is nevertheless a reform that should be carefully considered. Moreover, political parties and candidates should take seriously the suggestion that they focus their electioneering efforts on wooing young voters.

15 16

as pointless as those to the European Parliament, as we seem to see happening between congressional and presidential elections in the United States (see Chapter 4). A change in the electoral system to one yielding fewer wasted votes would be even better. These reforms are also required for reasons that have nothing to do with low voter turnout. Procedures that permit politicians to pervert the electoral process should not be tolerated in any self-respecting democracy.

appendix a

The Surveys Employed in This Book

This appendix contains a brief account of the surveys that formed the basis for the datasets used in Chapters 3 and 6. Because of space considerations, we give the minimum information needed to identify the surveys, credit their Principal Investigators if credited by their national archive, and locate their documentation. In addition, we provide the number of cases included in our file, which may be less than the number of cases contained in the corresponding archived dataset because we only analyzed data from respondents whose ages were known. The American national election studies (documentation available from the Inter-university Consortium for Political and Social Research at the University of Michigan). American National Election Study 1964; N = 1769; University of Michigan Survey Research Center (ICPSR 7235). American National Election Study 1968; N = 1552; University of Michigan Political Behavior Program (ICPSR 7281). American National Election Study 1972; N = 2688; Warren Miller, Arthur Miller, et al. (ICPSR 7010). American National Election Study 1976; N = 2234; Warren Miller, Arthur Miller (ICPSR 7381). American National Election Study 1980; N = 1612; Warren Miller, Center for Political Studies (ICPSR 7763). 225

226

Appendix A

American National Election Study 1984; N = 2237; Warren Miller, National Election Studies (ICPSR 8298). American National Election Study 1988; N = 2037; Warren Miller, National Election Studies (ICPSR 9196). American National Election Study 1992; N = 2485; Warren Miller, Donald Kinder, Steven Rosenstone, National Election Studies (ICPSR 9549). American National Election Study 1996; N = 1712; Steven Rosenstone, Donald Kinder, Warren Miller, National Election Studies (ICPSR 6896).

The British national election studies (documentation available from the Economic and Social Research Council Data Archive at the University of Essex). Political Change in Britain 1963–1970, 1964 electorate sample; N = 1769; D. Butler, D. E. Stokes (DA1091). Political Change in Britain 1963–1970, 1966 electorate sample; N = 1874; D. Butler, D. E. Stokes (DA1092). Political Change in Britain 1963–1970, 1970 pre-, post-election panel; N = 1843; D. Butler, D. E. Stokes (DA422). British Election study, February 1974, Cross-section survey; N = 2462; I. M. Crewe, B. Sarlvik, D. Robertson, J. Alt (DA359). British Election study, October 1974, Cross-section survey; N = 2365; I. M. Crewe, B. Sarlvik, D. Robertson (DA666). British Election study, May 1979, Cross-section survey; N = 1893; I. M. Crewe, D. Robertson, B. Sarlvik (DA1533). British General Election Study 1983; N = 3955; A. Heath, R. Jowell, J. K. Curtice, E. Field (DA2005). British General Election Study 1987; N = 3826; A. Heath, R. Jowell, J. K. Curtice (DA2568). British General Election Study 1992; N = 3534; A. Heath, R. Jowell, J. K. Curtice, J. A. Brand, J. C. Mitchell (DA2981). British General Election Study 1997; N = 4214; A. Heath, R. Jowell, J. K. Curtice, P. Norris (DA3887).


227

The Dutch national election studies (English-language documentation available from the Steinmetz Archive, University of Amsterdam). National Election Study 1971 (pre-, post-); N = 1261; P. C. Stouthard, F. Heunks, J. Thomassen, W. E. Miller, J. Rusk (P0136). Dutch Parliamentary Election Study 1972 (pre-, post-); N = 1526; L. P. J. de Bruyn and J. W. Foppen (P0136). Dutch Parliamentary Election Study 1977 (pre-, post-); N = 1856; NKO, G. A. Irwin, J. Verhoef, C. J. Wiebrens (P0354). Dutch Parliamentary Election Study 1981 (pre-, post-); N = 2305; NKO, C. van der Eijk, C. Niemoller, A. Th. J. Eggen (P0978). Dutch Parliamentary Election Study 1982 (pre-, post-); N = 1541; NKO, C. van der Eijk, C. Niemoller, A. M. J. Koopman (P0978). Dutch Parliamentary Election Study 1986 (panel and fresh sample); N = 3260; NKO, C. van der Eijk, G. A. Irwin, C. Niemoller (P0866 + P0999). Dutch Parliamentary Election Study 1989 (pre-, post-); N = 1754; SKON, H. Anker, E. V. Oppenbuis (P1209). Dutch Parliamentary Election Study 1994 (pre-, post-); N = 1812; SKON, H. Anker, E. V. Oppenbuis (P1209). Dutch Parliamentary Election Study 1998 (pre-, post-); N = 2101; C. Aarts, H. van der Kolk, M. Kamp (P1415). The German national election studies (English-language documentation available at the Central Archive for Social Research at the University of Colone). Federal Parliament Election 1965, Post-election survey, October and November 1965; N = 1305; DIVO (ZA0314). Federal Parliament Election 1969, Pre- and post-election survey (panel), October and November 1969; N = 1158; H.-D. Klingemann and F. U. Pappi (ZA0426). Election Study 1972, Pre- and post-election survey, September– December 1972; N = 2052; M. Berger, W. G. Gibowski, M. Kaase, D. Roth, U. Schleth, R. Wildenmann (ZA0635-7). Election Study 1976, Pre- and post-election survey; N = 2076; M. Berger, W. G. Gibowski, E. Gruber, D. Roth, W. Schulte, M. Kaase, H.-D. Klingemann, R. Wildenmann, U. Schleth (ZA0823-5).

228

Appendix A

Election Study 1980 (trend), Post-election component, November 1980; N = 1001; M. Berger, W. G. Gibowski, E. Gruber, D. Roth, W. Schulte, M. Kaase, H.-D. Klingemann, R. Wildenmann, U. Schleth (ZA1053). Election study 1983 (panel); N = 1622; M. Berger, W. G. Gibowski, D. Roth (ZA1276). Election Study 1987, Pre- and post-election survey (panel); N = 1954; M. Berger, W. G. Gibowski, E. Gruber, D. Roth, W. Schulte, M. Kaase, H.-D. Klingemann, F. U. Pappi, M. Kuchler (ZA1533). Election Study 1990 (panel); N = 2056; M. Kaase, H.-D. Klingemann, M. Kuchler, F. U. Pappi, H. A. Semetko (ZA1915). Post-Election Study 1994; N = 2046; WZB/ZUMA (ZA2601). Election Study 1998 (pre-, post-), N = 2010; MZES/WZB/ ZA/ZUMA (ZA3073). The Norwegian national election studies (English-language documentation available at the Norwegian Social Science Data Service at the University of Bergen. Study numbers are not available). Election Study 1965, pre-, post- (panel); N = 1623; H. Valen. Election Study 1969, pre-, post- (panel); N = 1595; H. Valen. Post-Election Study 1973; N = 1225; H. Valen, W. Martinussen. Post-Election Study 1977; N = 1730; H. Valen, SSB. Post-Election Study 1981; N = 1596; H. Valen, B. Aardal, SSB. Post-Election Study 1985; N = 2180; H. Valen, B. Aardal, SSB. Post-Election Study 1989; N = 2195; H. Valen, B. Aardal, SSB. Post-Election Study 1993; N = 2194; H. Valen, B. Aardal, SSB. Post-Election Study 1997; N = 2055; H. Valen, B. Aardal, SSB. The Swedish national election studies (English language documentation available at the Swedish Social Science Data Service at Gottenburg University). Election Study 1964 (trend + panel); N = 3109; B. Sarlvik, Statistics Sweden (SSD0007). Election Study 1968 (trend + panel); N = 3514; B. Sarlvik, Statistics Sweden (SSD0039).


229

Post-Election Study 1970; N = 4194; B. Sarlvik, Statistics Sweden (SSD0047). Election Study 1973 (trend + panel); N = 2596; B. Sarlvik, O. Peterson, Statistics Sweden (SSD0040). Election Study 1976 (trend + panel); N = 2686; O. Peterson, Statistics Sweden (SSD0008). Election Study 1979 (trend + RoPanel); N = 3758; S. Holmberg, Statistics Sweden (SSD0089). Election Study 1982 (trend + RoPanel); N = 3763; S. Holmberg, Statistics Sweden (SSD0157). Election Study 1985 (trend + RoPanel); N = 3907; S. Holmberg, M. Gilljam, Statistics Sweden (SSD0217). Election Study 1988 (trend + RoPanel); N = 3971; S. Holmberg, Statistics Sweden (SSD0227). Election Study 1991 (trend + RoPanel); N = 3747; S. Holmberg, Statistics Sweden (SSD0391). Election Study 1994; N = 3378; S. Holmberg, Statistics Sweden (SSD0570). Election Study 1998; N = 2931; S. Holmberg, Statistics Sweden (SSD0750).

appendix b

Aggregate Data for Established Democracies, 1945–1999

This appendix contains a brief account of the variables included in the dataset used for the analysis of elections at the country level in Chapters 5 and 7, including sources and references to scholarly work that employed the same variables and algorithms/formulas employed for derived variables. Variables with no source given were derived or calculated from Mackie and Rose (1991) and updates in the annual data issues of the European Journal of Political Research. Absentee – generally postal – voting (Franklin 1996, 2002). Our variable from Katz (1997), supplemented by the record of recent elections published in Electoral Studies, Vols. 16–20. Age – percent in various age groups – (Powell 1986; Gray and Caul, 2000). Our variable from United Nations Department of Economic and Social Affairs (1951–1998).∗ Coalition – whether the election outcome led to a coalition government. New variable from Banks (1997).+ Compulsory voting – whether a sanction is applied for failure to vote – (virtually everyone). Netherlands to 1970, Italy to 1993, Australia, Luxembourg. ∗ +

Gives rise to more than one variable through different operationalizations. The bestperforming version was chosen. Included in the Cross-National Time-Series Data archive (http://www.databanks. sitehosting.net).

231

232

Appendix B

Cohesive – whether effective party discipline is maintained (new variable from Powell 2000).+ Decisive – power of lower House shared with upper House or president (Jackman 1987; Blais 2000). Disproportionality – mean difference between percent seats and percent votes – Rae index (almost everyone).∗ District magnitude – seats – (Powell 1986; Jackman 1987; Jackman and Miller 1995; Radcliff 1996).∗ Electoral system – dummy variables – (Blais 2000; Gray and Caul 2000). Our variables from Katz (1997).∗ Electorate size – absolute, in millions (Franklin 2002). Electorate size – percent of maximum (variant on Radcliff 1992). GNP in constant $ per capita (Radcliff 1992; Blais 2000; Gray and Caul 2000). Our variable from Banks (1997).+ Executive responsiveness – whether legislature can dismiss the executive – (new variable from Banks, 1997).+ Majority status – absolute difference between size of largest party and 50 percent – (Franklin 2002). Margin of victory – percent – (Blais 2000; Gray and Caul 2000; Franklin 2002). Mean margin of victory – percent – (new variable coded for plurality elections in Britain, New Zealand, and France since 1958 – Canada coded 0 due to lack of data). Our variable from Caramani (2000), Federal elections Commission Web site, ICPSR Study # 7577, Jack Vowles. Number of parties in legislature (Crepaz 1990; Jackman and Miller 1995; Blais 2000). Party attachment – three-point scale (Dalton 1999). Our variable from election studies (see Appendix A). Party–group linkages – percent of variance in left voting explained by social structure – (Powell 1986). Our variable calculated as in Franklin, van der Eijk, and Oppenhuis (1995).∗ Percent illiterate (100-literacy) – percent of adult population – (Blais 2000). Our variable from Banks (1997).+ ∗ +

Gives rise to more than one variable through different operationalizations. The bestperforming version was chosen. Included in the Cross-National Time-Series Data archive (http://www.databanks. sitehosting.net).

Aggregate Data for Established Democracies

233

Percent university (tertiary) educated (Gray and Caul 2000). Our variable from Banks (1997).+ Polarization – distance between extreme parties from manifesto data – (Crepaz 1990). Our data from Party Manifestos Project data (Klingemann et al. 1944).++ Population density – population per square mile – (Blais 2000). Our variable from Banks (1997).+ Population size – (Blais 2000; Gray and Caul 2000). Our variable from Banks (1997).∗+ Rest day voting (Franklin 1996, 2002; Gray and Caul 2000). Time since most recent election of same type (Franklin 2002). Trust in government (officals) – (Dalton 1999). Our variable from election studies (see Appendix A). Unicameralism (Jackman 1987; Jackman and Miller 1995; Radcliff 1996). Union membership (Gray and Caul 2000). Our variable from Golden et al. (1998), Heikki (1993), ICFTU (1947–2000), and other sources.∗ Voting age – in years – (Jackman and Miller 1995; Radcliff 1996). Derived Variables The following interactions involve the proportion of the electorate that is new or that is facing one of its first three elections: register – registert−1 + registert−1 ∗ gap/50; register/50 if first election [where gap is years since previous election]. youngt−1 + new if voting at eighteen [where Young initiation: new is as above]. New cohorts proportion: new + newt−1 + newt−2 ; new ∗ 2 + newt−1 if second election. [This is used for making short-term interactions.]

New voters proportion:

∗ + ++

Gives rise to more than one variable through different operationalizations. The bestperforming version was chosen. Included in the Cross-National Time-Series Data archive (http://www.databanks. sitehosting.net). Variable provided by Hans-Dieter Klingemann. See Klingemann et al. (1994).

Appendix B

234

Cumulative new voters:

new + newt−1 [where new is as above and where indicates summation across remaining elections for the country]. [Used for long-term district magnitude and electorate size.]

The following interactions involve the proportion of the electorate that has been of voting age for a given proportion of the fifty-year average span of membership in an electorate (new cohorts are presumed to comprise 12/50 of an electorate, assuming elections every four years on average): Weight for executive responsiveness: (election–1963)/50 + 12/50 if Switzerlad and >1962; (election–1962)/50 + 12/50 if Malta and >1961; (election– 1945)/50 + 12/50 if France and 1970 (election–1994)/50 + 12/50 if Italy and >1993; else 1 Long-term compulsory voting: (compulsory–1) ∗ weight [where weight is as above] Weight for absentee voting: (election–1972)/50 + 12/50 if Canada and >1971; (election– 1970)/50 + 12/50 if Finland and >1969; (election–1995)/50 + 12/50 if Switzerland and >1994; – (election–1978)/50 – 12/50 if France and >1977; else 1

Aggregate Data for Established Democracies Long term absentee voting: Female empowerment:

235

(absentee–0.5) ∗ weight [where weight is as above] ((election-franchise)/50 + 12/50)/2 if >(franchise-1) [where franchise is year women gained the franchise]

> means “after year specified”; < means “before year specified.”

appendix c

Supplementary Findings

237

99

Voting age lowered

Australia Actual turnout r = 0.087

Predicted turnout 41 1946

election

1998

99

Women

Voting age

enfranchised

lowered

Belgium Actual turnout r = 0.422*


election

1999

99

Voting age lowered

Denmark

Actual turnout Predicted turnout

r = 0.481*

41 1945

election

1998

99

Women enfranchised

Electoral system reformed

Voting age lowered

Actual turnout

France

Predicted turnout

r = 0.714**

41 1945

election

1997

figure c.1. Expected and actual turnouts for Australia to Germany.

99

Compulsory voting abolished in Presidential elections Voting age lowered

Austria Actual turnout r = 0.793***


election

1995

99 Voting age lowered

Canada Actual turnout Predicted turnout

r = 0.503**

41 1945

election

1997

99

Voting age lowered

Actual turnout

Finland

Predicted turnout r = 0.850*** 41 1945

election

1999

99

Voting age lowered

Germany


r = 0.779***

41 1949

figure c.1 (continued)

election

1998

99

Voting age lowered

Iceland Actual turnout Predicted turnout r = 0.763*** 41 1946

1999

election

99

Israel


r = 0.022 41 1949

election

1999

99

Women

Japan

enfranchised


41 1946


r = 0.588** election

1996

99

Voting age Independence

lowered

election

Women

Malta

enfranchised


r = 0.970***

41 1947

election

1998

figure c.2. Expected and actual turnouts for Iceland to The Netherlands.

99

Voting age lowered

Ireland


r = 0.820*** 41 1948

1997

election

99

Voting age

Women

lowered

enfranchised

Compulsoy voting

Italy


abolished

r = 0.954*** 41 1946

election

1996

99

Voting age lowered

Luxembourg Actual turnout Predicted turnout

41 1945

r = 0.806***

election

1999

99

Voting age lowered Compulsory voting abolished

Netherlands

Actual turnout r = 0.875***



election

1998

Appendix C

242 99


Voting age lowered

New Zealand Actual turnout r = 0.622**


election

1999

99

Voting age lowered

Sweden r = 0.694**

Actual turnout Predicted turnout 41 1945

election

1997

99

Voting age lowered

United Kingdom r = 0.408


41 1945

election

1997

figure c.3. Expected and actual turnouts for New Zealand to the United States.


243

99

Voting age lowered

Norway Actual turnout r = 0.438*


1997

election

99

Voting age lowered

Switzerland r = 0.996***

"Golden rule" introduced Women


enfranchised

41 election

1947

1999

99

Voting age lowered

United States r = 0.897***


41 1948


election

1996

244

1. New voters difference 2. Second cohort change 3. Previous (third) cohort change 4+ All prior cohorts change

Norweigan averages

1. New voters difference 2. Second cohort change 3. Previous (third) cohort change 4+ All prior cohorts change 1965

1977 15.84 24.51 9.18 4.01

1973

−8.39 −23.29 −1.26 −6.46 −5.48 −7.28 2.55 4.02

1969

−6.37 3.47 −8.00 2.33

9.22 −14.29 6.62 −1.81 0.91 −1.03 6.55 4.45

8.40 10.47 2.73 4.10

1981

7.84 6.05 5.76 2.91

1983

1990

−6.82 −2.90 2.30 3.44

1985

1.14 3.76 6.08 3.93

1989

1997

1994

9.81 2.31 16.96 5.65

1994

−5.10 −5.79 −2.49 3.77

1993

9.05 7.61 5.61 3.81

Mean Mean absolute −3.43 −2.71 5.75 3.51 9.33 1.80 4.68

1997

12.43 7.50 9.39 4.96

Mean Mean absolute 26.48 1.37 13.78 1.84 −8.68 −1.55 5.50

1998

9.03 9.84 8.59 6.33

Mean Mean absolute −9.19 0.65 −7.79 −3.28 −6.17 4.03 7.47

1998

7.19 6.89 5.72 4.36

Mean Mean absolute

2.12 −15.89 −3.22 6.50 −12.07 1.13 7.13 −9.63 3.01 3.63 7.32

1992

−5.92 6.83 −22.47 0.28 −23.77 −4.21 −3.85 −20.12 26.78 5.95 5.91 6.07

1987

4.12 −4.73 −10.19 −6.57 −19.52 −13.08 −8.78 6.29 7.89 6.34 9.02 6.73

1989

−0.56 6.56 10.13 5.52

1987

−9.40 −5.51 −3.31 2.66

1986

3.42 2.23 −1.20 3.32

1983

1982

−5.50 2.46 1.90 2.10

1979

1981

1.92 4.47 13.21 8.46

1974

1980

6.97 3.45 4.44 3.94

1977

8.40 9.89 1.10 1.36

1970

1976

1972

1969

German averages

1972

−19.67 −10.98 1.46 3.16

1966

17.81 20.48 14.90 8.86

1971

1964


Dutch averages


British averages

table c.1 Cohort Effects in Six Countries (for Table 3.3 in Chapter 3)

245

−4.97 1.24 −0.58 2.74

−0.02 3.45 3.48 4.35


3rd

2nd −1.19 5.74 0.93 3.69

4th

−6.10 8.50 −8.50 3.70

−6.60 5.40 5.30 3.40

Overall averages

1976

−3.00 −1.00 3.00 1.00

1982

1972

−1.00 −3.00 −6.00 2.29

−6.00 −2.00 1.00 1.33 1968

1979

1976

6.90 7.80 8.10 3.80

1964

1973


U.S. averages


Swedish averages

1991

1984

1988

−4.00 −2.00 −7.00 1.00 −4.00 7.00 3.20 2.91

1988

3.03 1.02 −1.36 2.87

5th

7th

−4.81 −1.41 −4.38 −3.76 1.89 1.16 4.96 3.81

6th

5.30 −10.80 −3.70 −2.30 0.60 3.00 −3.80 11.80 −4.00 6.30 5.80 1.20

1980

−2.00 −4.00 −6.00 1.33

1985

1.49 2.95 12.20 4.60

8th

13.60 13.90 18.80 8.10

1992

11.00 5.00 6.00 2.25

1994

8.46 7.13 7.08 4.47

Mean Mean absolute −4.21 −0.99 −3.97 0.99 −4.56 2.04 3.56

9th

8.10 7.20 7.80 5.04

Mean Mean absolute

−12.20 −1.70 −16.50 2.55 −2.20 3.19 7.90

1996

5.00 3.75 5.38 2.33

Mean Mean absolute

−11.00 −0.31 −7.00 0.15 −10.00 1.77 4.31

1998

Appendix C

246 table c.2 Logistic Regression Results for Table 6.2, Model C

Constant Age Education Urban Respondent’s occupation Household head’s occupation Religion Church attendance Political interest Strength of party attachment Party sympathy New ∗ gender New ∗ education New ∗ urban New ∗ respondent’s occupation New ∗ household head’s occupation New ∗ religion New ∗ church attendance New ∗ strength Established ∗ majority status New ∗ majority status New ∗ margin Young initiation voter Voted at previous election

B

S.E.

−3.861 −0.793 0.675 0.417 1.383 0.605 0.398 0.622 1.354 0.616 3.506 −0.542 0.647 −0.575 −0.097 1.560 2.578 −0.197 0.216 −0.718 −1.961 −3.140 −0.765 1.620

0.463 0.347 0.151 0.120 0.167 0.219 0.296 0.144 0.200 0.107 0.283 0.163 0.145 0.235 0.351 0.345 0.480 0.073 0.095 0.179 0.428 0.430 0.147 0.108


247

table c.3 Logistic Regression Results for Table 6.3, Model E B Constant Age Established ∗ education Established ∗ urban Established ∗ respondent’s occupation Established ∗ household head’s occupation Established ∗ religion Established ∗ church Political interest Strength of party attachment Party sympathy New ∗ gender New ∗ education New ∗ urban New ∗ respondent’s occupation New ∗ household head’s occupation New ∗ religion New ∗ church attendance New ∗ strength Established ∗ majority status New ∗ majority status New ∗ margin Young initiation voter Voted at previous election

−3.836 −0.789 0.758 0.402 1.377 0.601 0.380 0.600 1.339 0.618 3.520 −0.541 1.027 −0.177 1.261 2.191 2.747 −0.237 0.224 −0.737 −2.676 −3.154 −0.771 1.621

S.E. 0.461 0.348 0.155 0.119 0.167 0.219 0.295 0.147 0.200 0.107 0.283 0.162 0.125 0.203 0.311 0.276 0.438 0.072 0.095 0.179 0.414 0.431 0.147 0.108

Converting Logistic Regression Coefficients to Differences in Proportions This is the procedure employed to create Tables 6.2 and 6.3 from the above logistic regression findings. It is moderately simple if done in a spreadsheet, more complex by hand. The raw material are the effects (b1,b2, . . . bk) and constant term (b0) from the logistic regression output and the means of the independent variables (X1,X2, . . . Xk). If you are using SPSS, you need to get the means from a separate DESCRIPTIVES procedure. Make sure that you only include the same cases as are included in the logistic regression, which performs listwise deletion of missing data. Thus you need to specify /MISSING = LISTWISE on the DESCRIPTIVES procedure.

248

Appendix C

Calculate the expected value of Y when all variables are at their mean values by using a spreadsheet to compute b0 + b1X1 + b2X2 + · · · + bkXk. We will call this value bX. Each difference in proportion requires you to calculate two proportions: one for the Y when the X is 1 and the other for the Y when the X is 0. When you subtract the proportion when X = 0 from the proportion when X = 1, the result is the OLS equivalent you are seeking. The two equations needed to generate these two proportions are as follows: FOR X1 = 1: Y = 1/(1 + EXP(−(bX + (1 − mean(Xi)) ∗ b1))) FOR X1 = 0: Y = 1/(1 + EXP(−(bX + (0 − mean(Xi)) ∗ b1))). EXP is short for “exponentiate” (the anti-log of the log normal). Every spreadsheet that I have used implements this function as EXP. Mean is simply the mean of each variable (X1 in this case) derived from the DESCRIPTIVE procedure, so mean(X1) is a number, not a function (this will be a different number for X2, X3, etc.). Remember to use /MISSING = LISTWISE. bX is the predicted Y when all Xs are at their mean values, as explained above (stays the same for X2, X3, etc.). b1 is the logistic effect (b) for the first independent variable. You will need to plug in values for b2, b3, etc., to get all the OLS equivalents. Make a spreadsheet with one row for each variable and columns for the following: variable name, mean(Xi), b(Xi), mean(Xi) ∗ b(Xi), Y(Xi = 1), Y(Xi = 0), Y(Xi = 1)-Y(Xi = 0). Your bX for the Y(Xi = 0/1) can be calculated at the foot of the mean(Xi) ∗ (b(Xi)) column with a formula that sums the column and adds b0 (the constant from the logistic regression output).

249

Predicted turnout change (sum of cells above) Actual turnout change Error in prediction

Time since previous election (0–7) Short-term majority status (0–17) Short-term margin of victory (0–10) Short-term mean margin (0–10) Short-term cohesiveness (0–0.6) Cumulative compulsory voting (−1 to 0) Cumulative executive responsiveness (−0.5 to 0.5) Cumulative female empowerment (0 to 0.5) Cumulative electorate size (millions) Cumulative proportion of young initiation (0 to 0.82) Previous turnout (41 to 98) −7.251 −1.314 −8 −1 0.749 −0.314

−7.923 0 1.358 −3.927

1.012

0.000

−8.154 −8.385

−5.096 −1.437

−2.847 −2.415

−0.194 −1.603

0.000

0.000

0.000 −0.130 0.362 0.647 0.000 0.000

0.000 −0.644 −0.855 −1.445 0.477 −0.155

−9 −3 −20 −13 1.306 0.543 11.846 4.615

−7.694 −2.457

−3.267 −0.915

−3.920 −0.392

0.000

0.410 Average −6.565 −3.927

0.000

−0.942 −0.161

0.000

0.000

−1.865 −1.859

1.370

0.000

−0.069 −1.920

0.000

0.000

0.000 0.000 1.719 −0.205 0.000 0.000

−3.808 −2.512

0.000

2.502

0.000

0.000 1.997 0.000

−0.199 −0.303

0.000

4.054

0.000 0.118 0.000

1.287 1.287 −1.193 −0.024 −2.145 0.706

−0.044 −0.606

0.000 0.871 0.000

−0.571 7.042 11.252

−1.929 0.000 −0.802 −2.526 −0.451 0.460

Australia Austria Belgium Canada Denmark Finland France

0.636 −0.643 −0.409 −0.463 −0.353 −0.574

b

table c.4 Predicted and Actual Turnout Change, 1960s to 1990s, Australia to Israel (for Table 7.3) Israel

0.000

0.000

0.000 1.666 0.000

0.000

0.000

0.000 1.248 0.000

0.000

−0.242

0.000

0.000

0.000 2.555 0.000

−5 −7.3 −9 −4.513 3.282 4.621

−9.513 −4.018 −4.379

−7 2.810

−4.190

−3.659 −1.359 −0.784 −1.176

−2.949 −2.271 −2.941

−1.858 −0.010 −0.116

0.000

0.000

0.000 1.674 0.000

0.000 0.000 0.644 −0.644 −0.624 −0.648 −1.113 −5.038 −2.097 −1.396 −0.672 −0.289

Germany Iceland Ireland

250 −0.009 −2.594 −1.829 −6.478 −4 −2.478

0.000 0.000 0.000 1.132

−3.369 0.000 1.191 −1.918 −4.318 0.000

0.000

−2.405 −2.744 −1.176 −9.230 −5.986

Average −6.565 −7.923 −10 −9 1.358 0.770 3.014 3.039

Predicted turnout change (sum of cells above) Actual turnout change Error in prediction Average absolute error in prediction (both subtables)

0.000

0.000

0.000

0.000

5 0.104

5.104

3.528

−2.882

−0.013

1.251

2.587

0.000

0.987

−2.366 0.000

0.000

0.000

−0.248

−0.700

0.956

−0.105

1.020

1.349

−1.335 −0.650

−22 2.959

−19.041

−6.272

−2.830

−0.549

0.000

0.000

−9.384

0.603

0.000

−0.207

−0.402

−3 −0.453

−3.453

−1.176

−2.956

−0.125

0.000

0.000

0.000

0.964

1.518

−0.836

−0.842

0.000

0.000

0.000

0.110

0.000

1.539

−1.278

1.176

−7 −3 3.568 2.119

−3.432 −0.881

−0.131

−2.256 −2.190

−0.140 −0.238

0.000

0.000

0.000

0.600

0.000

−0.371

−1.135

0.000

−0.715

−21 −2.686

−23.686

−9.016

−1.225

−6 −0.415

−6.415

−0.261

−2.225

−1.636

0.000

−9.921

−0.269

0.000

−10 −0.119

−10.119

−1.921

−2.294

−5.376

0.000

0.000

0.000

0.000

−1.622

−1.718 0.789

0.848

0.245

United States 0.000

−1.034

−0.685

0.000

0.000

0.000

0.000

−2.540

The New United Italy Japan Luxembourg Malta Netherlands Zealand Norway Sweden Switzerland Kingdom −1.931 −1.929 0.000 −1.287 0.000 0.000 0.000 0.000 0.000 1.931

Time since previous election (0–7) Short-term majority status −0.409 (0–17) Short-term margin of victory −0.353 (0–10) Short-term mean margin −0.571 (0–10) Short-term cohesiveness 7.042 (0–0.6) Cumulative compulsory voting 11.252 (−1 to 0) Cumulative executive 4.054 responsiveness (−0.5 to 0.5) Cumulative female 2.502 empowerment (0 to 0.5) Cumulative electorate size −0.044 (millions) Cumulative proportion of −3.808 young initiation (0 to 0.82) Previous turnout (41 to 98) 0.410

b 0.636

table c.5 Predicted and Actual Turnout Change, 1960s to 1990s, Italy to United States (for Table 7.3)

Bibliography

Abramson, Paul, John Aldrich, Phil Paolino, and David Rohde. 1992. “Sophisticated Voting and the 1988 Presidential Primaries,” American Political Science Review 86: 55–69. Achen, Christopher. 2000. “Why Lagged Dependent Variables Can Suppress the Explanatory Power of Other Independent Variables,” paper presented at the annual meeting of the Political Methodology Section of the American Political Science Association, Washington, DC. Aldrich, John H. 1995. Why Parties? The Origin and Transformation of Political Parties in America. Chicago: University of Chicago Press. Alvarez, R. Michael, and Jonathan Nagler. 1995. “Economics, Issues and the Perot Candidacy: Voter Choice in the 1992 Presidential Election,” American Journal of Political Science 39: 714–44. American Political Science Association Committee on Political Parties. 1950. “Toward a More Responsible Two-Party System,” American Political Science Review 44 (3–Supplement): 1–99. Andersen, Kristi. 1979. The Creation of a Democratic Majority, 1928–1936. Chicago: University of Chicago Press. Asher, Herbert B. 1992. Presidential Elections and American Politics: Voters, Candidates and Campaigns Since 1952. Pacific Grove, CA: Brooks/Cole. Banks, Arthur S., ed. 1975–1997. Political Handbook of the World. New York: McGraw-Hill. Beck, Nathaniel. 2001. “Time-Series–Cross-Section Data: What Have We Learned in the Past Few Years?” Annual Review of Political Science 4: 271– 293. Beck, Nathaniel and Jonathan N. Katz. 1995. “What To Do (and What Not To Do) With Time-Series – Cross-Section Data,” American Political Science Review 89: 634–647.

251

252

Bibliography

Beck, Paul Allen and Frank J. Sorauf. 1992. Party Politics in America, Seventh Edition. New York: Harper Collins. Berelson, Bernard R., Paul R. Lazarsfeld, and William N. McPhee. 1954. Voting: A Study of Opinion Formation in a Presidential Campaign. Chicago: University of Chicago Press. Beetham, David, et al., eds. 2002. International IDEA Handbook on Democracy Assessment. New York: Kluwer Law International. Blais, Andre and Richard Nadeau 1996. “Measuring Strategic Voting: A TwoStep Procedure,” Electoral Studies 15: 39–52. Blais, André, Elizabeth Gidengil, Neil Nevitte, and Richard Nadeau. 2001. “The Evolving Nature of Non-Voting,” paper delivered at the annual meeting of the American Political Science Association, San Francisco, CA. Blais, André and Kenneth Carty. 1990. “Does Proportional Representation Foster Voter Turnout?,” European Journal of Political Research 18: 167–181. Blais, Andre. 2000. To Vote or Not to Vote: The Merits and Limits of Rational Choice Theory. Pittsburgh, PA: University of Pittsburg Press. Blais, André and Agnieszka Dobrzynska. 1998. “Turnout in Electoral Democracies,” European Journal of Political Research 33: 239–261. Blondel, Jean, Richard Sinnott, and Palle Svensson. 1998. People and Parliament in the European Union. Oxford: Oxford University Press. Bloom, Jonathan. 2002. “Cambridge Gives 17-Year-Olds the Vote,” Boston Globe, 26th March. Boechel, Richard. 1928. Voting and Non-Voting in Elections. Washington DC: Editorial Research Reports. Bowler, Shaun and Todd Donovan. 1998. Demanding Choices: Opinion, Voting and Direct Democracy. Ann Arbor: University of Michigan Press. Boyd, Richard. 1981. “The Decline of U.S. Voter Turnout: Structural Explanations,” American Politics Quarterly 9: 133–159. Brady, David W. 1988. Critical Elections and Congressional Policy Making. Stanford, CA: Stanford University Press. Brady, Henry, Sidney Verba, and Kay Schlozman. 1995. “Beyond SES: A Resource Model of Political Participation,” American Political Science Review 89: 271–294. Brody, Richard A. 1978. “The Puzzle of Political Participation in America,” in The New American Political System, ed. Anthony King. Washington DC: American Enterprise Institute, pp. 287–324. Brug, Wouter van der, Cees van der Eijk, and Mark Franklin. 2003. “Designs for the Empirical Analysis of Electoral Preferences, Utilities and Choice,” paper presented at the ECPR Joint Sessions Workshops, Edinburgh. Burnham, Walter Dean. 1974. “Theory and Voting Research: Some Reflections on Converse’s ‘Change in the American Electorate’,” American Political Science Review 68: 1002–23. Burnham, Walter Dean. 1965. “The Changing Shape of the American Political Universe,” American Political Science Review 59: 7–28.

Bibliography

253

Burnham, Walter Dean. 1970. Critical Elections and the Mainsprings of American Politics. New York: Norton. Burnham, Walter. 1986. “Those High Nineteenth Century American Voter Turnout Figures: Fact or Fiction?” Journal of Interdisciplinary History 16: 613–644. Burnham, Walter. 1987. “The Turnout Problem” in Elections American Style, ed. A. James Reichley. Washington DC: Brookings Institution, pp. 97– 133. Burns, James MacGregor. 1963. The Deadlock of Democracy; Four-Party Politics in America. Englewood Cliffs, NJ: Prentice-Hall. Butler, David and Donald Stokes. 1969, 1974. Political Change in Britain. London: Macmillan. Cain, Bruce, John Ferejohn, and Morris Fiorina. 1987. The Personal Vote: Constituency Service and Electoral Independence. Cambridge, MA: Harvard University Press. Cain, Bruce. 1978. “Strategic Voting in Britain,” American Journal of Political Science 22: 639–655. Campbell, Angus. 1960. “Surge and Decline: A study of Electoral Change,” Public Opinion Quarterly 24: 397–418. Campbell, Angus, Philip Converse, Warren Miller, and Donald Stokes. 1960. The American Voter. New York: Wiley. Campbell, James. 1985 “Sources of the New Deal Realignment: The Contributions of Conversion and Mobilization to Partisan Change,” Western Political Quarterly 38: 357–376. Caramani. Daniele. 2000. Elections in Western Europe Since 1815: Election Results by Constituencies. London: Macmillan Reference (accompanying CD-ROM). Christy, Carol. 1987. Sex Differences in Political Participation: Processes of Change in Fourteen Nations. New York: Praeger. Clarke, Harold, Marianne Stewart, and Paul Whiteley. 2002. “Downs, Stokes and Modified Rational Choice: Modeling Turnout in 2001,” paper presented to the American Political Science Association annual meeting, Boston, MA. Cohen, Jacob and Patricia Cohen. 1983. Applied Multiple Regression Analysis for the Behavioral Sciences, Second Edition. Hillsdale, NJ: Lawrence Erlbaum Associates. Converse, Philip E. 1972. “Change in the American Electorate,” in The Human Meaning of Social Change, eds. Angus Campbell and Philip E. Converse. New York: Russell Sage Foundation. Converse, Philip. 1974. “Comment on Burnham’s ‘Theory and Voting Research’,” American Political Science Review 68: 1024–1027. Crepaz, Markus 1990. “The Impact of Party Polarisation and Postmaterialism on Voter Turnout,” European Journal of Political Research 18: 183–205. Dahl, Robert and Edward Tufte. 1973. Size and Democracy. Stanford, CA: Stanford University Press.

254

Bibliography

Dalton, Russell. 1999. “Political Support in Advanced Industrial Democracies” in Critical Citizens, ed. Pippa Norris. Oxford: Oxford University Press, pp. 57–77 Danesi, Martin. 2002. The Puzzle Instinct: The Meaning of Puzzles in Human Life. Bloomington: Indiana University Press. Downs, Anthony. 1957. An Economic Theory of Democracy. New York: Harper and Row. Egmond, Marcel van. 2003. “Influence of Political Context on Electoral Participation,” paper presented at the European Consortium for Political Research Joint Sessions Workshops, Edinburg. Egmond, Marcel van, Nan Dirk de Graaf, and Cees van der Eijk. 1998. “Electoral Participation in the Netherlans: Individual and Contextual Influences,” European Journal of Political Research 34: 281–300. Eijk, Cees van der. 2002. “Design Issues in Electoral Research: Taking Care of (Core) Business,” Electoral Studies 21: 189–206. Eijk, Cees van der, Mark Franklin, Tom Mackie, and Henry Valen. 1992. “Cleavages, Conflict Resolution and Democracy” in Franklin et al. (1992), pp. 406–431. Eijk, Cees van der and Kees Neimoller. 1983. Political Change in the Netherlands. Amsterdam: CT Press. Eijk, Cees van der and Mark Franklin. 1996. Choosing Europe? The European Electorate and National Politics in the Face of Union. Ann Arbor: University of Michigan Press. Erikson, Robert. 1981. “Why Do People Vote? Because They Are Registered.” American Politics Quarterly 9: 259–276. Erikson, Robert. 1988. “The Puzzle of Midterm Loss,” Journal of Politics 50: 1011–1029. Erikson, Robert and Kent Tedin. 1981. “The 1928–1936 Partisan Realignment: The Case for the Conversion Hypothesis,” American Political Science Review 75: 951–962. Evans, Diana and Mark Franklin. 2001. “Party Competition and Voter Motivation: Turnout in House Elections since 1830,” paper presented at the Midwest Political Science Association annual meeting, Chicago. Finifter, Ada W. 1974. “The Friendship Group as a Protective Environment for Political Deviants,” American Political Science Review 68: 607–625. Finkel, Steven, Edward Muller, and Karl-Dieter Opp. 1989. “Personal Influence, Collective Rationality, and Mass Political Action,” American Political Science Review 83: 885–903. Fiorina, Morris P. 1989a. Congress: Keystone of the Washington Establishment. New Haven, CT: Yale University Press. Flickinger, R. and D. Studlar. 1992. “Exploring Declining Turnout in Western European Elections,” West European Politics 15: 1–16. Fiorina, Morris. 1990. “Information and Rationality in Elections,” in Information and Democratic Processes, ed. John Ferejohn and James Kuklinski. Urbana IL: University of Illinois Press.

Bibliography

255

Fotos, Michael and Mark Franklin. 2002. “Na¨ıve Political Science and the Turnout Paradox,” paper delivered at the annual meeting of the Midwest Political Science Association, Chicago. Fowler, James and Oleg Smirnov. 2002. “A Cellular Automata Model of Elections: A Possible Solution to Puzzles of Voter Turnout and Platform Divergence,” paper prepared for the Midwest Political Science Association annual meeting, Chicago, Illinois, April 2002 (http://www.people. fas.harvard.edu/∼jhfowler/CAelection.doc). Franklin, Daniel P. and Eric E. Grier. 1997. “Effects of Motor Voter Legislation,” American Politics Quarterly 25: 104–117. Franklin, Mark. 1985. The Decline of Class Voting in Britain. Oxford: Oxford University Press. Franklin, Mark. 1991. “Getting Out the Vote: Social Structure and the Mobilization of Partisanship in the 1989 European Elections,” European Journal of Political Research 19: 129–147. Franklin, Mark. 1992. “The Decline of Cleavage Politics,” in Franklin, Mackie, Valen et al. (1992). Franklin, Mark. 1996. “Electoral Participation,” in Comparing Democracies: Elections and Voting in Global Perspectives, eds. Lawrence Leduc, Richard Niemi, and Pippa Norris. Thousand Oaks CA: Sage, pp. 214–233. Franklin, Mark. 1999. “Electoral Engineering and Cross-National Turnout Differences: What Role for Compulsory Voting?” British Journal of Political Science 29: 205–216. Franklin, Mark. 2001. “How Structural Factors cause Turnout Variations at European Parliament Elections,” European Union Politics 2: 309–328 Franklin, Mark. 2002. “The Dynamics of Electoral Participation” in Comparing Democracies 2, eds. L. Leduc, R. Niemi and P. Norris. Thousand Oaks CA: Sage, pp. 148–168. Franklin, Mark. 2003. “Too Much Democracy? How European Parliament Elections Undermine Turnout at National Elections in Europe.” Paper prepared for the biennial congress of the European Consortium for Political Research, Marburg. Franklin, Mark, Cees van der Eijk, and Erik Oppenhuis. 1996. “The Institutional Context: Turnout” in Choosing Europe? The European Electorate and National Politics in the Face of Union, eds. C. van der Eijk, M. Franklin et al. Ann Arbor: University of Michigan Press, pp. 306–331. Franklin, Mark and Diana Evans. 2000. “The Low Voter Turnout Problem,” in The U.S. House of Representatives: Reform or Rebuild? eds. Joseph F. Zimmerman and Wilma Rule. Westport, CT: Praeger, pp. 97–113. Franklin, Mark and Wolfgang Hirczy de Mino. 1998. “Separated Powers, Divided Government and Turnout in US Presidential Elections,” American Journal of Political Science 42: 316–326. Franklin, Mark and Matthew Ladner. 1995. “The Undoing of Winston Churchill: Mobilization and Conversion in the 1945 Realignment of British Voters,” British Journal of Political Science 25: 429–452.

256

Bibliography

Franklin, Mark, Patrick Lyons, and Michael Marsh. 2000. “The Tally of Turnout: Understanding Cross-National Turnout Decline Since 1945,” paper delivered at the annual meeting of the American Political Science Association, Washington, DC, August–September 2000. Franklin, Mark, Patrick Lyons, and Michael Marsh. 2001. “The Tally of Turnout: How the Changing Character of Elections Drives Voter Turnout Variations in Established Democracies,” paper delivered at the Congress of the European Consortium for Political Research, Canterbury, England, September 2001. Franklin, Mark, Richard G. Niemi, and Guy Whitten. 1994. “Two Faces of Tactical Voting,” British Journal of Political Science 24: 549–557. Franklin, Mark and Bernard Wessels. 2002. “Learning (Not) to Vote: The Generational Basis of Turnout Decline in Established Democracies,” paper delivered at the annual meeting of the American Political Science Association, Boston. Franklin, Mark, Tom Mackie, Henry Valen, et al. 1992. Electoral Change: Responses to Evolving Social and Attitudinal Structures in Western Countries. Cambridge: Cambridge University Press. Friedman, Jeffrey, ed. 1996. The Rational Choice Controversy: Economic Models of Politics Reconsidered. New Haven, CT: Yale University Press. Gay, Claudine. 2001. “The Effect of Black Congressional Representation on Political Participation,” American Political Science Review 95: 589–602. Golden Miriam, Michael Wallerstein, and Peter Lang. 1998. Union Centralization among Advanced Industrial Societies (Dataset). Ann Arbor, MI: InterUniversity Consortium for Political and Social Research. Golder, Matt. 2003. “Explaining the Success of Extreme Right Parties in Europe,” Comparative Political Studies 36: 432–466. Gosnell, Harold. 1927. Getting Out the Vote: An Experiment in the Stimulation of Voting. Chicago, IL: University of Chicago Press. Gray, Mark and Miki Caul. 2000. “Declining Voter Turnout in Advanced Industrial Democracies, 1950–1997,” Comparative Political Studies 33: 1091– 1122. Green, Donald P. and Ian Shapiro. 1994. Pathologies of Rational Choice Theory. New Haven, CT: Yale University Press. Greene, William H. 1990. Econometric Analysis. New York: Macmillan. Grofman, Bernard. 1993. “Is Turnout the Paradox that Ate Rational Choice Theory,” in Information, Participation, and Choice: An Economic Theory of Democracy in Perspective, ed. Bernard Grofman. Ann Arbor: University of Michigan Press, pp. 93–103. Groffman, Bernard. 2001. Political Science as Puzzle-Solving. Ann Arbor: University of Michigan Press. Grofman, Bernard, Christian Collet, and Robert Griffin. 1998. “Analyzing The Turnout-Competition Link With Aggregate Cross-Sectional Data,” Public Choice 95: 233–246.

Bibliography

257

Hall, Peter. 2002. “The Role of Government and the Distribution of Social Capital,” in Putnam (2002a), pp. 21–58. Hanks, Charles and Bernard Grofman. 1998. “Turnout in Gubernatorial and Senatorial Primary and General Elections in The South, 1922–90: A Rational Choice Model of The Effects of Short-Run and Long-Run Electoral Competition On Relative Turnout,” Public Choice 94: 407–421. Hanmer, Michael J. 2000. “Alternative Estimates of the Effects of Election Day Registration and Motor Voter Laws,” paper presented at the annual meeting of the Midwest Political Science Association, April 27–30, Chicago, IL. Heath, Anthony and Bridget Taylor. 1999. “New Sources of Abstention?” in Critical Elections: British Parties and Voters in Long-Term Perspective, eds. Geoffrey Evans and Pippa Norris, London: Sage, pp. 164–180. Heikki, Aintila. 1993. Trade Union Membership in Western Europe. Brussels: European Trade Union Institute. Hirczy, Wolfgang. 1992. “Electoral Participation,” University of Houston Ph.D. Dissertation. Ann Arbor, MI: University Microfilms. Hirczy, Wolfgang. 1995. “Explaining Near-universal Turnout: The Case of Malta,” European Journal of Political Research 27: 255–272. Huckfeldt, Robert, Eric Plutzer, and John Sprague, 1993 “Alternative Contexts of Political Behavior: Churches, Neighborhoods, and Individuals,” Journal of Politics 55: 365–381. Huckfeldt, Robert and John Sprague. 1987. “Networks in Context: The Social Flow of Political Information,” American Political Science Review 81: 1197– 1216. Huckfeldt, Robert and John Sprague. 1988. “Choice, Social Structure and Political Information: The Informational Coercion of Minorities,” American Journal of Political Science 32: 467–482. Huckfeldt, Robert and John Sprague. 1991. “Discussant Effects on Vote Choice: Intimacy, Structure, and Interdependence,” Journal of Politics 53: 122–158. ICFTU. 1947–2000. Report of the World Congress. Brussels: International Confederation of Free Trade Unions. Inglehart, Ronald. 1977. The Silent Revolution: Changing Values and Political Styles. Princeton, NJ: Princeton University Press. Inglehart, Ronald. 1990. Culture Shift in Advanced Industrial Society. Princeton, NJ: Princeton University Press. Inglehart, Ronald. 1997. Modernization and Postmodernization: Cultural, Economic, and Political Change in 43 Societies. Princeton, NJ: Princeton University Press. International IDEA. 1997. Voter Turnout from 1945 to 1997. Stockholm: International IDEA. International IDEA. 2001. See Beetham et al. (2002). International IDEA. 2002. See Lopez Pintor, Rafael et al. (2002).

258

Bibliography

Irwin, Galen. 1999. “The Dutch Parliamentary Election of 1998,” Electoral Studies 18: 271–276. Jackman, Robert. 1987. “Political Institutions and Voter Turnout in the Industrial Democracies,” American Political Science Review 81: 405–423. Jackman, Robert and Ross Miller. 1995. “Voter Turnout in the Industrial Democracies During the 1980s,” Comparative Political Studies 27: 467–492. Jennings, M. Kent and Richard Niemi. 1981. Generations and Politics. Princeton, NJ: Princeton University Press. Jennings, M. Kent and Richard Niemi. 1991. The Youth-Parent Socialization Panel Study. Princeton, NJ: Princeton University Press. Keman, Hans. 2002. Comparative Democratic Politics: A Guide to Contemporary Theory and Research. London: Sage. Kennedy, Peter. 1992. A Guide to Econometrics, Third Edition. Oxford: Blackwell. Key, V. O., Jr. 1964. Politics, Parties, and Pressure Groups, Fifth Edition. New York: Thomas Y. Crowell. Kleppner, Paul. 1982. Who Voted? The Dynamics of Electoral Turnout, 1870– 1980. New York: Praeger. Kleppner, Paul and Stephen Baker. 1979. “The Impact of Voter Registration Requirements on Electoral Turnout, 1900–16,” paper presented at the annual meeting of the American Political Science Association, Washington DC. Klingemann, Hans-Dieter, Richard Hofferbert, Ian Budge, et al. 1994. Parties, Policies and Democracy. Boulder, CO: Sage. Knack, Stephen. 1999. “Drivers Wanted: Motor Voter and the Election of 1996,” PS: Political Science and Politics 32: 237–243. Kriesi, Hanspeter. 1995. Le Système Politique Suisse. Paris: Economica. Leduc, Lawrence, Richard Niemi, and Pippa Norris. 1996. “Introduction” in Comparing Democracies: Elections and Voting in Global Perspective, eds. Lawrence Leduc, Richard Niemi and Pippa Norris. Thousand Oaks, CA: Sage, pp. 1–48. Lijphart, Arend. 1997. “Unequal Participation: Democracy’s Unresolved Dilemma,” American Political Science Review 91: 1–14. Lodge, Milton, Marco Steenbergen, and Shawn Brau. 1995. “The Responsive Voter: Campaign Information and the Dynamics of Candidate Evaluation,” American Political Science Review 89: 309–326. Lopez Pintor, Rafael and Maria Gratschew with Jamal Adimi, et al. 2002. Voter Turnout Since 1945: A Global Report. Stockholm: International Institute for Democracy Electoral Assistance. Lyons, Williams and Robert Alexander. 2000. “A Tale of Two Electorates,” Journal of Politics 62: 1014–1034. Mackie, Thomas and Richard Rose. 1991. The International Almanac of Electoral History. Washington, DC: CQ Press. Maddala, George. 1988. Introduction to Econometrics. New York: Macmillan.

Bibliography

259

Mair, Peter. 2001. “In the Aggregate: Mass Electoral Behavior in Western Europe,1950–2000,” in Hans Keman (2001), pp. 122–142. Marsh, Michael. 1991. “Accident or Design? Non-Voting in Ireland,” Irish Political Studies 6: 1–14. Mattila, Mikko. 2003. “Why Bother? Determinants of turnout in the European elections,” Electoral Studies 22: 449–468. Mayhew, David. 1974. “Congressional Elections: The Case of the Vanishing Marginals,” Polity 6: 295–317. Mayhew, David R. 1991. Divided We Govern: Party Control, Lawmaking, and Investigations, 1946–1990. New Haven, CT: Yale University Press. McDonald, Michael P. and Samuel L. Popkin, 2001 “The Myth of the Vanishing Voter,” American Political Science Review 95: 963–974. Merriam, Charles and Howard Gosnell. 1924. Non-Voting. Chicago: University of Chicago Press. Miller, Warren and Merril Shanks. 1996. The New American Voter. Cambridge, MA: Harvard University Press. Mitchell, Glenn, and Christopher Wleizen. 1995. “The Impact of Legal Constraints on Voter Registration, Turnout, and the Composition of the American Electorate,” Political Behavior 17: 179–202. Muller, Edward and Karl-Dieter Opp. 1986. “Rational Choice and Rebellious Collective Action,” American Political Science Review 80: 471–487. Nie, Norman, Sidney Verba, and John Petrocik. 1976. The Changing American Voter. Cambridge, MA: Harvard University Press. Niemi, Richard, Guy Whitten, and Mark Franklin. 1992. “Constituency Characteristics, Individual Characteristics and Tactical Voting in the 1987 British General Election,” British Journal of Political Science 22: 229–254. Norris, Pippa and Kenneth Newton. 2000. “Confidence in Public Institutions” in Pharr and Putnam (2000). Norris, Pippa, ed. 1999. Critical Citizens: Global Support for Democratic Government. Oxford: Oxford University Press. Norris, Pippa. 2001. “Women’s Power at the Ballot Box,” in Beetham (2001). Norris, Pippa. 2002. Democratic Phoenix: Reinventing Political Activism. New York: Cambridge University Press. Olson, Mancur. 1965, 1971. The Logic of Collective Action; Public Goods and the Theory of Groups. Cambridge, MA: Harvard University Press. Ordeshook, Peter C. 1996. “Engineering or Science: What Is the Study of Politics?” in The Rational Choice Controversy: Economic Models of Politics Reconsidered, ed. Jeffrey Friedman. New Haven, CT: Yale University Press. Ostrom, Elinor, Roy Gardner, and James Walker. 1994. Rules, Games, and Common Pool Resources. Ann Arbor: University of Michigan Press. Palfrey, Thomas R. and Howard Rosenthal. 1985. “Voter Participation and Strategic Uncertainty,” American Political Science Review 79: 62–78. Parry, Geraint, George Moyser, and Neil Day. 1992. Political Participation and Democracy in Britain. New York: Cambridge University Press.

260

Bibliography

Patterson, Thomas. 2002. The Vanishing Voter: Public Involvement in an Age of Uncertainty. New York: Knopf. Perea, Eva. 2002. “Individual Characteristics, Institutional Incentives and Electoral Abstention in Western Europe,” European Journal of Political Research 41: 643–673. Piven, Frances and Richard Cloward. 1977. Poor People’s Movements. New York: Vintage. Piven, Frances and Richard Cloward. 2000. Why Americans Still Don’t Vote: and Why Politicians Want it That Way. Boston, MA: Beacon Press. Plutzer, Eric. 2002. “Becoming a Habitual Voter: Interia, Resources, and Growth in Young Adulthood,” American Political Science Review 96: 41–56. Polsby, Nelson W. and Aaron Wildavsky. 1991. Presidential Elections: Contemporary Strategies of American Electoral Politics. New York: Free Press. Powell, G. Bingham Jr. 1980. “Voting Turnout in Thirty Democracies: Partisan, Legal and Socioeconomic Influences,” in Rose (1980). Powell, G. Bingham Jr. 1982. Contemporary Democracies: Participation, Stability and Violence. Cambridge, MA: Harvard University Press. Powell, G. Bingham, Jr. 1986. “American Voter Turnout in Comparative Perspective,” American Political Science Review 80: 17–43. Powell, G. Bingham. 2000. Elections as Instruments of Democracy. New Haven, CT: Yale University Press. Putnam, Robert. 1995. “Tuning In, Tuning Out: The Strange Disappearance of Social Capital in America,” PS: Politics and Political Science 28: 664– 683. Putnam, Robert. 2000. Bowling Alone. New York: Simon and Schuster. Putnam, Robert. 2002. “Conclusion,” in Putnam (2002a), pp. 239–416. Putnam, Robert. 2002a. Democracies in Flux: The Evolution of Social Capital in Contemporary Society. New York: Oxford University Press. Rallings, Colin, Michael Thrasher, and Galina Borisyk. 2003. “Seasonal Factors, Voter Fatigue, and the Costs of Voting,” Electoral Studies 22: 65–79. Radcliff, Benjamin. 1992. “The Welfare State, Turnout and the Economy: A Comparative Analysis,” American Political Science Review 86: 444–454. Ramseyer, Mark and Frances Rosenbluth. 1993. Japan’s Political Market Place. Cambridge, MA: Harvard University Press. Riker, William H. 1962. The Theory of Political Coalitions. New Haven, CT: Yale University Press. Riker, William H. and Peter C. Ordeshook. 1968. “A Theory of the Calculus of Voting,” American Political Science Review 62: 25–42. Rose, Richard. 1980. Electoral Participation: A Comparative Analysis. London: Sage. Rose, Richard. 2002. “Voter Turnout: A Global Survey,” in Rafael Lopez Pintor, Maria Gratschew, with Jamal Adimi et al. (2002). Rose, Richard and Ian McAllister. 1990. The Loyalties of Voters: A Lifetime Learning Model. London: Sage Publications.

Bibliography

261

Rosenstone, Steven J. and John Mark Hansen. 1993. Mobilization, Participation, and Democracy in America. New York: MacMillan. Rothstein, Bo. 2002. “Social Capital in the Social Democratic State,” in Putnam (2002a), pp. 289–332. Rusk, Jerrold G. 1974. “The American Electoral Universe: Speculation and Evidence,” American Political Science Review 68: 1028–1049. Schattschneider, E. E. 1960. The Semi-Sovereign People. New York: Holt, Rinehart, and Winston. Schelling, Thomas C. 1978. Micromotives and Macrobehavior. New York: W. W. Norton & Company. Schlozman, Kay, Nancy Burns, and Sidney Verba. 1994. “Gender and the Pathways to Participation: The Role of Resources,” Journal of Politics 56: 963–990. Schlozman, Kay, Nancy Burns, and Sidney Verba. 1999. “What Happened at Work Today? A Multistage Model of Gender, Employment and Political Participation,” Journal of Politics 61: 29–53. Schneider, William. 1980. “Styles of Electoral Competition,” in Rose (1980). pp. 75–100. Shepsle, Kenneth A. and Mark S. Bonchek. 1997. Analyzing Politics: Rationality, Behavior, and Institutions. New York: W. W. Norton & Company. Shugart, Matthew Soberg. 1995. “The Electoral Cycle and Institutional Sources of Divided Presidential Government,” American Political Science Review 89: 327–343. Summer, William. 1906, 1960. Folkways: A Study of the Sociological Importance of Usages, Manners, Mores, Customs, and Morals. New York: New American Library. Stanley, Harold and Richard Niemi. 1988. Vital Statistics on American Politics. Washington, DC: CQ Press. Teixeira, Ruy. 1992. The Disappearing American Voter. Washington, DC: Brookings. Tingsten, Herbert. 1937. Political Behavior. London: King and Son (reprinted 1963, Totowa NJ: Bedminster Press). Topf, Richard 1995. “Electoral participation,” in Citizens and the State, eds. Hans-Dieter Klingemann and Dieter Fuchs. Oxford: Oxford University Press, pp. 27–51. Uhlaner, Carole J. 1989. “Rational Turnout: The Neglected Role of Groups.” American Journal of Political Science 33: 390–422. United Nations Department of Economic and Social Affairs. 1951–1998. Demographic Yearbook, 1st to 48th Issues. Paris: United Nations Publishing Division. Verba, Sidney and Norman Nie. 1972. Participation in America: Political Democracy and Social Equality. New York: Harper and Row. Verba, Sidney, Norman H. Nie, and Jae-on Kim. 1978. Participation and Political Equality: A Seven-Nation Comparison. New York: Cambridge University Press.

262

Bibliography

Verba, Sidney, Kay Schlozman, and Henry Brady. 1995. Voice and Equality: Civic Voluntarism in American Politics. Cambridge, MA: Harvard University Press. Verba, Sidney, Kay Schlozman, Henry Brady, and Norman Nie. 1993. “Race, Ethnicity and Political Resources: Participation in the United States,” British Journal of Political Science 23: 453–497. Vowles, Jack, P. Amer, J. Karp, S. Banducci, R. Miller, and A. Sullivan. 2002. Proportional Representation on Trial: The 1999 Election in New Zealand and the Fate of MMP. Aukland: Aukland University Press. Wattenberg, Martin. 2000. “The Decline of Party Mobilisation,” in Parties Without Partisans, eds. Russell Dalton and Martin Wattenberg. New York: Oxford University Press. Wattenberg, Martin. 2002. Where Have All the Voters Gone? Cambridge, MA: Harvard University Press. Wernli, Boris. 2001. Contraintes Institutionnelles, Influences Contextuelles et Par´ ections Fédérales en Suisse. Bern: Haupt. ticipation aux El´ Whitely, Paul. 1995. “Rational Choice and Political Participation: Evaluating the Debate,” Political Research Quarterly 48: 211–233. Whiteley, Paul et al. 2001. “Turnout,” Parliamentary Affairs 54: 775–788. Wlezien, Christopher. 1996. “The President, Congress, and Appropriations, 1951–1985,” American Politics Quarterly 24: 43–67. Wlezien, Christopher, Mark Franklin, and Daniel Twiggs. 1997. “Economic Perceptions and Vote Choice: Disentangling the Endogeneity,” Political Behavior 19: 7–17. Wolfinger, Raymond E. and Steven Rosenstone, 1980. Who Votes? New Haven, CT: Yale University Press.

Author Index

Abramson, Paul, 29, 42, 43 Achen, Christopher, 78, 127, 132 Aldrich, John, 29, 38, 42 Alexander, Robert, 43 Alvarez, R. Michael, 43 Andersen, Kristi, 25, 111 Asher, Herbert, 100 Banks, Arthur, 96, 98, 103, 231–233 Beck, Nathaniel, 127 Berelson, Bernard, 44, 45, 48 Blais, André, xiii, 12, 14, 19, 26, 28, 29, 42, 43, 74, 75, 86, 114, 232, 233 Blondel, Jean, 152 Bloom, Jonathan, 65 Boechel, Richard, 2, 100 Bonchek, Mark, 47, 50 Borisyk, Galina, 114 Bowler, Shaun, 44 Boyd, Richard, 98 Brady, David, 3, 16, 19, 41, 162, 166 Brody, Richard, 19 Brug, Wouter van der, 35 Burnham, Walter Dean 87, 108 Burns, James MacGregor, 99 Burns, Nancy, 84 Butler, David, 21, 59, 65, 197, 204

Cain, Bruce, 29, 42 Campbell, James, 25 Campbell, Angus, 59, 100 Caramani, Danièle, 232 Caul, Miki, 14, 18, 26, 147, 171, 231–233 Christy, Carol, 60 Clarke, Harold, 31 Cloward, Richard, 2, 219 Cohen, Jacob, 128 Cohen, Patricia, 128 Converse, Philip 103, 106 Crepaz, Markus, 14, 28, 232, 233 Dahl, Robert, 114 Dalton, Russell, 2, 171, 232, 233 Danesi, Martin, xi Day, Neil, 3 Donovan, Todd, 44 Downs, Anthony, 2, 28, 37, 42, 45, 46, 100 Egmond, Marcel van, 114 Eijk, Cees van der, xii, xv, 21, 34, 35, 43, 52, 72, 92, 114, 197, 222, 232 Erikson, Robert, 25, 86, 102 Evans, Diana, xii, xv, 30, 74, 105, 106, 108, 127

263

264 FEC, 232 Finifter, Ada, 45 Finkel, Steven, 50 Fiorina, Morris, 28, 111, 218 Fotos, Michael, xii, xv, 37 Fowler, James, 42 Franklin, Daniel, 99 Franklin, Mark, 14, 15, 17, 21, 25, 28–30, 34, 35, 37, 39, 42–44, 48, 52, 59, 73–75, 92, 101, 103, 105, 106, 108, 113, 114, 116, 119, 127, 145, 147, 153, 178, 197, 203, 210, 220–222, 231– 233 Gardner, Roy, 55 Golden, Miriam, 233 Gosnell, Harold, 2, 28 Graaf, Nan Dirk de, 114 Gray, Mark, 14, 18, 26, 147, 171, 231–233 Green, Donald, 28, 40, 45 Greene, William, 119, 126, 197 Grier, Eric, 99 Grofman, Bernard, xi, xiii, xv, 14, 28, 49, 100 Hall, Peter, xiii, 58 Hanmer, Mchael, 99 Hansen, Mark, 3, 16, 17, 23, 41, 64, 101, 161, 166 Heath, Anthony, 19 Heikki, Aintila, 233 Hirczy, Wolfgang, xii, xv, 44, 92, 95, 98, 101, 113, 119, 145, 221 Huckfeldt, Robert, 41, 44, 45, 48 ICFTU, 27, 233 ICPSR, 255–226, 232 Inglehart, Ronald, 59 IDEA, International, 10, 11 Jackman, Robert, 14, 28, 44, 75, 103, 232, 233 Jennings, M. Kent, 204

Author Index Katz, Richard, 76, 127, 231, 232 Kennedy, Peter, 128, 134 Key, V. O., 100 Kim, Jae-on, 19, 41 Klingemann, Hans-Dieter, xiii, 233 Knack, Stephen, 99, 108 Kriesi, Hanspeter, 98 Ladner, Matther, 25 Lazarsfeld, Paul, 44, 45, 48 LeDuc, Lawrence, 98 Lijphart, Arent, xiii, 39, 86, 219, 220 Lipset, Seymore 16 Lodge, Milton, 44 Lyons, Patrick, 43, 119 Mackie, Thomas, 87, 88 Maddala, George, 126 Mair, Peter, 27 Marsh, Michael, xii, xv, 119, 152 Mattila, Mikko, 222 Mayhew, David, 101, 111, 218 McAllister, Ian, 59 McDonald, Michael, 26, 27, 84, 88, 147, 190, 214 McPhee, William, 44, 45, 48 Merriam, Charles, 2, 28 Miller, Warren, 12, 14, 28, 43, 60, 75, 232, 233 Mitchell, Glen, 101, 107 Moiser, George, 3 Morrow, 54 Muller, Edward, 50 Nadeau, Richard, 29, 42 Nagler, Jonathan, 43 Neimoller, Kees, 73, 210 Nie, Norman, 19, 40, 41, 59, 207 Niemi, Richard, 29, 42, 52, 98, 204 Norris, Pippa, xiii, 10, 14, 17, 19, 28, 34, 60, 74, 84, 98, 108, 114, 124, 158, 204 Olson, Mancur, 37, 39, 56 Opp, Karl-Dieter, 50

Author Index Oppenhuis, Eric, 34, 92, 232 Ordeshook, Peter, 2, 4, 27, 28, 30, 32, 38–44, 46, 49, 52, 56, 197 Ostrom, Elinor, 55 Palfrey, Thomas, 45, 46, 55 Paolino, Phil, 29, 42 Parry, Geriant, 3 Perea, Eva, 153 Petrocik, John, 59 Piven, Frances, 2, 219 Plutzer, Eric, 12, 41, 60, 63, 64, 160, 191, 203, 204 Polsby, Nelson, 100 Popkin, Samuel, 26, 27, 84, 88, 108, 147, 190, 214 Powell, G. Bingham, 14, 28, 39, 44, 75, 92, 96, 99, 103, 231, 232 Putnam, Robert, 12, 16, 49, 57, 58, 167, 203, 207 Radcliff, Benjamin, 75, 232, 233 Rallings, Colin, 114 Riker, William, 2, 4, 27, 28, 30, 32, 38–44, 46, 52, 56, 197 Rohde, David, 29, 42 Rose, Richard, 14, 15, 28, 59, 87, 88 Rosenstone, Steven, 3, 16, 17, 23, 41, 64, 101, 161, 166 Rosenthal, Howard, 45, 46, 55 Rothstein, Bo, 58 Rusk, Jerold, 103, 106 Schattschneider, E.E., 92, 100, 219, 221 Schelling, Thomas, 46, 47, 56 Schepsle, Kenneth, 47, 50 Schlozman, Kay, 3, 16, 19, 41, 84, 162, 166 Schneider, William, 92

265 Shanks, Merril, 12, 43, 60 Shapiro, Ian, 28, 40, 45, 197 Shugart, Matthew, 102 Sinnott, Richard, 152 Smirnov, Oleg, 42 Sprague, John, 41, 44, 45, 48 Steenbergen, Marco, 44 Stewart, Marianne, 31 Stokes, Donald, 21, 59, 65, 197, 204 Sumner, William, 20 Svensson, Palle, 152 Taylor, Bridget, 19 Tedin, Kent, 25 Teixeira, Ruy, 2, 99 Thrasher, Michael, 114 Tingsten, Herbert, 2, 28, 85, 108, 124 Topf, Richard, 26, 61 Tufte, Edward, 114 Twiggs, Dan, 197 Uhlaner, Carole, 48, 52–55 United Nations, 231 Verba, Sidney, 3, 16, 19, 40, 41, 59, 84, 162, 166, 207 Vowles, Jack, xiii, 232 Walker, James, 55 Wattenberg, Martin, 2, 167, 171, 176 Wernli, Boris, 3, 94, 97 Wessels, Bernard, xii, xv, 59 Whiteley, Paul, 4, 23, 31, 50 Whitten, B. Guy, 29, 42, 52 Wildavsky, Aaron, 100 Wlezien, Christopher, 99, 101, 107, 197 Wolfinger, Raymond, 3, 41

Subject Index

absentee voting, 74, 84–85, 110, 114, 115, 121, 122, 130, 136–137, 145, 148, 178, 231, 234–235 abstention, 4, 32, 43; See also turnout accidental. See consequences, accidental accountability, 101, 193, 218–220 act of voting. See decision to vote age, 16–18, 153, 158, 160–161, 207, 231 cohort, 31–33, 57, 59, 61–62, 64–65, 68–73, 75, 78–82, 88–89, 91, 104, 151, 168, 172, 186, 191, 199 effect, 12, 61–64, 66, 71, 73, 75, 78, 160, 193, 203–205, 214 older. See established cohorts structure, 147–148, 206, 212, 214, 215 younger. See newer cohorts aggregate data, 6, 35, 37, 58, 62, 66, 76, 118, 133, 135, 153, 211, 231–235 aggregate level, 5, 9, 16–18, 22–23, 25, 28–30, 35, 79, 80, 146–149, 167, 206 alienation. See disaffection analysis. See statistical analysis

arrangements. See electoral arrangements assessing the findings, 78–86, 136–142, 173–176, 178–183, 185–190, 195–197, 212–217 assumptions, 5, 32, 38–40, 48–49, 54–56, 58, 73, 83 attachment. See party identification attendence. See church attendence Australia, 3, 4, 111–112, 120, 180, 231, 238 Austria, 108, 116, 119, 238 autocorrelation, 108, 119, 134 autoregressive process, 127–128 average district magnitude 75, 76, 79, 121–122, 129–130, 142, 144, 232 baseline model, 3, 41; see also resources behavior of individuals, 209–210 Belgium, 3, 4, 25, 112, 120, 129, 130, 137–138, 177, 180, 238 benefits of voting, 2, 4, 22–24, 27–30, 37–40, 42–46, 48–56, 109, 117–118, 120–121, 149, 193, 195–196 bicameralism, 144

267

268 Britain, 10, 25, 30, 58, 67–72, 74–75, 91, 96, 108, 119, 179, 182–183, 197, 232, 242, 244 calculus of voting 23, 32, 37–56, 58, 121–122, 129 campaign effects, 44–45, 155, 157, 160, 162–163, 165–166, 207, 208 Canada 12, 43, 120, 129, 232, 238 cartel government, 97, 112, 212–213, 221 case study 7, 33, 91–92, 169 causes of turnout change, 4, 16, 18–20, 43, 79, 82, 91–95, 106, 112–115, 121, 183–184, 193, 208–209, 211 ceiling effect 76, 93, 94 census data 87–88 change. See turnout change character of election, 2–4, 6, 27, 29, 30, 32–34, 57–60, 65, 66, 73–75, 79, 80, 91, 109, 121, 124–125, 129, 131, 137–138, 142, 143, 146–149, 151–155, 158, 160–169, 171–176, 179, 184–188, 195, 196, 206–212, 215, 217, 220 characteristics. See individual characteristics church attendence 16, 45, 157, 159–160, 166 citizen. See individual voter civic virtue, 2–4, 6, 16, 18, 42, 48, 54, 171, 215 cleavage politics. See party–group linkages closeness. See margin of victory coalition bargaining, 112, 144, 214 coalition government, 142, 144, 231 cohesion. See party cohesion cohort effect, 186–190, 207, 210, 211, 244 level of analysis, 7, 32, 33, 68, 79, 167, 194, 209, 211

Subject Index size, 138–139, 177, 185–186, 188–190, 192–193, 198, 210, 215 See also age cohort collective action, 37, 41, 48, 54–56 comparability across election studies 68, 153, 155, 158 competitive. See electoral competition compulsory voting, 4, 15, 33, 34, 39, 73, 74, 80, 84–85, 93, 110, 114–116, 121, 122, 129–130, 137, 147, 171, 176, 178, 184, 191, 212, 221, 231, 234 confirmatory findings, 167–169, 211 congressional election, 30, 101–102, 106–107, 109–111, 119–149, 214, 220, 221, 223 consequences accidental, 4–5, 34, 221–222 incidental, 4, 34, 190–191, 213–215, 221–222 of low turnout, 5, 65, 219–222 consequential effects, 4, 32, 34, 184, 221 constituency. See district context, 38, 40–41, 43–46, 48, 52–56, 149, 197–198 institutional, 44–45, 95–105, 115–116, 121, 148, 184 political, 44 social, 45 contingent relationship, 16, 166–167, 180–185, 206–207 conventional approaches, 4, 80, 94, 107–108, 124, 160–161, 164, 167, 197, 202, 206 Costa Rica, 198 costs of voting, 28, 38, 39, 41–42, 46, 48, 51, 52, 54, 63, 109, 121, 122, 148, 191, 194, 196, 205 country differences, 9, 14–16, 25, 28, 29, 44, 79–81, 96, 100, 116, 126–129, 134, 136–137, 185, 208–209

Subject Index country differences in the causes of change, see contingent relationship dummies. See dummy variables level, 7, 32, 33, 209, 211, 231 country-by-country findings, 181–183, 238–242 cross-sectional analysis, 32, 115, 145, 148 cumulative effects, 13, 32–34, 62, 79–82, 85, 92–94, 102, 104–105, 107, 110–112, 114–120, 122–123, 125–126, 129–131, 138, 142, 144–148, 167–168, 184–186, 188, 191, 195, 199, 234; see also super-long-term effects data. See aggregate data, survey data, dataset dataset, 5, 32, 35, 56, 58, 62, 67–68, 73, 81, 89, 96, 106–107, 118–120, 146, 201, 209–211, 225–228, 231–235 decision to vote, 3, 12, 15–16, 22, 37, 38, 41, 48–55, 59, 64, 112 decisiveness of election outcome, 29, 31, 39, 48, 50–52, 130, 142, 144, 232 declining turnout. See turnout decline democracy, xii, 1, 2, 4, 171; see also established democracy, transitional democracy demographic change. See social structure Denmark, 112, 120, 180, 183, 238 differences in proportions, 156–157, 247–248 direct effects, 35, 161, 165, 211, 212 direction of effects, 76, 121 disaffection, 2, 6, 145, 146, 215 disenfranchisement of young voters, 213 disproportionality, 130, 142, 144, 232; see also proportionality

269 district. See electoral district, local contest divided government, 100–103, 105–107, 115, 120 dummy variable, 124–125, 128–129, 132, 136, 155 Dutch. See Netherlands early period, 173–178, 185, 193 education, 3, 16–20, 26, 30, 40, 61–63, 65, 80, 146–147, 157, 159, 162, 206, 213, 232–233 effects of independent variables, 34, 35, 59, 62, 65–66, 77–80, 82, 85, 101, 106, 112, 114–116, 120, 129, 130, 132–134, 142–146, 152, 189, 190; see also age effect, cumulative effect, direct effect, indirect effect, short-term effect, total effects from causal modeling election level of analysis, 7, 194, 209 outcome, 2, 23, 24, 28, 29, 32, 34, 37–40, 44, 47, 50, 52–55, 57, 111–112, 119, 122, 149, 162, 178, 214 study, 32, 60, 71, 73, 126, 151–153, 155, 164, 169, 225–228 see also character of election, electoral context electoral arrangements, 4, 6, 28, 30, 31, 33, 44, 48, 65, 80, 81, 84–85, 114, 116, 148, 153, 184 coalition, 48, 49, 51, 52, 54–56, 202; see also social group cohort. See age cohort competition, 6, 27–30, 33, 34, 57, 59, 74, 77, 84, 91–97, 108, 111–115, 118, 122, 146–148, 153, 178, 184, 192, 199, 207–209, 217–221, 223 context, 31, 59, 205–206 district, 30–31, 74–75, 121, 218

270 electoral (cont.) experience. See learning experience reform, 4, 65–67, 73, 79, 80, 82–85, 91–108, 110, 116, 118, 119, 123, 202, 213–214, 216, 220, 222–223; see also reforming U.S. electoral procedures salience, 12, 44, 74, 75, 99, 152, 161–162, 167, 179 system, 110, 115, 121, 136, 142–144, 218, 223, 232; see also electoral arrangements, majoritarian, proportional electorate size, 26, 27, 39, 62, 70, 79, 83–89, 114, 122, 124–126, 130, 138–139, 142, 144, 147, 176, 178, 184, 214–215, 232 empowerment. See female empowerment engagement. See social engagement established cohorts, 9, 26, 31, 59, 61, 62, 65, 66, 70, 72–73, 75, 76, 78–80, 82, 91, 105, 115, 147, 158–162, 166, 168, 186, 193–194, 199, 207, 210, 211, 215, 216 democracy, 6, 7, 10, 16–19, 25, 31, 32, 34, 56, 67, 78, 92, 126–129, 198–199, 231–235 estimated consequences of independent variables, 172–177 effects. See direct effects European Parliament elections, xi, xii, 25, 42, 52, 222 Union, xi, 95, 180, 222 evolution. See turnout evolution executive responsiveness, 28, 92–98, 103, 112, 119, 122, 124, 130, 138, 144, 147, 149, 171–172, 176, 179, 184, 188, 191, 193–194, 218–219, 232, 234 experience of cohort (years in electorate), 75, 76, 78

Subject Index of voting. See learning experience exploratory investigation, 152–167 fatigue. See frequency of elections, time since previous election female empowerment, 124–125, 130, 136, 140–141, 147, 176–178, 184, 191, 235 enfranchisement, 25, 60, 83–85, 93, 97, 124–125, 130, 141, 177, 188, 199, 212 turnout, 141, 158, 204; see also female enfranchisement 140–141, 158, 204 findings from past research. see research findings illustrative, 155, 167 implications. See implications of the findings of this research, 7, 37, 61, 67–73, 77–80, 91, 101–103, 106–110, 132–149, 152–153, 155–162, 171–175, 181–182, 185, 195–197, 201, 202, 211–212, 222, 237–248 Finland, 11, 120, 129, 132, 183, 238 first -past-the-post. See majoritarian -time voters, 43, 61, 63–65, 70, 77, 82, 85, 88, 91, 168, 233 fixed effects model, 128–129, 134, 136–137 footprint of past turnout, 12, 43, 60, 63, 64, 69, 82–85, 91, 105, 112, 120, 199 formative experience. See socialization ?? fractionalization of party systems, 6, 34, 176, 192, 214, 217, 221 France, 12, 25, 86, 96, 98, 100, 108, 116, 119, 129, 130, 177, 183, 232, 238 franchise extension, 60–62, 97, 110, 115, 121, 122, 124–125, 211, 213–214, 221

Subject Index free rider problem, 38, 39, 45, 47, 48, 51, 55–56 frequency of elections, 6, 61, 97–99, 114, 139, 194 gender. See female turnout generalizability, 153, 199 generalized least squares regression, 76, 132 generational change, 70 replacement, 5, 13, 21, 25, 32–33, 59–60, 62, 64–66, 68–71, 73–74, 82–83, 85–86, 88, 94–98, 112, 116, 120, 122–124, 168–169, 172, 177, 185, 205, 207, 209, 211, 214, 216–217, 221 Germany, 9, 34, 67, 71–73, 91, 108, 119, 153–168, 183, 194, 212–217, 238, 244 globalization, 179–180, 219 Great Britain. See Britain golden rule, 97–98, 221 Greece, 198 gross domestic product, 146, 232 group. See social group habit of voting, 2, 12, 21, 22, 24, 32, 40, 49, 53–56, 60, 61, 63–65, 72, 80, 84, 91, 112, 115–118, 120–121, 124, 127, 148–149, 152, 160, 167, 195, 197, 199, 203–205, 207–208, 216; see also inertia, socialization, turnout stability heteroskedasticity, 101–102, 108, 119, 126–128, 134, 136–137, 189 Holland. See Netherlands hybrid effect 125 hypothesis, 5–7, 32, 35, 56, 65–66, 68, 73–75, 77, 78, 115, 126, 129–131, 152, 178, 202 Iceland, 112, 120, 240 identification. See party identification

271 immunization, 5, 21–22, 60, 65, 216 implications of the findings, 146–149, 167–169, 191, 192, 195–199, 202, 217 important issues. See issue salience impossibility result, 38 incidental. See consequences, incidental income, 19, 30 independent variable, xi, 7, 15, 30, 33–35, 57, 58, 74–78, 80, 81, 91–115, 117, 120–129, 142–146, 151–153, 155, 162–167, 189, 209, 211, 215, 231–235; see also cumulative effects, effects, short-term effects indirect effect, 34, 80, 161–166, 176, 178, 211, 212 individual behavior. See behavior of individuals characteristics, 162–167 level of analysis, 5, 7, 9, 16–19, 22–23, 25–34, 68, 79, 80, 148, 151–153, 160, 167, 169, 206–207, 209, 211 voter, 2–6, 9, 16–18, 25, 27–30, 33, 34, 37–44, 46, 51, 52, 54–56, 58–60, 63, 79, 80, 151–169, 202, 205–206, 208–210, 213, 219–221 inertia, 21, 22, 56–57, 60, 66, 73, 77, 78, 121, 126–127, 132, 164, 185, 190, 197, 199, 215; see also habit of voting influences. See short-term effects, cumulative effects information, 38, 43–46, 48, 53–55 initiation. See young initiation institutions. See electoral arrangements, electoral systems, political institutions insulation from electoral verdicts, 105–111

272 interaction terms, 75–78, 81, 117–118, 122–125, 131, 136–137, 144, 152–153, 158, 177–178, 184, 186, 188–190, 192–195, 198, 211, 215 interest in politics, 19, 157, 158, 160–163, 207–208 Ireland, 113, 120, 183, 240 Israel, 111–112, 119, 180, 240 issue salience 2, 43; see also electoral salience Italy, 25, 80, 111, 116, 119, 129, 130, 137, 177, 183, 191, 212, 231, 240 Japan, 25, 111, 116, 119, 130, 137, 177, 240 lagged turnout. See past turnout later period, 173, 178–180, 185 learning experience, 3, 61, 63–66, 71, 75, 79–81, 168, 205, 213 see also socialization legislative elections, 30, 117, 120 revolts, 193 legitimacy, 2, 40, 42 level of analysis, 201; see also, aggregate level, cohort level, country level, election level, individual level, time-series analysis links. See party–group linkages literacy, 17, 146, 232; see also education local contests, 30–31, 103, 111, 189, 218, 223; see also district, electoral district logistic regression analysis, 155–158, 246–248 long-term effects. See cumulative effects Luxembourg, 111, 119, 130, 231, 240

Subject Index magnitude. See average district magnitude majoritarian electoral system, 31, 74–75, 113, 120, 121, 130, 136–137, 140–141, 143, 189 majority status, 74–76, 80, 81, 113, 120, 121, 130, 137, 144, 146–147, 149, 161, 163–164, 168, 176, 178, 184, 186, 191, 193, 207, 208, 211, 214–215, 232 Malta, 3, 4, 12, 18, 25, 33, 91–98, 103, 111, 113, 115, 119–126, 138, 145, 177, 179–180, 183, 190–191, 213, 217, 240 margin of victory, 2, 24, 29, 31, 40, 42, 45, 48–51, 54–57, 66, 74–78, 80, 81, 95, 101–113, 115, 119–121, 123, 130, 137, 142–144, 147, 149, 158, 161, 163–164, 168, 176, 178, 184, 188–189, 192–193, 202, 207, 208, 210, 211, 218, 220, 232 message-sending, 40, 42–43, 49, 52–55, 109, 196 midterm elections. See U.S. midterm elections missing data, 89, 198 mobilization of potential voters, 3, 16–18, 23–26, 29, 30, 41, 58, 114, 145–147, 155, 157, 161–162, 164–166, 191, 203, 208, 216 model specification, 3, 5, 6, 12, 30, 33–35, 38–47, 49–59, 65, 66, 69, 73–75, 77–80, 91, 101, 107–108, 116, 119, 126–127, 129, 131–136, 142, 144, 145, 147–148, 151–153, 157–162, 171–172, 175, 178–179, 190, 195–196, 198, 203–204, 209, 212 motivation. See voter motivation multicolinearity, 75, 125, 134, 136, 144, 186 multilevel models, 35

Subject Index national election, xi, 4, 6, 30–31, 34, 39 level. See country level of analysis Netherlands, 12, 67, 72–74, 77, 79, 111, 119, 129, 171, 191, 210, 212, 217, 231, 240, 244 new cohorts, 9, 13, 24–27, 31–34, 40, 43, 57, 59–66, 72–76, 78–82, 86–89, 91, 94, 104–105, 114–116, 121–126, 139, 142, 144, 146, 147, 152, 158–162, 166, 168, 177–178, 180, 185–186, 188–190, 192–195, 198–199, 203, 207–208, 210–211, 213–216, 223, 233 size 146, 184 voters. See first-time voters New Zealand, 80, 113, 116, 120, 183, 217, 232, 242 Norway, 9, 67, 72, 79, 111–112, 119, 180, 242, 244 number of cases, 6, 7, 33, 35, 68, 71, 76, 79, 136, 148, 194, 209, 210, 225–228 of parties, 95, 176, 232 occupation, 30, 45, 61, 157, 159–160 older age cohorts. See established cohorts OLS. See ordinary least squares operationalization of variables, 122, 124, 130, 137–138, 145–147, 153, 155, 163–164, 168, 192, 194–196, 204, 231–233 opinion poll. See sample survey ordinary least squares, 76, 155 organizational membership. See social groups outcome. See election outcome outlier, 92–98, 175, 183 panel corrected standard errors, 76, 127–128, 132, 137 paradox. See turnout paradox

273 parameter estimates. See effects parliamentary terms, 137 partisanship. See party identification party cohesion 6, 113, 121–122, 129–130, 137–138, 147, 176–178, 184, 189, 192–193, 210, 220, 231 –group linkage, 114, 146, 203, 232 identification, 21, 25, 65, 146, 152, 157, 164–167, 171, 208 manifesto, 42, 121, 192, 214 program. See party manifesto size, 81, 112–113, 115, 121, 137, 192–193 sympathy, 157, 160–163, 207–208 See also political party past research findings xii, 5, 9, 13, 27, 30, 32, 34, 37–41, 43, 44, 46, 55–56, 58–60, 66, 74, 75, 84, 98–99, 124, 134, 146–148, 152, 164, 166–167, 216, 231–235 past turnout, 77–78, 102, 107–108, 118, 119, 121, 126–128, 130, 132–134, 137, 175–176, 178, 195, 199, 208–210 path dependency. See contingent relationships period. See earlier period, later period Perot, Ross, 43, 196 platform. See party manifesto plurality elections. See majoritarian elections polarization, 95, 113, 121, 130, 142, 145, 233 policy stakes, 44, 48, 101–105, 149; See also public policy political causes, 4, 65, 97, 212–214, 216–223 context, 28 culture, 145 implications. See implications of the findings

274 political (cont.) interest. See interest in politics party, 2, 6, 12, 27, 30, 34, 38, 48, 52, 54, 59, 60, 66, 152, 167, 196, 213, 216, 223 system, 219 politicians, 2, 34, 218–220, 223 population density, 146, 233 Possum Hollow Rod & Gun Club, 47, 48 postal voting. See absentee voting PR. See proportional representation predicted turnout, 172, 175, 180–182 predictive success, 35, 58, 172, 174–175, 180–183, 190, 198, 216–217, 238–242 presidential election. See U.S. presidential elections prisoner’s dilemma, 46–47, 53–55 propensity to vote, 33 proportional representation, 31, 74, 80, 94, 95, 136, 140–141, 218, 220 proportionality of the electoral system, 15, 93, 113, 115–116, 121; see also disproportionality public policy, 2, 24, 29, 38, 39, 42, 44, 45, 47, 52, 54, 57, 95, 97, 99–100, 103, 111–114, 121, 144, 151, 179, 214, 218, 221, 222 puzzle-solving, xi, 4, 5, 7, 9, 31, 35, 59, 201–223 rational choice theorizing, 2, 27, 28, 37–46, 50, 117, 120, 129–131, 195–197 rationality. See voter motivation, voter rationality referendum, 97–99, 221 reform. See electoral reform reforming U.S. electoral procedures, 220 registered electorate, 14, 86–88

Subject Index regression analysis. See generalized least squares, logistic regression, ordinary least squares regression findings. See findings religion, 157–160 replacement. See generational replacement research constraints, 31–32, 212 design, 32, 81, 102, 116, 120–126, 147–148, 152–153, 201, 211 findings. See past research findings methods. See statistical methods question, 13, 31, 32, 118, 151, 212 see also past research findings residuals from the analysis, 146, 215 resources of individuals, 3, 6, 17–20, 25, 29, 30, 84, 145–147, 155, 157, 161–162, 164–167, 206–207 responsiveness. See executive responsiveness, voter responsiveness rigging the system, 219–220 robustness of findings, 129, 136, 179 safe seats in the U.S. congress, 106, 143, 144 salience of elections. See electoral salience sample survey, xii, 5, 28, 34, 62, 67, 68, 71, 74, 88, 151, 153–155, 196, 202, 225–228 self-interest. See rationality separation of powers. See U.S. separation of powers shooting the messenger, 220 short-term effect, 33, 34, 81, 105, 111, 116–118, 120, 122, 125, 129–131, 138, 142, 145–149, 167–168, 177, 184–186, 191, 195, 199 significance. See statistical significance simultaneous equation modeling, 163

Subject Index size. See electorate size social capital, 16–18, 26, 45, 51, 52, 57–58, 63, 167, 190–192, 203–204; see also social group social characteristics. See social structure social composition. See demographic change social context, 58, 202, 205–206, 208; see also social capital social engagement, 191, 203, 205, 216, 222–223 social group, 17, 20–21, 23, 40–43, 45–46, 48, 51–56, 61, 63, 64, 68, 84, 114, 123, 125, 196, 203–204 social linkage. See party–group linkages, social capital social status, 3, 17, 41, 80; see also education, income social structure, 6, 16–18, 25–27, 145–148, 206, 209–210, 219 socialization, 5, 20–25, 29, 32, 33, 54–56, 60, 63–66, 74, 80–82, 84, 104–105, 116, 123, 124, 160, 169, 172, 186, 189–191, 199, 207, 209, 214 society. See social context solutions to the voter turnout problem. See electoral reform spurious findings, 145, 148 stability. See turnout stability stakes. See election stakes, policy stakes standard error of the estimate, 12, 137, 139, 181–182 standing decision, 23, 54–56, 80, 121 statistical analysis, 7, 34, 64–66, 68, 73–76, 78, 100, 106–108, 131–133, 135, 143, 155–162, 201, 211 methods, 11, 67–68, 76, 127, 163; see also GLS regression, interaction terms, logistic regression, multilevel modeling, OLS regression, time-series

275 analysis, two-stage least squares, weighting the data significance, 35, 77–80, 88, 101, 116, 121, 125, 134, 142–146, 148, 157, 182, 186, 189–190, 201, 206, 210, 212–217 test, 35, 56, 58, 66, 73, 77, 79, 101–102, 129, 131, 140, 142, 145–146, 172, 179, 201, 202, 211–212 status. See social status strategic voting, 29, 30, 42, 196 super-long-term effect, 128, 130; see also cumulative effect survey data. See sample survey Sweden, 9, 58, 67, 72, 73, 91, 113, 180, 242, 245 Switzerland, xii, 4, 12, 18, 20, 25, 28, 33, 74, 91–99, 111, 113, 115, 121, 124, 129, 130, 137, 138, 148, 171, 177, 188, 190–192, 212, 217–221, 223, 242 sympathy. See party sympathy tactical voting. See strategic voting term limits movement. See U.S. term limits movement test. See statistical test theoretical framework, 5, 7, 12, 17, 30, 32, 33, 35, 37–38, 46, 69, 84, 91, 125, 153, 160, 176, 178, 185, 199, 201–202, 211 threshold requirement, 143 time series analysis, 9, 14, 16, 32, 78, 82, 100–103, 115, 137, 145, 208–209 data, 32, 34, 81, 126–129, 132, 194 -serial dependency, 78, 102, 126–128, 132, 136 since previous election, 61, 71, 75, 76, 88, 114, 119, 121, 130–131, 134, 137–139, 147, 176–178, 193–195, 233 see also frequency of elections

276 total effects from causal modeling, 83, 165–167, 176, 208 transition. See age effect transitional democracy, 6, 16, 17, 19, 172, 198–199 trust in government, 2, 6, 145–146, 171 turnout at previous elections. See past turnout change, 3, 9–14, 18, 30, 33–35, 43, 49, 57, 59–74, 76, 79, 81, 82, 86, 91, 94–98, 105–107, 110, 111, 118–127, 129, 148–149, 158, 173–175, 183–190, 193, 205–206, 208–210, 212, 217 decline, 1, 2, 4, 6, 7, 9–11, 13, 19, 25, 31–34, 58, 60–62, 64–67, 70, 71, 73, 79–80, 82–85, 93–94, 99, 101–103, 107, 111–112, 118, 120, 139, 148, 153, 167, 171, 173, 175, 178, 182, 185, 190–192, 205–207, 212, 214, 215, 217, 222 evolution, 7, 11, 31, 33, 65, 67–68, 76, 82–84, 91–111, 118, 120–121, 155, 168, 178, 199, 207–208 high, 4, 57, 60, 78, 94–95, 149, 164, 167, 214, 217 increase, 9, 31, 58, 60, 62, 64, 66, 70, 78, 83, 85, 93, 96–97, 139, 153, 177–178, 185, 206–207, 215 level, 16, 121, 126, 155, 217; see also turnout evolution 16, 121, 126, 155, 217 low, 5, 12, 43, 60, 61, 64, 98–108, 164, 189, 210, 217, 218, 221–223 measurement, 27, 76, 86–88, 123, 198 medium, 217–218 paradox, 28, 37, 46, 201–202 rate. See turnout level

Subject Index stable, 9–14, 205–206 trend in, 12 universal, 4, 206 two-stage estimation, 128, 134 unanimity, 50 U.S.A. See United States of America U.S. midterm elections, 3, 102, 106, 108–109, 117, 120, 121 presidential elections, 1, 3, 43, 99, 101–102, 105–109, 111, 117, 119–121, 223 separation of powers, 98–104, 112 term limits movement, 220 uncertainty, 43–46, 52–57, 149, 202, 218 union membership, xiii, 16, 18, 24, 114, 146, 147, 171, 233 unit of analysis 76; see also number of cases United Kingdom. See Britain United States of America, xii, 1, 3, 4, 12, 16, 18–20, 24–26, 28, 30, 33, 43, 60, 67, 71, 72, 74–75, 78, 86–88, 91–93, 96, 98–99, 101–102, 106, 109–113, 115, 119–126, 137, 143–144, 167, 176, 183, 192, 196, 213–221, 223, 242, 245 urban residence, 157, 159–160, 166 utility. See benefits of voting value. See benefits of voting variability. See turnout change variables, coding, 95–98, 123–125; See also independent variables variance explained, 33, 40, 127, 137, 141–142, 147, 153, 157, 160, 168, 178, 185, 189, 190, 196, 206, 211, 216 verdict. See election outcome victory. See election outcome virtue. See civic virtue volatility. See turnout change

Subject Index voter mobilization. See mobilization of potential voters motivations, 4, 5, 27–31, 37, 40, 42–45, 49, 51–56, 63, 93, 101, 121, 122, 185, 191, 196 rationality, 2, 4, 5, 22, 24–25, 27, 28, 32, 39, 41, 44, 46–48, 52, 74, 84, 91–118 registration, 99, 107–108, 119 responsiveness, 59, 115, 121, 122, 147, 179, 188, 192–193, 210, 219, 220 sophistication, 192 see also individual voter voting act. See decision to vote age, 26, 32, 33, 61–68, 71, 73, 74, 79, 80, 82, 83, 97, 125, 129, 130, 139–140, 178, 184, 192, 203, 205, 211, 213–214, 216, 233

277 age population, 14, 27, 86–88 benefits. See benefits of voting choice, 197–198, 222 wasted vote syndrome, 29, 31, 43, 74, 113, 143–144, 223 weekend voting, 15–16, 81, 115–116, 121, 122, 130–132, 144–145, 148, 233 weighting the data, 68, 76, 102, 120, 122, 155 why vote?, xi, 2, 9, 72, 91, 97, 193, 202–203, 205, 218 World Trade Organization. See globalization young initiation, 65, 66, 68, 71, 74–79, 82, 83, 114, 125, 130, 139–140, 147, 168–169, 176, 178, 184, 190–191, 197, 233 younger age cohort. See new cohorts

Voter Turnout and the Dynamics of Electoral Competition in Established Democracies since 1945

Electoral Authoritarianism: The Dynamics of Unfree Competition

Territory and Electoral Rules in Post-Communist Democracies

Memorialization in Germany since 1945

Memorialization in Germany since 1945

Religion in America Since 1945

The Marketplace of Democracy: Electoral Competition And American Politics

Votes and Violence: Electoral Competition and Ethnic Riots in India

War and Law Since 1945

Army Uniforms Since 1945

Europe Since 1945

The International Economy since 1945

Britain since 1945

U.S.Submarines since 1945

Britain since 1945

France since 1945

The Development of Economics in Western Europe Since 1945

Germany Since 1945 (Studies in Contemporary History)

Economic Growth in Europe since 1945

Christian Democracy in Europe since 1945

Combat Aircraft Prototypes Since 1945

Technology and the American Way of War Since 1945

The theory of the chemostat: Dynamics of microbial competition

Urban Planning Theory since 1945

Postwar: A History of Europe Since 1945

Combat Aircraft Prototypes Since 1945

American Film and Society since 1945

Postwar: A History of Europe Since 1945

American Film and Society since 1945

Contemporary France: Politics and Society since 1945

Postwar: A History of Europe Since 1945

Voter Turnout and the Dynamics of Electoral Competition in Established Democracies since 1945

Electoral Authoritarianism: The Dynamics of Unfree Competition

Territory and Electoral Rules in Post-Communist Democracies

Memorialization in Germany since 1945

Memorialization in Germany since 1945

Religion in America Since 1945

The Marketplace of Democracy: Electoral Competition And American Politics

Votes and Violence: Electoral Competition and Ethnic Riots in India

War and Law Since 1945

Army Uniforms Since 1945

Europe Since 1945

The International Economy since 1945

Britain since 1945

U.S.Submarines since 1945

Britain since 1945

France since 1945

The Development of Economics in Western Europe Since 1945

Germany Since 1945 (Studies in Contemporary History)

Economic Growth in Europe since 1945

Christian Democracy in Europe since 1945

Combat Aircraft Prototypes Since 1945

Technology and the American Way of War Since 1945

The theory of the chemostat: Dynamics of microbial competition

Urban Planning Theory since 1945

Postwar: A History of Europe Since 1945

Combat Aircraft Prototypes Since 1945

American Film and Society since 1945

Postwar: A History of Europe Since 1945

American Film and Society since 1945

Contemporary France: Politics and Society since 1945

Postwar: A History of Europe Since 1945

Recommend Documents