Mixed Electoral Systems: Contamination and Its Consequences

Mixed Electoral Systems

This page intentionally left blank

Mixed Electoral Systems Contamination and Its Consequences

Federico Ferrara Erik S. Herron Misa Nishikawa

MIXED ELECTORAL SYSTEMS

© Federico Ferrara, Erik S. Herron, and Misa Nishikawa, 2005. All rights reserved. No part of this book may be used or reproduced in any manner whatsoever without written permission except in the case of brief quotations embodied in critical articles or reviews. First published in 2005 by PALGRAVE MACMILLAN™ 175 Fifth Avenue, New York, N.Y. 10010 and Houndmills, Basingstoke, Hampshire, England RG21 6XS Companies and representatives throughout the world. PALGRAVE MACMILLAN is the global academic imprint of the Palgrave Macmillan division of St. Martin’s Press, LLC and of Palgrave Macmillan Ltd. Macmillan® is a registered trademark in the United States, United Kingdom and other countries. Palgrave is a registered trademark in the European Union and other countries. ISBN 1–4039–7097–1 Library of Congress Cataloging-in-Publication Data Ferrara, Federico. Mixed electoral systems : contamination and its consequences / Federico Ferrara, Erik S. Herron, Misa Nishikawa. p. cm. Includes bibliographical refernces and index. ISBN 1–4039–7097–1 1. Elections. 2. Representative government and representation. 3. Comparative government. I. Herron, Erik S. II. Nishikawa, Misa. III. Title. JF1001.F49 2005 324.6⬘3—dc22

2005049186

A catalogue record for this book is available from the British Library. Design by Newgen Imaging Systems (P) Ltd., Chennai, India. First edition: December 2005 10 9 8 7 6 5 4 3 2 1 Printed in the United States of America.

Ai miei nonni Bici, Sally, Liana e Fausto. For Lea, Carter, Thomas and Yvonne Herron. For Kimiaki, Miyoko and Yusuke Nishikawa.

This page intentionally left blank

C on t e n t s

List of Tables and Figures

ix

Preface and Acknowledgments

xi

1. A New Species of Electoral Institution? Relative Causal Inference and the Study of Mixed Electoral Systems

1

2. A Taxonomy of Mixed Electoral Systems

15

3. The Logic of Contamination: District-Level Party Systems

33

4. Going it Alone? Strategic Nominations in Mixed Electoral Systems

49

5. Interactive Ballots: The Microfoundations of Contamination

65

6. Voting from their Hearts? Contamination and Strategic Voting

81

7. Eyeing the Future: Contamination and Legislative Behavior

99

8. Better, Worse, or Just Different? Legislative Fractionalization in Mixed Systems

121

9. Conclusion

139

Notes

147

References

167

Index

179

This page intentionally left blank

List of Tables and Figures

Tables 2.1 2.2 2.3 2.4 3.1 3.2 3.3 4.1 4.2 4.3 4.4 5.1 5.2 6.1 6.2 6.3 7.1 7.2 7.3 7.4 8.1 8.2 8.3

Distribution of seats in mid and west Wales region (1999) PR seat distribution in the Ukrainian Rada (1998) Total seat distribution in the Ukrainian Rada (1998) List of mixed systems (1990–2004) Institutional variation in the sample countries Evidence of contamination in PR vote shares Mean effective number of electoral parties Institutional variation in sample countries Strategic entry under mixed electoral rules: Poisson regression results Coordination under mixed electoral rules: OLS results Assessing strategic entry in hypothetical mixed systems Contamination and party choice: Conditional logit estimates Contamination and party choice: Conditional logit estimates The incidence of strategic voting The effects of the opportunity factor on voting decisions— Japanese lower house, 1996 The effects of the opportunity factor on voting decisions— Japanese upper house, 1983 Faction affiliation and number of deputies in Ukraine’s Verkhovna Rada IRLS results from Ukraine’s Verkhovna Rada Parliamentary groups in the Italian Chamber of Deputies, June 2001 IRLS results from the Italian Chamber of Deputies Institutional variation in the world’s mixed systems Mixed systems: Evaluating alternative taxonomies: robust MM estimation How majoritarian is your mixed system (and does it matter)? Robust MM estimation

20 21 22 23 38 45 46 55 59 60 62 73 75 89 92 93 108 110 114 116 127 130 132

x / list of tables and figures

Figures 3.1 3.2 3.3 5.1 5.2 6.1a 6.1b 6.1c 6.2a 6.2b 6.2c 7.1 7.2 8.1 8.2

SMD candidate placement and change in PR performance: Japan, 2000 SMD candidate placement and change in PR performance: Italy, 1996 SMD candidate placement and change in PR performance: Russia, 1999 SMD candidate evaluations and party choice––Example 1: Labour “Leaner” SMD candidate evaluations and party choice—Example 2: Independent voter Predicted probabilities (marginal candidates) Predicted probabilities (front runners) Predicted probabilities (absent) Predicted probabilities (marginal candidates) Predicted probabilities (front runners) Predicted probabilities (absent) Variation in discipline, Ukraine Variation in discipline, Italy The ENPS in proportional, majoritarian, and mixed systems estimates from model 2, table 8.3 (DPI Data) The ENPS in proportional, majoritarian, and mixed systems estimates from model 3, table 8.3 (DPI Data)

42 42 43 76 77 94 95 95 96 96 97 111 117 135 136

Preface and Acknowledgments

Many years ago, we received some valuable advice from our mentor, Brian Silver: “good colleagues help you get rid of bad ideas quickly.” The evolution of ideas—discarding the bad and retaining the good—is linked to a metaphor underlying the book’s analysis: mixed systems as a new species of electoral institution. Just like the “electoral species” we describe in this book, our ideas have evolved. This book is a product of collaboration that began in 1998 and has grown stronger over time. Our initial work on contamination synthesized papers composed for graduate courses taught by Dennis Patterson and Mark Jones at Michigan State University. It has subsequently encompassed other ideas, conversations, papers, and dissertations, culminating in this volume: Mixed Electoral Systems: Contamination and Its Consequences. Each of us has brought our own strengths to the project. We could not have completed the book without the skills and efforts of all three authors. The book has been built from scratch, through a collective effort to refine and improve our good ideas and discard our bad ideas. While each of us has taken the lead on individual chapters, every chapter has three sets of fingerprints. We have benefited greatly from assistance provided by many people and organizations. However, we are responsible for the final version of the book’s prose, arguments, and analysis; others cannot be blamed for our shortcomings. Our friends at the University of Kansas, especially Ron Francisco and Paul Johnson, have offered staunch support, indispensable mentorship, and invaluable professional guidance, as well as helpful comments and critiques on many parts of the book. At Michigan State University, where two of the authors received their graduate degrees, Brian Silver, Dennis Patterson, Mark Jones, Bill Reed, Scott Gates, Ada Finifter, Robert Lowry, and Tse-min Lin provided feedback and encouragement for the research program. At Harvard University, Barry Burden, Andrea Campbell, Ben Goodrich, Sunshine Hillygus, Mike Kellermann, Gary King, Jason Lakin, Olivia Lau, Maggie Penn, Ken Shepsle, and the participants in Harvard’s Political Psychology and Behavior Workshop supplied constructive criticism and encouragement. We also received valuable feedback and ideas from Donna Bahry, Ken Benoit, Eric Browne, Jamie Carson,

xii / preface and acknowledgments

Mikhail Filippov, Justin Fox, Timothy Frye, Thomas Gschwend, Wonjae Hwang, Neal Jesse, Masaru Kohno, Michelle Kuenzi, Gerhard Loewenberg, Ko Maeda, Steven Reed, Thomas Remington, Todd Schafer, Susumu Shikano, Matthew Shugart, Mark Souva, Andrey Zakharov, members of the Japan Election and Democracy Study Project (Ken’ichi Ikea, Susan Pharr, Bradley Richardson, and Aiji Tanaka), the participants in the Kolloquium at the University of Mannheim’s Center for European Social Research (MZES), and colleagues at Ball State University. The analysis in the book would not have been possible without the help of many in acquiring needed data and information, including: Kathleen Bawn, Matt Carlson, Moshe Haspel, Svetlana Kravtsova, Olexander Lavrynovich, Valentin Mikhailov, Robert Moser, Taehyun Nam, Christina Schenstrom, Viktor Sheynis, Frank Thames, Rob Weiner, Paul Harris at the New Zealand Electoral Commission, J. Ray Kennedy and Art Levine at the International Foundation of Electoral Systems (IFES) in Washington, DC, Robert Dahl, Christian Nadeau, Ksenia Tyurkova and the staff of IFES (Moscow), David Earl and the staff of IFES (Ukraine), the National Democratic Institute and International Republican Institute offices in Russia and Ukraine, Vladimir Pigenko, Ellie Valentine, and the staff of the Parliamentary Development Project in Kyiv, and Evhen Lapin and Kostya Shevchenko at Lab F4. Finally, David Pervin at Palgrave supported this book from the outset and helped by guiding us to completion. His colleagues, William Fain, Heather Van Dusen and Maran Elancheran provided invaluable technical assistance. Many others provided material, technical, and psychological support as we composed this manuscript; our families and friends have our eternal thanks for their help and patience.

C h ap t e r 1 A New Species of Electoral Institution? Relative Causal Inference and the Study of Mixed Electoral Systems

Elections are the key to understanding politics in democratic societies. They bestow the power to make decisions about peace and war, wealth and poverty, and the laws that govern society. This book directly addresses how mixed electoral systems—a novel and increasingly important species of electoral institutions—affect the behavior of parties, voters, candidates, and legislators. It fits into a rich genealogy of scholarly research that stretches across generations. Over twenty years ago, William Riker (1982) chronicled the formulation, attempts at empirical falsification, repeated revisions, and eventual acceptance of Duverger’s Law and Hypothesis1 to illustrate that political science has evolved through the accumulation of knowledge.2 Riker’s choice of Duverger’s propositions was hardly coincidental. Electoral rules and their consequences are among the most widely investigated research problems in political science. The literature on electoral institutions—whether descriptive or analytical, formal or empirical, single-country or cross-national—is arguably the best developed of any subfield in the discipline. Much of the scholarly work on this topic has attempted to explain the heterogeneity in outcomes generated by the profoundly diverse families of electoral systems that are in use worldwide. The fundamental distinction among the most widely adopted types of electoral rules is that some systems elect candidates through majoritarian contests held in single-member districts (SMD)3 while others allocate seats to parties in multimember districts in rough proportion to their vote share (called proportional representation, or PR).4 Scholars have extensively evaluated how alternative electoral systems affect a citizen’s propensity to vote sincerely or sophisticatedly, how electoral rules influence a party’s nomination decisions,5 how a candidate positions herself on the ideological spectrum to maximize vote totals,6 how

2 / mixed electoral systems

electoral incentives constrain the behavior of office-seeking legislators,7 how strategic voting gives rise to district-level party systems,8 and how national party systems are shaped by the interaction of institutions and social cleavage structures.9 As is the case in any observational setting, the primary challenge facing the empirical literature on the consequences of electoral systems is the “fundamental problem of causal inference” (Holland 1986).10 Because only one set of electoral rules is employed in a given locality during a single election, we cannot directly observe how different rules affect outcomes. It is therefore problematic to isolate the relative causal effects exerted by electoral institutions from those exercised by other, sometimes unknown, determinants. Through single-country and large-N empirical studies, we have accumulated a substantial amount of information about the effects of SMD and PR on political competition. However, such research cannot answer a number of interesting and important counterfactual questions. Specifically, how would the outcomes observed in particular units (countries, districts, legislatures) vary if different electoral rules were employed? For instance, the United Kingdom uses an SMD system with the plurality formula to select representatives to the House of Commons. Generally, district-level competition in Britain is restricted to two viable parties. Based on copious formal and empirical research, we know that PR tends to produce multi-party competition at the district level while SMD exercises a reductive impact on the number of candidates (see Cox 1997). If Britain adopted PR, how many parties11 would we observe at the district level? From what we know about electoral systems, we would probably answer “more than two.” Depending on district magnitude (M),12 the answer could be “many more” or “a few more” than two. With what we know from formaltheoretical models (see Cox and Shugart 1996) we could even predict that the number of relevant parties would be constrained by strategic voting and entry to M ⫹ 1,13 under a series of assumptions. To many, this answer is probably satisfactory. In fact, by social science standards, it is quite remarkable that we can provide such an answer with this level of confidence. Unfortunately, it is impossible to predict exactly how many parties Britain would have if it used PR because Britain uses SMD. We cannot observe the effects that PR would have on the number of parties; we can only make inferences based on the differences in the average number of parties we observe in majoritarian and proportional systems. These observed differences, however, may also be due to a number of social, cultural, or otherwise country-specific baseline variables whose impact on the number of parties may be hard to disentangle from the effects of electoral institutions. Our inability to answer this type of counterfactual question reveals that we do not know the exact causal effects of electoral rules.

new species of electoral institution? / 3

A recent development has provided some hope to scholars that the relative causal effects of SMD and PR may be more accurately estimated. In the last decade, the electoral system of choice for dictatorships turned democratic and for established democracies undergoing reform has been a system that combines majoritarian and proportional principles of representation in a single election. Mixed electoral systems are now in use in countries and subnational units on every populated continent. In the standard type of mixed electoral system, voters cast two ballots:14 with the first, they vote for a candidate in a majoritarian constituency; with the second, they vote for a party list. The rise in number and prominence of mixed electoral systems has not gone unnoticed by the scholarly community. Over the last few years, a “cottage industry” dedicated to the study of mixed systems has emerged. A substantial amount of research continues to be conducted on how mixed systems arise,15 affect strategic voting and entry,16 influence legislative behavior,17 and shape national party systems.18 In addition, some researchers (see Benoit 2001b; Moser 1997, 1999; Moser and Scheiner 2004; Stratmann and Baur 2002) have regarded elections held under mixed electoral rules as opportunities to perform “controlled comparisons.” Through the comparison of the outcomes generated in the majoritarian and proportional components of the election, “controlled experiments” evaluate the relative causal effects of SMD and PR in ways that cross-national studies cannot: they measure at the same time the effects of each type of electoral institution on the same unit and thus control for all observed and unobserved baseline variables. These studies purport to assess the independent effects of SMD and PR on voting, district-level party competition, and legislative behavior. Of course, studies that treat mixed electoral systems as social laboratories in which one can compare the separate effects of SMD and PR rely on a critical assumption: The majoritarian and proportional tiers must be independent; that is, the two components must operate exactly as “pure” forms of SMD and PR elections would in isolation from one another. Recent studies (Bawn and Thies 2003; Cox and Schoppa 2002; Herron and Nishikawa 2001; Mizusaki and Mori 1998), however, have demonstrated that interaction, not independence, is generally observed. In this chapter, we illustrate how the violation of the independence assumption affects the conclusions drawn from “controlled comparisons.” We describe the idea of “contamination,” outline its implications for the empirical analysis of mixed electoral systems, and preview the empirical findings presented in this book. More than any other previous work, this chapter provides a comprehensive discussion of “controlled comparisons,” the logic underlying such research designs, and how contamination

4 / mixed electoral systems

undermines their potential to provide valid causal inferences about the effects of SMD and PR. We show that interaction between the majoritarian and proportional components of mixed electoral systems gives rise to incentive structures that differ from those we would observe if the two tiers operated independently. In subsequent chapters, we show that mixed systems are more appropriately classified and studied as a distinct species of electoral institution rather than a subtype of SMD or PR. Relative Causal Inference and the Study of Mixed Electoral Systems As we noted earlier, many scholars believe that “controlled comparisons” precisely identify how different electoral institutions affect particular outcomes of interest because such studies hold every possible intervening variable constant. On the surface, this research design has considerable appeal, as it appears to offer a unique opportunity to directly observe the effects of SMD and PR on the same unit of analysis, or on units that seem to be homogeneous. Assessing causality in randomized and non-randomized social science studies is not a simple enterprise.19 Suppose we want to assess the effect of a particular “treatment” (say, the use of SMD) that is assumed to be amenable to manipulation, relative to another treatment (say, PR) on a dependent variable (like the effective number of electoral parties). In this case, our dependent variable (Y) is measured by calculating the mean of the district-level number of parties for each country (i) included in the study. For each unit, we can potentially observe two outcomes: ● ●

the mean number of parties if SMD were used (YiSMD); and the mean number of parties if PR were used (YiPR).

Except for exposure to the “treatment,” the unit on which we measure the effects of the two treatments is the same. Thus, any difference between YiSMD and YiPR can be attributed to the effects of the electoral system employed. If we could realistically observe both YiSMD and YiPR at the same time, we would have an easy and reliable way to measure the causal effect of SMD on the effective number of parties relative to PR: YiSMD ⫺ YiPR. In practice, however, it is impossible to measure YiSMD and YiPR simultaneously: one of the two quantities is always a counterfactual. Generally, social scientists get around the problem of causal inference through the application of statistical solutions (Holland 1986). In our example, this solution implies measuring the number of parties in two different groups of countries: one group using SMD and another group

new species of electoral institution? / 5

using PR. The average causal effect is the quantity estimated by computing the mean for each of the groups ( ySMD and yPR) and subtracting one from the other. Whether this estimate is correct, however, depends on a number of assumptions, some of which cannot be verified directly. Most importantly, the two groups must be constructed so that relevant baseline variables affecting the number of parties are similarly distributed in each group (see Reiter 2000), or are independent of the treatment assigned. For example, in their seminal work on the origins of modern party systems, Lipset and Rokkan (1967) describe how ethnic and ideological cleavages influence the adoption of alternative electoral systems. In their view, preexisting fractionalized party systems led to the adoption of PR, while more homogenous systems were the precursor to SMD. If we do not account for cleavage structure, the difference ySMD ⫺ yPR may not only measure the relative effects of the electoral system employed but also the impact of ethnopolitical cleavages. If this concomitant is not balanced among the groups, it would be virtually impossible to disentangle its effects from that of SMD and PR. The presence of an omitted variable20 correlated with both the treatment and the response has the potential to bias the results of the analysis. In this case, omitted variable bias leads to the possible overestimation of the causal effects of SMD on the number of parties, relative to PR. Simply controlling for such variables may also prove insufficient: if the covariates are distributed differently in the treatment and control groups, our inferences may turn out to be highly model dependent (see King and Zeng 2004). Though observational studies by definition prohibit randomization, we can prevent bias by creating groups that are balanced on the baseline variables affecting the number of parties. However, we still do not observe some of the variables that might affect the number of parties. Some of these factors may be unknown or impossible to measure. Thus, we cannot know whether such variables are balanced among the sample countries employing majoritarian and proportional rules. The level of accuracy of the average causal effect we estimate depends on how we specify the model, on the magnitude of the impact of the unobserved baseline variables, and on how unbalanced such variables are. Hence, we cannot know for sure how many parties Britain would have if it employed PR instead of SMD. We only have part of the answer to the question we posed in the introduction (“probably more”) but cannot predict the precise magnitude of the change. In the absence of manipulation and randomization, counterfactual judgment and causal inference are problematic (see Sekhon 2004). The recent adoption of numerous mixed electoral systems has provided scholars with the hope that, by comparing the outcomes produced by the two components of the election, country effects can be definitively disentangled

6 / mixed electoral systems

from institutional effects. Reed (1999, 257–258), for instance, maintains that “Japan’s new electoral system . . . is almost ideally designed for the study of strategic voting,” as “no other existing electoral system provides political scientists with data on such neatly separated but simultaneous SMD and PR votes.” Similarly, Stratmann and Baur (2002, 506–507) argue that the presence of two “legislator types exist[ing] simultaneously in Germany” allows “one to test the differing incentive effects between PR and FPTP [first-past-the-post] electoral systems.” Even more forcefully, Moser and Scheiner (2004, 2) characterize mixed systems as a “social laboratory in which effects of different types of electoral systems can be studied in isolation from influences of the social context such as social cleavages, socioeconomic development, or culture.” In all of these cases, scholars expect to identify the true causal effects of SMD and PR on the variables of interest. On the surface, this idea is seductive. In most mixed electoral systems, where SMD and PR elections are simultaneous and separate, we could compare the outcomes produced by majoritarian and proportional rules in the same district during the same election. The district is the same, and therefore every possible baseline variable has the same values in both types of observations. The fundamental problem of causal inference appears to be solved. Instead of making a more or less educated guess about the different outcomes that would occur if one electoral system were used instead of another, we actually see what happens, because we can observe the effect of SMD and PR on each unit.21 Mixed systems appear to offer a valuable opportunity to study the effects of different electoral rules, but this approach also has potential downsides. In particular, under mixed electoral rules, voters, legislators, and parties do not necessarily make strategic decisions for each tier separately, as if the two components were different elections. When parties elaborate nomination strategies, they know that seat maximization depends on their performance in both tiers; during a campaign, voters are simultaneously exposed to messages tailored for both the majoritarian and proportional contests; in a legislature, members of parliament elected in one tier work side by side with colleagues elected through different rules. The validity of “controlled comparisons,” then, hinges on an affirmative answer to another counterfactual question: is the outcome observed in the SMD component of mixed electoral systems the same outcome we would observe under “pure” SMD? Are the decisions of voters, parties, and legislators in the SMD component independent of those made in the other component (and vice versa)? In other words, does each component of the election function as if it existed in isolation from the other? If it is not so, then “controlled comparisons” are not measuring the relative causal effects of SMD and PR; they are measuring something else.

new species of electoral institution? / 7

Before we discuss whether or not the two components of mixed electoral systems function as two separate elections, it is useful to preview how the violation of the independence assumption would affect “controlled comparisons.” For any research design to yield meaningful causal inferences, the treatment assigned to one unit should not affect the response for the units exposed to a different treatment, even if the units are completely homogenous and all the relevant baseline variables are balanced among the two treatment groups that are included in the sample (see Cox 1958; 19–21; Reiter 2000, 31). In our example, the presence of a parallel PR tier should not affect the number of parties in the SMD tier. The violation of the “Stable Unit Treatment Value Assumption” (Rubin 1990) resulting from interference—what we call “contamination”—is a source of hidden bias that produces distortions in the estimates of relative causal effects. The problem of interference between treatments has long been recognized by scholars who focus on experimental research. Cox (1958) explains that most experimental designs assume that the measurement obtained when a unit is exposed to a specific treatment is equal to (1) a quantity that solely depends on the unit; and (2) a quantity that depends on the treatment. If the value of the observation is also affected by the treatment applied to other units, it becomes difficult to determine the causal effects of each treatment. If interference is ignored or can only be roughly approximated, causal inferences are likely to be biased.22 In the study of mixed electoral systems, however, it is unthinkable to isolate the units to prevent potential interference or contamination. The “treatments” of SMD and PR are either administered to all subjects within a unit (voters, districts, or countries) or selectively to different units; like legislators, who work together. Ironically, this feature was considered a key advantage of “controlled comparisons.” An alternative is to amend the assumption above and concede that each observation (say, the number of SMD candidates) consists of (1) a quantity depending only on the unit; (2) a quantity depending on the treatment used (SMD); and (3) a quantity depending on the treatment that is not used (PR). Doing so, however, implies that we explicitly take into account the possible effects of contamination and that we can measure how the treatment that is not applied affects values of the dependent variable (e.g., how PR affects the number of SMD candidates). If contamination is overlooked, “controlled comparisons” at best yield causal estimates based on unverified assumptions and, at worst, yield causal estimates that are wrong.23 Specifically, if institutional features of the SMD and PR components of mixed electoral systems affect outcomes in each of the tiers, these rules do not produce the same incentives and outcomes as pure forms of majoritarian and proportional elections. Thus, “controlled

8 / mixed electoral systems

comparisons” do not measure the relative causal effects of SMD and PR; quite possibly they also measure the effects of SMD–PR interactions on each component of the election, therefore biasing causal inference about the independent effect of each set of electoral rules. The Idea of Contamination Outside of the study of mixed electoral systems, the notion that electoral institutions produce interactive outcomes is not controversial. For instance, many scholars have demonstrated that such interactive outcomes are observed in presidential and parliamentary systems (Amorim-Neto and Cox 1997; Cox 1997; Jones 1995; Shugart and Carey 1992). The closer presidential elections are to legislative elections temporally, the more marked is the tendency toward two-party competition for the latter, particularly if the presidency is a coveted, “non-divisible” prize (Cox 1997, 189).24 The interaction between presidential and congressional elections in the United States is well documented: in on-year congressional elections, more voters show up to the polls and the president-elect’s party often receives a considerable seat boost (Ferejohn and Calvert 1984). In midterm elections, fewer voters turn out and the president’s party tends to lose votes, often in proportion to its surge in the previous election (Campbell 1966; Campbell 1991). Similarly, lower and upper house elections also may interact. If different rules are employed and the elections are concurrent, party competition in the elections to one chamber of parliament may be affected by the electoral system used for the other, as “one would hardly expect that the party systems for house and senate elections would fully adapt to their respective electoral systems, in splendid isolation from one another” (Cox 1997, 21).25 If elections to distinct political institutions interact, it should be unsurprising that two sets of electoral rules employed to elect the same legislative assembly affect one another. Given that the elections in the two components of mixed electoral systems are held simultaneously, it is reasonable to expect party strategy to be shaped by features of both tiers, voting decisions in one component to be affected by the choices made for the other portion of the election, and the floor behavior of one legislator type to be influenced by the coexistence with legislators elected through different rules. It is quite natural to expect relevant actors to make, in each tier, decisions that are to some extent shaped by incentives generated by the entire electoral system. This interaction is what we mean by “contamination.” Contamination is present, at the micro-level, when the behavior of a voter, a party, a candidate, or a legislator in one tier of the election is demonstrably affected by the institutional rules employed in the other tier. At the aggregate level, contamination is

new species of electoral institution? / 9

observed when a particular outcome produced in one tier (like the number of parties) is affected by the institutional features of the other tier. While much of the existing scholarship assumes that political actors compartmentalize the SMD and PR segments of the election, we find strong evidence for contamination. Our findings suggest that the assumption of independence typical of “controlled comparisons” is tenuous. Moser (1999, 369) recognizes this potential complication and acknowledges that the two components of the election should affect one another to some extent. Nonetheless, contamination “does not ruin the experiment” (Moser 1999, 369) as outcomes in each tier should still reflect the incentives produced by the mechanics of SMD and PR. More recently, Moser and Scheiner (2004, 2), have asserted that the majoritarian and proportional components “tend to approximate their expected effects” in spite of contamination. The interaction between the two sets of electoral institutions is found not to produce “entirely new outcomes.” In support of this claim, Moser and Scheiner (2004) present empirical support for a number of propositions. Most importantly, disproportionality is greater in SMD than it is in the PR tier and the effective number of electoral parties/candidates is lower in the majoritarian component. These results lead them to conclude that mixed electoral systems preserve the independent effects of SMD and PR. Their analysis, however, does not speak to the core criticism that the assumptions underlying “controlled comparisons” are faulty and may overestimate (or underestimate) the causal effects of SMD and PR systems.26 Controlled comparisons may hold constant intervening variables as they assess the behavior of parties or candidates. But, if voting behavior and party strategy in SMD are shaped by specific features of the PR component, or by how majoritarian and proportional institutions are balanced, “controlled comparisons” are not measuring the relative causal effects of SMD and PR. In each component, the effect of the electoral rules employed for that portion of the election is snarled with the effects generated by SMD–PR interactions. Mixed electoral systems need not generate “entirely new” outcomes for “controlled comparisons” to yield invalid estimates. In fact, it is not the mere presence or the size of the contamination effect that may bias inferences drawn from “controlled comparisons.” What complicates those causal inferences is that the magnitude of contamination is largely unknown. Hidden bias caused by interference can be eliminated by isolating the effects of a treatment on units that are administered different treatments (Cox 1958). For instance, we show in this volume that the placement of SMD candidates allows parties to boost their PR performance. As a result, the effective number of candidates observed in mixed systems is generally higher than that typically found in majoritarian elections, as parties have

10 / mixed electoral systems

the incentive to nominate SMD candidates even in districts they do not expect to win. In the context of this example, if we know that the existence of a parallel PR tier inflates the effective number of SMD candidates by, say, one candidate, we can still estimate the relative causal effect of SMD and PR by calculating the difference between the outcomes observed in the two tiers of the election and by then adjusting the difference for the value by which PR is affecting outcomes in SMD. It goes without saying, though, that we do not observe the value by which contamination is inflating the number of candidates in the SMD tier. To estimate it, we have to answer another counterfactual question: what would the number of SMD candidates be in a district if only SMD elections were held there? The fundamental problem of causal inference, which some had hoped controlled comparisons would resolve, returns to haunt us. While we can test for contamination, we cannot know with certainty how much contamination is affecting the number of SMD candidates, relative to the number we would find if there were no simultaneous PR elections. In fact, we do not observe what would happen if no separate PR tier existed. Any estimate of how contamination renders the outcomes observed in the SMD component different from those we would observe in pure first-past-the-post elections presupposes that we already know the relative causal effects of SMD, and can therefore predict how many viable parties would be able to compete in the district under purely majoritarian rules. If we could do that, however, “controlled comparisons” would be unnecessary, as the purpose of such studies is the very assessment of the effects of SMD relative to PR. If the purpose of “controlled experiments” is to show that SMD elections are characterized by lower proportionality and by a lower number of parties than PR elections, Moser and Scheiner (2004) are correct when they affirm that mixed electoral systems are a “social laboratory” where this relationship can be established, particularly because contamination works against finding such a result. Alas, the problem with this argument is that we already know that SMD results in higher disproportionality and in a lower number of parties; surely, we knew that long before mixed electoral systems ever became fashionable (see, for instance, Rae 1971; Sartori 1985; Taagepera and Shugart 1989). “Controlled comparisons,” in this sense, are only interesting to the extent that they provide point estimates of the magnitude of these causal effects. Given contamination, however, the accuracy of such predictions is in question, as are the results obtained from any experiment where the treatments interfere with one another. Perhaps, the focus on mixed electoral systems as social laboratories is misplaced and the study of mixed electoral systems can yield more interesting and useful results if a different research path is undertaken.

new species of electoral institution? / 11

A Few Words about Our Methodology We have written this volume as a comparative analysis of mixed electoral systems and their consequences. This is in stark contrast with most of the literature on mixed systems, which generally focuses on a single country or on small groups of geographically proximate countries. Like Sartori (1991, 243–244), we are committed to the idea that comparative politics “is a field defined by a method” and we are therefore convinced of the necessity to compare in order to control whether “generalizations hold across the cases to which they apply.” Each of the analytical chapters that make up this book, however, evaluates our generalizations about how mixed electoral systems work on different sets of countries. While chapter 8 considers every election that has taken place under hybrid rules in the 1990s and could be reasonably described as “free and fair,” the number of cases drops dramatically when we consider district-level data. When we look at legislative behavior, the analysis is limited to Italy and Ukraine. When we investigate voting, we restrict our attention to Japan and New Zealand. Our selection criterion is a straightforward one: in each chapter, we include every case for which we could acquire meaningful data. In every case, we have done our best to enrich our analysis of the effects produced by mixed electoral institutions with whatever knowledge we had or could acquire of the individual countries we discuss. Our purpose, however, is never to “just” understand the outcomes that mixed systems bring about in a single country. To the extent it is even possible (Sartori 1991, 244), when the dearth of available data or the nature of the question we are asking limit our discussion to one or two cases, our analysis seeks to be “implicitly comparative” and to evaluate propositions that are designed to travel beyond those countries’ borders. Our approach follows two simple guiding principles. On the one hand, we try to be bold and ambitious when we formulate our hypotheses—in Popper’s words (1989, 36), “risky predictions”—and we are not too bashful to admit that it is causal mechanisms that we seek to describe and causal relationships that we seek to find. On the other hand, we discuss our empirical findings with circumspection and candor, as we recognize that the research designs and methodology we have chosen have limitations that may prevent us from conclusively drawing the causal inferences we seek. Given that causal effects are essentially the difference between “what is” (an observed value of the dependent variable when a “treatment” of sorts is applied) and “what could have been” (the value of the dependent variable we would have observed in the same unit if no treatment were administered), and given that the estimation of a counterfactual may run into a plethora of well documented complications that may easily bias our inferences (see King and Zeng 2004), we have no difficulty acknowledging that sometimes the

12 / mixed electoral systems

data may not provide more than a moderate level of corroborative evidence for our propositions. In the empirical chapters that follow, we employ an eclectic set of quantitative methods to evaluate our hypotheses. In each chapter, we have done our utmost to identify the estimation technique that best fits the structure of the data we have and to find state-of-the-art solutions to the many challenges posed by our data. We do not focus on providing evidence that our explanatory variables of choice exert a nonzero effect on the dependent variable under consideration. We privilege the discussion and presentation of quantities of direct substantive interest over simple hypothesis tests. To generate such quantities of interest, we follow the logic outlined in King, Tomz, and Wittenberg (2000). Each expected value or predicted probability we present is computed by simulating M (generally 1000) sets of coefficients from their posterior distribution—whose mean and variance equal the coefficients and variance–covariance matrices estimated by our statistical models—and by then calculating the model’s systematic component (see King 1989) for particular values of the independent variables. We calculate the mean of the resulting distributions to obtain our point estimates and the relevant quantiles to describe how uncertain such estimates are. Where Clarify (see Tomz, Wittenberg, and King 2003) or Zelig (Imai, King, and Lau 2004) do not support the models we use, we wrote our own functions. Conclusion The research program that evaluates the consequences of contamination is still in its infancy. However, in many cases, these studies have shown that mixed systems tend to produce outcomes that differ, often quite notably, from those that would be expected under pure forms of majoritarian or proportional rules. Research on contamination has concentrated on three consequences of mixed electoral institutions: those inherent to party strategy, voting behavior, and legislative voting. In each of the analytical chapters that follow, we show that mixed electoral systems present parties (chapters 3–4), voters (chapters 5–6), and legislators (chapter 7) with incentives that differ from those typically associated with majoritarian and proportional elections. We begin our analysis in chapter 2 with a brief discussion of alternative definitions of mixed electoral systems, a description of some of their most salient characteristics, and a review of the explanations accounting for their rapid diffusion. In chapter 3, we employ district-level data from seven mixed systems to investigate how candidate placement strategies in the majoritarian component of the election affect a party’s performance in the PR tier. We find that the placement of a candidate in a majoritarian race boosts,

new species of electoral institution? / 13

in the same district, the PR vote share of the party with which the candidate is affiliated. By nominating a candidate in SMD, a party can attract greater attention to its policy objectives and hence receive a substantial vote bonus in the PR component of the election by drawing the support of voters who would have otherwise voted for other lists. In turn, the boost that parties can expect to receive by running their own SMD candidates provides them with an incentive to participate in majoritarian races regardless of their chances of winning a district. We find that the effective number of candidates in the SMD tier of mixed electoral systems is not constrained by the logic of Duverger’s Law to the same extent that it is in pure first-past-thepost elections. In other words, multiparty competition typically associated with PR spills over to the majoritarian component. Chapter 4 assesses a puzzle that arises from the analysis presented in chapter 3. The existence of contamination implies that parties should employ “go it alone” strategies in SMD by nominating candidates wherever possible to maximize the boost in their PR performance that candidate placement guarantees. Parties, however, have not consistently adopted this strategy. In many mixed systems, parties do not run in every district; coalitions and stand-down agreements restrict nominations. Through an analysis performed on district-level data from 13 mixed systems, we identify a number of institutional incentives that induce parties to form preelectoral alliances in SMD in spite of contamination. Our key finding is that strategic entry is primarily shaped by how majoritarian and proportional institutions are combined. Mixed systems dominated by the majoritarian component exercise a reductive impact on the number of entrants. In contrast, mixed systems dominated by the proportional component discourage coordination and encourage the proliferation of SMD candidacies. In chapters 5 and 6, our attention shifts to voters. These chapters explore the microfoundations for the propositions advanced in chapter 3. In chapter 5, we employ individual-level data from the 1999 New Zealand Election Study to systematically address how a voter’s party choice in the proportional tier of the election is affected by the dynamics of the competition in SMD. We formulate a theory at the level of the individual voter that explains how campaign information originating from an SMD race spills over to the proportional component, where it is used by some voters in their choice of a party list. We find that strong SMD candidates have coattails, as the quality of SMD candidates has a considerable impact on party choice in PR. In chapter 6, we turn to voting behavior in the majoritarian component using survey data from Japan. In regular SMD systems, voters often abandon their most preferred candidate if the candidate is perceived as having little chance of winning and vote for a less preferred candidate to maximize the expected utility. If the effects of SMD and PR on voters work independently,

14 / mixed electoral systems

such strategic voting should occur in the SMD component of mixed systems to the same degree. We find, however, that the incidence of strategic voting is far lower in Japan’s SMD races than in pure SMD systems like the US or Britain. The final two chapters investigate how mixed electoral institutions affect party systems between elections. In chapter 7, we evaluate how mode of election influences legislative behavior, particularly party loyalty. Most of the research on this topic assumes that legislators “specialize” in either SMD or PR, creating conditions in which legislative voting behavior is exclusively shaped by the electoral incentives provided by the tier in which members of parliament are elected. Consistent with this logic, representatives elected through PR, who must please their party bosses to maximize their chances of reelection, should vote differently from their party less frequently than their colleagues elected in SMD. Through the analysis of electronic voting data from the Ukrainian and Italian legislatures, we show that representatives elected through SMD and PR do not differ in their propensity to stick to the party line. Rather, we suggest that a legislator’s strategic calculus is based on incentives that are more complex than those provided by the type of seat they occupy. Factional membership, variables related to “electoral path” (like relative seat safety and dual candidacy), and coalitional dynamics determining how candidates compete for ballot access substantially affect a legislator’s voting behavior. Chapter 8 examines the fragmentation of the party system by exploring how institutional features of mixed electoral systems affect the number of legislative parties. The literature on mixed electoral systems has largely neglected the consequences of different types of SMD–PR combinations on national patterns of representation. Few scholarly works have emulated the cross-national research that Amorim-Neto and Cox (1997), Ordeshook and Shvetsova (1994), and Benoit (2002) have conducted on the determinants of national party systems under SMD and PR. Using a new dataset (Golder 2005), we evaluate how different mixtures of the SMD and PR ingredients affect the overall fractionalization of a legislature as well as the party systems generated at the national level. The findings presented in this book show that the interaction of the majoritarian and proportional components of mixed electoral systems creates incentive structures that differ from those that shape the decisions of voters, parties, and legislators in pure SMD or PR.

C h ap t e r 2 A Taxonomy of Mixed Electoral Systems

The “father of taxonomy,” the eighteenth-century biologist Carl Linnaeus, organized the living world by collecting together and labeling similar organisms into increasingly detailed classes. Although he initially believed that the number of species—the most specific category of living creatures— was fixed, he later recognized that hybrids of existing species could produce new, previously unseen ones.1 The emergence and proliferation of mixed electoral systems has generated a similar quandary in the election studies literature. As we noted in chapter 1, some of the analytical literature on mixed electoral systems has emphasized their similarity to existing “species” of election rules. The contamination literature, however, suggests that mixed electoral systems constitute a new species of electoral rules, emerging from the miscegenation of majoritarian and proportional systems. How do we distinguish among different forms—or species—of election rules?2 Although Linnaeus classified plants by their sexual organs, psephologists are limited to less provocative features, most notably district magnitude, ballot structure, and electoral formula. The number of seats allocated to a district (M), has been emphasized because of its decisive role in determining proportionality (Taagepera and Shugart 1989) and its connection to Duverger’s propositions (Duverger 1954). Based on district magnitude, electoral institutions are generally divided into M ⫽ 1 and M ⬎ 1 systems. M ⫽ 1 systems encompass presidential elections3 and parliamentary elections in which representatives are selected in constituencies delimited geographically. M ⬎ 1 systems allow more than one official to be elected in a district and are typically characterized by more proportional outcomes. Ballots in M ⫽ 1 or M ⬎ 1 systems may be ordinal or categorical,4 and may allow voters to express a single preference or multiple preferences. The formulae employed to translate votes into seats also vary greatly. They may award seats based on plurality or majority support, as in majoritarian systems, or they may entail more complicated calculations using remainders or divisors,

16 / mixed electoral systems

as in proportional systems. Dividing election systems by these three features generates great diversity. However, students of electoral systems have generally simplified their taxonomies, collecting together “proportional,” “semi-proportional,” and “majoritarian” rules. The mechanics of mixed electoral rules complicate this straightforward approach to classification. This chapter begins to address a key question of our book: are mixed electoral systems—a hybridization of “pure” election rules—part of an existing “species” of election rules, or are they a new “species” of election rules? To answer this question, we must subject mixed systems to a series of empirical tests. Before we can begin these tests, however, we must explicitly state the conditions that are necessary for an electoral system to be classified as mixed. We identify our criteria for mixed systems and discuss their origins in this chapter, setting the stage for our empirical tests in subsequent chapters. Characteristics of Mixed Systems Institutional variation, rather than conformity, characterizes the world’s existing mixed electoral systems. Many scholars have assessed this variation and have developed different definitions and taxonomies. In their classification of all existing families of electoral rules, Reynolds and Reilly (1997, 19) consider systems to be mixed when they “use both PR lists and plurality– majority (‘winner takes all’) districts. . . .” In the first taxonomy developed solely for mixed systems, Massicotte and Blais (1999, 345) identify rules as mixed when they involve “the combination of different electoral formulas (plurality or PR; majority or PR) for an election to a single body” [emphasis in original]. In the first major book-length treatment of mixed systems, Shugart and Wattenberg (2001a, 10) provide a more restrictive definition: mixed systems are a form of “. . . multiple tier systems, with the specific proviso that one tier must entail allocation of seats nominally whereas the other must entail allocation of seats by lists” [emphasis in original]. These definitions are similar, but they contain important nuances that affect the classification of existing systems. The primary differences among these definitions are the treatment of formulae and spatial application of the rules. Although they note that the nominal tier need not be plurality or majority, Shugart and Wattenberg (2001a, 11) classify systems using single non-transferable vote (SNTV) (Japan’s Upper House and Taiwan) as a “distinctly different branch” of mixed systems. Massicotte and Blais incorporate the broadest combination of formulae into the mixed system family,5 including systems that apply different formulae to districts that do not overlap geographically (i.e., France’s Senate). Consequently, the list of systems classified as mixed varies across these definitions.

taxonomy of mixed electoral systems / 17

While scholars classify specific mixed electoral systems differently, they concur on an essential feature: seats are allocated by more than one formula. Accordingly, we adopt a rather minimalist definition. An electoral system is “mixed” if more than one formula is employed to distribute legislative seats. Corresponding with most definitions of mixed systems, we qualify the combination of formulae that creates a mixed electoral system. One formula is proportional and seats are allocated on the basis of the votes received by party lists. The other formula may be semi-proportional or majoritarian, but votes are cast nominally for individual candidates. We refer to the semi-proportional or majoritarian component as the “SMD tier”6 and to the proportional component of the election as the “PR tier.” Although most mixed systems employ first-past-the-post in the SMD tier, we should bear in mind that this tier may be associated with plurality or majority formulae and single-member or multimember districts. Systems combining PR and SMD, majority-runoff, SNTV or similar rules would be mixed. But under our definition, an election with regional PR that employs Largest Remainders-Hare in some districts and D’Hondt in other districts would not be considered mixed.7 Further, an election system with half of the seats selected through SMD-plurality and the other half through majority runoff would also not qualify as mixed. Mixed systems are characterized by great institutional variation and profound differences in the outcomes they typically produce. Hence, our definition encompasses many different formulaic structures whose mechanics “mix” different formulae and produce diverse combinations of majoritarian and proportional principles of representation. We identify five dimensions on the basis of which such variation can be understood. Though this list is not exhaustive, we believe the following characteristics capture many of the important institutional features that distinguish different types of mixed systems. ● ● ● ● ●

Geographic application of the rules; Ballot structure; Dual candidacy; Proportion of seats allocated to SMD or PR; and Seat/vote linkage.

First, the geographic application of the rules varies across the electoral systems that encompass multiple allocation formulae. Most mixed systems apply multiple formulae to the whole geographic space (a country or region), which is divided into overlapping constituencies. Russia’s Duma, for instance, divides seats evenly: 225 legislative seats are distributed nationally through PR, and 225 seats are distributed through SMD with plurality

18 / mixed electoral systems

rules. The Welsh Assembly, by contrast, distributes PR seats regionally and subdivides each region into a number of SMD plurality districts for the majoritarian ballot. In most mixed systems, majoritarian and proportional elections are held throughout the national territory. However, some mixed systems apply different formulae in specific regions of the country. The electoral rules in France (Senate), Greece, Niger, Madagascar, Panama, and Zimbabwe have historically (or currently) divided these countries by formula. In the election of the French Senate, for example, 206 deputies are selected in geographic districts using majority-plurality rules, whereas 98 deputies are elected in 15 multimember districts using the D’Hondt formula. Similarly, in Niger and Panama, some regions exclusively employ SMD and some use PR. Greece, Madagascar, and Zimbabwe used, but ultimately abandoned, their own “coexistence” systems.8 Second, mixed systems are distinguished by different ballot structures. Mixed electoral systems may be divided into “single” and “dual” ballot, depending on the number of votes that each voter may cast. In dual ballot systems, the most widely used variant, voters cast two votes. With the first, they generally select a candidate competing in the majoritarian component of the election;9 with the second, voters typically choose a party list.10 In single ballot mixed systems, voters only cast one vote, by which they select a constituency-based candidate. However, the vote is “nonexclusive” and “pooled” (see Cox 1997) because it affects seat allocation in both components of the election. The votes received by all candidates running under the same party label are aggregated for the purpose of distributing PR seats. Germany’s initial postwar rules provided a single ballot, linking the SMD and PR votes. After the 1949 election, ballots were decoupled—voters cast a separate vote for a political party and a candidate in the district, allowing German voters to split tickets. More recently, Albania, Italy, Mexico, Senegal, Seychelles, South Korea, Taiwan, and Tunisia have given citizens a single vote, but allocated seats with multiple formulae.11 Third, in addition to restricting the number of ballots, some rules forbid candidates from contesting seats in more than one tier. In Italy and Japan, for example, candidates may compete in both the majoritarian and proportional segments of the election. In these cases, dual candidacy consolidates incumbency advantage (McKean and Scheiner 2000) and guarantees the election of prominent party leaders. Ukraine, by contrast, adopted restrictions on dual candidacy after independent candidates complained about unfair advantages given to affiliated candidates. The Constitutional Court of Ukraine ruled that dual candidacy violated constitutional guarantees of equal rights for all citizens by allowing party-list candidates to participate in both components of the election while independent candidates could not. Dual candidacy was therefore prohibited in the 2002 parliamentary election.12

taxonomy of mixed electoral systems / 19

Fourth, while all mixed systems allocate seats through more than one formula, there is notable variation in the relative importance that different mixed electoral rules confer upon the components of the election. Some legislatures balance the proportion of seats distributed through SMD and PR. In other mixed systems, one component drives the outcome of the election. In some cases, a small share of seats is assigned through one formula, while the vast majority is distributed by different allocation rules.13 Israel, where all seats but one (the prime minister) were selected via proportional representation in 1996 and 1999, has been cited as an extreme form of mixed system (Shugart and Wattenberg 2001a).14 Similarly, Ecuador sets aside only 10 percent of the legislative seats for majoritarian elections, while 90 percent of the legislature is elected from party lists. At the other end of the continuum, 85 percent of the South Korean Kuk Hoe and 75 percent of the Italian Chamber of Deputies are elected through SMD. Most mixed systems do not exemplify the extremes, however. The median score for the percentage of seats allocated to PR in all mixed systems is 46 percent; the mean is 43 percent and the standard deviation is 20 percent.15 The linkage among tiers is the fifth feature that differs among countries with multiple formulae. This mechanism is often considered to be the most important form of variation in mixed systems (Ferrara and Herron 2005; Moser 2001; Nishikawa and Herron 2004). In fact, it is the basis of Shugart and Wattenberg’s (2001a) distinction between “mixed-member proportional” (MMP) and “mixed-member majoritarian” (MMM) systems. In MMP systems, the proportional ballot is used to compensate parties for poor performance in the nominal ballot. A party’s performance in the proportional election is used as a benchmark to determine overall seat allocation, guaranteeing the total number of seats that a party is assigned depends upon its performance in PR elections. District seats are factored into the calculation of PR seats, reducing the number of seats acquired in PR by strong SMD performers (see the example below). This feature characterizes mixed systems in Germany, Bolivia, New Zealand, Scotland, Wales, and elsewhere. Italy and Hungary employ vote transfers rather than seat transfers to link the tiers. A party’s PR vote is increased by the number of votes received by its losing SMD candidates (as in Hungary) or reduced by the number of votes received by its winning SMD candidates (as in Italy).16 Hungary also uses three formulae to allocate seats: majority runoff, regional PR, and vote transfers to a top-off list. Table 2.1 shows an example of compensatory seat distribution in the Mid and West Wales region’s four-seat PR tier in the 1999 elections. In the first and second columns, the parties are identified along with their regional PR vote totals. In the first round, the parties’ vote totals are divided by the number of seats they received in SMD, plus one. The first seat is allocated

Table 2.1 Distribution of seats in mid and west Wales region (1999) PR Vote

SMD Seats

Seat 1

Seat 2

Seat 3

Seat 4

Total Seats

Conservatives Green Ind. Turner Labour Liberal dems. NLP Plaid Cymru Soc. labour

36,622 7,718 1,214 53,842 31,683 705 84,554 3,019

0 0 0 2 2 0 4 0

1 0 0 0 (17947) 0 (10561) 0 0 (16911) 0

1 (18311) 0 0 0 0 0 0 0

0 (12207) 0 0 1 0 0 0 0

0 0 0 0 (13461) 0 0 1 0

2 0 0 3 2 0 5 0

Data Source: BBC (http://news.bbc.co.uk).

20

Party

taxonomy of mixed electoral systems / 21

to the party with the highest total after dividing through by the divisor: the Conservative Party. In the second round, the Conservative Party’s divisor is raised by one, its index is recalculated, and the next seat is allocated. Again, the Conservative Party wins the seat. In the next round, the Conservative Party’s total is adjusted and the seat is allocated to the Liberal Democratic Party. The final seat is given to Plaid Cymru. This process is repeated for each PR district. In noncompensatory mixed systems, no formal linkage exists between tiers of allocation. Japan, Lithuania, Russia, and other countries employ noncompensatory MMM systems.17 Each party acquires the sum of the seats it received in the majoritarian and proportional components of the election. For example, eight parties in Ukraine overcame the 4 percent threshold in the PR component in 1998. They were thus eligible for seat allocation in the PR tier, which uses the Largest Remainders-Hare formula. To determine how many seats each party receives, a quota is calculated by dividing the number of eligible votes by the number of seats to be distributed. Votes for parties that did not pass the threshold are excluded from seat distribution. Each party’s vote total is divided by this quota, and the resulting whole number determines the number of “full quotas” the party receives. Table 2.2 shows the PR vote received by parties passing the 4 percent barrier, their quotas and the total number of seats they received in the Rada. Based on the initial seat distribution, 221 of the 225 seats are allocated to parties. The last four seats are allocated based on the fractional component of the parties’ number of quotas. The first seat is awarded to Hromada because it has the largest remainder (0.99). Subsequently, seats are allocated to the Progressive Socialist Party (PSP), the Social Democratic Party of Ukraine (United) (SDPU(o)) and the Green Party.18 Table 2.3 displays the total seat allocation in Ukraine’s 1998 election. PR seat allocation, as noted earlier, is unaffected by SMD seat acquisition in Table 2.2 PR seat distribution in the Ukrainian Rada (1998)

KPU Rukh S&P bloc Greens NDP Hromada PSP SDPU (O) Total

% PR

# of Quotas

24.65 9.40 8.56 5.44 5.01 4.68 4.05 4.01

84.31 32.15 29.27 18.59 17.14 15.99 13.84 13.72

Full Quotas

Data Source: Central Electoral Commission of Ukraine.

84 32 29 18 17 15 13 13 221

Remainders 0 0 0 1 0 1 1 1 4

# of Seats 84 32 29 19 17 16 14 14 225

22 / mixed electoral systems Table 2.3 Total seat distribution in the Ukrainian Rada (1998)

KPU Rukh S&P bloc Greens NDP Hromada PSP SDPU(O) APU Other Parties Independents Total

% PR

# of PR Seats

24.7 9.4 8.6 5.4 5.0 4.7 4.0 4.0 3.7

84 32 29 19 17 16 14 14 0

% SMD 14.7 6.3 4.5 0.8 4.2 3.7 1.0 1.9 3.3 46.9

225

# of SMD Seats

Total Seats

37 14 5 0 11 8 2 3 9 20 111 220

121 46 34 19 28 24 16 17 9 20 111 445

Note: Five seats were not filled in SMD until after the election. Data Source: Central Electoral Commission of Ukraine.

noncompensatory systems. For some parties, SMD seat allocation is relatively proportional to the total vote received in SMD: Rukh received 6.3 percent of votes and 6.4 percent of seats. But, allocations are notably disproportional for other parties: the Socialist and Peasant Bloc received 4.5 percent of the vote, but only 2.3 percent of the seats; the Green Party received 0.8 percent of the vote and no seats; the APU was overrepresented, receiving 4.1 percent of the seats, but 3.3 percent of the vote. Of course, SMD rules are not designed to proportionally allocate seats, so these disparities are not surprising. Noncompensatory systems, however, do not “rescue” parties that underperform in SMD. For example, the Green Party received 5.4 percent of the party-list vote, but only 4.2 percent of the overall seats. It is likely that the Green Party’s poor SMD performance would have been attenuated by the acquisition of additional PR seats in a compensatory system. While the allocation disparities in Ukraine are relatively small, larger gaps can occur in noncompensatory mixed systems. Disproportionality is magnified in systems where most of the legislative seats are allocated through SMD. Given that the use of multiple allocation formulae is our guiding principle in the classification of mixed electoral systems, what countries are part of the universe of mixed systems? Table 2.4 identifies countries that employed mixed electoral rules for national legislative elections during the period 1991–2004. The table also shows the categorization of these rules by three major taxonomies. Inconsistencies across these classifications are due to various factors. First, the coding rules for mixed systems differ, as noted earlier. Reynolds and Reilly (1997) define systems with proportional

taxonomy of mixed electoral systems / 23

outcomes (with seat or vote transfers) as MMP. Rules that do not link tiers for allocation are called Parallel. Massicotte and Blais (1999) identify systems with seat linkages or vote transfers as Corrective. Those without formal linkages are called Superposition if the rules are applied across the country’s entire geographic space. Coexistence systems apply one set of rules in a region of the country and another set of allocation rules in the rest of the country. Supermixed systems combine aspects of compensatory and noncompensatory seat allocation mechanisms. Shugart and Wattenberg (2001a) divide mixed systems on the basis of seat linkages: those countries with seat transfers are labeled MMP; those without are labeled MMM. Italy and Hungary, which use vote transfers, are labeled MMM with partial compensation. Second, the timing of publication affects coding. In some cases, notably Azerbaijan, Bulgaria, Croatia, Kyrgyzstan, Macedonia, and Ukraine, mixed systems were adopted and later abandoned. In other cases, such as Lesotho, mixed systems were adopted recently. Third, some scholars, like Shugart and Wattenberg (2001), focus exclusively on electoral democracies. Table 2.4 List of mixed systems (1990–2004)1 Countries2

Reynolds and Reilly (1997/2002)3

Massicotte and Blais (1999)

Shugart and Wattenberg (2001)

Albania Andorra Armenia Azerbaijan Bolivia Bulgaria Cameroon Chad Croatia Ecuador France4 Georgia Germany Guatemala Guinea Hungary Israel Italy Japan5 Kazakhstan Kyrgyzstan Lesotho Lithuania Macedonia

Parallel—TRS Parallel—Block Parallel—FPTP Parallel—TRS MMP –– Parallel—FPTP TRS Parallel—FPTP Parallel—PB –– Parallel—TRS MMP Parallel—FPTP Parallel—FPTP MMP PR MMP Parallel—FPTP –– –– –– Parallel—TRS TRS

Corrective Superposition Superposition Superposition Corrective –– Supermixed Supermixed Superposition Supermixed Coexistence Superposition Corrective –– Superposition Supermixed — Corrective Superposition — — — Superposition Superposition

–– –– MMM –– MMP –– –– –– –– –– –– MMM MMP –– –– MMM/Partial Compensation Direct PM MMM/Partial Compensation MMM –– –– –– MMM MMM

Continued

24 / mixed electoral systems Table 2.4 Continued Countries2

Reynolds and Reilly (1997/2002)3

Massicotte and Blais (1999)

Shugart and Wattenberg (2001)

Mexico6 New Zealand Niger Panama Philippines Russia Senegal Seychelles Somalia South Korea Taiwan Tajikistan Thailand Tunisia Ukraine7 Venezuela

MMP MMP List PR and FPTP List PR and FPTP Parallel—FPTP Parallel—FPTP Parallel—PB Parallel—FPTP Parallel—FPTP Parallel—FPTP Parallel—SNTV –– Parallel—FPTP Parallel—PB Parallel—FPTP MMP

Corrective Corrective Coexistence Coexistence Corrective Superposition Superposition Superposition –– Superposition Superposition –– –– Corrective Superposition Corrective

Limited MMM MMP –– –– MMM MMM –– –– –– MMM –– –– MMM –– MMM MMP

PB ⫽ Party Block Vote (voters choose parties and not candidates) TRS ⫽ Two round system (or majority-runoff ) 1 Sub-national units such as Scotland and Wales, and many German länder and Russian regions use mixed systems to select local representatives. We include only state-level units on this table. In addition, some proportional systems include special seats for ethnic minorities (i.e., Slovenia). These systems are not included in our definition as mixed. 2 Azerbaijan, Bulgaria, Croatia, Macedonia, Kyrgyzstan, and Ukraine adopted and later abandoned mixed electoral systems. 3 The classification is based upon Reynolds and Reilly’s book (1997) and the modified manuscript (2002) found on the International IDEA website at: http://www.idea.int/esd/data/world.cfm. 4 Coexistence applies to the Senate. Municipal elections are categorized as Fusion systems in which multiple allocation rules are applied at the district level. 5 We include data from lower and upper house elections. 6 The Mexican lower house is included. 7 Ukraine’s parliament voted in early 2004 to abandon its mixed system. However, the status of this institutional reform was unsettled at the time this manuscript was written.

Countries that are considered to be “illiberal” democracies or authoritarian states are not included in their analysis. By our liberal count of mixed systems, 40 states used these rules at the national level during the period 1990–2004. This figure constitutes slightly more than 20 percent of the countries in the world. In addition, mixed systems are found on every populated continent, including 9 in Africa, 12 in Asia, 1 in Oceania, 12 in Europe, 3 in North America, and 3 in South America. The expansion of these rules is a global phenomenon that occurred in a relatively short time frame; most mixed systems were selected during the decade of the 1990s. What accounts for the diffusion of this family of electoral institutions?

taxonomy of mixed electoral systems / 25

Origins of Mixed Systems The creative minds of institutional designers have shown a seemingly limitless capacity to innovate. In the eighteenth and nineteenth centuries, intellectuals from many disciplines sought formulae that would better translate individual preferences into aggregate outcomes.19 Many of these innovators became giants in electoral design, with their names immortalized in the elections nomenclature. The Marquis de Condorcet advocated pairwise competition among candidates; Charles de Borda created a weighted rank-order method of voting and decision making; Thomas Hare invented the Single Transferable Vote; Victor D’Hondt developed party-list proportional representation; and many others expanded the range of choices available to electoral systems designers (Farrell 2001; Reilly 2001). Electoral pioneers also created new methods of casting and counting ballots in the twentieth century. One of the most notable inventions of that century was the mixed electoral system. Various combinations of party lists and personalized ballots were tested and discarded in Denmark, Iceland, France, Italy, and Romania before mixed systems became a permanent item on the electoral systems menu after World War II (Massicotte and Blais 1999). Germany adopted the world’s oldest, continuously employed mixed electoral system in 1949. Although the rules have changed many times, the principle of allocating seats by more than one formula has been preserved over the years. Mixed electoral systems were regularly classified as anomalies until the 1990s. When they were empirically rare, mixed systems were not identified as a separate category (Bogdanor 1983; Butler 1982; Grofman and Lijphart 1986; Nohlen 1984; Rae 1967; Rose 1983; Sartori 1968, 1986). They were often classified as consensus-type systems, even during the initial period of expansion in the early 1990s (Lijphart 1994).20 The nearly simultaneous collapse of communism in East Central Europe and Eurasia and the wave of electoral reform that swept through many long-standing democracies created demand for election rules that promised to solve the problems facing new and established democracies alike. Mixed systems were touted as a solution to many shortcomings of pure SMD and PR rules and quickly proliferated across the globe. By the turn of the twenty-first century, Germany was no longer an exception, but the vanguard of a new electoral order. Germany’s mixed electoral system was adopted while the state was under occupation by Allied forces in the wake of World War II. Prior to the codification of a mixed electoral system at the national level, German Länder adopted rules for local elections. Proportional representation was employed in some occupation zones, while a mixture of proportional and majoritarian rules was used in other regions. Subsequently, the Parliamentary Council,

26 / mixed electoral systems

representing parties that had won seats in local elections, debated and adopted a federal election law. Germany’s 1949 law allocated 60 percent of Bundestag seats to SMD races and 40 percent to elections decided by party list votes in the Länder. Voters would cast a single ballot for a candidate in the majoritarian race, but this vote also would be used to award PR seats. In 1953 and 1956, the law was modified to allocate an equal proportion of seats to PR and SMD, to include a federal threshold of 5 percent (or three SMD districts) for participation in PR seat allocation, to provide voters two ballots, and to modify other aspects of the rules. Seat allocation remained compensatory; a party’s total number of Bundestag positions was determined by the party-list vote. Seats gained in majoritarian constituencies were subtracted from the total number of seats allocated to a party based on its PR vote, producing a fully proportional distribution of seats in a parliament composed of some members elected in local constituencies and others representing national constituencies (Bawn 1993; Scarrow 2001). By the 1957 election, Germany’s politicians had adopted what would become the most durable mixed electoral system to date. While Germany is generally identified as the archetypal mixed system, many crucial features of its system—seat linkages, dual ballots, a balanced division of seats into party list, and SMD-plurality districts—were modified by new members of the mixed systems “club.” The growing heterogeneity of institutional rules collected under the umbrella of “mixed systems” prompts us to evaluate how mixed systems spread so far and so quickly. Why have institutional designers selected mixed systems in both established democracies and transitional states? The scholarly literature has not reached consensus about the mechanisms by which election rules are modified over time, although some efforts to find generalizable patterns exist (see Benoit 2004). The proliferation of mixed systems in the 1990s and 2000s prompted further assessments of electoral system choice, primarily in country-specific case studies. While these case studies were often part of explicit efforts to understand the adoption of mixed electoral rules, they did not produce a generalizable model. However, we can extract general explanations for the design of mixed systems from the single-country analyses. One potential explanation for the selection of mixed electoral rules is a consensus on normative goals, or the normative preferences of strong political entrepreneurs. In other words, mixed systems may have been selected because of attachments to the “best of both worlds” principle. Some scholars have applied this explanation to Germany, stating that the mixed system was selected “to meet the chief objections levied both against single-member constituencies and against proportional representation”.21 Analyses of the design process in Russia have also suggested that political

taxonomy of mixed electoral systems / 27

entrepreneurs with a strong commitment to multipartyism and mixed systems swayed the decision-making process. The influence of such actors was enhanced by the uncertainty inherent in periods of transition (McFaul 1999; McFaul 2001). In Bolivia, parties attempted to tinker with the election rules to their benefit, but ultimately adopted a mixed system to improve representation and promote political modernization (Mayorga 2001). In these cases, the preference for mixed systems evolved primarily from interpretations of the “common good.” By combining majoritarian and proportional representation rules, mixed systems would promote party formation in new democracies, encourage both national and local representation, ensure accountability to local and national constituencies, and provide voters with better opportunities to express their preferences than in “pure” systems. The proliferation of these systems could be seen as the triumph of a design process taking place under a Rawlsian “veil of ignorance,” or the result of opportunism with a positive externality—the choice of a system to maximize benefits to all. Alternatively, mixed systems may have been selected through idiosyncratic processes that were contextually driven. Some factors contributing to Germany’s adoption of a mixed electoral system suggest a unique confluence of events, most notably the failed Weimar Republic’s use of a highly proportional system, the lack of experience with proportional representation among the occupying powers (especially the United Kingdom and United States), and the lessons learned from rule-making in the Länder. German election rules were crafted through bargaining, but “the system’s features evolved from political compromises arising out of the postwar constellation of partisan forces” (Scarrow 2001). In the postcommunist world, the selection of alternative electoral rules may have reflected the distribution of power in the transition process. Institutional design was initiated exogenously; the opportunity to design new electoral systems was a consequence of authoritarian collapse. However, the decision-making process that produced many mixed electoral systems was endogenous. Specifically, the power wielded by communist parties relative to the opposition may have driven the choice for majoritarian, proportional, or mixed rules. Some scholars argue that the specific configuration of the electoral rules emerged from bargaining processes that involved communist and opposition parties whose preferences were assumed to be fixed cross-nationally: communists generally preferred SMD and opposition members preferred PR. The relative strength of communist parties and the opposition across postcommunist space yielded majoritarian rules where communists dominated and PR where the opposition prevailed. Mixed systems were the result of compromises in bargaining environments characterized by stalemate (Dawisha 2000).

28 / mixed electoral systems

Interest in electoral reform in many long-standing democracies may have arisen due to mounting concerns about representation and political scandals. In the late 1980s and the early 1990s, the Japanese government was under constant pressure from the general public to carry out political reform because of repeated incidences of corruption. It reformed the electoral rules in part because some scholars suggested that intra-party competition, flowing from the nomination of multiple candidates to districts, promoted corruption. Japan’s dominant party, the Liberal Democratic Party (LDP), tried to adopt a mixed system that would give the party a substantial seat bonus. LDP leaders expected the system to produce two-party competition in which the LDP could maintain its majority party status for the foreseeable future. Although a standard SMD system was more attractive to most LDP members, a mixed system with an SMD component was proposed because it was more likely to pass through both houses of parliament. Having lost its majority in the Upper House of the Diet in 1989, the LDP needed the support of other parties to pass an electoral reform bill. In other cases, mixed systems were not adopted until a “shock” to the system created an opportunity for change. In New Zealand, a Royal Commission recommended MMP rules to correct problems of representation. The ruling Labour Party did not support this plan, but by 1993 New Zealanders selected mixed electoral rules in a referendum. Why was the referendum held in the face of opposition by the political elite? The decision to submit the choice of an electoral system to the people of New Zealand was prompted by a political error; the prime minister mistakenly contradicted his party’s preferences by promising to hold the binding vote in a televised debate (Denemark 2001). This explanation suggests that randomness may also influence the process of election system choice. Political actors would not have selected a mixed system but for some exogenous event, such as a public blunder, that forced their hands. Mixed systems may also have spread due to bargaining among selfinterested politicians. Much of the scholarly work on the origins of mixed electoral systems in transition countries points out that irrespective of whether the new electoral rules are introduced by majority rule, presidential decree, or inter-party bargaining, the choices reflect the interests of the parties involved, which are in turn based on their expected distributive gains. The findings presented by Bawn (1993), Remington and Smith (1996), and Benoit and Schiemann (2001) run counter to the thesis that new institutions in transition countries are shaped by historical precedent or determined by structural, social factors independent of the preferences of relevant actors. This research also indicates that mixed electoral systems are likely to arise during transitions from authoritarian rule under uncertainty because (1) the diverse combinations of SMD and PR that can be realized

taxonomy of mixed electoral systems / 29

in mixed systems offer the opportunity for parties to compromise and achieve mutually acceptable solutions through bargaining and (2) mixing SMD and PR allows institutional designers to hedge their bets and minimize the risks of being excluded from power when the uncertainty about their popularity prevents them from identifying which “pure” electoral system is most likely to provide the distributive gains they seek. Mixed electoral systems, however, were not only introduced in countries undergoing a transition from autocracy. To assess self-interest as a motivation for the adoption of mixed systems, three questions must be addressed. First, what makes mixed systems preferable to the status quo? Second, what motivates parties that are clearly advantaged by the electoral rules already in use to initiate or go along with the reform? Third, how do dominant parties attempt to preserve their position of strength in the design of a new electoral system? Shugart (2001) attempts to answer the first question by describing the “efficiency” of mixed electoral systems. On the inter-party dimension, efficiency is defined as the combination of sufficiently broad representation and high preelectoral identifiability of potential governing coalitions. On the intra-party level, efficiency denotes accountability to parties and local constituents. Shugart argues that electoral systems in pre-reform Italy, Japan, New Zealand, and Venezuela were inefficient. Italy’s hyperrepresentative system (see Katz 1995 and 2001), Japan’s pork-ridden hyperpersonalistic system (see Reed and Thies 2001), New Zealand’s pluralitarian system (see Denemark 2001; Nagel 1994), and Venezuela’s hypercentralized system (see Crisp and Rey 2001) were reformed through the introduction of mixed electoral rules that significantly improved efficiency by increasing identifiability, proportionality, responsiveness to party leaders, and accountability to constituents, respectively. Mixed electoral systems may be perceived to be preferable to the existing electoral rules because of the higher efficiency that different SMD–PR combinations may guarantee. Shugart also argues that political crises make it rational for dominant parties to initiate the transformation of electoral rules. Corruption scandals, the growth of an antiestablishment, separatist party, and economic stagnation led to a referendum on the electoral law in which Italian voters vigorously signaled their desire for change (Katz 2001). Voters compelled governing parties to craft an electoral system that was expected to increase the accountability of elected officials, provide voters with clear and identifiable choices among different potential governing coalitions, and introduce mechanisms facilitating alternation in government. In New Zealand, manufactured majorities of the 1980s generated widespread opposition to the existing electoral rules and eventually resulted in referenda that abrogated plurality and instituted MMP rules (Denemark 2001; Nagel 1994). Pervasive

30 / mixed electoral systems

corruption became a salient issue in Japan, providing incentives to parties to seriously engage in the design of an electoral reform that purportedly aimed to increase the accountability of the members of the Diet and reduce their dependence on pork spending (Reed and Thies 2001). Finally, in Venezuela, economic crises and political upheaval produced demands for reform that could make elected officials less corrupt and more responsive to voters’ preferences (Crisp and Rey 2001). In all four cases, Shugart (2001, 178) maintains that “the appeal of mixed-member system[s] is that, almost uniquely among electoral systems, they appear to generate a balance not only between majoritarianism and proportionality, but also between representation centered around legislators personally identified with a locality and representation centered around political parties.” The answer to the third question, addressing the behavior of political parties during the reform process, can be found in arguments analogous to those made about the introduction of mixed electoral systems in transition countries. Parties should behave consistently with their office-seeking or policy-seeking preferences. In this context, dominant parties that fare well under the old electoral rules should attempt to impose provisions that perpetuate the effects of the electoral system that is being reformed. In Italy, political parties like Christian Democracy (DC) attempted to maximize the impact of incumbency and thus enhance the reformers’ immediate chances of reelection. A compensatory mixed system allowing multiple candidacies was ideally suited to accomplish this objective, as it altered the electoral rules decisively without drastically modifying “the pattern of political representation” (Navarra and Sobbrio 2001, 45). Similarly, while the mixed system in Japan “may indeed produce alternation in power between two parties,” new electoral rules permitting dual candidacy maximize the electoral advantages of incumbency and provide representatives an incentive to behave in the same particularistic fashion as under the single non-transferable vote (McKean and Scheiner 2000, 477). Again, these arguments imply that mixed electoral systems are likely to arise through bargaining because the plentiful varieties of SMD–PR combinations allow political parties with diverse interests to achieve their distributive goals. Similar to other institutions, the adoption of mixed systems is the result of endogenous social choices made under different procedural rules by actors whose preferences are derived from expectations of what they can gain through those new institutions (see Brady and Mo 1992; Riker 1980; Shepsle 1986). Under the assumption of institutional selection based on self-interest, mixed systems would spread when political actors calculate that the combination of majoritarian and proportional elections facilitates seat maximization, their chances of participating in government, or hedges bets against likely losses in pure SMD or PR systems. This calculation of

taxonomy of mixed electoral systems / 31

self-interest could lead to mixed systems as the best choice, or as a hedging solution under conditions of uncertainty. To summarize, mixed electoral systems may have spread due to normative preferences, the idiosyncrasies of local political conditions, or decisions by self-interested political actors. While all of the explanations may evidence some shortcomings when they are applied to real-world election reform, explanations based on self-interest cover the widest range of cases temporally and spatially. Do self-interested actors always pursue their interests in environments characterized by perfect information? Empirical evidence suggests that these assumptions sometimes must be relaxed. The story of New Zealand’s electoral reform implies that actors do not always adopt rules they perceive to be beneficial, as the impetus for change in that case was a public misstatement of party position. While the party in power did not favor change to a mixed system, the damage from a failure to meet its promise was potentially greater. Uncertainty also characterizes the political environment in postcommunist states undergoing electoral reform. In some versions of the Russian electoral reform narrative, political actors in the president’s camp debated the likely costs and benefits of adopting SMD or a mixed system. Mixed rules facilitated hedging bets in the face of uncertainty. Actors selected rules that they believed would benefit them, but their expectations were not fully met by the outcomes. Despite these qualifications, self-interest provides a plausible, if untestable, explanation for election rule design in new and old democracies that is generalizable beyond a single world region or outcome. Conclusion This chapter explored how the extant literature classifies mixed systems and explains their origins. Prior to the widespread diffusion of mixed electoral rules in the 1990s, such systems were classified as anomalies, or aberrant forms of proportional representation. The adoption of mixed systems across the globe, in both new and established democracies, prompted scholars to rethink how they were treated in the taxonomic literature. While efforts to classify mixed systems extended our understanding of their mechanics, they did not converge to a single definition of mixed electoral rules. We attempted to rectify this shortcoming by outlining forms of institutional variation in mixed systems. Building on existing scholarship, we noted that one feature of mixed systems—the allocation of seats by more than one formula in the election for a legislative body—is a necessary condition for rules to qualify as mixed. Mixed electoral systems, however, vary on a number of important dimensions, such as geographic application of the two formulae, number of ballots, provisions allowing dual candidacy,

32 / mixed electoral systems

proportion of seats allocated through a given formula, and presence of seat or vote linkages. Many of the empirical tests we perform in the remainder of this volume aim to explain how these institutional features affect the outcomes produced by mixed electoral systems by shaping the strategic environment faced by voters, candidates, parties, and legislators. We also reviewed some of the most common explanations that were provided by the literature on the recent proliferation of mixed electoral systems. Alternative explanations, such as those focusing on political entrepreneurs with normative preferences, idiosyncratic conditions of regime transition or political debate, and political actors pursuing their office-seeking or policy-seeking objectives, have been proffered for electoral reforms that ushered in the adoption of mixed systems. While explanations emphasizing instrumental rationality may not fully explain the origins and diffusion of all mixed electoral systems, their parsimony and applicability across a wide range of cases is attractive. Regardless of the precise nature of the selection process, mixed electoral systems have become popular because they seem to provide advantages over other election rules. Mixed systems promise to contribute to party system development: the use of PR should encourage party formation, but the combination of SMD and a threshold in PR should keep the number of parties in check. Mixed systems provide voters with a wider range of choices: in most mixed systems, voters are granted two votes, facilitating a more nuanced expression of preferences. Moreover, the combination of legislators elected on a party list and in local constituencies should facilitate better representation. Mixed systems also benefit parties and candidates, especially by providing candidates a “safety net” through dual candidacy. Some of the advantages offered by mixed electoral systems have undoubtedly influenced constitutional engineers and helped them win popular support for the outcome of the reform. However, do mixed systems produce these benefits? The following chapters challenge the conventional wisdom about mixed electoral systems by assessing the interaction of the majoritarian and proportional components of the election, by investigating its consequences for the behavior of relevant political actors, and ultimately by evaluating whether hybridization has resulted in speciation: the conception of a distinctly different variety of electoral systems.

C h ap t e r 3 The Logic of Contamination: District-Level Party Systems

Even before mixed systems rose to the status they currently enjoy—the most fashionable set of electoral institutions—Maurice Duverger (1985, 72) discussed how these electoral rules are characterized by a spill-over effect: “Politically, this combination of the English system and PR exerts a pressure toward a two-party system as a result of the first vote, which influences citizens in how they cast their second vote.” Duverger envisioned a kind of interactive effect whereby the two-party competition that is typical of majoritarian races would spill over to the proportional tier. In both components of the election, the party system should approximate “bipolar” arrangements, in spite of the tendency of proportional representation to encourage the proliferation of parties. While Duverger’s conjecture was mostly based on Germany’s postwar experience with mixed electoral rules, the rapid diffusion of these electoral systems has generated the data to evaluate such spill-over effects more systematically. Most analytical work on mixed electoral systems has studied how the mechanics of SMD–PR combinations affect the strategic behavior of parties and voters. As we noted in chapter 1, the adherents of the “controlled comparisons” approach minimize the implications of the interaction between the two components of the election. Others have posited that the two tiers of mixed electoral systems do not function independently, but have reversed the causal arrow of the relationship hypothesized by Duverger. Instead of focusing on the reductive impact of majoritarian races on the number of parties contesting on the proportional ballot, researchers have studied how the multiparty competition characteristic of PR spills over to the SMD contests (Cox and Schoppa 2002; Ferrara 2004; Herron 2000; Herron and Nishikawa 2001). This chapter builds upon previous work by evaluating how contamination is manifested in district-level party competition. We begin by discussing the logic of contamination, explaining how the interaction between the two

34 / mixed electoral systems

components of a mixed system alters the strategic environment facing political actors and undermines the achievement of Duvergerian equilibria in both tiers. First, we investigate how a party’s nomination strategy in the majoritarian component affects its performance in PR. If the two components of the election are independent, we should not find a systematic relationship between a party’s decision to nominate an SMD candidate and the number of votes garnered by its PR list. Such an interactive effect, however, is found in many mixed systems even after we exclude plausible competing explanations. This initial test provides the first piece of evidence demonstrating the existence of contamination. After showing that a party’s PR performance receives a “boost” from the placement of SMD candidates in the corresponding district, we evaluate one of the implications of this finding. Specifically, if parties can improve their PR vote by running a candidate in SMD, however hopeless, they have a clear incentive to compete autonomously in as many districts as possible. We present evidence that mixed systems, in fact, inhibit the development of two-party competition. Even in the majoritarian component, votes are dispersed more widely than they are generally under first-past-the-post. The Logic of Contamination As Duverger (1954) noted, the behavior of parties and voters is shaped by an electoral system’s mechanical and psychological effects. The mechanical effect describes how votes are translated into seats. Some electoral rules are permissive, facilitating seat acquisition by small parties. Other rules are restrictive, punishing marginally performing parties by preventing them from gaining a share of seats that reflects the level of support they enjoy in the electorate. The psychological effect describes the strategic responses of political actors to the mechanical effect. Under certain conditions, voters may recognize that a vote for a small party is a “wasted” vote. Instead of voting for their most preferred party or candidate, voters may abandon their sincere choice and cast a ballot for a more viable contestant. Parties who lose support over successive elections may subsequently merge with other parties or withdraw from electoral competition altogether. SMD-plurality systems place a relatively high burden on parties: they only win a seat when their candidate gains the most votes in a district. Parties whose support is diffused across the national territory, but do not enjoy plurality support in any district, may fail to win representation in a parliamentary election. As voters, candidates, and contributors recognize the impotence of small parties, these parties are abandoned. Over time, SMD should generate two-candidate competition at the district level (Cox 1997; Taagepera and Shugart 1989).

logic of contamination / 35

Strategic behavior should play a more limited role in elections held under proportional rules, but it is still evident. When district magnitude (M) is unlikely to produce proportional results (i.e., when M ⫽ 2), small parties have little hope of winning a seat. Just as in SMD races, voters have the incentive to desert hopeless competitors; weak parties have little incentive to run. Under a series of assumptions, strategic voting and entry should constrain the number of competitors to M ⫹ 1 for district magnitudes less than or equal to five (Cox 1997; Cox and Shugart 1996; Reed 1990). When district magnitude exceeds five, though, strategic desertion tends to dissipate because voters perceive that they can influence the outcome even if they vote sincerely for a party that is not among the most competitive. Particularly where PR seats are allocated in one national district, voters are expected to behave expressively and foster the fractionalization of the party system. The strategic environment created by the mechanical effect is more complicated in mixed electoral systems. By combining two formulae that generally produce different outcomes—particularly with regard to the degree to which small parties are underrepresented—mixed systems provide parties and voters with contradictory messages that might hamper successful coordination. For parties, the incentive to merge, coalesce, or form coalitions that majoritarian rules provide is mitigated by the prospect of receiving adequate representation through the proportional component. For voters, the decision-making processes for each of the two ballots are likely to be inextricably linked, potentially blurring the difference between voting behavior in the PR and SMD components. In a mixed electoral system, voters1—particularly those who identify with medium- or small-sized parties—face three main courses of action. Under the assumption that the choice set is unrestricted in both tiers, voters could: ● ●

●

Vote strategically in SMD and sincerely in PR; Vote sincerely in PR and select their favorite party’s affiliated candidate in SMD, even if that candidate is not viable; or Vote strategically in SMD and select the corresponding party in PR.

First, some voters may respond to the mechanical effect by splitting their ticket: they may cast a strategic vote in SMD and vote sincerely in PR, as they are expected to do in pure majoritarian and proportional systems. If the voter’s preferred party is anticipated to finish third or lower, she would desert that party’s candidate in SMD—and select the least disliked among the two most viable alternatives—but vote for the third-ranked party in the proportional segment. This behavior is consistent with the “controlled comparison” approach, which assumes that the two components of a mixed

36 / mixed electoral systems

system can be studied as if they were different elections. Under this scenario, two-candidate competition should develop in majoritarian races, while multiple parties will maintain their viability in PR. A second possibility is that rather than voting strategically on the majoritarian ballot and sincerely on the proportional ballot, some voters may behave consistently across the two components. That is, if the voter prefers a party whose affiliated candidate is considered a long shot to win the single-member district, she would vote for the losing candidate in order to be consistent with the choice made in the PR election. Various explanations may account for this pattern of behavior: the voter may not understand the difference between the two components, may be motivated by longer term instrumental considerations, or may attach greater importance to the proportional election. Alternatively, her decision-making process may be simply characterized by what Elster (1999) calls a “spill-over” mechanism: the choice made in the PR component increases the probability that she will replicate that voting decision in the SMD race. In any case, if this voter type were to dominate the electoral landscape, multiparty competition would emerge in both tiers. The third scenario also involves consistency across the two segments of the election. Instead of basing candidate choice on the party-list vote, however, this type of voter’s decision-making process is driven by the strategic logic that is characteristic of SMD: in this case, having cast a strategic vote in the majoritarian component, the voter may behave accordingly in PR, deserting her preferred party. If this type of voter is prevalent, Duverger’s bipolar competition should characterize both the proportional and the majoritarian components of the election.2 Mixed electoral systems also create a more complex strategic environment for political parties. While party strategy encompasses the organization of political campaigns as well as the improvement of the party’s material and technical resources, we focus on one dimension of strategic decision making: how political parties nominate candidates.3 In principle, the logic of strategic entry should induce parties to make different decisions in the proportional and majoritarian components of the election. Assuming minimal costs for ballot access and district magnitudes greater than five (Cox 1997), proportional rules encourage small parties to compete. In SMD, however, parties expecting to receive an insufficient number of votes to win a particular district should withdraw, enter alliances, or lend their support to candidates with similar ideological orientations. According to this logic, a party commanding the support of, say, 10 percent of the electorate in a given constituency would run a PR list but strategically withdraw from the SMD race. Nonetheless, parties that become invested in the electoral competition by committing to the PR portion of the ballot may benefit from

logic of contamination / 37

extending participation to the majoritarian tier. By running a candidate in SMD, particularly if the candidate is experienced, popular, telegenic, or charismatic, the party can advertise its platform more widely, enhance voter awareness of its program, increase its visibility, attract greater attention to its policy objectives, and potentially gain more votes in the PR component of the election by drawing the support of voters who would not have voted for the party otherwise.4 In this sense, parties may elaborate their SMD nomination strategies based not only on their potential candidate’s estimated chances of winning the district, but also with an eye to how candidate selection may affect their performance in the PR component. If parties anticipate candidate placement in majoritarian elections to boost their vote total in PR, they have a clear incentive to nominate candidates, irrespective of their chances of winning an SMD race, to as many single-member districts as possible.5 Even minor, nonviable parties, which are generally expected to coalesce or altogether withdraw from competition in plurality elections, should run in SMD to maximize their vote share in PR. This incentive structure should lead the majoritarian component of the election to differ from pure forms of SMD in two relevant ways. First, the number of entrants, which is typically restricted to few candidates when district magnitude equals one and there is no runoff (Cox 1997), should be noticeably higher in the SMD contests of mixed electoral systems than in pure SMD elections. Second, though short-term instrumentally rational voters with enough information to distinguish the candidates’ viability should still desert hopeless candidates, it is plausible to expect a higher number of entrants to result in greater dispersion of votes away from the top two competitors. By inhibiting the tendency of SMD to restrict the effective number of district-level candidates to two, contamination should produce multiparty competition in single-member districts and hence undermine the acquisition of Duvergerian equilibria in SMD elections. In short, we expect contamination to influence the strategic calculus of relevant political actors in mixed electoral systems, as the interaction of the SMD and PR components creates an incentive structure that differs from independent SMD and PR elections. We test this argument empirically with two propositions. The first hypothesis is that a party’s PR performance is improved by placing a candidate in corresponding majoritarian districts. In turn, by undermining the development of two-candidate races at the district level, contamination should lead to non-Duvergerian equilibria in the majoritarian districts of mixed electoral systems. Because of the incentives provided by contamination, we expect mixed systems to encourage multipartyism.

38 / mixed electoral systems

Data and Variables Our analysis assesses contamination in a variety of important and diverse mixed electoral systems. In this chapter, we report results of tests on data from elections for the German Bundestag, Italian Chamber of Deputies, Japanese House of Representatives and House of Councilors, Lithuanian Seimas, New Zealand’s House of Representatives, Russian Duma and Ukrainian Verkhovna Rada. By presenting the results of statistical tests performed on such a heterogeneous group of countries, we seek to show that our theoretical expectations have generality and that contamination is observable in many kinds of mixed electoral systems irrespective of contextual factors. We consider every major variant of mixed electoral systems. Our analysis includes two MMP systems (Germany and New Zealand), four noncompensatory MMM systems (Japan, Lithuania, Russia, and Ukraine), and one partially compensatory MMM system (Italy). As we note in table 3.1, the electoral rules employed in the sample countries also vary on a number of other dimensions, such as the share of seats allocated through PR, the presence of a threshold in the PR tier, PR district magnitude, and the formula utilized to allocate seats in the majoritarian component. Another relevant difference among the sample countries is party system institutionalization. Most political parties in Germany and New Zealand have existed for decades. Although the Italian and Japanese party systems underwent substantial changes and realignments in the 1990s, most of the new parties have emerged as coalitions of old parties or as splinter groups Table 3.1 Institutional variation in the sample countries Formula in SMDs

656

Plurality

155 70

Plurality Majorityrunoff Plurality

Country

System

Germany

MMP

1953

50

Italy Lithuania

MMM MMM

1993 1992

25 50

5% (or 3 SMDs) 4% 5%

Japan (House of MMM Representatives) Japan (House of MMM Councilors) New Zealand MMP

1994

40

None

18

1983

38

None

50

1993

44

Russia Ukraine

1993 1997

50 50

5% (or 1 SMD) 5% 4%

MMM MMM

PR seats (% of total)

PR district magnitude (mean)

Year adopted

PR threshold

Sources: Massicotte and Blais 1999; Shugart and Wattenberg 2001.

120

Plurality (SNTV) Plurality

225 225

Plurality Plurality

logic of contamination / 39

from existing organizations (the most notable exception is Forza Italia). In contrast, the three transition countries in the sample (Lithuania, Russia, and Ukraine) had inchoate party systems during the 1990s. In these states, party organizations have been ephemeral, partisan affiliations among voters have been weak (White, Rose, and McAllister 1997), elite attachments to parties and legislative factions have been changeable (Haspel, Remington, and Smith 1998; Smyth 1997), the electorate’s volatility has been high, and independent candidates have played a prominent role in majoritarian races.6 By testing our propositions on a sample of countries distinguished by great variation in the levels of democratic consolidation and party system institutionalization, we seek to demonstrate that our theoretical expectations are applicable in various institutional and social settings. Evaluating Contamination Our first hypothesis states that the performance of political parties in the PR component of mixed electoral systems is improved by placing a candidate in corresponding majoritarian districts. We evaluate this proposition on district level election returns from Germany (1953), Italy (1994 and 1996), Japan (1996 and 2000 for the House of Representatives; 1995 for the House of Councilors7), Lithuania (1996), New Zealand (1996), Russia (1995 and 1999), and Ukraine (1998). We employ two main variables in this assessment along with a series of controls. ●

●

PR Vote. Our dependent variable is the percentage of the valid PR votes received by the party, aggregated at the level of SMD constituencies. SMD Candidate.8 The primary independent variable is candidate placement in SMD. Each party is coded 1 if an affiliated candidate participated in the SMD race and 0 if the party did not field a candidate. If this explanatory variable exerts a positive, statistically significant effect on party performance, it will indicate the existence of contamination.9

Our analysis accounts for a major rival explanation. While we anticipate that contamination, in part, explains party performance in PR, one might argue that careful candidate placement is a more adequate explanation. That is, party leaders nominate candidates in districts where they know that their party enjoys a high level of popularity, rendering candidate placement at least partially endogenous to the model. While this argument assumes that political parties clearly comprehend the distribution of support in electoral districts, an assumption that is unlikely to hold in some of the cases evaluated here, we account for popularity in our statistical tests.

40 / mixed electoral systems ●

●

●

Incumbency. We assume that incumbents who won the previous election have demonstrated popularity in their district.10 For each party, the variable is coded 1 if the candidate running in the district is an incumbent who won her seat in the previous election and 0 if the candidate is not an incumbent.11 Demographics. In countries where a district’s demographic or geographic characteristics may be relevant predictors of a party’s performance, we include other control variables. The variables Urban and Education, which designate respectively the proportion of the population in the region that is made up of urban dwellers and the proportion that have completed higher education, are used to estimate the models for the Lithuanian, Russian, and Ukrainian elections.12 Reformist parties in Lithuania, Russia, and Ukraine generally perform well in urban areas where voters are highly educated, while communist and agrarian parties enjoy stronger support in rural areas. Similarly, the variable Primary is added to the Japanese models.13 This variable measures the proportion of the population employed in the agricultural sector. In Japan, the Liberal Democratic Party (LDP) is quite strong in rural areas, as farmers constitute one of the LDP’s key constituencies. Regional Controls. We employ regional dummy variables for Italy and Ukraine. In Ukraine, the most critical regional variables are East (historically more pro-Russian and supportive of leftist parties) and West (where Ukrainian nationalist parties are most popular).14 For Italy, we employ the dichotomous variables North and Center.15 This choice is justified by the existence of “three electoral Italies” (Cartocci 1998, 52–55): the right-wing bloc is particularly popular in the north, the center of the country is typically a stronghold of the Left, while the races in much of the south are far more competitive (see Bartolini and D’Alimonte 1996).

The empirical analysis of district-level multiparty data, such as those employed in this chapter, has been debated extensively. Recent scholarship has noted that Ordinary Least Squares (OLS) is not the optimal method of analysis for compositional data. The vote shares received by each of the parties we consider are highly dependent on one another. Each party receives a proportion of the district PR vote that falls between 0 and 1; moreover, the sum of the votes received by all parties in competition must sum to 1. If we estimate the impact of candidate placement on each party’s PR vote separately—treating the dependent variables as unbounded and independent—we potentially generate nonsensical predictions. The predicted vote of any given political party may fall outside the unit interval; alternatively, the sum of the predicted vote shares of each of the parties in

logic of contamination / 41

competition may exceed one (Honaker, Katz, and King 2002; Katz and King 1999; Tomz, Tucker, and Wittenberg 2002). Different solutions exist to address this problem. We employ the estimator proposed in Tomz, Tucker, and Wittenberg (2002). We transform the vote shares into an unbounded scale by calculating, in each district, the natural logarithm of each party’s proportion of the valid PR vote relative to the vote received by a reference party. Subsequently, we estimate our statistical models via Seemingly Unrelated Regression (SUR), which accounts for the fact that the regression equations for each of the parties have correlated errors.16 Transforming vote shares into log ratios complicates the substantive interpretation of the coefficients. For this reason, we present quantities of interest computed in two steps. We first simulate how candidate placement in the majoritarian component affects the log ratio of each party’s PR vote for particular values of the covariates included. We subsequently convert the values of the dependent variables back into vote shares. We selected the political parties according to two basic criteria. First, their nomination strategies must be characterized by enough variation in the placement of SMD candidates to yield meaningful results. Some parties were excluded because they did not run candidates in enough districts, others were not included because, consistent with expectations, they placed candidates in most SMD races, precluding an effective comparison between their PR performance in the presence and in the absence of affiliated candidates. In the analysis of the 2000 Japanese election, for instance, the Japan Communist Party (JCP) is excluded because there is insufficient variation in its placement strategies (the JCP ran candidates in every SMD). Similarly, all major parties in New Zealand and Germany generally nominate candidates to every plurality contest, so the existence of contamination could only be assessed on the PR performance of minor parties.17 Second, for countries where political parties run regional, rather than national, PR lists (Italy and Japan), we only included parties that fielded lists in enough PR constituencies to provide a sufficiently large sample. This choice ensures that we have an adequate number of observations for each party. Figures 3.1, 3.2, and 3.3 display the results of tests conducted on data from elections held in Japan (Lower House, 2000),18 Italy (1996), and Russia (1999), respectively. Each figure shows the effect of SMD candidate placement on the parties’ PR performance, assuming that (1) all other parties are running an SMD candidate (SMD Candidate is coded 1 for all other parties); (2) incumbency and regional controls are held constant at 0; and (3) the variables Urban, Education (in Russia), and Primary (Japan) are at their median values. The first differences of each party’s expected PR vote given an increase in SMD candidate from 0 to 1 were simulated through the

% Change in PR vote given candidate placement

42

6

4

2

0

–2

–4

DPJ

Komeito

LDP

Jiyuto

Shaminto Hoshuto Jiyurengo

90% Confidence Int.

Figure 3.1 SMD candidate placement and change in PR performance: Japan, 2000.

% Change in PR vote given candidate placement

3.0 2.5 2.0 1.5 1.0 0.5 0.0 –0.5 PDS

Verdi

Popolari

Rif.Com. Forza Italia

AN

90% Confidence Int.

Figure 3.2 SMD candidate placement and change in PR performance: Italy, 1996.

logic of contamination / 43

% Change in PR vote given candidate placement

4

3

2

1

0

–1 NDR KPRF Yab. LDPR KRO OVR ANSF

DN

RSP

SPS

MED

90% Confidence Int.

Figure 3.3 SMD candidate placement and change in PR performance: Russia, 1999.

procedures described in chapter 1 (see also King, Tomz, and Wittenberg 2000). We present the 90 percent confidence intervals of our estimates. The results are consistent with our theoretical expectations. In Japan, SMD candidate placement appears to affect the PR performance of each of the six parties that were considered in our analysis: DPJ, Jiyuto, New Komeito, Shaminto (Social Democratic Party), Hoshuto (New Conservative Party), and Jiyurengo. The substantive effect of running an SMD candidate on PR performance appears to be relatively large. For DPJ (which received 25 percent of the national vote), New Komeito (13 percent), and Shaminto (9 percent), PR performance is on average 4 percentage points higher in districts where they contested the corresponding SMD race. The confidence intervals reveal that such effects could be as high as 6 percentage points and as low as 2–3 percentage points. The effects are slightly less pronounced for Jiyuto and Hoshuto; running SMD candidates appears to result in an increase in PR performance that ranges from about 1 to about 2 or 3 points. Finally, the much smaller Jiyurengo receives approximately a half of 1 point PR vote boost by running SMD candidates.

44 / mixed electoral systems

The analysis of the 1996 Italian election (figure 3.2) yields analogous results, though the substantive effect of SMD candidate placement seems to be a bit smaller. Holding incumbency and regional controls constant, Forza Italia, Rifondazione Comunista, and PDS received on average between 0.5 and about 2 percentage points more PR votes in districts where their candidates competed in the corresponding SMD races. The Greens (Verdi) received about a half a point PR bonus; the Popolari and Alleanza Nazionale appear to receive a similar boost, but the confidence intervals of the first differences estimated for those parties also include 0. When different “patterns of competition” are considered,19 the effect of SMD candidate placement on the PR vote for Lista Dini, Lega Nord, Lista Pannella, and Movimento Sociale—parties that did not run lists in every constituency— is generally positive and statistically significant at the .05 level, as it is in models estimated with OLS. Though the Russian results (figure 3.3) are not as consistent as those from Japan and Italy, they are also generally supportive of our propositions. Of the 11 parties in the analysis, SMD candidate placement appears to exert a positive impact on the PR performance of 6 (Bloc of Andrey Nikolaeyev and Svyatoslav Fyodorov (AN-SF), Spiritual Heritage (DN), Communist Party of the Russian Federation (KPRF), Congress of Russian Communities (KRO), Russian Socialist Party (RSP), and Yabloko). The effect is most pronounced for KPRF and OVR, about 2 percentage points, though the confidence intervals are relatively wide. Yabloko’s PR bonus is approximately 1 point, while the smallest parties (KRO, ANSF, RSP) seem to gain half a percentage point in their PR vote in districts where they nominate their own SMD candidates. In table 3.2, we summarize the results for each of the statistical tests we performed. Given the similarity of the SUR and OLS estimates, we provide a summary of the OLS results.20 Aside from the cases we just discussed, we find robust support for the logic of contamination.21 Contamination appears to consistently affect PR performance in elections for the Italian Chamber of Deputies, both houses of the Japanese Diet22 and the Lithuanian Seimas. The PR fortunes of the minor parties in New Zealand and Germany also appear to be influenced by their nomination strategies in single-member districts. Though our results are weaker for Ukraine and Russia, contamination is still evident. In the overwhelming majority of the cases we subjected to empirical tests, voting behavior in the proportional elections of mixed systems appears to be systematically affected by the nomination strategies employed by parties in majoritarian races. Of course, skeptics may object to these findings by noting that the demographic, regional, and political proxies of popularity we consider are imperfect measures. Other dimensions of popularity, in

logic of contamination / 45 Table 3.2 Evidence of contamination in PR vote shares Country (year of election) Japan (House of Representatives 1996) Japan (House of Representatives 2000) Japan (House of Councilors 1995) Italy (1994) Italy (1996) Lithuania (1996) Russia (1995) Germany (1953) New Zealand (1996) Ukraine (1998) Russia (1999)

Number of parties analyzed

# of cases with evidence of contamination

% of cases with evidence of contamination

6

6

100%

6

6

100%

4

4

100%

11 11 11 10 4 4 13 11

10 10 10 8 3 3 8 6

91% 91% 91% 80% 75% 75% 62% 55%

fact, may affect both candidate placement and the PR vote. In turn, the potential for omitted variable bias undermines our ability to make definitive causal inferences about the magnitude of the PR vote boost a party can reap by placing an SMD candidate. To overcome this serious challenge, we asked ourselves the question: what information do parties use to estimate the level of support they enjoy in different localities? In some cases, they may have access to opinion polls; in others, historical election returns (if sufficiently recent to constitute reliable barometers of their popularity) may be available. Given that we do not have access to district-level polling data, we made use of past election returns to verify the robustness of our estimates. First, we ran OLS models that include, for each party in the three elections discussed earlier, the share of votes they received in the election held at t ⫺ 1 (Italy 1994, Japan 1996,23 and Russia 1995). The results seem quite robust to the inclusion of this additional covariate, though of course many parties failed to compete in the two successive elections we evaluate. Second, we conducted nonparametric preprocessing of the data via propensity score matching (see Ho et al. 2004). Essentially, this procedure involved “pruning” the data so that, for each party, the “treatment” group (districts where SMD candidate ⫽ 1) and the “control” group (districts where SMD candidate ⫽ 0) have the same distribution of the PR vote at t ⫺ 1. The results were promising but of limited use. In Japan’s House of Representatives, matching worked only for Jiyurengo; the impact of candidate placement is similar to that shown in figure 3.1. In Italy, matching failed to

46 / mixed electoral systems

produce balanced samples for the largest parties (PDS and Forza Italia). For smaller parties such as Verdi and Rifondazione Comunista, the point estimates of the effect of candidate placement remain stable: the estimated Average Treatment Effects are 0.9 and 0.5 percent, respectively; the 90 percent confidence intervals exclude 0. Finally, matching in Russia succeeded in producing balance for only one of the parties whose PR performance we had found to be affected by SMD candidate (KRO). Many of the parties that ran in 1999, in fact, did not participate in the 1995 election. For KRO, SMD candidate placement still appears to systematically affect the PR vote. All in all, while still worthy of further investigation, our first hypothesis finds consistent empirical corroboration. In most settings, parties win more PR votes where they run SMD candidates. Contamination and Duverger’s Law The second hypothesis we investigate in this chapter predicts that contamination should lead to non-Duvergerian equilibria in the single-member districts of mixed electoral systems. In other words, the majoritarian component should be distinguished by multiparty competition. Given that our expectations solely regard the development of party systems at the district level, rather than at the national level, we evaluate this proposition by comparing the mean effective number of electoral parties characterizing elections under pure plurality (specifically, those that took place in the typical Westminster democracies and the United States) to the SMD component of the mixed electoral systems included in our sample (see table 3.3).24 Consistent with our expectations, table 3.3 shows that the effective number of parties (ENEP) observed in the SMD component of mixed systems generally exceeds the ENEP found in pure plurality systems. For

Table 3.3 Mean effective number of electoral parties SMD systems U.S. House of Reps. 1900–1990 U.S. Senate 1914–1990 United Kingdom 1922–1997 Canada 1935–1993 India 1957–1991 New Zealand 1972–1993

Mixed systems, SMD elections 1.81 1.91 2.27 2.40 2.49 2.56

Italy 1994–2001 Germany 1990–2002 Japan 1996–2000 New Zealand 1996–2002 Russia 1993–1999 Lithuania 1996 (Majority) Lithuania 2000 (Plurality) Ukraine 1998

Sources: SMD systems: Cox and Schoppa (2002); Mixed systems: Authors’ calculations.

2.65 2.73 2.83 3.09 5.07 5.63 5.80 5.98

logic of contamination / 47

countries where competitive elections are a relatively new phenomenon (Lithuania, Russia, Ukraine), one might argue that the multiparty competition characterizing their SMDs is caused by the lack of an institutionalized party system. Electoral volatility and voter fickleness hinder strategic behavior, as voters may not have the information necessary to assess the viability of each candidate and may not have precise, ordered party preferences (Moser 1997, 1999). Also, Lithuanian SMD elections operated under two-stage majority in 1996, which is considered to be more favorable to the proliferation of parties (Cox 1997; Benoit 2001b). It is interesting, however, that the mean effective number of parties observed in Lithuanian SMD races increased with the introduction of plurality rules in 2000. In addition, even in countries such as Italy and Japan, where the introduction of mixed electoral rules was judged to have been most successful in the promotion of two-party competition (Reed 1999, 2001), the number of viable candidates competing in majoritarian elections is, on average, higher than that observed in each of the countries employing standard SMD rules.25 Moreover, the case of Germany, where MMP rules have been in use since the end of World War II, suggests that such differences may not be attributed to the relative novelty of mixed electoral systems. Finally, the comparison of the figures calculated for New Zealand, whose “pluralitarian” (Shugart 2001) electoral system was replaced with MMP, reveals that the number of parties in single-member districts has expanded considerably under the reformed rules. Of course, we may only draw limited inferences from such a comparison. We cannot tell whether the differences observed between the mean effective number of parties in majoritarian elections held under “pure” and mixed rules rise to any level of statistical significance. More importantly, rather than comparing, say, the elections of the German Bundestag to those of the U.S. Senate, the more relevant counterfactual question is: would German SMDs be characterized by a lower number of parties if the entire Bundestag were elected through plurality? Although our analysis suggests that the answer to this question is affirmative, the numbers shown in table 3.3 cannot conclusively establish whether this is the case. At a minimum, these results suggest that party systems emerging in single-member districts in the presence of mixed electoral institutions tend to diverge from the expectations of Duverger’s Law. The SMD component of each mixed system evaluated here does not appear to have as strong a reductive effect on the number of parties as standard majoritarian systems, as the interaction of SMD and PR rules mitigates the tendency of majoritarian systems to produce the results predicted by Duverger. Mixed electoral rules present parties and voters with incentives that are dissimilar from those found in pure PR and SMD; in turn, such institutions generate

48 / mixed electoral systems

outcomes that deviate from those that would be observed if mixed systems were simply the combination of independent SMD and PR elections. Conclusion In this chapter, we have shown that mixed systems are distinguished by incentives and outcomes that differ from those of pure SMD and PR systems. We demonstrated that contamination is evidenced in district-level data across multiple elections held in seven countries that employ three different varieties of mixed electoral systems and are characterized by profoundly dissimilar levels of democracy, social cleavage structures, and dominant cultural norms. Under the assumption of no contamination, PR performance should not be affected by candidate placement in majoritarian contests. However, even after controlling for popularity, we find that a party’s majoritarian nomination strategies systematically affect its PR performance. In turn, the SMD races of mixed electoral systems are distinguished by multiparty competition, or at a minimum by a number of viable candidates that exceeds the number generally observed in pure majoritarian systems. The logic outlined in this chapter, which defines many of the puzzles we address in the remainder of this volume, constitutes the basis of our contention that mixed systems are theoretically and empirically distinct from SMD and PR.

C h ap t e r 4 Going it Alone? Strategic Nominations in Mixed Electoral Systems*

In the study of the consequences of mixed systems on party strategy, a critical question remains unanswered: How does the strategic environment created by mixed electoral rules affect preelectoral coordination? As we demonstrated in chapter 3, placement of a candidate in an SMD race seems to boost the vote share of the PR list with which the candidate is affiliated. An implication of this finding is that, in an attempt to maximize their PR vote share, parties have an incentive to “go it alone” and nominate candidates in as many districts as possible, thereby eschewing cooperative agreements that would limit how they distribute candidacies in majoritarian elections. The proposition that contamination discourages coordination, however, is challenged by an empirical observation: parties frequently fail to place candidates in every single-member district and often engage in substantial preelectoral coordination in the SMD component of mixed electoral systems. Particularly in cases such as Hungary and Italy, many parties strategically support one another’s candidates, enter stand-down agreements, or form broad, explicit alliances. Why would parties forego the benefits of contamination, either by exchanging support for one another’s nominees or by forming coalitions that jointly nominate single-member district candidates? In other words, how much does contamination matter to a party’s nomination decisions? Although some studies have attempted to explain preelectoral coordination under hybrid rules, all of them have focused on one or two countries. Herron (2002c), for instance, explains why Russian and Ukrainian parties generally do not run candidates in most districts. Possible negative effects of candidate placement, limited organizational resources, and the possibility * This chapter’s analysis was first published as Federico Ferrara and Erik S. Herron. 2005. “Going it Alone? Strategic Entry Under Mixed Electoral Rules.” American Journal of Political Science 49(January): 16–31. © 2005 Midwest Political Science Association.

50 / mixed electoral systems

of reaching strategic agreements with other parties may discourage elites from nominating candidates to each SMD race. Ferrara (2004) focuses on institutional incentives and argues that coordination in Italy is promoted by the majoritarian character of its electoral system (see also D’Alimonte 2001; Katz 2001). Similarly, Benoit (2001a) attributes the Hungarian party system’s convergence toward bipolar competition to the incentives built into the new electoral law. In this chapter, we move beyond country-specific explanations of preelectoral coordination by concentrating on the institutional incentives that may discourage parties from running candidates in every single-member district. We conduct empirical tests on district-level data of single-member district elections taking place in various mixed electoral systems. We find that single ballot mixed systems, seat linkages connecting the SMD and PR tiers, and the national allocation of PR seats promote the proliferation of candidacies in SMD races; in contrast, preelectoral coordination is more likely to take place in highly majoritarian mixed systems. Preelectoral Coordination Under Mixed Electoral Rules Strategic coordination encompasses a heterogeneous set of behaviors that may include decisions to form parties, register parties for participation in elections, and engage in specific nomination strategies in the two tiers of a mixed electoral system. This observation begs the question: is the number of participating parties determined endogenously or exogenously? The distinction has implications for the analysis (see Shepsle 1991) as the incentives influencing parties to coordinate constituency nominations could also affect entry decisions. Because our analysis focuses on electoral coordination in the SMD tier in a cross section of mixed systems, we assume that the number of parties that can offer candidates in SMD is fixed.1 While interelection dynamics may encourage political actors to defect from parties (Laver and Benoit 2003), to create new parties (Hug 2001), to merge existing parties, or to disband parties, we are concerned with decisions to coordinate nominations in the SMD tier by parties that have committed to participation by gaining entry to the party-list ballot.2 We expect elements of the “formulaic structure” (Cox 1997, 60) of mixed systems (such as ballot structure, electoral formula, the majoritarian character of the electoral rules, and PR district magnitude) to affect a party’s strategy and its decision to either coordinate with other parties or “go it alone” by placing autonomous candidates in SMD races. On the basis of ballot structure, mixed electoral systems may be divided into two broad categories. In dual ballot mixed systems, voters cast two separate ballots: one for a candidate and one for a PR list.3 Parties and voters can make

going it alone? / 51

distinct strategic decisions in each tier. By contrast, in single ballot mixed systems, voters only cast a single “nonexclusive” vote (Cox 1997, 41), by which they simultaneously select a single-member district candidate and a party list. The votes received by all SMD candidates affiliated with a particular party are then pooled for the purposes of allocating PR seats. We hypothesize that parties are more likely to coordinate in the single-member district races of dual ballot mixed systems than in single ballot mixed systems. We expect to observe such a systematic difference for a number of reasons. Because SMD votes determine PR votes in single ballot mixed systems, parties have the incentive to place candidates in as many districts as possible if they wish to maximize the number of PR seats they receive. Failing to nominate candidates in SMD causes vote loss in PR. Further, in these systems, voters are less likely to consider a vote cast for a hopeless SMD candidate to be a wasted vote, particularly if the party with which that candidate is affiliated has a reasonable chance of winning a PR seat in that constituency. The pooling of SMD votes to allocate PR seats removes the incentive for voters to comply with expectations derived from Duverger’s Law in single-member districts. In turn, anticipating fewer desertions, more parties should be expected to nominate single-member district candidates. Finally, preelectoral coordination involves significant costs for parties, particularly if it is realized through the establishment of explicit coalitions, rather than through stand-down agreements. By entering an alliance, a party must sacrifice some autonomy. While parties are able to preserve their independence and cultivate their electoral support by running autonomous PR lists in dual ballot mixed systems, parties can only maintain a separate identity by nominating their own SMD candidates in single ballot mixed systems (Katz 2001). In this sense, formulaic structures allowing voters to cast two separate ballots render preelectoral coordination relatively less costly. The electoral formula employed to elect representatives in the majoritarian component is another relevant source of incentives influencing preelectoral coordination. The literature explaining strategic entry decisions in single-member district races held under alternative, pure electoral rules is well developed. Cox (1997)4 evaluates strategic voting and entry in elections held under majority-runoff rules. Contrary to the incentives of single-member plurality contests, electoral rules requiring candidates to win a majority of the vote (at least in the first round) should not produce two-party competition. Though voters still have the incentive to desert first round candidates whom they perceive to be unlikely to qualify for the second round,5 and hopeless candidates have an incentive to withdraw or coordinate with other parties, majority elections have a weaker reductive impact on the number of viable candidates and thus provide fewer incentives to establish preelectoral alliances.

52 / mixed electoral systems

Strategic voting and preelectoral coordination may be mitigated by the majority–plurality rules in mixed electoral systems. In Hungary, for example, because advancement to the second round may produce “bargaining leverage,” the electoral law discourages the formation of alliances before the election (Benoit 2001b, 478). Though political parties in Hungary’s SMDs may coordinate (Benoit 2001a), coordination typically takes place only after the first round: when “two [allied candidates] step forward, one steps back.” Consistent with these expectations, we hypothesize that more coordination is observed in single-member district races of mixed electoral systems if the majoritarian component of the election operates through plurality, rather than majority-runoff or majority-plurality. Mixed systems not only combine different electoral rules in the election of a single legislative assembly; they also combine two distinct principles of representation. In MMP systems, the seat linkage compensates for the disproportionality of SMD elections. In MMM systems, the number of seats that a party receives in one component of the election is not directly linked to the seats it wins in the other component. We hypothesize that the existence of a seat linkage inhibits preelectoral coordination. In MMP systems, parties face little incentive to coordinate. Even parties that have no hope of winning any SMD are not damaged by the electoral formula because the compensatory mechanism guarantees that their total seat share will be proportional to their vote share. Hence, there is no reason for parties to form preelectoral alliances in the SMD elections of MMP systems, because the outcome is inconsequential for the maximization of their seat total (Moser 2001). Rather, parties should place candidates in as many districts as their resources allow. By running their own SMD candidates, parties may increase their visibility, advertise their platform, and costlessly exploit the boost that contamination effects are likely to give to their PR performance. The disincentive to coordinate is far more pronounced in MMP systems, where non-competitiveness in SMD races is not penalized by the mechanics of the SMD–PR mixture. While the method of seat linkage is an important factor determining whether the majoritarian or proportional components drive election results, it is not the only institutional feature that affects the overall proportionality of the electoral outcome. In particular, three critical elements of the formulaic structure confer more importance upon one tier of the election or the other: the percentage of total seats distributed through SMD, the existence of a PR threshold and its relative permissiveness, and PR district magnitude. The incentives that such institutions provide to parties and voters should figure prominently in a party’s calculus and in its formulation of an electoral strategy. We posit that, holding the presence of a seat linkage constant, as the share of seats allocated in single-member district races increases, parties are more likely to engage in preelectoral coordination.

going it alone? / 53

In systems where a large proportion of seats is set aside for the majoritarian component of the election, forming broad preelectoral alliances may help parties of all sizes achieve their goals. For large parties, effective preelectoral coordination with other parties and the establishment of alliances that are competitive in most SMD races nationwide may prove to be the difference between forming a government and organizing the opposition. In these systems, it may be instrumentally rational for party leaders to forego the benefits of placing their own candidates in every SMD to maximize, through the formation of preelectoral alliances, their chances of commanding a legislative majority in conjunction with other parties. Similarly, smaller parties whose candidates are not viable in SMDs should expect to be punished by the mechanics of MMM systems where the share of seats available in the PR tier is too small to offset the disproportionality that majoritarian elections are likely to yield. For these parties, failing to coordinate with other parties may cause their seat share to diverge substantially from their vote share. Nonetheless, entering preelectoral coalitions with larger parties in SMD races6 allows small and medium-sized parties to win more seats than they would as a result of the boost in PR performance that the placement of autonomous candidates provides. Consequently, an increased seat share may also confer upon these parties additional blackmail power and leverage they can employ to extract policy concessions from their allies after the election, particularly if their participation is indispensable to the survival of a coalition government. Heavily majoritarian mixed electoral systems provide parties an incentive to enter preelectoral coalitions. The larger the deviation from proportionality that the mechanics of a mixed system are expected to produce, the more likely it is that parties will coordinate in the SMD component of the election. While large parties should attempt to maximize their chances of winning a legislative majority by concluding strategic agreements with parties they wish to include in a future governing coalition, small parties should readily renounce some of their autonomy in an effort to protect their survival, maximize their seat share, and increase their bargaining strength. Contamination provides some benefits to parties that command the resources to nominate their own candidates in every single-member district. In turn, the SMD contests of mixed systems should tend toward the multiparty competition that SMD elections normally discourage. In highly majoritarian hybrid systems, however, the benefits of coordination should exceed those of running independently of other parties. We should therefore expect the SMD races of mixed electoral systems to increasingly approximate two-party competition as the relative salience of the majoritarian component increases. Endeavoring to bolster PR performance in systems where the PR tier is far less relevant than the majoritarian tier is not an optimal electoral strategy.

54 / mixed electoral systems

In addition, the incentive for small parties to join preelectoral alignments should be stronger in the presence of a PR threshold that is high enough to threaten their survival by precluding access to the seats distributed in the proportional component of the election. For instance, in the election of the Chamber of Deputies in Italy, small parties find it relatively easier to win SMD seats (by obtaining larger parties’ backing for some of their SMD candidates) than PR seats. While only five parties managed to overcome the 4 percent threshold in the PR tier of the 2001 elections, as many as 10 parties that ran PR lists won representation in the majoritarian component. We hypothesize that as the percentage of votes that must be received to participate in the distribution of PR seats increases, preelectoral coordination also increases. The final institutional feature that should affect strategic entry decisions is the district magnitude of the PR constituency where the district is located. We expect that preelectoral coordination is lower in single-member districts where the corresponding PR district magnitude is higher. Political parties, in fact, may attempt to improve their PR performance by nominating candidates to as many SMDs as possible. In countries where all of the available PR seats are distributed in a national PR constituency, parties may expect to win a nontrivial number of additional seats through the PR bonus spawned by candidate placement in SMDs. In contrast, however, when PR district magnitude (M) is low and where only a limited number of parties have realistic aspirations of winning PR seats, smaller, nonviable parties may find it inconsequential to boost their PR performance by nominating their own SMD candidates. Rather, they may be more sensitive to incentives to either withdraw or back other, more viable candidates. The above discussion yields the following predictions: ●

●

●

●

Little or no preelectoral coordination should be observed in the SMDs of MMP systems without a PR threshold; In contrast, substantial coordination should take place in dual ballot MMM systems where most of the legislative seats are allocated through single-member plurality and where the PR threshold is high; Parties should engage in some coordination in MMM systems where the majoritarian and proportional components are evenly balanced; and The electoral strategies enacted locally by political parties should be informed by the marginality of a district and the participation of an incumbent candidate in an SMD race. Data and Variables

We report results of tests conducted on constituency-level data from 14 elections in 13 countries or regions (table 4.1). For each country or

going it alone? / 55

subnational region, we analyze the most recent election for which data are available.7 The measurement of political parties’ strategic entry decisions in the majoritarian component of mixed electoral systems involves a difficult choice of indicators that may reveal the extent to which parties employ “go it alone” strategies in SMDs or engage in some form of preelectoral coordination with other parties. Benoit (2001c) measures preelectoral coordination through the use of the “average coalition size,” or the average number of parties supporting each candidate competing for the available seat(s). Though this measure is optimal in assessing whether parties conclude explicit preelectoral agreements with other parties, its use is particularly problematic here because it does not capture the significance of informal deals or strategic withdrawals by which parties agree to support one another’s candidates without necessarily forming official alliances.8 We evaluate our propositions on two dependent variables. While both variables are measures of strategic entry, they allow us to assess two different aspects of the problem under investigation.

Table 4.1 Institutional variation in sample countries

Country

System

Election analyzed Ballots

SMD seats (% of total)

Bolivia MMP Germany MMP

2002 2002

2 2

52 50

Hungary

MMM

1998

2

46

Italy (lower) Italy (upper) Japan Lithuania

MMM

2001

2

MMM

2001

MMM MMM

Mexico New Zealand Russia Scotland South Korea Ukraine Wales

PR threshold

PR district magnitude (mean)

Formula in SMDs Plurality Plurality

75

3% 15 5% (or 598 3 SMDs) 5% 8 (district) 56 (nat’l) 4% 155

1

75

None

4.6

Plurality

2000 1996

2 2

62.5 50

None 5%

16.4 70

MMM MMP

2003 2002

1 2

60 56

200 120

MMM MMP MMM

1999 2003 2000

2 2 1

50 57 85

2% 5% (or 1 SMD) 5% None 3%

Plurality Majorityrunoff Plurality Plurality

225 16 46

Plurality Plurality Plurality

MMM MMP

2002 2003

2 2

50 67

4% None

225 12

Plurality Plurality

Sources: Massicotte and Blais 1999; Shugart and Wattenberg 2001.

Majorityplurality Plurality

56 / mixed electoral systems ●

Number of Candidates. We first measure preelectoral coordination as a count of the candidates participating in a SMD race who are affiliated with a party that ran a list in the PR component of the election.9 This is a rough measure of preelectoral coordination, as it does not reveal much about both official and informal agreements among political parties. But, this variable allows us to identify the kinds of mixed systems that encourage parties to run candidates in as many districts as possible and those that induce them to forego the PR vote boost generated by “go it alone” strategies in SMD.

A limitation of the indicator we just described is that, by merely considering the absolute number of affiliated candidates, it fails to take into account the number of parties that could have placed SMD candidates. For instance, less coordination takes place in a district where 5 out of 5 parties participating in PR elections ran their own candidates than in a SMD where 20 PR lists support 10 candidates. To avoid over- or underestimating pre-electoral coordination, we make use of a second dependent variable that measures coordination more directly by considering the number of candidacies relative to the number of parties in the election. ●

Average Number of Parties. For each SMD in our sample, we calculate the average number of parties supporting each candidate by dividing the number of parties running lists in the corresponding PR constituency by the total number of candidates participating in the SMD race. We include independent candidates in the denominator, as parties may back non-affiliated candidates with whom they share ideological affinities, policy preferences, or common adversaries.10 We expect the average number of parties per candidate to increase as the share of seats distributed through SMD increases, as the PR threshold increases, and when SMD seats are assigned through the first-past-the-post formula. In contrast, we anticipate the average number of parties per candidate to be lower where PR seats are allocated nationally, in single ballot mixed systems, and in MMP systems.

We include a number of explanatory variables in our analysis to evaluate our hypotheses. ●

Single Ballot. Our first hypothesis seeks to explain the difference between the placement strategies of political parties in single and dual ballot mixed systems. We created the variable “single ballot,” which is coded 1 if voters only cast one ballot, by which they simultaneously select a SMD candidate and a PR list, and 0 otherwise.

going it alone? / 57 ●

●

●

●

●

●

●

Plurality. We test the impact of the formula employed in the SMD component through the use of the dichotomous variable “plurality,” which is coded 1 for the SMDs of each of the countries included in the sample with the exception of Hungary (which employs two-stage majority-plurality) and Lithuania (which used majority runoff in the 1996 election). Seat Linkage. We coded the variable 1 if the electoral system under consideration has a seat linkage (MMP) and 0 if it has no linkage (MMM). This variable allows us to assess the difference between the strategic entry decisions made by parties in MMP and MMM systems. %SMD. The explanatory variable “%SMD,” reflects (in percentages) the share of total seats in the majoritarian tier of the election. It is employed to test if parties are encouraged to engage in preelectoral coordination in SMD races when the relative salience of the majoritarian component is high. PR Threshold. We assess the effect of the relative permissiveness of the PR threshold on candidate placement strategies in SMDs by regressing the dependent variables on the variable “PR threshold.” We coded PR threshold as the percentage of votes that a party must receive to participate in the distribution of PR seats. PR District Magnitude (Logged). Our final hypothesis suggests that high PR district magnitudes encourage parties to run more SMD candidates in the hope that the PR vote bonus they receive by placing autonomous candidates in SMD races will translate into additional PR seats. To assess whether this pattern is observable in the data, we employed the natural logarithm of PR district magnitude. We used the natural log of PR district magnitude, rather than the raw value, because we expect a nonlinear effect. As PR district magnitude increases, the magnitude of its marginal effect on the number of affiliated candidates should gradually attenuate.11 Reformed PR. We control for the electoral system that was employed to elect a particular legislative body prior to the introduction of a mixed system in some of our tests. Sartori (1994, 75) predicts that replacing a PR system with mixed electoral rules is likely to fail to reduce the fractionalization of the party system. To control for the effect of past electoral rules, we included in some models the variable “Reformed PR,” which is coded 1 if the introduction of the mixed system used in the election under consideration was adopted to replace a “pure” PR system (as in Italy and Bolivia). Postcommunist. Much of the literature on countries that have recently completed transitions from communist rule has noted that many of those countries have inchoate party systems. Many parties have arisen

58 / mixed electoral systems

●

and quickly disappeared over the course of few general elections, party affiliations among voters are weak, and the electorate’s volatility is high (see Moser 1997 and 1999; White, Rose, and McAllister 1997). To control for the effects of an inchoate party system on strategic entry, we employ the variable “postcommunist” (coded 1 for Hungary, Lithuania, Russia, and Ukraine and 0 for each of the other sample countries). Marginality. Consistent with the literature on “pure” SMD systems, we control for the marginality of a district and for the presence of an incumbent candidate. “LN (Margin)” is the natural logarithm of the margin of victory (in percentages) by which a candidate won the SMD in question in the previous election.12 Incumbency was coded 1 if the candidate who had won a SMD in the previous election was competing for reelection in the same district.13 Tentatively, we expect high margins of victory and incumbency to exercise a reductive impact on the number of SMD candidacies by discouraging the entry of potential competitors (see Cox 1997; Jones 1999).

Analysis and Discussion We assess four models of strategic entry in table 4.2, where the dependent variable is the number of SMD candidates who were affiliated with a PR list.14 In table 4.2, the first model includes six explanatory variables that measure institutional features of mixed electoral systems: single/double ballot, SMD formula, seat linkage between the tiers, percentage of seats allocated to SMD, PR threshold, and the natural logarithm of PR district magnitude. In subsequent models, we add control variables such as “postcommunist,” “reformed PR” (models 2, 3, and 4), margin of victory at t ⫺ 1 (model 3), and incumbency (model 4). The poisson estimates support our expectations. In each of the models, the coefficients for single ballot, seat linkage, and district magnitude are highly significant and positive in sign, indicating that single ballot mixed systems, MMP systems, and systems where PR seats are allocated in large constituencies encourage the proliferation of SMD candidacies. Plausibly, these institutional features of mixed electoral systems induce parties to seek to maximize the PR vote bonus they are likely to receive by running autonomous candidates in SMDs. Similarly, the parameter estimates for plurality formula, percentage of seats allocated in SMDs, and PR threshold, which are statistically significant and negative in sign, suggest that employing the first-past-the-post rule in SMD, setting aside a large proportion of

going it alone? / 59 Table 4.2 Strategic entry under mixed electoral rules: Poisson regression results Model 1 Constant Single Ballot Plurality Formula Seat Linkage MMP ⫽ 1 MMM ⫽ 0 % SMD PR Threshold LN(PR District Magnitude) Postcommunist

Model 2

Model 3

Model 4

3.16* (.07) 0.67* (.02) ⫺0.49* (.04) 0.15* (.02)

2.94* (.13) 0.77* (.02) ⫺0.75* (.04) 0.34* (.04)

2.46* (.18) 0.82* (.03) ⫺0.71* (.05) 0.39* (.05)

2.61* (.20) 0.68* (.04) ⫺0.84* (.05) 0.46* (.05)

⫺0.02* (.00) ⫺0.03* (.01) 0.04* (.01) —

⫺0.02* (.00) ⫺0.13* (.01) 0.19* (.02) 0.26* (.05) 0.56* (.03) —

⫺0.01* (.00) ⫺0.09* (.02) 0.15* (.02) 0.29* (.05) 0.38* (.05) ⫺0.02* (.001) —

⫺0.01* (.00) ⫺0.07* (.02) 0.11* (.02) 0.37* (.06) 0.44* (.05) —

Reformed PR

—

LN (Margin t ⫺ 1)

—

Incumbent

—

—

Pseudo R2 N

0.13 2767

0.17 2767

0.16 2533

⫺0.01 (.02) 0.12 2061

Note: Dependent variable is # of affiliated candidates in SMDs. Standard errors in parentheses. * p ⬍ .05.

the available legislative seats for the majoritarian tier, and instituting less permissive PR thresholds exercise a reductive impact on the number of entrants in SMD races, as they provide incentives to both large and small parties to enter preelectoral coalitions in the majoritarian component of the election. The consistency of the results shown in table 4.2 to our research hypotheses and the robustness of the estimates to alternative specifications of the models strongly support the propositions we put forth to explain preelectoral coordination.15 The results shown in table 4.3 take into consideration the number of parties in competition rather than exclusively concentrate on the raw count of candidates participating in SMD contests. The dependent variable is, for every district, the ratio of the number of parties participating in the PR

60 / mixed electoral systems Table 4.3 Coordination under mixed electoral rules: OLS results Model 1

Model 2

Model 3

⫺1.64* (.12) ⫺1.66* (.04) 1.40* (.08) ⫺0.64* (.04)

⫺1.90* (.14) ⫺1.62* (.05) 1.42* (.08) ⫺0.63* (.05)

⫺1.70* (.19) ⫺1.55* (.08) 1.30* (.13) ⫺0.63* (.05)

LN (Margin t ⫺ 1)

0.04* (.00) 0.15* (.02) 0.11 (.02) —

0.04* (.00) 0.15* (.03) 0.11 (.04) —

Incumbent

—

0.04* (.00) 0.17* (.02) 0.09 (.02) 0.04* (.01) —

Constant Single Ballot Plurality Formula Seat Linkage MMP ⫽ 1 MMM ⫽ 0 % SMD PR Threshold LN(PR District Magnitude)

Adj. R2 N

0.55 2767

0.56 2533

0.02 (.04) 0.46 2061

Note: Dependent variable is average number of parties per candidate. Standard errors in parentheses. * p ⬍ .05.

portion of the election to the number of SMD candidates. Because our dependent variable is continuous in this analysis, we employ OLS. However, since the dependent variable can only take on positive values, we also evaluate our research hypotheses with tobit. Our expectations change for this series of models; we expect positive coefficients for percent SMD, PR threshold, and plurality formula. We expect negative coefficients from single ballot, unlinked tiers, and national PR. The results of our statistical analysis again support our expectations. With the exception of PR district magnitude, the coefficients for each of the variables of interest are significant at the .01 level; signs are in the hypothesized direction. Plurality formulae, strongly majoritarian systems where most of the available seats are distributed in SMDs, and high PR thresholds are associated with a higher average number of parties supporting each SMD candidate. In contrast, MMP systems and single ballot mixed systems appear to be characterized by a lower average number of parties per SMD

going it alone? / 61

candidate, as they presumably provide parties with little incentive to withdraw or engage in preelectoral coordination in SMDs.16 Additional tests yielded estimates that are similar to those presented in table 4.3. Tobit and robust regression estimates conform to those produced by OLS. When we relaxed the assumption of independence within countries, the parameter estimates for single ballot, SMD formula, seat linkage, and percentage of seats allocated in SMD were significant in each of the three models. These additional tests, altering some of the standard OLS assumptions, generate results that largely match the findings in table 4.3. While the evidence for our last two hypotheses is less conclusive, our remaining hypotheses receive strong empirical support in the data. The uniformity of the results we obtain by running a variety of statistical tests on two dependent variables gives us confidence that the institutional features we identified affect strategic entry decisions in mixed electoral systems. However, the substantive impact of those explanatory variables remains to be addressed. In other words, do the predictions we made with regard to preelectoral coordination in SMD races hold? We can use the equations estimated through poisson and OLS (model 1 in table 4.2 and model 1 in table 4.3, respectively) to assess strategic entry in mixed electoral systems characterized by different combinations of majoritarian and proportional rules. Table 4.4 shows the expected values of the affiliated number of candidates and average number of parties per candidate we estimated through Clarify’s stochastic simulation techniques (see Tomz, Wittenberg, and King 2003) for six hypothetical mixed systems: a dual ballot MMP system with 5 percent threshold, an MMP system with no threshold, two dual ballot MMM systems where the majoritarian and proportional components are evenly balanced (one that uses plurality and one that uses majority), a heavily majoritarian MMM system where 70 percent of the available seats are allocated in SMDs, and a single ballot mixed system. The estimates that are shown in table 4.4 are consistent with our predictions and indicate that, for the most part, our institutional explanatory variables exert a nontrivial, substantively significant effect on the two measures we employ as indicators of strategic entry. The number of affiliated candidates participating in an SMD race is higher in MMP systems (6.8 or 7.79, depending on the existence of a PR threshold), in systems where voters only cast one ballot (where over 11 affiliated candidates run in each district), and in systems where SMD elections take place under two-stage majority (9.55). In contrast, the number of SMD candidacies decreases dramatically in highly majoritarian mixed systems (3.82). Finally, more balanced mixed systems employing first-past-the-post are characterized by an intermediate number of entrants.17 The estimates computed for the average number of parties per SMD candidate in the six different ideal types of mixed electoral systems also

Table 4.4 Assessing strategic entry in hypothetical mixed systems Highly proportional mixed systems

Majoritarian mixed systems

Single ballot mixed systems

1.

2.

3.

4.

5.

6.

No Plurality Yes 50 5 250 6.80 (.13) 6.55 to 7.06

No Plurality Yes 50 0 250 7.79 (.31) 7.22 to 8.42

No Plurality No 50 5 250 5.88 (.11) 5.66 to 6.10

No Majority No 50 5 250 9.55 (.26) 9.06 to 10.09

No Plurality No 70 5 250 3.96 (.07) 3.82 to 4.11

Yes Plurality No 50 5 250 11.52 (.35) 10.86 to 12.21

2.24 (.04) 2.17 to 2.32

1.50 (.08) 1.36 to 1.65

2.88 (.04) 2.81 to 2.95

1.48 (.06) 1.37 to 1.60

3.6 (.03) 3.54 to 3.66

1.22 (.06) 1.11 to 1.33

Note: Standard errors in parentheses. M ⫽ 1000 sets of simulated parameters.

62

Single ballot Formula Seat linkage %SMD PR threshold PR magnitude # Affiliated candidates 95% confidence intervals Avg. parties per candidate 95% confidence intervals

Balanced mixed systems

going it alone? / 63

support our predictions. Virtually no coordination seems to characterize SMD races in single ballot mixed systems (there is just over 1 party per candidate), where parties have a strong incentive to nominate their own candidates to as many SMDs as possible.18 In addition, MMM systems appear to encourage more preelectoral coordination than MMP systems. Finally, the average number of parties per candidates is particularly high in markedly majoritarian mixed systems (3.6), revealing that substantial preelectoral coordination takes place under rules that assign most of the legislative seats through single-member plurality. Conclusion To what degree do mixed electoral systems encourage parties to “go it alone” in the majoritarian component of the election? The results presented in this chapter support the proposition that the institutional features of mixed electoral systems generate outcomes that differ from those generally observed under “pure” SMD and PR. Nonetheless, the institutional heterogeneity that characterizes the world’s mixed systems also produces variation in outcomes by creating different strategic environments for political actors. Our key finding is that the incentives produced by mixed systems are influenced substantially by the manner in which majoritarian and proportional institutions are combined. Where the proportional component is dominant, the interaction between the SMD and PR tiers of the election renders parties more likely to “go it alone” in SMDs, as candidate placement in SMD provides them with the opportunity to boost their PR performance. Consequently, SMD races should operate differently from purely majoritarian elections, at least from the standpoint of strategic entry. Even in those systems where the majoritarian and proportional components are separated and evenly balanced, parties are likely to nominate more candidates than they would under pure SMD rules, hampering the evolution of two-party competition in singlemember districts. However, mixed systems may also create a strategic environment conducive to cooperation. In particular, in systems where the majoritarian component dominates the election, parties have an incentive to enter into formal or informal nomination pacts. Under those circumstances, the incentives provided by contamination may be outweighed by the benefits yielded by preelectoral coordination.

This page intentionally left blank

C h ap t e r 5 Interactive Ballots: The Microfoundations of Contamination

The existence of contamination has important implications for the study of mixed electoral systems. Given that the interaction between their mechanical components is likely to generate outcomes that diverge from those observed under pure SMD or PR rules, mixed systems do not constitute a controlled laboratory setting that allows researchers to isolate and compare the independent effects of SMD and PR. Our study of SMD candidate placement’s effects on a party’s PR performance (chapter 3) and our empirical analysis of party nomination strategies in the SMD component of the election (chapter 4) have uncovered evidence supporting the proposition that mixed electoral systems are a different species of electoral institutions. However, our research thus far also suffers from a critical lacuna. Specifically, the interaction between the SMD and PR components of the election discussed in chapter 3 operates at the level of the individual voter. Candidate placement in SMD should boost PR performance because running a candidate in SMD is expected to attract, in the PR tier, the support of voters who would have otherwise voted for other parties. Thus far we have made only casual attempts to explain how the interaction of the SMD and PR rules shapes individual party choice, or how voters should react in PR to campaign messages that originate from the SMD component of the election. This approach is reflected in other work on mixed systems. Herron and Nishikawa (2001, 69) mostly focus on party strategy and argue that by running candidates in SMD a party “can increase voter awareness and potentially gain more votes for the PR portion of the election.” The mechanism by which candidate placement in SMD may attract the support of voters in PR who are undecided or typically vote for different parties is left unexplained. Similarly, Cox and Schoppa (2002, 1031) emphasize “sticky voting” and argue that, under mixed electoral rules, voters should be less susceptible to appeals to behave strategically and avoid wasting their vote on

66 / mixed electoral systems

a small party or a nonviable candidate. They “expect many to swing toward voting expressively, casting both their SMD and PR ballots for their firstchoice party” (Cox and Schoppa 2002, 1031). Small parties may attract voters by running SMD candidates wherever possible, “boost[ing] their PR vote totals by using the human face of their SMD candidates.” Again, the individual-level mechanism generating this effect is underspecified. Ignoring the individual voter implies that we cannot know the direction of the relationship between candidate placement and PR performance: does the nomination of an SMD candidate improve PR performance, or are parties more likely to nominate candidates in districts where they already enjoy high popularity? In chapter 3, we accounted for this rival explanation by controlling for some indicators of a party’s popularity. However, in the absence of individual-level evidence, we cannot definitively refute this competing interpretation of the district-level results. In this chapter, we attempt to fill two major shortcomings in the study of contamination. First, we formulate a theory at the level of the individual voter that specifies how the interaction between the two tiers of the election shapes party choice. Second, we provide individual-level evidence for contamination. We evaluate our propositions with data from the 1999 New Zealand Election Study (NZES). How Contamination Shapes Party Choice At the individual level, the proposition that SMD candidate placement boosts a party’s PR performance implies that when a party runs an SMD candidate in a particular district it increases the probability that voters will cast a ballot for its list in the PR component of the election. This hypothesis, in turn, is based on the assumption that campaigns matter and “conversion” is possible, at least to the extent that conversion means changing one’s vote choice from one party to another because of increased knowledge or attitude shifts that may take place over the run-up to election day (see Finkel 1993, 6). For SMD candidate placement to have any effect on PR voting decisions, some voters must pay attention to the unfolding campaign and employ the information they receive from the SMD component of the election when choosing a party in PR.1 In the context of the study of mixed electoral systems, we posit that SMD candidate placement affects the probability of voting for a PR list because it provides voters with additional information they can use to sift through the uncertainty about party positions to arrive at a voting decision. What is the mechanism by which exposure to campaign messages from the SMD component of the election affects the probability of voting for a party in PR? We submit that party nomination strategies in the majoritarian tier,

interactive ballots / 67

as well as the quality of the candidates who are nominated, may introduce information asymmetry that affects voting decisions in the proportional segment of the election. Gelman and King (1993) argue that campaigns matter because they provide information to voters who, in the absence of such information, may have voted differently. When campaigns are balanced, the information provided by the parties or candidates in competition is likely to even out, leaving outcomes to reflect the initial predispositions of the electorate. However, when campaigns are asymmetrical (in resources, news coverage, or organizational prowess) they are much more likely to have an impact on the electoral outcome. Asymmetry is likely to be particularly accentuated in lower profile, multiparty races. Under mixed electoral rules, parties face several options as they formulate nomination strategies in SMD. One possibility is to “go it alone” and run candidates in as many districts as their resources permit. Another is to nominate candidates only where they stand a reasonable chance of winning or placing well. Also, parties may engage in various forms of preelectoral coordination, either by supporting one another’s candidates or by forming broad coalitions that run a single candidate per district (see chapter 4).2 As voters gather the information that eventually guides their party choice in the PR tier, different nomination strategies adopted by the parties competing in the parallel SMD tier generates additional information that potentially introduces asymmetry in the PR election. If voters incorporate cues they receive from the campaign for the SMD race into the information upon which they base their voting decision in PR, rather than compartmentalize the messages originating in the two components of the election, different campaign strategies in SMD may affect the choice of a PR list. It is generally thought that candidates benefit from running under a party label because such labels reduce voters’ information costs (Downs 1957) and establish a candidate’s viability (Cox 1997). However, if voters seek “information shortcuts” based on data that are “new and easy to process” (Popkin 1994, 62; see also Bartels 1996), it is plausible that the nomination of an SMD candidate may also provide voters information to evaluate a party against the available alternatives. Voters can use the quality of the candidates that parties run in SMD as an indicator of the kind of members of parliament they elect by voting for a particular PR list. Given that many mixed systems allow “dual candidacies,”3 parties should be expected to field their best candidates in SMD contests, where success depends on personal popularity. Consequently, voters can use their evaluations of the competence, experience, and trustworthiness of individual candidates to form expectations about the representation they are likely to get from the parties with which such candidates are affiliated.

68 / mixed electoral systems

SMD nominations may render the campaign asymmetrical and affect party choice in the PR component of the election. Nominating a “good” SMD candidate may increase the probability that voters will select the party that the candidate represents; in contrast, withdrawing from a district race deprives voters of an additional cue, while running a “bad” candidate may turn potential voters off, especially if the candidate’s poor credentials, unappealing personality, or lackluster performance come to be associated with the party. Different SMD nomination strategies, as well as the quality of the candidates, may determine some shifts in the election results.4 The information conveyed by SMD candidates, and the dynamics of the SMD race, may alter the probability that a voter will cast her ballot for a given party in the proportional tier of the election.5 The logic predicting that SMD nomination strategies have an impact on PR voting decisions should be similar to that underlying the welldocumented phenomenon of “coattail voting” in the United States. In mixed systems, we expect the mechanism by which campaign information from the SMD tier affects a voter’s party choice in PR to be similar, though its effect may be less pronounced. SMD candidates provide external cues that voters may use as they evaluate parties. Hence, good SMD candidates may have coattails. Under conditions of uncertainty and low information, it is reasonable to expect a voter’s decision making in each component of the election to be influenced by the information generated in the other tier. Therefore, running a candidate whose favorability ratings are high due to widespread name recognition, an extensive resume, charisma and personal appeal, or a history of dedicated constituency service may increase a party’s support in the proportional segment of the election. The discussion above suggests that SMD candidate placement, and the quality of SMD candidates, may alter the probability that an individual will vote for a particular party in PR. However, which kinds of voters are likely to be most responsive to the information they receive from the SMD campaign? We speculate that weak partisans, independents, and undecided voters, who are generally the most sensitive to campaign information (see Hillygus and Jackman 2003), are also the most likely to be influenced by “contamination.” We propose three mechanisms by which running a strong SMD candidate helps a party’s PR performance. In each of these cases, SMD candidates are believed to provide cues that contribute to a voter’s perception of “party images,” which involve considerations of “how well parties represent people like themselves . . . and how well parties perform different tasks of government” (Popkin 1994, 60). First, a strong candidate may reinvigorate partisan identification, strengthen the support of voters who already identify, albeit weakly, with the party and discourage them from switching to other parties. A particularly

interactive ballots / 69

competent, charismatic, or empathic SMD candidate may provide weak partisans information reinforcing their belief that the party is the best qualified and best equipped to deal with the issues they care about. In this case, running a “good” SMD candidate reduces an individual voter’s propensity to defect. PR performance, therefore, is boosted by activating the predispositions of voters who already sympathize with the party. Second, a strong SMD candidate may attract new votes in PR by gaining the support of independent and undecided voters who have no declared party affiliation. Effective candidates may provide voters who are shopping around for a party information that factors into their estimate of the proximity of their preferences to the party’s ideal point on particular issues or their assessment of which of the parties is most credible in its promises to champion such issues. Good candidates advertising the policies of obscure parties, or of parties about which individual voters have little information, may bring in additional votes for their party list by persuading independents that the party is indeed the most appropriate choice. Third, strong SMD candidates may contribute to their party’s efforts to take votes away from competitors. Voters who have weak partisan affiliations with an ideologically proximate party may be induced to defect by SMD candidates whose message and personal qualities reinforce their doubts about the sincerity of their own party’s stances and its willingness and ability to accomplish its official programmatic goals. In spite of some ideological differences, such voters may be convinced, more easily than the party’s true believers, to switch their vote to a party that appears more likely to get the job done. In this case, placing the right SMD candidate may result in enhanced PR performance through “conversion.” Data and Variables In 1993, New Zealand replaced the first-past-the-post electoral system that had been in use for the previous 138 years with an MMP system styled after Germany’s (see Denemark 2001; Shugart and Wattenberg 2001a). The new electoral law allocates 56 percent of the seats in 67 SMDs under the plurality rule and the remaining seats through closed-list proportional representation in a single national constituency. Though the ratio of PR to SMD seats appears to balance the majoritarian and proportional principles of representation, under MMP the proportional component drives the outcome of the election. The total number of seats a party is assigned is entirely dependent upon its performance in the PR tier. Each party receives a number of PR seats that equals the number of seats to which it would be entitled if the entire House of Representatives were elected through PR, less the number of SMDs it won.6

70 / mixed electoral systems

We test the propositions put forth in this chapter on data from the NZES, which surveyed approximately 6,000 people after the 1999 legislative elections. The NZES is ideally suited for the study of contamination’s effect on individual voting decisions, as respondents are asked a variety of questions regarding their preferences, their evaluations of both parties and candidates, and their understanding of the MMP system. Party choice in the PR component of the election is evaluated through the use of a conditional logit model, which is a variant of multinomial logit (see McFadden 1973).7 Contrary to multinomial logit, however, conditional logit allows for choice-specific explanatory variables, which take on different values for each of the outcomes of the dependent variable (see Long 1997). Also, conditional logit estimates only one coefficient for each independent variable; the parameters estimated measure the effect that the change in the value of an independent variable for a given party has on the probability of choosing that particular outcome.8 ●

●

●

PR Vote. The dependent variable is the “PR Vote.” This variable has six categories, corresponding to the parties that voters were asked to evaluate: Labour, National, Alliance, New Zealand First, ACT New Zealand, and the Green Party. These parties are the most important in New Zealand politics; in 1999, they were also the only ones to win seats.9 Each outcome is coded 1 if the respondent indicated casting a vote for that party and 0 otherwise. Candidate Ratings. Consistent with the theory of contamination, the central hypothesis we assess is that a voter’s evaluation of an SMD candidate affiliated with a particular party list affects the probability that the voter will cast a ballot for that party in PR. For each possible outcome of the dependent variable, the respondents were asked to rate, on a 0–10 scale, the SMD candidate nominated in their district by those parties. Higher ratings for a particular SMD candidate should result in higher probabilities of voting, in the PR portion of the election, for the party with which the candidate is affiliated. SMD Vote. The second measure of the impact of contamination is the vote cast in the SMD portion of the election. For each party, this variable is coded 1 if the voter selected that party’s candidate in the SMD race.10

Of course, finding a statistically significant relationship between candidate ratings (or the SMD vote) and the probability of casting a PR vote for a particular party does not necessarily prove that voting decisions are shaped by the interaction between the two tiers of the election. Rather, a voter may rate a candidate highly because of her ideological proximity to the party that

interactive ballots / 71

the SMD candidate represents, because she identifies with that party, or because of the strength of her party affiliation. In turn, these factors are likely to impact both the PR and the SMD vote choice. Failure to consider these factors would almost certainly result in the overestimation of the impact of contamination. To account for this complication in the research design, we follow King, Keohane, and Verba (1994) and treat the endogeneity problem as omitted variable bias. The models presented in the next section include several independent variables measuring factors that explain party preference (partisanship, party evaluations, ideology, and positions on issues) and allow us to isolate the impact of that SMD candidate ratings have on the PR vote independently of the effects of party. ●

●

●

●

●

Partisanship. For each outcome of the dependent variable, partisanship is coded 4 if the respondent indicated “very strong identification” with the party, 3 for “fairly strong,” 2 for “not very strong,” 1 for “feel a little closer,” and 0 for no identification. This variable controls for the possibility that party choice is driven by long-term, primarily affective orientations toward the parties in competition. Party Rating. For each outcome, this variable measures how much a respondent “dislikes” or “likes” the party (on a 0–10 scale). We include this variable because it may measure considerations that factor into a voter’s evaluation of a party and whose effect may not be adequately captured by the other controls. Proximity. For each outcome of the dependent variable, proximity corresponds to the distance between the respondent’s ideology and that of the party. Each respondent was asked to identify her ideology on a 0–10 left–right scale and to do the same for each of the parties under consideration. We calculate the absolute value of the difference between the respondent’s ideology and her subjective evaluation of the party’s position. Ideological distance should be negatively correlated with the probability of voting for a particular party. Evaluation of Leader. For each party, we include a measure (on a 0–10 scale) of how much voters like its leader. Consistent with Vowles (2003), we expect the evaluations of party leaders (who are typically their party’s candidate for prime minister) to affect the likelihood of choosing a particular PR list. Issue Voting. In the NZES, voters were asked to identify the party whose views on particular issues are closest to theirs. We include the economy, taxes, health, education, race relations, and unemployment. For each party, we code a particular issue 1 if the voter indicated that the party best represents her views and 0 otherwise.11 Unfortunately, including these variables reduces the number of categories of

72 / mixed electoral systems

the dependent variable to five, as voters were not given the option of identifying the Green Party as being the closest to them on the issues (they were, however, given the possibility of marking “other”).12 Analysis and Discussion The results of models estimated through conditional logit are shown in table 5.1. In model 1, we exclude the issue voting variables to assess contamination on a dependent variable that includes the Green Party as a possible PR choice. Model 2 is a basic model that excludes variables measuring the impact of the SMD campaign on the voting decisions made in PR. Measures of SMD candidate ratings (models 3 and 5) and the SMD vote (models 4 and 5) are added to the restricted model to compare the explanatory power of alternative specifications. Under the null hypothesis of no contamination, the coefficients for candidate ratings and the SMD vote should be statistically undistinguishable from 0; moreover, the fit of the models should not significantly improve as a result of the inclusion of such variables. As predicted, variables measuring partisanship, party ratings, and evaluations of party leaders exert a positive, statistically significant effect on the probability of voting for a party list, while voters are less likely to vote for a party as the distance between their ideology and their subjective evaluation of the party’s position increases. Also, the probability of casting a PR ballot for a given party increases if voters identify that party as being the closest to their views on the economy, taxation, education, race relations, and unemployment. As for evidence of contamination, table 5.1 highlights three notable results. First, including variables measuring the effects of contamination significantly improves the fit of the models. We conducted likelihood ratio tests (see Long 1997) to compare model 2 with models 3, 4, and 5.13 In each of these cases, the test statistic is significant at the .01 level, suggesting that adding SMD candidate ratings and the SMD vote significantly increases the model’s ability to explain party choice in the PR segment of the election. Second, as models 3 and 4 reveal, the probability of voting for a party in the proportional component of the election significantly increases as the voter’s evaluation of the party’s SMD candidate becomes more favorable. Moreover, the vote cast by the respondent in the SMD race has a positive effect on the probability of voting for the same party in PR. Models 3 and 4 show that SMD candidate ratings and the SMD vote matter to PR vote choices independently of a voter’s assessment of a party, the intensity of partisanship, the voter’s closeness to the party on both specific issues and more general left/right positions, and personal judgments about party leaders.

73 Table 5.1 Contamination and party choice: Conditional logit estimates Dependent variable: PR Vote 1 SMD Candidate rating SMD vote

.08* (.02) ––

Partisanship

Economy

.41* (.03) .58* (.04) ⫺.21* (.03) .23* (.03) ––

Taxes

––

Health

––

Education

––

Race relations

––

Environment

––

Unemployment

––

Pseudo R2 Likelihood Ratio Test (LR ␹ 2u ⫺ LR ␹ 2r )i No. of observationsii

.684 ––

Party rating Proximity Party leader

10107

2 ––

3

––

.06* (.02) ––

.33* (.03) .50* (.03) ⫺.20* (.03) .20* (.02) .28* (.07) .36* (.07) .08 (.08) .23* (.08) .49* (.08) ⫺.17 (.09) .36* (.07) .711 ––

.34* (.03) .46* (.04) ⫺.22* (.03) .23* (.03) .28* (.08) .38* (.08) .02 (.09) .20* (.09) .42* (.10) ⫺.16 (.10) .38* (.08) .714 7.55*

14781

9860

4 — .26* (.07) .28* (.03) .49* (.03) ⫺.19* (.03) .20* (.02) .26* (.07) .35* (.07) .06 (.08) .21* (.08) .49* (.08) ⫺.19 (.09) .35* (.07) .713 14.87* 14550

5 .03 (.03) .25* (.08) .31* (.04) .45* (.04) ⫺.21* (.03) .23* (.03) .27* (.08) .37* (.08) .003 (.09) .19* (.09) .42* (.10) ⫺.18 (.10) .38* (.08) .717 16.42* 9735

Note: *p ⬍.05; Standard errors in parentheses. i

The likelihood ratio test compares the fit of the basic restricted model (model 2), as measured by the likelihood ratio ␹2, with each of the unrestricted models, (3, 4, and 5). To ensure comparability between the models, we calculate the quantities (LR ␹ 2u ⫺ LR ␹ 2r ) by using estimates from models that include the same observations (rather than those shown). From each of the models, we exclude every observation having missing values for SMD vote and candidate ratings (see Long 1997, 97–98).

ii The number of observations corresponds to the number of respondents multiplied by the number of outcomes of the dependent variable. Single observations where values of the independent variables are missing are excluded; respondents who have not voted for any of the six parties are also excluded.

74 / mixed electoral systems

These results provide evidence for the proposition that voters employ external cues from the SMD portion of the election in their choice of a PR list. If voters like an SMD candidate, the probability they will vote for the party that nominated that candidate increases; conversely, running bad candidates may lose parties some PR votes. Finally, model 5 indicates that, when indicators of candidate ratings and the SMD vote are considered together, candidate evaluations exert an effect on party choice in PR that is not distinguishable from 0. Given that candidate ratings are highly correlated with the SMD vote, they exert no independent, direct effect on the choice of a party list when included in the same model as the variable that measures SMD voting decisions. But, this finding is significant because it hints at how contamination affects party choice in PR. Specifically, Cox and Schoppa (2002) predict that nominating a good SMD candidate encourages straight ticket (or “sticky”) voting. The results presented in model 5 support this scenario: parties that run good SMD candidates provide voters with information that increases the likelihood of winning both their SMD and PR votes. The parameter estimates obtained through conditional logit are consistent with the chapter’s theoretical expectations. Still, it is possible that the relationships may be spurious. If variables that are omitted from the models affect both candidate evaluations (and the SMD vote) and the PR vote, we risk erroneously attributing the results to the impact of contamination. To exclude this rival explanation, we perform separate tests on different categories of voters: those who made their voting decisions prior to the campaign, those who decided during the campaign, those who “seriously thought” about voting for another party or not voting, and those whose choice of a PR list was never in doubt (table 5.2). We expect that SMD candidate evaluations influence party choice only for voters who decided during the campaign and those who at some point considered other options, but not for others. In fact, we noted that SMD candidate placement should influence PR vote choices through the effect of campaign information. Hence, we should not expect SMD candidate ratings to influence the PR voting decisions of voters who made up their minds before the campaign, for whom campaign information was irrelevant, and of those whose support for a particular party was never in question. For these voters, the assessment of parties alone should have a significant impact on the PR vote. If the coefficients expressing the effects of SMD candidate ratings on PR decisions are significant for each of these types of voters, the results would indicate that the relationships found earlier may be spurious. The results presented in table 5.2 conform to expectations. SMD candidate evaluations have a positive, statistically significant impact on the PR vote choice only for voters who decided during the campaign and those who at any

interactive ballots / 75 Table 5.2 Contamination and party choice: Conditional logit estimates Dependent variable: PR Vote Decided during the campaign

SMD Candidate rating Partisanship Party rating Proximity Party leader Economy Taxes Health Education Race relations Environment Unemployment Pseudo R2 No. of observations

Considered voting for different party

Yes

No

Yes

No

.06* (.03) .19* (.05) .47* (.05) ⫺.23* (.04) .19* (.03) .31* (.11) .20* (.10) .02 (.12) .19 (.12) .22 (.14) ⫺.15 (.14) .34* (.11) .56 3717

.05 (.04) .47* (.05) .47* (.03) ⫺.20* (.05) .28* (.04) .23 (.14) .61* (.13) ⫺.04 (.14) .20 (.14) .69* (.16) ⫺.22 (.16) .43* (.13) .82 6143

.09* (.03) .28* (.05) .49* (.06) ⫺.19* (.04) .19* (.04) .31* (.11) .29* (.10) .05 (.12) .19 (.12) .32* (.14) ⫺.12 (.14) .41* (.11) .62 4039

.01 (.04) .41* (.05) .44* (.05) ⫺.25* (.05) .28* (.04) .24 (.13) .49* (.12) ⫺.04 (.14) .14 (.14) .50* (.15) ⫺.20 (.16) .35* (.13) .79 5778

Note: *p ⬍ .05; Standard errors in parentheses. The number of observations corresponds to the number of respondents multiplied by the number of outcomes of the dependent variable. Single observations where values of the independent variables are missing are excluded; respondents who have not voted for any of the six parties are also excluded.

time considered voting for a different party. Conversely, the party choices of voters who decided before the campaign began and of those who never pondered changing their PR voting decisions are exclusively shaped by their partisanship, party ratings, evaluations of party leaders, and their closeness to a party’s ideology. As predicted, information from the SMD campaign only influences the PR vote choices made by voters whose decision-making process might have actually been influenced by campaign information. For these

76 / mixed electoral systems

P(Y = m|X)

0.6

0.5

0.4

0.3

0

2

4 6 Labour candidate ratings

8

10

Labour 90% Confidence Int. Alliance 90% Confidence Int.

Figure 5.1 “Leaner”.

SMD candidate evaluations and party choice—Example 1: Labour

voters, the choice of a party list in the PR component of the election is affected by external cues provided by the parallel SMD race.14 The substantive impact of the variables of interest remains to be addressed. Specifically, we outlined three mechanisms by which SMD nomination strategies may help a party boost its PR performance. Do these predictions hold? We use the estimates obtained from model 1, table 5.1, to simulate change in the predicted probabilities of voting for each of the parties in competition as we vary the values of candidate ratings for one of the parties and hold every other variable constant.15 The 90 percent confidence intervals of these predicted probabilities are presented in figures 5.1 and 5.2. When assessing these results, two cautionary notes are in order. First, given that the independent variables are choice specific, we can only take into account few of the hundreds of possible combination of values that each voter may assign to the parties for the variables under consideration. Second, interpreting the results in this fashion is problematic, as it involves making the unrealistic assumption that, as candidate ratings increase, all of the other variables remain constant.16

interactive ballots / 77

P(Y = m|X)

0.6

0.5

0.4

0.3

0

2

4

6

8

10

Labour candidate ratings Labour 90% Confidence Int. Alliance 90% Confidence Int.

Figure 5.2 SMD candidate evaluations and party choice—Example 2: Independent voter.

Consider a left-wing voter who is leaning toward Labour (partisanship is coded 1 for Labour) but whose ideology is equidistant from Labour and Alliance (1). Ideological distance is higher when the hypothetical voter compares her position to that of the Greens (2), NZ First (3), ACT (5), and National (6). Further, assume that the voter rates the various parties, their leaders, and their SMD candidates by assigning them the same values (7 for Alliance, 5 for the Greens, 4 for NZ First, and 2 for ACT and National). Labour and its leader are given a rating of 7. How do the probabilities of voting for each of the parties change as the voter’s rating of the SMD candidate fielded by Labour changes? When Labour’s candidate is rated equally to Alliance’s (7), the voter has a 55 percent probability of voting for Labour and 37 percent of voting for Alliance (see figure 5.1). Moreover, as the evaluation of the Labour candidate increases, the probability of voting for Labour also increases (to 62 percent for ratings of 10). As predicted, running good candidates reduces the probability that weak partisans will switch in the proportional segment of

78 / mixed electoral systems

the election. However, if Labour nominates a bad candidate, the voter becomes more likely to vote for Alliance and defect from the party with which she weakly identifies. Consistent with our theoretical expectations, Alliance has the opportunity to convert the hypothetical Labour “leaner” and induce her to vote against her party identification in PR by running a better SMD candidate.17 Now consider a similar scenario (the values of the independent variables are the same), with an independent voter (figure 5.2). In this case, when the leftist voter who is ideologically indifferent between Labour and Alliance rates the SMD candidates fielded by the two parties equally, she has the same probability of voting for each party (about 0.45). As the ratings of the Labour candidate decrease, the likelihood that she will vote for Labour (Alliance) appreciably decreases (increases). By contrast, as the ratings of the Labour candidate improve, the probability that she will vote for Labour in PR becomes considerably larger. In this case, candidate ratings are the difference between voting for a party or another. Running good candidates can win over independent and undecided voters. The hypotheses in this chapter find robust empirical support in the data. The conditional logit coefficients indicate that evaluations of SMD candidates and the SMD vote exercise a statistically significant impact on party choice in the PR segment of the election. Perhaps more importantly, SMD candidate evaluations have a substantively significant effect on party choice in PR. Under some circumstances, the nomination of strong SMD candidates attracts voters, in the PR component of the election, who might have otherwise voted for other parties. In addition, fielding good SMD candidates may help a party attract independent and undecided voters. As these results show, under mixed electoral rules PR voting decisions are influenced by the dynamics of the simultaneous SMD races. Conclusion This chapter’s findings complement the analyses performed on district-level data in previous chapters by specifying a micro-mechanism that supports the observed macro-level patterns. Specifically, running a candidate in an SMD race improves a party’s PR performance because campaign information from the SMD component of the election spills over to the proportional tier, where voters use it to inform how to cast their ballot. Voters are simultaneously exposed to campaign messages crafted for both tiers of the election; quite plausibly, they do not compartmentalize that information and do not engage in separate decision-making processes when choosing how to cast their two ballots. Once again, these findings suggest that the

interactive ballots / 79

majoritarian and proportional components of the election are interactive, not independent. Our findings also suggest that the quality of the candidates nominated in SMD matters to individual voting decisions in PR. Nominating competent, experienced, or charismatic SMD candidates may attract, in the PR tier, the support of voters who have not yet made up their minds. For these voters, external cues generated in the SMD campaign may help overcome uncertainty and low information to reach a voting decision in the PR segment of the election. Attractive SMD candidates have coattails that translate into additional PR votes for the parties they represent. To what extent are these findings generalizable to other mixed systems? The extent to which campaigns affect voting decisions is likely to vary across countries, across elections over time, and even across districts within a particular country. Factors like a population’s overall levels of partisanship, the competitiveness of the race, the public’s exposure to campaign information, the issues over which the election is fought, and the amount of resources expended by contenders are likely to affect the magnitude of the changes in vote intentions. Predictably, the incidence of contamination will also be affected by such contextual dynamics. New Zealand, for instance, is characterized by a highly personalized style of politics. In that country, where representatives elected through PR are still considered “second-class MPs” (see Barker et al. 2001), it is reasonable to expect SMD races to impact PR voting decisions more markedly than in other contexts. However, the fact that district-level evidence of contamination was found in a number of other mixed systems suggests that these patterns may be typical of SMD–PR combinations rather than idiosyncratic.

This page intentionally left blank

C h ap t e r 6 Voting from their Hearts? Contamination and Strategic Voting

Although the number of entrants in SMD races is likely to vary across different mixed systems, depending on how majoritarian and proportional institutions are combined, the number of candidates participating in the election is generally larger in the SMD component of mixed electoral systems than it is in pure forms of majoritarian elections. As we have shown in previous chapters, parties may garner additional votes in the PR segment of the election by running strong SMD candidates. What are the implications of a higher “supply” of candidates for the applicability of Duverger’s Law to the majoritarian elections that take place under mixed electoral rules? In “pure” SMD systems employing first-past-the-post, strategic voting is generally expected to produce two-party competition at the district level, or Duvergerian equilibrium, as voters tend to desert candidates who are unlikely to win. A voter is considered to be “strategic” or “sophisticated” if she maximizes expected utility by casting a ballot for a candidate who does not rank first in her preference ordering.1 Strategic voting under SMD has been the subject of extensive theoretical studies (see Farquharson 1969; McKelvey and Ordeshook 1972). Also, empirical research has evaluated the incidence of strategic voting in a variety of settings (see Abramson et al. 1992, 1995; Kim and Fording 2001). This chapter evaluates the likelihood that the majoritarian component of mixed electoral systems will also produce district-level two-party competition. If Duverger’s law applies to the SMD tier of mixed electoral systems, Duvergerian equilibria should characterize those races, as short-term, instrumentally rational voters should desert nonviable, marginal candidates. In this context, the presence of a greater number of entrants that is generally observed in mixed systems should have no effect on voting behavior. On the other hand, as we discussed in chapter 3, it is possible that a higher number of entrants may result in multiparty competition. In this chapter,

82 / mixed electoral systems

we illustrate and empirically evaluate the mechanism that undermines two-candidate competition in the SMD tier. Through the use of individual level data, we show that (1) strategic voting is less common in mixed systems than in pure forms of plurality elections and that (2) an individual voter’s propensity to vote for a frontrunning candidate decreases if the voter has the option of voting for a candidate representing her most preferred party. Consequently, the SMD component of mixed electoral systems should be characterized by multiparty competition because a higher number of entrants is likely to produce greater dispersion of votes away from the top two finishers. When given the opportunity, voters are more likely to “vote from their hearts” and cast an SMD vote that is consistent with their party preferences. We also demonstrate that previous studies, which have primarily employed district-level election returns, have overestimated the incidence of strategic voting in mixed electoral systems and show that the source of reductive impact on the number of parties is the parties’ strategic entry decisions rather than strategic voting. Given that sophisticated voting is an individuallevel phenomenon, analyses employing district-level data inevitably suffer from the ecological inference problem. In this chapter, we focus on voting behavior in the nominal tier of the MMM systems used in the Japanese Lower House (1996) and Upper House (1983) elections. We use public opinion data from the 1996 Japanese Election and Democracy Study (JEDS96) and the 1983 Japan Election Study (JES83). In Japan, the House of Representatives is elected through an MMM system that combines PR with first-past-the-post elections; majoritarian elections in House of Councilors take place in multimember districts under the single non-transferable vote (SNTV), while 40 percent of the seats are allocated through PR.2 How Aggregate Data Overestimate Strategic Voting Most of the existing literature has concluded that strategic voting is common in mixed electoral systems. Through the study of aggregate data from the 1996 elections of the Japanese Lower House, Kohno (1997) and Reed (1999) found evidence that voters tend to behave strategically in SMD. Though they are careful to point out some of the limitations in the research design they employ, the authors evaluate strategic voting by comparing a party’s performance in the two components of the election. These scholars assume that citizens vote sincerely in PR; hence, the difference in the vote shares garnered by a party in SMD and PR is believed to reveal the extent to which that party gains or loses votes in the majoritarian tier because of strategic voting. Kohno (1997) finds that large parties like the Liberal Democratic Party (LDP) tends to gain in SMD, while smaller

voting from their hearts? / 83

parties generally receive fewer votes in majoritarian elections. Similarly, Reed (1999, 257) concludes that there is “strong evidence for strategic voting: candidates with a good chance of winning receive more SMD than PR votes, and those with little chance of winning receive more PR votes.” Analyses of district-level election data overestimate the occurrence of strategic voting because they fail to consider whether voters have the opportunity to vote sincerely.3 The logic of contamination suggests that the presence of simultaneous PR elections should induce parties to nominate candidates in the SMD component of mixed electoral systems, undermining two-candidate competition. However, as we have shown in chapter 4, it does not necessarily follow that each party contesting the election will invariably place candidates in every district. Voters consequently face different patterns of candidate placement across districts. In one district, voters may only be able to choose between two candidates, while in others they may find five or six candidates on the ballot. In this context, some voters have the opportunity to express an SMD vote that is consistent with their party preferences while others do not. If a party does not run a candidate in a given district, voters who identify with that party are forced to either abstain or vote for one of the available choices. Failing to account for the opportunity to vote sincerely is likely to result in the overestimation of strategic voting. In fact, many of the voters who do not have the opportunity to vote for a candidate representing their preferred party and may transfer their vote to a major party candidate, may be erroneously classified as strategic voters.4 In reality, some of these voters may have behaved sincerely had the choice set included a candidate nominated by the party with which they identify. The data in the JEDS96 and JES83 suggest that a substantial number of voters did not, in fact, have such an option. In the 1996 Lower House elections and in the 1983 Upper House elections, about 90 percent of the respondents had the opportunity of voting for a candidate nominated by the party they rated highest on feeling thermometers.5 Between 8 and 9 percent, however, were forced to either abstain or vote for other candidates. Classifying the portion of these voters that cast a ballot for a front-runner as sophisticated may overestimate the incidence of strategic voting. It is important to note that voters who transfer their ballots under these conditions are not classified as sophisticated voters. These voters transfer their votes to a front-runner since they do not have the opportunity to vote for their most preferred candidate. The key characteristic of sophisticated voting is that voters “desert” their most preferred candidates so that they can maximize their expected utility. One of the goals of this chapter is to establish that these voters are more likely to vote for their most preferred candidates (i.e., vote sincerely) if they had the opportunity. That is, the nature of their behavior differs from that of

84 / mixed electoral systems

sophisticated voting. These voters are not necessarily interested in maximizing their expected utility. Strategic Voting Under Mixed Electoral Rules In “pure” SMD systems, two separate dynamics contribute to the development of two-party competition. Realizing that only one candidate can actually win the seat available in their district, voters who are motivated by shortterm, instrumentally rational considerations should desert candidates who are not expected to win and should coordinate on the race’s front-runners. If enough information is available to distinguish between the top two candidates and the rest of the field, prospective third-place finishers should receive very few votes, leading to two-party competition in the district. At the same time, parties anticipate the effect of strategic voting on the performance of their candidates. Parties expecting to receive insignificant vote shares should withdraw from competition, enter alliances with other parties, or lend their support to candidates with similar ideological orientations. In this sense, SMD should reduce the number of candidates receiving positive vote shares to two by (1) reducing the number of entrants and by (2) providing voters with the incentive to coordinate away from hopeless candidates. Cox (1997) extends this reasoning to multimember districts and concludes that strategic voting and entry constrain the number of candidates to M ⫹ 1. Consistent with this account, Reed (1990) shows that the M ⫹ 1 party competition achieved under the old SNTV rules employed in Japan was primarily the result of elite coalition building, rather than strategic voting. In mixed electoral systems, this dynamic partially breaks down. Parties, in fact, do not have as strong an incentive to withdraw from competition in the SMD component, because participation in SMD races offers them the opportunity to boost their PR performance. For this reason, SMD rules should fail to reduce the number of entrants to two.6 As a result, voters are generally presented with a greater number of candidates in the nominal tier of mixed electoral systems than they are in plurality elections; more of them will find a candidate who adequately represents their ideological preferences. In mixed electoral systems, therefore, the acquisition of Duvergerian equilibria may rely to a greater extent on the ability and willingness of voters to coordinate away from a crowded field of hopeless candidates, rather than on the preelectoral winnowing typical of strategic entry under majoritarian rules. Two-party competition can only develop if the inflated number of entrants is inconsequential to a voter’s decision-making calculus. We argue that there are good reasons to expect a greater supply of candidates to result in multiparty competition. In some cases, a larger number of

voting from their hearts? / 85

entrants may reduce the ability of voters to coordinate on the two frontrunners. A crowded field complicates voter calculations of candidate viability, rendering the identification of likely winners more problematic. As a result, greater uncertainty about the relative chances of the contenders may prevent voters from effectively coordinating on the winner and the first runner-up and may instead induce them to cast sincere votes for their preferred candidates.7 In this sense, for mixed systems to produce two-party competition in SMD in the face of a higher number of entrants, more coordination should take place among voters than in pure SMD systems, where the number of entrants is filtered by the incentives to withdraw from competition. Assuming that each candidate represents the preferences of a relatively large portion of the electorate, as the number of candidates on the ballot increases the amount of strategic voting that is required for Duvergerian equilibrium to emerge also increases. When parties fail to coordinate, more coordination must take place among voters for a two-party system to be established. We contend, however, that voters are less likely to coordinate in the SMD component of mixed systems where the number of entrants is higher. As the number of candidates on the ballot increases, more voters are provided with an opportunity to vote for a candidate that closely represents their policy views. Voters who identify with small parties, in particular, may indeed vote for a front-runner if their party fails to run a candidate in their district. If their preferred party fields an SMD candidate, though, the probability that such voters will vote for a less preferred likely winner should, on average, decrease, especially for voters who have strong partisan attachments. In mixed electoral systems that encourage the proliferation of SMD candidacies, more voters have the opportunity to vote for a candidate of their choice. Consequently, it is reasonable to expect that a lower number of voters should vote strategically for a front-runner, while more will vote for marginally performing candidates. This is the simple mechanism by which mixed electoral systems should produce multiparty competition in SMD: a larger number of entrants translates into greater dispersion of SMD votes, as the entry of additional candidates should generally decrease a voter’s propensity to vote for the two most viable candidates. We evaluate this proposition empirically on the 1996 post–election survey data from the Japanese Lower House elections and the 1983 Upper House elections. For the Japanese Lower House elections, where first-pastthe-post is employed in majoritarian races, we hypothesize: the likelihood of voting for marginally performing candidates, relative to voting for one of the top two candidates, increases when voters have an opportunity to vote for a candidate who is nominated by their most preferred party. For the Japanese Upper House elections, where SNTV is employed in the nominal tier, we hypothesize: the likelihood of voting for marginally performing candidates, relative to voting for

86 / mixed electoral systems

one of the top M ⫹ 1 candidates, increases when voters have an opportunity to vote for a candidate who is from their most preferred party. The logic outlined earlier should hold in pure SMD systems as well. In this sense, we expect the effective number of SMD candidates to be higher in mixed systems than in regular majoritarian elections simply because the number of candidates on the ballot is generally higher and voters tend to vote sincerely. Two characteristics of mixed systems, in contrast to pure SMD, lead us to expect that an increased number of entrants should more substantially influence the effective number of parties. First, while third parties in SMD are generally squeezed out from competition (see, for instance, Rosenstone, Behr, and Lazarus 1996), in mixed systems the presence of separate PR elections allows minor parties to cultivate their electoral support and exercise greater influence on policy making. When such parties run candidates in the majoritarian tier of mixed systems, they are likely to receive a greater number of votes than they would in pure SMD, particularly from those voters who support them in the PR segment of the election. Another factor is that mixed electoral systems may make it easier for minor parties to persuade voters to “stick with them.” Cox and Schoppa (2002, 1031) argue that many voters who are generally swayed by appeals to not “waste their vote” on hopeless candidates in pure SMD systems are less likely to vote strategically in mixed systems, where the campaign messages to which they are exposed are also more likely to be mixed. Large parties may still exhort such voters to avoid wasting their votes on a losing candidate, but small parties may emphasize how the presence of simultaneous PR elections renders such desertions unnecessary. Cox and Schoppa (2002, 1031) expect “many to swing toward voting expressively” and predict that mixed systems should largely discourage strategic voting. In Japan, dual candidacies strengthen voters’ tendencies to stick with their most preferred candidates. Voters are less likely to engage in strategic voting and desert their most preferred candidates when their SMD candidates are also running in PR. This tendency is especially true when parties use these candidates’ SMD performance to determine the ranking in PR (Suzuki 1999). In Japanese Lower House elections, for example, some parties use the ratio of the SMD vote share of a candidate to the largest vote share in the same district to determine the ranking. Under such circumstances, voters are motivated to vote sincerely since candidates’ SMD vote shares will decide whether or not their candidates win a seat in PR. A Typology of Voters One of the empirical implications of the hypotheses advanced in this chapter is that the number of strategic voters should be lower in the SMD races

voting from their hearts? / 87

of mixed electoral systems than in pure majoritarian elections. To measure the frequency of strategic voting, we must investigate how voters actually behave in the majoritarian races of mixed electoral systems. We classify voters into several types, based on how they voted in the SMD tier of the Japanese mixed system and show that the occurrence of strategic voting is low relative to pure majoritarian systems. The first group of voters includes strategic voters. These are voters who have the opportunity to vote for a candidate placed by their preferred party; however, the candidate is trailing other candidates and, as a result, these voters transfer their ballots to one of the front-runners. The second group is made up of sincere voters. Sincere voters are similar to strategic voters as they have the opportunity to vote for a marginal candidate affiliated with their most preferred party. The difference between the first and second groups, however, is that sincere voters actually vote for their favorite candidate instead of transferring their votes to a less preferred, likely winner. The third group is composed of straightforward voters. These voters happen to prefer a front-running candidate, who also represents their preferred party, and cast their ballots for that candidate. The fourth group consists of voters who do not have the opportunity to vote for a candidate from their most preferred party and vote for one of the front-running candidates. These voters may be alternatively categorized as strategic voters, as they may split their tickets and vote for a major party candidate in SMD and a minor party in PR, or as sincere voters, since they might have voted for their most preferred candidate among those who were competing in their district. We classify these voters as non-opportunity voters to distinguish from both strategic and sincere voters. Anomalous voters make up the fifth category in our typology. Some patterns of voting can be classified as anomalous. For example, the voter’s preference ordering may be Candidate 1⬎ Candidate 2⬎ Candidate 3⬎ Candidate 4. The order of the candidates’ popularity may be Candidate 4⬎ Candidate 3⬎ Candidate 2⬎ Candidate 1. The voter may cast her ballot for Candidate 2, even though Candidate 2 may be considered a marginal candidate. It may be especially rational to vote for Candidate 2 when the race between Candidate 2 and Candidate 3 is close. For simplicity, such a voter is classified as an anomaly because she did not transfer her ballot to one of the front-runners. By resorting to this residual category, we avoid more complex classifications and can simply focus on the incidence of strategic voting. The final group includes voters who did not vote in their SMD race. A voter may abstain for a variety of reasons. For instance, a voter may abstain because the candidate who is nominated by her preferred party is not present in the district. It is also possible that the voter is indifferent among different candidates and is thus not motivated to turn out. Although

88 / mixed electoral systems

diverse factors motivate abstention, voters who abstain from voting are classified as a single group in this analysis. To construct our typology, we use opportunity and marginality factors as well as information about how voters actually cast their ballots. Opportunity and marginality factors are based on feeling thermometer questions and election returns. Feeling thermometers are used to ascertain voter preferences, while election returns are used to gain information concerning which parties placed SMD candidates and to distinguish the candidates who were leading the race from those who were trailing.8 Finally, we employ the respondent’s voting decisions as reported in the surveys. To classify voters into these six categories, we first identify respondents who have the opportunity to vote for a candidate affiliated with their most preferred party and is performing marginally. In the Lower House elections, candidates who did not finish first or second are considered marginal since the SMD component of the Lower House elections produces only one winner in each district. Similarly, candidates who are not one of the top M ⫹ 1 candidates, where M is the district magnitude, are considered marginal in the majoritarian segment of the Upper House elections.9 If a respondent voted for a less preferred candidate who is also one of the frontrunners in her district, she is classified as a strategic voter. If the respondent cast her ballot for a marginal candidate, she is classified as a sincere voter. When respondents voted for a less preferred candidate who is not a frontrunning candidate, they are classified as anomalous voters. Second, we identify respondents who do not have the opportunity to vote for a candidate representing their most preferred party. If these respondents cast their ballots for one of the front-runners, they are categorized as no opportunity voters. If they vote for one of the marginal candidates, they are classified as anomalous voters. Next, we identify respondents who happen to prefer a front-running candidate. If these respondents cast their ballots for their most preferred candidate, they are categorized as straightforward voters. If not, they are categorized as anomalous voters because they wasted their vote on a less preferred candidate. Finally, we identify respondents who did not turn out to vote through two sets of questions contained in the surveys. The surveys ask whether or not respondents went to the polls at all. Respondents who reported that they did not go are categorized as abstainers. Another question asks what competitor respondents voted for in the SMD component. If respondents answered that they did not vote, they are also categorized as respondents who abstained; these are the voters who went to the polls and only voted for the PR election. Table 6.1 shows the frequency of strategic voting and other types of voting behavior based on our typology. The results show that only 3 percent of the electorate can be categorized as strategic in the 1996 Lower House

voting from their hearts? / 89 Table 6.1 The incidence of strategic voting Lower house Strategic No-opportunity Sincere Straightforward Abstained Anomalous Total N

3% 5% 9% 50% 22% 11% 100% 963

Upper house 1% 4% 6% 55% 30% 4% 100% 1329

Source: JEDS96 and JES83.

Election, while only 1 percent voted strategically in the 1983 Upper House election. Sincere voters comprise 9 percent of the electorate in the Lower House and 6 percent in the Upper House. The percentage of no opportunity voters is 5 percent in the Lower House and 4 percent in the Upper House, indicating that a substantial number of voters cast their ballots for one of the front-runners when they did not find a candidate from their most preferred party. The largest category is straightforward voters. Half (50 percent) of the electorate belongs to this group in the Lower House, while more than half (55 percent) belongs to the category in the Upper House. The values for abstention in the Lower House and the Upper House are 22 percent and 30 percent, respectively. Anomalous voters comprise 11 percent of the respondents in the Lower House and 4 percent in the Upper House. The results suggest that the incidence of strategic voting in the majoritarian segment of Japan’s MMM systems is substantially smaller than in pure majoritarian systems. The literature on strategic voting suggests that strategic voters generally make up between 5 percent and 15 percent of the electorate in SMD systems (Black 1978; Kim and Fording 2001). Cain (1978) suggests that about 15 percent of the electorate voted strategically in the 1970 British General Election. Blais and Nadeau (1996) report that the percentage of strategic voting in the 1988 Canadian election was at least 6 percent. Even if we only focus on voters who turned out in the Japanese elections, strategic voting is still less common than in countries employing regular SMD rules. Black (1978) reports that 12 percent of those who voted in the 1968 and 1972 Canadian elections voted strategically. This percentage seems to be substantially larger than the percentages obtained from the Japan’s Lower House election. Strategic voting among those who turned out in the 1996 Lower House Election was only about 3 percent. The results above seem to support the claim that mixed systems discourage strategic voting. Yet, they do not show how the opportunity of voting for an SMD candidate nominated by the preferred party affects the probability of

90 / mixed electoral systems

voting strategically. In the next section, we test the significance of the opportunity factor statistically. We show that the frequency of strategic voting is reduced significantly when voters have the opportunity to vote for a candidate who represents their most preferred party and that voters in mixed systems do not necessarily try to maximize their expected utility like strategic voters in pure SMD. Data and Variables Voters have three main choices in an SMD election. The first is to vote for a marginal candidate. The second is to vote for a front-running candidate. The third is to abstain from voting. Accordingly, we code our dependent variable as having three outcomes: voting for a marginally performing candidate, voting for a front-running candidate, and abstaining from voting. The dependent variable is based on the typology described in the previous section. Respondents who were coded as strategic or non-opportunity voters cast a ballot for a front-running candidate. If respondents belong to the category of sincere voters, they are categorized as voters who cast their ballots for a marginal candidate. Anomalous voters may have voted for a front-runner or a minor party candidate. Straightforward voters are excluded from the analysis, because we are assessing how the opportunity factor affects the decisions of voters who do not identify with the parties to which front-running SMD candidates are affiliated. In the case of the Lower House, we consider voters whose preferred party did not run candidates in their district and voters who prefer candidates ranked third or lower. In the Upper House, we include voters whose preferred party withdrew from their district race and those who prefer a competitor who is ranked lower than M ⫹ 1. ●

Opportunity. The key independent variable measures whether or not a voter’s most preferred party ran a candidate in the SMD where the voter resides. If respondents have the opportunity to vote for a candidate representing their preferred party, they are coded 1. Otherwise, they are coded 0.

As was the case in the previous section, a respondent’s preferences over candidates are determined based on their party preferences. We also add a number of control variables, which constitute factors that are generally found to affect voting behavior. ●

Citizen’s Duty. The first control variable is voter’s sense of citizens’ duty. If a voter feels that it is her duty as a citizen to participate in

voting from their hearts? / 91

●

elections, she may be less likely to abstain from voting. The two surveys ask respondents whether or not they feel they have a duty to vote. There are five categories for responses, including “agree,” “somewhat agree,” “neither,” “somewhat disagree,” and “disagree”. Partisanship. The second control variable measures whether or not respondents support a political party. Voters who identify with a party are more likely to vote because they more intensely prefer one candidate to others. Similarly, respondents who strongly support a small party may be less likely to vote strategically. Stronger partisan attachments may discourage voters from deserting their most preferred candidate even when the candidate is trailing in their district. Other standard socioeconomic variables are also included in the models.

Analysis and Discussion We use multinomial logit and multinomial probit models to evaluate our propositions. The outcome variable, SMD vote choice, is a multichotomous qualitative variable.10 We test our hypotheses with both multinomial logit and probit models to verify if these two estimators produce robust findings. Table 6.2 shows the results of statistical tests performed on post–election surveys conducted after the 1996 elections of the Japanese House of Representatives. Table 6.3 shows results from the 1983 elections of the House of Councilors. The results in tables 6.2 and 6.3 generally support our propositions. The opportunity to vote for a candidate nominated by one’s preferred party increases the likelihood that voters who do not identify with the frontrunning candidates’ parties will vote for marginal candidates. As the negative coefficients on the opportunity variable in both the lower and upper portions of the tables indicate, an individual voter’s propensity to vote for a front-runner or abstain, relative to casting a ballot for a marginal candidate, decrease when the voter’s preferred party runs a candidate in their district. Although there are minor differences between the coefficients estimated through multinomial logit and multinomial probit, the results are consistent with our expectations.11 Our analysis supports the conjecture that the opportunity to cast an SMD vote consistent with one’s party preferences decreases the probability of voting for a likely winner. It is especially significant that the results hold for both the Lower House and Upper House elections, which use different formulae to allocate seats in the nominal tier. At the aggregate level, these findings imply that a higher number of entrants produces greater fragmentation of the SMD vote.

92 Table 6.2 The effects of the opportunity factor on voting decisions—Japanese lower house, 1996

Front runners

Opportunity Duty Party support Sex Age Education Income Union City size Intercept

Abstention

Opportunity Duty Party support Sex Age Education Income Union City size Intercept Pseudo R2 N

Note: *p ⬍ .05; Standard error in parentheses. Source: JEDS96.

MNL model

MNP model

⫺2.402* (0.482) ⫺0.268 (0.313) ⫺0.610 (0.455) 0.362 (0.440) 0.019 (0.021) ⫺0.180 (0.177) ⫺0.030 (0.120) ⫺0.264 (0.617) 0.146 (0.149) 1.945 (2.000)

⫺1.909* (0.432) 0.147 (0.284) ⫺0.039 (0.119) 0.296 (0.344) 0.005 (0.025) ⫺0.176 (0.136) ⫺0.571 (0.097) ⫺0.193 (0.490) ⫺0.389 (0.119) 0.803 (3.248)

⫺2.006* (0.516) ⫺0.807* (0.274) ⫺1.093 (0.510) 0.354 (0.469) ⫺0.024 (0.023) ⫺0.094 (0.184) 0.028 (0.129) 0.051 (0.602) 0.191 (0.159) 5.417 (1.903) .38 166

⫺1.554* (0.785) ⫺0.094 (0.1099) ⫺0.455 (0.463) 0.152 (0.397) ⫺0.027* (0.019) ⫺0.112 (0.149) ⫺0.025 (0.103) 0.007 (0.526) ⫺0.091 (0.109) 3.164* (1.453) .11 166

93 Table 6.3 The effects of the opportunity factor on voting decisions—Japanese upper house, 1983 MNL Model Front runners

Opportunity Duty Party support Sex Age Education Income Union City Size Intercept

Abstention

Opportunity Duty Party support Sex Age Education Income Union City size Intercept Pseudo R2 N

Note: *p ⬍ .05; Standard error in parentheses. Source: JES83.

MNP Model

⫺2.788* (0.508) ⫺0.152 (0.144) ⫺0.110 (0.630) ⫺0.013 (0.515) 0.002 (0.020) ⫺0.070 (0.368) ⫺0.018 (0.107) ⫺0.636 (0.663) 0.511* (0.246) 0.508 (1.952)

⫺2.089* (0.409) ⫺0.097 (0.141) 0.313 (0.221) ⫺0.117 (0.453) ⫺0.004 (0.018) ⫺0.099 (0.356) 0.005 (0.102) ⫺0.492 (0.564) 0.313* (0.221) 0.802 (2.100)

⫺1.618* (0.516) 0.228 (0.156) ⫺1.254* (0.605) ⫺0.603 (0.552) ⫺0.042 (0.022) ⫺0.911* (0.402) 0.134 (0.112) 0.074 (0.622) 0.105 (0.253) 4.179 (2.068) .43 145

⫺1.475* (0.528) 0.121 (0.105) ⫺0.488 (0.528) ⫺0.485 (0.525) ⫺0.028 (0.0185) ⫺0.545 (0.370) 0.092 (0.115) ⫺0.110 (0.480) 0.121 (0.105) 2.838 (1.706) .17 145

94 / mixed electoral systems

We also interpret the substantive effect of the opportunity variable by simulating the probability distributions we obtain by varying the value of the opportunity variable from 0 (voters do not have an opportunity to vote for a competitor representing their most preferred party) to 1 (voters have such an option). With predicted probability distributions, uncertainties surrounding the predicted probabilities can also be interpreted effectively. Since the estimation of coefficients under multinomial probit can be unstable, we use the coefficients from the multinomial logit models to simulate probability distributions. Figures 6.1 and 6.2 display the results from the Lower House and Upper House elections, respectively. Both are composed of three sets of figures, representing the predicted probabilities voting for a marginal candidate, voting for a front-runner, and abstaining. Each of these figures contains two distributions: one represents the choice of voters with the opportunity to vote for a candidate from their most preferred party, whereas the other represents that of voters deprived of such an opportunity. The simulated probability distributions strongly support the expectation that voters who have the option of voting sincerely are, in fact, less likely to vote for a front-runner. Figures 6.1a through 6.1c show the probability distributions we estimated for the Lower House elections. These figures indicate that voters who identify with a minor party have about a 0.50 probability of voting for a marginal candidate when their party is contesting the election in their district. When a candidate representing their preferred party is present, the probability of voting for a front-runner is about 0.30, while the probability 13

Kernel density

Not Present

Present

0 .75

0 Pr(choice = marginal)

Figure 6.1a Predicted probabilities (marginal candidates).

voting from their hearts? / 95 28

Kernel density

Present Not Present

0 .20

.70 Pr(choice = front runner)

Figure 6.1b Predicted probabilities (front runners).

Kernel density

10

Not Present

Present

0 .50

0 Pr(choice = absent)

Figure 6.1c Predicted probabilities (absent).

of abstaining is approximately 0.20. By contrast, when these voters do not have the option of voting for a candidate affiliated with their party of choice, the probability of voting for a front-runner or abstaining increases considerably (to about 0.45 and 0.35, respectively). In addition, the probability distributions do not overlap much in figures 6.1a and 6.1b, while overlap is more substantial with abstention. This indicates that the impact of the opportunity variable is not as large with regard to the probability of abstaining.

96 / mixed electoral systems

The Upper House results are shown in figures 6.2a through 6.2c. These figures display similar patterns to those observed in the Lower House election. We find a significant impact of the opportunity factor on the likelihood of voting for a marginal candidate (figure 6.2a) and for a front-runner (figure 6.2b). Figure 6.2a suggests that the probability for voting for a marginal candidate is about 0.67 when voters have the opportunity to cast an SMD vote consistent with their party preferences, whereas the probability 8

Kernel density

Not Present

Present 0 0

.88 Pr(choice = marginal)

Figure 6.2a Predicted probabilities (marginal candidates). 30

Kernel density

Present

Not Present

0 .62

.08 Pr(choice = front runner)

Figure 6.2b Predicted probabilities (front runners).

voting from their hearts? / 97

Kernel density

8

Present

Not Present

0 0

.65 Pr(choice = absent)

Figure 6.2c Predicted probabilities (absent).

is 0.17 when voters do not have such an option. According to figure 6.2b, the probability of voting for a front-runner is noticeably higher when voters do not have the opportunity of casting an SMD vote for their preferred party (0.30) compared to when voters have such an opportunity (0.10). As was the case in the Lower House election, the two predicted probability distributions overlap to a greater extent in figure 6.2c, suggesting that the impact of the opportunity factor is not significant to the likelihood of abstention. These results indicate that the voting behavior of those who do not support the front-runners’ parties is significantly affected by the strategic entry decisions made by the parties with which they identify. If their party of choice withdraws from the race and fails to place a candidate in their district, those voters are quite likely to vote for one of the front-runners. Conversely, when their preferred party fields an SMD candidate, these voters become far more likely to vote sincerely rather than coordinate on the likely winners. This mechanism illustrates why the majoritarian tier of mixed electoral systems fails to reduce the number of parties to the same extent that pure SMD rules typically do. Under pure SMD, the low number of entrants generally induces voters to concentrate on the front-runners, often producing minimal deviations from two-party competition. Voters whose favorite party is not contesting the election, in fact, are considerably more likely to vote for a likely winner than waste their vote on another marginal candidate. Under mixed electoral rules, however, contamination generally encourages parties to nominate candidates irrespective of their chances of

98 / mixed electoral systems

success. In the presence of a higher number of entrants, voters are less likely to coordinate on the top two finishers and their votes are dispersed more widely. Consequently, in cases where the strategic entry dynamic reducing the candidate supply in SMD breaks down, multiparty competition, not Duvergerian equilibrium, is likely to emerge. Conclusion This chapter provides the missing theoretical and empirical links in our investigation of contamination’s influence on party strategy and voting behavior. Because running SMD candidates (especially strong SMD candidates) improves a party’s PR performance, parties have an incentive to nominate candidates in SMD races. This dynamic undermines the Duvergerian logic of strategic entry that is prevalent in first-past-the-post elections, especially under institutional rules where the majoritarian component does not dominate the election. In turn, a greater number of entrants in the SMD component of the election reduces the propensity of voters to coordinate on the front-runners. When given a chance, more voters tend to “vote from their hearts” and cast an SMD ballot that is consistent with their party preferences. As a result, mixed systems should generate multiparty competition in SMD or, at a minimum, a greater number of parties than we generally observe under regular majoritarian rules. The deviations from two-party competition are likely to be more marked in systems that discourage preelectoral coordination and strategic withdrawals from SMD contests. It is in such systems that “controlled comparisons” are most likely to yield biased estimates of the independent effects of PR and SMD.

C h ap t e r 7 Eyeing the Future: Contamination and Legislative Behavior

Victory or defeat in an election is determined by the formal results, announced during the evening or early morning hours after the polls have closed. Confident candidates and their zealous supporters wait impatiently for the ballots to be tallied in polling stations, aggregated nationally by the proper authorities, and reported by the news media. The losers graciously concede to their opponents and the winners make jubilant speeches. While party tacticians immediately begin planning for the next campaign, victorious parties and candidates also turn to a pressing responsibility: governing the country. As in much of this book, elections research often focuses on the relatively short period of active campaigning, from candidate and party nomination until the final results are published. But, what happens between elections profoundly affects the strategies of political actors in the campaign period. Policy choices in parliament determine different winners and losers: those who benefit and those who suffer from resource allocation decisions. The process of legislating often reveals the strengths and weaknesses of political parties, potentially affecting their performance in the subsequent round of electoral contests.1 The rules by which candidates are elected and their stands on specific policy items are closely connected. A member of parliament (MP) may be influenced by particular policy preferences or ideological predispositions when making decisions in the legislature. But, the desire to be reelected also serves as a powerful motivator (Mayhew 1974). In this chapter, we argue that contamination not only influences the behavior of voters, candidates, and parties in the campaign period, but also has significant implications for the actions of legislators as they govern. By altering the strategic environment inherent to pure SMD and PR, contamination encourages behaviors that violate standard assumptions about how legislators should act.

100 / mixed electoral systems

The Scholarship on Legislative Voting Behavior For over a century, scholars of American politics have assessed the voting proclivities of congressional representatives. In 1885, Woodrow Wilson noted that the “. . . Senate is just what the mode of its election and the conditions of public life in this country make it.” That is, legislators are strongly conditioned by the electoral process. But, the connection between elections and voting in early work on the American Congress assumed that parties in the United States were similar—organizationally and operationally—to political parties in European parliamentary systems (Rohde 1991). The failure of American legislators to evidence the kind of party loyalty found in European parliaments was linked to the decline of the party in the U.S. Yet, some researchers noted that cohesive party voting in the legislature was more likely to be manifested when “an electorate . . . votes for parties rather than individual personalities and from an institutional arrangement that vests party leaders with an authority recognized by rank and file members . . .” (Collie and Brady 1985, 283).2 In the comparative politics literature, scholars emphasizing the influence of elections on MPs’ behavior typically concentrate on the varying effects of majoritarian and proportional election rules (Bowler, Farrell, and Katz 1998; Cain, Ferejohn, and Fiorina 1987). Electoral rules are thought to exert a strong influence on voting: legislators in PR systems are more likely to toe the party line than legislators elected in SMD because they rely upon the party for ballot access, financial resources, and career advancement opportunities. Legislators elected under SMD, whose political future is largely contingent upon the popularity they enjoy in their districts, are expected to stray from the party line more frequently. The establishment of a tighter “local constituent–legislator” bond, in fact, often requires lower party discipline, as the party’s positions are likely to occasionally conflict with the parochial interests of specific localities (Carey 2003; Loewenberg and Patterson 1975; Weaver and Rockman 1993). These general expectations have been qualified by research that has attempted to probe the relationship between electoral systems and the importance of personal reputation (Carey and Shugart 1995). While PR is thought to encourage party-oriented behavior, the institutional features of proportional systems vary substantially. Party lists can be open or closed, constructed nationally or regionally, include one or multiple allocation tiers, use different electoral formulae, and employ varied levels of thresholds. Many existing PR systems encourage national-level, party-focused behavior among legislators, but some combinations of rules classified as PR promote a focus on local constituency needs. For example, Brazil’s use of open-list PR and several multimember constituencies create

eyeing the future / 101

incentives for MPs to pursue resource redistribution that is concentrated narrowly on local constituencies (Ames 1995a). Similarly, Italy’s own, prereform open-list system induced legislators to consolidate their incumbency advantage by passing a multitude of leggine, pork-barrel legislation generally enacted in committees, and by offering their constituents extensive facilitation services (Golden 2003). All party-list systems are not the same, but PR systems with closed, national lists are most likely to encourage the cohesive behavior generally attributed to PR systems. Similarly, constituency-based systems are not always characterized by weak party cohesion and deference to local concerns in the policy-making process. For example, parties in the United Kingdom enjoy relatively strong legislative cohesion despite the use of majoritarian electoral rules. Party control over nominations and MPs’ career paths in Great Britain provides a powerful incentive for ambitious legislators to comply with party demands. In a study of free votes, Hibbing and Marsh (1987) found that when parties loosened the chains controlling their members’ voting behavior, party attachments continued to drive a substantial proportion of the votes. But, other factors—including constituency features—also intervened to influence MPs’ decisions. Freed from the formal demands of the party, some legislators sought to satisfy local needs, as the conventional wisdom suggests should occur in legislatures selected by majoritarian rules. Prior to the adoption of its MMP system, New Zealand was similarly characterized by strong party discipline; legislators cultivated their personal vote by spending a great deal of time in their districts and by regularly performing constituency service (Anagnoson 1983). Analyses of legislative behavior in mixed electoral systems have tended to overlay the expectations developed for SMD and PR on mixed systems. Shugart (2001) argues that a mixed system “encourages some legislators to specialize in representing localities (through the nominal vote) and some to specialize in the priorities of party leaders (through list allocation).” Mixed systems, in this sense, should result in an ideal blend of local- and nationaloriented behavior by striking a balance between accountability to local constituents and responsiveness to the preferences of party bosses. Legislators elected to the same legislative assembly under different rules, in fact, are thought to face different sets of incentives: those elected on a party list have a clear incentive to please their party leaders if they wish to maximize their chances of reelection via higher placement on their party’s rankordered list. Legislators elected in SMD, by contrast, should display more independent behavior. Research on legislative behavior in the German Bundestag has suggested that two “legislator types” are generated by the separate incentives of SMD and PR (Stratmann and Baur 2002), and that legislators elected in districts

102 / mixed electoral systems

are more likely to favor pork-barrel projects than are their PR counterparts (Lancaster and Patterson 1990). The search for a “mandate divide” has driven researchers to investigate voting behavior in many mixed systems. Some scholars report evidence of distinct patterns among deputies occupying SMD or PR seats (Judge and Ilonszki 1995; Thames 2001). Other scholars find less conclusive evidence of mandate effects; the influence of seat type is substantially mitigated by controls for party affiliation (Haspel, Remington, and Smith 1998; Smith and Remington 2001). Consistent with the argument in this book, other research has questioned the existence of a clear division in mixed electoral systems, noting that the behavior of elected deputies is complicated by the competing incentives present in the SMD and PR components (Bawn and Thies 2003; Ferrara 2004b; Herron 2002b). This chapter builds upon earlier findings by outlining how the logic of contamination extends to legislative behavior and by providing empirical evidence from roll-call voting in two distinct mixed systems, Ukraine and Italy. We show that the type of seat a legislator occupies is not the sole determinant of the incentives for voting. Rather, contamination helps to account for legislative voting patterns. Electoral Path and Legislative Behavior The impact of electoral institutions on legislative voting is based on the ability of parties to control access to parliamentary seats. If parties exert substantial control over access, politicians must show some degree of party loyalty to gain and retain positions in the legislature.3 If parties have limited control over access, politicians may be less inclined to conform to party expectations. The extent to which parties may control access to a legislative assembly depends on three factors: ● ●

●

Whether the electoral rules allow voters to cast nominal votes; Whether ballot access depends upon party nominations or primary elections; and The degree to which such nomination decisions are centralized or left to regional party organizations.

When seat allocation in a purely proportional system is national and based on closed lists, local concerns and pork-barrel politics are often subordinated to a nationally focused agenda. The importance of personal reputation is minimized because votes are cast for parties, not candidates. Further, central party headquarters has complete control over who gets on the list and in what order. In an SMD system, parties cannot always prevent candidates from using the party label. Parties with centralized leaderships and strict nomination requirements can substantially influence legislative

eyeing the future / 103

behavior even if SMD rules are employed, but decentralized nomination procedures create strong incentives for pork-barrel politics. How does the combination of SMD and PR rules realized in a mixed electoral system affect the behavior of elected officials? It is possible, as Shugart and Wattenberg (2001b) hypothesize, that mixed systems would stimulate the emergence of two types of legislators: PR representatives who tend to conform with their party and SMD legislators who give precedence to the advocacy of local constituency interests. If this were true, the type of seat that legislators occupy should accurately account for legislative behavior. By simply controlling for seat type, and including variables reflecting alternate explanations, researchers should find clear evidence that deputies occupying PR seats conform to party preferences in roll-call voting more frequently than legislators occupying SMD seats. But, as we noted earlier, much existing scholarship finds weak evidence of this connection. Particular institutional characteristics of mixed electoral systems may weaken or altogether break the link between seat type and behavior. First, even if legislators specialized in the majoritarian or proportional components of the election, SMD deputies may not be less loyal than their colleagues elected in PR. Stringent internal party rules governing how candidates access the ballot may keep SMD representatives on a rather short leash, forcing them to cultivate their personal vote through constituency service more than legislative activity, especially if the latter conflicts with the party platform. Centralized nominations and the absence of residency requirements should blur the observed differences in the behavior of legislators elected via majoritarian and proportional rules. Second, most mixed systems permit candidates to contest seats on both ballots simultaneously.4 In fact, the rules often provide a great deal of flexibility to parties and candidates: a candidate could run solely in a constituency, solely on the party list, or in both races. In some mixed systems, such as Italy and Japan,5 political parties use dual candidacy to maximize the probability that their most prominent leaders will be elected. These provisions have been controversial in some countries; the PR list can be used to “save” a candidate rejected by his constituency. Regardless of its normative implications, though, dual candidacy blunts the incentive for legislators to “specialize.” Rather, the prospect being nominated to both a party list and an SMD race allows incumbents to hedge their bets, focusing some of their attention on demonstrating their partisanship to party leaders and some on showing their dedication to local constituents. This strategy would allow MPs to maximize their chances of reelection by keeping open both available avenues to continued legislative tenure. The incidence of dual candidacy suggests that mixed systems may discourage the specialization of deputies and that a narrow focus on seat type is likely to yield an incomplete account of the strategic environments faced

104 / mixed electoral systems

by winning legislators who care about their political future. We argue that it is not the type of legislative seat that determines behavior, but the candidate’s electoral path to the seat that affects cohesion with the party. By electoral path we mean: ● ● ● ●

How politicians get on the ballot; Which portions of the election they contest (PR, SMD, or both); How comfortably they were elected to their seats; Whether the safety of their seats depends on their personal popularity or their status within the party.

If the process by which SMD and PR legislators are nominated is similar and legislators have the possibility of either contesting both tiers or easily moving from one to the other, the behavior of the two “legislator types” is likely to be quite similar empirically. If legislators look forward to the next electoral cycle and make policy choices with an eye to their competitiveness in subsequent contests, the relative safety of their legislative seats should exercise some influence on their behavior. Identifying the relationship between seat safety and the voting behavior of SMD deputies can be challenging. In pure SMD systems, legislators elected in marginal districts are generally found to be less responsive to their party leaders, as they focus on increasing their popularity with local constituents (Deckard 1976; Sullivan and Uslaner 1978; Fiorina 1984). Under mixed electoral rules, however, deputies elected in safe SMDs have been shown to vote differently from their parliamentary factions more frequently than those elected in marginal districts (Haspel, Remington, and Smith 1999). Greater electoral marginality could result in higher party discipline under hybrid rules. If nominations are centralized, incumbents who narrowly defeated their opponents must seek nomination to a more secure constituency, perhaps lobby for greater financial and logistical support from party headquarters, or try to avoid placement in a less winnable district where the prospects of retaining their seat would be substantially compromised. SMD deputies who barely won their plurality race may feel that they have more to prove to their party leaders than their colleagues who won their respective districts by a wide margin. At the same time, legislators who have comfortably won an SMD race because of their personal popularity may induce their party leaders to nominate them to the same district, which they have demonstrated their ability to deliver, and to possibly tolerate slightly lower levels of partisanship. Similarly, a candidate who is placed high on the party list may feel more secure, freer to diverge from party preferences, while a candidate in a more

eyeing the future / 105

tenuous list position may want to show strong party loyalty as she strives to climb the ranks in later elections. The safety of a PR deputy’s seat depends on at least three factors. First, her ranking on the party list is critical: candidates whose names are toward the top of the list are more likely to win seats. Second, the district magnitude of the PR constituency to which she is nominated is important in countries employing regional PR lists, as small district magnitudes may render even high list rankings relatively unsafe. Third, the popularity of the party in that particular PR constituency also affects the likelihood that a candidate will be elected, which increases with the vote share received by her list.6 A PR deputy, in this sense, may maximize safety by improving her ranking on a list in any PR system. If the PR system employs regional rather than national seat allocation, she may also increase safety by improving her ranking on the list relative to district magnitude, or earning placement in a PR constituency where her party is anticipated to perform better. Tentatively, therefore, we expect deputies elected in unsafe seats to be more likely to conform with the voting patterns of their chosen legislative faction than those elected in safe seats. Such a pattern should be especially marked for PR deputies. Data and Variables Our empirical analysis focuses on legislatures in Ukraine and Italy. These systems share mixed electoral rules, but vary in important ways. Ukraine’s mixed electoral system, used in 1998 and 2002, is noncompensatory, uses a single national PR district, allocates half of its seats to PR and operates in an inchoate party system environment. Italy’s system is partially compensatory, uses regional PR seats, allocates one-quarter of its seats to PR and operates in an institutionalized party system environment. While deputies face different institutional incentives, Ukraine and Italy share mixed rules and dual candidacy in the period under analysis. In addition, both legislatures record votes electronically, providing a rich dataset with which to evaluate hypotheses. To assess MP voting behavior, we use electronically recorded votes in Ukraine’s Verkhovna Rada and Italy’s Chamber of Deputies. Ukraine’s votes were taken from May 1998 until July 1999. Italy’s votes were taken from June 2001 until May 2002. We assess every vote cast during those time periods. In Ukraine, we exclude unanimous votes and votes cast by legislators whose identities changed during the period under analysis. In Italy, we exclude secret votes, votes that were annulled, and votes that were declared void because of lack of quorum (for which the Chamber of Deputies provides no data). In Ukraine, we consider the behavior of 428 deputies; in Italy we analyze the legislative votes cast by 580 deputies who were members of parliamentary groups other than the gruppo misto.7

106 / mixed electoral systems

We use Iteratively Reweighted Least Squares8 to assess the data in this chapter. This method is more appropriate than Ordinary Least Squares (OLS), given the nature of the data under consideration, because the presence of notable outliers in the data may bias OLS estimates. Because of variation in Ukraine and Italy’s institutions and party system features, our coding varies slightly from country to country. However, we assess the same dependent variable and covariates that represent the same concepts. ●

●

●

●

●

Party Loyalty. The key dependent variable in our analyses is “party loyalty.” We measure the propensity of a legislator to vote consistently with her party as a percentage of the total votes in which the legislator could have participated. We operationalize this concept by calculating the percentage of votes where the legislator’s behavior (yea, nay, abstained, not present, or did not vote) matched the modal behavior of the parliamentary group of which she is a member.9 Seat. Most studies of legislative voting in mixed electoral systems privilege seat type as the key independent variable. We include this dichotomous variable as well, coding legislators by the seat they occupy: PR (0) or SMD (1). Seat Safety. As noted earlier in the chapter, in addition to mandate, we focus our attention on factors that could better represent electoral effects. The issue of seat “safety” is the first concern. We employ continuous interactive variables that measure the relative seat safety of SMD and PR deputies. For SMD legislators, we test the impact of the natural logarithm of their margin of victory on party loyalty; PR deputies are assigned a score of 0. For PR legislators, seat safety is calculated by dividing the effective vote received by their party list in the constituency where they were declared elected by the number of Hare quotas needed for them to be guaranteed a seat.10 To be guaranteed a seat, candidates ranked first on a party list only need their party to garner a number of effective votes that equals one quota, candidates ranked second need two quotas, and so on. We take the natural logarithm of those quotients; SMD deputies are assigned a score of 0.11 Dual Candidacy. To address the issue of dual candidacy and seat safety, we use a series of variables that reflect the combination of safety and seat type (PR, SMD, Dual). We created dichotomous variables to measure various types of dual candidates: those who were elected in safe SMDs and were ranked high on a party list, those who were elected in safe SMDs and were ranked low on a party list, those who ran in a marginal district and had a high list ranking, and those who ran in a marginal district and had a low list ranking.12 Incumbency. In most of the statistical tests, we control for incumbency (1 ⫽ incumbent, 0 ⫽ otherwise).13 Because freshman legislators may

eyeing the future / 107

●

view their seats as more precarious, they may be more beholden to the wishes of party secretaries than their senior colleagues. Party Affiliation. Consistent with most of the literature on the relationship between mandate and party loyalty in mixed electoral systems, we control for partisan affiliation to account for the possible variation in the cohesiveness of parliamentary groups. We create dummy variables that measure membership in each of the groups formed in the respective legislatures.14 Due to the large number of factions in both countries, we suppress some of the results for these variables in the tables, but address them in discussion.

Empirical Analysis: Party Discipline in Ukraine’s Verkhovna Rada The Verkhovna Rada is Ukraine’s unicameral parliament. Ukraine conducted legislative elections in 1994, 1998, and 2002.15 The elections in 1994 were held in SMD constituencies with a majority-runoff formula. Electoral districts were reconfigured for 1998 and the parliament was divided in half: 225 deputies were selected in SMD races with plurality rules and 225 deputies were selected through a single national district using closed-list PR with a 4 percent threshold. These rules were preserved in 2002, with some changes. In 2004, mixed electoral rules were abandoned for pure proportional representation. Our analysis, however, focuses on electronically recorded votes from the parliament elected in 1998. Influence over candidate nomination varied across parties in the 1998 contest, but parties were able to control use of their labels in two ways. First, parties controlled the construction of national lists for the PR component. By law, the list was to be constructed by the party’s highest managerial body.16 While internal party politics affected list construction, the central leadership had the final word on the contents. Second, parties asserted control over nominations in SMDs. While candidates could run as independents (as many candidates did in 1998), political parties were able to influence who used the party label in the campaign. The 1998 Ukrainian ballot included information about the candidate’s party membership, but did not include information about the candidate’s sponsorship. That is, the ballot provided no information about who nominated the candidate: a party or a voter group.17 If the candidate was a member of a party, however, the party name appeared on the ballot. Parties thus controlled their labels through membership requirements. By law, a party’s official candidate nominations were produced by local party organizations, subject to the party’s central requirements.18 With central control over party list construction, Ukrainian politicians were pressured to conform to party expectations for high list positions or

108 / mixed electoral systems

to use financial pressure to gain high list positions. Constituency candidates also faced some pressure from the party, given that party membership affected ballot language. The ease of nomination and success of independent candidates, however, relieved party-induced pressure for SMD legislators to conform to party norms. For success in SMD, candidates could build local support networks with or without party resources.19 Thus, we would expect Ukrainian candidates to feel some pressure from national party organizations for PR list access, and some pressure from local constituencies for support in SMD races. Dual candidates, however, would feel competing pressures from both the party and regional networks. Once elected to parliament, deputies formed factions that were initially based on electoral parties. Legislators who joined parliamentary factions gained access to committee chairmanships, staff and offices, automobiles, and other benefits (D’Anieri, Kravchuk and Kuzio 1999). Groups of 14 or more deputies were permitted to form factions.20 Table 7.1 lists factions and membership for legislators who occupied the same legislative seat from the beginning of the period under analysis (May 12, 1998) until the end (July 16, 1999).21 Table 7.1 illustrates that faction membership substantially changed over the 14 months in our investigation. All of the initial factions survived in some form and new factions were created. While membership was stable in some factions, it was volatile in others. Twenty-one deputies from Hromada (HRO) moved to the Fatherland (FAT) faction and the same number of People’s Democratic Party (PDP) deputies defected to Regional Revival (RR). Table 7.1 Faction affiliation and number of deputies in Ukraine’s Verkhovna Rada Faction Communist Party of Ukraine People’s Democratic Party Hromada Rukh/Rukh-Kostenko Rukh-Udovenko Left-Center Social Democratic Party of Ukraine (United) Greens Progressive Socialist Party Fatherland Regional Revival Reforms-Congress Independents Working Ukraine Peasant Party No Faction Total

May 12, 1998

July 16, 1999

117 85 44 43

116 30 16 27 14 24 26 23 14 28 25 20 19 16 15 15 428

34 26 24 15 — — — — — — 40 428

eyeing the future / 109

Deputies initially aligned with the PDP also formed the core of Working Ukraine (WU) and Reforms Center (RC). Factional defection benefited the established factions as well, although all initial factions except the Social Democratic Party of Ukraine (United) SDPU(O) experienced a net loss in membership. Discipline varied across factions. Many were characterized as occupying an ill-defined ideological center, labeled as occupants of a “swamp” as well as “parliamentary sludge.” Members of some centrist factions, such as the Green Party (GRE) and PDP, were accused of poor participation in plenary sessions and inconsistent views on policy matters. By contrast, left-wing factions (CPU, Left-Center [LC], Peasant Party [PEA], Progressive Socialist Party [PSP]) and center-right factions (Rukh-Kostenko [R-K], RukhUdovenko [R-U]) were considered to engage in more active and consistent behavior (D’Anieri, Kravchuk, and Kuzio 1999, 159–160).22 Table 7.2 shows the results of the empirical analysis we conducted on data from Ukraine’s Rada. In the first model, we evaluate the impact of seat type and our two continuous measures of seat safety. The second model adds variables for dual candidacy and incumbency. We subsequently disaggregate the effects of dual candidacy in models 3 and 4 by classifying deputies who ran in both the SMD and PR components on the basis of the safety of their list position of the SMD they contested. Model 4 also includes measures of partisan affiliation. On the whole, the results appear to conform to our theoretical expectations. As model 2 indicates, dual candidates appear to display lower levels of party loyalty than “specialists.” However, model 3 suggests that this result is driven by the voting behavior exhibited by the dual candidates whose position on their party’s PR list is low. Even dual candidates whose SMD is safe seem to keep their options open by voting consistently with their parties. By contrast, dual candidates whose position in the proportional component is safe show no greater propensity to vote with their party. What about the behavior of deputies who run in only one component of the election? None of the four models provides much evidence that electoral marginality or list position have any effect—positive or negative—on party discipline. The coefficients measuring the effects of margin of victory and the ratio between the votes received by a deputy’s list and the number of full quotas she needs to be guaranteed a seat are quite small in size and characterized by relatively high standard errors. Perhaps more importantly, when factional affiliation is not accounted for in our statistical analysis, SMD deputies seem less inclined to stick to the party line than are legislators who “specialize” in the PR component. This effect seems quite pronounced in models 1–3, where SMD representatives are as much as 10 percent less likely to vote with the majority of their parliamentary groups.

110 / mixed electoral systems Table 7.2 IRLS results from Ukraine’s Verkhovna Rada

Constant

Model 1

Model 2

Model 3

Model 4

68.63* (1.37) ⫺9.58* (2.33) ⫺0.11 (0.78) 0.97 (1.29) —

66.61* (1.62) ⫺8.76* (2.16) ⫺0.71 (0.92) 1.68 (1.36) —

84.88* (1.02) ⫺0.96 (1.16) 0.02 (0.48) ⫺1.15 (0.71) — 2.01 (1.77) 0.33 (1.68) 0.46 (1.81) ⫺0.55 (1.11) ⫺0.66 (0.85) ⫺32.05* (1.64) ⫺10.32* (2.22) ⫺25.21* (1.64) ⫺8.80* (1.72) 57.75* 428

Safe SMD/High PR

—

66.51* (1.57) ⫺9.65* (2.32) ⫺0.19 (0.78) 1.44 (1.30) 3.41* (1.59) —

Safe SMD/Low PR

—

—

Unsafe SMD/High PR

—

—

Unsafe SMD/Low PR

—

—

Incumbent

—

NDP

—

1.59 (1.66) —

⫺2.95 (3.34) 8.24* (3.24) ⫺5.67 (3.45) 5.65* (2.16) 1.27 (1.62) —

PSP

—

—

—

Rukh-Kostenko

—

—

—

Left Center

—

—

—

16.48* 428

11.51* 428

10.21* 428

Seat SMD * Ln(Margin) PR * Ln(PRVote/Quotas) Dual Candidacy

F Statistic N

Note: * p ⱕ .05. Controls for all factions are included in model 4, but most results are suppressed. The CPU was the excluded category.

Once factional affiliation is accounted for, however, the effects of seat type and dual candidacy are washed away (model 4). Party effects are as anticipated: the excluded category (CPU) shows the highest cohesion, followed by other parties of the left. The CPU has a relatively high membership barrier and a tradition of discipline among its members. In this sense, the results conform with the electoral path argument: politicians who strive to gain access to the CPU party list may have to show even greater loyalty to the party than competitors who contest seats on new party lists. The electoral path to nomination varies across parties, with greatest loyalty evidenced among legislators in parties with a more competitive membership and nomination environment. However, while the strong effects of party factions could be due to party discipline, they also could be due to

eyeing the future / 111

self-selection. Since like-minded politicians may band together in party factions, their voting behavior may not be induced by the party organization, but may naturally flow from common views on policy matters. Recent research suggests that personal and party preferences drive behavior in Ukraine’s parliament (Thames 2004). Figure 7.1 illustrates the lack of difference between the deputies elected to the Rada through SMD and PR. For both legislator types, we graph how party discipline changes with relative seat safety. For example, a value of 5 on the x-axis indicates that an SMD legislator won her district by 5 percent; the same value for a PR legislator means that her list gained 5 times the quotas she needed to be guaranteed a seat. For each level of seat safety, we simulate 1,000 sets of coefficients from our parameters’ posterior distribution, whose mean equals the estimates presented in model 4 and whose variance is the estimated variance–covariance matrix of the robust model (see King,

70

Party discipline

65

60

55

50

0

2

4

6

8

SMD Legislators: Margin of victory PR Legislators: List vote / Quotas needed SMD Legislators: 90% Confidence Int. PR Legislators: 90% Confidence Int.

Figure 7.1

Variation in discipline, Ukraine.

10

112 / mixed electoral systems

Tomz, and Wittenberg 2001). We set all variables related to dual candidacy and party affiliation equal to 0 (with the exception of the Rukh faction, assigned a value of 1)23 and estimate the model’s systematic component X␤. For each legislator type, we plotted the 90 percent confidence intervals of the estimates. Figure 7.1 indicates that SMD and PR “specialists” behave quite similarly. Rukh deputies, in particular, vote with the majority of their faction between 55 and 60 percent of the time, whatever the component of the election in which they ran. For SMD deputies, levels of party loyalty remain quite stable as their margin of victory changes from 0 to 10. Party discipline, instead, is quite high for PR deputies who were elected thanks to the largest remainder system (where list vote/quota * list position ⬍ 1, party discipline is above 60 percent). It drops to less than 60 percent for legislators whose party received twice as many votes as those necessary for them to win the seat outright, and finally stabilizes as seat safety increases further. Interestingly, for certain levels of seat safety, SMD legislators appear even more loyal to their parties than PR deputies, though such differences never rise to conventional levels of statistical significance. In any case, the results presented here indicate that seat type is a poor predictor of legislative voting. Electoral path, particularly faction membership and the degree to which parties control nominations and are able to enforce discipline, tends to overshadow mandate. Empirical Analysis: Party Discipline in Italy’s Chamber of Deputies Italy adopted mixed electoral systems for the election of both houses of its bicameral legislature in 1993. In the 630-member Chamber of Deputies, 75 percent of the members are elected through plurality races held in 475 SMDs. The remaining 155 seats are distributed through PR. Voters cast two ballots: with the first, they select an SMD candidate. Each candidate participating in a plurality race must be affiliated with a PR list and may also run in up to three different PR constituencies. With the second ballot, voters select a PR list.24 Two heterogeneous coalitions dominated the Italian party system during the period under analysis (see table 7.3).25 On the center-left, the Ulivo (Olive Tree) was led in the 2001 elections by the Democratici di Sinistra (DS, Left Democrats) and by an alliance of centrist parties called Margherita (the Daisy). The Ulivo also comprises small parties such as the Comunisti Italiani and the Girasole (the Sunflower, which is in turn made up of the Green Party, or Verdi, and the Socialisti Democratici Italiani, SDI) and avails itself of the support of Rifondazione Comunista (RC), with which it generally concludes stand-down agreements. On the center-right, the Casa delle Libertà (House of Freedoms), is centered on the alliance

eyeing the future / 113

between Forza Italia and the post-fascist Alleanza Nazionale (AN). It also includes small, socially conservative Catholic parties and the formerly separatist Northern League (Lega Nord). In their attempt to establish preelectoral coalitions that are broad and inclusive enough to be competitive in most of the country’s SMDs, large parties that seek to maximize their chances of forming a government must persuade smaller political forces to join the alliance. In fact, the defection of small parties, including those that have little hope of overcoming the 4 percent threshold in the PR tier, may cost the coalition valuable districts and ultimately the general election. Commonly, the most effective means by which parties such as DS and Forza Italia secure the participation a minor party in the alliance is the nomination of that party’s candidates to a share of safe districts that is commensurate to its anticipated electoral strength. Coalitions, basing their estimates on results from previous elections and opinion polls, assign to each of the parties of which they are composed a share of safe and winnable districts that is roughly proportional to their predicted performance in the PR tier.26 This mode of candidacy distribution leads to a phenomenon that D’Alimonte (2001) terms “the proportionalization of the plurality tier.” The “proportionalization of the plurality tier” has determined the emergence of highly centralized nomination procedures, because, as D’Alimonte (2001) notes, the proportional allocation of candidacies “requires a high degree of centralized decision-making in the allocation of districts to parties and in the selection of candidates.” It is evident that the 1993 reform did little to reduce the dominance of party organizations, which have maintained a firm grip over the nomination of both SMD and PR candidates. The process by which SMD candidates are selected in Italy remains far more centralized than that typical of SMD systems, as party leaders exercise direct control over nominations rather than leaving the choice of candidates to local branches (as in Britain) or to voters through the organization of primary elections (as in the United States). In addition, the absence of any residency requirements has allowed coalitions to frequently move legislators from one district to another. As a result, the incumbency advantage that is typical of elections held under plurality has not developed in Italy’s majoritarian races (Verzichelli 2002). Few SMD deputies, in fact, are “rooted” in a particular constituency. Only about 100 candidates competed in the same SMD in each of the three elections since the reform (Di Virgilio 2002), while only 234 incumbent SMD deputies ran, in 2001, in the same district they had won in 1996. Most SMD candidates, in contrast, are “parachuted” into a district where they have no ties with local constituents.27 Although the objectives of the reform, and of the move away from a proportional system, included reducing the power exercised by party leaders and increasing the accountability of elected officials to their constituents,

114 / mixed electoral systems

the intra-coalitional dynamics that emerged after 1993 have effectively prevented voters from exercising any influence over ballot access. Shugart and Wattenberg (2001b) maintain: “this result is not an indictment of the mixed-member system per se, as it stems from the nature of multiparty alliances that negotiate their single-seat district nominations prior to elections.” However, this explanation fails to consider that the procedure by which SMD candidacies are distributed to political parties is a direct consequence of how the Italian mixed system works. The mere existence of separate PR elections, in fact, provides small parties with the opportunity to consolidate their electoral support, strengthen their blackmail power, efficaciously extract concessions from their coalition partners by obtaining safe SMDs, and hence neutralize the potential of markedly majoritarian electoral rules to threaten their survival. While most legislators serving in the Chamber of Deputies are elected in SMD races, where they run under a coalition’s banner, the parties that make up such preelectoral alliances typically reclaim their independence after the elections and form autonomous parliamentary groups. The standing orders for the Chamber of Deputies provide that 20 representatives may form a parliamentary group (gruppo parlamentare), though organized parties that are present throughout the national territory may be allowed to constitute a group even if they have less than 20 members (the RC, for instance, was granted parliamentary group status despite the fact that it only won 11 seats). The remaining deputies join the gruppo misto (mixed group). Table 7.3 displays the composition of the seven parliamentary groups that were formed at the beginning of the fourteenth legislature, immediately after the 2001 elections.

Table 7.3 Parliamentary groups in the Italian Chamber of Deputies, June 2001

Alleanza Nazionale (AN) Democratici di Sinistra (DS) Forza Italia (FI) Lega Nord Margherita Rifondazione Comunista (RC) Unione Democratico-Cristiana e di Centro (UDC) Gruppo Misto

Total Deputies

SMD Deputies

PR Deputies

99a 136 178b 30 83c 11 40

73 104 132 30 56 0 40

26 32 46 0 27 11 0

41

38

3

a Deputies Beccalossi and Lo Porto resigned on July 11, 2001 and July 31, 2001, respectively; they were replaced by deputies Garnero Santanché (7/12/2001) and Cannella (8/1/2001). b Deputy Colletti died on 11/3/2001. c Deputy Rocchi (formerly a member of the gruppo Misto) joined the group on 7/18/2001.

eyeing the future / 115

Our evaluation of the incentives that flow from parties’ nomination procedures and by the newly introduced electoral rules suggests that legislators who are elected through SMD and PR should behave similarly. PR deputies, who are elected exclusively by virtue of their position on a party list, should abide by the dictates of their party leaders if they wish to maintain their position or gain higher list placement in future elections. SMD legislators should maximize their prospects of reelection by attempting, through reliably disciplined behavior, to induce party leaders to nominate them to a safer district. As a result, consistent with the observed behavior of legislators elected under pure plurality in polities where nominations are centralized, the voting behavior of SMD representatives in the chamber should be characterized by strong party loyalty. We replicated the analysis presented earlier with data from the Italian Chamber of Deputies. The results are displayed in table 7.4. Models 1–4 reveal that seat type exerts unusual effects on party discipline. As predicted, in none of the Models do PR deputies display higher levels of party loyalty than their colleagues elected in SMD. Quite the contrary, SMD deputies appear to be systematically more responsive to their party leaders, even when controls for relative seat safety, dual candidacy, incumbency, and partisan affiliation are included. Also, the results shown in table 7.4 paint a mixed picture about the relationship between the various indicators of seat safety and party discipline. On the one hand, dual candidates and incumbents are substantially less likely to behave consistently with their parliamentary group. Also, when controls for faction affiliation are considered, PR deputies whose list position allowed them to be easily elected tend to deviate from the party line more frequently than those whose low list position renders their seats more precarious. On the other hand, SMD deputies seem to be unaffected by the margin by which they won their respective districts. In spite of these inconsistencies, however, one can conclude that legislators occupying safe seats are less likely to behave in conformity to their parliamentary group than deputies for whom seat safety may be presumed to be a stronger concern. Figure 7.2, which provides a clearer substantive interpretation of the coefficients shown in model 4, was generated following the logic outlined in King, Tomz, and Wittenberg (2000) and the procedures described in the previous section. As in the Ukrainian case, we plot our estimates of how party loyalty changes with the relative seat safety of both SMD and PR deputies. The patterns revealed here are quite similar—if considerably more marked—to those uncovered in the Rada. SMD deputies nominated by Forza Italia, whatever their margin of victory, vote with their parliamentary groups about 90 percent of the time; a surprising result, particularly given the expectation that mixed systems should bring about an ideal blend of

116 / mixed electoral systems Table 7.4 IRLS results from the Italian Chamber of Deputies

Constant Seat (SMD ⫽ 1; PR ⫽ 0) SMD * Ln(Margin)a PR * Ln(PRVote / Quotas)

Model 1

Model 2

Model 3

Model 4

77.59* (1.34) 5.99* (2.05) ⫺0.80 (0.67) ⫺3.48 (2.17)

82.92* (1.49) 3.88 (2.06) ⫺0.60 (0.68) ⫺3.31 (2.19) ⫺18.17* (1.90)

83.19* (1.52) 3.07 (2.05) ⫺0.18 (0.69) ⫺1.50 (2.23)

72.86* (4.09) 2.74 (1.73) ⫺0.44 (0.58) ⫺6.06* (1.99)

⫺6.65* (1.35)

⫺39.30* (5.51) ⫺31.59* (4.10) ⫺38.15* (3.33) ⫺8.01* (2.64) ⫺6.77* (1.33)

31.13* 580

38.77 580

⫺26.72* (4.58) ⫺18.85* (3.42) ⫺52.44* (2.77) ⫺7.54* (2.20) ⫺4.47* (1.12) 6.72 (4.22) 10.08* (4.22) 16.87* (4.24) 15.50* (4.77) 3.59 (4.26) 3.88 (4.66) 44.70 580

Dual Candidacy Safe SMD / High PR Safe SMD / Low PR Unsafe SMD / High PR Unsafe SMD / Low PR Incumbent AN DS Forza Italia Lega Nord Margherita UDC F Statistic N

3.51* 580

Note: *p ⬍ .05; Standard Errors in parentheses. In model 4, dual candidates are assigned a value of 0 for the variables Ln(Margin) and Ln(PRVote/Quotas).

national and local-oriented behavior. In contrast, PR deputies display high levels of party loyalty (about 95 percent) when they are elected thanks to the largest remainder rules, but their levels of cohesion drop sharply—to about 75 percent—for deputies whose party received a number of effective votes that equals more than twice the Hare quotas they needed to qualify for a full seat. Interestingly, for most levels of seat safety, PR deputies are less, not more, loyal to their parliamentary groups. Models 3 and 4 evaluate the impact of seat safety on the voting behavior of different types of dual candidates. Each of the four dichotomous

eyeing the future / 117 100

Party discipline

90

80

70

60

1

2

3

4

SMD Legislators: Margin of victory PR Legislators: List vote / Quotas needed SMD Legislators: 90% Confidence Int. PR Legislators: 90% Confidence Int.

Figure 7.2

Variation in discipline, Italy.

variables are statistically significant and negative in sign, suggesting that dual candidates in general are less likely to stick to the party line than are “specialists.” This is not surprising given the features of Italy’s system. In Italy, dual candidates are generally party leaders: the placement of a candidate with universal name recognition at the top of a PR list serves the purpose of maximizing PR votes as well as guaranteeing the leadership parliamentary seats. Once elected, party leaders are most likely to participate in government and to be occupied by the many activities inherent to leading a political party. Therefore, they are also more likely to miss parliamentary votes, particularly if the outcome of the vote is either not in doubt or is not of much relevance to them. It is also possible that by virtue of the safety of their seats and of their status within the party, they may exercise wider discretion when they cast their votes. However, the differences in the magnitude of the coefficients reveal that dual candidates of different types may face different incentives. In particular, those who ran in unsafe SMDs and

118 / mixed electoral systems

were ranked low on their party list are significantly less likely to deviate from their group’s modal behavior than those who were safely elected in either the majoritarian or the proportional components of the election (or both). It should be noted that the differences found between SMD and PR legislators, between specialists and dual candidates, and between candidates occupying safe and not-so-safe seats are entirely due to different patterns of attendance that characterize different classes of deputies in the Chamber of Deputies. The impact exerted by mode of election and relative seat safety on party loyalty disappears when party discipline is measured as a percentage of the votes that a legislator actually cast and when, therefore, nonvoting is excluded from consideration. The comparison of the results obtained by testing the effects of mandate and electoral path on different measures of party discipline suggests that no legislator is particularly inclined to vote differently from his/her parliamentary group. Rather, legislators occupying seats that are more precarious may have a stronger incentive to demonstrate their loyalty to party leaders through more assiduous participation in legislative sessions. On average, PR deputies and representatives occupying safe seats are less likely to participate in parliamentary sessions. Whether or not nonvoting amounts to “voting with one’s feet” (and we suspect that it does in many cases), the fact remains that deputies occupying relatively less safe seats display systematically lower rates of absenteeism. This is consistent with the greater pressures that are placed upon these deputies to demonstrate their loyalty to their party organizations. Conclusion Our empirical analysis of legislative voting in the Verkhovna Rada and the Chamber of Deputies supports the notion that mixed systems do not simply “split the difference” (Bawn and Thies 2003) between SMD and PR. The incentive structure faced by Ukrainian and Italian legislators is not exclusively affected by the component of the election in which they “specialize,” but depends upon their coexistence with deputies elected under different rules, the prospect of competing in both SMD and PR, and the nomination strategies that parties employ to maximize their vote share in both segments of the election. “Electoral path” appears to matter to party unity more than simple mandate. In Ukraine, it is partisan affiliation that most influences party discipline. In Italy, though deputies are elected under different rules, have profoundly diverse ideological views, and belong to parties with different PR–SMD splits, they all share the incentive to vote consistently with the instructions issued by their party secretaries. Centralized nominations minimize local accountability, blur the difference

eyeing the future / 119

between the behavior of SMD and PR legislators, and strengthen party discipline by providing legislators with electoral incentives to comply with the dicta of their party bosses and the freedom to ignore their constituents. Members of parliament differ in many ways, yet they share at least one common interest: the desire to be reelected. Although the factors influencing voting behavior may be multifaceted, ideological, or policy-based, motivations are irrelevant if legislators lose their seats. Legislators may not be guided solely by their desire of being reelected, but electoral considerations are central concerns. As our analysis has begun to show, mixed systems provide legislators with electoral incentives that are more complex than those we would expect by simply applying the conventional wisdom about proportional and majoritarian systems. The influence of contamination is visible well beyond the campaign period. The interaction of the SMD and PR components of the election is highly consequential for coalitional dynamics and the competition for ballot access, which in turn affect patterns of representation and legislative behavior.

This page intentionally left blank

C h ap t e r 8 Better, Worse, or Just Different? Legislative Fractionalization in Mixed Systems

Since the adoption of mixed systems in many of the new democracies that emerged from the rubble of the Berlin Wall and in a number of countries with established democratic records, many scholars have theorized about the consequences that mixed systems exert on the configuration of national party systems. The normative judgments made by practitioners and academics alike about the “goodness” or “badness” of mixed electoral institutions have mostly been based on predictions about the kinds of party systems that these election rules should engender. Few, however, have investigated this issue cross-nationally. If mixed systems are to be judged based on their representativeness and their ability to produce cabinet stability, coherent and identifiable governing majorities, and a balance between local and national interests (Shugart and Wattenberg 2001b), no such evaluation can be complete without a thorough assessment of the kinds of legislatures they spawn. As this volume suggests, there is considerable disagreement about how mixed systems should be studied and where they fit in the established majoritarian/proportional dichotomy. Do mixed systems produce intermediate outcomes, depending on how much of the proportional and majoritarian ingredients are thrown into the mix? Alternatively, do they produce outcomes that are distinct from those observed in “pure” SMD and PR systems? This chapter assesses each of these questions empirically by studying the variation in legislative fractionalization observed in over 130 free or quasi-free countries after 300 elections held between 1990 and 2001. We build on earlier work (Nishikawa and Herron 2004) to assess which of the existing taxonomies of mixed electoral systems best explains the variation in the effective number of legislative parties. In addition, we compare the outcomes produced by mixed systems to those observed under “pure” majoritarian and proportional rules and explore how legislative fragmentation

122 / mixed electoral systems

is affected by the great institutional variation that characterizes this new family of electoral systems. Again, the empirical results challenge the view that mixed systems simply synthesize the independent effects of SMD and PR. Better, Worse, or Just Different? A substantial amount of research has investigated the effects exerted of electoral institutions on legislative fractionalization. Much of this literature has uncovered important differences between proportional and majoritarian systems. In the former, small district magnitudes tend to reduce the number of parties; in the latter, more permissive rules generally usher in greater fragmentation. Such differences may be mitigated or exacerbated by sociological factors, such as ethno-linguistic cleavages, and institutional variables such as the proximity of presidential elections. As for mixed electoral systems, the evidence is considerably more scant. Few scholarly works on mixed systems have emulated the cross-national analysis that Amorim-Neto and Cox (1997), Cox (1997), Ordeshook and Shvetsova (1994), Benoit (2002), Mozaffar, Scarritt, and Galaich (2003), and others have conducted on the determinants of national party systems under SMD and PR.1 Studies evaluating the consequences of mixed systems on nationwide patterns of representation are for the most part single-country and descriptive in nature. While the dynamics that led to the adoption of mixed electoral systems in the dozens of countries that have adopted them clearly differed, mixed systems were invariably heralded by their proponents as election rules that would combine the most beneficial consequences of both majoritarian and proportional systems while tempering their well-known shortcomings. Similar optimism also characterizes prominent scholarly work on the subject. Shugart and Wattenberg (2001b) state that the mixture of majoritarian and proportional institutions offers “the best of both worlds,” in part because of its anticipated moderating effects on national party systems. Mixed systems, as noted in chapter 2, are a diverse bunch, and many consider their greatest strength to be the seemingly limitless variety of SMD–PR combinations they may encompass. Institutional designers may strike different balances between majoritarian and proportional principles, tailoring mixed systems to fit their country’s unique political context. Polities suffering from excessive legislative fractionalization may adopt a mixed system to decrease the number of parties without sacrificing adequate representation of minority interests. Similarly, countries frequently governed by manufactured majorities may introduce a mixed system to enhance their representativeness without renouncing cabinet stability, local accountability, and a tradition of bipolarism.

better, worse, or just different? / 123

The key claim advanced by the proponents of the “best of both worlds” approach is that mixed systems ideally balance majoritarian and proportional institutions, moderating some of their pathological effects and combining their most desirable consequences. At the inter-party level, mixed systems are expected to simultaneously encourage two-party competition, or at a minimum the formation of two blocs cemented by the imperative to win as many SMD races as possible, and provide an acceptable level of representation to small parties that will primarily focus their resources on the proportional segment of the election. The result should be representative but not terribly fragmented legislatures where most parties also coalesce to form two clearly identifiable potential governing coalitions that are cohesive enough to rule in conditions of stability. The quantities of the majoritarian or proportional ingredients that are blended into a mixed system should determine the types of outcomes that such system brings about. Shugart and Wattenberg (2001b) intuitively argue that mixed systems should produce “intermediate” outcomes that lie somewhere in between the majoritarian and proportional extremes. Where mixed systems are more majoritarian (because of the absence of compensation, the distribution of most legislative seats via SMD races, the presence of high PR thresholds, and regional rather than national allocation of proportional seats), we should observe outcomes that more closely approximate those of majoritarian systems. More representative mixed systems, in contrast, should resemble proportional systems to a greater degree. MMM systems where the two components have roughly equal relevance, instead, should most effectively reach equilibrium between the priorities constituted by adequate representation and coherent, stable governance. It is the ability of mixed systems to ideally synthesize the effects of SMD and PR that creates the “best of both worlds” and has led scholars and political actors to advocate their adoption in countries like Britain and Canada, where major electoral reform has been discussed in recent years. The crucial empirical implication of Shugart and Wattenberg’s (2001b) reasoning is that we should expect a monotonic, quasi-linear, negative relationship between the share of seats allocated in SMD and the number of legislative parties. MMP systems and mixed systems where most legislators are chosen through proportional representation should be characterized by high party system fragmentation. Many small parties, encouraged by the prospects of a highly proportional outcome, are likely to contest the election on their own; at the same time, the electoral systems will fail to meaningfully reduce the number of parties. As the share of seats allocated in SMD increases, however, the number of parties should decrease. Even if a more majoritarian mixed system failed to discourage the entry of small parties, a larger SMD component in all likelihood would introduce greater disproportionality,

124 / mixed electoral systems

reward parties that win SMD races beyond the share of votes they have garnered, penalize small parties, and hence reduce the overall fractionalization of the legislature. Generally speaking, the more markedly majoritarian a mixed system is, the stronger should be the tendency toward two-party competition. Of the few existing studies that have attempted a cross-national assessment of the consequences that mixed systems have on the configuration of national legislative assemblies, two support the propositions advanced by Shugart and Wattenberg (2001b). Kostadinova (2002), who investigates the national party systems in postcommunist Eastern Europe, finds that mixed systems are distinguished by a number of legislative parties that are comprised between the numbers observed in countries that have adopted SMD and PR. Perhaps more interestingly, the number of parties seems to increase with the share of seats set aside for the proportional segment of the election. More recently, Moser and Scheiner (2004) have presented similar results. The “best of both worlds” conjecture is not universally accepted. Monroe (2003, 443) cautions that “the two components of mixed systems interact to produce complex strategic incentives to many players, and there is no reason to expect that the results are in all cases and on all dimensions something intermediate between the pure types.” Also, Bawn and Thies (2003) note that Shugart and Wattenberg (2001b) have failed to fully articulate why mixed systems should produce the best, and not the worst, of the majoritarian and proportional worlds. Bawn and Thies (2003) conclude that whether mixed electoral rules bring out the most beneficial or the most pernicious features of SMD and PR largely depends upon how political actors adapt to the introduction of hybrid rules. Perhaps more important, their results demonstrate—consistent with the theme that recurs throughout this volume—that any analysis of the incentive structure to which mixed electoral systems give rise must explicitly and systematically consider the implications of mixing proportional and majoritarian institutions, rather than focusing on each component separately and taking for granted that mixed systems will simply synthesize their consequences. The argument that emphasizes the interactive features of mixed electoral systems has important implications for the analysis of legislative fractionalization. Chapters 3–6 have shown that the dynamic leading majoritarian rules to reduce the number of parties partially breaks down in mixed electoral systems. The prospect of receiving a vote boost in the PR component by running their own SMD candidates encourages parties to “go it alone” when possible and profitable and discourages the formation of large alliances. In principle, this should have no impact on legislative fractionalization. More majoritarian mixed systems still introduce greater disproportionality: regardless of how many candidates run for an SMD seat, and however widely SMD votes are dispersed, highly majoritarian systems should still

better, worse, or just different? / 125

reduce the number of parties. Nonetheless, the research on contamination suggests that parties may not necessarily coalesce in two blocs, as foreshadowed by Shugart and Wattenberg (2001b). Sartori (1994, 74–75) makes a similar point as he dismisses the “best of both worlds” argument. He maintains that, rather than combining the best features of SMD and PR, mixed systems are a “bastard-producing hybrid which combines their defects.” The SMD component, in fact, is likely to encourage “shots in the dark,” while the PR tier, regardless of how majoritarian the electoral system is, maintains the potential to produce high legislative fragmentation. The kind of party system that is likely to develop, therefore, depends upon the structure of preexisting party competition. Countries that move from a majoritarian to a mixed system will in all likelihood not suffer from excessively fractionalized legislatures, “for the habit of two party voting will not be lost overnight.” In countries that have either inchoate or highly fragmented party systems, however, even the introduction of a highly majoritarian mixed system “will maintain the fragmentation at damaging levels.” Anecdotal evidence suggests that the interactive qualities of mixed electoral systems may prevent even markedly majoritarian rules from appreciably reducing legislative fractionalization. The case of Italy provides a simple illustration of this result. In Italy, political parties adapted to the introduction of MMM rules where 75 percent of the seats are distributed in the SMD component by forming heterogeneous coalitions that jointly nominate and support SMD candidates. But, large parties can only secure the participation of a small party in the alliance by nominating that party’s candidates to a share of safe and winnable districts that is commensurate to its anticipated electoral strength. The “proportionalization of the plurality tier” (D’Alimonte 2001)—that we described in greater depth in chapter 7—effectively doomed the hopes that the new MMM system would reduce the fractionalization of the legislature. Parties with no hope of clearing the 4 percent threshold in the proportional tier generally manage to win a considerable number of SMDs. The 4 percent PR threshold, in fact, has effectively excluded small parties from the Chamber of Deputies. Six parties overcame the threshold in 1994, eight parties won PR seats in 1996 and only five managed to do so in 2001. In 2001, however, as many as ten parties won SMD seats. Nonviable parties like the Lega Nord and the CCD-CDU, which received less than 8 percent of the proportional vote combined, won an astonishing 70 out of a total of 475 SMDs. This unexpected outcome2 may well be a reflection of the idiosyncrasies that distinguish Italian politics. However, this outcome is also consistent with the manner in which the Italian mixed system works. Other things being equal, the more majoritarian a mixed system is, the higher is

126 / mixed electoral systems

the elasticity of the relationship between seats and votes (see Rogowski and Kayser 2002). Because even small vote swings may determine significant shifts in overall seat distribution, large parties should be expected, in the presence of highly majoritarian rules, to aggressively seek to build coalitions that are broad and inclusive enough to be competitive in SMDs nationwide. In chapter 4, we demonstrated that more preelectoral coordination is observed in the majoritarian component when the outcome of the election most hinges on the results of the SMD races. As the share of seats allocated in SMD increases, parties become more likely to form coalitions, back one another’s candidates, coordinate district nominations to prevent head-to-head competition, or simply sit out the race in some districts. As the number of seats allocated in SMD increases, and as the outcome of majoritarian races grows more and more important to the overall election result, larger parties may be inclined to offer substantial concessions to attract the support of small but potentially decisive parties. One way “coalition-makers” may lure small parties into the alliance is to offer the coalition’s backing, whether tacit or explicit, for some of those parties’ SMD candidates. Past a certain threshold, therefore, the number of parties may paradoxically increase as the system becomes more majoritarian. The necessity to form larger, more inclusive coalitions may result in increasing the bargaining leverage, and seat share, of small parties. While plurality systems over time tend to weed out small parties perennially excluded from a legislature (see Rosenstone, Behr, and Lazarus 1996), in mixed systems, the presence of separate PR elections allows smaller parties to win representation, cultivate their base, consolidate their electoral support, receive public financing, and establish some “blackmail power” they can use to extract preelectoral or postelectoral concessions from prospective coalition partners. In previously fractionalized party systems, the result may well be continued fragmentation. Data and Variables This chapter’s empirical analysis addresses differences between PR, SMD, and mixed systems, asking how institutional factors and alternative classifications predict levels of legislative fractionalization. Table 8.1 displays, for all of the mixed systems considered in this chapter, some of their most salient institutional features. The analysis begins with an extension of the results presented in Nishikawa and Herron (2004), who analyze the data provided in the World Bank Database of Political Institutions (DPI, see Beck et al. 2001). Their time-series cross-section data include 90 elections held between 1990 and 2001 in 53 countries classified by Freedom House as “free.” We modify the models and evaluate them on an expanded data set, which includes more

better, worse, or just different? / 127 Table 8.1 Institutional variation in the world’s mixed systems Country 1

Ballots 2

Formula 3

Albania Andorra Armenia Bolivia Bulgaria Croatia Ecuador Georgia Germany Guatemala Hungary Israel Italy Japan Korea Kyrgyzstan Lithuania Macedonia Madagascar Mexico New Zealand Niger Panama Philippines Russia Senegal Taiwan Ukraine Venezuela

2 2 2 2 2 2 2 2 2 2 2 2 2 2 1 2 2 2 Coex. 1 2 Coex. Coex. 2 2 1 1 2 2

Maj. FPTP FPTP FPTP Maj. FPTP FPTP Maj. FPTP FPTP Maj. FPTP FPTP FPTP FPTP Maj. FPTP Maj. FPTP FPTP FPTP FPTP FPTP FPTP FPTP Block SNTV FPTP FPTP

Compensation Small party seats None None Seat Linkage None None None None Seat Linkage None Vote Transfers None Vote Transfers None None None None None None 8% Over-Repr. Seat Linkage None None Max. 3 PR seats None None None None Seat Linkage

M in PR 4

Threshold

40 14 56 14 7 80 12 14 656 22 58 120 155 16 46 15 70 34 2 200 120 8 3 52 225 70 41 225 8

4 None 5 3 4 5 0 7 5 5 5 1.5 4 None 3 5 5 5 None 2 5 None None 2 5 0 None 4 None

% SMD 74 50 58 52 50 37 10 36 50 80 46 1 seat 75 63 85 75 50 71 71 60 56 10 37 80 50 50 70 50 47

1

Entries refer to the last lower house election considered. Bulgaria, Croatia, Kyrgyzstan, and Macedonia since abandoned their mixed systems.Chad, Lesotho, and Thailand introduced their mixed systems after 2000. Azerbaijan, Cameroon, and Tunisia were excluded because they are nondemocratic. 2 In Madagascar, Niger, and Panama, SMD and PR constituencies do not overlap. 3 FPTP indicates “First-Past-the-Post;” SNTV denotes the “Single Non-Transferable Vote.” 4 Average district magnitude of the PR component.

elections and observations for each country judged by Freedom House to be “partly free” at the time elections were held. The resulting time-series cross section includes 275 elections held in 123 countries, 26 of which have employed a mixed system during the 1990s. We also use Golder’s (2005) recently released data set Democratic Electoral Systems Around the World, which includes, for the time period under consideration, 114 countries, 303 elections, and 26 mixed systems.3 How to best analyze time-series cross-sectional data is the subject of a lively debate in the political science literature (see Beck and Katz 1995 and 1996; Bartels 1996; Zorn 2001).4 In this chapter, we follow the suggestions

128 / mixed electoral systems

advanced in Goodrich (2004), who proposes the use of a “Consecutively Parsed Estimator” (CPE) for the analysis of two-dimensional data with continuous dependent variables.5 The CPE, which simply involves the separate estimation of Within Effects and Between Effects models, allows us to make full use of the data’s within- and across-unit variation and is shown to produce better, more reasonable standard errors than other possible solutions. In this case, however, given the “sticky” nature of electoral institutions and the rarity with which drastic electoral reforms are undertaken, the independent variables of interest are mostly fixed within the sample countries. Goodrich (2004) suggests that the Between Effects estimator is optimal for data where the explanatory variables suffer from insufficient within-unit variation.6 ●

●

●

●

●

ENPS. Our key dependent variable is the Effective Number of Legislative Parties (ENPS, see Laakso and Taagepera 1979).7 As Sartori (1976, 120) explains, the number of parties that are able to win legislative representation is important because it “immediately indicates, albeit roughly, an important feature of the political system: the extent to which power is fragmented or non-fragmented, dispersed or concentrated.” PR. The dichotomous variable PR is coded 1 if a country employed proportional representation, the single transferable vote (as Ireland and Malta), or the single non-transferable vote (as Vanuatu and Jordan), which research has shown to produce proportional results akin to the d’Hondt formula (see Cox 1991; 1996). Ethnic Fractionalization. We employ Fearon’s (2003) measure of Ethnic Fractionalization, an index constructed with local and general sources that ranges from 0 (no fractionalization) to 1.8 Postcommunist. We included a dichotomous variable coded 1 if a country recently completed a transition from communist rule to account for the fact that some postcommunist countries still have inchoate party systems and may produce more fractionalized legislatures than established democracies employing similar electoral rules. Proximity. We control for the Proximity of legislative and presidential elections. This variable is coded 1 for concurrent elections and 0 for elections taking place in parliamentary systems or at the presidential midterm. High proximity should reduce the number of legislative parties.9 Analysis and Discussion

The empirical analysis proceeds in two steps. We first assess the differences between “pure” and mixed systems with regard to legislative fractionalization

better, worse, or just different? / 129

and evaluate how well alternative typologies of mixed electoral institutions explain the observed differences. Then, we provide a fuller account of the consequences that mixed systems have on legislative fragmentation by considering other aspects of their institutional variation. We focus primarily on the share of seats allocated in the majoritarian and proportional components. In table 8.2, we present the results of statistical tests conducted on both the Database of Political Institutions and Golder’s (2005) Democratic Electoral Systems data. Models 1–3 evaluate the effects of different types of mixed electoral systems, employing the typologies of Massicotte and Blais (1999), Shugart and Wattenberg (2001a), and Reynolds and Reilly (1997).10 We coded mixed systems that were not explicitly mentioned by the three typologies according to the definitions that the authors provided. For instance, Madagascar, classified as Parallel by Reynolds and Reilly (1997), was placed in Massicotte and Blais’s (1999) Coexistence category but was not coded as a mixed system when using Shugart and Wattenberg’s (2001a) typology, which excludes systems with nonoverlapping constituencies. Similarly, Guatemala and Kyrgyzstan, which are not considered by Massicotte and Blais (1999) and Shugart and Wattenberg (2001a), were coded as Superposition and MMM, respectively. Andorra, Ecuador, and Senegal, which are not mentioned by Shugart and Wattenberg (2001a), are also considered MMM. The tests confirm some of the conventional wisdom about the effects of sociological and institutional variables on legislative party systems. On average, PR produces between 1 and 1.5 more parties than majoritarian systems, which constitute the excluded category in each of the six models. In addition, ethnically fractionalized countries generally have more parties than relatively homogenous societies, though this effect appears to be substantively small and, particularly when we consider the DPI data, is characterized by wide confidence intervals. Finally, the proximity of presidential elections appears to have a small reductive impact on the number of parties. The high uncertainty around its point estimates depends primarily upon the fact that the creation of a “meaned” data set for the purposes of estimating the Between Effects models dilutes the effects of this variable, which are more marked in analyses that pool the data. Turning our attention to mixed electoral systems, systems with linked tiers—those labeled Corrective and MMP—produce levels of legislative fractionalization that are close to, and sometimes higher than, proportional systems. Conversely, systems with no or weak formal linkages, classified as Superposition, MMM, and Parallel, are distinguished by a number of parties that more closely approximates majoritarian systems. As for the different classifications, the estimates suggest that the three typologies are essentially equivalent. Compensatory systems, however defined, generate

130 / mixed electoral systems Table 8.2 Mixed systems: Evaluating alternative taxonomies: robust MM estimation Dependent variable Effective number of legislative parties Model 1

Constant Ethnic fractionalization Proximity Post-Communist PR Massicotte and Blais (1999) Superposition Correction Coexistence Shugart and Wattenberg (2001) MMM MMP Reynolds and Reilly (1997) Parallel MMP N

Model 2

Model 3

DPI

DES

DPI

DES

DPI

DES

2.33* (.32) .34 (.52) ⫺.48 (.32) .70 (.38) .96* (.28)

2.02* (.36) 1.32* (.58) ⫺.69* (.35) .28 (.40) 1.46* (.33)

2.31* (.32) .50 (.53) ⫺.43 (.33) .68 (.39) .94* (.28)

2.09* (.38) 1.55* (.61) ⫺.59 (.37) .33 (.41) 1.31* (.34)

2.28* (.32) .45 (.52) ⫺.45 (.33) .64 (.38) .99* (.28)

1.97* (.36) 1.48* (.58) ⫺.69* (.35) .17 (.39) 1.50* (.33)

.41 (.47) 1.05* (.51) 1.52 (.82)

.83 (.51) 1.63* (.55) 1.81* (.85)

––

––

––

––

––

––

––

––

––

––

––

––

–– –– ––

––

.53 (.41) 1.50* (.71)

.83 (.48) 1.66* (.76)

––

––

––

––

––

––

––

––

––

––

––

––

.56 (.42) 1.60* (.56)

1.07* (.45) 1.91* (.59)

123

106

123

106

123

106

––

Notes: *p ⬍ .05; Standard errors in parentheses. DPI: World Bank Database of Political Institutions. DES: Democratic Electoral Systems Around the World.

between 0.5 and 1 more parties, on average, than noncompensatory or partially compensatory mixed systems.11 We could speculate (by looking at the DPI estimates) that the classification proposed by Massicotte and Blais (1999), which describes systems with any form of compensation as Correction, perhaps excessively blurs the differences between different types of mixed systems. At the same time, the inclusion of

better, worse, or just different? / 131

systems with nonoverlapping constituencies in Reynolds and Reilly’s (1997) Parallel category may slightly exaggerate the differences that exist between mixed and majoritarian electoral systems, as suggested by the relatively large and statistically significant coefficient for Parallel mixed systems in the analysis of Golder’s (2005) data. Nonetheless, the results are essentially a wash in this respect. The main inference suggested by the empirical results presented in this section is that mixed systems produce outcomes that fall “somewhere in between” SMD and PR. The number of parties observed under mixed electoral rules is generally larger than that found in majoritarian systems and smaller than that generated by broadly defined proportional systems. The high uncertainty around the point estimates, however, prevents us from establishing conclusively that mixed systems engender party systems that are clearly distinct from those found under majoritarian or proportional institutions. MMP or Corrective systems may be characterized by a higher number of parties than “pure” SMD, but the differences between compensatory mixed systems and PR never rise to conventional levels of statistical significance. In this sense, the results support Sartori (1994) and Bawn (1993), who consider MMP systems to be subspecies of PR. Conversely, Superposition, MMM, and Parallel systems are not meaningfully different from either PR or SMD. While the large confidence intervals may be mostly a function of the small number of observations, the fact remains that the results do not warrant the inference that mixed systems produce different outcomes from pure types of electoral institutions. The data support one of Shugart and Wattenberg’s (2001b) predictions: mixed systems do, in fact, appear to produce intermediate outcomes. When mixed systems produce fully proportional results, they tend to be just as, or more, representative than proportional systems. When their two components are more balanced, mixed systems seem to equilibrate SMD’s tendency to under-represent small parties and PR’s propensity to create hyper-representative, highly fractionalized legislatures more effectively. Where the proportional component is dominant, fragmentation is higher; where the mix has a stronger majoritarian flavor, it generally produces more majoritarian outcomes. Nevertheless, the fact that such outcomes rarely differ—in any substantive way—from those found under majoritarian and proportional rules contradicts the claim that mixed systems are a panacea for the many ills of “extreme” electoral systems. The results shown earlier lend some credence to Shugart and Wattenberg’s (2001b) assertion that mixed systems produce outcomes that lie somewhere in between majoritarian and proportional systems. The models presented in table 8.2, however, only consider one aspect of the high institutional variation that characterizes the world’s mixed systems. In this

132 / mixed electoral systems

section, we examine the consequences of other sources of such variation. Specifically, we concentrate on seat linkages, the average district magnitude of the PR component, and especially the share of seats allocated in each segment of the election. Table 8.3 presents the results of three Between Effects models estimated on the DPI and DES data. Essentially, the three models constitute an assessment of three distinct views of how the majoritarian character of a mixed system affects the configuration of a country’s party system. Models 1–2 evaluate Shugart and Wattenberg’s (2001b) and Kostadinova’s (2002) Table 8.3 How majoritarian is your mixed system (and does it matter)? Robust MM estimation Dependent variable Effective number of legislative parties Model 1 DPI Constant Ethnic fractionalization Proximity Post-Communist PR ⫻ Ln(Avg. M) MMM ⫻ Ln(PR M) MMP Coexistence MMM ⫻ Share SMD MMM ⫻ Ln(Share SMD ⫹ 1) MMM ⫻ Share SMD2 N

2.53* (.29) ⫺.07 (.51) ⫺.33 (.31) .49 (.37) .32* (.09) .53* (.18) 1.45* (.67) 1.49 (.79) ⫺2.50* (1.27) — — 123

Model 2 DES

DPI

2.36* 2.55* (.33) (.20) .91 ⫺.09 (.59) (.51) ⫺.56 ⫺.32 (.35) (.31) .26 .53 (.40) (.37) .47* .31* (.11) (.09) .47 .58* (.25) (.19) 1.64* 1.44* (.73) (.66) 1.67 1.48 (.86) (.79) ⫺.48 — (1.54) — ⫺3.76* (1.76) — — 106

Notes: *p ⬍ .05; Standard errors in parentheses. DPI: World Bank Database of Political Institutions. DES: Democratic Electoral Systems Around the World.

123

Model 3 DES

DPI

2.37* 2.60* (.28) (.29) .91 .02 (.59) (.52) ⫺.56 ⫺.36 (.36) (.32) .25 .92* (.40) (.42) .47* .25* (.11) (.10) .30 .81* (.37) (.22) 1.63* 1.33* (.73) (.67) 1.67 1.38 (.86) (.80) — ⫺12.11* (4.65) ⫺.85 — (2.18) — 11.37* (5.29) 106

123

DES 2.36* (.37) .97 (.66) ⫺.55 (.40) .33 (.46) .46* (.13) 1.31* (.37) 1.59* (.81) 1.62 (.95) ⫺9.72 (6.58) — 5.77 (7.07) 106

better, worse, or just different? / 133

expectations: in systems without seat linkages, legislative fractionalization should decrease as either a linear or logarithmic function of the share of total seats reserved for majoritarian races. These effects are tested through the use of interactive terms multiplying the dummy variable MMM by the proportion of seats allocated in the SMD tier (Share SMD), and by the natural logarithm of Share SMD increased by 1. In model 3, consistent with the reasoning introduced in the discussion of the theoretical expectations, we assess whether the relationship between Share SMD and the number of parties may be best captured by a quadratic, “U-shaped” function, whereby legislatures elected through both highly proportional and highly majoritarian MMM systems would be distinguished by relatively high levels of fragmentation while mixed electoral rules that balance the two components more evenly would produce a smaller number of parties. To generate more useful results, we improve the somewhat rudimentary model specifications presented in the previous section by including an interactive term multiplying the dichotomous variable PR by the natural logarithm of Average District Magnitude, which is considered by much of the literature to be the most crucial determinant of the number of legislative parties (Cox 1997; Taagepera and Shugart 1989). For the same reason, we also evaluate how the district magnitude of a MMM system’s proportional component affects the number of legislative parties. Other things being equal, higher PR district magnitudes should encourage the proliferation of parties, but should not do so in a linear fashion. Rather, the marginal effect of district magnitude should gradually decrease and eventually taper off.12 The results in table 8.3 support some of the observations made in the previous section. Proportional systems are characterized by a greater number of parties than majoritarian systems; such difference appears to increase as a logarithmic function of average district magnitude. In addition, the signs of the coefficients on the proximity of parliamentary and presidential elections, ethnic fractionalization, and postcommunist countries are mostly consistent with expectations, though in all cases the estimates are characterized by too great a level of uncertainty to reject the null hypothesis of no effect on the number of legislative parties. As for mixed electoral systems, some of the propositions advanced in the literature are buttressed by the empirical analysis. MMP systems are, on average, found to generate about 1.5 more parties than majoritarian systems. MMM systems, moreover, evidence a positive relationship between the number of parties and the average district magnitude of the proportional side of the election. The effects of the share of legislative seats that mixed systems assign in their majoritarian and proportional components are less conclusive. On one hand, Golder’s (2005) data offers no evidence that the effective number of parties is affected by changes in the proportion of seats distributed in the

134 / mixed electoral systems

SMD tier. Certainly, these results are inconsistent with the view that legislative fractionalization decreases monotonically as a linear or logarithmic function of the majoritarian character of a mixed system. The coefficient estimated for the interactive term MMM ⫻ Share SMD (model 1), for instance, is small: the effect of increasing the proportion of SMD seats from, say, 0.25 to 0.75 is less than one quarter of an effective party. Moreover, the uncertainty around each of the estimates is high: standard errors are about as large as the estimated parameters. The DPI data produces contradictory results. Each of the three functional forms seems to fit the data relatively well. Every one of the interactive terms multiplying MMM by Share SMD and its logarithmic and quadratic transformations is statistically significant at the .05 level. Substantively, the effect of the share of seats allocated in majoritarian elections is much larger: model 1 now suggests, for example, that when Share SMD increases by 0.5, the number of parties diminishes by 1.25. The three models nonetheless prompt different conclusions about the effects that highly majoritarian mixed systems have on party system fragmentation. While the DPI estimates shown in models 2 and 3 are useful for the purposes of conducting simple hypothesis tests, they offer no obvious substantive insight into the relationship between Share SMD and the number of legislative parties in MMM systems. We employ the logic outlined in King, Tomz, and Wittenberg (2001) to illustrate how our predictions change when we use models 2 and 3 as our guide. We draw M ⫽ 1000 sets of simulated parameters for majoritarian, proportional, and mixed systems characterized by different proportions of seats assigned to SMD races (increments of .01 between 0.1 and 0.9) from a multivariate normal distribution with mean that equals the vector of coefficients estimated in models 2–3 and variance equaling the estimated variance–covariance matrix of the robust MM models. We hold other variables constant at their medians13 and calculate, for each of these systems, the expected value and 90 percent confidence interval of the number of legislative parties. Figure 8.1, generated with the estimates of the logarithmic model, indicates that the number of parties is quite high when most of the seats are distributed in the PR segment of the election: when Share SMD equals 0.1, the estimated ENPS is above 4. Consistent with Shugart and Wattenberg (2001b), as mixed systems become more and more majoritarian legislative fractionalization decreases sharply, to about 3 effective parties when Share SMD ⫽ 0.5 and to the number generally observed under “pure” majoritarian institutions (around 2.5) when Share SMD ⬎ 0.7. This figure suggests that highly majoritarian mixed systems produce highly majoritarian results. The pressure toward two-party competition seems greater in systems where most of the seats are allocated in SMD races.

Effective number of legislative parties

better, worse, or just different? / 135

5 Mixed systems 4

3

2

Proportional

Majoritarian

1

Density of % SMD in mixed systems

0 0.0

0.2

0.4

0.6

0.8

1.0

Share of seats allocated in SMDs E(ENPS|X)

90% Confidence Int.

Figure 8.1 The ENPS in proportional, majoritarian, and mixed systems estimates from model 2, table 8.3 (DPI Data).

Figure 8.2 paints a strikingly different picture. When we employ the estimates from model 3, the simulated quantities of interest support a scenario in which the number of parties in mixed electoral systems varies as a “U-shaped” function of the proportion of seats allocated to the SMD component. As in the previous simulation, the number of parties is high (above 4) when a small number of seats are assigned via SMD. At first, increases in the majoritarian bent of the mixed system are accompanied by a sharp decrease in the ENPS, which reaches a minimum of about 2.3 when half of the legislative assembly is elected through majoritarian rules. Past this minimum, however, the number of parties increases again as the electoral institutions become more majoritarian, to approximately 3.5 when Share SMD ⫽ 0.8. The differences between the predictions made by the two models, particularly with respect to the kinds of party systems that are likely to emerge under highly majoritarian hybrid rules, are quite substantial. Do mixed systems where the majority of seats are assigned in SMD races approximate “pure” SMD systems in the kinds of legislatures they produce? Or do such heavily majoritarian mixed systems counterintuitively increase

Effective number of legislative parties

136 / mixed electoral systems

5

Mixed systems

4

3

2

Proportional

Majoritarian

1

Density of % SMD in mixed systems

0 0.0

0.2

0.4

0.6

0.8

1.0

Share of seats allocated in SMDs E(ENPS|X) 90% Confidence Int.

Figure 8.2 The ENPS in proportional, majoritarian, and mixed systems estimates from model 3, table 8.3 (DPI Data).

the number of parties? Clearly, observing a “U-shaped” rather than a quasi-linear relationship between the share of SMD seats and the effective number of parties has wide-ranging implications for policy making and governance. Moreover, these divergent findings profoundly affect the answers to questions about what mixed systems are, how they work, and how they should be studied. Unfortunately, this important question will not be resolved here. Cross-validation, in fact, reveals that the three models’ in-sample predictions fit the data equally well. More important, as this discussion has demonstrated, the limited number of countries employing MMM systems (about 20 in all), coupled with the diversity that characterizes them on sociological, political, and institutional dimensions, renders any estimate of the impact of the share of seats allocated in the majoritarian tier highly model-dependent (see Ho et al. 2004; King and Zeng 2004). The relatively scarce overlap between the distributions of many baseline variables—some included and some omitted—for different values of Share SMD requires any causal inference about the effect of such variable to be based on extrapolation to areas where there is little or no data (see King and Zeng 2004). In this sense,

better, worse, or just different? / 137

counterfactual inferences about the effects of changes in the share of seats that an MMM system reserves for its majoritarian component are highly sensitive to the functional form employed and other modeling assumptions. Such conclusions may be substantively quite different, but they are virtually indistinguishable for validity, verisimilitude, and the firmness—or more appropriately, precariousness—of their evidentiary bases. The paucity of mixed systems and their marked heterogeneity complicates the application of possible solutions such as nonparametric preprocessing based on propensity score matching (see Ho et al. 2004). What lessons can be drawn from the erratic and inconsistent results shown in this chapter Ironically, these results are more exemplary of the difficulty of drawing meaningful inferences when the available observations are so few and so different from one another than they are illuminating of the relationship between legislative fractionalization and the majoritarian nature of a mixed electoral system. At a minimum, however, these results show that the idea that highly majoritarian mixed systems produce SMD-like results, far from being a given, remains an unverified assumption, and one that the data do not support strongly. When the analysis focuses on Golder’s (2005) DES data, the salience that the mechanics of an MMM system confer upon each of its component parts does not appear to affect the composition of national legislative assemblies. When the DPI data are examined, a quadratic functional form performs just as well as the linear and the logarithmic specifications. It could be that these results are so muddled because, as Sartori (1994) predicted, even different varieties of SMD–PR combinations cannot alter significantly the preexisting structure of party competition.14 It could also be that, in the presence of previously fragmented party systems, markedly majoritarian mixed systems may strengthen small parties. Alternatively, it could be that the nature of the data renders the estimates highly sensitive to single, perhaps unusual observations. As Ho et al. (2004) put it, the empirical analysis shown here has merely demonstrated that it is possible to find a specification consistent with a hypothesis that differs from those advanced by Shugart and Wattenberg (2001b) and others. However, given the centrality of this proposition to the “best of both worlds” conjecture and the prominence that this approach has acquired within the literature on mixed system, pointing out the absence of a clear relationship between alternative PR–SMD mixtures and the structure of legislative party competition is significant. Conclusion Where do mixed systems fit in the established distinction between majoritarian and proportional rules, which dominated electoral studies for nearly half a century? The first set of findings presented in this chapter generally

138 / mixed electoral systems

support the view that mixed systems may produce outcomes that lie “somewhere in between” those observed under SMD and PR. Systems whose mechanics guarantee fully proportional outcomes usher in legislatures that are at least as fragmented as those elected through proportional representation. Fewer parties, in contrast, are able to win representation under rules that balance the two components more evenly. The fact that such outcomes do not appear to be substantively and statistically distinct from those found under pure types of election rules, however, casts some shadows over the claim that mixed systems simultaneously lead to vast improvements in the representation of small parties, the coherence of governing majorities, and the balance between local and national interests. Also, the analysis suggests that the three most widely used typologies of mixed electoral systems capture the differences in the number of legislative parties equally well. Further empirical tests, which evaluate in greater depth the institutional variation found in the world’s mixed systems and explicitly take into account their interactive features, muddy the waters. The “somewhere in between” hypothesis finds little support when the analysis focuses on the share of seats that a mixed system allocates in its proportional and majoritarian components. Rather, mixed systems may produce more complex aggregate outcomes than those uncovered in the literature thus far. While the nature of the data employed prevents us from drawing causal inferences of which we can be reasonably confident, the findings in this chapter are significant because they suggest that mixed systems may not simply produce outcomes that lie somewhere in between the majoritarian and proportional extremes. Highly majoritarian mixed systems may fail to reduce legislative fractionalization. This result appears particularly plausible in polities with a history of multiparty competition.

C h ap t e r 9 Conclusion

Mixed electoral systems are not social laboratories. When combined in a single election, SMD and PR contests do not take place in isolation from one another. It is hardly a surprise, therefore, that the two components of the election do not function as if they existed in isolation. We have found evidence of contamination virtually everywhere we have looked: the placement of SMD candidates affects a party’s PR performance, the prospect of receiving a PR vote boost encourages multiparty competition in SMD, nomination strategies in the majoritarian component are shaped by the institutional rules of both tiers, party choice is affected by perceived candidate quality, candidate choice is less strategic and more sincere than under first-past-the-post, legislators engage in voting behavior that is inconsistent with their seat type, and highly majoritarian mixed systems may fail to reduce legislative fragmentation. The violation of the condition of independence between the two components of the election compromises the validity of causal inferences drawn from “controlled comparisons.” Measuring the relative causal effects of SMD and PR with higher levels of precision is a worthwhile endeavor. However, our findings reveal that mixed electoral systems are not the appropriate setting in which such effects can be established. What We Know about Mixed Systems: Putting the Puzzle Together Our basic argument challenges the conventional wisdom about the consequences of mixed electoral systems. Our constant refrain has been that contamination is the foundation of the incentive structure in mixed electoral systems. In contrast to much of the existing literature, we posited that the two components of mixed rules do not operate independently, but are characterized by profound interaction that alters the decision-making processes and ultimately the choices of relevant political actors. Understanding the outcomes produced by mixed electoral systems requires that we consider

140 / mixed electoral systems

the implications of mixing majoritarian and proportional institutions. Assuming independence in the mechanics of mixed systems may not only bias our inferences; it may also blind us to how mixed systems really work. We began this book by outlining the logic of contamination and its potential to undermine the contemporary research on mixed electoral systems. “Controlled comparisons” are quite attractive at first glance. Measuring the effects of SMD and PR in the same country, the same district, or the same legislature appears to resolve potential bias and model dependence that notoriously plagues counterfactual inference: the effects of both now seem to be directly observable. However, controlled comparisons are only viable if the two components of the election are independent: mixed SMD works like pure SMD, mixed PR works like pure PR. If mixing SMD and PR alters how they would have operated in the absence of the other, controlled comparisons can tell us little that we did not already know. After offering our definition of mixed electoral systems in chapter 2, we have systematically shown in successive analytical chapters how contamination influences real-world outcomes. Chapter 3 sketched out the argument in its simplest form: by nominating an SMD candidate, a party may gain more votes on the proportional side of the ballot. Under the independence assumption, no relationship should be evidenced. But, our empirical analysis of district-level data uncovered evidence of the interaction between tiers. The following three chapters refined this logic and investigated its implications in much greater depth. First, if political parties increase their PR vote by nominating candidates in corresponding SMD races, parties should field candidates in as many constituency-based contests as their resources permit. Yet, many political parties fail to nominate candidates in all or even most districts because they pursue preelectoral coordination. If contamination matters, as we claimed in chapter 3, why would parties fail to take full advantage of the “bonus” that accrues in PR from SMD nominations? Chapter 4 showed that variation in the institutional features of mixed systems influences nomination strategies in SMD races. In mixed systems dominated by majoritarian rules, parties confront strong incentives to enter into formal or informal pacts that coordinate nominations and reduce the total number of candidacies. While highly majoritarian mixed systems may attenuate the incentive to “go it alone,” the fact that different combinations of SMD and PR result in different levels of elite coordination provides additional evidence of contamination. The analysis in chapter 3 relied on assumptions about voting behavior and only assessed voters indirectly through district-level election returns. In chapters 5 and 6, we employed public opinion data to uncover individuallevel evidence of contamination and specify with greater precision the micro-mechanisms that explain the results found in the aggregate data.

conclusion / 141

Chapter 5 assessed how party choice is affected by a voter’s subjective evaluations of SMD candidates and discussed how voters integrate the information generated in the SMD campaign into their decision-making processes for the proportional ballot. Chapter 6 provided evidence for the mechanism by which contamination leads to multiparty competition in the majoritarian tier. The presence of a higher number of SMD candidates gives more voters the opportunity to select a candidate affiliated with their favorite party. Wider choice, in turn, discourages strategic behavior and spawns greater dispersion of votes away from the top competitors. Viewed as a whole, chapters 3–6 tell a cohesive story of how Duverger’s logic breaks down in the majoritarian races of mixed electoral systems. The existence of a parallel proportional election keeps small- and medium-sized parties alive; the possibility of boosting their PR performance via candidate placement alters the strategic calculus that guides their decision to enter or abstain from competition in SMD. Strategic entry decisions in the majoritarian component are not solely driven by expectations about winning or losing a district because the placement of hopeless but attractive candidates may yield significant benefits in PR. Though markedly majoritarian mixed systems may offset these advantages, in most cases the presence of a higher number of candidates on the SMD ballot weakens the reductive pressure of majoritarian rules on the number of parties. In turn, greater competition hampers voter coordination on the most viable candidates and greater choice reduces the strategic desertion of less viable contestants. In sum, mixed electoral rules modify the psychological effect that their two components would produce in isolation. More complicated mechanics produce more complex outcomes. Contamination affects political actors as they seek office and as they choose parties and candidates. But, do SMD–PR interactions resonate beyond the campaign period? We investigated the incidence of contamination between elections in chapter 7 by focusing on the voting behavior of legislative representatives. Using electronic voting data, we illustrated how the implications of contamination extend beyond the straightforward logic presented in chapters 3–6. The “specialization” of deputies seems to be discouraged by the prospects of competing in both tiers and the possibility of moving from one to the other in successive elections. These features of mixed systems induce hedging in a legislator’s floor behavior. In addition, we described how the interaction of proportional and majoritarian rules may shape nomination procedures and blur the difference between the behavior of the two “legislator types” that most scholars and observers expect mixed systems to produce. The fact that we observed no difference in the propensity of SMD and PR deputies to stick to the party line raises doubts about the ability of mixed electoral systems to strike

142 / mixed electoral systems

an ideal balance between commitment to the needs of local constituents and dedication to national priorities. Chapter 8 also generated questions about the potential for mixed systems to create more coherent party systems than those observed under “extreme” electoral institutions. In this chapter, we asked: how well do mixed systems provide representation to minority parties while still preventing excessive legislative fragmentation? Our analysis of legislative fractionalization supports the notion that mixed systems produce intermediate levels of fragmentation, though such outcomes rarely differ from either of the two extremes. However, we do not find conclusive evidence for the proposition that highly majoritarian mixed systems are particularly effective in reducing the number of legislative parties, undermining the argument that institutional designers may increase or decrease fractionalization by giving their mixed systems a more proportional or majoritarian flavor. Our results can be summarized to generate two major lessons about how mixed systems work and how we should study them. First, mixed electoral systems are unsuitable for the study of the independent consequences of SMD and PR. When mixed systems began to proliferate across the globe, many researchers viewed them as a laboratory to resolve unanswered questions about majoritarian and proportional institutions. Existing scholarship had revealed that majoritarian rules are more conducive to strategic behavior, a smaller number of parties, and greater independence of legislators from party bosses than proportional representation. But, we did not know exactly how much more. Mixed systems promised to reveal how much more by substituting counterfactual inference with direct measurement. However, the findings presented in this book demonstrate that “controlled comparisons” are based on faulty assumptions. Second, because the two tiers of mixed systems interact, the classification of mixed electoral systems as an intermediate form of majoritarian and proportional systems whose outcomes synthesize the independent effects of SMD and PR is untenable. This “degreeism”—as Sartori (1991) defines it, or “the abuse of the maxim that differences in kind are best conceived as differences of degree”—is likely to mislead scholars as they assess the consequences of mixed systems. Rather than lying somewhere along a continuum that ranges from national-level, proportional seat allocation to British-style plurality, mixed electoral systems constitute a distinct, qualitatively different class of electoral institutions. Whether they combine “the best of both worlds” (see Shugart and Wattenberg 2001b) or are a “mismarriage” (Sartori 1994, 74) remains to be seen. Mixed systems present the literature with a new challenge. The emergence of an altogether different, novel family of electoral institutions offers the opportunity to investigate a wide range of new electoral and legislative

conclusion / 143

outcomes. Moreover, the striking institutional variation that characterizes the many mixed systems crafted in recent years allows psephologists to evaluate how alternative SMD–PR combinations affect strategic voting, party competition, and legislative behavior. What We Still Need to Know: Framing New Puzzles As each of the analytical chapters has made clear, many questions about how mixed electoral systems work remain unanswered. This book proposes a research path to search for answers to these questions; concentrating on the interactive qualities of these two systems offers the greatest potential to get those answers right. Chapters 3 and 5, for instance, provide a schematic view of the psychological processes that lead voters to incorporate the cues they receive in the SMD campaign into their choice of a party on the PR ballot. Candidate quality, whether actual or perceived, seems to affect PR voting decisions. In this area, it would be useful to emulate research on American campaigns to evaluate how attributes such as incumbency status, occupational background, and national prominence of SMD candidates also affect a party’s PR performance. Moreover, future research may investigate in greater depth how the influence of contamination varies along with levels of political knowledge, interest, and involvement. Finally, it may be valuable to establish how contextual and institutional factors affect the integration of information shortcuts from one tier of the election to the other. This type of research could have important implications for the strategies that political parties should implement under mixed electoral rules. More generally, it would enhance our understanding of voter decision making in any setting where multiple elections are held concurrently under different rules. The results presented in chapters 4 and 6 also point to future analyses on strategic entry and voting in mixed electoral systems. Assessing the evolution of coordination strategies employing longitudinal data would provide valuable information about how quickly politicians and voters adapt to changing electoral incentives. Similarly, evaluating whether the low incidence of strategic voting found in Japan extends to other mixed electoral systems would provide greater insights into voter coordination. Studying strategic behavior in other mixed systems, particularly at the subnational level, would enhance our understanding of institutional effects as well as the interaction between local and national elections. Chapters 7 and 8 only scratched the surface of the complex systemic consequences of mixed electoral systems. Both also provide empirical evidence that is inconsistent with the “best of both worlds” approach.

144 / mixed electoral systems

The findings in chapter 7 have implications that go beyond party discipline, but scholars should consider other aspects of legislative activity to provide a fuller account of how mixed systems shape politics between elections. It might be worth investigating, for instance, how electoral path affects representatives’ choices of committees with more uniform or more targeted externalities, how the incentives to perform constituency service and pass pork-barrel legislation are affected by the mechanics of mixed electoral systems, and how parties respond to their legislators’ voting behavior through nomination decisions. This type of research would provide deeper insights into the propensity of legislators to “specialize” as well as the kind of representation voters are likely to get under mixed electoral institutions. Chapter 8 focuses our attention to the inter-party dimension of political competition. More specifically, the inconclusive findings we presented raise questions about how mixed systems may affect the development and configuration of a country’s party system. Future studies may investigate the potential of mixed electoral systems to alter preexisting patterns of competition and party system fragmentation by evaluating their consequences in countries with completed democratic transitions, those that have moved away from pure proportional rules, and those that reformed pure majoritarian rules. Another set of important research questions addresses how government functions under mixed electoral rules. To what extent do different kinds of mixed systems promote alternation, cabinet stability, and the emergence of legislative coalitions? In addition, how do different variants of mixed electoral systems bring a polity closer to the “consensus” or “majoritarian” ideal types described by Lijphart (1996)? In some countries, like Japan, mixed systems seem to have produced consensus-based politics, coalition government, and power sharing; in other cases, such as Italy, they seem to have concentrated power in the hands of the chief executive, limited dialogue between government and opposition, and generated greater policy instability. Finally, in transitional countries like Russia and Ukraine, mixed electoral institutions have failed to produce more coherent, institutionalized party systems. Do these outcomes vary systematically with the institutional features of mixed electoral systems? Do they depend on the structure of party competition that preceded the introduction of hybrid rules? Answering these questions might help us identify the conditions under which mixed systems are likely to bring out the best and the worst of the majoritarian and proportional worlds. Survival or Extinction? The Future of Mixed Electoral Systems The critical test for any species—living or otherwise—is the test of survival. Many species have populated the earth for a short period of time, failed to

conclusion / 145

successfully adapt to changing environmental or demographic conditions, and ended in extinction. In a similar way, election rules have been created, adapted, and faced the test of survival. Some election rules occupy a small but secure niche: the Single Transferable Vote (STV) in Ireland and Malta, for example. Other rules seem to be on the decline: the SNTV was abandoned in Taiwan, South Korea, and Japan, resists only in Jordan and Vanuatu, and has been provisionally instituted in post–Taliban Afghanistan. What is the prognosis for mixed electoral systems? The survival of mixed systems largely depends on two factors. First, it likely hinges on the ability of mixed systems to fulfill their promise, which varies with the context in which they were adopted. Second, and perhaps more importantly, it is contingent upon how well mixed systems serve the interests of dominant parties. On both counts, the performance of these electoral systems has been mixed. Mixed systems have not always performed as advertised. In Japan, mixed systems have not produced as dramatic a departure from SNTV as expected. In Italy, the introduction of a mixed system appears to have been successful in producing alternation, the rise of bipolar competition, and an increase in the stability of governments. However, it also failed to meaningfully reduce legislative fragmentation, decrease the power of party organizations, and give voters control over ballot access. In New Zealand, more permissive rules have resulted in greater representation for small parties, but this has come at the cost of increased government instability and preelectoral uncertainty about the shape of potential governing coalitions. Moreover, the increased popular participation that MMP’s proponents predicted has not materialized. Lastly, while mixed systems may have stimulated party development in the postcommunist world, those party systems remain inchoate. While the disappointment with mixed electoral rules may be a function of unrealistic expectations that accompanied their introduction, there is little evidence to suggest that the outcomes they produce are necessarily better than the alternatives. Their inability to significantly modify certain aspects of party competition may reflect the opportunity they offer elites to change an unpopular electoral system while preserving enough of its features to perpetuate the advantages they enjoyed. Whatever the reason, if mixed systems are perceived as a failure, pressure may develop to abandon these rules. Mixed electoral systems have generated considerable discontent among political elites, both winners and losers. In some cases, dissatisfaction appears to stem from the unpredictability of the outcomes they produce. Politicians seem to have engaged in the same fallacy that characterizes much of the scholarly research on this topic: assuming the independence of the two components of the election. In this regard, the words of former

146 / mixed electoral systems

Ukrainian president Leonid Kuchma are emblematic: I often ask myself if I didn’t make a mistake by agreeing to a mixed electoral system for the last parliamentary elections . . . The logic of the decision [was that] . . . elections on the party list would speed the development of blocs and coalitions, bring greater clarity, specificity, and simplicity to the party palette. It did not turn out this way.1

Kuchma’s dissatisfaction has been shared by other prominent political leaders. Vladimir Putin’s proposed elimination of the majoritarian component of Russia’s electoral system is motivated by his desire to exert greater control over the Duma. Similarly, Silvio Berlusconi’s Forza Italia recently proposed a reform that would bring the Italian mixed system much closer to PR. While Berlusconi has defended his proposal as an attempt to remove some of the “technical difficulties” the current electoral law allegedly imposes, many observers have speculated that the reform is designed to eliminate the 1.8 million vote gap between the governing coalition’s PR performance and the votes garnered by its less popular SMD candidates. The jury is still out on whether or not mixed systems combine the best of the majoritarian and proportional worlds. While mixed systems were viewed as an optimal solution to the challenges faced by institutional designers in both old and new democracies throughout the 1990s, citizens and political elites have frequently expressed second thoughts about the selection of mixed electoral systems. Mixed systems have been abandoned in Azerbaijan, Bulgaria, Croatia, Kyrgyzstan, Macedonia, and Ukraine; Russia is next in line to abandon its mixed system. If mixed electoral systems generate the best of both worlds, why have they been discarded so readily? Perhaps everything is not for the best in this best of possible worlds.

Notes

1

A New Species of Electoral Institution? Relative Causal Inference and the Study of Mixed Electoral Systems

1. Duverger’s law is: “the simple-majority [plurality] single-ballot system favors the two-party system” (Duverger 1954, 217). Duverger’s Hypothesis predicts: “the simple-majority system with the second ballot and proportional representation favors multi-partyism” (Duverger 1954, 239). 2. By “accumulation of knowledge,” Riker (1982, 753) means “not only that the corpus of propositions is growing, but also that each one of the series of revisions is more general and precise than its predecessor.” 3. SMD systems may allocate seats according to either the majority or plurality formulae. However, we will employ the acronym SMD to describe “singlemember districts with the plurality formula.” 4. Of course, not all elections held in multimember districts are proportional. The single-nontransferable vote, limited vote, bloc vote, and cumulative vote are examples of electoral systems where candidates are elected in multimember constituencies through the application of the plurality formula. 5. See, e.g., Browne and Patterson (1999), Christensen and Johnson (1995), and Cox and Niou (1994). 6. See, e.g., Downs (1957), Cox (1990), and Shepsle (1991). 7. See, e.g., Bowler, Farrell, and Katz (1999), Cain, Ferejohn, and Fiorina (1987), Carey (2003), Carey and Shugart (1995), Ames (1995a), and Golden (2003). 8. See, e.g., Rae (1971), Cox (1987; 1997), Cox and Shugart (1996), Taagepera and Shugart (1989), and Tsebelis (1986). 9. See, e.g., Amorim-Neto and Cox (1997), Ordeshook and Shvetsova (1994), and Benoit (2002). 10. We will limit the discussion to the comparison of SMD and PR. However, the arguments that follow can be easily extended to any other type of electoral system. 11. When we refer to the number of parties or candidates, we generally do not mean the number of parties or candidates that participate in the election (i.e., that are on the ballot) but the number of relevant or viable parties. Normally, the literature employs the indicator proposed by Laakso and Taagepera (1979) as a measure of the “effective number of electoral parties.” This is equal to N ⫽ 1/⌺pi2, where pi is the proportion of the total vote received by party i in a given district. When we refer to the number of parties/candidates on the ballot, we will generally use the expression “number of entrants.” 12. District magnitude (M) is the number of seats distributed in an electoral district.

148 / notes 13. For district magnitudes smaller than or equal to five (see Cox 1997). The M ⫹ 1 rule is an extension of Duverger’s Law and suggests that the number of candidates running would be one larger than the number of seats available in districts (Reed 1990). 14. There are exceptions to this observation. In single ballot mixed systems (used in Mexico, the Italian upper house, Taiwan, Senegal, and South Korea) voters only cast a ballot for an SMD candidate that are pooled to allocate PR seats. 15. See Bawn (1993), Benoit and Schiemann (2001), Navarra and Sobbrio (2001), Remington and Smith (1996), and Shugart (2001). 16. See Bawn (1999), Benoit (2001a), Cox (1997), Cox and Schoppa (2002), Ferrara (2004), Gschwend, Johnston, and Pattie (2003), Herron (2002a), Herron and Nishikawa (2001), Johnston and Pattie (2002), Karp et al. (2002), McAllister and White (2000), Moser (1997 and 1999), Moser and Scheiner (2004), Reed (1999; 2001); Shvetsova 2002. 17. See Bawn and Thies (2003), Haspel, Remington, and Smith (1998), Herron (2002b), Lancaster and Patterson (1990), Stratmann and Baur (2002), and Thames (2001). 18. See Kostadinova (2002), Nishikawa and Herron (2004), Sartori (1994), Shugart (2001), and Shugart and Wattenberg (2001b). 19. The assessment of causality in randomized and non-randomized social science studies was formalized by Rubin (1974; see also Holland 1986; Reiter 2000). The discussion is an adaptation of the model presented by Reiter (2000). 20. In this instance, cleavage structure could affect both the choice of an electoral system and the number of parties. 21. For other problems of interests, such as the investigation of the impact of electoral rules on legislative behavior, we cannot observe the impact of SMD and PR on the same legislator (some are elected through SMD, others through PR). However, the fact that we can evaluate the difference between the behavior of legislators elected to the same assembly through different rules appears to be a marked improvement (in terms of balancing intervening variables and isolating country effects) from the comparison of legislative voting in, say, the United States and the Netherlands. 22. Cox (1958, 20–21) describes several examples in which interference may constitute a problem. In the most straightforward example, Cox (1958) discusses experiments where adjacent agricultural plots are tested for the effectiveness of an inoculation protecting the crops from a particular virus. To avoid interference (the inoculation affecting the control plot where no treatment is applied), it is often useful to separate and isolate the plots. 23. At least in magnitude. 24. This is the case when plurality rules are used. 25. Cox (1997: 21) argues: “If a party can run and elect candidates under the more permissive system, it may decide to run candidates in the other system as well— not to win seats, perhaps, but to keep its electoral organization in good trim, to establish its blackmail potential, or for other reasons. In this case, the party system in each chamber should be influenced by that of the other, in such a way as to lessen observed differences.” 26. The question is: “are the outcomes observed in the SMD component of mixed electoral systems the same we would observe under pure SMD?”

notes / 149

2

A Taxonomy of Mixed Electoral Systems

1. http://www.ucmp.berkeley.edu/history/linnaeus.html. 2. To be sure, some efforts to classify election rules exist, most notably Reynolds and Reilly (1997). But, even this detailed and comprehensive taxonomy of election rules does not approach the question as systematically as Linnaeus did in biology in the eighteenth century. 3. By definition, presidential elections should be M ⫽ 1 systems. The collective presidency in Bosnia-Herzegovina could be defined as M ⫽ 1 because the casting and aggregation of votes is dependent on the voters’ and candidates’ ethnicity. A single president is elected for the Serbian, Croatian, and Bosniak populations. Alternatively, since the top vote getter is designated as the chairperson, we could conceive of the election as M ⫽ 3. 4. Ordinal ballots, for example, are used to select the lower house of Australia’s parliament. Voters rank order candidates for districts where M ⫽ 1. In Ireland, voters rank candidates for election in districts where M ⬎ 1. In Japan’s abandoned Single Non-Transferable Vote system, voters cast a single nominal ballot in districts where M ⬎ 1. 5. The authors indicate that these systems may be part of a broader family of mixed systems, but focus their volume on rules with nominal tiers and M ⫽ 1. 6. SNTV, used in Japan’s Upper House elections, is included in this label. 7. LR-Hare and D’Hondt are two distinct formulae used to distribute PR seats. The Largest-Remainders Hare approach allocates seats in the following manner. First, a quota is calculated by dividing the valid votes for political parties by the total number of seats. Party votes are divided through by the quota. The resulting integers indicate the “full seats” each party receives. Any seats unallocated through “full seats” are subsequently allocated by the remainders. The D’Hondt method divides party votes by a sequence of divisors (1, 2, 3, 4, etc.). Seats are allocated to parties with the highest totals based on a series of calculations. More information about these methods is available in Farrell (2001), Lijphart (1994) and at http://www.aceproject.org. 8. Massicotte and Blais (1999) also note historic examples in Canada at the provincial level. 9. Some mixed electoral systems, such as Ecuador and Senegal, use the block vote. 10. Some mixed electoral systems, such as Lithuania and Japan (Upper House), use open-list PR, allowing voters to select a party and cast preference votes for candidates in the party-list component. 11. In the Italian case, the single ballot applies only to the upper house. Single ballots are also used in some German Länder. 12. Azerbaijan also prohibited dual candidacy. 13. In these extreme cases, the allocation formula must actually differ for the systems to be mixed. That is, if a country has regional PR and one region has M ⫽ 1, the system is still PR if the voters select a party. If, however, voters select a candidate in that district, it could be an extreme form of mixed system. 14. It could also be interpreted as a modified presidential system. 15. Calculations are based on data from 32 national level mixed systems (Massicotte and Blais 1999). 16. This system is employed in the Italian Senate and is known as the “scorporo totale.” In the election of the Chamber of Deputies, vote transfers through the

150 / notes

17. 18.

19. 20. 21.

mechanism of the “scorporo parziale,” only involve the subtraction, from the vote of each PR list, of the votes that were necessary for their SMD candidates to win their respective districts (the votes received by the second-placed candidate ⫹ 1). Ukraine and Russia were in the process of abandoning their mixed systems at the time this book was being written. Russia uses a hybrid of both national and regional lists to determine seat allocation within parties in PR, but this is distinct from the use of regional lists to allocate seats to parties. In 1999, parties could place 18 candidates on their national PR lists. Other candidates were placed on regional lists. The party’s overall seat allocation was not affected by this division. Rather, it determined the identity of the candidate who would receive the seat (IFES 1999). Arrow (1951), however, shows that this effort was quixotic; no electoral system can simultaneously produce logical results with fair rules. Lakeman and Lambert (1955) also viewed mixed systems as a form of PR. U. W. Kitzinger (1960) cited in Bawn (1993, 986).

3

The Logic of Contamination: District-Level Party Systems

1. Specifically, voters who go to the polls. See chapter 6 for more discussion of this issue. 2. It is possible that one of these voter types dominates. It is also possible that voter types vary within countries, with one or two dominant “types.” This is an empirical question for future research. 3. An underlying assumption is that parties generally strive to win elections. It is possible to consider other goals, including participating in government and implementing good policies (Fenno 1973; Laver and Shepsle 1996). However, we only consider elections to build a simpler model. 4. The participation of some candidates may have a negative effect on PR performance as well—a noncharismatic individual could hurt the party in PR. Also, some parties might benefit from the lack of information about their actual platform—voters may be more likely to support the image of the party rather than the substance. If the SMD candidate increases knowledge about the party’s positions, it could damage the party’s performance. Despite these possible scenarios, parties presumably field candidates that they expect will help their overall performance. Moreover, the effect of increasing awareness of the party generally should be positive. 5. As we discuss in chapter 4, this may be influenced by the institutional features of the system as well as financial, personnel, or technical resources. 6. Independent candidates are less prevalent in Lithuania than in Russia and Ukraine. 7. The House of Councilors combines PR and Single Non-Transferable Vote in its election. 8. SNTV candidate in the case of the Japanese House of Councilors. 9. In one of the cases assessed in this chapter, Italy, the coding of this independent variable is rendered problematic by the fact that a great deal of preelectoral coordination takes place in SMD races. In Italy, parties tend to run autonomous lists in PR elections. However, the most significant parties

notes / 151

10. 11.

12.

13.

14.

15. 16.

coalesce in the majoritarian component and jointly support SMD candidates that run under the coalition’s, not the party’s, banner. Most political parties, therefore, forego the benefits that placing their own candidates in each SMD race should be expected to yield in the form of a PR vote boost. Cox and Schoppa (2002, 1039) explain coordination by conjecturing that the Italian electoral system neutralizes contamination by allowing parties to run common candidates in single-member districts without favoring any of them in the PR tier. Moreover, they argue that the Italian electoral law, which requires all lists that are affiliated with a candidate in a single-member district to also jointly support the same candidates in that region, produces incentives to form “systematic alliances” in the SMD tier. While it is true that the electoral law of 1993 requires all the PR lists that are jointly backing a candidate in the plurality tier to also jointly back all single-member district candidates in that PR district, it is also a fact that in most cases candidates in the SMD tier are affiliated with only one PR list, not with all the parties that compose the alliance under whose banner they run. In the majority of cases, thus, only one party officially backs a single-member district candidate; the other parties in the coalition support that candidate informally. Contrary to what Cox and Schoppa (2002) seem to imply, the affiliation of a particular candidate to a specific PR list is common knowledge among voters. This considered, there is no reason to expect contamination to be inhibited. The placement of a candidate in the SMD tier should still be expected to boost the PR performance of the list with which the candidate is affiliated. We do not include this variable in the tests employing data from the 1994 Italian elections and from the 1953 election of the German Bundestag. Generally, Incumbent was coded 1 if the candidate was an incumbent from any SMD, not only the district in which s/he ran in the election under analysis. In the analysis of the 1996 Italian elections, however, Incumbent was coded 1 only if the candidate who won that district in the previous election was competing for reelection. Both variables are based on data obtained from the 1989 Soviet census and are aggregated at different levels. For Russia and Ukraine, the data are at the oblast level. This unit is roughly comparable to a U.S. state. Lithuanian data are a combination of regional- and city-level data. Kensei (2003) is used to create the Primary variable. Other demographic factors, such as education, age, and population density, also account for party support in Japan. However, these indicators are not included because they are highly correlated with the Primary variable. The question of regionalism in Ukraine has been debated by many scholars. The simple division of Ukraine into four regions may not obtain to all research questions. But, less nuanced divisions of Ukrainian regions have been relevant in elections research (Barrington and Herron 2004). The y-intercept, therefore, expresses the effect of South. For district i, the dependent variable Yi is a vector of the log ratios for J⫺1 parties (party J is the reference party employed to calculate the log ratios). It is assumed to be multivariate normal with mean vector ␮i and variance matrix ⌺. We then model the systematic component ␮i as a linear function of the covariates Xi, which includes SMD candidate and the controls we discussed earlier: ␮i ⫽ [Xi1␤1, Xi2␤2, . . . , Xi(J-1) ␤(J-1)].

152 / notes 17. In Germany, we only assessed the Bavarian Party (BP), Deutsche Partei (DP), DRP and Zentrum (Z); in New Zealand, we only evaluated the Christian Coalition (CC), the McGillicuddy Serious Party, United New Zealand, and Progressive Greens (PG). 18. The election returns collected by Kokusei Joho Center (2001) are used to create the dependent variables. This data source is used since it contains the PR vote shares for each party at the city, ward, and village levels, which makes the calculation of PR vote shares for each single member district possible. Some districts are dropped from our analysis since some of these units are split into more than one single-member district. 19. Given that Italy has regional lists, many PR constituencies are only “partially contested”; that is, not every party fields a list everywhere. If all of the parties that meet our criteria for inclusion were considered together, the number of observations would drop dramatically—from 475 to about 150—because SUR listwise deletes every observation where at least one of the parties did not compete. Tomz, Tucker, and Wittenberg (2002) propose a solution to this problem that involves the separate assessment of different “patterns of competition.” Accordingly, we present here the results of models that consider only those parties that ran PR lists everywhere. In separate tests, we added to the models the other parties that satisfy our selection rules. 20. Results in table 3.1 based on 95% confidence intervals. 21. Also see Herron and Nishikawa (2001) for more OLS results. 22. Contamination is also evidenced in the results of the Upper House election in 2001 (Nishikawa 2003). This election was held under a mixed system with SNTV and open list PR. The Japanese government adopted a new rule in 1998 and changed the existing closed PR rule to open list PR. In open list PR, voters can choose to vote for a candidate or for a party in PR. If voters select a candidate, their votes are counted toward the party to which the candidate belongs. Their votes also determine the ranking of the candidates on the PR list. Nishikawa (2003) reports that contamination is seen among voters who choose to vote for a party. That is, by placing candidates in the SNTV races, parties gain extra vote shares from voters who cast their ballots based on parties rather than individual candidates. 23. PR vote shares in single-member districts in the Japanese 1996 election are obtained from Asahi Senkyo Taikan published by Asahi Shimbun (1997). 24. The effective number of electoral parties (ENEP), which measures the number of significant parties or (more appropriately in this case) candidates competing in a given district, is equal to 1/⌺pi2, where pi is the proportion of the vote received by candidate i in that district (Laakso and Taagepera 1979; Taagepera 1997). 25. As noted earlier, the Japanese Upper House uses SNTV for its majoritarian component. Although SNTV has a weaker reductive impact on the number of parties than SMD plurality, we can still observe that the SNTV component of the Japanese Upper House elections creates a larger number of viable candidates than pure SNTV. For example, the effective number of parties is larger in the SNTV component of the Japanese Upper House elections than that in the Japanese Lower House SNTV elections. More specifically, the SNTV component of the mixed system in the Upper House elections (1983–1992) produced

notes / 153 the average of 4.95 where M ⫽ 3. On the other hand, the regular SNTV system in the Lower House elections (1983–1993) produced 4.45. The mixed system is deviating more from what the M ⫹ 1 rule predicts (Nishikawa 2002).

4

Going it Alone? Strategic Nominations in Mixed Electoral Systems

1. An important related question is how contamination influences the number of parties in the system. That is, would contamination increase the number of parties contesting elections relative to a pure PR system? Depending on the rules of entry and the benefits allocated to parties and candidates (in terms of funding limits, access to free media time, etc.), the incentives of mixed electoral systems may inflate the number of parties contesting in PR relative to a pure PR system. Candidates who covet particular district seats may find advantages in coalescing together under a party banner to gain access to resources that would benefit them in the SMD campaign. The absence of an SMD tier in a pure PR system precludes this incentive from influencing parties in pure PR. Creating pseudoparties that serve seat maximizing goals for politicians in SMD rather than PR is more likely to occur in countries with inchoate party systems, and anecdotal evidence of such a phenomenon exists in these systems. For instance, in the 1995 Russian parliamentary election, the “party” Pamfilova-Gurov-Lysenko nominated 33 candidates, but only 2 candidates gained seats—Ella Pamfilova and Vladimir Lysenko. The party’s primary purpose was to enhance the electoral prospects of its namesakes in the district races by creating a “party” that could access benefits restricted to registered political parties. Monetary penalties for micro-parties that contested but failed to perform well in PR were increased for the 1999 Russian elections, diminishing the benefits of such a strategy. Some research has discussed these incentives (Herron 2000), but it is a worthy question for future analytical work. However, this incentive does not fundamentally affect our research question. 2. In some cases, candidates in the SMD tier of mixed systems are affiliated with parties that have not gained entry to the PR ballot. But, these instances are rare. 3. We should note that mixed systems generally employ closed list PR, but Lithuania uses open list PR in its proportional tier in the election under analysis. 4. See also Shugart and Taagepera (1994); Shugart and Carey (1992, 206–225). 5. Irrespective of whether the runoff takes place under “top two” (Cox 1997) or majority–plurality rules. 6. This generally involves agreements to jointly back one another’s candidates in different SMDs. 7. In the case of Japan, we analyze data from the 2000 election; in the case of Hungary, we only have data for the 1998 election; in the case of Lithuania, we consider the 1996 elections. 8. For example, some Russian political parties engage in preelectoral coordination that is manifested by the creation of “blocs” of parties. But, many parties also agree behind the scenes to coordinate district nominations (Herron 2002c). 9. We exclude independent candidates because our purpose is to assess party strategy. 10. The incidence of independent candidacies is high in Ukraine and Russia. Therefore, the distinction between total number of candidates and number of

154 / notes

11.

12.

13.

14.

affiliated candidates is only meaningful in those two countries. In Ukraine, 1,537 out of a total of 3,084 candidates ran as independents (or declared affiliation with parties that did not run PR lists) in 2002; in the 1999 Russian elections, 1,178 out of a total of 2,227 candidates were independents. In both countries, independents not only participate in large numbers; they also win an unusually high number of SMD races (in Ukraine, 92 of 225 SMD races were won by independents in 2002; in Russia, 112 of 225 districts were won by independents in 1999). Given their importance, Russian and Ukrainian political parties should be expected to aggressively court successful independents to persuade unaffiliated deputies to join their parliamentary factions. Parties may withdraw from SMD races they cannot win and endorse an independent front runner they wish to include in their faction as a bargaining strategy. When PR district magnitude increases from 2 to 10, for example, we anticipate a marked increase in the number of affiliated candidates competing in SMD elections. When M ⫽ 2, small and medium-sized parties might find it useless to run their SMD candidates, as the PR vote bonus will not translate into PR seats. When M ⫽ 10, however, more parties can hope to win a PR seat, and therefore should be expected to run their own candidates rather than withdraw or engage in preelectoral coordination. Past a certain value of M, though, almost every party has the incentive to run its own candidates. Hence, we should not expect much difference between the impact that PR district magnitude exerts on nomination strategies in systems like Hungary (where M ⫽ 56) and in systems like Russia or Ukraine, where M ⫽ 225. We use the natural logarithm because we expect the relationship between margin of victory and the dependent variables measuring preelectoral coordination to be curvilinear. We do not have data on the margin of victory for South Korea. In some cases we had to deal with the problems posed by redistricting. In the case of Germany, the Central Election Commission provides 1998 SMD results adjusted for the redistricting that occurred prior to the 2002 elections. In the cases of New Zealand, Russia, and Ukraine, we assigned to each district the margin of victory at t ⫺ 1 provided that the core of the district remained the same (even if its boundaries underwent some changes). For districts that were split, we assign to both of the new districts the margin of victory of the preexisting SMD (this change only regards Ukraine). For districts that were merged into one, we took the mean margin of victory in the preexisting districts. In the coding of “incumbency,” we also had to account for redistricting. In the countries where reapportionment was significant (Germany and Ukraine), we assigned a value of 1 to each SMD where a candidate who had won a plurality race in the same region in the previous election was competing for reelection. We do not have data on incumbency for Hungary, Mexico, and South Korea. Because the dependent variable is a count, and thus can only take on nonnegative integer values, standard regression analysis could produce biased and inefficient parameter estimates. The more appropriate functional form to apply in this case is a poisson distribution (tests for overdispersion were negative, indicating that poisson is the appropriate distribution). Results we estimated through OLS and through Iteratively Reweighted Least Squares robust regression, which minimizes the influence of the outliers in our data (see Huber 1981), are identical to those shown in table 4.2.

notes / 155 15. When we include independent candidates in the dependent variable, and control for postcommunist countries to account for the large number of SMD candidacies typical of such countries, the estimates are very similar (in size, sign, and significance levels) to those shown in table 4.2. 16. When we exclude independents from the measure of the average number of parties, two notable changes occur. First, consistent with our hypotheses, the coefficient for PR district magnitude becomes negative and significant; more surprisingly, though, the parameter estimate expressing the effect of the percentage of seats allocated through SMD also becomes negative. In this case, however, there is a discrepancy between the results generated by OLS and those produced by robust regression, which generates estimates that are consistent with each of our hypotheses by limiting the influence of notable outliers. These are instances (particularly in Russia and Ukraine) where, as expected, parties yielded to independents in some districts and hence nominated very few candidates. This considerably inflates the average number of parties per candidate and leads to overestimates of the amount of coordination taking place in a given district (if there is only one affiliated candidate for 26 PR parties in a district, it is unlikely that all 26 parties support that candidate; it is more likely that parties distributed their support among the numerous independents in the race). Therefore, the results generated by regression techniques that are robust to the distortions introduced by outliers are in our view more credible. Overall, our decision to exclude or include independents in our dependent variables, depending on the aspects of coordination we seek to explain, is not consequential to the predictions generated by the models. Alternative operationalizations of the dependent variables yield similar results. 17. We also estimated the expected values of the number affiliated candidates from table 4.2, model 2, to assess the difference in the number of entrants observable in postcommunist countries and in countries where mixed electoral rules were introduced to replace a pure PR system. There are substantive differences. In the balanced type of mixed system (like the one in table 4.4, column 3), for example, the expected value of the number of affiliated candidates 4.28 (standard error ⫽ 0.23) for cases where the variables “postcommunist” and “reformed PR” both equal to 0. This quantity increases to 5.52 affiliated candidates in postcommunist countries (s.e. ⫽ 0.11) and increases even more, to 7.5 candidates (s.e. ⫽ 0.37), in countries where mixed systems replaced PR. 18. A notable exception is constituted by the Italian upper house elections. Most of the major parties, in fact, coalesce into broad coalitions and nominate common candidates to the SMDs of both the Chamber (dual ballot) and the Senate (single ballot). However, it is plausible that preelectoral coordination takes place in the Italian Senate elections because the existence of a separate PR tier in the election of the Chamber (where all of these parties run autonomous lists) reduces the costs of forming electoral coalitions by allowing individual parties to maintain their separate identity (see Ferrara 2004; Katz 2001).

5

Interactive Ballots: The Microfoundations of Contamination

1. In the 1999 New Zealand elections, the NZES data suggest that campaign effects may have had considerable impact on voting decisions made. Of the

156 / notes

2. 3. 4.

5.

6.

4,730 respondents who reported voting in the PR component of the election, 49% indicated that they made their decision during the campaign, 37% recounted changing their vote choice during the campaign, only 60% identified themselves as partisans, 14% said they did not lean toward any particular party, and as many as 23% voted against their stated party preference. These descriptive statistics are consistent with the high electoral volatility and voter responsiveness to campaign events found by Vowles (2002a and 2003) in his study of the 1999 and 2002 elections. In turn, this result is unsurprising, given the decrease in levels of political interest (Vowles 1994) and the general decline in partisan affiliations (Vowles 2002b) recently observed in New Zealand. Less partisan, less interested voters are also more likely to be swayed by political campaigns (see Zaller 1992; Bartels 1993). In New Zealand, consistent with the theory of contamination, parties generally “go it alone.” In 1999, all the major parties ran candidates in between 90 and 100% of the districts. The Green Party is the only exception. A candidate can simultaneously run in an SMD and be on a party list. Research on presidential coattails has shown that evaluations of presidential candidates matter to a voter’s decision making in congressional elections independently of the effects of partisanship. In particular, individual voters who seek to minimize the information costs of formulating voting decisions in congressional races maximize “cognitive efficiency” by substituting simple cues for an “extensive deliberative process” (see Mondak and McCurley 1994, 153). Their evaluation of a presidential candidate may provide an external cue upon which to base heuristic judgments of the candidates competing for a House seat. As a result, both the presidential vote and the intensity of a voter’s rating of a presidential candidate are found to influence congressional votes (see Mondak 1990; Calvert and Ferejohn 1983; Jacobson 1976). This discussion begs the question: what makes a candidate more or less “attractive,” “strong,” or “good”? What are the traits that make up “candidate quality”? The literature on campaigns and elections focuses on three broad dimensions that are believed to form the basis upon which voters evaluate candidates: competence, which includes perceptions of intelligence and experience; integrity, which pertains to judgments about a candidate’s honesty and morality; and the empathy displayed by the candidate for her constituents (Funk 1999, 702; see also Popkin 1994). In New Zealand, which are the traits that are most likely to affect a voter’s evaluation of both the candidate and the party with which she is affiliated? Prior to the introduction of its mixed system, New Zealand’s politics was highly personalized. Legislators performed extensive constituency service and spent considerable amounts of time in their home districts in the hope that personal popularity “would protect them from national swings against their party” (Anagnoson 1983, 157). By all indications, that has not changed since the system was reformed in 1993 (see Barker et al. 2001). Though the identification of the sources of high candidate evaluations is a complex and important normative question (as it speaks to which candidates a party should run), from an empirical standpoint the issue is quite simple: high quality candidates are those whom voters like best. The extent to which voters understand that the proportional tier alone determines the outcome of the election is very much in dispute. The debate has focused primarily on the causes of ticket-splitting in Germany. Jesse (1988) and

notes / 157 Schoen (1999) claim that voters often split their tickets because of their misunderstanding of the electoral rules, as they believe that both of the votes they cast affect the number of seats a party receives. Bawn (1999), Karp et al. (2002) and Johnston and Pattie (2002), however, demonstrate (on data from New Zealand, Wales, and Scotland) that ticket-splitting in MMP is consistent with high information and rational decision making. In the NZES, 60% of the respondents who cast a PR vote answered correctly to the question: “Which [of the two votes] is most important in deciding which party will get the largest number of seats in Parliament?” 7. Formally, conditional logit estimates the following: p(yi ⫽ m兩zi) ⫽

exp(zim ␤) J

兺exp(z ␤)

j⫽1

ij

where p(yi ⫽ m|zi) is the probability that voter i will choose party m given the values of the dependent variable z; zim is the value of the independent variable for respondent i and party m; ␤ is the coefficient that measures the impact of z on the probability of observing a particular outcome of the dependent variable; and zij is the value of the independent variable for respondent i and party j. 8. Though conditional logit is considered to be the appropriate method to model party choice in multi-party elections (see Alvarez and Nagler 1998; Iversen 1994; Macdonald, Rabinowitz, and Listhaug 1998), its use is problematic because it imposes the independence of irrelevant alternatives (IIA) condition. Essentially, IIA implies that the ratio of the probability of choosing one party to the probability of selecting another does not change if more parties enter the competition (or if any of the existing parties exits). Clearly, there are good reasons to believe that IIA is violated, as the entry of an additional competitor is likely to take more votes away from parties that have similar ideological orientations. To avoid imposing the IIA condition, an alternative is to estimate party choice through multinomial probit (see Alvarez and Nagler 1998). In addition, as Whitten and Palmer (1996, 255) claim, “the IIA assumption may limit our ability to draw inferences . . . about the electoral effects of party entry and exit, but it does not cast doubt over the reliability of parameter estimates for assessing the determinants of voting behavior in any particular election.” In any case, tests employing multinomial probit yield similar results. 9. Labour (center-left) received 38.7% of the PR vote, National (center-right) 30.5%, Alliance (left) 7.7%; NZ First (center) 4.3%; ACT New Zealand (neoliberal offshoot of Labour) 4.3%; the Green Party 5.2%. Labour received 49 seats and formed a minority coalition government with Alliance. No other party received more than 2.5% of the vote. 10. This variable is similar to that used to test the impact of presidential coattails in American congressional elections (see Mondak 1990). Also, its effect on the probability of voting for a given party measures the incidence of “sticky voting,” or the voters’ tendency to vote for their preferred party in both SMD and PR (Cox and Schoppa 2002). In MMP systems, where SMD votes are inconsequential to the distribution of seats, the incidence of sticky voting

158 / notes

11. 12.

13.

14.

15.

should be higher than in MMM systems, as voters have less of an incentive to vote strategically and desert hopeless candidates in SMD. We also ran tests that include variables that consider how much importance voters place on such issues (on a 1–5 scale). The results do not differ from those shown later. In other tests (not shown here), we control for the possibility that retrospective economic evaluations may affect a voter’s probability of voting for an opposition party and a government party. Retrospective evaluations should matter because popular satisfaction with the perceived outcomes of economic policies should affect the performance of incumbent parties (see Fiorina 1981). When such measures are included, the estimations presented later do not change. Furthermore, the coefficients estimated for such variables do not rise to conventional levels of statistical significance and do not improve the overall fit of the model. The likelihood ratio ␹2 of the restricted model (model 2) is subtracted from that of the unrestricted models. The resulting test statistic is distributed ␹2 with degrees of freedom that correspond to the difference in the number of parameters estimated by the two models. When the variable measuring the SMD vote is included, the results do not change. The SMD vote has a significant impact on the PR vote only in models considering voters who made their decision during the campaign and those who thought about voting for a different party. For each of the J ⫽ 6 parties, we estimate the model’s systematic component, ␲j ⫽

exp(zj ␤) J

兺exp(z ␤)

j⫽1

j

through the stochastic simulation techniques described in chapter 1. 16. Calculating predicted probabilities in this fashion is likely to result in underestimation of the impact of contamination, as it does not take into consideration how SMD candidate evaluations affect the PR vote indirectly (e.g., by improving party images). Hence, those presented in figures 5.1 and 5.2 are somewhat conservative estimates of the effects of contamination on party choice. If we allow the possibility that viewing an SMD candidate favorably may induce a voter to develop a more favorable image of the party that the candidate represents, or to perceive the distance between her stances and the party’s ideology to be smaller, the effects that it exerts on voting decisions in the PR component may be even larger. 17. For values of partisanship greater than 1, the two lines do not intersect, revealing that the conversion of partisan voters is considerably more arduous.

6

Voting from their Hearts? Contamination and Strategic Voting

1. Expected utility equals, for each candidate, the utility that the voter attaches to that candidate’s election multiplied by the probability that the candidate will actually win. 2. In 2001, elections in the proportional tier of the Upper House took place under open-list PR.

notes / 159 3. Jesse (2000) also uses the term “opportunity” to classify different patterns of voting environment, but it is different from our use of the term. In Jesse (2000), the presence of opportunity means that voters have a chance to change an election outcome if they engage in strategic voting. In this chapter, the presence of opportunity means that voters have a chance to vote for a candidate who is from their most preferred party. 4. The survey data indicates that most of these “non-opportunity” voters transfer their votes to front-runners, rather than marginal candidates. For instance, the JEDS96 survey reveals that only 18% of the Democratic Socialist Party of Japan (DSPJ) partisans residing in districts where the DSPJ did not run candidates transferred their votes to candidates from the Japan Communist Party (JCP). JCP candidates are usually marginal candidates. 5. Thermometer questions are used widely to measure voters’ preferences. For example, see Black (1978), Cain (1978), and Abramson et al. (1992). 6. Especially in mixed electoral systems that are not markedly majoritarian, as we note in chapter 4. 7. More generally, however, an increase in the number of entrants should result in greater dispersion of votes away from the top two finishers regardless of the information that is available to voters. Studies of strategic voting in Britain and other countries indicate that the number of strategic voters generally ranges from 5 to 20% of the electorate (Cain 1978; Johnston and Pattie 1991; Heath et al. 1991; Niemi, Whitten and Franklin 1992; Black 1978, Kim and Fording 2001). If we assume that the number of voters who are willing to cast a sophisticated vote is fixed, perhaps because strategic voters are those who have weak partisan affiliations, are more highly educated, or are better informed (Cox 1997, 84), a higher number of candidates should result in lower vote shares for the first- and second-placed candidates. In fact, if at least 80% of the voters behave expressively, the district vote should display higher fragmentation when sincere voters split among, say, five candidates compared to a situation where there are only two or three candidates. 8. The JEDS96 survey provides thermometer questions regarding six political parties. Some parties, such as Shinshakaito and Jiyurengo, are not included even though they placed candidates in some districts. 9. The classification is based on the M⫹1 Rule. See Reed (1990) and Cox (1997) for a more detailed discussion of the M ⫹ 1 rule. 10. As discussed in chapter 5, multinomial probit may be more appropriate to estimate models of party choice, as multinomial logit models impose the independence of irrelevant alternatives (IIA) assumption. Unlike the analysis presented in chapter 5, however, the dependent variable in this case has only three categories, and the results generated by multinomial probit may therefore be more reliable (though still not perfectly stable). 11. In any case, the parameter estimates are well within two standard errors of one another.

7

Eyeing the Future: Contamination and Legislative Behavior

1. In addition to changes in legislative output during the governing period, party membership often changes (Herron 2002a; Laver and Benoit 2003).

160 / notes 2. Scholars have paid particular attention to the link between constituency preferences and legislative behavior in the U.S. SMD system. See Hill and Hurley (1999) for a detailed review of the literature. Public preferences across districts are likely to be more heterogeneous than the party can capture in a national platform. Thus, if public opinion in the districts guides legislator behavior, legislators in SMD systems are likely to exhibit less cohesive voting than legislators without geographic constituencies in PR systems. 3. Political science research has drawn distinctions among parties, factions, and fractions (Sartori 1976). In this chapter, we use the terms faction and party interchangeably when referring to parties in parliament. 4. Dual candidacy was permitted in both Ukraine and Italy. However, a Constitutional Court decision eliminated dual candidacy for the 2002 Ukrainian elections. 5. See McKean and Scheiner (2000). 6. Katz (2001, 17) notes that the system of negative vote transfers in Italy potentially creates a conflict of interests between members of the same party who are running in plurality and PR elections within the same constituency. In fact, the probability that a PR candidate will be elected decreases if plurality candidates affiliated with her list win their SMDs (as the votes of the first runner up ⫹ 1 are subtracted from the party vote). We account for the impact of negative vote transfers when measuring the relative seat safety of PR deputies. In 2001, however, both coalitions partially avoided the scorporo by running “dummy” PR lists to which many of their SMD candidates were affiliated. The center–right coalition affiliated 426 of its 475 candidates to the Lista per l’Abolizione dello Scorporo; the center–left associated 235 SMD candidates to Paese Nuovo. 7. Measuring the party loyalty of deputies belonging to the gruppo misto is useless, given the diversity of the ideological positions within that group. 8. See Huber (1981). 9. The only Italian deputy who switched parties during the time period under consideration is deputy Rocchi, who joined Margherita from the gruppo misto. His voting behavior is considered beginning from the date when he joined Margherita. Votes in which the legislator’s absence was justified (where the data provided by the Chamber of Deputies indicate the deputy to be in missione) were excluded. 10. For both countries, the Hare quota in each constituency equals the total number of effective votes received by the parties that cleared the 4% threshold nationally divided by the number of seats available in that constituency. Ukraine employed a single national district; Italy has several regional PR districts. 11. The use of natural logarithms is consistent with the expectations about the impact of relative seat safety on party discipline. Specifically, while legislators who won their district by a wide margin (or who were ranked high on a party list) may be provided with more discretion than their colleagues who won tight plurality races (or were ranked low on their PR list), straying too much from the party line may still induce party leaders to demote them to more marginal districts or list positions or drop them from nomination altogether. Hence, the relationship between margin of victory (or list ranking) and party loyalty should not be linear. As seat safety increases, its negative marginal effect on party loyalty should gradually attenuate.

notes / 161 12. In this study, a candidate is defined as running in a “safe” district if the margin of victory is greater than the lower bound of the effective threshold for the district. This is the smallest proportion of the vote that will guarantee electoral victory under the best circumstances. In an SMD with four candidates in competition, a candidate could be guaranteed victory if he wins more than 25% of the vote, assuming that the other three candidates split the remainder evenly. The upper bound is more than 50%, if two competitors receive no votes and the remaining competitor receives all other votes (Lijphart 1994). In this example, the margin of victory would need to exceed 25% to be classified as safe. As the number of candidates in competition increases, the lower bound of the effective threshold decreases as does the margin of victory needed for the district to be labeled safe. The mean margin of victory in Ukraine for deputies in safe SMD seats was 20.23% with a standard deviation of 14.54. The mean margin of victory for deputies in unsafe SMD seats was 2.67% with a standard deviation of 2.08. Within PR, the safety of the seat is determined by list position. Legislators could be placed low or high on the party list, affecting their confidence in future electoral success and, consequently, their behavior. As the legislator’s placement climbs closer to the top of the party list, the likelihood of obtaining a seat increases, assuming that the party passes the threshold. A legislator placed low on the party list may conform closely to party norms to climb to a more advantageous list position in the next election. A safe PR position is defined as a number on the list that is guaranteed a seat in parliament if the party passes the electoral threshold in the PR component. In Ukraine, the PR threshold is 4%, a vote total that grants a party at least 9 out of the 225 seats allotted to the PR portion of the electoral system. Therefore, any candidate rated one through nine on the list is guaranteed a place in parliament as long as the party exceeds the threshold. To distinguish seat safety between SMD and PR, “safe” and “unsafe” seats will refer to SMD. The corresponding terms for PR will be “high” and “low” list positions. Consistent with Bartolini and D’Alimonte (1996), an SMD is considered safe in Italy if the margin of victory exceeded 8%; a list ranking is considered high if the ration of the party’s effective PR vote to the Hare quotas needed for that legislator to be guaranteed election in the constituency in which she was nominated is greater than or equal to one. The mean margin of victory for SMD legislators in Italy is 12.24% with a standard deviation of 9.96. For PR legislators, ratio of the party’s effective vote to the quotas needed for a seat ranges from 0.20 to 3.71, with a mean of 1.02 and standard deviation of 0.69. 13. In Ukraine, there are 144 incumbents, 188 dual candidates, and 120 deputies who changed their party affiliation at least once during the period under analysis. In Italy, there are 289 incumbents, 37 legislators elected in multiple districts, and 98 dual candidates. 14. The comparison category for Ukraine is the Communist Party of Ukraine. The comparison category for Italy is Rifondazione Comunista. 15. See Birch (2000) for an analysis of Ukrainian elections. 16. “Law on the Election of People’s Deputies of Ukraine,” Chapter 4, Article 21 (NAU 1999). 17. Independent candidates in Ukraine were nominated by “voter groups.” 18. “Law on the Election of People’s Deputies of Ukraine,” Chapter 4, Article 22 (NAU 1999).

162 / notes 19. See Smyth (1997) for a discussion of this issue in Russia. 20. The minimum number of members was lowered from 25 to 14 by the May 13, 1998 parliamentary resolution “On the Introduction of Changes in the Regulation of the Verkhovna Rada.” The text can be found in NAU (1999). 21. The number of legislators in the tables sums to 428 rather than 450 because we excluded all seats that were not occupied by the same legislator during the first three sessions of the Rada. 22. If deputies left their initial faction, their seats did not revert to the faction. This rule provided deputies with great mobility. Factional change can be motivated by many factors, among which electoral factors are important. Factions initially formed through partnerships between parties may split. The Socialist Party and Peasant Party united into a single bloc during the election to improve their odds of winning seats in PR, but the Peasant Party faction (PEA) subsequently emerged as an independent group. Factions can also serve as breeding grounds for new parties that will participate in future elections. Individual legislators may also weigh the possible electoral benefits of factional switching. Deputies with a low ranking on a party list may change factions and attempt to improve the likelihood of winning a seat in the next election. They may also be induced by financial considerations. 23. Deputies in this category are those who were initially members of Rukh, but became members of the Rukh-Kostenko branch after the party faction split into two. 24. The district magnitude of the 26 PR constituencies ranges from 1 to 11. The maximum number of candidates appearing on a local party list is equal to onethird of the district magnitude of the PR constituency, rounded to the next highest integer. If a list wins more seats than it has candidates, seats are filled by candidates who lost their SMD races. 25. Consistent with the expectations that the new electoral system would promote two-party competition in SMD, in the general elections of 1994 political parties formed broad preelectoral coalitions. Coalitions are composed of parties that run autonomous lists in the PR component of the Chamber of Deputies elections. However, such parties coalesce and jointly back candidates in the SMDs of both the Chamber and the Senate. The composition of these alliances continuously evolved in each election since 1994 and between elections. Substantial realignments and a good deal of fence jumping have progressively modified the configuration of these alliances, whose boundaries remain quite porous. 26. Once the bargaining at the coalitional level results in the assignment of a quota of safe, unwinnable, and marginal districts to each of the coalition’s components, it is the parties’ responsibility to nominate the actual, flesh-and-bone candidates. See Di Virgilio (1998 and 2002) and Cartocci (1998). 27. It goes without saying that the extent to which party leaders control the nomination of both PR and SMD candidates varies with the degree of party centralism and factionalism. In relatively unified, highly centralized parties like Forza Italia and the Northern League, the influence of party secretaries over candidate selection can certainly be anticipated to be stronger than in parties that are characterized by more divisive internal splits, like the Democratici di Sinistra and Alleanza Nazionale. In any event, the fact that it is central party organizations

notes / 163 that are vested with the power to nominate candidates for both of the components of the election is anticipated to have important repercussions on legislative behavior. Consistent with the logic outlined in this chapter, Verzichelli (1997) attributes the high legislative turnover observed each election since the reform to shifts in the internal power relations within coalitions and to centralized nominations, which allow parties the power to select candidates and run them wherever they see fit.

8

Better, Worse, or Just Different? Legislative Fractionalization in Mixed Systems

1. The exceptions are Kostadinova (2002), Moser and Scheiner (2004), and Nishikawa and Herron (2004). 2. But see Navarra and Sobbrio (2001) for evidence that this outcome was not so unexpected. 3. The two data sets display a few notable differences. Countries like Senegal and Georgia, considered “partly free” by Freedom House at various times during the 1990s, are excluded by Golder (2005) as nondemocratic. At the same time, Golder (2005) considers countries like Kyrgyzstan, which Freedom House deemed “not free” in 1995 and 2000. Differences in the values of the dependent variables are also present. Having little basis to judge which data set includes more accurate figures, we run the analysis on both and present the results alongside one another. To ensure consistency, the dependent variable employed in the analysis of Golder’s (2005) data is the ENPS that considers each legislator not affiliated with any party group to constitute one party. However, the results are identical when independents are grouped into a single category. 4. Many studies of legislative party systems analyze two-dimensional data by employing a pooled OLS model. OLS, however, has been shown to be inappropriate for data characterized by both temporal and spatial variation. While the coefficient estimates are unbiased, the standard errors estimated through OLS are likely to be grossly inaccurate in the presence of nonspherical errors (that is, heteroskedastic error processes across units or temporal dependencies within units). Given that the assumption of spherical errors is untenable in most practical applications, OLS is likely to vastly underestimate the variability of the regression coefficients (Beck and Katz 1995). Many proposed solutions exist to the problems posed by two-dimensional data. Beck and Katz (1996) recommend the estimation of a pooled model containing the lagged dependent variable on the right side to account for temporal, within-unit dependencies and the use of panel-corrected standard errors to correct for heteroskedasticity. However, as Goodrich (2004) points out, the problem with any pooled model is it imposes the restriction that the between- and within-unit effects of the independent variables are equal to one another. If this assumption—which is also characteristic of the random effects estimator and its variants—turns out to be invalid, the model will provide biased parameter estimates. Moreover, Goodrich (2004) shows that Beck and Katz’s (1996) proposed solution may produce unreasonable standard errors if we fail to include fixed effects for each of the units. Doing so, however, involves restricting our attention to the withinunit variation to evaluate the effects of the key covariates.

164 / notes 5. In similar empirical studies, for instance, Jones (1999) and Nishikawa and Herron (2004) do not settle on a single estimation technique and assess the results generated by alternative estimators. This includes pooled OLS, Between Effects, Random Effects Generalized Least Squares, a Random Effects Maximum Likelihood Estimator, and Heteroskedastic Feasible Generalized Least Squares. Tests conducted using Random Effects MLE and Heteroskedastic Feasible Generalized Least Squares produce similar results. 6. Formally, the Between Effects estimator estimates the following model: Yi ⫽ ␤0 ⫹ Xi ⫹ ␧i where Yi is the mean value of the number of legislative parties observed in country i, ␤0 is a constant and Xi is a vector of the mean values of the covariates for unit i. The presence of notable outliers and “bad” leverage points uncovered in various diagnostic tests, however, imposes the need to estimate the Between Effects models with a robust method. We present the results of the analysis performed via MM estimation, which provides efficient estimates while its high breakdown point renders it more effective at unmasking and downweighting unusual observations (Yohai 1987). Nonetheless, other robust methods produce identical results. 7. ENPS is calculated as 1/⌺si, where si, is the proportion of total seats won by party i. 8. Countries not considered by Fearon (2003), were coded based on the CIA World Factbook (as in the cases of Andorra and Vanuatu); Dominica and St. Kitts and Nevis were coded employing the figures that appear in Cox (1997). 9. Proximity is calculated as: 2⫻

冏

冏

Lt ⫺ Pt⫺1 1 ⫺ , Pt⫹1 ⫺ Pt⫺1 2

where Lt is the date of the legislative election and P are the dates of the preceding (t ⫺ 1) and succeeding (t ⫹ 1) presidential elections (see Cox 1997, 210). Also, we tested the impact of an interactive term multiplying Proximity by the Effective Number of Presidential Candidates with Golder’s (2005) data. The results are similar to those presented here. 10. In the case of Reynolds and Reilly (1997), we also consulted the updated 2002 version of their classification, which may be found on the International IDEA website at: http://www.idea.int/esd/data/world.cfm. Also, when coding mixed systems based on Massicotte and Blais’s (1999) typology, we assign Hungary to the Corrective category, while we assign Ecuador to the Superposition category. 11. Cross-validation performed by dividing the data into k ⫽ 3, k ⫽ 4, and k ⫽ 5 groups also indicates that the three classification result in in-sample predictions that have virtually the same prediction error. 12. Controlling for PR threshold does not change the results shown. Also, while it would be interesting to assess the impact exerted by electoral formula and ballot structure within mixed electoral systems, the lack of sufficient variation in such variables limits the usefulness of including such variables.

notes / 165 13. For majoritarian systems, we set the interaction between PR and district magnitude equal to 0. To facilitate a comparison between mixed and proportional systems, we set both the average district magnitude for proportional systems to 46, which is the median value for the average district magnitude of the proportional component in MMM systems. 14. The estimated within-effects models (not shown here) appear to support this view. Increasing the share of seats assigned via SMD in a given country is associated with a decrease in the number of parties, but the estimates are characterized by too great a level of uncertainty to conclusively support this claim.

9

Conclusion

1. The quote is from Leonid Kuchma’s now defunct website.

This page intentionally left blank

References

Abramson, Paul R., John H. Aldrich, Phil Paolino, and David Rohde. 1992. “ ‘Sophisticated’ Voting in the 1988 Presidential Primaries.” American Political Science Review 86: 55–69. Abramson, Paul R., John H. Aldrich, Phil Paolino, and David Rohde. 1995. “Third Party and Independent Candidates in American Politics: Wallace, Anderson, and Perot.” Political Science Quarterly 110: 349–367. Alvarez, Michael and Jonathan Nagler. 1998. “When Politics and Models Collide: Estimating Models of Multiparty Elections.” American Journal of Political Science 42: 55–96. Ames, Barry. 1995a. “Electoral Rules, Constituency Pressures and Pork Barrel: Bases of Voting in the Brazilian Congress.” Journal of Politics 57(2): 324–343. Ames, Barry. 1995b. “Electoral Strategy Under Open-List Proportional Representation.” American Journal of Political Science 39: 406–433. Amorim-Neto, Octavio and Gary W. Cox. 1997. “Electoral Institutions, Cleavage Structures, and the Number of Parties.” American Journal of Political Science 41: 149–174. Anagnoson, J. Theodore. 1983. “Home Style in New Zealand.” Legislative Studies Quarterly 8: 157–175. Arrow, Kenneth. 1951. Social Choice and Individual Values. New York: Wiley and Sons. Asahi Shimbun Senkyo Hombu. 1997. Asahi Senkyo Taikan. Tokyo: Asahi Shimbunsha. Barker, Fiona, Jonathan Boston, Stephen Levine, Elizabeth McLeay, and Nigel S. Roberts. 2001. “An Initial Assessment of the Consequences of MMP in New Zealand.” In M.S. Shugart and M.P. Wattenberg, eds., Mixed-Member Electoral Systems: The Best of Both Worlds? Oxford: Oxford University Press. Barrington, Lowell and Erik S. Herron. 2004. “One Ukraine or Many?: Regionalism in Ukraine and its Political Consequences.” Nationalities Papers 32:53–86. Bartels, Larry M. 1993. “Messages Received: The Political Impact of Media Exposure.” American Political Science Review 87: 267–285. Bartels, Larry M. 1996. “Uninformed Voters: Information Effects in Presidential Elections.” American Journal of Political Science 40: 194–230. Bartolini, Stefano, and Roberto D’Alimonte. 1996. “Plurality Competition and Party Realignment in Italy: The 1994 Elections.” European Journal of Political Research 29: 105–142. Bawn, Kathleen. 1993. “The Logic of Institutional Preferences: German Electoral Law as a Social Choice Outcome.” American Journal of Political Science 37(4): 965–989.

168 / references Bawn, Kathleen. 1999. “Voter Responses to Electoral Complexities: Ticket Splitting, Rational Voters, and Representation in the Federal Republic of Germany.” British Journal of Political Science 29: 487–505. Bawn, Kathleen and Michael Thies. 2003. “A Comparative Theory of Electoral Incentives: Representing the Unorganized under PR, Plurality, and MixedMember Electoral Systems.” Journal of Theoretical Politics 15: 5–32. Beck, Nathaniel and Jonathan N. Katz. 1995. “What to Do (and Not to Do) with Time-Series-Cross-Section Data.” American Political Science Review 83: 634–647. Beck, Nathaniel and Jonathan N. Katz. 1996. “Nuisance vs. Substance: Specifying and Estimating Time-Series—Cross-Section Models.” Political Analysis 8:1–36. Benoit, Kenneth R. 2001a. “Two Step Forward, One Steps Back: Electoral Coordination in the Hungarian Election of 1998.” Trinity College, Ireland: Unpublished Manuscript. Benoit, Kenneth R. 2001b. “Evaluating Hungary’s Mixed-Member Electoral System.” In M.S. Shugart and M.P. Wattenberg, eds., Mixed-Member Electoral Systems: The Best of Both Worlds? Oxford: Oxford University Press. Benoit, Kenneth R. 2001c. “District Magnitude, Electoral Formula, and the Number of Parties.” European Journal of Political Research 39: 203–224. Benoit, Kenneth R. 2002. “The Endogeneity Problem in Electoral Studies: A Critical Reexamination of Duverger’s Mechanical Effect.” Electoral Studies 21: 35–46. Benoit, Kenneth R. 2004. “Models of Electoral System Change.” Electoral Studies 23: 363–389. Benoit, Kenneth R. and John W. Schiemann. 2001. “Institutional Choice in New Democracies: Bargaining over Hungary’s 1989 Electoral Law.” Journal of Theoretical Politics: 13: 153–182. Birch, Sarah. 2000. Electrons and Democratization in Ukraine. New York: St. Martin’s Press. Black, Jerome. 1978. “The Multicandidate Calculus of Voting: Application to Canadian Federal Elections.” British Journal of Political Science 22: 609–638. Blais, Andre and Richard Nadeau. 1996. “Measuring Strategic Voting: A Two-Step Procedure.” Electoral Studies 15: 39–52. Bogdanor, Vernon. 1983. “Introduction.” In V. Bogdanor and D. Butler (eds.) Democracy and Elections: Electoral Systems and their Consequences. Cambridge University Press. Bogdanor, Vernon. 1983. “Conclusion.” In V. Bogdanor and D. Butler, eds., Democracy and Elections: Electoral Systems and Their Political Consequences. Cambridge: Cambridge University Press. Bowler, Shaun, David M. Farrell, and Richard S. Katz. 1999. “Party Cohesion, Party Discipline, and Parliaments.” In S. Bowler, D. Farrell, and R. Katz, eds., Party Discipline and Parliamentary Government. Columbus, OH: The Ohio State University Press. Brady, David and Jongryn Mo. 1992. “Electoral Systems and Institutional Choice: A Case Study of the 1992 Korean Elections.” Comparative Political Studies 24: 405–430. Browne, Eric C. and Dennis Patterson. 1999. “An Empirical Theory of Rational Nominating Behavior.” British Journal of Political Science 29: 259–289.

references / 169 Butler, David. 1982. “Electoral Systems.” In D. Butler, H. Penniman, and Austin Ranney, eds., Democracy at the Polls: A Comparative Study of Competitive National Elections. Washington, DC: American Enterprise Institute. Cain, Bruce. 1978. “Strategic Voting in Britain.” American Journal of Political Science 22: 639–655. Cain, Bruce, John Ferejohn, and Morris Fiorina. 1987. The Personal Vote: Constituency Service and Electoral Independence. Cambridge, MA: Harvard University Press. Calvert, Randall L. and John A. Ferejohn. 1983. “Coattail Voting in Recent Presidential Elections.” American Political Science Review 77: 407–419. Campbell, Angus. 1966. “Surge and Decline: A Study of Electoral Change.” In Angus Campbell, Philip E. Converse, Warren E. Miller, and Donald E. Stokes, eds., Elections and the Political Order. New York: Wiley. Campbell, James E. 1991. “The Presidential Surge and its Midterm Decline in Congressional Elections, 1868–1988.” Journal of Politics 53: 477–487. Carey, John M. 2003. “Discipline, Accountability, and Legislative Voting in Latin America.” Comparative Politics 35: 191–211. Carey, John M. and Matthew S. Shugart. 1995. “Incentives to Cultivate a Personal Vote: A Rank Ordering of Electoral Formulas.” Electoral Studies 14: 417–439. Cartocci, Roberto. 1998. “Omens of an Early Winter: The Proportional Vote and the Changing Italian Party System.” European Journal of Political Research 34: 35–61. Christensen, Ray and Paul E. Johnson. 1995. “Toward a Context-Rich Analysis of Electoral Systems: The Japanese Example.” American Journal of Political Science 39: 575–598. Clarke, George, Thorsten Beck, Philip Keefer, Albert Groff and Patrick Walsh. 2001. “New Tools in Comparative Political Economy: The Database of Political Institutions.” World Bank Economic Review 15: 165–176. Collie, Melissa P. and David W. Brady. 1985. “The Decline of Partisan Voting Coalitions in the House of Representatives.” In Lawrence C. Dodd and Bruce I. Oppenheimer, eds., Congress Reconsidered. Washington, DC: CQ Press. Cox, David R. 1958. Planning of Experiments. New York: Wiley. Cox, Gary W. 1987. “Electoral Equilibrium Under Alternative Voting Institutions.” American Journal of Political Science 31: 82–108. Cox, Gary W. 1990. “Centripetal and Centrifugal Incentives in Electoral Systems.” American Journal of Political Science 34: 903–935. Cox, Gary W. 1996. “Is the Single Non-Transferable Vote Superproportional? Evidence from Japan and Taiwan.” American Journal of Political Science 40: 740–755. Cox, Gary W. 1997. Making Votes Count: Strategic Coordination in the World’s Electoral Systems. Cambridge: Cambridge University Press. Cox, Gary W. and Emerson Niou. 1994. “Seat Bonuses Under the Single Nontransferable Vote System.” Comparative Politics 26: 221–236. Cox, Gary W. and Matthew S. Shugart. 1996. “Strategic Voting under PR.” Journal of Law, Economics, and Organization 12: 299–324. Cox, Karen and Len Schoppa. 2002. “Interaction Effects and Mixed-Member Electoral Systems: Theory and Evidence from Germany, Japan, and Italy.” Comparative Political Studies 35: 1027–1053.

170 / references Crisp, Brian F. and Juan Carlos Rey. 2001. “The Sources of Electoral Reform in Venezuela.” In M.S. Shugart and M.P. Wattenberg, eds., Mixed-Member Electoral Systems: The Best of Both Worlds? Oxford: Oxford University Press. D’Alimonte, Roberto. 2001. “Mixed Electoral Rules, Partisan Realignment, and Party System Change in Italy.” In M.S. Shugart and M.P. Wattenberg, eds., Mixed-Member Electoral Systems: The Best of Both Worlds? Oxford: Oxford University Press. D’Anieri, Paul, Robert Kravchuk and Taras Kuzio. 1999. Politics and Society in Ukraine. Boulder: West¨view Press. Dawisha, Karen. 2000. “The Unintended Consequences of Electoral Reform.” Working Paper, Johns Hopkins School of Advanced International Studies. Deckard, Barbara S. 1976. “Electoral Marginality and Party Loyalty in the House Roll Call Voting.” American Journal of Political Science 20(3): 469–482. Denemark, David. 2001. “Choosing MMP in New Zealand: Explaining the 1993 Electoral Reform.” In M.S. Shugart and M.P. Wattenberg, eds., MixedMember Electoral Systems: The Best of Both Worlds? Oxford: Oxford University Press. Di Virgilio, A. 1998. “Electoral Alliances, Party Identities, and Coalitional Games.” European Journal of Political Research 34: 5–33. Di Virgilio, A. 2002. “L’offerta elettorale: la politica delle alleanze si istituzionalizza.” In R. D’Alimonte and S. Bartolini, eds., Maggioritario Finalmente? La Transizione Elettorale 1994–2001. Bologna: Il Mulino. Downs, Anthony. 1957. An Economic Theory of Democracy. New York: Wiley. Duverger, Maurice. 1954. Political Parties. New York: Wiley. Duverger, Maurice. 1985. “Duverger’s Law: Thirty Years Later.” In Bernard Grofman and Arend Lijphart, eds., Electoral Laws and their Political Consequences. New York: Agathon. Elster, John. 1999. Alchemies of the Mind: Rationality and the Emotions. Cambridge: Cambridge University Press. Farhquharson, Robin. 1969. Theory of Voting. New Haven: Yale University Press. Farrell, David M. 2001. Electoral Systems: A Comparative Introduction. New York: Palgrave. Fearon, James. 2003. “Ethnic and Cultural Diversity by Country.” Journal of Economic Growth 8: 195–222. Fenno, Richard F. 1973. Congressmen and Committees. New York: Wiley. Ferejohn, John A. and Randall L. Calvert. 1984. “Presidential Coattails in Historical Perspective.” American Journal of Political Science 28: 127–146. Ferrara, Federico. 2004a. “Electoral Coordination and the Strategic Desertion of Strong Parties in Compensatory Mixed Systems with Negative Vote Transfers.” Electoral Studies 23: 391–413. Ferrara, Federico. 2004b. “Frogs, Mice and Mixed Electoral Institutions: Party Discipline in the XIV Italian Chamber of Deputies.” Journal of Legistative Studies 10: 10–31. Ferrara, Federico and Erik S. Herron. 2005. “Going It Alone? Strategic Entry Under Mixed Electoral Rules.” American Journal of Political Science 49: 16–31. Finkel, Steven. 1993. “Reexamining the ‘Minimal Effects’ Model in Recent Presidential Elections.” Journal of Politics 55: 1–21. Fiorina, Morris. 1981. Retrospective Voting in American Elections. New Haven: Yale University Press.

references / 171 Fiorina, Morris. 1984. Representatives, Roll Calls, and Constituencies. Lexington, MA: Heath. Funk, Carolyn L. 1999. “Bringing the Candidate into Models of Candidate Evaluations.” Journal of Politics 61: 700–720. Gelman, Andrew and Gary King. 1993. “Why Are American Presidential Polls so Variable When Voters are so Predictable?” British Journal of Political Science 23: 409–451. Golden, Miriam. 2003. “Electoral Connections: The Effects of the Personal Vote on Political Patronage, Bureaucracy, and Legislation in Postwar Italy.” British Journal of Political Science 33: 189–212. Golder, Matthew. 2005. “Democratic Electoral Systems Around the World, 1946–2000.” Electoral Studies 24: 103–121. Goodrich, Ben. 2004. “Problems with and Solutions for Two-Dimensional Models with Continuous Dependent Variables.” Harvard University: Unpublished paper. Grofman, Bernard and Arend Lijphart. 1985. Electoral Laws and their Political Consequences. New York: Agathon Press. Grofman, Bernard, and Arend Lijphart (eds.). 1986. Electoral Laws and Their Political Consequences. New York: Agathon Press. Gschwend, Thomas, Ron Johnston, and Charles Pattie. 2003. “Split Ticket Patterns in Mixed-Member Proportional Election Systems: Estimates and Analyses of their Spatial Variation at the German Federal Election, 1998.” British Journal of Political Science 33: 109–127. Haspel, Moshe, Thomas F. Remington, and Steven S. Smith. 1998. “Electoral Institutions and Party Cohesion in the Russian Duma.” Journal of Politics 60: 417–439. Heath, Anthony, Robert Jowell, John Curtice, Geoff Evans, Julia Field, and Sharon Witherspoon. 1991. Understanding Political Change: The British Voter 1964–1987. Oxford: Pergamon Press. Herron, Erik S. 2000. Mixed Signals: Incentive Structures and Party Strategies in Mixed Electoral Systems. Dissertation, Michigan State University. Herron, Erik S. 2002a. “Causes and Consequences of Fluid Faction Membership in Ukraine.” Europe–Asia Studies 54: 625–639. Herron, Erik S. 2002b. “Electoral Influences on Legislative Behavior in MixedMember Systems: Evidence from Ukraine’s Verkhovna Rada.” Legislative Studies Quarterly 27: 361–381. Herron, Erik S. 2002c. “Mixed Electoral Rules and Party Strategies: Responses to Incentives by Ukraine’s Rukh and Russia’s Yabloko.” Party Politics 8: 719–733. Herron, Erik S. and Misa Nishikawa. 2001. “Contamination Effects and the Number of Parties in Mixed-Superposition Electoral Systems.” Electoral Studies 20: 63–86. Hibbing, John R. and David Marsh. 1987. “Accounting for the Voting Patterns of British MPs on Free Votes.” Legislative Studies Quarterly 12(2): 275–297. Hill, Kim Quaile and Patricia Hurley. 1999. “Dyadic Representation Reappraised.” American Journal of Political Science 43: 109–137. Hillygus, D. Sunshine and Simon Jackman. 2003. “Voter Decision Making in Election 2000: Campaign Effects, Partisan Activation, and the Clinton Legacy.” American Journal of Political Science 47: 583–596.

172 / references Ho, Daniel, Kosuke Imai, Gary King, and Elizabeth A. Stuart. 2004. “Matching as Nonparametric Preprocessing for Reducing Model Dependence in Parametric Causal Inference.” Harvard University: Unpublished paper. Holland, Paul W. 1986. “Statistics and Causal Inference.” Journal of the American Statistical Association 81: 945–960. Honacker, James, Jonathan Katz and Gary King. 2002. “A Practical Statistical Model for Multiparty Electoral Data.” Political Analysis 10(1): 84–100. Huber, Peter. 1981. Robust Statistics. New York: Wiley. Hug, Simon. 2001. Altering Party Systems: Strategic Behavior and the Emergence of New Political Parties in Western Democracies. Ann Arbor: University of Michigan Press. Imai, Kosuke, Gary King, and Olivia Lau. 2004. “Zelig: Everyone’s Statistical Software.” Available online: http://gking.harvard.edu/zelig/docs/zelig.pdf Iversen, Torbèn. 1994. “Political Leadership and Representations in Western European Democracies: A Test of three Models of Voting.” American Journal of Political Science 38: 45–74. Jacobson, Gary C. 1976. “Presidential Coattails in 1972.” Public Opinion Quarterly 40: 194–200. Jesse, Eckhard. 1988. “Split-Voting in the Federal Republic of Germany.” Electoral Studies 7: 109–124. Jesse, Neal G. 1999. “Candidate Success in Multi-Member Districts: An Examination of Duverger and Cox.” Electoral Studies 18: 323–340. Jesse, Neal G. 2000. “A Sophisticated Voter Model of Preferential Electoral Systems.” In Shaun Bowler and Bernard Grofman, eds., Elections in Australia, Ireland, and Malta under the Single Transferable Vote. Ann Arbor: University of Michigan Press. Johnston, Ron and Charles Pattie. 1991. “Tactical Voting in Britain in 1983 and 1987: An Alternative Approach.” British Journal of Political Science 21: 95–108. Johnston, Ron and Charles Pattie. 2002. “Campaigning and Split-Ticket Voting In New Electoral Systems: The First MMP Elections in New Zealand, Scotland, and Wales.” Electoral Studies 21: 583–600. Jones, Mark P. 1995. Electoral Laws and the Survival of Presidential Democracies. Notre Dame: University of Notre Dame Press. Jones, Mark P. 1999. “Electoral Laws and the Effective Number of Candidates in Presidential Elections.” Journal of Politics 61: 171–184. Judge, David and Gabrielle Ilonszki. 1995. “Member-Constituency Linkages in the Hungarian Parliament.” Legislative Studies Quarterly 20: 161–176. Karp, Jeffrey A., Jack Vowles, Susan Banducci, and Todd Donovan. 2002. “Strategic Voting, Party Activity, and Candidate Effects: Testing Explanations for Split Ticket Voting in New Zealand’s New Mixed System.” Electoral Studies 21:1–22. Katz, Richard. 1995. “The 1993 Parliamentary Electoral Reform.” In C. Mershon and G. Pasquino, eds., Italian Politics: Ending the First Republic. Boulder, CO: Westview Press. Katz, Richard. 2001. “Reforming the Italian Electoral Law, 1993.” In M.S. Shugart and M.P. Wattenberg, eds., Mixed-Member Electoral Systems: The Best of Both Worlds? Oxford: Oxford University Press. Katz, Jonathan and Gary King. 1999. “A Statistical Model for Multiparty Electoral Data.” American Political Science Review 93: 15–32.

references / 173 Kim, Heemin and Richard C. Fording. 2001. “Does Tactical Voting Matter? The Political Impact of Tactical Voting in Recent British Elections.” Comparative Political Studies 34: 294–311. King, Gary. 1989. “Variance Specification in Event Count Models: From Restrictive Assumptions to a Generalized Estimator.” American Journal of Political Science 33: 762–784. King, Gary, Robert O. Keohane and Sidney Verba. 1994. Designing Social Inquiry. Princeton: Priceton University Press. King, Gary, Michael Tomz, and Jason Wittenberg. 2000. “Making the Most of Statistical Analyses: Improving Interpretation and Presentation.” American Journal of Political Science 44: 341–355. King, Gary and Langche Zeng. 2004. “When Can History be Our Guide? The Pitfalls of Counterfactual Inference.” Available online at: http://gking.harvard. edu/files/counterf.pdf Kohno, Masaru. 1997. “Voter Turnout and Strategic Ticket-Splitting Under Japan’s New Electoral Rules.” Asian Survey 37: 429–440. Kokusei Joho Center. 2001. Shugiin Senkyo Kekka. Tokyo: Kokusei Joho Center. Kostadinova, Tatiana. 2002. “Do Mixed Electoral Systems Matter? A Cross-National Analysis of Their Effects in Eastern Europe.” Electoral Studies 21: 23–34. Laakso, Markkuu and Taagepera, Rein. 1979. “Effective Number of Parties: A Measure With Application to West Europe.” Comparative Political Studies 12: 3–27. Lakeman, E. and J.D. Lambert. 1955. Voting in Democracies. London: Faber and Faber. Lancaster, Thomas and W. David Patterson. 1990. “Comparative Pork Barrel Politics: Perceptions from the West German Bundestag.” Comparative Political Studies 22: 458–477. Laver, Michael and Kenneth R. Benoit. 2003. “The Evolution of Party Systems Between Elections.” American Journal of Political Science 47: 215–233. Laver, Michael, and Kenneth A. Shepsle. 1996. Making and Breaking Governments. Cambridge: Cambridge University Press. Lijphart, Arend. 1994. Patterns of Democracy. New Haven: Yale University Press. Lipset, Seymour Martin, and Stein Rokkan. 1967. “Cleavage Structures, Party Systems, and Voter Alignments: An Introduction.” In S.M. Lipset and S. Rokkan, eds., Party Systems and Voter Alignments: Cross-National Perspectives. New York: Free Press. Loewenberg, Gerhard and Samuel C. Patterson. 1975. Comparing Legislatures. Boston: Little, Brown. Long, Scott. 1997. Regression Models for Categorical and Limited Dependent Variables. Thousand Oaks: Sage. Massicotte, Louis and Andre’ Blais. 1999. “Mixed Electoral Systems: A Conceptual and Empirical Survey.” Electoral Studies 18: 344–366. Mayhew, David. 1974. Congress: The Electoral Connection. New Haven: Yale University Press. Mayorga, Rene’ Antonio. 2001. “The Mixed-Member Proportional System and Its Consequences in Bolivia.” In M.S. Shugart and M.P. Wattenberg, eds., MixedMember Electoral Systems: The Best of Both Worlds? Oxford: Oxford University Press. McAllister, Ian and Stephen White. 2000. “Split Ticket Voting in the 1995 Russian Duma Elections.” Electoral Studies 19: 563–576.

174 / references Mac Donald, Stuart Elaine, George Rabinowitz and Ola Listhaug. 1998.“On Attempting to Rehabilitate the Proximity Model: Sometimes the Patient Just Can’t Be Helped.” Journal of Politics 60: 653–690. McFadden, Daniel. 1973. “Conditional Logit Analysis of Qualitatative Choice Behavior.” In P. Zerembka, ed., Frontiers in Econometrics. New York: Academic Press. McFaul, Michael. 1999. “Institutional Design, Uncertainty, and Path Dependency During Transition: Cases from Russia.” Constitutional Political Economy 10: 27–52. McFaul, Michael. 2001. “Explaining Party Formation and Nonformation in Russia.” Comparative Political Studies 34(10): 1159–1187. McKean, Margaret and Ethan Scheiner. 2000. “Japan’s New Electoral System: la plus ça change . . .” Electoral Studies 19: 447–477. McKelvey, Richard D. and Peter Ordeshook. 1972. “A General Theory of the Calculus of Voting.” In J.F. Herndon and J.L. Bernd, eds., Mathematical Applications in Political Science, VI. Charlottesville: University Press of Virginia. Mizusaki, Tokifumi and Hiroki Mori. 1998. “Tokuhyo Data Kara Mita Heiritsusei No Mekanizumu.” Senkyo Kenkyu 13: 50–59. Mondak, Jeffery J. 1990. “Determinants of Coattail Voting.” Political Behavior 12: 265–288. Mondak, Jeffery J. and Carl McCurley. 1994. “Cognitive Efficiency and the Congressional Vote: The Psychology of Coattail Voting.” Political Research Quarterly 47: 151–175. Monroe, Burt L. 2003. “Mixed-Member Electoral Systems: The Best of Both Worlds? Edited by Matthew Soberg Shugart and Martin P. Wattenberg.” Perspectives on Politics 1: 442–443. Moser, Robert G. 1997. “The Impact of Parliamentary Electoral Systems in Russia.” Post-Soviet Affairs 13: 284–302. Moser, Robert G. 1999. “Electoral Systems and the Number of Parties in Postcommunist States.” World Politics 51: 359–384. Moser, Robert G. 2001. Unexpected Outcomes: Electoral Systems, Political Parties, and Representation in Russia. Pittsburgh, PA: Pittsburgh University Press. Moser, Robert G. and Ethan Scheiner. 2004. “Mixed Electoral Systems and Electoral System Effects: Controlled Comparison and Cross-National Analysis.” Electoral Studies 23: 575–599. Mozaffar, Shaheen, James R. Scarritt and Glen Galaich. 2003. “Electoral Institutions, Ethnopolitical Cleavages, and Party Systems in Africa’s Emerging Democracies.” American Political Science Review 97: 379–390. Nagel, Jack. 1994. “What Political Scientists Can Learn from the 1993 Electoral Reform in New Zealand.” PS:Political Science and Politics 27: 525–529. Navarra, Pietro and Giuseppe Sobbrio. 2001. “Election Re-Running and the Nature of Constitutional Choices: the Case of Italian Election Reform.” Constitutional Political Economy 12: 31–50. Niemi, Richard G., Guy Whitten and Mark V. Franklin. 1992. “Constituency Characteristics, Individual Characterstics and Tactical Voting in the 1987 British General Election.” British Journal of Political Science 22: 229–254. Nishikawa, Misa. 2002. The Electoral Consequences of Mixed Electoral Systems: Contamination Effects and Strategic Behavior in Japan. Dissertation: Michigan State University.

references / 175 Nishikawa, Misa. 2003. “Mixed Electoral Systems with Open List Proportional Representation and Their Impact on Party Systems: The 2001 Japanese Upper House Election.” Japanese Journal of Electoral Studies 18: 12–25. Nishikawa, Misa and Erik S. Herron. 2004. “Mixed Electoral Rules’ Impact on Party Systems.” Electoral Studies 23: 753–768. Nohlen, Dieter. 1984. “Two Incompatible Principles of Representation.” In Arend Lijphart and Bernard Grofman, eds., Choosing an Electoral System. New York: Praeger. Ordeshook, Peter and Olga Shvetsova. 1994. Ethnic Heterogeneity, District Magnitude, and the Number of Parties. American Journal of Political Science 38: 100–123. Patterson, Dennis. 2003. New Rules and Old Outcomes: Assessing the Political Impact of Japan’s Double Ballot, Parallel Election System. Unpublished Manuscript. Popkin, Samuel. 1994. The Reasoning Voter. Chicago: Chicago University Press. Popper, Karl. 1984. Conjectures and Refutations: The Growth of Scientific Knowledge. London: Routledge. Rae, Douglas. 1971. The Political Consequences of Electoral Laws. New Haven: Yale University Press. Reed, Steven R. 1990. “Structure and Behavior: Extending Duverger’s Law to the Japanese Case.” British Journal of Political Science 20: 335–356. Reed, Steven R. 1999. “Strategic Voting in the 1996 Japanese General Election.” Comparative Political Studies 32: 257–270. Reed, Steven R. 2001. “Duverger’s Law Is Working in Italy.” Comparative Political Studies 34: 312–327. Reed, Steven R. and Michael F. Thies. 2001. “The Causes of Electoral Reform in Japan.” In M.S. Shugart and M.P. Wattenberg, eds., Mixed-Member Electoral Systems: The Best of Both Worlds? Oxford: Oxford University Press. Reilly, Benjamin. 2001. Democracy in Divided Societies: Electoral Engineering for Conflict Management. Cambridge: Cambridge University Press. Reiter, Jerome. 2000. “Using Statistics to Determine Causal Relationships.” American Mathematical Monthly (January): 24–32. Remington, Thomas F. and Steven S. Smith. 1996. “Political Goals, Institutional Context, and the Choice of an Electoral System: The Russian Parliamentary Election Law.” American Journal of Political Science 40: 1253–1279. Reynolds, Andrew and B. Reilly. 1997. The International IDEA Handbook of Electoral System Design. Stockholm, Sweden: International Institute for Democracy and Electoral Assistance. Riker, William H. 1980. “Implications from the Disequilibrium of Majority Rule for the Study of Institutions.” American Political Science Review 74: 423–477. Riker, William H. 1982. “The Two-Party System and Duverger’s Law: An Essay on the History of Political Science.” American Political Science Review 76: 753–766. Rogowski, Ronald and Mark A. Kayser “Majoritarian Electoral Systems and Consumer Power: Price Evidence from the OECD Countries.” American Journal of Political Science 46: 526–539. Rohde, David W. 1991. Parties and Leaders in the Postreform House. Chicago: University of Chicago Press. Rose, Richard. 1983. “Elections and Electoral Systems: Choice and Alternatives.” In V. Bogdanor and D. Butler, eds., Democracy and Elections: Electoral Systems and their Political Consequences. Cambridge: Cambridge University Press.

176 / references Rosenstone, Steven J., Roy L. Behr, and Edward H. Lazarus. 1996. Third Parties in America: Citizen Response to Major Party Failure. Princeton, NJ: Princeton University Press. Rubin, Donald B. 1974. “Estimating Causal Effects of Treatments in Randomized and Nonrandomized Studies.” Journal of Educational Psychology 66: 688–701. Rubin, Donald B. 1990. “Formal Modes of Statistical Inference for Causal Effects.” Journal of Statistical Planning and Inference 25: 279–292. Sartori, Giovanni. 1968. “Political Development and Political Engineering.” In J.D. Montgomery and A.O. Hirschman (eds.) Public Policy. Cambridge: Cambridge University Press. Sartori, Giovanni. 1976. Parties and Party Systems. New York: Cambridge University Press. Sartori, Giovanni. 1985. “The Influence of Electoral Systems: Faulty Laws or Faulty Method?” In Bernard Grofman and Arend Lijphart, eds., Electoral Laws and Their Political Consequences. New York: Agathon Press. Sartori, Giovanni. 1991. “Comparing and Miscomparing.” Journal of Theoretical Politics 3: 243–257. Sartori, Giovanni. 1994. Comparative Constitutional Engineering. New York: New York University Press. Scarrow, Susan. 2001. “Germany: The Mixed-Member System as a Political Compromise.” In M.S. Shugart and M.P. Wattenberg, eds., Mixed-Member Electoral Systems: The Best of Both Worlds? Oxford: Oxford University Press. Schoen, Harald. 1999. “Split-Ticket Voting in German Federal Elections, 1953–1990: An Example of Sophisticated Balloting?” Electoral Studies 18: 473–496. Sekhon, Jasjeet. 2004. “Quality Meets Quantity: Case Studies, Conditional Probabilities and Counterfactuals.” Perspectives on Politics 2: 281–293. Shepsle, Kenneth A. 1986. “Institutional Equilibrium and Equilibrium Institutions.” In H. Weisberg, ed., Political Science: The Science of Politics. New York: Agathon. Shepsle, Kenneth A. 1991. Models Multiparty Electoral Competition. New York: Harwood. Shugart, Matthew S. 2001. “Electoral ‘Efficiency’ and the Move to Mixed-Member Systems.” Electoral Studies 20: 173–193. Shugart, Matthew S. and John M. Carey. 1992. Presidents and Assemblies: Constitutional Design and Electoral Dynamics. Cambridge: Cambridge University Press. Shugart, Matthew S. and Rein Taagepera. 1994. “Plurality Versus Majority Elections of Presidents: A Proposal for a ‘Double Complement Rule.’ ” Comparative Political Studies 27(3): 323–348. Shugart, Matthew S. and Martin P. Wattenberg. 2001a. “Mixed-Member Electoral Systems: A Definition and Typology.” In M.S. Shugart and M.P. Wattenberg, eds., Mixed-Member Electoral Systems: The Best of Both Worlds? Oxford: Oxford University Press. Shugart, Matthew S. and Martin P. Wattenberg. 2001b. “Are Mixed-Member Systems the Best of Both Worlds?” In M.S. Shugart and M.P. Wattenberg, eds., Mixed-Member Electoral Systems: The Best of Both Worlds? Oxford: Oxford University Press.

references / 177 Shvetsova, Olga. 2002. “Gaining Legislative Control Through Strategic District Nomination: The Case of the Russian Left in 1995.” Legislative Studies Quarterly 37: 635–657. Smith, Steven S. and Thomas F. Remington. 2001. The Politics of Institutional Choice: The Formation of the Russian State Duma. Princeton: Princeton University Press. Smyth, Regina. 1997. Building Democracy by Winning Votes? Candidates and Parties in Transitional Russia. Dissertation: Duke University. Stratmann, Thomas and Martin Baur. 2002. “Plurality Rule, Proportional Representation, and the German Bundestag: How Incentives to Pork-Barrel Differ Across Electoral Systems.” American Journal of Political Science 46: 506–514. Sullivan, John L. and Eric M. Uslaner. 1978. “Congressional Behavior and Electoral Marginality.” American Journal of Political Science 22 (3): 536–554. Suzuki, Motoshi. 1999. “Shugiin Shinseidosenkyo niokeru Senryakutekitohyo to Seitoshisutemu.” Leviathan 25: 32–51. Taagepera, Rein. 1997. “Effective Number of Parties for Incomplete Data.” Electoral Studies 16: 145–151. Taagepera, Rein and Matthew S. Shugart. 1989. Seats and Votes: The Effects and Determinants of Electoral Systems. New Haven: Yale University Press. Thames, Frank C. 2001. “Legislative Voting Behavior in the Russian Duma: Understanding the Effects of Mandate.” Europe–Asia Studies 53: 869–894. Thames, Frank C. 2004. “Party and Personal Preference in Post-Soviet Legislatures.” Social Science Quarterly 85: 478–496. Tomz, Michael, Joshua A. Tucker, and Jason Wittenberg. 2002. “An Easy and Accurate Regression Model for Multiparty Electoral Data.” Political Analysis 10: 66–83. Tomz, Michael, Jason Wittenberg, and Gary King. 2003. “Clarify: Software for Interpreting and Presenting Statistical Results.” Available at ⬍ http://gking. harvard.edu/stats.shtml ⬎ Tsebelis, George. 1986. “A General Model of Tactical and Inverse Tactical Voting.” British Journal of Political Science 16: 395–404. Verzichelli, L. 2002. “Da un ceto parlamentare all’altro: il mutamento del personale legislativo italiano.” In R. D’Alimonte and S. Bartolini, eds., Maggioritario Finalmente? La Transizione Elettorale 1994–2001. Bologna: Il Mulino. Verzichelli, M. 1997. “La classe politica della transizione.” In R. D’Alimonte and S. Bartolini, eds., Maggioritario per caso. Bologna: Il Mulino. Vowles, Jack. 1994. “Dealignment and Demobilisation? Nonvoting in New Zealand, 1938–1990.” Australian Journal of Political Science 29: 96–114. Vowles, Jack. 2002a. “What Happened at the 1999 Election.” In Jack Vowles, Peter Aimer, Jeffrey Karp, Susan Banducci, Raymond Miller, and Ann Sullivan, eds., Proportional Representation on Trial: The 1999 New Zealand General Elections and the Fate of MMP. Auckland: Auckland University Press. Vowles, Jack. 2002b. “Offsetting the PR Effect? Party Mobilization and Turnout Decline in New Zealand, 1996–99.” Party Politics 8: 587–605. Vowles, Jack. 2003. Before and After: Campaign Influences on Voting Choices at the 2002 New Zealand Elections. Presented at New Zealand Political Studies Association Conference, University of Auckland, April 12–13.

178 / references Weaver, R. Kent and Bert A. Rockman. 1993. Do Institutions Matter? Government Capabilities in the United States and Abroad. Washington: Brookings Institution. White, Stephen, Richard Rose, and Ian McAllister. 1997. How Russia Votes. Chatham, NJ: Chatham House Publishers. Whitten, Guy D. and Harvey D. Palmer. 1996. “Heightening Comparativists’ Concern for Model Choice: Voting Behavior in Great Britain and the Netherlands.” American Journal of Political Science 40: 231–260. Yano Tsuneta Kinenkai. 2003. Kensei 2003-Data de Miru Kensei. Tokyo: Yano Tsuneta Kinenkai. Yohai, Victor J. 1987. “High Breakdown Point and High Efficiency Robust Estimates for Regression.” Annals of Statistics 15: 642–656. Zaller, John R. 1992. The Nature and Origins of Mass Opinion. New York: Cambridge University Press. Zorn, Christopher. 2001. “Estimating Between- and Within-Cluster Covariate Effects, with an Application to Models of International Disputes.” International Interactions 27: 433–445.

Index

Albania, 18, 23, 127 Andorra, 23, 127, 129, 164 Armenia, 23, 127 Azerbaijan, 23, 127, 146, 149 Ballot structure, 3, 15, 17, 18, 19, 25–26, 31, 50–51, 54, 55, 56, 58, 60–61, 62, 63, 107–108, 112, 127, 148, 149, 164 Berlusconi, Silvio, 146 Bolivia, 19, 23, 27, 55, 57, 127 Borda Count, 25 Britain, see United Kingdom Bulgaria, 23, 127, 146 Cameroon, 23, 127 Campaign information, 13, 68–69, 74–75, 78–79 Candidate nominations and candidate quality, 66–69, 78 and district marginality, 54, 58 in Hungary, 49, 52, 57, 58, 153, 154 in Italy, 44, 113–115, 162 in Japan, 28, 43 and legislative behavior, 101–104, 118–119, 141–142, 144 in New Zealand, 55, 154 and PR party performance, 37, 39–46, 48, 66, 76–78, 140 in Russia, 44, 153 strategic, 1, 6, 13, 49–63 in Ukraine, 107–108, 110, 112

Candidate ratings, 68, 70–72, 73–74, 75, 76–78, 156 Causal inference, 2, 4–8, 9–10, 11, 45, 47, 131, 136, 137, 138, 139, 140, 142 Chad, 23, 127 Citizen’s Duty, 90–91, 92, 93 Coattail effect in mixed electoral systems, 13, 68, 79 in the US, 68, 156, 157 Coexistence systems, 18, 23–24, 129, 130, 132 Communism, 25, 27, 57–58, 59, 128, 130, 132, 133, 145 Compensatory systems, 19–23, 26, 30, 38, 52–53, 129–131 Condorcet pairwise competition, 25 Congressional elections, US, 8, 100, 156, 157 Contamination defined, 7, 8–10, 12 and district-level party competition, 33–48, 148 and legislative behavior, 99–119 and the number of parties, 121–138 and PR performance, 39–46 and pre-electoral coordination, 13, 49–63 and voting, 65–79, 81–98 Controlled comparisons, 3–10, 33, 35, 98, 139, 140, 142 Corrective systems, 23–24, 129, 13, 164

180 / index Corruption, 28, 29–30 Cox, Gary W., 51, 84, 122, 148 Croatia, 23, 127, 146 Democracy and election rules, 25–31, 146 level of democracy, 2, 23–24, 39, 48, 127, 128, 144 Denmark, 25 Disproportionality, see Duverger’s Law District magnitude, 2, 15, 35, 36, 37, 38, 50, 52, 54, 55, 57, 58, 59, 60, 62, 88, 105, 122, 127, 132, 133, 147, 148, 154, 155, 162, 165 Dual candidacy, 14, 17, 18, 26, 30, 31, 32, 67, 86, 103, 105, 106, 108, 109, 110, 112, 115, 116, 117, 118, 149, 155, 160, 161 Duverger, Maurice, 15, 33, 34, 36 Duvergerian equilibrium, 34, 37, 46, 81, 84, 85, 98 Duverger’s Hypothesis, see Duverger’s propositions Duverger’s Law, see Duverger’s propositions Duverger’s propositions, 1, 13, 15, 46–48, 51, 81, 98, 141, 147, 148 Ecuador, 19, 23, 127, 129, 149, 164 Education, 40, 41, 71, 72, 73, 75, 92, 93, 151 Effective number of candidates, see Effective number of parties Effective number of parties, 4, 9, 10, 13, 37, 46–7, 86, 121, 128, 130, 132, 133, 135, 136, 147, 152, 164 Electoral path, 14, 102–105, 110, 112, 118, 144 Electoral rules mechanical effect of, 34, 35, 65, psychological effect of, 34, 141 study of, 1–8 Electoral volatility, 39, 47, 58, 156

Ethnic heterogeneity, measurement of, 128 and the number of parties, 129, 130, 132 Experimental research, 7, 10, 148 First-past-the-post system, see Single Member District (SMD) systems France, 16, 18, 23, 25 Freedom House, 126, 127, 163 Georgia, 23, 127, 163 Germany electoral system, 6, 18, 19, 23, 25–26, 33, 38, 39, 41, 44, 45, 46, 47, 55, 69, 127, 152, 154, 156 Weimar Republic, 27 Greece, 18 Guatemala, 23, 127, 129 Hungary, 19, 23, 49, 52, 55, 57, 58, 127, 153, 154, 164 Hybrid rules, see Mixed electoral systems Iceland, 25 Incumbency as indicator of party popularity, 40 in Italy, 18, 30, 101, 113, 115, 161 in Japan, 18, 30 and legislative behavior, 103, 104, 106–107, 109, 110, 116 and pre-electoral coordination, 54, 58, 59, 60 Independent candidates in Lithuania, 39, 150 in Russia, 39, 150, 153, 154, 155 in Ukraine, 18, 19, 39, 108, 150, 153, 154, 155, 161 Institutional design alternative approaches to, 25 in Germany, 25–26, 27 in Italy, 29–30

index / 181 in Japan, 28, 29–30 in New Zealand, 28, 31 in postcommunist countries, 26–27 in Venezuela, 29–30 Interaction effect, see Contamination Israel, 19, 23, 127 Italy and Alleanza Nazionale, 42, 44, 113, 114, 116, 162 and candidate nominations, 44, 113–115, 162 and Casa delle Libertà, la, 112 and Democrazia Cristiana, 114, 116 and dual candidacy, 18, 103, 105, 115, 116–17, 118, 160, 161 and electoral reform, 25, 27, 29, 30 and Forza Italia, 39, 42, 44, 46, 113, 114, 115, 116, 146, 162 and Girasole, il, 112 and gruppo misto, il, 105, 114, 160 and Lega Nord, 44, 113, 114, 116, 125 and Lista Dini, 44 and Lista Pannella, 44 and Margherita, la, 112, 114, 116, 160 and Movimento Sociale, 44 and party discipline, 112–118 and PDS, 42,44, 46 and Popolari, 42, 44 and pre-electoral coordination, 49, 50, 54 and Rifondazione Comunista, 44, 46, 112, 114, 161 and Ulivo, l’, 112 and Unione Democratico-Cristiana e di Centro (UDC), 114, 116 and Verdi, i, 42, 44, 46, 112 Japan and Democratic Party of Japan (DPJ), 42, 43 and dual candidacy, 18, 30, 86, 103 and electoral reform, 28, 29–30

and Hoshuto, 42, 43 and Japan Communist Party (JCP), 41, 159 and Jiyuto, 42, 43 and Liberal Democratic Party (LDP), 28, 40, 42, 82 and New Komeito, 42, 43 and SNTV, 16, 38, 82, 84, 85, 145, 149, 150, 152–153 and strategic entry in, 55 and strategic voting in, 81–98 Kuchma, Leonid, 146 Kyrgyzstan, 23, 127, 129, 146, 163 Legislative behavior and centralized nominations, 101, 102–103, 104–105, 107–108, 110–111, 112, 113–114, 115, 118, 160 and district marginality, 104, 106, 109, 160 and incumbency, 101, 103, 104, 106–107, 109, 110, 113, 115, 116, 161 in Italy, 102, 103, 105, 112–118 and majoritarian systems, 100–101, 103, 114, 118 and mixed systems, 14, 99–119, 141–142 in New Zealand, 101 and party affiliation, 102, 107, 108–110, 112, 115, 118 and personal reputation, 100, 102 and proportional systems, 99, 100–2 in the United Kingdom, 101 in the United States, 100 in Ukraine, 102, 105, 106, 107–112 Lesotho, 23, 127 Linnaeus, Carl, 15, 149 Lithuania, 21, 23, 38, 39, 40, 44, 45, 46, 47, 55, 57, 58, 127, 149, 150, 151, 153

182 / index M ⫹ 1 rule Definition, 148, 159 and SNTV systems, 84 and strategic behavior, 2, 35, 84, 85–86, 88, 90 Macedonia, 23, 127, 146 Madagascar, 18, 127, 129 Majoritarian systems, see Plurality rule, Majority-plurality, Majority runoff, Electoral rules Majority-plurality, 18, 52, 57, 153 Majority runoff, 17, 19, 24, 38, 51, 52, 55, 57, 107, 153 Mandate, 102, 106, 107, 112, 118 Marginal candidates Definition, 81, 87 and strategic voting, 87–88, 90, 91, 94–96, 97, 159 Mexico, 18, 24, 55, 127, 148, 154 Mixed electoral systems as the “best of both worlds,” 26, 122–125, 137, 142, 143, 146 and causal effects of SMD and PR, 3, 4–8, 9–10, 11, 139 compensatory, 19–22, 23, 26, 30, 38, 52, 105, 129, 130, 131 and contamination, see Contamination and controlled comparisons, see Controlled Comparisons criteria for classification, 15–24 definitions of, 17–19 and D’Hondt formula, 17, 18 diffusion of, 12, 22–24, 31–32 and Largest Remainder Hare formula, 17, 21, 149 mechanical effect of, 46–48 Mixed-Member Proportional (MMP), 19, 23, 29, 38, 47, 52, 54, 56, 57, 58, 60–61, 63, 69, 70, 101, 131, 133, 145, 157 Mixed-Member Majoritarian (MMM), 19, 21, 23, 38, 52, 53, 54, 57, 61, 63, 82, 89, 123,

125, 129, 131, 133, 134, 136, 137, 158, 165 non-compensatory, 21–22, 23, 38, 105, 130 Parallel systems, 23, 129, 130, 131 psychological effect of, 34–37, 141, 143 and seat linkage, 23, 26, 50, 52, 58, 59, 61, 132–133 single ballot, 18, 26, 50, 51, 56, 58, 59, 60, 61, 62, 63, 147, 148, 149, 155 as social laboratory, see Controlled Comparisons and spill-over effects, see Contamination Supermixed systems, see Supermixed Superposition systems, see Superposition taxonomy of, 15–32 and voting, see Voting Multimember districts, 1, 17, 18, 82, 84, 100, 147 Multinomial Logit, 70, 91, 94, 159 Multinomial Probit, 91, 94, 157, 159 Multiparty competition, see Party Systems New Zealand and Alliance Party, 70, 76, 77, 78, 157 and Association of Consumers and Taxpayers (ACT), 70, 77, 157 and electoral reform, 28, 29, 31, 47, 69 and Green Party, 70, 72 and Labour Party, 28, 70, 76–78, 157 and National Party, 70, 77, 157, and New Zealand First (NZF), 70 Niger, 18, 24, 127, Nomination strategies, see Candidate nominations and Pre-electoral coordination

index / 183 Number of parties/entrants, see Effective number of parties Open List systems, 100–101, 149, 152, 153, 158 Ordinary Least Squares (OLS), 40–41, 44–45, 60–61, 106, 152, 154, 155, 163, 164 Panama, 18, 24, 127 Parallel systems, 23, 129, 130, 131 Parliamentary systems, 8, 100, 128 Partisanship, 71, 72, 73, 75, 77, 79, 91, 103, 104, 156, 158 Party discipline, see Legislative behavior Party systems fractionalization, 5, 14, 35, 57, 121, 122, 124–125, 126, 128–129, 132–134, 137, 138, 142 institutionalization, 38, 39, 47, 57–58, 105, 144, 145, 153 local, 2, 33–48, 85 national, 2, 3, 5, 8, 14, 32, 57, 121–138, 142, 144 Philippines, 24, 127 Plurality rule and Duverger’s Law, 2, 34, 37, 46–47, 147, and legislative behavior, 104, 112, 113, 115, and the M ⫹ 1 rule, 2 and strategic entry, 37, 41, 51, 52, 54, 57, 58, 59–60, 62, 63, and strategic voting, 82, 84 Pre-electoral coordination definition of, 50 measurement of, 55–56 in Hungary, 49, 52, 55, 57, 58 in Italy, 49, 50, 54, 55, 57 in mixed systems, 49–63 in Russia, 55, 58 in Ukraine, 55, 58

Presidentialism concurrent presidential elections, 8, 122, 128, 129 and two party competition, 8, 156 presidential elections, 8, 15, 128, 129, 133, 149 Proportional representation (PR) definition of, 1 and Duverger’s Law, 2, 47–48 and Effective number of parties, 4–8 and the M ⫹ 1 rule, 2 and strategic entry, 3, 153, 155 and strategic voting, 3 Putin, Vladimir, 143 Quantities of direct substantive interest, 12 Representation, 3, 14, 17, 27, 28, 29, 30, 32, 34, 35, 52, 54, 67, 69, 119, 122–123, 126, 128, 138, 142, 144, 145 Romania, 25 Russia and AN-SF, 44 and Communist Party of the Russian Federation (KPRF), 43, 44 and Congress of Russian Communities (KRO), 43, 44, 46 and Russian Socialist Party (RSP), 43, 44 and Spiritual Heritage (DN), 43, 44 and Yabloko, 44 Scotland, 19, 24, 55, 157 Seat safety definition of, 161 measurement of, 161 for PR legislators, 104–105, 106, 109, 110, 118 for SMD legislators, 104, 106, 109, 110, 113, 114, 115, 116, 125

184 / index Seemingly Unrelated Regression (SUR), 41, 44, 152 Senegal, 18, 24, 127, 129, 148, 149, 163 Seychelles, 18, 24 Single Member District (SMD) systems, see Plurality Rule, Majority-Plurality, Majority Runoff, Electoral Rules Single Non-Transferable Vote (SNTV), 16, 17, 24, 38, 82, 84, 75, 127, 145, 149, 150, 152–153 Single Transferable Vote (STV), 145 Social cleavages, 2, 5, 6, 48, 122, 148, Sophisticated voting, see Voting South Korea, 18, 24, 55, 127, 145, 148, 154 Strategic voting, see Voting Supermixed systems, 23 Superposition systems, 23–24, 129, 130, 131, 164 Taiwan, 16, 18, 24, 127, 145, 148 Thailand, 24, 127 Threshold, 21, 26, 32, 38, 52, 54, 55, 56, 57, 58, 59, 60, 61, 62, 100, 107, 113, 123, 125, 126, 127, 160, 161, 164 Time Series Cross Section (TSCS) models, 126–127 Tunisia, 18, 24, 127 Ukraine and candidate nominations, 44 and the Communist Party of Ukraine, 109, 110 and dual candidacy, 18, 105, 108, 109–110, 112, 160, 161

and electoral reform, 18, 107–108, 146 and Hromada, 21, 22, 108 and party discipline, 107–112 and the People’s Democratic Party, 21, 22, 110 and pre-electoral coordination, 55, 58 and Rukh, 21, 22, 108, 109, 110, 112, 162 and the Social Democratic Party of Ukraine (United), 21, 22, 109 United Kingdom, 2, 5, 14, 27, 46, 101, 113, 123, 159 United States, 8, 14, 27, 46, 68, 100, 113, 148 Urbanization, 40, 41 Venezuela, 24, 29, 30, 127 Voting “anomalous,” 87, 88, 89, 90 and ideology, 71, 72, 75, 77, 158 and issues, 69, 71–72, 79 and opportunity, 83–97, 141 and party identification, 68, 71, 78 sincere/expressive, 1, 34, 35, 36, 82–83, 85, 86, 87, 88, 89, 90, 94, 97, 159 sophisticated/strategic, 1, 2, 3, 6, 13–14, 35, 51, 52, 81–98, 143, 159 split-ticket, 18, 35, 87, 156–157 “sticky,” 65, 74, 157 “straightforward,” 87, 88, 89, 90 Wales, 19, 20, 24, 55, 157 World Bank, 126, 130, 132 Zimbabwe, 18