. 45..” 4‘,th . 3 .2 ,. 0.! JUN“. R .9. .fl. 3.3.3.3? ., fl :5 bk... . . Fit; . . v. 1.1 . . e ,. $1. .3 a.» % t ‘1‘ . . 1... , .1353?! 11.. ‘ v: .15}: 3‘: f l .V- . A, puffing. .fir en? . ’.-N 3 u a)». h 3‘ v... 3.... 1 "3. lg . immwfififi .. 4:415? .... E. 5.91.. +5.)... h 1”! . . ,. FUN“ .xravthitu .. \ 1...... . .2... 2 21. . 2.3.1.1..- ..Ibuaufl... 91., I: .3. .‘l I I «hm gang"? .. I - «war... ti! o. x iiiflfliéi 5%" ....{.l-.-4.r!vh .. 3.2.. -3... 2 . :1! , .43.!!- .,I!.l.¢?t¢!‘ h... . I. 1". £3. mm“ 5.355319%:- :‘Ii,.-K- o (fin , . .. 53! x)... 5 . n... $ng t I: T 53., l :l. . . {no}! . .. .. «11‘ .111 mmfi 2 200% LIBRARY Michigan State Universny fl This isto certify that the dissertation entitled POLITICAL STRUCTURE ACROSS NATIONS: HOW THE DIMENSIONALITY OF POLITICS AFFECTS ELECTORAL BEHAVIOR presented by SHANE P. SINGH has been accepted towards fulfillment of the requirements for the _____«_DQ_<_:_t_cg_If_a.l__ degree in Political Science \LSIBLXERK RAW“ \ “m‘ Major Professor‘s Sianature v\[ is l on. \ Bate MSU is an Affirmative Action/Equal Opportunity Employer PLACE IN RETURN BOX to remove this checkout from your record. TO AVOID FINES return on or before date due. MAY BE RECALLED with earlier due date if requested. DATE DUE DATE DUE DATE DUE ALB“? 1Q fi,UV" va We”? 15‘ 25113 5(08 K:lProj/Acc&Pres/ClRC/DateDue.indd POLITICAL STRUCTURE ACROSS NATIONS: HOW THE DIMENSIONALITY OF POLITICS AFFECTS ELECTORAL BEHAVIOR By Shane P. Singh A DISSERTATION Submitted to Michigan State University in partial fulfillment of the requirmnents for the degree of DOCTOR OF PHILOSOPHY Political Science 2009 ABSTRACT POLITICAL STRUCTURE ACROSS NATIONS: HOW THE DIMENSION ALITY OF POLITICS AFFECTS ELECTORAL BEHAVIOR By Shane P. Singh This project examines the relationships among electoral institutions, political struc— ture, and party and voter behavior. While the comparative literature establishes a clear link between electoral systems and political outcomes, it ignores the role played by a country’s underlying political structure. I conceptualize political structure as the degree to which political conflict in a nation is captured by a single dimension and create a. new, quantitative measure of this concept, which I term ”political di- mensionality." I then examine the eercts of political dimensionality on political and social outcomes from a cross-national perspective. The dissertation begins by developing the method used to generate the new cross- national measure of political dimensionality. Essentially, the method determines the dimensional structure of political space through an iterative, least squares procedure. Using voter preference data, I apply the method to 81 cases from the Comparative Study of Electoral Systems and produce a quantitative measure of dimensionality for each. Next, I develop a. model predicting that majoritarian electoral systems discourage the emergence of new dimensions; because small parties are unable to exploit new issues to overcome barriers to representation, electorally entrenched parties have little incentive to take strong stances on emergent issues. Using a sophisticated empirical strategy, 1 find this indeed to be the case. I then theorize that proximity voting is less likely in political systems that do not conform to a single dimension. This is true for a host of reasons. including the simple idea that identifying the most proximate party is difficult in complex political space. I again find empirical support for this prediction. I then examine the nature of representation across various dimensional constructs and electoral institutions. I find evidence that elite-voter congruence is greater in low-dimensional systems. but. I do not find that the nature of representation varies across electoral systems. The final portion of the dissertation looks in depth at. two countries: Australia and Peru. The chapter on Australia finds that voters and parties are coherently located along a single dimension and that these locations are a strong predictor of vote choice. In Peru, the country’s proportional electoral system has created a complex underlying political space. Using data over three elections, I examine how dimensionality and voting behavior evolved within this arena. In sum, this dissertation provides a new measure of political dimensionality, a concept frerplcntly mentioned in comparative political research but rarely quantilicd. I explore the measure’s relationships with several micro and macro political factors across several countries. In the end, a clearer picture of political dimensionality across countries emerges. I find that dimensionality is dependent on a country‘s institutional configurations, while it also effects the behavior of parties and voters. Thus. it is important to consider the dimensionality of politics when conducting cross—national research. ACKNOWLEDGMENTS First and foremost, I thank Dr. William G. J acoby for guidance with this dissertation and help and advice throughout my graduate career. The idea for this dissertation first developed in 2005 during his “Scaling and Dimensional Analysis” course, and without his assistance and input this project would never have come to fruition. Also at Michigan State, I would like to thank Dr. Saundra K. Schneider and Dr. Eric C.C. Chang for guidance throughout graduate school and my dissertation writing process. Outside of Michigan State, I wish to thank Dr. Heather Stoll of the University of California, Santa Barbara and Dr. Benjamin Nyblade of the University of British Columbia, who both provided me with helpful comments on Chapter 4 of this project and generously shared their data. In addition, Dr. Jay K. Dow of the University of Missouri provided helpful assistance with Chapter 6. Finally, I wish to thank Dr. Matthew M. Singer of the University of Connecticut, who provided thorough and useful comments on this project and specific comments on Chapter 8. I would also like to acknowledge the comments given by participants at academic conferences where I presented portions of this project. Chapter 4 was presented the 2008 Meetings of the Midwest Political Science Association in Chicago, Chapter 5 was presented at the 2009 Meetings of the Southern Political Science Association in. New Orleans, Chapter 6 was presented at the 2009 Meetings of the Midwest Political Science Association in Chicago, and Chapter 7 was presented at the MSU Political Science Seminar Series in East Lansing, the 2007 Meetings of the Southern Political Science Association in New Orleans, and the 2007 Meetings of the Midwest Political Science Association in Chicago. iv TABLE OF CONTENTS List of Tables ................................................. ix List of Figures .................................................. xi 1 Introduction .................................................. 1 1.1 The Cross-National Focus ............................... 4 1.2 The Within-Country Focus ............................... 5 1.3 Conclusion ........................................... 6 2 Methodology and Measures .................................... 8 2.1 Previous Cross-National Measures of Dimensionality .......... 9 2.2 A New Measure of Dimensionality ........................ 13 2.3 The Spatial Proximity Model and Unfolding Analysis .......... 16 2.4 A Detailed Exposition of Unfolding ........................ 18 2.4.1 The Basics ................................... 18 2.4.2 Nonmetric Unidimensional Unfolding - A Newer MethOdéd 2.4.3 The Unfolding Method in Comparison to OLS ........ 23 2.5 Unfolding in Comparative Elections Research ............... 23 2.6 Conclusion .......................................... 27 3 Political Dimensionality across Nations ........................... 28 3.1 Cross-National Unfolding Results ........................ 29 3.2 Intuitive Results ...................................... 32 3.3 Counterintuitive Results and Alternative Dimensions ......... 35 3.4 What are the Dimensions? An Empirical-Substantive Exploration . . ....................................................... 38 3.5 Relationship with Previous Measures of Dimensionality ........ 42 3.6 Conclusion .......................................... 44 3.7 Appendix to Chapter 3 ................................. 47 4 Electoral Systems and the Dimensionality of Politics ................ 66 4.1 Measuring Dimensionality ...... ‘ ........................ 68 4.1.1 A New Measure of Dimensionality ................. 69 4.2 Electoral Systems and Dimensionality in Theory ............. 70 4.2.1 Previous Empirical Tests ........................ 73 4.3 A New Test of Dimensionality and Electoral Systems ......... 74 4.3.1 Accounting for other Factors ..................... 77 4.3.2 Estimation Strategies and Results ................. 82 4.3.3 An Endogeneity Problem? ....................... 85 4.4 Conclusion .......................................... 89 4.5 Appendix to Chapter 4: A Combined Measure of Socioethnic Fractionalization ......................................... 91 5 Electoral Behavior and the Dimensionality of Politics: A Cross- National Examination of Proximity Voting .......................... 92 5.1 Proximity Voting in Theory .............................. 95 5.1.1 Individual-Level Factors ......................... 96 5.1.2 Country- and Election-Specific Factors ............. 98 5.1.3 Dimensionality ................................ 100 5.2 Research Design and Methods .......................... 101 5.2.1 Constructing the Dependent Variable .............. 101 5.2.2 Measuring Dimensionality ....................... 102 5.2.3 Individual-Level Variables ....................... 103 5.2.4 Country— and Election-Level Variables ............. 104 5.2.5 Model Specification and Methodology .............. 105 5.3 Results ............................................ 109 vi 5.4 Conclusion ......................................... 113 6 Electoral Systems, the Dimensionality of Politics, and Party-Voter Correspondence across Nations ................................ 115 6.1 Party System Variance ................................ 117 6.2 Party-Voter Correspondence ........................... 118 6.2.1 Electoral Rules and Party-Voter Correspondence . . . . 119 6.2.2 Dimensionality and Party-Voter Correspondence ..... 121 6.3 Expectations, Variables, and Measurement ................ 122 6.4 Model Specification and Methodology .................... 124 6.5 Results ............................................ 128 6.6 Conclusion ......................................... 133 7 The Dimensionality of Politics and Voter Behavior in Preferential Systems: The Case of Australia ................................ 135 7.1 Australian Political Dimensionality and Voter Behavior in Theoryiéé 7.2 A New Look at Dimensionality and Voting Behavior .......... 140 7.3 What is the Underlying Dimension? ...................... 141 7.4 Determinants of Party Preferences ....................... 143 7.4.1 Shifting Preferences ........................... 147 7.5 Ideal Point and Vote Choice: A Corroborating Test .......... 148 7.6 Conclusion ......................................... 152 8 The Dimensionality of Politics and Voter Behavior under Proportional Representation: The Case of Peru .............................. 154 8.1 The 2001 and 2006 Congressional Elections in Peru ......... 156 8.2 Previous Research, Variables, and Expectations ............ 158 8.2.1 Proximity Voting in Peru ........................ 162 8.3 Methodology ........................................ 163 vii 8.4 Results ............................................ 165 8.5 Conclusion ......................................... 171 9 Conclusion ................................................ 173 9.1 So What? .......................................... 174 9.1.1 Real-World Importance ......................... 174 9.1.2 Academic Importance .......................... 175 9.2 Summary of the Project ................................ 177 9.3 Shortcomings ....................................... 181 9.4 Final Thoughts: Flatland and the Dimensionality of Politics . . . . 182 Bibliography ................................................ 184 viii LIST OF TABLES 3.1 Salience of the Left-Right Dimension and Fit ..................... 42 3.2 Correlations among Measures of Dimensionality .................. 44 3.3 Party Names .............................................. 60 3.4 Variance Explained and Party Locations ......................... 63 4.1 Summary Statistics ......................................... 82 4.2 District Magnitude and Political Dimensionality: OLS Estimates ....... 83 4.3 District Magnitude and Political Dimensionality: ZSLS Estimates ...... 88 4.4 PCA of Fractionalization Measures ............................. 91 5.1 Summary Statistics ........................................ 107 5.2 Included CSES Elections ................................... 108 5.3 Proximate Voting across Elections ............................ 112 6.1 Included CSES Elections by Model Number ..................... 125 6.2 Summary Statistics ........................................ 127 6.3 Electoral Rules and Party System Variance: Expert Placements ..... 129 6.4 Electoral Rules and Party System Variance: Unfolded Placements . . . 129 6.5 Determinants of PartyNoter Variance Correspondence ............ 130 6.6 Determinants of PartyNoter Median Correspondence ............. 130 7.1 Party Descriptions ......................................... 137 7.2 Variable Definitions and Descriptive Statistics ................... 144 7.3 Determinants of Ideal Point .................................. 146 7.4 Ideal Point and Vote Choice ................................. 149 8.1 Main Peruvian Parties in the 2001 and 2006 National Elections ..... 157 8.2 Summary Statistics ........................................ 161 8.3 Vote Choice in the 2001 Peruvian Congressional Elections ......... 168 8.4 Vote Choice in the 2006 Peruvian Congressional Elections ......... 169 8.5 Changes in Voting Probabilities, 2001 ......................... 170 8.6 Changes in Voting Probabilities, 2006 .......................... 170 LIST OF FIGURES 2.1 Two Voters on a J Scale ..................................... 17 2.2 A J Scale and Voter Preferences .............................. 19 3.1 Dot Plot of Dimensionality .................................... 31 3.2 Dot Plot of Correlations ...................................... 41 3.3 Salience of the Left-Right Dimension and Fit ...................... 43 3.4 Relationships among Measures of Dimensionality ................. 45 3.5 Asia ..................................................... 48 3.6 Asia continued .............................................. 49 3.7 Central and Eastern Europe .................................. 50 3.8 Central and Eastern Europe continued ............................ 51 3.9 Iberia .................................................... 52 3.10 Latin America ............................................. 53 3.11 Oceania ................................................. 54 3.12 Post-Soviet States ......................................... 55 3.13 Scandinavia .............................................. 56 3.14 Western Continental Europe ................................. 57 3.15 Western Continental Europe continued .......................... 58 3.16 Other Western Countries .................................... 59 4.1 Dimensionality, Electoral Institutions, and the Number of Parties ...... 71 4.2 Political Dimensionality Across Electoral Institutions ................ 75 4.3 Political Dimensionality and District Magnitude .................... 76 4.4 Political Dimensionality Over Time in New Zealand, Japan, and Peru . . 78 4.5 The Conditional Effect of Electoral Permissiveness on Political Dimensionality ................................................ 85 xi 6.1 Hypothetical Party and Voter Positions ......................... 119 6.2 The Effect of Voter Positions on Party Positions (Expert Placements) . 132 6.3 The Effect of Voter Positions on Party Positions (Unfolded Placements) . . ........................................................... 133 7.1 Party Locations on the Underlying Dimension ................... 140 7.2 Density Plot of Individual Ideal Points .......................... 141 7.3 Two Hypothetical Voters in Relation to the Six Parties ............. 148 7.4 Predicted Probability of Voting ALP or Liberal across Ideal Points . . . . 150 7.5 Predicted Probability of Voting Democrat, Green, National, or One Nation across Ideal Points ..................................... 152 8.1 Voter and Party Positions in Peru ............................. 164 xii Chapter 1 Introduction It is well—established that institutional structure guides and constrains human behav— ior. As such, it is known that the electoral system of a nation shapes the actions of voters and parties. In this dissertation project I show that electoral institutions also affect the lines of political conflict in a nation, or the dimensions along which parties and voters align. Dimensional configurations, in turn, affect the behavior of parties and voters. Thus, the actions of voters and parties are constrained by the electoral system to which they are subject, with the underlying political configuration in a nation acting as a catalyst between institutions and behavior. Politicians, pundits, journalists, and academics often invoke phrases such as “left” or “right,” to describe the location of an individual or an organization in some space that is assumed to be familiar to their audience. Moreover, individuals routinely describe themselves as “liberal,” “moderate,” or “conservative.” These phrases are heard on television, in movies, in classrooms, and in evr—iryday conver— sations. Moreover. on the internet, members of social networking websites often divulge their political leanings. When people use such terminology, they tacitly assume that the political space to which they refer is of a certain dimensional construct. While many people have a good idea about the meaning of the terms they use to locate themselves or others in space, such typologies are vague in that they do not precisely define “location” or 1? “space. Research which examines the true nature of underlying political space is valuable in that it identifies and quantifies the dimensions which are so—often referred to in. daily interactions. Much academic research either explicitly or implicitly references dimensionality and spatiality when formulating or testing theories. For example, researchers study- ing the formation and dissolution of coalitions often consider the spatial locations of parties. In addition, the veto players tradition of explaining policy change has re— cently gained much attention (see Tsebelis 2002). Studies using veto players theory must assume the policy space in which actors live, and their positions within this space. Election procedures and voting behavior have also long been explained and ex- amined with spatial theory (see, for example, Cox 1997; Downs 1957; Hotelling 1929; Rabinowitz and Macdonald 1989). According to Ordeshook (1997), spatial construc- tions in which issues are represented as lines, candidates (or parties) and voters are represented as points on these lines, and where voters make decisions based on their dish-nice from candidates, are well—accepted as representations of elections. Hinich and Munger (.1998) note spatial theory‘s crucial role in elucidating the effects of electoral systems on democratic outcomes. In turn, they stress the importance of continued empirical studies grounded in the spatial theory of voting. Green and Shapiro (1994) assert that theoretical work of spatial theory in relation to voting behavior has outpaced its empirical counterpart. That is, despite the heavy reliance on dimensionality in political science, much of the previous empirical and theoretical work either assumes the dimensionality of a given space, or preselects a host of issue stances to be analyzed with a data reduction technique. In the words of Laver and Hunt (1992, 22-23), “While the theorist can wave a magic wand and (16‘1“ e a POIiCY system t0 be 0110-, two-, or three—(iin‘iensional, the empirical. analyst to dealing with a particular case is left with no hint as to how to determine the actual dimensionality of the space in question.” Thus, there is a clear need for empirical methods that assess dimensionality. Existing empirical approaches take two approaches. First, they ask experts or mass survey respondents to assess dimensionality. Second, they take a large data set and attempt to reduce it, testing whether the observed variation in the data is dependent on some latent dimension or dimensions. Following the call for agnostic empirical dimensional research, this dissertation describes and (plantifies the dimensionality of conflict within nations without pre- supposing anything about a nation’s underlying structure. I conceptualize dimen- sionality as the amount of variance in party and voter locations in a nation captured by a single dimension. This dimension may be thought of as the political “super dimension” of Gabel and Huber (2000), which constrains party positions over several issues. The method I use measure this concept is unfolding analysis, which is based on an underlying geometric model of spatial proximity. Unfolding recovers the dimensions of the space it is applied to and locates stimuli (in this case, parties) and individuals (in this case, survey respondents) along these dimensions with interval-level mea— sures. Associated with these results are statistics indicating the “goodness of fit.” of the model, or how much variance in voter preferences the recovered dimensional construct explains. Using these statistics, I derive the measure of dimensionality in- troduced in this research, which I term “political dimensionality.” The methodology used to produce this measure will be developed and explored in detail in Chapters 2 and 3. 1.1 The Cross-National Focus The cross—national portion of this project grounds itself in previous theoretical work, empirically analyzing predictions and gauging their validity. In doing so, specific at- tention is given to electoral institutions, party systems, and voter behavior. In the end, a clearer picture of the interplay between electoral institutions, the (1111101181011- ality of politics, and various political outcomes is provided. In Chapters 4, 5, and 6 I examine political dimensionality and its relationship with other political variables across several countries. Chapter 4 explores the re— lationship between electoral institutions and political dimensionality. In line with previous theory, I predict that restrictive electoral institutions - those with hurdles to representation for small parties - lead to simple dimensional constructs. This is because they encourage entrenched parties to ignore emerging issue dimensions as they need not worry about small parties riding them to power. Alternatively, in permissive systems - those in which small parties can gain par- liamentary representation with a small fraction of the vote - political dimensionality should be high. This is because major parties cannot simply ignore emerging issue dimensions out of fear of losing seats to niche competitors. Instead, they must en— gage such parties on up and coming issues, thereby bringing new dimensions to the political forefront. Empirical tests provide evidence for these predictions. In fact, politics in nearly all countries that employ restrictive electoral systems conform well to a single dimension. Likewise, most nations with permissive electoral setups tend to have non-unidimensional political space. Moving dimensionality to the right side of the equation, in Chapter 5 I examine how dimensional configurations and other individual- and election-level factors affect electoral behavior. Under the assumption that proximity voting is less likely in countries with complex political space, I find such behavior to occur less in countries where political variation does not arise from one dimension. In addition, I find that strong party identity and political efficacy have a positive relationship with proximity voting, while party system fractionalization and compulsory voting rules relate negatively to proximity voting. These findings shed light on the institutional and individual bases for proximity voting and add to the general understanding of the nature of voting behavior. In Chapter 6 I‘expand upon the cross-national examination of representation, examining how it varies with the dimensionality of politics and electoral institu- tions. I expect that party-voter correspondence will be high in nations with simple dimensional constructs. Alternatively, in countries where political space is not well— captured unidimensionally, representatives are less likely to accurately reflect the desires of constituents. To test these expectations, I examine how well party po- sitions mirror both the median and spread of voter preferences, conditional on the electoral institutions and political dimensionality of nations. Using data from a wide sample of nations and the new measure of dimensionality, I find that the positions of parties correspond more closely to those of voters in countries with low—dimensional political space, whereas electoral systems play a smaller role in the nature of political representation. 1.2 The Within-Country Focus I also conduct two country-level studies in which the insights derived from the cross~ country analyses are applied to individual nations. This allows me to examine nation-level idiosyncracies that cannot be captured outside of the error term of large-n statistical models. In Chapter 7, I examine the dimensionality of politics and voter prefertmces in Australia. Evidence from the unfolding model provides that Australian parties and voters are organized along a unidimensional continuum. Individual-level variables, derived from previous theory, are then used to predict voter ideal points on this O1 continuum. From the ideal points, voter preferences over each party are ascertained. Because this approach allows for a full examination of voter preference orderings. it is important to the study of voting behavior and representation under preferential elec- toral institutions. Moreover, the findings verify the intuition from the cross-national analyses; due to Australia’s majoritarian electoral system, political variation tends to arise from one dimension and voters generally follow the proximity logic. In Chapter 8 I conduct an in depth analysis of Peruvian voter behavior in the 2001 and 2006 congressional elections. Because political variation in Peru does not arise from a single dimension, voters are less likely to correctly identify the party closest to them in political space. As such, I expect that proximity voting should be minimal in Peru. Using an alternative-specific multinomial probit model, I find that proximity voting did not occur in Peru in 2001. Moreover, in 2006, while proximity considerations did enter the voting calculus, they played only a minor role as compared to the effects of other factors. 1 .3 Conclusion The aim of this dissertation is to fill a gap in previous comparative political research, while introducing new data to the discipline. While dimensionality as a concept is commonly referenced, it is rarely quantified or examined in relation to other political variables. In this work I create a new cross-national measure of dimensionality and explore its relationships with several micro and macro political factors across several countries. In the end, a clearer picture of political dimensionality across countries emerges. I find that dimensionality is dependent on a country’s institutional configurations, while it also effects the behavior of parties and voters. Thus, it is important to consider the dimensionality of politics when conducting cross—national research. I also apply my ”(ms-national findings at the country level, looking in depth at Australia and Peru. In Australia, I finds that voters and parties are coherently located along a single dimension and that. these locations are a strong predictor of vote choice. In Peru, I find that the country’s proportional electoral system has created a complex underlying political space. Using data over two elections, I examine how dimensionality and voting behavior interact within this arena. In sum, this dissertation provides a new measure of dimensionality, a concept fre— quently mentioned by politicians, pundits, journalists, and academics alike. I show that electoral institutions affect the lines of political conflict in a nation, or the di- mensions along which parties and voters align. Dimensional configurations. in turn, affect the behavior of parties and voters. Thus, the actions of voters and parties are constrained by the electoral system to which they are subject, with the underly— ing political configuration in a nation acting as a catalyst between institutions and behavior. Chapter 2 Methodology and Measures This research uses a new approach to measure dimensionality cross-nationally. Rather than using expert surveys or party manifestos, I examine voter evaluations of par- ties. A data reduction technique, unfolding analysis, measures the dimensionality of the space from which these evaluations are generated. This approach is useful in that it provides an objective indicator of dimensionality, as well as party and voter locations along the recovered dimensions. Underlying political space is measure in many ways. For example, social cleav- ages are often conceptualized and measured under the label of “socioethnic hetero- geneity.” However, such cleavages may or may not prove salient at the national level, depending on the incentives provided to important political actors. Exam- ining dimensionality, rather than heterogeneity, allows researchers to measure how socioethnic structure manifests itself as important national-level (:lirnensions. To measure. dimensionality, previous research generally focuses on the number of ide- ological or issue dimensions in a country. More specifically, it. examines the. issues or ideological cleavages that are important to the various national-level political parties. The measure developed here departs from this approach to measuring dimen- sionality by gauging the makeup of the political space in which parties and voters are. located. If all parties in a given system choose to squeeze the salient issue di- mensions into one “super dimension.” the measure will reflect. this. If a small, niche party brings to government an emerging issue dimension that proves salient to the point that entrenched parties wish the compete along it, the measure will reflect this as well. Such an approach to measuring dimensionality is important in that it gauges political space after political parties reduce it and stake out pesitions within. T hus, the parliamentary behavior of parties is captured by this measure; any decision to ignore or absorb an issue or ideological divide is accounted for. The new measure also adds to the existing literature by expanding the number of countries beyond previous measures and focusing on recent years. With 42 countries covered, it spans the widest range of nations to date. Using data from 1996—2006, the measure focuses on a relatively recent time period, as opposed to previous cross-national measures which generally expire in the late 1990s. 2.1 Previous Cross-National Measures of Dimen- sionality “Dinnmsionality” as a concept is measured in various ways throughout previous lit- erature. While some measures count salient political issues, others gauge the number of ideological dimensions purported to arise from such issues. Expert opinions, party manifestos, and citizen evaluations are all used in various measures of dimensionality. and each measure covers different countries and time periods. Ethnic, social, religious, and linguistic heterogeneity are frequently studied and used as dependent or independent variables (see, for example, Alesina, Dcvleeschauwer, Easterly, Kurlat, and Wacziarg 2003). Socioethnic heterogeneity has the potential to create cleavages, introduce new issues into the political dialogue, and shape party systems, but its likelihood of doing so is moderated by societal factors, such as elec- toral institutions (Amorim Neto and Cox 1997; Geys 2006; Ordeshook and Shvetsova 1994). As noted by Laver and Hunt (1992, 17—18), while certain religious or ethnic divides may define the political behavior of individuals, these cleavages may or may not prove to be salient at the legislative level. Moreover, some cleavages may sim- ply not be important enough to gain national-level political attention, or elites may exclude them from of national politics out of self-interest (Taagepera 1999, 545). Measures of dimensionality gauge the number of salient dimensions of political conflict within nations. rather than the amount of conflicting societal groups. One such measure, what Stoll (2009) terms as raw dimensionality, counts the number of salient ideological conflicts in a nation. This she contrasts with effective dimension— ality. which counts only ideological conflicts that may be considered orthogonal to one another.1 The most well-known measure of issue dimensionality is that of Lijphart (1999), which Stoll (2009) classifies as efi‘ective. Lijphart assesses dimensionality based on his subjective, but “straightforward and uncontroversial” judgements of the salience of seven issue dimensions across 36 nations, providing a single measure for the years 1945-1996 (79). As noted by Taagepera (1999), Lijphart’s method of determin- ing dimensionality is highly subjective and therefore “less than satisfactory” (546). To improve on the measurement of dimensionality, recent work has turned to the Comparative Manifestos Project (CMP) (Budge, Klingemann, Volkens, Bara, and Tanenbaum 2001). The CMP is a reliable and widely-used data source, which hand- codes party manifestos. One measure of dimensionality arising from the CMP is that of Richman (2005), which simply gauges the portion of the party manifestoes that are coded in the 1Orthogonality implies that movement along one dimension causes no movement along another. Thus, in two dimensions, orthogonal axes are situated at 900 angles to each other. Terming a measure efiectwe does not presuppose that is uses scaling analyses to empirically determine orthogonality, though it certainly may. 10 “left—right” category. Nyblade (2004) creates more involved measures from CMP data. He first examines 43 CMP issue categories in 17 West European countries over the years 1945 to 1998. Applying the common Laakso and Taagepera (1979) effective parties formula (EN EP)2 to weight the issues by their salience, Nyblade creates a measure of the effective number of issues (EN 1). He then creates a measure of the effective number of issue dimensions (ENID) by weighting issues by the vote share of parties with analyzed manifestos and their similarity, in addition to salience. Thus, when two or more parties consider the same issue important, dimensionality decreases. The similarity measures are derived by pooling all countries and election years and calculating the angular separation3 between all issue pairs. Stoll (2009) classifies the ENID measure as efi’ectruc because it collapses raw dimensions into effective dimensions. Stoll herself (2005; 2009), also using CMP data, creates various measures of raw dimensionality in 24 Western countries from 1945-1998. These measures can be thought of as indices of ideological dimensionality. The measure used in her 2009 work is based 011 the salience of seven ideological dimensions. To gauge salience, Stoll looks at the proportion of space each party manifesto in a given nation devotes to each dimension, and takes an average across manifestos. She then applies Molinar’s (1991) formula4 to weight each conflict by its salience. Other studies use scaling methods to count dimensions. For instance, McAllister and Studlar (1995) use principal components analysis to examine the number of opinion dimensions among elites and voters in Australia, recovering about two salient dimensions for each. Similarly, Jackman (1998) employs a factor analysis to uncover ZENEP is measured as Tij, where “j is the proportion of votes or seats .' e“. 9:1 J obtained by the jth party 3Angular separations are essentially correlations that are bound between 0 and 1 rather than -1 and 1. r 4Molinar’s formula was originally created as an alternative to the Laakso and Taagepera (1979) effective number of parties index. 11 the structure of elite and mass preferences in Australia, finding four cleavages in both groups. Ray and N arud (2000) find a two-dimensional representation of Norwegian political space using factor analysis. Cross-nationally, Moreno (1999) uses the World Values Survey (WVS) and prin- cipal components analysis to comparatively analyze dimensional structure. He finds political space in all nations to be of a 2- or 3-dimensional structure, though in each area of the world (Latin America, Post-Communist and Eastern Europe, and the “First \Vorld”) the dimensions are substantively different. Warwick (2002) performs a principal components analysis of party positions from Laver and Hunt’s (1992) expert data on 16 West European nations, finding they arise from three dimensions, 1 which account for 89% of the variance (105). He also applies principal components analysis to party positions from the CMP, this time finding that three dimensions capture only 14.9% of the variance ( 111). In addition to counting dimensions, a substantial amount of research locates parties (but not voters) along presupposed dimensions with the use of expert surveys (Castles and Mair 1984; Dodd 1976; Huber and Inglehart 1995; Laver and Hunt 1992; \Varwick 2005). Though based on subjective opinion, the validity of this approach is high in that the measures produced relate strongly to one another. Gabel and Huber (2000, 98) find that the expert party locations of Huber and Inglehart and Castles and Mair correlate very highly with the locations reported by party supporters in the Eurobarometer and World Values Survey (from .88 to .94). Warwick (2005) employs a. new, original expert survey of 13 West European Nations. Respondents are asked to identify the salience of up to four dimensions in each nation and to place parties along them. The respondents generally identify three important dimensions, with the left-right dimension proving to be the most salient in all but Austria, Italy, and the Netherlands (34). IDiscouragingly, several of these measures make a priori assumptions about the 12 importance of specific issues, and whether or not to include them in expert question- ing. To counteract this, Gabel and Huber (2000, 95) develop a “vanilla method.” which includes all issue categories of the CMP in a factor analysis, predicting party positions via regression scoring. Similar approaches to placing parties are used in the work of Klingemann (1995) and Budge, Robertson, and Hearl (1987). However, even the vanilla method is influenced by the CMP‘s decisions to include and exclude certain issue categories in its coding scheme. In addition, as n‘ianifestos do not nec- essarily correspond to the true parliamentary behavior of parties, such approaches are sensitive to any distortions found in the documents. 2.2 A New Measure of Dimensionality To gauge dimensionality, I simply examine the n'iakeup of the political structure in which parties and voters live. I am not concerned with social cleavages, issues, or ideological dimensions themselves, but the ability of a particular, dimensionally reduced structure to capture the political variation in party and voter positions. 1 tr 1111 the new measure “political dimensionality.” Put simply, if variation in political outcomes is well-captured by a single dimension, political dimensionality is low. Alternatively, in a country in which political variation does not depend strongly on a. sole underlying continuum, political dimensionality is high. The measure of political dimensionality has numerous practical and theoretical strengths. First, it covers more countries than any previous measure of dimensional- ity and does so across a wider range of developed and developing polities. l\/Ioreover. it" f0(Inses on recent years and can be applied to any data set in which respondents rate political stimuli. The measure also locates parties and voters in the recovered SpaC-O. This provides valuable information on the relative locations of representatives ‘11 ( 1‘1 the represented. By relying on voter evaluations of parties, I avoid the subjectivity inherent in 13 measures created from party manifestos, expert surveys, and voters’ opinions of dimensionality. Such techniques for determining dimensionality must presuppose the importance of some latent cleavages. Manifestos are coded according to some researcher-defined ideological divides and, when asking survey rr-‘spondents to gauge the salience of cleavages, the survey designer must choose which clea ages to include. The measure created here, on the other hand, does not presuppose the importance or insignificance of any societal divide. Respondents are also more capable of providing evaluations of parties than plac- ing parties at ideal points along assumed dimensions, and thus the response rate for this type of question is relatively high (Narud and Oscarsson 1999, 12). Further- more, evaluative responses have more comparability than “left—right” responses over time and across nations (Bobbie 1996). Finally, there is less research bias in evalu— ative response measures, which do not presuppose the importance or obscureness of any given issue or dimension (Narud and Oscarsson 1999, 12-13). The new measure is also useful in that it accounts for the public and parliamen— tary behavior of political parties. It is well-known that parties shape their agendas ac-(«ering to political ambition. That is, they selectively engage and ignore various iss ue dimensions based on a calculus of their electoral prospects (Budge et a1. 1987; Cantillon 2001; Meguid 2005; Przeworski and Sprague 1986; Taagepera 1999). This measure of dimensionality gauges the space parties compete in after they choose which issues to actively compete along on which issues to ignore or absorb based on Various political incentives. Accordingly, it is sensitive to any such bel'iavior by DOIitical parties that may shape or reduce political space. Previous measures, on the other hand, capture either the number of important issues in a country or the amount of issues or ideological cleavages deemed impor- t‘1, _ . . . . . . C I"; by polltical parties. Manifestos do not necessarily reveal true party strategics 14 or relate highly to observed parliamentary behavior.5 Moreover, they are not neces- sarily updated each time parties adopt a new parliamentary strategy, and thus are not ideal for gauging the actual space in which parties locate. Voter evaluations, instead, change with the observed behavior of political parties. In the United States for example, the party system is dominated by the De- mocrats and the Republicans, with the Greens capturing very modest electoral sup- port among environmentally concerned and far-left voters. Imagine such a voter rates the Greens 9, the Democrats 5, and the Republicans 2. If the Democrats see value in taking strong positions on environmental issues and introducing bills ac— cordingly, this voter may change her opinion of the party, perhaps then assigning a 7' to the Democrats rather than a 5. Only the measure developed here is sensitive to such shifts. In determining dimensionality, the unfolding procedure also recovers interval- 1<1vel locations of parties and voters in space, which can be seen as an alternative to those produced by expert surveys or factor/principal components analysis; Hinich. Khrnelko, and Ordeshook (1999) and Carkoglu and Hinich (2006. 371) note that principal components analysis and factor analysis are inferior for the study of the structure of voter preferences because they are not based on a formal mathematical model of preference. The unfolding method used here, on the other hand, is di— r< 1c tly based on the Downsian proximity model of voting. Other studies that apply unfolding cross—nationally (Dow 2001; Listhaug, Macdonald, and Rabinowitz 1990; N ar 11d and Oscarsson 1999; Rabinowitz, Macdonald, and Listhaug 1991) focus on "1‘ ":18 uring party and voter positions and place less emphasis on cross-national di- 1110 nsionality. Moreover, these studies generally cover very few nations. \ 5 t i Budge and Farlie (1983), for example, note that parties downplay diver g1 11g posi- .' OHS on politic al 1ssues in their official manifestos but tend to emphasize diffcic 111c cs In the media. 2.3 The Spatial Proximity Model and Unfolding Analysis Jacoby (1991, 27) conceptualizes dimensionality as the number of salient sources of variation among objects. As the left—right ideological dimension forces a range of issues to logically and consistently relate to one another (Converse 1964; Gabel and Huber 2000), this number is often very low. The method of gauging dimensionality employed here, a spatial proximity model estimated with unfolding analysis, deter— mines how much variation in party and voter positions is due to a single dimension. Spatial models in which parties and voters are represented as points on dimen- sions, and where voters make decisions based on their distance from candidates, are well—accepted as representations of elections (Ordeshook 1997). The spatial prox- imity model, in the tradition of Hotelling (1929) and Downs (1957), assumes that voters choose the party or candidate closest to them in some 71-dimensional space. 6 it is the most commonly used model for ex— 21 ‘- VVhile this model has its opponents, arnining vote choice in political science, economics, and related fields. Figure depicts a simple spatial model of four parties and two voters. By proximity theory Voter 1 will prefer Party B and Voter 2 will prefer Party D. Unfolding analysis determines how well a given set of preferences conform to Spatial proximity theory. It is important to note that the actual voting behavior of survey respondents is not considered with unfolding analysis. Rather, the corre- SI’Olldence of their survey responses to a proximity model of preference is gauged. The only assumption made is that individuals tend to express more positive affect for I>arties with ideal points similar to their own. Developed in one dimension by Coombs (1950; 1964) and later generalized to ‘T 3111 1 Most notable is Directional Theory (see for example, Macdonald, Rabinowitz, i n: Listhaug 1998), whlch pos1ts that voters, rather than selectmgothe most prox- as ate .party, choose parties that take extreme pos1t10ns in a d1rect1on they prefer. 811111mg that they remain within some “region of acceptability.” 16 several dimensions by Bennett and Hays (1960), the unfolding model employed here is a nonmetric and unidimensional model. The model locates stimuli (in this case. parties) and individuals (in this case, survey respondents) along the recovered dimen- sion with interval-level values. Associated with these results are statistics indicating the “goodness of fit” of the model, or how much variance in voter preferences it ex1'1lains. As the proportion of voter preferences that are single-peaked increases,7 so does the goodness of fit.8 In sum, the unfolding model has three important applications: 1) it determines whether there is a common dimension (or dimensions) underlying individuals’ pref— erence orderings, 2) it gauges the extent to which preferences are “single-peaked” along the recovered dimension(s), and 3) it recovers metric information about both individuals and stimuli in the recovered space (Melver and Carmines 1981, 71-75). I Volter I Viter I .1. .1. .1... .1. A B C D Figure 2.1: Two Voters on a .1 Scale r 7 “Single—peaked” preferences are transitive. If voter A prefers X to Y and Y to 4.1 and he also prefers X to Z, his preferences are transitive. 8Unfolding analysis, as developed by Coombs (1964; 1950) assures that, with scalable data, preferences are single—peaked along the recovered dimension. Note, 1OV’Vever, that unidimensionality does not alone guarantee single-peakedness; Niemi and Weisberg (1974) show that Guttman scales can provide a perfect unidimensional solution even in the presence of multiple-peaked preferences. 17 2.4 A Detailed Exposition of Unfolding 2.4.1 The Basics Coombs’ original unidimensional unfolding method was popularized in his 1964 book, but first introduced in his 1950 work. In this model each individual‘s preference ordering over a number of stimuli (for example, parties) is called an 1 scale. The unfolding model asks whether there is a common latent attribute, a J scale (a joint scale), underlying individuals’ preference orderings. If individuals employ a common, unidimensional criterion in evaluating stimuli, the various I scales will be consistent with a single J scale. Stated differently, if all preferences are single-peaked along an underlying J scale, individuals can be aligned along a unique, single dimension (McIver and Carmines 1981). Figure 2.1 depicts a single J scale, along which three hypothetical voters maintain I scales. Voter 1’s 1 scale is BCAD and voter 2’s is DCBA. Because these individuals can be placed along the dimension, their 1 scales are consistent with the underlying .7 scale. If the axis is “folded” at voter 1’s ideal point, the preference ordering BC AD is recovered. If the axis is “folded” at voter 2’s ideal point, the order DCBA is recovered. However, imagine an individual, voter 3, with I scale ADBC. There is no location along the underlying J scale which corresponds with this I Scale. Figure 2.2 further illustrates this, as well as unfolding’s peakedness criterion: any preference order that can be generated from the common underlying continuum can be depicted with a S‘111.£§111r:~1111ake(l curve. Single-peaked curves, as defined by Riker (1982, 124). can be: - always upward sloping - always downward sloping - sloping upward to a point and then sloping downward 18 o sloping upward to a plateau and then sloping downward 0 horizontal and then downward sloping, or o upward sloping and then horizontal. The preference curves for both voter 1 and voter 2, which are single—peaked, reach a. single maximum at their ideal points and then decline monotonically. Alternatively, voter 3’s curve, which “peaks” at both sides of the x-axis, is not single—peaked. Therefore, while the preference profiles of voter 1 and voter 2 can be. described with reference to a single underlying dimension. the profile of voter 3 does not fit on the unidimensional continuum. Voter 3 may be evaluating the parties using some different criteria or appealing to separate dimensions when formulating his preferences.9 \1 oter . 1 Vote . a) \ r 2 o C: 93 % . /O/\ 9: ,x‘ \ o C / ’\_\ ‘21—: o E - \ IVbtén', I 3 I . o o ' 0 Voter Voter I 1 I 2 l I I“ C I Party Party Party Party B C D > Figure 2.2: A J Scale and Voter Preferences \ 9 _ 1 1( . He may also be reporting his preferences untruthfully, thus adding to the random ’ 139. inherent in survey data. 19 2.4.2 Nonmetric Unidimensional Unfolding — A Newer Method Newer, con‘iputer-based unfolding methods, built on the work of Coombs, have de- veloped over the past four decades. Here I outline the nonmetric unidimensional unfolding procedure used in this work, which is based on optimal scaling (Jacoby 1999; Young 1981) and Poole’s (1984) conditional global minimum (CGM) algo- rithm. This method was developed by William G. Jacoby, and I thank him for assistance with the information put forth in this section. The Metric Procedure This procedure begins with an n by k matrix of preference data, A. Smaller values 111ean more preference for a stimuli, and larger values mean less: A is a dissimilarities matrix. Each entry in A, 6,- j1 depicts i’s preference for stimulus j. The unfolding procedure produces a matrix D, with values (in. There is an error when (5,], g (SI-q but dip > dig (subject 75 prefers stimulus p to q but the unfolding results say otherwise). Thus, the objective is to minimize the differences l’1etween the. 6-,-’s and 7.} the (l ,1 j’s, or the sum of squared errors, shown in Equation 2.1. Z 2013‘— dijlg (‘2-1) Poole’s (1984) CGM algorithm provides a method for minimizing Equation 2.1. I 11 ragine that each individual’s ideal point has A? vectors attached to it (one for each of t. he, stimuli) and the length of each vector is equal to the subject’s preference for that, stimulus. Since this is a unidimensional situation, each vector can point to the. left or the right of the ideal point. Poole proved that when each vector is pointed tovVard the correct location of the stimulus j, the sum of squared errors is minimized. II‘O find optimal locations for the stimulus points, each stimuli is 111oved along t1 11 x - . . . . 9 ( 11me11s10n, and the Ideal pomts are held fixed. There are n + 1 intervals on the 20 dimension (the areas between the individuals’ ideal points and the two outer areas). At each stimulus location, the variance in the terminal points of the attached vectors is calculated. The final location is the one associated with the lowest variance value. Next. the procedure is reversed, and the stimulus points are held fixed10 while ideal points are tried in each of the k + 1 intervals along the dimension. Incorporating Nonmetric Data If a degree of preference is expressed rather than a precise amount, data are consid- ered nonmetric. For example, if, on a 10 point scale, survey respondent A rates the “Progressives” with a 9 and the “Regressives” with a 1, we know that she prefers the former. And, if survey respondent B rates the “Progressives” with a 6 and the “Regressives” with a 5, we again know that he prefers the “Progressives.” How- ever, we cannot make the assumption that respondent A likes the “Progressives” more than respondent B, as these two individuals may be using different criteria for evaluation. With data such as “US. dollars spent” or “votes received,” a metric procedure suffices.11 However, a metric procedure in not equipped to handle survey responses gauging personal opinions. Alternating least squares optimal scaling (ALSOS) (Jacoby 1999; Young 1981) provides a fix to this problem. ALSOS finds an interval-level set of values that re- spect original assumptions about the measurement level of the data and provide the best. fit. (minimize the sum of squares) for the spatial proximity model. After in- corporating ALSOS into the metric procedure, the steps in the unfolding procedure 1 lsed in this research are as follows: . * . . . . o . . 1 . Start With Am (m indexes the iterations), a matr1x of row—conditional monotomc transformations of A. 1 , . . . . . . . . OT he first time through, the locations of either the stimulus pomts or ideal pomts II111st be specified by the researcher. . 1111 fact, dollars spent is used as an input to a metric procedure estimated in dCoby and Schneider (2009). 21 2. Run A?” through the CGM algorithm and find initial coordinates for the ideal points and stimulus points, producing the estimated matrix, Dikn- Calculate the g1 )odness of fit for this iteration. 3. Run D* an optimally scaled version of D*, through the CGM algorithm 112+1’ and find new coordinates for the ideal points and stimulus points. Calculate the goodness of fit for this iteration. 4. If the goodness of fit for this iteration is worse than for the first or is unchanged, stop. Otherwise, go to the next step. 5. Run Dikn. +2, an optzmally scaled version of Dikn+1’ through the CGM algorithm. Find new coordinates for the ideal points and stimulus points. Calculate the good- ness of fit for this iteration. (5. When the goodness of fit. stops improving across iterations, stop. Use these final estimations of the stimulus and ideal points. The goodness of fit is defined as Stress2. 3:125:1(62-3- - if)? Z?=1Zf=1(d*j— d:- )2 (2.2) which takes 011 lower values as the fit improves. A more intuitive measure of fit is the R2 value, which is equivalent to the squared correlation between the original. though optimally scaled, data values and the distances between the ideal points and stimulus points estimated with the unfolding model.12 This R2 value is the measure of political dimensionality used in this research. As R2 tends toward zero, political dimensionality increases. As R2 tends toward one, political dimensionality does as well.13 12R2 is also equal to 1- Stress%. 3This holds under the intuitive assumption that individuals feel positively about parties with ideal points close to their own. 22 2.4.3 The Unfolding Method in Comparison to OLS Ordinary least squares (OLS) regression is a method familiar to nearly all quan— titative social scientists. As such. it is useful to briefly build the. intuition of the unfolding method developed here in relation to this technique. With OLS, one de- termines how well a. given set of observed points corresponds to an underlying linear model. The observed points are derived from each observation’s values on two or more variables. After being fit to a linear model, a. predicted point is reported for each observation. A goodness of fit statistic. R2, is reported, which captures the strength of the relationship between the predicted points and the observed points. A value of 1.0 indicates a perfect linear relationship. The unfolding method also starts with a set of observed points. These are each respondent’s evaluations of the six parties. It then determines how well this set of points corresponds to an underlying unidimensional spatial proximity model. Like with OLS, a predicted point is reported for each observation (respondent) and each party. These are the ideal points. A goodness of fit statistic is also associated with the. unfolding analysis. And, like the OLS R2 statistic, it captures the strength oft he relationship between the predicted points and the observed points. A value of 1.0 indicates that voter preferences in a given nation are entirely generated by a single dimension - that the politics of the nation are unidimensional. 2.5 Unfolding in Comparative Elections Research A limited body of research uses unfolding to study elections comparatively, pro- viding interesting observations, predictions, and insights. In an early treatment of (lii‘neusionality, Butler and Stokes (1969) use unfolding to examine 1.)references of British voters over three parties: Labour, Liberal, and Conservative. In doing so, the authors test whether voters perceive the underlying political space of Britain 23 as unidimensional. Their findings demonstrate multidimensionality, meaning that the preferences of British voters, at least in the 19608, could not be explained with reference to a single underlying dimension. Studying a classic multiparty system, Norpoth (1979) examines party prefer- ences in the West German electorate. He finds that a single dimension, defined by social class and religion, underlaid German politics from the early 1960s to mid- 19703. Because voter preferences along a single dimension shift regularly across elections, Norpoth surmises that they are heavily susceptible to cues from party elites. And, due to the precarious coalitional politics that arise from Germany’s relatively inclusive, tiered electoral system, these cues are unlikely to come from a "fixed configuration of parties” (724). Lin, Chu, and Hinich (1996) examine the 1992 election in Taiwan, which initiated the nation’s process of democratic consolidation (455). They posit that a single di- mension with political liberalism on one end, and conservatism on the other, existed before Taiwan’s transition. However, after regime change, the makeup of Taiwanese political space shifted. To identify the new dimensions of conflict, the authors exam- inc survey data collected just before the 1992 election. They find two dimensions of conflict, along which both political candidates and ordinary citizens are located. The first reflects national identity, with strong Taiwanese identification 011 one end, and strong Chinese identification on the other. The second dimension they identify as relating to socioeconomic justice (469). The authors demonstrate that individuals’ locations on these dimensions are related to a number of demographic characteristics (472-473). They then show that vote choice in Taiwan is dependent on an individ- uals distance from a given party in the two-dimensional space, which itself can be predicted from demographic characteristics. Hinich et a1. (1999) examine the 1998 national election in Ukraine. Using pref- erence data, they find a “traditional” left-right dimension and a. dimension gauging 24 preference for reform, along which both voters and parties are located. The authors then use demographic and opinion variables to predict the recovered voter ideal points. Assuming that voters select the party closest to them in the two—dimensional space, the authors use the predicted ideal points to forecast vote shares for several of the competing parties. There are differences between the authors’ forecasts and the actual vote returns, though they are not stark. The authors also find that ideal point locations predict voting intentions. A fair amount of research uses the unfolding model to study European nations cross-nationally. For example, Listhaug et al. (1990), studying European party systems. find that parties in these nations tend to gravitate toward the extremes, leaving an “empty center” in the political space. Rabinowitz et a1. (1991) find that in Norway and Sweden, parties again tend toward the outer portions of the political space. Studying these same nations, Narud and Oscarsson (1999) use multidimen- sional unfolding to examine the makeup of policy space and the locations of voter and leaders within. They find the Norwegian system to have a “multidimensional character,” while the Swedish system conforms well to a single dimension (28). Dow (1998) uses unfolding to study voter and candidate behavior in Chile’s 1989 national election. Finding that a single dimension represents Chilean political space well, he shows that the nation’s “binomial” electoral system14 encourages politicians to move toward the extremes of political space. Dow then shows that voters’ proximities to the various candidates along the recovered dimension, as well as a battery of demographic variables, are significantly and strongly related to their reported vote choice (463-466). Carkoglu and Hinich (2006) use unfolding to examine the preferences of voters 14Chile’s electoral system applies the d’Hondt electoral formula in two—member districts. The result is that each party (or coalition of parties) can nominate two candidates per district, and the only way a party wins both seats is if it receives more than twice the vote of its closest competitor. If party A receives 60% of the vote and party B receives 29%, party A gets both seats. If party B instead receives 31% of the vote, each party gets one seat. 25 and political space in Turkey. They find two main dimensions of voter attitudes in Turkey, one defined by secularism vs. Islamism and another defined by the Turkish nationalism vs. Kurdish identity. Voters are distributed throughout the range of this two-dimensional space, though the bulk tend toward the center (379-380). The authors go on to show that the location of voters’ ideal points in this space is dependent on several demographic characteristics. In a cross-national study of four countries, Dow (2001) examines whether there are differences in the dispersion of parties across majorit arian and proportional elec- toral systems. Representing each nation’s parties in two dimensions, the author finds no cases of strong centrist behavior. However, in the two majoritarian nations examined, parties are more inclined to gather around the median voter than in the proportional systems. While this project uses unfolding to study party and voter behavior in a fashion similar to the work outlined above, a different strain of previous political science research uses unfolding to examine other tepics. For example, to examine the spa- tial makeup of party factions and their associated leaders in Colombia, Hoskin and Swanson (1973; 1974) use a multidimensional unfolding model. Additionally, Ja— coby (1982) uses unfolding to study party identification in the US. and Jacoby and Schneider (2009) employ unfolding to evaluate spending priorities in the US. states. In addition, an extensive body of research, based on earlier work by Poole and Rosenthal (see, for example, 1997), uses unfolding in the form of NOMINATE scores to measure the behavior of legislators in floor votes. Note that \Neisherg (1972) and Wood and Jacoby (1984) also used unfolding to model legislative vot- ing. In addition, unfolding has been employed in fields as diverse as anthropology, psychology, and engineering. 26 2.6 Conclusion In this chapter I introduce a new way to conceptualize and measure dimensionality across countries. While previous measures are concerned with the number of issues or ideological divides in a country, the technique of measuring dimensionality intro- duced here, unfolding analysis, explicitly gauges the space in which political parties compete. Such a technique is important for examining the dimensional configura— tions that result after political parties adjust their strategies according to societal conditions and institutional rules. Unfolding is also useful in that it provides party and voter locations across several countries. Such information provides a way to examine the correspondence between voters and elites. Moreover, the locations are not obtained by relying 011 subjective expert opinions or mass survey data, but are instead arrived at by evaluating voter preferences. In short, unfolding allows researchers to examine how well a set of party prefer- ences conform to a single underlying continuum. In doing so, it locates parties and voters along this continuum. If the relative distances between parties and voters correspond highly with the input preference data, there is evidence that a single di— mension captures much of ’the variance in party and voter locations. Alternatively, if voters and parties cannot be located along the recovered unidimensional continuum in manner consistent with the preference data, it is shown that the political space of that country cannot be captured unidimensionally. 27 Chapter 3 Political Dimensionality across Nations I11 this chapter I introduce the new measure of political dimensionality derived from the unfolding analysis. The necessary voter preference data is obtained from the Comparative Study of Electoral Systems (CSES).1 The measure covers several coun— tries across the years 1996—2006. While it varies intertemporally within some coun— tries, there is also much variation across nations. Comparing the measure to other variables indicates that, for most countries, the most salient political dimension is the common left-right, socioeconomic continuum. However, in certain nations where national politics are defined by atypical forces, the substance of the dimension is different. This is generally the case in countries where politics are defined by relations with a former colonial power. I also report the party and voter locations associated with the new measure in this chapter. In many nations, especially those without viable fringe parties, the placement of the political parties is intuitive. In other nations party placements along a single dimension do not correspond to expectations. For example, parties that embody separatist issues are often placed at nonintuitive locations. This is 1Available at http://www.cses.org/ 28 likely because such parties base their existence on a second dimension that is highly unrelated to, or even orthogonal with, the prime dimension in their home country. Finally, I compare the new measure to previous measures of cross-national dimen— sionality, finding there is little relation with these measures. This is not unexpected, as the measures all purport to 111easure something intrinsically different. However. the results of the comparisons must be taken lightly as the overlap of countries and years among the new measure and previous indices is minimal. 3.1 Cross-National Unfolding Results The. CSES is invaluable for this research in that it asks consistent survey questions across dozens of countries, regions, and levels of socioeconomic development, provid— ing a broad sample with high comparability. Questions A3020 from CSES Module 1 and 83037 from CSES Module 2 ask voters to rate most or all of the competing political parties in their home nation. The exact question wording is: “I’d like to know what you think about each of our political parties. A ftcr I read the name of a political party, please rate it on a scale from 0 to 10, where 0 means you strongly dislike that party and 10 means that you strongly like that party. If I come to a party you haven’t heard of or you feel you do not know enough about, just say so. ” Because higher values indicate more preference, the data were 'I'cflmtcd (sub- tracted from 10) to meet the criteria of the unfolding procedure (see Chapter 2). I also Obtain data from the 1999 and 2005 New Zealand Election Studies (NZES).2 Th ese waves of the NZES ask a question that is nearly identical to A2020 and B3037 \ 2 Available at http://www.nzes.org/ 29 of the CSES.3 I use all nations, most voters,4 and nearly all parties'5 covered in these studies as of November, 2007. In the end, there are 42 countries covered. As some nations are surveyed multiple times, I end up with 81 cases.6 A total of 100,820 individuals’ evaluations were used in the creation of the index, with an average of 1244.691 per case. I estimate a spatial proximity model for all 81 country-years using the unfolding procedure. For each country-year, the voters and parties are aligned along a single dimension, each with an interval-level value demarcating their location. The R2 values, gauging the salience of the recovered dimension, range from a low of 0.541 in New Zealand in 2002 to a high of 0.955 in Great Britain in 1997. The standard deviation of these values is 0.098, and the mean is 0.746. Figure 3.1 depicts the R2 value for each country-year dyad. Table 3.4, in the appendix to this chapter, provides the numerical R2 values. After the analyses, I centered the ideal points (voter locations) and stimulus points (party locations) in each country, constraining the ideal points to have a mean 3111 1999, Question B3 of the written survey inquires: “We would like to know what you think about each of these political parties. Please rate each party on a scale from 0 to 10, where 0 means you strongly dislike that party and 10 means that you strongly like that party. If you havent heard about that party or dont know enough about it, please tick ‘don’t know.”’ Question B6 from the 2005 wave is worded almost identically. 401in survey respondents with no missing party evaluations could be analyzed. While this meant losing many respondents (23.7% from Module 1 and 24.4% from N1 Odule 2) the. number of respondents per country-year across both modules was still very high. Moreover, comparisons of means indicates that there are no important systematic differences in the age, education, gender, socioeconomic status, or self- I‘Oported ideologies of the respondents that could rate all parties and those who could not- This is unsurprising, as affect toward parties is an “easy" political emotion and ‘foters are very capable of evaluative responses (Campbell, Converse, Miller, and Stokes 1960; Narud and Oscarsson 1999) For some country—year dyads, certain parties, though included in the CSES questioning, simply did not receive enough ratings to be analyzed without losing a )i‘ery high percentage of available data. For Hong Kong in 1998 and 2000 this includes ’19 Citizen’s Party, which, in questioning, was combined with The Neighbourhood :uld Workers’ Service Centre. For the United Kingdom in 1997 and 2005 this includes 'he Scottish National Party. , 6111 Belgium, the CSES surveyed \Valloon and Flanders separately in both 1999 and 2003. ' 30 New Zealand 2002 Philippines 2004 Czech Republic 2002 S au12000 Fin and 2003 Nonway2001 Germany 1998 Brazrl 2002 Russia 2000 Pen12001 Taiwan 2001 Mexico 2003 Peni2000 Slovenia 1996 Romania 2004 Mexico 1997 Slovenia 2004 Poland 1997 Israel 1996 Russia 1999 Ireland 2002 Mexico 2000 Korea 2000 Netherlands 2002 Thailand 2001 Poland 2001 Albania 2005 New Zealand 2005 Spau11996 Russia 2004 Korea 2004 Iceland 1999 New Zealand 1999 Bul aria 2001 en12006 NonNay1997 Belarus 2001 Netherlands 1998 Lithuania 1997 Canada1997 Switzerland 1999 Romania 1996 S aki2004 taly 2006 Sweden 1998 Ukraine 1998 Ja an 1996 Belgium-Flan ers 2003 . Germany 2002 Belgium-Walloon 2003 Demnmk2001 France 2002 Chile 2005 Denmark 1998 Australia 2004 Iceland 2003 Japan2004 Ponugal2002 New Zealand 1996 Taiwan 1996 Switzerland 2003 United States 1996 Canad32004 Hong Kong 1998 Kyrgyzstan 2005 Israel 2003 United States 2004 . Ponugal2005 Belgium-Walloon 1999 Hong Kong 2000 Ausnafia1996 ' Hun ary 1998 United King om 2005 Czech Republic 1996 Sweden 2002 Hong Kong 2004 Belgium-Flanders 1999 Hungary 2002 Taiwan 2004 . ‘ Chfle1999 United Kingdom 1997 Figure 3.1: —* ———————— .- ------ —-—I Inn—.- ——————————————— u-n—I Inn—I. ————————————————— I ——-—. ———————————————— I m—-. ——————— I. --------- ———*-——- ----------- I ——-—. ———————————————— —---. ——————————————— —-——. ——————————————— -———* ——————————————— ————+ —————————————— I —-——+ —————————————— l I—————.- —————————————— I _----. —————————————— I ————— .——-—_——-————I——I I———-—. —————————————— I I—I—-——+ ————————————— I —————— O-—---—--------1 ——-—--* ------- — ————— I —————— .I—-——————I——-———I —————— .--------—---—I -— ————— . ————————————— I — ————— *——————-— ————— I — ————— + ——————————— I-II - ------ O- ------------ 1 — —————— . ———————————— Il -- ————— . ———————————— It - —————— -o ———————————— 'l I:— —————— a. ———————————— I - —————— -o— ——————————— l -- —————— —o— ——————————— 1 I— ——————— .II ——————————— vl — ————— --._——--- ————— d ———————— .————-———-————I ———————— o———————————l b— ------ ‘ ----------- 1 ———————— *--———---—— E ———————— . —————————— II F ———————— .- ——————————— L -------- . ----------- b-.. ----- _‘ —————————— I h- ———————— ' ——————— ——I — ————————— . ————————— I —————————— .------_—-l h- -------- *—_————-—d ————————— --o-——-——--—---# C ---------- .- ———————— 4 r----- ----- o -------- « l- —————————— o -------- i L— ———————— —-. ———————— d —————————— ‘————————I C —————————— -o -------- 'l b ---------- * ————————— P- ---------- -.- ------- I l- —————————— -o- ——————— - l- ——————————— .- ——————— < ———————————— .--—-———-—I ——————————— ’———-———I C ————————— ——’ ——————— u h- ——————————— . ——————— I I— ——————————— . ------- I h ——————————— . ———————— ———————————— .—————-—-I — ——————————— —. ———————— h- ——————————— + —————— I ———————————— +--———_l — ———————————— . —————— I -——-—————————. —————— I k- ———————————— . —————— I I—- ———————————— I. ——————— — ———————————— a —————— w h- ———————————— + ————— I I— ————————————— .r ————— I ————————————— .———-—I r ............. . ..... . i-I —————————————— .--———I r— ————————————— I—.——--I — —————————————— I.————I l-I ——————————————— .———I ——————--— ———————— .——I I I I I I .5 6 7 8 9 1 Fit of Underlying Dimension Dot Plot of Dimensionality 31 of zero. Also, if necessary, I flipped the ideal points and stimulus points to correspond with an intuitive left-right party ordering. For example, in the U .S. in 2004, the recovered configuration placed the Republicans on the left and the Democrats on the right. In order to make the locations correspond with common parlance, I reversed the signs on the ideal points and party locations. These alterations do not affect. the relative locations of voters and parties or the fit of the recovered dimension: they merely adjust the unfolding results to correspond with common unidimensional perceptions of parties. Figures 3.5-3.16, in the appendix to this chapter, display a total of 81 density plots of voter ideal points. Overlaid on each plot are the rough party locations and median voter for each country. The coding scheme for the party labels is borrowed directly from the CSES and each party is identified in the appendix in Table 3.3. In addition, Table 3.4, also in the appendix, displays the precise location of each party and median voter. 3.2 Intuitive Results In'most nations the placements of parties along the recovered continuum corresponds directly to popular perceptions. For example, in the UK. in both 1997 and 2005, the Labour and the Tories are on the left and right, respectively, with the Liberal Democrats toward the center. This corresponds with the findings of Butler and Stokes (1969), who also applied unfolding to the British case. In the United States in 1996 and 2004, while both parties fall near the center, the Democrats are to the left of the Republicans. The Reform Party falls to the right of both major parties in both election years. Also corresponding with intuition is Portugal, which unlike the US. and U.K. (".lects its parliament with a proportional system. In 2002 and 2005, the Communist \Vorkers’ party and the (also-communist) Unitary Democratic Coalition are together 32 on the far left, while the Christian conservative, Partido Popular is on the far right. Additionally, the larger Socialist and Social Democratic parties fall towards the center in the expected order (the Socialists to the left of the Social Democrats). Dow (1998) concludes that Chile’s “binomial” electoral system causes parties to abandon the center of political space. The results from both 1999 and 2005 correspond with this finding; while the voters are distributed evenly in a. “bell shape” on both sides of the median, the parties fall into two groupings on the left and the right of the dimension. On the left are the Communist Party (PC), Socialist Party (PS), and the Party for Democracy (PPD), a progressive social-democratic party. On the right are the Christian Democrats (PDC), the National Renewal Party (RN), and the Independent Democratic Union (UDI), the latter two being laissez—faire conservative parties. The relative order of the parties found by Dow (1998) across several different election districts in the 1989 Chilean election is identical to those recovered here, with the exception of the PC, which he finds to be right. of the PPD and the 1337 In the 2005 election to the lower house the PS, PPD, and PDC, received 15, 21, and ‘20 seats respectively. The Chamber of Deputies in Chile has 120 seats, meaning 61 are needed to form a government. With the addition of the 7 seats won by the Social Democrat Radical Party, these parties entered into a coalition dubbed the Concertacio’n. The theory of minimal winning coalitions presumes that parties strive for maxi— mal political power, and will thus include the fewest parties necessary when forming a government (Riker 1962). And, if policy is important to parties, they will be in- clined to enter into coalitions with parties with similar ideal points (Laver 1998). According to the positions recovered by the unfolding model, the formation of the Coherrrtacion in Chile fits perfectly with these two propositions. Not. only did the \ 7Only three parties are included in the 1999 CSES Chile study: the PPD, and U DI, and the PC. The relative order of the parties in this year matches that of 2005. 33 parties enter into a minimal winning coalition, the three parties included in the CSES data that entered into the coalition are aligned sequentially along the recov- ered continuum. The findings from Sweden and Norway match well with those of Narud and Oscarsson (1999), who also employ an unfolding model. They find the politics of Norway to be characterized by several dimensions, while the politics of Sweden are. well-represented unidimensionally. In the present analyses, the variance explained (R2) by a single dimension in Sweden is 78% in 1998 and 88% in 2002. In Norway, alternatively, the variance explained is 73% in 1997 and 60% in 2001, suggesting that Sweden does indeed conform better to a single dimension. Moreover, in Sweden in 2002, I recover a party ordering identical, over the parties studied, to that of Narud and Oscarsson’s (1999) analysis of Sweden for the year 1994 and Rabinowitz et al.’s (1991) analyses in the years 1979 and 1982. However, the order I recover in Sweden in 1998 differs from the previous studies slightly, in that the Liberal and Center parties are placed to the left of the Social Democrats. This may be because the aforementioned authors use multidimensional unfolding models which gz‘iuge variation along a second dimension, while I force all parties to align along a single continuum. In Norway the recovered party ordering is identical to Narud and Oscarsson’s (1999) in both 1997 and 2001, with the exception the Center party.8 The inability to place the Center party unidimensionally is unsurprising; Narud and Oscarsson find the Center party, an agrarian organization wary of the EU, to be the part y most deviant from a single recovered continuum (20). Rabinowitz et a1. (1991) also locate parties in Norway for the years 1969, 1973, 1981, and 1985. Though they do so multidimensionally, their recovered party locations along a single continuum align closely with mine for each year studied.9 :BNote that the Liberal Party is not included in 1997 CSES study. “)The gravitation of parties to the extremes in these nations, as noted by Rabi- 34 3.3 Counterintuitive Results and Alternative Di- mensions Intuitive configurations of parties are not apparent in all CSES elections. The rea— sons for these atypical placements vary. Comparing the party locations derived from expert and mass surveys and data reduction techniques, Gabel and Huber (2000) find the placements of extreme left and right parties to be the most precarious. In addition, special-issue parties are often placed in sporadic locations. Finally, in na— tions where political discourse does not conform to the common left—right distinction, unidimensional party placements are not readily interpretable. For example, in Spain in 1996 and 2000, in which the variance explained by a sole dimension is 70% and 58% for each year respectively, the recovered ordering of the parties makes little sense. In 1996, while the Socialist Workers’ Party, as expected, falls to the left of the more conservative People’s Party, the smaller United Left is placed at the far right. Again, in 2000 the larger parties are placed as expected and the United Left is placed on the right. Moreover, the Basque Nationalist Party (PNV), which is defined more by its link to the Basque community than its political positions, flips from the left side to the right side of the continuum across the two years. As expected, in 2004, when the PNV is not included in the CSES questioning, the variance explained by a single dimension rises to 77% and the parties align from left to right as expected. A similar pattern is observed in Canada in 1997 and 2004. Like the PNV in Spain, the Bloc Quebecois (BQ) in Canada is known more for its allegiance to the province of Quebec than its left-right political positions. As expected, the left- wing New Democratic Party and the center—left Liberals fall on the left side of the nowitz, Macdonald, and Listhaug (1991), is not apparent for all parties. Though, for Sweden 1n both 1998 and 2002 it does appear that all parties but the Social Democrats abandon the center. 35 dimension, and the Conservative Party10 falls on the right in both years. However, the BQ flips from the right to the left over the two survey years in a fashion similar to that of the PNV in Spain. When parties form for the sole purpose of representing one region, they are less likely to take strong stances on salient national issues. This, in turn, increases the dimensionality of politics in a. nation and lowers the likelihood that all parties in a nation will compete along the same underlying dimension. While countries such as Canada and Spain have strong, national parties com— mitted to regional separatism, other nations’ politics are defined by their historical ties. For example, in Taiwan and Hong Kong political parties are well-known for their stance on relations with the People’s Republic of China. In Hong Kong in 1998 and 2000 the parties all fall near each other on the recovered spectrum, and their ordering is based on their stance toward Beijing. On the left are the Democrats and the Frontier, both pro—Democracy parties, while the Democratic Alliance and the Liberal party, both pro-Beijing in orientation, fall on the right. In 2004 party polarization increases, and the parties again are placed from left to right according to their stance on independence (though the Frontier is excluded from the CSES in this year). In addition, for the 1998 and 2000 survey years, the CSES asks a question (A3033) in Hong Kong gauging individuals’ orientations toward the People’s Re- public, with higher values corresponding to an anti-Beijing position. As expected, in 1998, this variable relates negatively to ideal point (r = -.425), meaning that indi- viduals with a more pro—Hong Kong stance tend to gravitate toward the anti—Beijing Democratic and Frontier parties. In 2000, though still negative, this relationship weakens dramatically(r = -.045).11 10The Reform party, which morphed into the Canadian Alliance, then merged With the Progressive Conservatives to form the Conservative Party. In 1997, Reform and the Progressive Conservatives are evaluated separately, both falling on the right, as expected. 1This question was not asked in the 2004 survey year. 36 In Taiwan, a similar pattern emerges. In 1996, the pro—independence Democratic Progressive Party (DPP) falls on the far left of the spectrum, while the strongly pro-Beijing New Party (NP) falls on the far right. The more moderate Kuomintang (KMT) falls in between the two parties. In 2001 more parties are included in the survey. The New Party again falls on the right, but the remaining parties are placed on the left of the dimension. Finally, in 2004, the NP and KMT fall on the right, while the DPP and the pro-independence Taiwan Solidarity Union are on the left. It appears that the recovered dimension is the main dimension found by Lin, Chu, and Hinich (1996): a nationalism dimension defined by Chinese vs. Taiwanese identity. Like in Hong Kong, the CSES asks a question in 1996 and 2004 in Taiwan (A3033 in 1996 and 83046 in 2004) gauging individuals’ orientations toward Beijing, but with higher values corresponding to a pro-Beijing position. However, in both 1996 and 2004, this variable is essentially uncorrelated with individual ideal points (correlations of -.131 and .038, respectively). The politics of Taiwan and Hong Kong are unique due to their relationship with the People’sRepublic, and thus are not captured with a traditional L-R continuum. Another trend, apparent throughout numerous countries, is the inability to place Green or ecological parties in a consistent and meaningful place on the recovered continuums. Parties of this sort, though primarily defined by their concern for the environment, often take stances on issues that are traditionally associated with left- wing politics. These stances include a commitment to social justice, peace, and non-violence. For country-year dyads such as Australia in 2004, Mexico in 1997 and 2000, and Finland in 2003, the Green parties fall toward the left of the continuum. However, in Germany in 2002, the Greens are placed as centrists, and in Mexico in 2003, New Zealand in 1999, Australia in 1996, and Belgium-Walloon in 2003, the Green parties fall on the right side of the continuum. The inconsistency of Green party locations among the various recovered continuums likely stems from the fact 37 that these parties fall along a second dimension, likely defined by “postmaterialist” or “New Politics” issues (McAllister and Studlar 1995; Inglehart 1977). Other parties that take stances associated with issues that do not fall clearly on the traditional left—right dimension also fall at various, nonintuitive locations throughout the recovered dimensions. For example, in Germany in 2002, the Re- publikaner party, a far-right entity often associated with neo—Nazism, falls on the far-left of the continuum with the Socialists. Similarly, in Belgium-Walloon in 1999, the National Front (F N), a segregationist, far-right party is placed on the left side of the recovered dimension. In Finland in 2003, the Swedish People’s Party (SPP), which primarily represents the interests of the Swedish-speaking minority, is located at the far right of the dimension. Though the party tends toward liberal, free-market economic positions, its placement on the far right overstates its conservatism. l\»’lore likely, the SPP, the Republikaner, and the Walloon FN do not align along the left- right dimension, but instead along external dimensions defined by special interests (in the case of the SPP) or xenophobia and nationalism (in the cases of the Repub- likaner and the F N). 3.4 What are the Dimensions? An Empirical- Substantive Exploration Policy positions may be constrained by ideology, which can force positions across a range of issues to relate to one another in a consistent manner (Converse 1964). As noted by Huber and Powell (1994) the most common dimension in developed democracies is the left-right ideological continuum. This dimension can uniformly assimilate the various issues presented to the electorate. In this vein, Gabel and Huber (2000, 96) define the left—right dimension as “the ‘super issue’ that most con- strains parties’ positions across a broad range of policies.” As expected, throughout 38 the bulk of the countries examined here, communist and socialist parties fall on the left, while Christian democrat and free market liberal parties fall to the right. The R2 value for each nation-year may be conceptualized as the strength of this “super—issue” left-right dimension. As this value is usually well. above .50, the single dimensions recovered across each election usually explain most of the political variation in a given country. While the party placements give insight as to the substantive interpretation of these dimensions, an appeal to individual-level data also sheds light on their “real-world” meanings. In countries where parties align along the left-right dimension as one would ex- pect based on their known socioeconomic issue positions, R2 values are relatively high. In fact, the U.K., the US, Portugal, and Chile, each with party labels that correspond to intuition, all have R2 values of 80% or above. Alternatively, in cases such as New Zealand in 1999 and Finland in 2003, parties do not align as expected and the variance explained is lower (72% and 59% respectively), meaning that polit- ical space is likely not unidimensional. In nations such as Hong Kong and Taiwan, where relationships with China primary define politics, a single dimension may cap- ture much variance, but be substantively unrelated to the common left—right issue dimension. To examine the substance of the recovered dimension in each nation, I test how strongly they correlate with individuals’ self-placements along the common left-right continuum. Recall that the estimation of the spatial proximity model places all indi- viduals studied at a certain ideal point along the recovered dimension. In addition, the CSES and NZES ask individuals to place themselves along the common left- right dimension.12 Correlating answers to this question with individual ideal points 12Actual CSES question wording: “In politics people sometimes talk of left and right. Where would you place yourself on a scale from 0 to 10 where 0 means the left and 10 means the right?” In Module 1, this is question A3031. In Module 2 it is B3045. The wording in the NZES is almost identical. In the 1999 N ZES this is question B6, and in 2005 it is B9. 39 in each nation provides a gauge of how strongly individuals’ self-placements along the left-right dimension relate to their positions on the latent left—right dimension recovered by the unfolding analyses. I first generate a variable. correlation, which is the (‘toefficienn r, from a bivariate correlation between individuals’ ideal points and their self-reported left—right posi- tions in each nation. The CSES does not ask for individuals’ self-reported left-right. positions in Thailand, noting that left—right evaluations are not relevant in Thai pol- itics.13 Thus, there is no correlation generated for Thailand, dropping the number of observations from 81 to 80. The measure ranges from a low of —0.210 in Hong Kong in 2004 to a high of 0.760 in Sweden in 2002. Its mean is 0.296 and its standard deviation is 0.230. This variable is displayed in Figure 3.2. The variable fit is the R2 value from the unfolding routine. High values of this variable indicate that the recovered continuum in a nation is strong. If the recovered continuums truly capture the left-right, socioeconomic dimension, fit should be pos- itively related to the correlation variable. As expected, a bivariate OLS regression of correlation on fit initially returns a coefficient of .649, significant at p = .012. This result shows that a .10 increase in variance explained by a single dimension corresponds to an increase of .065 in the correlation between self-reported left-right. positions and ideal points from the unfolding analysis. As explained in Section 3.3, certain nations are sin'iply not defined by the classical left-right dimension, and including them in this test is thus misleading. In the sample at hand, these nations include Thailand, Taiwan, and Hong Kong. While. left-right positions are not gauged in Thailand, the average of the absolute value of the correlation variable across each election covered is .114 in Hong Kong and .087 in Taiwan. Clearly these nations are not organized along the traditional left—right. dimension. l3See Module 1 Variable, Descriptions Codebook. 40 Hong Kong 1998 alwan 2001 Taiwan 2004 Slovenia 1996 Mexrco 2000 _ Korea 2000 Belgium-Walloon 2003 Chile 2005 Israel 1996 Czech Republic 2002 Kyrgyzstan 2005 Hong Kong 2000 Peru 2001 _ Mexrco 2003 United States 1996 Peru 2006 Albania 2005 Romania 1996 Hong Kong 2004 Japan 2004 BraZIl 2002 Taiwan 1996 Romania 2004 Bulgaria 2001 eru 2000 Russra 2004 Belarus 2001 _ France 2002 _ PhIwmees 2004 Belgrum— alloon 1999 Mexrco 1997 Spain 1996 Germany 1998 . Ireland 2002 Belguum—Flanders 2003 New Zealand 2002 Lithuania 1997 , Spain 2000 Belgium-Flanders 1999 Australia 1996 Slovenia 2004 Japan 1996 Korea 2004 Canada 1997 . Finland 2003 United States 2004 Russia 1999 . _ Chile 1999 United Kingdom 2005 Nonrvay 2001 Netherlands 2002 Ukraine 1998 Netherlands 1998 Switzerland 1999 Germany 2002 , _Russua 2000 United Kingdom 1997 Iceland 1999 Iceland 2003 Czech Republic 1996 ew Zealand 1999 Portugal 2002 Portugal 2005 New Zealand 2005 Spain 2004 Australia 2004 Swrtzerland 2003 Norway 1997 New Zealand 1996 Denmark 1998 Sweden 1998 Hunga 2002 Denmar 2001 Israel 2003 Sweden 2002 Correlation Between Ideal Points and L—R Placements -. ——————————————————— I.--‘ ———————————————— I ———-o— ——————————————— « ———+ ——————————————— I ————.— ——————————————— I ———+-———-_--——-—---l ————. ——————————————— I _-——.-——-— ——————— --—l ———-+ —————————————— I ————+ —————————————— I P-—-+ —————————————— I -————c —————————————— 1 I-—————-. —————————————— I l"""—-—+ -------------- ——--—-o— ————————————— , —-—--+ ————————————— I — ----- * ————————————— d I'- ————— -.— ———————————— I p ----- *- ———————————— I II— ————— —-.— ———————————— I - —————— .- ———————————— '4 ——————— *—_——---—‘-—-1 ——————— .-———-———-——-—I — —————— . ———————————— I —————— .———————-————I C ______ .._ ___________ . P— ————— * ----------- I F ——————— * ——————————— I — ——————— .n ——————————— I — ——————— * ——————————— I - ——————— O- ——————————— 'l I—- ——————— .n ———————————— — ———————— .— —————————— I II— ———————— .II —————————— I r— ———————— .- —————————— 4 - -------- . ---------- I L -------- +0 ————————— I h ————————— . —————————— h- ————————— . ————————— I L ————————— .I ————————— I h ————————— . —————————— l— ————————— l. --------- I - ————————— -o- ———————— . _ ————————— + ———————— I he ————————— + ————————— - —————————— o- ———————— l h —————————— . ————————— b —————————— ‘ ———————— I r— —————————— -.—- ———————— P —————————— + ———————— _ —————————— + ------- I II- —————————— + ——————— d — ——————————— .— ——————— I II— ——————————— .- ——————— I I- ——————————— .. ——————— I t ——————————— + —————— I ———————————— *-----—I l———-- ————————— . ------ I h- ----- —-————. ------ I P ———————————— . ------ I b ———————————— . ------ I II- ———————————— ‘ —————— I i.- ———————————— + ————— I — ———————————— —o— ————— 4 - ————————————— o ————— l l.- ————————————— c. ————— I _——— ——————— ——-‘—---—I F- ————————————— -.--I-——I—I _ ————————————— +——-—Il - ————————————— +-———I .— ————————————— +———-—J -———— --------- -.——-—I III- —————————————— +———I — —————————————— +-——I In- ——————————————— .n—I—u—I .- ——————————————— .I—I—I—I - ———————————————— *—-I III- ———————————————— .I——I _ ———————————————— ‘——I —-———— ———————————— .u-I Figure 3.2: Dot Plot. of Correlations 41 w The results of an OLS regression without Taiwan and Hong Kong (71 drops from 80 to 74) are shown in Table 3.1. The relationship is also depicted in Figure 3.3. The coefficient on fit jumps to .933 and is now significant at p = .000. In nations with low dimensional political space, the correspondence between individuals’ recovered ideal points and their self—reported left-right positions is high. Thus, there is strong evidence that the primary dimension recovered in the nations studied, with the ex— ception of Hong Kong, Thailand, and Taiwan, is the traditional left-right continuum. In the language of Gabel and Huber (2000), it is a “super dimension,” capable of capturing and organizing several of the issues presented to the electorate. Table 3.1: Salience of the Left-Right Dimension and Fit Variable Coefficient (p—value) Fit 0.933 (0.000) Intercept -0.370 (0.043) n, 74 R2 0.173 Prob > F 0.000 3.5 Relationship with Previous Measures of Di- mensionality As discussed in Section 2.1, a handful of researchers have previously measured dimen- sionality across nations. In this section I provide a description of the relationship of the new measure of political dimensionality with three existing cross-national measures: those of Stoll (2009), Nyblade (2004), and Lijphart (1999). Obtaining these previous measures was easy, thanks to the courtesy of Benjamin Nyblade and Heather Stoll, who both happily provided me with their data sets. Lijphart's data was readily available in his 1999 book, Patterns of Democracy. Directly assessing the relationship with each measure is diflicult. as they sample 42 .8 l .5 l .4 1 .2 I Correlation Between Ideal Point and L-R Position 0 I O O O mi .1 \‘q «>4 . .8 Fit From Unfolding Routine Figure 3.3: Salience of the Left-Right Dimension and Fit (.lifferent nations and years. Stoll and Nyblade sample across countries and over time, while Lijphart creates a single measure purported to describe dimensionality from 19453 to 1996. And, while Lijphart samples nations throughout the world, Stoll and N yblade focus on westernized nations. Furthermore, because the years covered in Stoll and N yblade’s data do not always correspond to those in the CSES, for several nations I proxy with the closest year sampled. If nearby years are not proxied, the resulting overlap is 6 country-year cases with Lijphart’s measure, 7 with Nyblade’s n'ieasures, and 11 with Stoll’s. Also, as noted in Section 2.1, each measure aims to gauge a different type of dimensionality. Stoll seeks to measure the amount of “raw” ideological dimensions in a nation, while Lijphart is concerned with the number of salient issue dimensions. Nyblade’s Effective Number of Issue Dimensions (ENID) not only gauges how many salient issues arise in a country, but down-weights this number when multiple parties deem the same issue important. C(‘inversely, the new measure of dimensionality introduced here is only concerned with the space in which parties and voters align. I subtract the R2 value from the unfolding analyses from 1 so lower values cor- respond to lower dimensionality, as is the case with the other three measures. I also average across the Flemish and Walloon regions of Belgium to create single mea- sures for each year, dropping the observations on my new variable from 81 to 79. Correlations between the measures are given in Table 3.2, and a scatterplot matrix illustrating the relationship between all four measures is depicted in Figure 3.4. The measure of dimensionality from the unfolding procedure is essentially uncorrelated with the previous measures. Lijphart’s measure is positively correlated with Stoll’s, and moderately negatively correlated with Nyblade’s. Table 3.2: Correlations among Measures of Dimensionality Variables Lijphart Nyblade Stoll My Measure Lijphart 1.000 (41) Nyblade -0.236 (25) 1.000 (25) Stoll 0.404 (41) 0.114 (25) 1.000(41) My Measure 0.166 (41) —0.135 (25) -0.001(41) 1.000 (81) Number of observations in parentheses. The lack of congruence among the measures must be viewed with caution. First, the relationships among the previous measures are only evaluated across the coun- tries and years also covered in my CSES sample, perhaps skewing their true rela- tionships. Second, because the data for certain years is proxied with nearby years, the results may be misleading, especially for the measure of Lijphart, which does not consider any year beyond 1996. 3.6 Conclusion The new index of dimensionality introduced here covers more countries than any previous measure. Thus, it serves as a valuable tool in cross-national research, especially that which makes use of the Comparative Study of Electoral Systems. 44 Figure 3.4: Relationships among Measures of Dimensionality 45 2 3 0 I I l i—4 O O O .. , o a) o m a) o oanoom oo o —3 Lijpharts oo o o a: o coa- o 0 Measure 0 o o o -2 O 0 O O moo O O 3 Q Q m m 0 ~1 , "l . O O O L; 2‘ o o StolI's oo o o o o 0 Measure 0 Cb o 9 0 ° 6 o o 0 0°C 0 o p g 0 o b Q3 ‘1’ 00 ° §° oo 1‘ ~35 o . o o °oo tel/blades o 0 r3 0 o e 9 Q, easure ooooq, o e 9 Q: 0 ° °o °ooB°°0 ‘b o o o 0 ~25 .5« 0 ° 0 ° ° 00 Q) o g o o 9 980608 0 0o 0% My 3 o 8 B @3 0 ° a 0 °e Measure 0 O O O O O I I I I I 1 2 3 2.5 3 3.5 The measure is valuable for researchers who wish to examine the dimensionality of the space in which parties and voters locate, either as an outcome variable or as a causal or independent factor. The index indicates politics conform almost perfectly to a single dimension in several countries, while political variation in others cannot. be well-captured unidimensionally. Substantively, there is evidence that the main political dimension in each country is the common socioeconomic continuum. This can be thought of as a “super dimen- sion” that bundles together multiple issues. In some nations this super dimension is very s1 rong, while in others politics are defined primarily by alternate factors. Voter and party locations are provided along the recovered dimension in each of the 42 countries and 81 elections. Party locations are generally intuitive, follow- ing common perceptions. However, in many cases fringe and niche parties fall at unexpected locations in the recovered political space. In sum, the new index of dimensionality provides a clear, quantitative. measure of the unidimensional conformance of political space across countries. In addition, party and voter locations along the recovered dimension are provided. Because the. new measure covers numerous countries and years corresponding to the CSES (and can be updated with subsequent waves of the survey), it is a useful tool for comparative political, social, and economic research. 46 3.7 Appendix to Chapter 3 This appendix provides figures and tables detailing the results of the spatial prox— imity model, estimated with unfolding, for 79 country years from the Comparative Study of Electoral Systems and two from the New Zealand Election Study (1999 and 2005). Figures 3.5-3.16 display a total of 81 density plots, each illustrating the the dis- tribution of voter ideal points in a particular country-year dyad. Table 3.3 provides the names and ideal points of the parties in each dyad. The coding scheme is taken directly from the CSES, with two differences. First, in Belgium in 2003, the CSES incorrectly labeled parties G and H as the Centre Democrate Humaniste and the New Flemish Alliance, respectively. I reverse this labeling to fix the error. Second, in Brazil in 2002 the CSES treats parties B and G, the Brazilian Social Democratic Party (PSDB) and the Labor Democratic Party (PDT), together. I refer only to the larger party, in terms of votes received and seats in parliament, the PSDB. Table 3.4 provides the location of each party on the recovered continuums, as well as the location of the median voter and the R2 value for each country year. 47 Hong Kong 1998 .15 Hong Kong 2004 A 0 0| 3+ Japan 2004 L o I ca 0 01 .25 .15 .05 .25 .15 .05 Hong Kong 2000 —i0 —'5 6 E. 1? Japan 1996 —i 5 —i o :5 6 5 Philippines 2004 i l Figure 3.5: Asia Korea 2000 :4 —’2 (i .5. 4i Thailand 2001 —i o —'5 6 é 10 Taiwan 2001 —i o :5 6 5 Korea 2004 .4 Taiwan 1996 —1 Taiwan 2004 Figure 3.6: Asia continued... 49 o)— Albania 2005 Bulgaria 2001 a, - V: -l (\l. —1 ('3. —r “,3 . (\I —r 8 A ". 'l O 1 O J .5 (i is 1’0 1’5 :5 6 2'5 Czech Republic 1996 Czech Republic 2002 or a 04 a 3‘3 ~ 52 — v— .J ‘— -i L0 to O ‘ O _ O "l O 1 -10 —’5 6 é —i0 —'5 6 Hungary 1998 Hungary 2002 N -i 2 _r a _ m — u, 0. c, _ O '1 o _l —4 —2 o 2 i 6 —’4 —'2 6 é 3 Figure 3.7: Central and Eastern Europe 50 Poland 1997 Poland 2001 ID '— LO . Q _ 0.! _ r'. ~ m d ‘8. 4 F’ ’ LO Q _ O - o . —i5 —i0 35 0 5 1’0 —5 o 5 Romania 1996 Romania 2004 N 8 ID _ (‘1 ‘ 53 _ ID Q " ID Q - O 4 0 d I I I I l I I I I I I _10 —5 0 5 10 15 —15 —1o —5 0 5 Slovenia 1996 Slovenia 2004 N - a ‘ 0! _. :2 _ . m , ID 0 “ ID . Q _ O 4 o _. -15 —1o —’5 0 5 —4 —’2 0 z i 5 Figure 3.8: Central and Eastern Europe continued... 51 Portugal 2002 l A | N O N ¢ — Spain 1996 -I .4 1 Spain 2004 Portugal 2005 Spain 2000 _4 —2 0 2 4 Figure 3.9: Iberia O .05 .1 .15 .2 O .05 .1 .15 .2 0.05.1.15.2.25 LIIIII 0.05 .1 .15 .2 Brazil 2002 Chile 1999 l I l_1 k ..; fl Q a o —4 is 0 5 1'0 —’5 o 5 Chile 2005 Mexico 1997 ’53. . I): _. Q _ o — —i0 -’5 0 5 1’0 —io —’5 0 5 Mexico 2000 Mexico 2003 Q .. A N. A o -l J5 £4 12 6 5 1 —i0 :5 0 Te Peru 2000 Peru 2001 “N1 _ 04 1 53 _ Q -I O —4 —2 0 5 31 5 —io —'5 0 5 10 Peru 2006 I 5 I I I I —5 0 5 10 1 Figure 3.10: Latin America 53 .15 .05 .15 .05 Australia 1996 O-1 New Zealand 1996 10 New Zealand 2002 Figure 3.11: A— 54. Australia 2004 10 New Zealand 1999 -5 New Zealand 2005 fi — —I Oceania Belarus 2001 1 Lithuania 1997 "-1 ‘ _4 q q Russia 2000 —’5 0 5 10 Ukraine 1998 --10 -5 0 5 10 Kyrgyzstan 2005 Russia 1999 01.. Russia 2004 Figure 3.12: Post-Soviet States CJ-‘I C"! 01.1 0.05.1 .15.2 .1 .15 O .05 Denmark 1998 —10 .E c3 5 1'0 Finland 2003 -10 —'5 (i 5 Iceland 2003 do :5 6 5 1’0 NonNay 2001 -I 10 —5 0 5 Sweden 2002 -’ I I I I —5 o 5 10 0.05.1 .15.2 0.05.1.15.2 0.05.1.15.2.25 Denmark 2001 I 1A _ fi I I 0 -5 0 5 Iceland 1999 q I I I o —5 o 5 Norway 1997 — I I I -5 0 5 Sweden 1998 1 A E I I I I ~10 -5 0 5 Figure 3.13: Scandinavia Belgium—Flanders 1999 (\l .1 :9 - g;. o _. —i0 :5 5 5 1’0 Belgium—Walloon 1999 (\I _ m -1 I!) Q . o 1 —i0 —’5 5 5 10 Germany 1998 m, a “I 1 O 1 —15 :10 —’5 I) 5 Belgium—Flanders 2003 —5 0 5 10 15 Belgium—Walloon 2003 .15 Germany 2002 Figure 3.14: Western Continental Europe France 2002 (5,1 I i 0 5 10 Netherlands 1998 Switzerland 1999 10 Italy 2006 l .5 I N O N A .. Netherlands 2002 I 01 01- Switzeland 2003 .15 Figure 3.15: W'estern Continental Europe continued... Canada 1997 Canada 2004 cq d “I 1 m -4 4‘! u *1 v-. A g3 a O - O - is 6 is 1‘0 .H is 6 5 Great Britain 1997 Great Britain 2005 «,3 cs I 2 J 8 d /\/\/\ 8 ‘ //\’~ 0 -‘ o .. -10 :5 (I 5 1’0 4’0 _’5 0 5 10 Ireland 2002 Israel 1996 m, 4 fl _ a! 3 Nw/\_\ F. — W ,_ to . . c2 ‘ O -* O -‘ ‘ _i 5 —i 0 —'5 5 5 1’0 :5 0 5 1’0 15 Israel 2003 United States 1996 :9 cs ' In F. s "- I ’3. ‘ m ’3 : /\/\’\ O -i o .., -5 0 5 —1o :5 0 5 10 United States 2004 lo Q a o —I —10 35 5 5 10 Figure 3.16: Other Western Countries 59 .0me use: so tosfiucob .52 020 has: Nmowm maze; fig SR 55:5 an: FEE NmQNm “EVE $2-2m 05m: was sawém E55 sens 3-85 55 88 secs meow E533 Amrfiofl BEGOE =<é~oQ Ema coca wcox wmom 3.825 AnTEoQ 35:05 :<-EoQ Eon. mama wcovH wqom gem name Be as; 5038 >3-qu ago Em SR .8955 mom age 508% smog poo cam 82 .3550 mom .5 om> um: 04 mm: mm 2m max 88 tease mommrgm EGQIHSU SGQHU 2z Em 5 >30 «.mm BS 88 aésmsm 5:9» 5555 3m 5 saw m>o RS 82 b-5228 SOULSO 301Gb Ema-00m :HOQIQMQ flflmudhwdw QHEOU HOON maid—om 20 $5ng EEOEQZ 2080 AHA/w 120an «MOON mzmppwddx Gmmpo mama ECCG‘GZ Epmnfig sagas. maafi .mzduumflaeq mam GE .33 mom as man 5 mm mm SR 532 a 3.8m 3 3.8% U high an magma m been Q 3.8m O ~3st m 3.8% 4 3.8% new? mecca—CO $2sz 5.5% ”MM. 033. 60 0&00 4x0: :0 4002:3450 n43 L02 4mm 3:4 Q4m m»? 304 40:040n4 8050 3. 000 A: 002 at: 32 5.55858 mm 22 0mm n4< 2D > 034 <90 moon 28020502 m0m mm 40 own .480 Q>-> 04034 wag 28020502 05280 8 285 000 80 28 .88 oat: 224m SET/m Hm Qmm 42m 24¢ coon 02x02 7:20.24 Hm 2m>m OMS 245 SE 33 00202 32 34-80Q 54-800 80Q-80 809-00m :082-2 304 0800424 2-8800 302 32-002 4803002 00m 2-8800 .3830 meow 40.030.23.22 4m<2 >24. QEm QQE Q2 QD gem 00:02 0002 £4-80Q 0003809 >24. 02m Om oocm 00:02 000 02. 002 004 550 32 8%.. gem 00m 00.. 550 see-52 000 8% ages 000 000 8 72 8 E 88 as: 80.82 :D-32 004m 3:2m .8404 405:4 moom 30:04 30.4042 3032 0am 03:4 000>< @064 200.84 :0000 80Q-m00m 80-.4-m 80404 300 0804 43-2-22 meow 480404 3024 50-5004 mm 00:02; n: meow 4804004 3024 50-204 a: 00:02:44. E @004 4804004 4 3.40m E 300nm .0 3.80m 04 500nm m— bamm Q zaamm O 3:0n4 m 280nm < 3.40m 80> 2.5550 €083 2.. 538.. 61 Epomom pauooaoa SZQQMH vOON magnum ©0033 EEO-Hem asaom HGMUOEQQ mam: mmpcum @0053 0000-: 80 583 383 .20 5000:: :00 05030 32 2.0 EoQ-uom mpzmmeamnm SEQ-3500* 00m £23m HES-400 mama 0420-42 D 0.30 000 0-00 2080 00 000 :53 05:00. 02 000- 000 00.8 000 38 $300. 02 000. 000 00.8 000 Sam 80200. 02 000 0.200 032 £300. 2.45.0 U QED-.220 Suwanee-20m SEQ-Dom mh BQOmQ MOON Ucmszfigm Coo-40 ESQ-0:0 SOC-00.2fm ESQ-vow 90720002 OOOH UCEHONfiBm 0.5900 dad SEQ-0&0 mownm-m—MQ CCU SEQ-00m NOON :wUQBW @0003de 93:00 Ema-.220 £04 0000020042 2000—loom mam: cofioam 0:0 0: 0000 00 :53 5300 >20 00 00 000.0 00 803 5303 >20 00 00 0000 00 002 5.000 00000 3:40 000 :02 0003 000 000 38 0560 0.23 0003 02 000 000 000 000 05.60 000 300 3020 0000 0002 00 :53 30.90 96300. 800-033 000 000 000:0 0002 083 0.000 903% 800-033 000 000 00:0 0002 0%: 0030 000-20 000 0200 200 00 020 000 0003 050050 020 .0200 0200 00 000 0007.0 02: 05:50 00 00-000 000 030-000 00 803 003.50 8800 00 000 00-000 00 000-000 003 005.50 30 0002 000 000 000 m 00 00-000 :03 33.30 2 2000.2 m 200.02 .0 2000A 2 2000mm m 500% Q 283% 0 2000mm m 20.20am < 2003-Km 030? 2.50000 20.800,» mum. 030.40 If]! 62 @wa 38: no 3255200 ddd wad- dad d3: ddd d3- ddd- ddd dddd Edd; 2;. 2d 2d- ddd 8d d3- wdd ddd dddd gmém Ed ”3 3d Ed- ddd- wwd dddm deem wdom d3 ddd 3d 3d ddd 31- 3d dddm 98M yam Ed H: Ed NE 8d. Sd- ddd dddd deem mcom dmd- add. $5. Ed 5d- dmd d3 add- ddd Ed dddd $55.50 8d. 2d- 5d- 3d 3:: 23. dd 5d dd”: 5:550 ddd d; Rd 3d 3d Ed mm: as Sd :d- ddd dddd 8:5 Ed ad Ed. 5d- ”dd- wdd- 3d 8d ddd dddd 253E ddd d5- ddd- Ed- 3; d3 Ed- 8d 2d dwd add fidafia 3d- :3. add- and dad 92. 3d. dwd d2: gamma dds d3- Ed- ddd- ddd Ed- Ed dddm 23:3 530 Ed Es- ad ddd- and $3 Sd 5d 82 053% :88 odd- 3%- 33 9%. 3d 2d 5d dmd dddd 220 dmd- ddd ddd- Ed- add $9 220 as- dmd- 23 and. 8d- mmd Edd 58:8 Ed 2d. odd mad 2:... ddd- Ed 32 888 was odd- 3.» 3d- :3. as- a: 2d Ed add $3me Mud ddd Ed- ad dad- ddd- 8d dddd 3am Ed dad 5d ed $4- dwd dddm 82338338 Rd. SE 2d 32 82 ad. ad 82 saddeséawg ddd Ed :3 odd ddd :3- Ed dddd madcaaéawdom d3- d2: dds and. RE. 5d- Sd- ddd 32 £25.82me ad 2d 8d ddd wdd d3- :d Ed add 2:2me d: ddd- d3 ddd- :3: ddd Ed- ad Edd 3:..st d2 .93 Rd 9% Ed- 5d- dwd dddd figmé Ed dud :d 3.» odd 5d Qd XS. 8.“ and- Ed dddm £932 H m U m m a o m < 58> swam: 3.x Sew 5:580 mzofiaooq .mtmm 98 @935qu 8:325 ”Tm @738 63 omda 38: no cwsidcoo ddd- Ed- Edd- ddd ddd- ddd Ed ddd SE 223 Ed Ed- Ed- Ed- ddd Ed Ed- ddd dddm dosage: 2:- dd Ed ddd ddd Ed Ndd ddd- Ed dddd Eda ddd- Ed ddd- ddd ad- ddd- ddd Ndd add add dd.» add EE- ddE Ndd- Ed 2d ddd dddm Pam ddd- Edd Ed- ddd- ddd d3- ddd- Ed ddd add .3382 Ed. Ed- ddd ddd Ed :d d:- Ed .32 5:02 Ed- ddd Ed. Edd Ed Edd Ed- ddd Ed- Ed dddd EddddN 52 dd- 3». Ed EE- ddd ddd- ddd- ddd- Ed ddd mddm EddoN 52 add a! dds ddd- 2d- Ed- Ed ddd Ed 32 953$ 52 m3 Ed Ed. Edd- ddd- 5d ad add 8.2 ddsddm Ez- ddE Ed dd- ddd- Edd- Ed- Hdd- Ed- ddd- ddd- ddd mddm ddqdzdfidz ddd ddd- Ed- ddd- :d odd ad- Ed Ed 3: mdadEfidz ddd- dds- Ed Edd- ddd ddd- dmd- ddd dddm 852 EE- ddd- dds- Ed ddd 3d. Ed. Edd dddm 8%: EE- ddE- ddd- ddd ddd 8d Ed- ddd BE 8%: ddd- Edd- Ed Ed- ddd- Ed- . Ed Ed SE 35:53 Ed- Ed. Edd- ddE- Ed- Ed- odd- d3 ddd dddd dddidedm d2- EdE Ed ddd- ddd Ed 8d- Ed dddd 88m Ed- ddd Ed. Edd- Edd Ed Ed- Edd dddd 88x Ed- d3- Ed- Edd add ddd- Ndd dddm Eadd Ed- 8d. Ed. 8d. Ed- Es Ed Ed 82 :33 dds- dds- dds- dds EE- ddE ddd Ed dddm 39: Ed- ddd dds ddd- fidd- dmd ddd- ddd dddd dad-E Ed «dd dds add Rd- add ddd ddE Edi ddd- ddd- ddd- ddd- ddd d2- Ed Edd mddd Edd: ”dd- ddd- d3 Ed- 2.“ Ed 5d dddd Edda EE ddd- Ed- Ed- ddd dad- Ed ddE Edda H m U m m Q o m d. 38> 5:82 3.x 58> fiasco 9,203 dd 24¢ 64 Ed Ed- Ed ddd- Ed dddd 88% deg: ddd ad- Ed- Ed Ed EE 83E RES Ed Ed Ed- Ed- Ed dddd 583:9 8:5 dd:- Ed Ed- ddd- ddd SE 583d 825 Ed Ed- Ed Ed Ed Ed- Ed- Ed E2 midi: Ed- Ed- Ed Ed- Ed Ed Ed- ddd SE BEE-d Ed ddd- ddd Ed Ed- ddd ad dddd Ease ddd ddd- Ed- 3d. Ed- ddd ddd SE 535 ddd fidd- ddd ddd Ed ddE @23- Ed- Ed- ddd Ed- ddd ddd- Ed Edd dds-“SEE Ed. :2 Ed Ed- Ed ddd Ed Edd dddzadad ddd Ed- ddd ddd ddd Ed- ddd- Ed dddd 88% Ed. ddd- Ed Ed. Edd Ed- 8.? Ed EE 58% Ed Ed- Edd Ed 31. Ed dddd 58d 2:. EE- Ed Ed- Ed Ed Ed dddd 5de EE- EE- ddd Ed Ed ddd Ed Ex: 5de d5 d3 ddE ddE ddd- Ed ddd ddd- ddd dddd 388% Ed- Ed- 23. Ed- Ed 9d. Ed Ed 82 2585 Ed SE SE. EE EE- Ed- ddd Ed dddd Edam Ed ddd ddd Ed- ddd fidd- ddd Ed dddd diam Ed- Ed- Ed 3d. Ed Ed- Ed- Ed 82 Edda EE- ddd- :1.- Ed Ed- Ed- Ed- ddd ddd dddd dedeom ddd ddd Ed Ed Ed- Ed Ed- Ed 82 dedeom Ed- ddd ddd- Ed Ed- 8d- Ed dddd amnion EE- Ed- ad- ddd Ed Ed ddd Ed dddd Essa E. dd. ddd Ed- ddd- :d- Ed Ed- 3d. ddd add 338 m U m m Q U m < 88> .8682 Sam 30> E880 :Czchdv “an mid-H- Chapter 4 Electoral Systems and the Dimensionality of Politics Is the political dimensionality of a nation related to its electoral system? Among other things, electoral institutions are known to shape the behavior of elites and voters, to affect the nature of representation, and to play a role in economic outcomes. Yet, their relation- ship with dimensionality is scarcely explored in existing comparative political research. As such, in this chapter I explore the relationship between dimensionality and electoral rules across several countries using the measure of dimensionality introduced in this project. The dimensionality of underlying political structures is commonly assumed, both in everyday language and academic research. When an individual describes himself as “right- wing,” he implicitly locates himself on a latent continuum, when a media report labels a candidate “liberal,” a political dimension with liberals on one end and conservatives on the other is implied, when a researcher studies the proximity between a voter and a party. she must assume a space in which the actors operate, and when studying multi-player policy creation. one must. assume the dimensions that define actors’ preferences. Accordingly, it is important to examine how dimensional constructs vary with external societal factors. A sizeable body of previous work in comparative political science focuses on explaining the number of political parties across countries. The first research tradition, pioneered by Duverger (1954) and Sartori (1976), assigns primacy to the electoral institution employed in a given nation. The second tradition, spearheaded by Downs (1957) and Lipset and Rokkan (1967), attributes explanatory power to social and ideological cleavages within nations. Later work by researchers such as Taagepera and Grofman (1985), Lijphart (1999). Taagepera (1999), and Stoll (2009), explicitly links multipartism to the number of ideological or issue dimensions in a country. I argue that political dimensionality is itself a product of electoral systems. Sartori (1976) and Lijphart’s (1984; 1999) famous analyses of democratic institutions and party systems posit such a relationship. Electoral systems that are restrictive to small party entry discourage the emergence of new dimensions as large, electorally entrenched parties have no incentive to take strong stances on emergent issues. Permissive systems, conversely, induce existing parties to take stances on nascent issues, thereby facilitating the rise of new dimensions. With the term “political dimensionality” I refer to the underlying makeup of the polit— ical space in which parties compete. The new measure of dimensionality employed in this research operationalizes this conceptualization of political space, determining how well a single dimension captures the space in which parties and voters align. Thus, if entrenched part ics choose to absorb or ignore a certain issue dimension. the measure reflects such be- havior. This is contrasted with previous indices of dimensionality, which count the number of salient issues or ideological conflicts among political parties in a country, but do not account for the transformation of such issues into political dimensions. I find that political dimensionality is functionally related to electoral institutions. In line with previous theory, I empirically demonstrate that restrictive electoral systems lead to low-dimensional political structures, while political space cannot. be captured unidimen- sionally in permissive systems. This relationship holds when other factors are controlled for. including the number of relevant political parties, and when the potential endogeneity of electoral systems is taken into account. 67 4.1 Measuring Dimensionality As discussed in Chapter 2,previous cross-national measures of dimensionality generally take specific issues into account rather than gauging the space in which parties compete. The most well-known measure of dimensionality is that of Lijphart (1984; 1999). This index is based on his subjective expert judgement of the salience of seven issue dimensions across 36 nations, averaged from 1945-1996, and has been employed as a variable of interest. in subsequent research (for example, Taagepera 1999; Taagepera and Grofman 1985). Two more recent studies make use of the Comparative Manifestos Project (CMP) (Budge, Klingemann, Volkens, Bara, and Tanenbaum 2001) to measure dimensionality. First. N yblade (2004) creates a measure of the effective number of issues (ENI) from 43 CMP issue categories in 17 West European countries from 1945 to 1999. He applies the common Laakso and Taagepera (1979) formula to weight the issues by their salience, in addition to weightng them by party vote shares. From this he creates a reduced measure, the effective number of issue dimensions (ENID). The ENID reduces the ENI when multiple parties consider the same issue to be important.1 Stoll (2005; 2009) creates postwar (1945-1998) measures of “raw” dimensionality across 24 Vt'estern countries based on seven ideological dimensions. Her measures gauge the amount of salient ideological dimensions in a polity, according to what conflicts parties deem important in their manifestos. In the measure employed in her 2009 work, to gauge the salience of each dimension, Stoll looks at the proportion of space devoted to each in the party manifestos of a given nation, takes an average across manifestos, and applies Molinar’s (1991) weighting formula.2 1Issue similarity measures are derived by pooling all countries and election years and calculating the angular separation between all issue pairs. Angular separations are essentially correlations that are bound between () and 1 rather than —1 and 1. 2Molinar’s (1991) formula was originally created as an alternative to the well- known Laakso and Taagepera (1979) effective number of parties index. 68 4.1.1 A New Measure of Dimensionality The measure of dimensionality in this analysis comes from the goodness-of-fit statistic from the unfolding routine, described in Section 2.3. A value of 1.0 indicates that voter preferences in a given nation are entirely generated by a single dimension. I subtract this value from 1 so that higher values correspond with poor adherence to a single dimension. The measure covers 79 elections.3 This measure gauges what I call “political dimensionality,” as opposed to issue dimen— sioi‘iality or ideological dimensionality. Instead of gauging the number of salient issues or the number of ideological dimensions in a country, it captures the space in which parties and voters locate themselves.4 Additionally, because it is based on voter perceptions of the parties. rather than coded party manifestos or expert opinions, it avoids the subjectivity inherent in such measures. Moreover, unlike manifesto—based measures, it is sensitive to any public conduct by existing parties that affects political dimensionality: voter percep- tions are influenced by the behavior of parties, while manifestos do not necessz-u'ily reveal true party strategies.5 If major parties in a given country purposely ignore a given issue dimension, the mea- sure will reflect this. Additionally. if an emerging issue dimension proved salient to all parties at the national level, the measure captures the associated increase in dimension- 6 ality. Nyblade’s ENID measure is the closest to the measure employed here in that it 3Because I average across Flanders and Walloon in both 1999 and 2003 to get a single measure for Belgium in this macro—level examination, the number of cases drops from 81 to 79. 4The measure maps parties and voters into the same space. That is, party move- ments throughout this space affect voters’ evaluations of the parties, as posited by spatial voting theory. 5Budge and Farlie (1983), for example, note that parties downplay diverging posi- tions on political issues in their official manifestos. but tend to emphasize differences in the media. 6’To use the example given by Stoll (2009), racial considerations became salient. in the United States in the 1930s and grew in importance through the 19608, after which they diminished in importance. There were only two major parties and a handful of very minor parties in the U.S. throughout this period. In the 19605 the space in which parties competed was likely multidimensional, as the major parties consistently campaigned on racial issues. Conversely, in recent years the salience of race has decreased and racial issues have not been brought to the political forefront. by major parties. 69 accounts for party overlap on issue dimensions. However, the EN ID is based on the word- ing of party manifestos, which, as noted, do not necessarily correspond to the strategic actions of political parties. While previous indices capture issue dimensionality or ideological dimensionality, they do not. explicitly measure the space in which parties and voters locate. Thus, they are less sensitive to the incentives electoral systems provide to political parties. The measure used in this research, alternatively, captures such incentives and thus provides a way to examine the link from electoral setups to dimensionality. 4.2 Electoral Systems and Dimensionality in The- ory T aagepera (1999, 532) posits reciprocal relationships within the three dyads depicted in Figure 4.1. Link 1 is exhaustively examined in previous research. Authors such as Cox (1997) and Norris (2004), for example, show that multipartism is related to district magni- tude, while Colomer (2005) shows reciprocity in this relationship. That is, countries with many parties tend to institute permissive electoral systems. Regarding link 2, Lijphart (1999, 88) finds a strong and significant. ('()]‘l‘(‘l&—tll()ll between the effective number of parties and the number of issue dimensions in a nation. while Taagepera and Grofman (1985) show the number of parties in a nation to equal to its number of issue dimensions plus one. Stoll (2009) also examines whether the number of effective parties in a nation is related to its dimensionality. Using her new measure of raw ideological dimensionality and Nyblade’s (2004) ENI measure, she finds evidence for this relationship in non-majoritarian electoral systems. Taagepera (1999) notes that causality may precede from the number of parties to issue dimensionality. In multiparty systems, small parties may appeal to narrow constituencies, or "favored minorities” (Myerson 1993). As such, small interest groups can select political parties as their parliamentary agents. And, as the number of parties rises, so does the amount of new issues brought to government. While this logic is intuitive in relation 70 [ Number of Parties )6 2 >[ Dimensionality J Electoral Institution Figure 4.1: Dimensionality, Electoral Institutions, and the Number of Parties to issue dimensionality, I contend that, independent of the number of parties, electoral systems affect the underlying dimensionality of the space in which parties and voters are located through the incentives they provide to major parties and voters. As noted by Lijphart (1984, 127), “In the majoritarian model of democracy, the po- litical parties typically differ from each other along a single issue dimension, the socioe- conomic or left-right dimension. In contrast, partisan differences in the consensus model are multidimensional.” Sartori (1976, 342) explains that multidimensionality will only en‘ierge in nations in which there exists an extra dimension defined by issues that existing parties are not willing to absorb; parties either ignore emerging issue dimensions or risk political suicide by taking a stance on electorally unimportant issues. That is, for multi— dimensionality to emerge, there must be a new issue dimension that is salient to the point that existing parties want to compete along it. Budge, Robertson, and Hearl (1987, 39) note the tendency of major parties to emphasize issues on which they have an advantage, rather than taking stances on issues they deem unimportant. In the American context, Petrocik, Benoit and Hansen (2003) note that candidates and parties “own issues” and campaign on those that provide them with an advantage. In addition, Przeworski and Sprague (1986) show how parties modify their agendas in anticipation of political gains within the electorate. Thus, to maximize electoral success, electorally entrenched parties may choose to sim- ply ignore a nascent dimension, signalling to voters that it “lacks merit” (Meguid 2005, 349) and keeping the dimension out of the political forefront. This behavior is likely in 71 restrictive systems, under which ignoring an emerging issue dimension is of low risk to ma— jor parties. Voters concerned with such a dimension will likely not risk casting a “wasted vote” for a small party that embodies it, but instead vote for the major party they most prefer. Alternatively, in permissive systems, if an issue important to a voter is abandoned by the major parties, voting for a smaller party that actively engages the issue dimension becomes an attractive option. Consider, for example, the cases of Germany and the United States. In both countries environmental issues have gained considerable attention in recent decades, and in both countries the Green Party has built its platform around the environmental issue dimen- sion. In the U .S., which employs a restrictive first-past-the-post (F PTP) electoral system, voters are left with little recourse if the major parties choose to ignore environmental issues; the hurdles which the Green Party must overcome to gain representation in the U.S. are insurmountable and voters recognize this when casting their vote. Alternatively, under Germany’s relatively permissive mixed electoral system, ignoring environmental is- sues is a politically dangerous strategy for the major parties; voters concerned with this dimension may punish such behavior by voting for the Greens without fear of casting an inconsequential vote.7 Thus, in permissive electoral systems entrenched parties will often choose to compete along emerging issue dimensions, and such behavior will increase the dimensionality of political space. Because voters choose based on issue dimensions they deem important (Bélanger and Meguid 2008), major parties will engage salient. emerging dimensions to decrease the electoral gains of niche parties (Meguid 2005, 349). Such a strategy induces voters to abandon the niche party in favor of an existing major party, although the niche party‘s “pet” dimension still gains political exposure; as explained by Meguid (2005), when ’ A study of the Chewa and Tumbuka ethnic groups in Zambia and Malawi by Posiier (2004) also helps to illustrate this logic. Though both countries use restric- tive, first-past-the-post electoral systems, in Malawi the two ethnic groups make up a sizeable proportion of the population. Contrast this with Zambia, where each group makes up less than 10 percent of the total population. Thus, only in Malawi do the groups have a realistic chance of winning seats in the restrictive FPTP elections. As expected, in Malawi the socioethnic cleavage between the parties is politicized, as politicians have an incentive to “ride it” into office. Conversely, in Zambia politicians must and do look for other societal cleavages to politicize. 72 mainstream parties engage smaller parties on their niche issue dimensions, the salience of such dimensions is enhanced. In sum, permissive electoral systems induce parties to take stances on emerging issue dimensions, thus increasing the dimensionality of political space. while restrictive electoral institutions lead major parties to simply ignore emerging issues. Cantillon (2001) draws similar conclusions about the dimensions of politics from a formal modeling perspective. In equilibrium, she finds that entrenched parties in restric- tive electoral systems often choose to replace, ignore, or lump together emerging issue dimensions. In permissive systems, alternatively, low barriers to party entry may lead to a two—dimensional political space; incentives to lump issues together are lower under such electoral rules.8 Previous theoretical work is unified in its predictions that systems of low restrictivcness facilitate multidimensional political space, while majoritarian institutions lead to low- dimensional constructs. Only a scarce amount, though, empirically examines the direct. link between institutions and the dimensionality of politics. Moreover, the studies that do examine this relationship focus on ideological and issue dimensionality. rather than the dimensionality of the space in which parties locate themselves. 4.2.1 Previous Empirical Tests Taagepera (1999) posits a specific relationship between the number of issue dimensions and electoral system permissiveness, which I reproduce in Equation 4.1: I = (2.15 x M3/1“) .— 1, (4.1) where I is defined as the. number of issue dimensions in a nation, which he quantities with Lijphart’s (1984) subjective judgements, and M is the arithmetic mean of the district magnitudes in a nation. Taagepera finds moderate empirical support for Equation 4.1 across 22 stable democracies. 8This discussion is based on Stoll (2005, 195). More recent work examines this relationship with updated data and measures. For example, Stoll (2005, 198) tests the relationship between dimensionality and institutions using her self—produced measure of dimensionality. She conducts a difference in means test of dimensionality between four plurality and 20 more—permissive electoral systems, but finds no support for the hypothesis that dimensionality is higher in permissive systems. Richman (2005), also using CMP data and a loose measure of dimensionality,9 tests how tightly the parties of 25 nations simultaneously adhere to a single left-right ideological dimension. He also finds no evidence that restrictive systems lead to unidimensionality. 4.3 A New Test of Dimensionality and Electoral Systems The findings of Richman (2005) and Stoll (2005) provide no evidence that dimensional- ity is higher in permissive electoral systems. In fact, only Taagepera’s (1999) moderate findings support this hypothesis. However, none of these studies explicitly examine the dimensionality of the space in which parties locate. I instead use the new measure of political dimensionality as the outcome variable in several empirical tests. Theory puts forth that, conditional on the electoral system employed, parties selectively compete along emerging issue dimensions, thus structuring the space in which they align. The measure introduced here allows me to test whether these theoretical arguments are’empirically realized. Using the new measure, Figure 4.2 depicts a bar chart of mean political dimensionality across three classes of electoral institutions. “Majoritarian” refers to institutions with single-member districts, “proportional” indicates that the district magnitude is greater than one and the country employs a proportional electoral formula, and “mixed” refers to systems in which elections are conducted across two tiers, one with single-member districts 9Richman’s dependent variable is simply the portion of the party manifestoes that are coded in the “left—right” category. 74 and another with multi-member districts.10 As expected, Figure 4.2 illustrates that nations with majoritarian systems are better- captured by a single dimension than their mixed and proportional counterparts. Due to the combination of majoritarian and proportional elements, nations with mixed systems have lower political dimensionality than proportional systems, but are not captured by a single dimension as well as majoritarian systems. Difference in means tests indicate that. the difference in political dimensionality between majoritarian and proportional systems is significant at p = .025, and the difference between majoritarian and mixed systems is significant at p = .055. The difference between mixed and proportional systems is statistically indistinguishable from zero (p : .720). 2.--,“ 1 l .25 l .2 1 Mean Political Dimensionality .15 J Majoritarian Mixed Proportional Figure 4.2: Political Dimensionality Across Electoral Institutions 10Italy switched from a mixed electoral system to a fully proportional setup during the 2006 election process. I code its electoral system as mixed because it is unlikely that the new system could have affected dimensionality in such a short time frame. In practice, the empirical analyses were not sensitive to this coding decision. 75 The district magnitude of electoral systems provides a continuous measure of electoral permissiveness (see, for example, Lijphart 1984). Figure 4.3 depicts a scatterplot with di- mensionality on the vertical axis and logged11 mean district magnitude on the horizontal axis. As theoretically expected. the figure demonstrates that political dimensionality sys- tematically increases with district magnitude. The slope of the superimposed regression line. which is statistically significant at the .01 level, is .029. The correlation between the two variables is .410. Political Dimensionality o o.— 0 1 2 3 4 5 Mean District Magnitude (Logged) Figure 4.3: Political Dimensionality and District Magnitude 11As is common practice in the literature, the log of district magnitude was taken; an increase in DM from 1 to 2 is expected to have greater effects than a change from 30 to 31. Furthermore, a loess nonparametric regression (see Jacoby 2000) of dimensionality on mean district magnitude indicated a clear logarithmic relationship between the two variables. 76 4.3.1 Accounting for other Factors Figures 4.2 and 4.3 provide preliminary evidence that political dimensionality increases along with the permissiveness of electoral systems. However, these exercises do not con- trol for other factors that may affect dimensionality across nations. Examining political dimensionality over time in nations that have reformed their electoral systems provides a way to hold other factors constant. The data at hand cover three nations that underwent recent electoral reform: New Zealand, Japan, and Peru. In 1993 New Zealand, through a national referendum, switched from first—past-the— post (FPTP) majoritarian elections to a mixed member proportional (MMP) system.” It conducted its first election under MMP in 1996. Between the CSES and the N ZES, survey data in New Zealand is available for the years 1996, 1999, 2002 and 2005.13 According to theory, dimensionality in New Zealand should increase after the switch to MMP. Though there is no data before the switch, Figure 4.4 shows that, as the new electoral system set in. diinensionality steadily rose in New Zealand through 2002, but began to level off to its 1999 level in 2005. The circumstances of Japan’s electoral reform differ in multiple ways from those of New Zealand. In 1994 Japan adopted a mixed systems for elections to its lower house. While New Zealand’s mixed system is compensatory, in that it awards seats from the PR tier to achieve as proportional a result as possible, the two tiers in Japans new mixed system were designed to operate independently (Gallagher 1998). Also, while New Zealand abandoned pure FPTP elections for a mixed system, Japan formerly used the single non-transferable vote (SN TV). SNTV, though not a “fully propor- tional" system, can sometimes lead to proportional outcomes (Lijphart 1999, 163). T hus. Japan’s switch from one send-proportional system to a marginally more proportional sys- tem. should not work to increase or decrease political dimensionality in a drastic manner. 1‘2 In MMP elections seats are allocated from the multi—member tier of the electoral system in a fashion designed to achieve as proportional a result as possible; seats won in the majoritarian tier are subtracted from each party’s seat winnings in the multi-member tier. 13The N ZES also had waves in 1990 and 1993, but, because of dissimilarities in data, it could not be analyzed here. 77 Data for Japan is available in 1996 and 2004, both falling after the switch. Figure 4.4 shows that political dimensionality is essentially unchanged in Japan from 1996 to 2004. This result is unsurprising, considering the relative innocuousness of Japan’s electoral reform. Peru’s unicameral legislature contains 120 members and is elected with open-list pro- portional representation via the d’Hondt electoral formula. In 2000, Peru used a single national district to elect its lower house. Thus, the Peruvian electoral system was very permissive to small party entry. In 2001 and 2006, the country was electorally apportioned, and legislators were elected from 25 different districts, meaning the average district mag— nitude was 4.8. Thus, the electoral system became more restrictive to small party entry, though still relatively permissive when compared to majoritarian or plurality systems. Figure 4.4 shows that political dimensionality in Peru did not immediately respond to the electoral reform, but by 2006, as expected, political dimensionality dropped in response to the more restrictive electoral rules. LO ‘1 _ 3‘". - 1% S -— lD ‘é’ “2 4 a) .E O C". e l \- cxg _l I I 1996 2006 Year —0— New Zealand --O-- Japan —---o- Peru Figure 4.4: Political Dimensionality Over Time in New Zealand, Japan, and Peru To assure that the relationship between electoral permissiveness and dimensionality is not spurious, I take other theoretically—related variables into consideration in cross- 78 national analyses. First, the number of political parties in a nation is posited to affect. issue and ideological dimensionality (Lijphart 1999; Stoll 2009; Taagepera and Grofman 1985). Though I contend that the number of parties is unrelated to the new measure of political dimensionality, I include the variable to be sure of its null effectl‘1 To measure the number of parties, I use the Laakso and Taagepera (1979) effective number of electoral parties measure (ENEP).15 This index accounts for parties that do not win representation in addition to those with parliamentary seats. Thus, parties that campaign on emerging issue dimensions but do not win seats are accounted for. I obtained the measure from the CSES.16 Because of missing data on this variable, the number of observations drops from 79 to 74.17 The causal logic derived above evokes an interactive effect of socioethnic fractional- ization and electoral permissiveness on political dimensionality. Because small parties in electorally restrictive but socioethnically fractionalized nations may win seats by empha- sizing an emerging issue within a district densely populated by a minority group (see, for example, Chhibber and Kollman 1998), entrenched parties in such systems cannot simply ignore nascent dimensions as they may in socioethnically homogenous nations. In such countries, minority issues may gain national prominence regardless of the electoral system employed. Thus, electoral permissiveness should have a. relatively weak effect on political dimensionality. 1 8 14Because the number of parties has a well-established empirical relationship with electoral permissiveness, collinearity is a concern when including it in a regression equation with district magnitude. However, the bivariate correlation between logged ENEP and logged MDM is only .304 in the sample at hand. Moreover, variance inflation factors from the multiple regressions below indicate that the variances of the coefficient estimates for either variable were never inflated by a factor larger than 1.40. 15 , ." ’ ‘ ‘E ‘ 1 r‘ ... - .' r '\" I") '7 ENEP is measured as W, where 1)] IS the proportion of \otts obtained by I 3:1 J the j“ party. I take the log of this measure to reign in outliers such as Belgium and Brazil. 16CSES Macro Data, available at http : / / www.cses.org / download / contributions / contributionsmirror. htm 17The missing country-years are Belarus in 2001, Hong Kong in 1998 and 2000, Kyrgyzstan in 2005, and the Philippines in 2004. 8Previous work on party systems also posits a conditional effect of electoral insti- 79 Canada embodies this logic. Though it employs a first-past-the—post electoral system, relative to other restrictive systems in the sample political variation in Canada conforms poorly to a single dimension. This is likely due to its socioethnically heterogeneous makeup, which is the highest of any nation sampled. Québécois issues, for example, gain national prominence in Canada due to the clustering of this minority group into certain electoral districts. Were Canada to switch to a system of proportional representation, the ability of the Bloc Québécois, a federal level Quebec-nationalist party, to bring Quebec-centric issues to government would not increase greatly, and the larger Canadian parties would likely devote a similar amount of attention to Quebécois issues.19 In relatively socioethnically homogenous nations, alternatively, an increase in electoral permissiveness should have a strong, positive effect on political dimensionality through the mechanisms put forth in Section 4.2. That is, increased electoral permissiveness allows small, issue-centric, parties to come to power. Thus, large parties must adopt emerging dimensions to prevent this ascension and avoid a loss of parliamentary seats. This effect is offset in ethnically fractionalized nations, as their heterogeneous social character affects political dimensionality, even under restrictive electoral systems. Three existing measures of fractionalization cover all of the countries in CSES sam- ple: Alesina, Devleeschauwer, Easterly, Kurlat, and Wacziarg’s (2003) measures of ethnic and linguistic heterogeneity and Kok Kheng’s (2001) measure of ethnic fractionalization. Recent work shows that the conclusions drawn from statistical analyses are sometimes dependent on the measure of fractionalization chosen (Stoll 2008). To avoid this potential pitfall, I use each of the three measures to create a single index of socioethnic fraction- alization, thereby not assignng primacy to any single measure. I conduct a principal tutions and socioethnic fractionalization (Amorim Neto and Cox 1997; Geys 2006; Ordeshook and Shvetsova 1994). However, the outcome variable is the number of political parties rather than political dimensionality. The driving theory is that. cleavages can only materialize into multiple parties if the electoral institution in place is relatively permissive. 19India, though not included in the sample, also provides an example of this logic. The Philippines also embody this logic, though excluded from the sample due to missing data on the EN EP variable. Note, however, that the Philippines, despite its restrictive electoral system, has the second-highest political dimensionality score in the sample. This is likely due to the fact that it is a very heterogeneous country, with the third highest social fractionalization score in the sample. 80 components analysis, depicted in Table 4.4 in the appendix to this chapter. The first component captures roughly 76% of the total variance in these measures, and I use ob- servations’ scores on this component to gauge their level of socioethnic fractionalization. The ethnic and linguistic measures of Alesina et a1. load20 highly on the component, as does Kok Kheng’s index. I also expect that bicameral nations are home to politics of higher dimensionality than their unicameral counterparts. Because countries with two houses of government may 21 more dimensions may come to the political have two veritably distinct party systems, forefront. Accordingly, I include a dummy variable for bicameralism, coded 1 for bicameral nations and 0 for unicameral. Transparency and openness should also increase political dimensionality. When citizens in a nation are free to exchange ideas and their votes are fairly counted, more issues will circulate and gain electoral prominence. Moreover, when the press is allowed to report. freely, more issues will enter the political discourse, including those from outside countries. To account for openness I use Freedom House scores, which gauge political rights and civil liberties.22 This index ranges from 1 to 10, and I reverse it so that high values correspond to more freedom. Finally, I account for economic development. To gauge develomnent, I use GDP per capita, adjusted for purchasing power and measured in thousands of constant international dollars. I obtain this measure from the World Bank’s Development Indicators.23 Each variable is summarized in Table 4.1. ‘OThe "loadings’ are the coeffiCIents of each variable 111 a multiple regression With the component as the dependent variable and the three measures of fractionalization as covariates, assuming each measure is standardized to a mean of zero and unit variance. 21This is especially true if the electoral rule for the upper and lower house differs, such as in Australia. 22Freedom House has a separate scale designed to gauge media freedom. Because this measure correlates with the overall freedom measure at r = .91 in my sample. I opt to use only the broader freedom measure in the interest of saving degrees of freedom and avoiding collinearity issues. 23Taiwan’s per capita GDP is taken from the VVTO and Wu (2004). 81 Table 4.1: Summary Statistics Variable Mean Std. Dev. Min. Max. 72. Political Dimensionality 0.256 0.098 0.045 0.459 79 MDM 21.946 37.514 0.800 150 79 Majoritarian 0.152 0.361 0 1 79 Proportional 0.481 0.503 0 1 79 Mixed 0.367 0.485 0 1 79 ENEP 4.636 1.570 2.174 9.761 74 Socioetlmic Fractionalization1 0 1.512 -2.253 3.364 79 Bicameral 0.620 0.488 0 1 79 Freedom -1.848 1.202 —6 —1 79 Per capita GDP 19.088 9.483 1.722 39.451 79 British2 0.190 0.395 0 1 70 Lat1112 0.177 0.384 0 1 79 IMeasure from principal components analysis depicted in Table 4.4. ZUsed only as an instrumental variable. 4.3.2 Estimation Strategies and Results I estimate the relationship between political dimensionality and the explanatory variables with linear least-squares regression.24 As several polities appear more than once in the data, the standard assumption that observations are independent must be relaxed. Thus, I cluster the standard errors by country to account for intranational correlation. There are a total of 74 observations. Results are depicted in Table 4.2. \ 24Because the dependent variable is bounded between 0 and 1, a link function technically should be used to map from the covariates to the dependent variable. I therefore estimated a model using a logistic link function. Because the results betv'veen this model and its linear counterpart did not substantively differ, I con— tlnue under the assumption of linearity to ease presentation, interpretation, and. estimation. 82 .memmficmoaa o: weigh 866.958 .mooto woewceom Umomomfio .3338 £258 4:3 woerBmm— oooo ooo.o ooo.o a A sea ammo osmo Boo mm 3. E : noooo ammo Sooo ammo Aoooo momo Eases Gooo mooo- goo oooo- as; Nooo- has same 5a some oooo so: :oo noooo :oo- goose :33 too Aooi mooo as: oeoo season Aoofi oooo- “moi :oo- Aooi omoo- so mmzm Aoeoo oooo- sea .8m x a: so: goo Eoo Cooo oooo- Aooi mooo oososeaososm ososocom Aoooo mooo Aoooo goo so 2oz floooo oooo 38:88.5 Aoooo Hooo €52 Ava—13.3 .300 33.875 umoD 33.973 «moo m3m€¢> o E52 a 7:82 H E52 mofimEEmm mQO utmozecommcgifl Hezizom Use ESSEwE/H “2.5me “ma @3st l 3 8 The first model gauges electoral permissiveness with categorical variables for propor— tional, mixed, and majoritarian systems, with the majoritarian regressor excluded as the reference category. It is clear that mixed and proportional systems have more complex dimensional constructs than their majoritarian counterparts. In mixed systems, a single dimension explains 8% less political variation than in majoritarian systems (1) 2 .023), and 11% less variation is explained in proportional systems as compared to majoritarian systems (p :- .000). In Model 2 I use district magnitude rather than categorical variables to capture elec- toral permissiveness. At the 1% significance level, logged district magnitude relates pos- itively to dimensionality. The amount of variance in dimensionality captured by district magnitude and the other covariates is just under 30%. To gauge the interactive effect. of socioethnic fractionalization and electoral permis- sivcm'ss, I include a multiplicative term in Model 3. Coefficients on continuous variable interaction terms and their constituent parts are not readily interpretable (Brambor, Clark, and Golder 2006; Braumoeller 2004).25 Thus, I graphically display the conditional effect of district magnitude across the range of socioethnic fractionalization in Figure 4.5.26 As expected, the marginal effect of district magnitude is largest in socioethnically ho- mogenous nations. Additionally, in highly socioethnically diverse nations, the effect of district magnitude on dimensionality is statistically indistinguishable from zero. This con— firms the expectations put forth above: only in nations of relatively low social diversity does the permissiveness of the electoral system affect political dimensionality. In fraction- alized nations, the effects of electoral institutions are washed out and multiple political issues may gain prominence regardless of the electoral system employed. The. effect of bicameralism on political dimensionality is positive, but the p—valne for its coefficient never reaches conventional levels of statistical significance, coming close to the 10% threshold in Models 1 and 2. Thus, the effect of bicameralism on dimensionality 251n fact, the coefficients on the constituent variables are equal to their marginal effect when the other constituent variable equals zero. 26This figure was produced with the help of code from a web supplement to Brambor et a1. (2006), available at http: //homepages.nyu.edu/~mrg217/interaction.html. 84 is essentially inconclusive, though there is weak evidence that the existence of two houses may lead to increased political dimensionality. The coefficient on the freedom variable is also positive but never reaches statistical significance, though it is near—significant (at the 10% level) in Model 2. Thus, its effect on political dimensionality is likely null. The control variable, per capita GDP, relates negatively to political dimensionality and, at the 10% level, is significant across Models 1 and 2. OLS Estimation ZSLS Estimation N N .. 53 . 3 4 0 LO Q ‘i -2 o 52 i 32 0 i 5: Socioethnic F radionalization Socioethnic Fractionalization ME. of MDM(In) on Pol. Dim. —— ME. of MDM(In) on Pol. Dim. — —— - 95% Confidence Interval — — — - 95% Confidence Interval Figure 4.5: The Conditional Effect of Electoral Permissiveness on Political Dimen— sionality 4.3.3 An Endogeneity Problem? A growing body of literature models electoral institutions as endogenous to various social. political, and economic phenomena (for a review, see Benoit- 2007). Factors such as the number of parties (Colomer 2005), the organization of economic interests (Cusack, Iversen, and Soskice 2007), and political ambition (Bawn 1993; Benoit 2004; Benoit and Hayden 2004) have been used to explain institutional choice. Relevant to the research at hand, Taagepera (1999) specifically posits that the number of issue dimensions in a nation may influence its choice of electoral system. In the case of endogeneity a single—equation least~squares model will return biased and inconsistent parameter estimates.‘27 Accordingly, I use instrumental variables to create a proxy of district magnitude that is exogenous to political dimensionality. To statistically examine whether endogeneity is a problem, I employ a Hausman test28 of the null hy— pothesis that district magnitude can be treated as exogenous, or that the equations can be consistently estimated with OLS. The regressor was not found to be problematically endogenous, with a p-value of .471. However, because previous theory points to a potential emlogcneity problem, I continue under the assumption of endogeneity, modeling electoral institutions with instrumental variables. Good instruments are correlated with the endogenous right-hand-side variable in each equation, but uncorrelated with the error term. Therefore, I obtain variables that are the— oretically related to electoral permissiveness, but unrelated to dimensionality, as predictors of the mean district magnitude variable. As noted by Blais and Massicotte (1997), colo- nial background influences the selection of a particular electoral setup in multiple ways. Thus, following Persson and Tabellini (2003, 129), I employ two dummy variables as in— ? . . . . . . “7Est1mates of the structural coeffic1ents 1n nonrecurslve models Wlll be biased and 111cOHS1stent 1f estimated with ordinary least squares because the OLS assumption that the dlsturbances are uncorrelated with the explanatory variables is violated (Gujarati 2003, 725). To illustrate the endogeneity problem, imagine the system of equations Y1 = 310+B12Yz+711X1+N1 Y2 = 320+321Y1+722X2+u2 If 11] increases, Y1 increases. This, in turn, leads to an increase in Y2, which itself increases Y1. Thus, the disturbances will be related to the right- hand side variables Y1 and Y2. The relationships between 11 and Y2 and the dependent variables are inflated (biased) because the slope of each variable gets "credit” for increases in the disturbances (Kennedy 1998,138). The estimates are inconsistent because this bias does not decrease as n —> oo (Gujarati 2003, 726). 28Included in the wreg? pac kage for Stata by Baum, 1 c. ll‘dlIt r, and Stillman (200 i) 86 struments, one capturing nations with British-influenced institutions and another nations with Iberian heritage. Both variables are binary, coded 1 if the country meets the criterion, and 0 otherwise. Because Commonwealth nations often adopt a variant of Great Britain’s FPTP electoral setup, a British history leads nations to have relatively restrictive elec- toral systems. Similarly, Latin American nations generally opt for versions the permissive electoral systems employed by their colonizers. Each instrument is sumn‘iarized in Table 4.1. To test the validity of the instruments in both equations, I employ a Hansen-J test29 of the null hypothesis that the instruments are uncorrelated with the error terms and that excluded instruments are correctly withheld from the equations. The null hypothesis is not rejected at any conventional significance level, with a p—value of .445. “J Included 111 the wregQ package for Stata by Baum et a1. (2003) 87 @3583me E EOE .5355pr E mmfigé E37035 .885 pudendum @2833 33:8 £358 53» vBmESmm 08s 25s a A sens was come we E E .2 Sci Sec 88 V was amass as: wood. $8 V mood- n50 Same as so: was so: Sod 888$ g3 83 $3 53 $5885 35 Bed. $3 83. E: mmZm 83a 33.. see sow x 25 :52 83V ass 53 08s cosszessssa oefisoeom 33 a: $3 $3 a: 29>. fies—97$ .300 33.875 .mooU m. 682 a Eco: master mofizcsmm mqmm rfizefizmcmfiwm Eoflzom use owzésmfiz SwBflQ “ma. @3de 88 I use two—stage least squares (2SLS) to proxy district magnitude and, in turn, predict political dimensionality. I again employ country-clustered standard errors. The results are displayed in Table 4.3. The resulting coefficients are very similar to the OLS results, though significance levels rise slightly. The interaction between electoral permissiveness and socioethnic fractionalization, which is depicted graphically in Figure 4.5, also behaves similarly, though the wider 95% confidence interval demonstrates that the marginal ef- fects of the instrumented district magnitude variable are estimated with less precision. Nevertheless, even when the potential endogeneity of electoral institutions is taken into account. there remains a clear and significant link from electoral permissiveness to political dimensionality.30 Moreover. in all five model specifications, the effective number of parties has no sta- tistically discernable effect on political dimensionality. The measure of dimensionality employed here, which captures the structure of the space in which parties align, rather than the number of ideological or issue dimensions, as in Stoll (2009) and Lijphart (1999), is unaffected by the number of political parties. Instead, the incentives provided to po- litical parties by the electoral system to which they are subject are what determine the underlying makeup of political space. 4.4 Conclusion Political dimensionality is often referenced in the media, and individuals tend to think of government and ideology in dimensional terms. Reporters and pundits regularly refer to politicians and parties as “right-wing” or “left—wing,” and individuals often label them- selves as “liberal” or “conservative.” Such labels assume a unidimensional and bipolar underlying political construct; even the most fundamental descriptions of politics rest on dimensional assumptions. Scholars rely on dimensionality when studying party systems, the locations of voters, JOBecause sufficient historical data on dimensionality is unavailable. I am 1111- able. test whether complex dimensional configurations lead constitutional framers to adopt pern11ss1ve electoral systems. 89 politicians, and parties within countries, and the formulation of public policy. Research on the congruence of party and voter locations either implicitly or explicitly assumes a. certain dimensional construct when locating political actors in Space. Additionally, studies of the formation of public policy or governing coalitions in the “veto players” tradition (Tsebelis 2002) must assume the number of dimensions political actors align upon when coming together to make decisions. Cruisequently, it is important to study what leads to certain dimensional configurations. Previous theory posits that entrenched parties will compete over fewer issues when electoral systems are restrictive, thus lowering political dimensionality. Conversely, in permissive systems, parties are inclined to adopt emerging issues out of the fear of losing parliamentary seats, thus increasing the dimensionality of the underlying political space. In this research I employ a new measure of dimensionality derived directly from voter preference data. By quantifying the space in which voters and parties live, I am able to systematically explore theoretical predictions of dimensionality’s relationship with politi- cal and social factors. I find that electoral institutions strongly affect a nation’s political diu1c11siouality, especially in socioethnically homogenous countries. Due to the differing in- ccntives pI‘(_)Vl(l(.‘d to political parties under each institutional structure, permissive electoral systems lead to complex underlying political constructs, while the politics of restrictive systems tend to conform well to a single dimension. 90 4.5 Appendix to Chapter 4: A Combined Mea- sure of Socioethnic Fractionalization I employ a principal components analysis to distill the common variance in three measures of social heterogeneity: Alesina et al.’s (2003) measures of ethnic and linguistic hetero— geneity and Kok Kheng’s (2001) measure of ethnic fractionalization. The eigenvalue for the first component is 2.285. Because it captures over three-fourths of the variance in the three measures, the remaining components are disregarded. Table 4.4 summarizes the analysis. Table 4.4: PCA of Fractionalization Measures Loading on Component Alesina et a1. (2003) Ethnic Heterogeneity 0.6116 Linguistic Heterogeneity 0.5154 Kok Kheng (2001) Ethnic Fractionalization 0.6002 'n. 79 Eigenvalue of Component 2.285 Variance Captured by Component 0.762 91 Chapter 5 Electoral Behavior and the Dimensionality of Politics: A Cross-National Examination of Proximity Voting The defining characteristic of democracy is the right of citizens to choose their represen- tatives. As such, a vast amount of political research examines what leads voters to a particular choice. The purpose of this research, alternatively, is not to examine what vot- ers choose, but instead how they reach their decisions. More specifically. I examine what individual, institutional, and societal factors lead voters to follow or violate the proximity model of voting. I also expand on previous studies by examining the relationship between proximity voting and a country’s underlying political structure, or “political dimensional— ity.“ I again conceptualize political dimensionality as the degree to which political conflict in a nation can be captured by a single dimension. The proximity model of voting (Downs 1957; Hotelling 1929) is generally accepted as an accurate portrayal of voter behavior in the political science literature. Put simply, the model predicts that voters choose the candidate or party closest to them on some ideo- 92 logical continuum in any given election. Elegant and intuitive, the theory has withstood numerous empirical and theoretical tests throughout the past several decades (see, for example, Blais, Nadeau, Gidengil, and Nevitte 2001; Westholm 1997). Nevertheless, many voters deviate from the proximity model of voting. This may be because they are unable to cast a well-reasoned ballot due to some individual-level cl'iaracteristics or societal-level factors. On the other hand, they may choose to strategically vote for a party that is not closest to their ideal point. Finally, they may employ an entirely different criterion when casting their ballots. The directional theory of Rabinowitz and 1\v‘lacdonald (1989), for example, challenges proximity notions, predicting that voters choose parties that are on their side of issue space, but take extreme positions.1 In addition, the discounting theory of voting puts forth that individuals “discount” campaign promises, taking into account the actual policies they expect governments to produce once in office (Merrill and Grofman 1999; Tomz and Van Houweling 2008). Recognizing that voters have incomplete information about candidates and parties, Lau and Redlawsk (1997) define a correct vote as one that a citizen would make if he or she had full information. The authors operationalize this definition two ways. First, in an experimental setting, information was divulged to subjects only after they voted, and rcspomlents were then asked whether they would switch their vote. Second, the authors used survey data to compare whether voters chose the candidate closest to them on the issues. If so, the voter was deemed to have “voted correctly.” In later studies, Lau, Redlawsk, and coauthors expand their examination of correct voting using the second operationalization. Lau, Andersen, and Redlawsk (2008a), for example, examine the effects of individual-level factors and campaign characteristics on the probability of casting a correct vote in American presidential elections. Lau, Patel, Fahmy, and Kaufman (2008b) conduct a similar analysis, studying correct voting across cmmtries. Finally, Lau and Redlawsk (2008) examine how the propensity to vote correctly in the United States varies with age. 1Parties must be within a region of acceptability for voters to allow them to enter the1r decision calculus. 93 The idea of a “correct vote” is normatively pleasing, and those concerned with the representational characteristics of democracy would hope that voters tend to choose the party that suits them best. However, outside of the laboratory, operationalizing a “correct" vote is very difficult. A strategic vote is surely “correct,” but it does not. necessarily follow either spatial model of voting (directional or proximity). Abstaining 111ay also be the “correct” vote if the nonvoter is unconfident in the democratic process, or perhaps implicitly supporting the status quo by choosing to stay at home (Lau et al. 2008b). Thus, this project does not directly consider whether voters choose “correctly,” but rather whether they vote according to the proximity model. Therefore, I am unable to normatively evaluate the various countries and democratic institutions studied. However, I develop a clear picture of how the nature of voting behavior varies with individual characteristics and across institutional contexts. Moreover, determining how voters choose is a first step to finding out if they do so “correctly.“ Previous comparative work on proximity voting is quite limited. As such, country—lwel explanations of proximity voting violations are scarce, though the number and spread of parties, the age of institutions, the electoral system, wealth, and information availability have all been empirically linked to the nature of voting. In this study I add to these macro- lcvel explanations by considering the dimensionality of political space. I make the simple prediction that following the proximity voting model is more difficult when political space is complex. I quantify dimensionality with the measure developed in Chapter 3. I also add to the existing literature by introducing a new measure of party and voter positions - also developed in Chapter 3. Because these measures are less reliable when political dimensionality is high, I also employ expert—derived measures of party and voter locations to assure the robustness of the empirical findings. In sum, I test the predictive power of the proximity model across several nations. At the individual level, I find party identification and political efficacy to have the strongest links to proximity voting. At the election level, the number of parties, compulsory voting, and dimensionality all affect one’s likelihood of following the proximity logic in the voting booth. Thus, in addition to personal characteristics. institutional and election—specific fac- tors affect a voter’s decision calculus. These findings elucidate the nature of voting across countries and provide important insights to those interested in normatively examining the representational qualities of democracy. 5.1 Proximity Voting in Theory As articulated by Downs (1957), voters derive the highest utility from the election of the party closest to their ideal point. Formally, on a single dimension, the utility of voter 1' for party j is given as uij = _(v21 — pj)2a (0-1) where e,- is voter i’s ideal point and pj is location of party j. If a given voter chooses the party that maximizes uij, he is a proximity voter. If, however, he chooses another party, he is a proximity voting violator. The theory of proximity voting has withstood numerous theoretical and empirical tests (see, for example, Blais et a1. 2001; Kramer and Rattinger 1997; Pierce 1997; VVestholm 1997) and has become the foundation of voting behavior research (Adams, Merrill, and Grofman 2005, 17). Though the utility function given in Equation 5.1 is generalizable to more than one di- mension, for simplicity this research constrains parties and voters to locate along the same continuum. This continuum is thought of as the political “super dimension” of Gabel and Huber (2000), which constrains party positions over several issues. Previous research on the tendency of voters to follow the proximity logic imposes this same constraint (Beatright 2008; Wessels and Schmitt 2008). Chapter 4 shows that a single dimension is strongest under restrictive electoral systems. Numerous individual- and (xnrntry-lcvcl factors affect the utility calculus depicted in Equation 5.1. That is, voters may find it appealing to abandon the party closest. to them due to some external constraints or individual-level considerations, thus casting a strategic vote. Alternatively, certain personal or societal factors may lead voters to mistakenly Choose a party that is not the most proximate. Adding the parameter a?” to Equation 5.1 95 accounts for individuals’ unique considerations over each competing party or candidate, giving uij = ’(W — leg + 37z'j- (53-?) Thus, there are two separate mechanisms that may lead to proximity voting violations. First, under certain conditions, voters may not have the ability to make an informed choice. Alternatively, they may decide to abandon the party closest to them for rational reasons, thus casting a strategic vote. If, in the words of Key (1966, 7), “voters are no fools,” does this lack of foolishness lead them to violate 0r conform to the rules of proximity voting? 5 . 1 . 1 Individual-Level Factors Previous academic work associates a host of characteristics with an individual‘s vote choice, including partisanship, economic evaluations, issue positions, candidate evaluations, and socioeconomic status.2 Less research, however. examines what individual factors affect how people vote. Here I consider what micro-level factors shape an individuals decision calculus. People who are knowledgeable and interested in politics are conscious of political par- ties and their positions. Moreover, such individuals likely have well—thought out stances on salient issues and are likely to be able to identify the party they are closest to. As such, Tomz and Van Houweling (2008) show that highly educated voters are more likely to follow the proximity logic. However, these same individuals are also likely to know when to vote tactically and abandon the proximity logic. Thus, the effect of political knowledge and interest on proximity voting is unclear. In addition to education and interest, previous work links age and income to one‘s ability to cast an informed vote. First, growing old leads to a decline in cognitive abil- ities (Salthouse 2004), and old age can thus can hamper an individual’s ability to vote 2Dalton (2000) provides a comparative review of this literature. 96 accurately. As such, Lau and Redlawsk (2008) find “correct” voting to decline with age. Regarding income, if a voter has her basic needs met, she can spend time researching elec- toral choices, rather than worrying about her next meal. Thus, richer individuals should be less prone to random, uniformed voting. However, such individuals may again be more likely to cast a tactical vote, due to their clear view of political reality. In fact, Boatright (2008) tests the effect of income on proximity voting and finds inconsistent effects across Republican and Democratic groups in the U.S. Campbell, Converse, Miller, and Stokes (1960) put forth that Americans" partisan identifications. instilled at a young age, can affect voting behavior throughout their life. Examining this relationship cross-nationally, Green, Palmquist, and Schickler (2002, 165) note that party ID is “related to, but not identical with, the vote.” Strong partisans tend to vote based on a psychological attachment to a party, even if their preferred party does not necessary fall closest to their ideal point. Thus, such voters may violate the principles of proximity voting. Indeed, Tomz and Van Houweling (2008), in an experimental setting, find that strong partisans are less likely to vote according to the proximity logic than independents. Alternatively, individuals may identify with a party because it falls close to their ideal point. In this case, having a party identification will increase the odds of a proximity vote. Again, the relationship between partisan identification and proximity voting is unclear. Regarding efficacy, individuals who feel the political process is valid are likely to cast informed votes, whereas those who see politics as distant, non—responsive, or meaningless are prone to choosing randomly, if they decide to vote at all. Thus, individuals who see some value in politics have a higher likelihood of voting proximately. Clearly, previous theory is scattered as to its individual-level predictions about proxim- ity voting behavior. A. single personal characteristic often has numerous factors associated with it, and these factors exert opposing forces on voters, either pulling them away from or toward proximity voting considerations. 97 5.1.2 Country- and Election-Specific Factors Incentives and constraints vary with societal conditions, and previous theory examines how the tendency for voters to employ the proximity logic shifts with such conditions. However, like with individual-level factors, separate directional predictions are often associated with a single variable. For example, an emerging body of research explicitly links the nature of voter l‘)chavior to electoral systems. Kedar (2005) theorizes that voters in majoritarian systems will choose the party closest to their ideal point, as they can safely assume that this party, if victorious, will not be impeded in the implementation of its policy goals. Contrarily, voters in proportional systems, which are often characterized by power sharing among parties, are likely to discount future outcomes and vote tactically for extreme parties, as they are aware that policy will be watered down by institutionalized bargaining. Several emerging papers firirther-analyze the findings of Kedar (2005), showing that voters do, in fact, cast strategic votes based on coalition preferences (Bargsted and Kedar 2008; Bowler, Donovan, and Karp 2008; Duch, May, and Armstrong 2008; Meffert and Gschwend 2008). In this vein, Karp and Banducci (2002) examine directional and proximity voting in New Zealand, a country which experienced a switch from plurality to mixed-member proportional rules. and find that the proximity model of voting held less predictive power under proportional rules. According to this logic, proximity voting should be observed less in systems in which one party does not generally win an electoral majority, i.e. proportional systems. On the other hand, proportional systems are permissive in that they do not erect high hurdles for parties to gain parliamentary representation. As such, voters in proportional systems are free to choose sincerely; they do not have to cast a tactical vote for a less liked or less proximate party out of fear of “wasting a vote.” In restrictive systems, on the other hand, voters must often choose parties that they do not necessarily consider their first choice in order to prevent the election of a more—disliked party.3 Thus, relative to restrictive systems. 3A prominent recent example is the 2004 U.S. presidential election, in which some voters abandoned their first preference, Ralph Nader, for the Democrat, John Kerry. 98 proportional systems may actually lead to an increase in proximity voting considerations. As noted by Budge and Farlie (1978), voting cannot be adequately examined without regard to party competition, and vice versa. As such, Wessels and Schmitt (2008) examine the impact of the number of parties, the range of their positions, and the dispersion of their distribution on the tendency of voters to choose proximately. They posit that more choices, and more differentiation among these choices, makes it easier to find a party suitable to one’s preferences. Alternatively, it is plausible that increasing the number of parties makes it harder to “correctly” discern which party is most proximate.4 Nevertheless, they find that as the effective number of parties rises, proximity considerations become stronger. The same is true for the range of party locations. However, when parties spread out too sporadically within this range, they find that proximity considerations no longer explain one's vote choice. Also examining party systems, Lachat (2008) argues that high levels of party system dispersion reinforce voters’ reliance on ideological criteria when voting. This is because parties place greater emphasis issue positions in polarized systems. Thus, the issues associ- ated with ideology should become “more easily accessible to voters” as dispersion increases (688). This should, in turn, increase the likelihood of a proximity vote. Lachat finds sup- port for his expectations with data from Western European democracies. Similarly, in the British setting, Green and Hobolt (2008) show that as polarization increases, voters are more likely to choose according to ideological or spatial criteria, as opposed to competence considerations. A country’s experience with democracy and its overall wealth may also affect. the na- ture of voting. Todosijevic (2005) hypothesizes that voters have difficulty recognizing the exact positions of parties when democracy is still young and finds support for this predic— . . r . . . a tron 111 Hungary.” Thus, older countries Wlth an umnterrupted history of democracy may The assumption was that a vote for Nader would make it more difficult for Kerry to defeat the incumbent, Republican George W. Bush. 4This is the logic given in Lau and Redlawsk (1997) and the recent studies by Lau and co—authors that have built from it. r 0 r no o . . . ”In fact, I‘odosuewc finds some ev1dence that Hungarian voters follow the direc- tional model of voting. 99 experience higher levels of proximity voting. Regarding wealth, through the same mech- anisms discussed at the individual level, richer countries should have a higher incidence of proximity voting. That is, citizens in countries that provide a basic level of comfort will have the time and resources to research electoral choices and make an informed deci— sion when voting (Lau et al. 2008b). However, regardless of wealth and democratic age, without a free and fair press, voters will not have the ability to select the most proxi- mate party; information availability is a necessary condition for proximity voting. Thus, proximity voting should be observed more in nations with a free media. Finally, even in strong, free, and rich democracies, those who are forced to participate will likely not make an informed decision. Compulsory voting forces disinterested and uniformed citizens to vote (Jackman 2001). Thus, proximity considerations will be lower when voting is coerced, as voters may choose essentially at random when in the polling both. Hines (2006) tests whether “correct” voting, as defined by Lau and Redlawsk (1997), is related to compulsory rules, but finds no evidence for the assertion when other factors are accounted for. 5. 1 .3 Dimensionality With the term “political dimensionality” I refer to the makeup of the political space in which parties compete and voters locate. The logic behind the relationship between political dimensionality and proximity voting is straightforward: as political space becomes more. complex, it becomes more difficult for voters to locate the most proximate party. More choices may make voting for a proximate party more difficult simply because it is easier to get it wrong. The same is true for the number of political dimensions. Alternatively, if a single political “super dimension” (Gabel and Huber 2000) effectively constrains party positions over several issues, proximity voting should be relatively “easy.” In a unidimensional setting, voters can discern between the parties (or groups of parties) to their left and their right, and pick the one with the ideal point closest to their own with little. effort. However, in several dimensions, the decision calculus becomes much more. involved, and therefore less voters are able to select the most proximate party. Thus, the 100 political dimensionality of a nation will be inversely related to the tendency of citizens to cast a proximity—based vote. Previous research does not consider the relationship between political dimensionality and the likelihood of voters choosing proximately. 5.2 'Research Design and Methods I’i'evioiis theory indicates that proximity voting is affected by individual— and national—level factors. That is, the at” term in Equation 5.2, which accounts for individual considerations over the utility derived from each party, is affected by both personal factors and the political environment in which elections take place. In general, the directionality of the relationships between proximity voting and these factors is tenuous. Using CSES data, I attempt to elucidate the nature of these links. 5.2.1 Constructing the Dependent Variable The dependent variable simply gauges whether individuals voted proximately in legisla- tive. lower house, elections. There are three measures needed to construct this variable: the location of the voter, the locations of the parties, and intended vote choice. I use the unfolded party and voter locations produced in Chapter 3 to obtain the former two measures. To gauge vote choice, I rely on self-reported vote data from the CSES. If the respondent voted for the party she was closest to, I assign a 1. If not, I assign a 0. I do not consider respondents who reported abstaining or did not report their vote choice.6 To assure the robustness of the findings, I also code a dependent variable based on expert party placements and self-reported voter locations. The CSES asks voters to lo— cate themselves along a left-right continuum7 and also gives expert-provided locations of political parties along this continuum. I apply the same coding scheme to these questions, ()In mixed electoral systems voters cast a vote in a multimember district and a vote in a Single member district. In such systems I consider the single member district vote. In addition, I do not consider second-round voting in run-off systems. 7The question states: “In politics people sometimes talk of left and right. Where would you place yourself on a scale from 0 to 10 where 0 means the left and 10 means the right?” 101 again assigning proximity voters a 1 and non—proxii‘nity voters a 0. It is also possible to obtain measures of party positions based on individuals’ percep- tions of the political parties, as the CSES asks respondents to locate parties along the left-right continuum. Measuring party positions with such responses, however, can be problematic. Respondents may place their most-preferred party closer to their own posi- tion. regardless of that party’s true position (Adams et a1. 2005, 170). Additionally, they may shape their responses to meet the proximity voting criterion (Boatright 2008). Such ratioiializaticn‘is of perceptions lessens the reliability of individual placements (Macdon- ald, Rabinowitz, and Listhaug 2007). Thus, I choose not to construct a third dependent variable with such data. 5.2.2 Measuring Dimensionality No previous research on the nature of voting (and very little research in general) examines whether voting behavior is related to dimensionality. To test this, I measure dimensionality using the index developed in Chapter 3. This measure captures how well a single dimension captures the variance in party and voter locations in a nation. I subtract it from 1.0 so that higher values correspond with poor adherence to a single dimension. I also multiply the value by 100. Thus, it captures the percent of variance in party and voter locations not captured by a single dimension.8 The variable changes over elections and is thus observed across 79 elections and 43 countries,9 listed in Table 5.2.10 8The unfolding method used to create the dimei'isionality measure is itself based on a spatial proximity model. More specifically. it determines how well voter pref— erences fit a proximity model in one dimension. However, it operates independently of voting behavior. Thus, a finding that preferences fit well to a proximity model does not necessarily mean that voting will follow a proximity logic. ( , . . 'l The Flanders and Walloon regions of Belgium are treated separately for both 1999 and 2003 in this analysis, as they have distinct party systems and dii‘iiensional configurations. 1” Due to incongrucnce of data, the elections in New Zealand in 1999 and 2005 are. not included here, as in Chapters 4 and 6. 102 5.2.3 Individual-Level Variables To test the individual-level theory discussed above, I use several variables from the mass portion of Modules 1 and 2 of the CSES. Though there are no questions across both modules that directly gauge political interest and participation, the CSES does report individuals” education levels and their ability to correctly answer three trivia-type political questions. Thus, to gauge overall education, I create a dummy variable differentiating university graduates from others. As the trivia-type questions vary widely in both difficulty 11 Moreover, correct responses to and content, their cross-national comparability is low. these questions are strongly associated with one’s level of education. Accordingly, I opt to exclude such questions and proxy political sophistication with the education variable. I interact this variable with dimensionality with the expectation that more sophisticated individuals will be able to identify the most proximate party, regardless of the underlying makeup of political space. To capture whether or not an individual identifies with a. party, I include a. variable gauging partisan closeness.12 This is again a dummy variable, coded 1 if an individual has a party ID and 0 otherwise. I gauge political efficacy with a CSES question which inquires as to whether the respondent feels that his vote makes a difference in the political process.13 The variable is split into five categories, with higher values corresponding to more. political efficacy. To gauge household income the CSES separates respondents into quintiles. I use this measure to capture respondents’ economic well-being. In addition, I include age to account :for the changes in the nature of voting that may occur as one gets older. The variable 11For example, in Australia in 1996 one question asks respondents, “True or false, no one may stand for Federal parliament unless they pay a deposit”. In the United States in 1996, a question asks respondents to name the Vice President. Clearly, the difficulty and substance of such questions is divergent. 12The question asks: “DO you usually think of yourself as close to any Particular political party?" 3Question wording: “Some people say that no matter who people vote for, it won‘t make any difference to what happens. Others say that who people vote for can make a difference to what happens. Using the scale on this card, where would you place yourself?” 103 is measured simply as the respondent’s age in years. Fii'ially, I include a gender control variable, coded 1 for females and 0 for males. 5.2.4 Country- and Election-Level Variables Aside from dimensionality, several election-level factors are theorized to affect the nature of voting behavior in legislative elections. I rely on data from several sources to operationalize these variables. First, regarding the characteristics of party systems, the all'loul'lt of parties and the variance in their positions may have an effect on the nature of voting in each legislative election. To measure the number of parties in each election, I use Laakso and Taagepera's (1979) effective number of electoral parties measure (ENEP).14 I obtained this i‘iieasure from the CSES Macro Data.15 Under the assumption that a shift from 2 to 3 parties has a greater impact than a shift from 7 to 8 parties, I take the log of this index. 1 gauge the dispersion of the parties using a measure from Ezrow (2007), which draws on the work of Alvarez and Nagler (2004). The measure takes the divergence of each party from the weighted mean position into account, as well as the vote share of each party.16 Of course, party positions are measured with either the unfolding results or the CSES expert positions depending on which dependent variable is employed.17 The theory above puts forth that voters may abandon the most pI'OXllnaiP. party by discounting outcomes in anticipation of institutionalized bargaining (for example, Kedar 2005). The pern'iissiveiiess of electoral systems captures the propensity for such bargaining. In countries where more parties can gain representation, the chance that they will have to enter into coalitions to form a government, or work together to formulate policy, increases. 1.1.1.1 _ . . 1 . . . ie formula is iven as , where U.‘ is tli , ro or .10 (f v ite‘ )b .-i e ( g 2".” 7’. j e p p t n ) ( s ( ldll d I .7=1 .7 by the jf’ ' party. r 0 1"Available at http:/ / www.cses.org / download / contributions / contributionsiiiirror.htm leor each election, the formula is “23:12),- (pJ- _ 132, where ej is party J5 vote proportion, pj is the position of party j, and 13 is the weighted mean party position. 17Some measures of dispersion are weighted by the overall distribution of voters. However, because this is a micro-level examination of voting behavior, the overall voter distribution in each election is irrelevant. 104 To gauge permissiveness I rely on the logged mean district magnitude, which provides a continuous measure of electoral permissiveness (see, for example, Lijphart "1984). I obtained the data from the CSES. To capture whether peeple are legally coerced to the ballot box, I include a dummy variable for compulsory voting. The variable is coded 1 if voting is mandatory and 0 otherwise, and again comes from the CSES Macro Data. To gauge to freedom and openness I use Freedom House scores, which account for political rights and civil liberties. The index ranges from 1 to 10, and I reflect it over zero so that higher values correspond to more freedom. 5.2.5 Model Specification and Methodology The. data are observed at three levels: individuals (level-1) participate in elections (level- 2), which themselves are nested in countries (level-3). Regarding the specific variables, the number and spread of the parties and political dimensionality vary both over time within countries and spatially across countries. Freedom and district magnitude vary greatly across countries but are relatively stable within the countries that are observed at multiple elections. The only variable that is static over time across all countries is the compulsory voting dummy. Nevertheless, because there is very little within-country temporal variance in the macro-level variables and most countries are observed at only one time period, I opt for a simpler two-level model in which I consider individuals (level-1) to be nested within elections (level—2). In effect, I allow the intercept to vary across each election in each country. As the dependent variable is dichotomous, I use a logistic link to map from the independent variables to the probability of voting proximately. The resulting model is given in Equation 5.3: logit[Pr(PVij 2‘— IIXU, (1)] : X113 + Cj, (5.3) where PVU- represents the probability of individual 13 in election j voting proximately and X.”- is a matrix of the explanatory variables and the overall constant. 3 is a vector of coefficients on the explanatory variables, and 9 represents the election—specific random intercepts. These capture any election—specific factors not accounted for by the covariates that. may affect the likelihood of voters choosing proximately (Ralie-Hesketh and Skrondal 2005). Several of the independent variables are theoretically and empirically linked in previous research. There is a well-established link from electoral permissiveness to the number of parties and Chapter 4 shows that dimensionality is also a function of electoral systems. Moreover. Chapter 6, in addition to previous research (see, for example, Cox 1990; Dow 2001; Ezrow 2008), uncovers a relationship between the spread of parties and electoral institutions. Thus, collinearity among these variables is an issue. However, because each variable is theoretically linked to the nature of voting I opt to include them; rather than risking a misspecified model I choose to suffer from the possibly-inflated standard errors that arise from collinearity. Also, the theoretical links from country-level wealth and individual income to the nature of voting are identical; economic comfort gives individuals more time and resources to research parties and ponder their vote. Measuring wealth solely at the country level aggregates away the individual-level variance in incomes within nations. Thus, I gauge wealth only at the individual level to test this link. Each variable is summarized in Table 5.1. For each measure of proximity voting, about 45% of respondents follow a Downsian logic. Because the party and voter placements from the unfolding results cover a much wider scale than the expert placements, the dispersion variable has a higher mean for the former (both are non—negative). lV-Iissing data arises in the sample at both the individual and election levels. At the individual level, the CSES asks a nearly identical battery of questions across each election. However, in some of the elections included in this study, questions needed for the variables of interest were not included, thus leading to missing data for entire elections. In other election years, there is random missing data across individuals, though it is not serious 106 Table 5.1: Summary Statistics Variable Mean Std. Dev. Min. Max. in. Individual-Level Variables Proximity Voting (Expert) 0.429 0.495 0 1 57610 Proximity Voting (Unfolded) 0.455 0.498 0 1 59855 Age 45.653 16.876 15 102 120461 Gender 0.523 0.499 0 1 125214 Education 0.152 0.359 0 1 124017 Income 2.935 1.374 1 5 102504 Party Identity 0.458 0.498 0 1 116710 Efficacy 3.809 1.299 1 5 117369 Election-Level Variables Dispersion (Expert) 1.860 0.583 0.710 3.819 63 Dispersion (Unfolded) 3.687 1.269 0.525 7.443 69 ENEP 4.795 1.733 2.174 9.761 7" MDM 21.506 37.579 0.800 150 79 Political Dimensionality 25.277 9.967 4.465 45.924 79 Compulsory 0.241 0.430 0 1 79 Freedom -1.854 1.199 —6 —1 79 enough to warrant dropping the entire election from the data set. Regarding election-level variables, because party vote shares and locations are needed to construct the measures of dispersion, these variables are missing when either party placements or aggregate election results are unavailable. The effective number of parties data are also unavailable for some elections. In addition, as the theory and mechanisms above correspond to legislative voting, I do not consider any elections which were purely executive. 18 Finally, as noted, to construct the dependent variables, data on the placements of the parties. the placements of the individuals, and individuals’ intended vote choices are needml. Thus, elections in which any of this data are unavailable were dropped from the analysis. Table 5.2 displays all elections covered in Modules 1 and 2 of the CSES and indicates which were included in the sample. Though several elections had to be discarded, there remain 56 elections in the sample using expert locations and 60 in the sample using unfolded locations. These elections cover a broad range of countries with 18In the CSES sample, Belarus in 2001, Chile in 1999, and Russia in “2000 and 2004 conducted only presidential elections. 107 respect to electoral and party systems, level of development, and world region. Table 5.2: Included CSES Elections Country Election Year Expert Unfolded Albania 2005 X X Australia 1996 X X Australia 2004 X X Belarus 2001 Belgium-Flanders 1999 X X Belgium-Flanders 2003 Belgium-Walloon 1999 Belgium-Walloon 2003 Brazil 2002 X X Bulgaria 2001 X X Canada 1997 X X Canada 2004 X X Chile 1999 Chile 2005 X X Czech Republic 1996 X X Czech Republic 2002 X X Denmark 1998 X X Denmark 2001 X X Finland 2003 X X Ffance 2002 Germany 1998 X X Germany 2002 X X Hong Kong 1998 Hong Kong 2000 Hong Kong 2004 X X Hungary 1998 X X Hungary 2002 X X Iceland 1999 X Iceland 2003 X X Ireland 2002 X X Israel 1996 X X Israel 2003 X X Italy 2006 X X Japan 1996 X Japan 2004 Korea 2000 X X Korea 2004 X X Kyrgyzstan 2005 Lithuania 1997 Mexico 1997 X X Mexico 2000 X X Mexico 2003 X X Continued on next page 108 Table 5.2 (cont’d) Country Election Year Expert Unfolded Netherlands 1998 X X Netherlands 2002 New Zealand 1996 X X New Zealand 2002 X X Norway 1997 X Norway 2001 X X Peru 2000 Peru 2001 Peru 2006 X X Philippines 2004 Poland 1997 X X Poland 2001 X X Portugal 2002 X X Portugal 2005 X X Romania 1996 X X Romania 2004 X X Russia 1999 X X Russia 2000 Russia 2004 Slovenia 1996 X X Slovenia 2004 X X Spain 1996 X X Spain 2000 X Spain 2004 X X Sweden 1998 X X Sweden 2002 X X Switzerland 1999 X X Switzerland 2003 X X Taiwan 1996 X X Taiwan 2001 X X Taiwan 2004 Thailand 2001 Ukraine 1998 X X UK. 1997 X X UK. 2005 X X United States 1996 X X United States 2004 X X 5.3 Results The results of the estimation of Equation 5.3 are given in Table 5.3. As logistic regression coefficients are not. easily interpreted, I report the factor change in the odds of voting 109 proximately associated with a unit change in each variable.19 Because previous theory is quite scattered in its directional predictions regarding the included covariates, I adopt a strict standard for accepting the significance of each. First, the p—value on the variable must be less than .10, two-sided, in both models. (In fact, most variables are significant at p < .01). Second, the variable must maintain consistent directionality across the models. The effects of education, income, and freedom do not meet these criteria. as they remrse directionality across the models. In addition, age, gender, district magnitude, and disper- sion are considered insignificant. as they do not reach the p < .10 threshold in one. or both of the models. Regarding the significant results, the estimation sheds light on some interesting effects of the individual- and election-level explanatory variables. At the individual-level, identi- fying with a party increases the odds of a proximate vote by a positive factor (in Model 1, 1.36; in Model 2, 1.19). Also, as expected, seeing elections as worthwhile has a positive effect on proximate voting across both models. At the election level, the effective number of parties is negatively associated with proximate voting. That is, an increase in the number of parties makes it more difficult. for voters to “correctly” identify the party closest to them. Compulsory voting rules also negatively impact proximity voting. W' hen individuals are forced to vote. they are less likely to cast an informed ballot and thus less likely to follow the proximity logic. Finally. across both analyses, political dimensionality has a strong and significant neg- at ive effect. on proximity voting. The interaction term with education gives little evidence that this effect varies with an individual‘s level of sophistication. In fact, Model 1 reports a weak negative interaction between sophistication and dimensionality. This counterintu- itive result suggests that complex political space hampers the proximity considerations of college-educated individuals more so than their counterparts. Based on the analysis done with expert placements, a 25% increase in the dimension— Pr(:r/=1)-'If+1/Pr(y=1)l’r Pr(y=0)|m+1 Pr(y= 0)|.’r with the estimated (is: for {3,9 the odds ratio is equal to (43k. 19Odds ratios are given as and have. a direct association 110 ality variable corresponds to a .778 factor change in the odds of being a proximity voter.20 Thus, all else equal, the odds of a person living in New Zealand in 2002, in which political variation is not unidimensional, voting proximately are only 78% of those of an individual residing in Australia in 2004, in which politics conformed well to a single dimension. As the complexity of political space increases, it becomes more difficult for individuals to discern between parties, and ultimately, to choose proxin'iately. ‘) r- 1 . . - . I o - . . / “0 Io get this value, multiply the original coefficient on dimensionality. llltflflfl) : 0 ° — r _. -.()l(l()5, by 25 and use the resulting value as an exponent of e: e ‘2”1 :2 .778. 111 dosmicmm awn: 3.52:2: e. 50: was?» $33. .mmmo:_:.Ho.§a E mofigfi fiomwwmioam. 112 Snags- e353- Bosses: a: £3 23 a was weed is» 00 ©m. wfloflomfim m0 Hon—ESZ 39.. E2. 2 as; $3 an: a? .Eeeoafisa x cassava EH93 :OmuUdHQHGH 253 news 803 mm: 588$ :53 8; A33 ammo homage 82: £3 82: case $205555 3.3 as: 82: 83 G: 292 Sci 32 803 was 9: amzm :53 $3 603 S3 sausage mG—QMMHQNV ~0>®1HIGOMHU®~W :53 $3 $03 3:. 585m 82: no: 303 $2 3:52 sea 82: :3 953 m2: oases as: See :2: SE cassava 32a 83 A33 e85 580 A33 83 8:: 83 as. 8333/ ~934§E>6£ Ams~m>iav mva E mmcwno Amen—$73 mUfiO E @9530 maaoamoflm «530.ch “N 3632 maaoamoflm “Seam— ; Eve: 98:85 $0an wcflo> opeEWanm ”mm. wish In each model, {4; gives the estimated standard error of the election-specific intercepts and ,0 indicates how much variance in proximity voting is attributable to election-specific factors.21 In Model 1, p is .113, indicating that about 11% of the variance in proximity voting propensity is due to such factors. In Model 2 p is .086. Likelihood ratio tests indicate that the values of p for each model are significantly greater than zero ()2 = .00), meaning that a simple pooled logistic regression would not. suffice; the. random-intt-ircepts approach is useful. 5.4 Conclusion As shown throughout volumes of previous literature, the Downsian proximity logic of vot- ing is powerful. In fact, about one half of CSES survey respondents follow the model. The purpose of this research is to examine what causes the other half of voters to abandon the party closest to them. I find that partisan identifiers are more likely to vote proximately. as are those who see the political process as valid. At the election level, when political dimensionality or the number of parties is high, voters are more likely to abandon the proximity model of voting. These findings indicate that the environment to which voters are subject affects how they formulate their decisions. That is, individually—held characteristics are not solely important to the nature of voting behavior. Party systems and political dimensionality also play an important role in the rational (or irrational) cognitive processes that enter into one‘s vote choice. Thus, the character of voting in the world’s democracies is affected by societal conditions; certain environments encourage proximity voting, while others induce voters to abandon the proximity logic. This has broad implications for the study of voting behavior and democratic represen- tation. Clearly, voters are affected not only by their personal considerations come election 71) . . . . . . 21p = , where 7r2 / 3 IS the assumed variance of the residuals in a logistic 'z/)+7r / 3 regression. As the variance in the intercepts accounts for election-specific factors not captured by the level-2 covariates, ,0 indicates how much variance in proximity voting is due to unobserved election dependence. 113 day, but are also influenced by the political context of the election at hand. Thus, fu— ture cross-national studies of voting behavior, or studies over time in one country, can better-explain voting by looking beyond individuals to broader institutional and societal conditions. In addition, studies of the representational quality of democracy may take into account the societal factors that affect the nature of voting; political parties are likely to adjust their strategies based on the constraints and incentives placed on voters. While it is still unclear which societal conditions lead to a “correct” vote, it is apparent that the tendency of voters to follow the proximity model systematically varies with both macro- and micro—level factors. 114 Chapter 6 Electoral Systems, the Dimensionality of Politics, and Party-Voter Correspondence across Nations In democracies, the relationship between the constituent and the representative is of funda- mental iiiiportai‘ice. Yet the nature of representation is not uniform throughout the world. In this chapter I expand upon the cross-national examination of representation, examining how it varies with the dimensionality of politics across nations. I expect that party-voter ciirrespondence will be high in nations with simple dimensional constructs. Alternatively, iii ('toiiiitries where political space is not defined by a single dimension, the probability of parties and voters converging on the same ideal points decreases. In addition, previous work puts forth that. political institutions place constraints on leaders and citizens that. shape their behavior, and thereby the character of representation. To test. these expecta- tions. I exairiiiie how well parties mirror both the median and spread of voter preferences, (T()ll(llll()llitl on the electoral institutions and the. dimensionality of politics across nations. Using data from a wide sample of nations and a new measure of dimensionality, 1 find 115 that the positions of parties correspond more closely to those of voters in countries with low-dimensional political space, whereas electoral systems play a smaller role in the nature of representation. Popular sovereignty, the idea that the highest political authority is the will of the people, is a core value of contemporary republican democracy. As such, the representative must strive to carry out the desires and interests of his constituents in government. In doing so, the representative may act as either an “accountable. guardian" or an “instructed delegate.” Instructed delegates merely communicate the wishes of their constituents in goverimmnt, while accountable guardians use their intellect and reasoning skills to make the decisions best for their constituents, without necessarily considering their wishes. Whereas Hobbes preferred the instructed delegate vision of representation, so as to con- strain the “vanity” of elected delegates (Mansfield 1971). Burke saw the representative as an elected official. not expected to sacrifice "his unbiased opinion, his mature judgement.“ and "his enlightened conscience” to constituents. This sacrifice, Burke felt, would betray. rather than serve, constituents (Hoffman and Levack 1949, 114). Madison also preferred the accountable guardian vision of representation, reasoning that representatives deliber- ate. to reach the common good. which is more, productive than simply reflecting the will of the people (Pitkin 1967. 193). In what Huber and Powell (1994) term the “Majority Control” vision of democracy. elections exist to create strong, unconstrained, single-party majority governments. In this vision. governments are likely to produce policy in line with the preferences of the median voter, as they will not be constrained by other actors in parliament. In the "l’roportionate Influence" vision, conversely. elections exist to produce legislatures that rcfiect the preferences of all citizens. Though these visions or democracy differ as to who is represented, they both follow the “instructed delegate" model of representation in that they assess governments according to how well they match the preferences of citizens. \Vith the host of institutional configurations in place across countries, the nature of representation is unlikely to be uniform across the globe. As such, in this chapter I examine how well parties represent both the median and spread of voter preferences. conditioi'ial 116 on the electoral institutions and dimensional configurations of nations. I again quantify dimensionality with the measure developed in Chapter 3. In sum, I examine what. societal conditions lead to representation of the “instructed delegate” brand, and. conversely, what causes political parties to stray from the positions of voters. While previous research has found the nature of representation to be conditioned by electoral rules (Ezrow 2007'. McDonald. Mendes, and Budge 2004), I find evidence that the dimensionality of politics explains more variation in party-voter correspondence across nations. More specifically, quantitative results indicate that parties generally reflect the positions of voters in when political variation in a nation is unidimensional. However, when political space is complex. there is no discernable link between party and voter positions. Thus, the delegate model of representation does exist across countries, but only when political dimensionality is low enough that parties are able to accurately discern the positions of voters before responding to their desires. 6.1 Party System Variance To increase their share of votes, parties position themselves in issue space with regard to the location of voters (Adams, Clark, Ezrow, and Glasgow 2004; Kollman, Miller. and Page 1992; Laver 2005), and it is established that the strategies parties adopt to do so differ across institutional configurations. For example, Downs (1957) posits that. parties in two-party systems converge to the center in order to maximize vote shares, while in multiparty systems parties will spread out. offering the electorate a range. of ideological choice. Sartori (1976) confirms this expectation. emphasizing that multiparty systems place centrifugal incentives on parties. Likewise, Cox (1990), through an inductive formal analysis. finds that proportionality of the electoral systems relates positively to the amount. of outward pressure on parties. Dow (2001) tests this theory empirically and finds that parties do indeed tend to gravitate to the median voter in systems which erect barriers for smaller parties. Ezrow (2008), alternatively, finds that the predictions of the above theory are not con- 117 sistently empirically realized. That is, he finds no clear systematic relationship between electoral permissiveness and the divergence of party positions. He attributes this result to a body of theory by Schofield and coauthors, which posits that parties stake out positions by appealing to specific constituents (Schofield and Sened 2005), or to put themselves in a fa- vorable position for post-electoral bargaining (Schofield, Martin, Quinn, and Nixon 1998). Adams and Merrill (2009), using a formal approach, find that proportional representa— tion leads parties to take extreme positions only when the electorate positively evaluates characteristics such as honesty and competence - when they have enhanced “valence im- ages.” Alternatively, when valence images deteriorate, parties are induced to moderate their policies. 6.2 Party-Voter Correspondence Examining party system variance is useful because as the dispersion of parties changes, so may the character of representation, or the correspondence between the positions of parties and voters. This is illustrated in Figure 6.1, which displays the positions of three parties, A, B, and C, in relation to two voter distributions in a hypothetical election.l Clearly, if the voter ideal points are distributed consistent with Distribution A, correspondence between party and voter positions will be greater than if the voters are characterized by Distribution B. Early on, Miller and Stokes (1963) found that, though voter-representative correspon- dence was imperfect and differed across policy domains, the roll call behavior of represen- tatives in the United States was generally influenced by their perceptions of the preferences of their constituency.2 Following in this tradition, Erikson, MacKuen, and Stimson (2002) and Stimson, MacKuen, and Erikson (1995) study “dynamic representation,” the phenom— enon of public policy changing in response to shifts in public opinion. The authors find 1This figure is based on that found in Ezrow (2007, 184.). 2It should be noted that Achen (1977) points out the perils of comparing cor— relation coefficients across districts. the. approach taken my Miller and Stokes, thus calling into question their conclusions. 118 /. -\ l— Voter Distribution A l I \ Voter Distribution B Figure 6.1: Hypothetical Party and Voter Positions that dynamic representation does exist in the United States, varying in character across institutions of government (the House, Senate, and Courts). From a comparative perspec- tive. Adams, Haupt, and Stoll (2009) find that centrist and right-wing parties adjust their positions in response to both public opinion and the global economy, while parties of the left. tend not to adjust their behavior based on the prevailing public mood. 6.2.1 Electoral Rules and Party-Voter Correspondence Other research examines the nature representation in relation to electoral systems. Cox (1997, 230), for example, shows that, in equilibrium, parties have incentives to disperse across the percentiles of the voter distribution as district magnitude, and thus proportion- ality, increases. This, in turn, should increase representative—voter correspondence. This finding is similar to that of Austen-Smith and Banks (1988), who find that. with three parties, one will locate at the median with the two others dispersed equally in opposite directions. Thus, in multiparty systems, it is expected that the median voter will be represented by a party (Powell and Vanberg 2000, 396). i\"IcDonald et al. (2004) examine how well the median party in parliament corresponds to the median voter in the electorate. They find that correspondence is accurate across sev- eral nations, though proportional systems are better at “conferring the median mandate.” 119 This com-.lusion also echoes those of Powell and Vanberg (2000), who find that “median correspondence” is higher under proportional representation (PR), especially when the electoral threshold is low. Blais and Bodet (2006) also examine the ideological correspondence between parties and citizens in PR systems. They find that, while PR leads to more parties than majoritar- ian democracy, its centrifugal pressures increase the distances of parties in the legislature from the median voter. However, the frequency of coalition governments in PR systems pulls governments to the center and thus decreases the distance between parties and voters. making the net impact of PR. marginal. This latter finding corresponds with the predic- tions of Huber and Powell (1994), Lijphart (1999, 288), and Powell (2000), who suggest that the bargaining associated with the government formation process leads to govern— ments that ideologically match the median voter. Powell (2009) examines the association between election rules and congruence over time, finding that while PR genm'ally leads to closer median congruence, in recent years this advantage has waned. Instead of examining the locations of the median party and voter, Ezrow (2007) investi- gates the correspondence of the variance in party and voter positions across nations, finding it to be weaker in proportional systems. This is attributed to the fact that restrictive, non- prt)pm'tirmal systems punish smaller parties, thereby inducing them to adopt aggressive vote—seeking strategies and move towards the thickest distributions of voters. Dow (2001) attributes this aggressive behavior to “winner—take—all” feature of such electoral arrange- ments. Conversely, proportional systems motivate parties to merely find ideological niches large enough to obtain the amount of votes needed to overcome a low electoral threshold (Ezrow 2007, 184).“ I 5Taking this work in a direction outside of the scope of this project, Tavits (2007) finds that. voters reward parties for shifts on pragmatic issues, such as economic policy, but punish parties for shifts on principled issues, such as core values, as these latter shifts signal inconsistency. Alternatively, Adams et al. (2004) find that. public opinion shifts lead parties to adjust their ideologies, but only if these shifts are in a dliregtgon) harmful to the party (for example, if a rightist party, did opinion move to t '10 .0 ti; . 120 6.2.2 Dimensionality and Party-Voter Correspondence While previous comparative studies of representation focus on party systems and institu— tional constructs, in this research I also examine how the dimensionality of political space affects the nature of representation across countries. The relationship between dimension- ality and representation is simple: if dimensionality increases, so does the overall space for parties and voters to locate. This, in turn, lowers the probability that Vt.)lT(—§I‘S and parties will adopt common ideological positions. Thus, even if representatives do wish to act as “instructed delegates” and perfectly represent voters, determining voter locations when political space is complex will be rela- tively difficult. If voters are spread out along a left-right, socioeconomic dimension. it is relatively easy for parties to locate the voters and make adjustments to best match their positions. However, if a new dimension is introduced, parties must work to determine where voters are in a more complex political space. This task becomes more and more forn'iidable as the number of dimensions continues to rise; it is harder to find a needle in a. haystack than a pincushion. It is well established that parties locate themselves in a fashion designed to increase their vote shares (Adams et al. 2004; Kollman et al. 1992; Laver 2005) and voters tend to choose parties that are close to them ideologically (Downs 1957). Thus, a repeated game emerges in which voters choose parties or answer opinion surveys and parties adjust accordingly. Once they perceive of these adjustments, voters again choose parties and respond to opinion polls, and parties again adjust accordingly. Thus, the lack of correspondence in more complex dimensional settings also arises from “incorrect” cues given by voters. On a single dimension, it is relatively easy for voters to choose the party closest to their ideal point, and thus parties’ adjustments to voter opinions will be reasonably accurate. However, when political variation does not conform to a single dimension, voters may have a difficult time locating the party most proximate to them and may vote or answer opinion surveys “incorrectly” (see Chapter 5). Thus, the cues parties take when making adjustments will be wrong, and the disparity between party and voter positions will be exacerbated. Because both voters and parties will have 121 difficulty locating each other as the dimensionality of political space increases, it is more difficult for parties to accurately (and adequately) represent voter preferences in nations where political variation is not unidimensional. 6.3 Expectations, Variables, and Measurement The above theory makes clear directional predictions about the positioning of parties and the nature of representation across democracies. Following authors such as Ezrow (2007) and McDonald et al. (2004), I put forth the following hypotheses: Hypothesis 1 : In proportional electoral systems, the correspondence between the variance of voter preferences and the variance of parties will be lower than in majoritarian systems. Hypothesis 2: In preportional electoral systems, the correspondence between the median voter and the median legislative party will be higher than in majoritarian systems. The. electoral institution of a nation is the conditioning variable of interest in Hypothe- ses 1 and 2, while dimensionality is left unconsidered. Thus, I also put forth the following hypotheses, derived from the theory in Section 6.2.2: Hypothesis 3: There will be a positive link between the variance of voter preferences and the variance of parties only when political variation is well-captured by a single dimension. Hypothesis 4 : There will be a positive link between the median voter and the median party only when political variation is well-captured by a single dimension. I determine party and voter positions using the empirical method described in Section 2.3. To assure the robustness of the analyses, I also obtain expert party locations from the CSES. The CSES asks experts to locate parties along a left-right continuum and asks respondents to locate themselves on this same continuum,4 which makes gauging 4The question states: “In politics people sometimes talk of left and right. Where would you place yourself on a scale from O to 10 where 0 means the left and 10 means the right?” 122 the relative positions of voters and parties straightforward. It is also possible to obtain measures of party positions based on individuals’ perceptions of the political parties, as the CSES also asks respondents to locate parties along the left-right continuum. Such measures, however, can be problematic. Respondents may place their most-preferred party closer to their own position, regardless of that party’s true position (Adams, Merrill, and Grofman 2005, 170). Additionally, they may shape their responses to meet a proximity voting criterion (Boatright 2008). Thus, I choose not. place parties using individual-level party placements. These measurement strategies assume that positions along the left-right continuum provide a meaningful representation of preference. While this assumption is not always justifiable, a substantial amount of research has established the left—right scale as a rea— sonable distillation of voter and party ideologies at the national level, even in instances of milltidimensionality (Powell and Vanberg 2000, 385: see also Huber 1989; Inglehart 1984). I also choose to use a unidimensional approach to facilitate the testing of the previous work outlined here, which also models parties and voters along a single dimension. To gauge the dispersion of parties, I use a formula from Ezrow (2007), which itself draws on the work of Alvarez and N agler (2004). The measure takes the divergence of each party from the weighted mean position into account, as well as the vote share of each party.‘5 \ ‘ . 7” .. .___2 a -Q' ‘9‘? J " For each election, the formula is , (23:1 15(1)] p) , where 2.] is party 3 s yotc proportion, 6 To measure p, is the position of party j, and )5 is the weighted mean party position. voter polarization I take the standard deviation of voters’ self-reported positions on the. aforementioned 0-10 left-right scale (expert placements) or their ideal points (unfolded placements). I calculate median voter and median party position across each CSES election using the same party and voter placements. ,. ”Vote shares were obtained from the CSES Macro Data, available at. ht tp: / / w ww.cses.org / download / contributions / contributionsmirrorhtm Each party’s contribution to the mean is weighted by its share of the vote. The formula for the weighted mean is thus 3110)]- x 19]). where vj is party j’s vote I pr(‘)portion. 123 Following Ezrow (2007) I gauge the disprOportionality of electoral systems using Gal- lagher’s (1991) least squares index. The formula is W where 1:]- is the per- centage of votes for party j and 33- is the percentage of seats won.7 The index theoretically ranges from 0 to 100, with 0 indicating perfect proportionality. To gauge dimensionality, I use the index developed in Chapter 3. A value of 1.0 indi- cates that voter preferences in a given nation are entirely generated by a single dimension. I subtract. this value from 1 so that higher values correspond with poor adherence to a single dimension. I also multiply the value by 100. The resulting variable thus measures the percent of variance in party and voter locations not captured by a single dimension, with a value of 0 indicating that voter preferences in a given nation are entirely generated by a single dimension. As noted, I term this new measure “political dimensionality.” Ezrow (2008) posits that the spread of parties increases with the overall number of par- ties. Therefore, I use the effective number of parliamentary parties (EN PP) as a control variable. The variable is created using the common Laakso and Taagepera (1979) index.8 Because there is broad disparity in the wealth of the countries in the dataset, I also control for each country’s overall level of development using GDP per capita, adjusted for pur— chasing power and measured in thousands of constant international dollars. I obtain this measure from the World Bank’s Development Indicators.9 Each variable is summarized in Table 6.2. 6.4 Model Specification and Methodology I collected on data on 81 elections covered in the Comparative Study of Electoral Systems and the New Zealand Election Study (see Chapter 3). Because the above tl'lCOI‘y is based 7Vote and seat shares were obtained from the CSES Macro Data. 8The index is given as Tlfi)’ where s,- is the proportion of seats obtained by . 3‘7” 721 2. the ith party. (“)Taiwan's per capita GDP is taken from the VVTO and Wu (2004). 1‘24 on legislative party behavior, I do not examine presidential elections. I thus dropped the purely executive elections of Belarus in 2001, Chile in 1999, and Russia in 2000 and 2004. In addition, data on the percentage of votes received by parties was unavailable for certain elections. As these values are needed for the dispersion and disproportionality measures, the elections were dropped. Moreover, expert opinions on the positions of parties and voters’ self-reported positions were not available across all countries and elections. The elections covered across each model thus vary due to the variables included and the measures used. Table 6.1 lists the countries and election years included in each model, and the variables used in the analyses are summarized across all available elections in Table 6.2. Table 6.1: Included CSES Elections by Model Number Country Election Year 1 2 3 4 5 6 7 8 9 10 Albania 2005 X X X X X X X X X X Australia 1996 X X X X X X X X X X Australia 2004 X X X X X X X X X X Belarus 2001 Belgium-Flanders 1999 X X X X X X X X X X Belgium—Flanders 2003 X X X X X X X X Belgium-Walloon 1999 X X X X X Belgium—Walloon 2003 X X X X X X X X Brazil 2002 X X X X X X X X X X Bulgaria 2001 X X X X X X X X X X Canada 1997 X X X X X X X X X X Canada 2004 X X X X X X X X X X Chile 1999 Chile 2005 X X X X X X X X X X Czech Republic 1996 X X X X X X X X X X Czech Republic 2002 X X X X X X X X X X Denmark 1998 X X X X X X X X X X Denmark 2001 X X X X X X X X X X Finland 2003 X X X X X X X X X X France 2002 X X Germany 1998 X X X X X X X X X X Germany 2002 X X X X X X X X X X Hong Kong 1998 X X X X Hong Kong 2000 X X X X Hong Kong 2004 X X X X Hungary 1998 X X X X X X X X X X Hungary 2002 X X X X X X X X X X Continued on next page 125 Table 6.1 (cont‘d) Country Election Year 1 2 3 4 5 6 7 8 9 10 Iceland 1999 X X X X X Iceland 2003 X X X X X X X X X X Ireland 2002 X X X X X X X X X X Israel 1996 X X X X X X X X X X Israel 2003 X X X X X X X X X X Italy 2006 X X X X X X X X X X Japan 1996 X X X X X X X X Japan 2004 X Korea 2000 X X X X X X X X X Korea 2004 X X X X X X X X Kyrgyzstan 2005 Lithuania 1997 X Mexico 1997 X X X X X X X X X X Mexico 2000 X X X X X X X X X X Mexico 2003 X X X X X X X X X X Netherlands 1998 X X X X X X X X X X Netherlands 2002 X X X X X X X X X X New Zealand 1996 X X X X X X X X X X New Zeal and 1999 X X X X X X X X X X New Zealand 2002 X X X X X X X X X X New Zealand 2005 X X X X X X X X X X Norway 1997 X X X X X Norway 2001 X X X X X X X X X Peru 2000 X X X X X X X X X X Peru 2001 X X X X X Peru 2006 X X X X X X X X X X Philippines 2004 Poland 1997 X X X X X X X X X X Poland 2001 X X X X X X X X X X Portugal 2002 X X X X X X X X X X Portugal 2005 X X X X X X X X X X Romania 1996 X X X X X X X X X X Romania 2004 X X X X X X X X X X Russia 1999 X X X X X X X X X X Russia. 2000 R ussia 2004 Slovenia 1996 X X X X X X X X X X Slovenia 2004 X X X X X X X X X X Spain 1996 X X X X X X X X X X Spain 2000 X X X X X Spain 2004 X X X X X X X X X X Sweden 1998 X X X X X X X X X X Sweden 2002 X X X X X X X X X X Switzerland 1999 X X X X X X X X X X Continued on next page 126 Table 6.1 (cont’ ) Country Election Year 1 2 3 4 5 6 7 8 9 10 Switzerland 2003 X X X X X X X X X X Taiwan 1996 X X X X X X X X X X Taiwan 2001 X X X X X X X X X X Taiwan 2004 X X Thailand 2001 X X X X X Ukraine 1998 X X X X X X X X X X United Kingdom 1997 X X X X X X X X X X United Kingdom 2005 X X X X X X X X X X United States 1996 X X X X X X X X X X United States 2004 X X X X X X X X X X Table 6.2: Summary Statistics Variable Mean Std. Dev. Min. Max. n Party Polarization (Unfolded) 3.714 1.261 0.643 6.609 71 Voter Dispersion (Unfolded) 2.760 1.166 1.018 6.558 81 Median Party (Unfolded) —0.008 4.292 -7.247 10.091 81 ;\'Iedian Voter (Unfolded) -0.142 0.904 —3.714 2.657 81 Party Polarization (Expert) 1.864 0.575 0.710 3.819 65 Voter Dispersion (Self Placements) 2.412 0.449 1.318 3.873 76 Median Party (Expert) 4.979 1.158 1.000 7.500 72 Median Voter (Self Placements) 5.079 0.744 1.000 8.000 76 Disproportionality 6.828 4.464 0.080 21.97 75 Political Dimensionality 25.354 9.848 4.465 45.924 81 MDM 21.589 37.111 0.800 150 81 Majoritarizn1 0.148 0.357 0 I 81 Mixed 0.358 0.482 0 1 81 Proportional 0.494 0.503 0 1 81 ENPP 3.865 1.593 1.173 9.051 78 Per Capita GDP 19.295 9.456 1.722 36.451 81 The main explanatory variables in the analyses are the variance and mean of voter loca- tions. To test the link between voter and party positions across electoral systems, I interact, these variables with disproportionality. I also create interactions with the new dimension— ality variable to examine the nature of representation across (.liniensional configurat ions. According to the theory developed in this project. the interaction terms should reveal a positive link from voter positions to party positions in countries with low dimensional political space. Conversely, in countries where political variation is not unidimensional. there should be little or no relationship between voter and party locations. 127 I estimate the relationships between party positioning and the explanatory variables with ordinary least-squares (OLS) regression. As several countries appear more than once in the data, the standard OLS assumption that observations are independent is unmet. I therefore cluster the standard errors by country to account for intranational correlation. 6.5 Results Before testing Hypotheses 1-4, I revisit the contending findings of Dow (2001) and Ezrow (2008) regarding the relationship between electoral permissiveness and the spread of party systems. I gauge electoral rules in three ways: Gallagher’s disproportionality index, mean district magnitude, and a three-category dummy variable for majoritarian, mixed, and proi)ortional systems. As is standard practice in the comparative literature, I take the logarithm of district magnitude. Results are shown in Tables 6.3 and 6.4. As found by Ezrow (2008), the models estimated with unfolded party locations show an absence of a relationship between electoral rules and party system dispersion. The models estimated with expert placements, alternatively, show signs of a relationship more in line with Downsian predictions and the findings of Dow (2001). Though the coefficient is only marginally significant. Model 1 ing’l‘able 6.3 shows that an increase in disproportionality leads to a decrease in party system dispersion. Likewise, Model 3, estimated with a categorical variable (proportional systems are left out as the reference category) shows that parties are significantly more clustered in majoritarian systems. Party system variance is also lower in mixed systems, though this result again slightly misses conventional levels of statistical significz-ince. 128 .mofiazsoo so cosmic 22.8 Eagfieam .mozogémbfi E 8357; ”8286.038 mmwd mwod mamd m A 30.5 wood sees :38 mm C. C. we .2 82: £3 803 23 89: mass 2888 $3 $3 was: cam: mm; $235.82 ea: ES- E: 352 at; gas sesaesoaoama $2573 EmmoEmoU 33.875 anemoEmoO fies—gig acmmoEmoU moi—~25» moswmawx/ Bomuommawo "c mtsumswwz 32¢me "m homecomfiompammm 2v $58955 onomcD neonate> 839mm beam was 33m Egopoflm “flaw mien. .mQEuESOU GO Umhmamflflo magnum Uhdwfidum .mmmwfififlwpdfl A: m051~>i Uwfiuflmioarfi wood 3.3V mid m A no.5 mmfid mad mmod mm we we ma 2 803 Ros 82: am: :53 mass 88.0.8... as: same. was: A Seq wand- caCSCOHwE 33 ES 3: 202 an: m3? Escocsaoaao Am:_m>iv 3205000 $3.873 ecomoEmoO 3:373 EmmoEmoO mEmEm> moEwEmS Buiowmawo um eczemammz «$.5me ”m ham—wcofipomohammfl ”H meEwomE 959$ “roughest, EEmzm beam 9%. $3M €8.3on “Muo 83$. 129 Table 6.5: Determinants of Party/ Voter Variance Correspondence 7: Expert Placements 8: Unfolded Placements Variable Coefficient (p-value) Coefficient (p—value) Voter Dispersion 1.356 (.084) —1.214 (.002) Disproportionality 0.075 (.414) -0.048 (.595) Dispersion >< Disprop. -0.036 (.343) 0.023 (.428) Political Dimensionality 0.101 (.047) -0.069 (.026) Dispersion X Dimen. -0.043 (.042) 0.016 (.171) Enter) (1057 (171) .41059 (497) Per Capita GDP -0.013 (.320) -0.014 (.288) constant —1.146 (.559) 8.222 (.000) n 59 68 It? (1126 (i417 Prob > P 0.112 0.000 Two-sided p—values in parentheses. Standard errors clustered on countries. Table 6.6: Determinants of Party / Voter Median Correspondence 9: Expert Placements 10: Unfolded Placements Variable Coefficient (p-value) Coefficient (p-value) Median Voter 1.745 (.009) 3.517 (.077) Disproportionality 0.342 (.466) -0.025 (.847) Median Voter x Disprop. -0.079 (.386) -0.087 (.260) Political Dimensionality 0.252 (.001) -0.102 (.085) Median Voter x Dimen. -0.050 (.002) -0.178 (.032) Per Capita GDP -0.005 (.771) -0.026 (.640) constant -3.297 (.288) 3.116 (.252) 'n 61 72 112 (1125 (1171 Prob > F 0.000 0.002 rTwo-sided p—values in parentheses. Standard errors clustered on countries. To test variance correspondence, I regress the spread of parties on the spread of voters, ilwt "l l" * 't' 'l't 10 d' " 'l' r *l ' ° ' ‘ * l in f " r (( oia ( ispropoi 4011.1 1 .y, imensiona. ity, t 1e interaction teims, tie num )(r 0 par- ties, and per capita GDP. Additionally, to test correspondence between the median party '10 Electoral rules and dimensionality are systematically linked, as shown in Chapter 4. In addition, electoral rules are clearly linked to the number of parties that gain representation. Collinearity is thus an issue. However, because each variable is theoretically linked to the dependent variable, I opt to include them; rather than risking a misspecified model I choose to suffer from the possibly-inflated standard errors that arise from collinearity. 130 and the median voter, I regress the median party position on the median voter position, electoral disproportionality, dimensionality, the interaction terms, and per capita GDP. Because the number of parties is not theoretically linked to the median party position, 1 do not include it in this equation. The models indicate that parties do adjust their positions in response to those of voters. However, this phenomenon occurs independent of electoral rules; the interaction term l')etween disproportionality and voter positions is insignificant across each model. Thus, no support is found for Hypotheses 1 and 2. Yet, there is indeed an interactive relationship of dimensionality and voter positions on party positions. In Models 7, 9, and 10 the coefficient on political dimensionality and its interaction with voter positions is statistically significant (p <.10, two-sided). Coefficients on continuous variable interaction terms and their constituent parts are. not. readily interpretable (Brambor, Clark, and Golder 2006; Braumoeller 2004).” Thus, I graphically display the conditional effect of voter positions on party positions across the range of political dimensionality in Figures 6.2 and 6.3.12 Figure 6.2, which is produced with the coefficients from the models estimated with the expert party positions, confirms the theoretical expectations. The left panel of the figure shows that when politics are well— captured by a. single continuum, an increase in the spread of voters corresponds with an increase in the spread of parties. However, when political variation is not unidimensional, this link deteriorates. The same is true for the link from the median voter to the median party; the right panel of the figure shows that when politics are well—captured by a single continuum, a rightward shift of the median voter corresponds to a rightward shift of the median party. And, when political variation is not unidimensional, this link is again no longer significant. Thus, Figure 6.2 provides strong support for Hypotheses 3 and 4. Figure 6.3, which is produced using the models estimated with the unfolded party and voter positions, however, does not provide support for these hypotheses. First, the 11In fact. the coefficients on the constituent variables are equal to their marginal effect when the other constituent variable equals zero. 12 These figures were produced with the help of code from a web supplement to Brambor et al. (2006), available at http: / / homepages. nyuedu / ~mrg2 1 7/ interactionht ml. 131 (1) —4 N _. '- 1 O 'T _ N a I <0 _ m d . l l l l I I I T I l T 1 I 0 10 20 30 4O 50 0 10 20 30 40 50 Political Dimensionality Political Dimensionality ME. of Voter Var. on Party Var. ——- ME. of Voter Med. on Party Med. ~ - - - 90% Confidence Interval — - — - 90% Confidence Interval Figure 6.2: The Effect of Voter Positions on Party Positions (Expert Placements) left panel indicates an unexpectedly negative relationship between voter dispersion and party system dispersion across all levels of political dimensionality. The right panel shows that there is indeed a positive link between the position of the median voter and the median party when dimensionality is low. However, when political dimensionality is not unidimensional. there is a strong and significant negative link between the position of the median voter and the median party. No matter how complex political variation in a nation is, it is unexpected that parties will move in a direction opposite of the voters. Thus, the quantitative results provide only partial evidence that political dimension— ality plays a role in the nature of representation in countries. In the models estimated with expert party positions and self-provided voter locations, it is shown that dimension- ality affects the degree to which. parties mirror the preferences of voters. However. in the models estimated with the unfolded party and voter positions, theoretical expectations are 132 6 1h 2‘0 Q) 4T0 5'0 0 10 20 30 4o 50 Political Dimensionality Political Dimensionality M.E. of Voter Var. on Party Var. -— M.E. of Voter Med. on Party Med. ~ ~— — - 90% Confidence Interval — — — - 90% Confidence Interval I’igure 6.3: The Effect of Voter Positions on Party Positions (Unfolded Plact-érnents) 1H it met. Regarding electoral rules, there is little evidence that representation varies with (‘lcctoral disproportionality once the dimensionality of politics is accounted for. 6 .6 Conclusion V’V hile the representative in the guardian model of democracy uses his superior intellect and reasoning skills when governing, the delegate model of democracy puts forth that rep- resentatives should reflect the desires of voters. Thus, in the delegate model of democracy it is the duty of the representative to transmit the true interests of the represented in government. Following this model, there should be strong correspondence between the I’Oxitions of parties and voters across the range of issues and political matters in a country. Examining the positions of parties and voters on every politically significant topic, 133 however, is altogether impossible. As such, researchers place voters and parties in a polit- ical space which summarizes their overall preferences. In this work parties and voters are organized unidimensionally along a left-right continuum. The strength of this continuum in explaining political variation in a nation is taken into account with the measure of “political dimensionality” produced in Chapter 3. The findings provide evidence that the positions of parties correspond more. closely to those of voters in countries with low-dimensional political space, whereas the nature of representation does not appear to be directly influenced by electoral rules. However, as Chapter 4 shows a relationship between electoral systems and the dimensionality of politics, there is likely an indirect link to party-voter correspondence; dimensionality acts as a catalyst in the relationship between electoral rules and the nature of political repre- sent ation. 134 Chapter 7 The Dimensionality of Politics and Voter Behavior in Preferential Systems: The Case of Australia When studying voter preferences in systems that employ preferential voting,l researchers must consider all parties on the ballot. Voters in such systems are required to assign a ranking to competing parties or candidates, rather than simply choosing their most preferred alternative. From these rankings, various preference allocation schemes are used to nominate one or more candidates to parliament. Empirical spatial analysis allows for inferences as to how voters will order parties or candidates on their ballots, as the approach recovers voter and party positions in the same space. In this research I study the nature of voter preferences in Australia. As a stable and transparent democracy, Australia is an attractive nation in which to study voting behavior. hiUI'COVGI‘, because it employs the alternative vote (AV). a preferential electoral formula, for its lower house elections, it is interesting to examine voter preferences over each party 1The most common form of preferential voting is the alternative vote (AV), also known as “the alternative transferable vote.” “instant-runoff voting,” or sim- ply “preferential voting.” The alternative vote is used for elections in Australia, Ireland, Papua New Guinea, and Fiji. Preferential voting is also used in various Eon-AV forms in countries such as Sri Lanka, Malta, India, Northern Ireland, and Scotland. in Australia. In contrast, much of the previous literature on Australian voting accounts solely for the first preferences of voters. To proceed, I review previous literature on the dimensionality of politics and voting behavior in Australia. Then, I use the unfolding method described in Chapter 2 to assess dimensionality in Australia in 2004 and recover party and voter ideal points. The model asserts that voters rate Australian parties according to a unidimensional proximity logic. Results indicate a good data-to-model fit, and evidence of a strong socioeconomic political dimension in Australia is found. Using voter ideal points obtained from the analysis, I then examine how several social and demographic characteristics affect voter preferences over six2 Australian political parties. Finally, I show these ideal points to relate strongly to actual vote choice over major and minor parties in Australia. These findings have im- portant implications for the future study of voting in Australia and other countries that employ preferential voting institutions. In preferential systems, studying voter preferences over all available alternatives is more informative than solely examining voters“ first pref- ereimes. Thus, this study adds to the literature by introducing a new way to study voting behavior in preferential systems, and by reexamining the nature of voter behavior and the dimensionality of politics in Australia. 7 .1 Australian Political Dimensionality and Voter Behavior in Theory The Australian House of Representatives is perennially divided between “The Coalition,” which is composed of the Liberal and National parties,3 and the Labor party. As two- party competition is unidimensional and bipolar by definition (Jackman 2003, 280). it stands to reason that Australian political space should be as well. However, the existence of several viable parties, each with unique platforms and leaders, may plausibly lead to 2The parties examined received the most votes for the lower house in the 2001 Federal Election. 3“The Coalition” is quite stable. In fact, the last time the Liberal and National parties directly competed with one another for seats was the 1.987 Federal Election. 136 multidimensionality in Australian politics. Table 7.1 provides a short description of each of the six parties under examination. Table 7.1: Party Descriptions Party General Description Australian Social liberal ideology. Democrats Support gay and indigenous rights and the welfare system. Maintain interventionist economic policies. Australian Social progressive ideology. Greens Promote universal health coverage, progressive taxation, Australian Labor Party (ALP) Liberal Party of Australia. National Party of Australia One Nation Party and a broad welfare state. Strongly oppose militarization and war. Democratic socialist ideology. Combines both the environmental and socially liberal issues of the left with a support of workers. Officially affiliated with labor unions. Conservative liberal ideology. Takes a relatively hands-off approach to economic affairs. Supports free trade, family values, and small government. Conservative agrarian ideology. Supportive of rural peoples. Currently urges free enterprise and conservative social values. Much policy overlap with the Liberals. Nationalist conservative ideology. Opposes immigration (mostly ethnic) without approval through referenda. Pro-gun and pro—free speech, favors a strict justice systen‘r. Favors a drastic reduction in taxes on individuals. A sizeable body of research examines what factors lead Australian voters to choose among the political parties. Some of this work puts forth that postmaterialist Australian voters align with minor parties such as the Greens or the Democrats (Papadakis 1990: Weakliem and Western 1999), and several voting studies include postmaterialism as a variable of interest (see, for example, Blount 1998; Western and Tranter 2001). As such. Charnock and Ellis (2004) conclude that the left-right economic dimension performs well 137 at predicting vote choice, but deem it inadequate for describing minor party locations. McAllister and Studlar (1995, 205) also find minor party voting to be associated with a postmaterialist, “new politics” dimension, but also provide evidence of a prevalent so- cioeconomic dimension. Using 1990 survey data, they employ factor analysis to examine various issue attitudes of Australian voters and elites. Though they find a new politics dimension among both groups, it explains only 12.3% and 11.3% of the variance in issue attitudes among voters and elites respectively. The social and economic dimensions, on the other hand, explain 41.0% of the variance among voters and 52.7% among elites. Huber and Inglehart ( 1995) employ an expert survey to examine Australian political dimensionality. The survey asks respondents to not only locate the parties on a single dimension. but also to state whether there is a second dimension of political conflict within the nation (77). The respondents did not identify a clear second dimension in Australia and 65% agreed that political conflict in Australia takes place along a single economic. or "class conflict” dimension (87). Recognizing the importance of the socioeconomic dimension, several studies use social and economic variables to predict Australian voting behavior. For example, Cameron and Crosby (2000, 354) and Wolfers and Leigh (2002) find macroeconomic factors to be fairly good predictors of election results. At the individual level, Gow (1990) notes that electoral choice is highly dependent on voters’ perceptions of the economy. McAllister and Bean (2000, 395) explain, “By far the most important and consistent influences in defection [from Liberal to Labor] were the positions on the four major economic issues that were ('lebated during the [1998 federal] election.” Regarding class, though the social—structural approach was once paramount to the understanding of Australian electoral behavior (Jackman 2003), the importance of class as an explanatory variable has diminished. Jaensch (1995, 130) notes that class has weakened as an explanatory factor because more significant cleavages have formed. In addition, he 3xplains that voters have “de-aligned” themselves from “social cleavages as a basis for electoral behavior.” However, class, along with other demographic factors, such as religion, gemler, age, 138 ethnicity, and geographic location are still present in the voting literature. For example, Charnock (1997) finds that, in the 1996 federal election, religion and union membership significantly affected voting behavior, while age and place of residence were less conse- quential. Leigh (2005) concludes that vote choice is affected by nationality, income level, and age, with age impacting the voting choices of women more than men. Further examining the gender divide, Renfrow (2003) notes that women are “more liberal than men in their political attitudes and their vote choice.” However, she con— cludes that a merger of men’s and women’s social and economic standings may lead to a convergence of political attitudes and behavior (310-312). This conclusion is also reached by Studlar, McAllister, and Hayes (1998). who find gender differences in voting behavior in Australia are inconsequential once factors such as age, education, marital status, and employment. are accounted for. Such conclusions correspond with Jaensch’s (1995) and Cbarnock’s (1997) findings of an insignificant gender effect. The “iVIichigan model” of Campbell, Converse, Miller, and Stokes (1960), which con- nects psychological attachments to political parties with voter behavior, is also prevalent in the Australian voting literature. Aitkin and Stokes (1977) first. examined party identifi- cation in Australia, emphasizing its stability over time (Jackman 2003, 275). Accordingly, .laensch (199." , 18) notes, “In Australian elections, the main explanatory factor is party identification” and “the keystone [of party support] in Australia appears to be stability.” In fact, the strength of partisanship in Australia may be due to the nation’s unique voting rules. W'hile compulsory voting obligatorily reinforces party identification at each election, the alternative vote allows one to maintain an attachment to a minor party without fear of “wasting a. vote” (Marks 1993, 141). In general, previous research on Australian voting and dimensionality links individu- als’ vote choices to demographic characteristics, psychological attachments, and economic factors. A second line of research finds minor party voting to be associated with postma- terialist ideals. To disentangle the effects of various factors on Australian voting behavior, I revisit previous research using a new approach to the study of voting behavior. 139 7 .2 A New Look at Dimensionality and Voting Behavior In this analysis of Australian voting, I examine whether voter positions along a single dimension are related to vote choice. To do so, I test whether a single dimension captures the. party preferences of Australians using the unfolding method developed in Chapter 2 and the party and voter locations produced in Chapter 3. As shown in Chapter 4, political variation in countries with majoritarian electoral systems generally conforms well to a single dimension. And, as expected, Australian political space is well captured by a single dimension. Figure 7.1 displays the recovered party positions. The R2 value is 0.809, meaning that a. single dimension explains about 81% of the variance in individual evaluations of the parties. The parties are aligned with the Democrats on the far left and One Nation on the far right. The numerical values in Figure 7.1 represent the locations of each party on the cmrtinnum. On the left the Democrats and Greens are very close to each other, with Labor roughly two units toward the center. On the right, One Nation is roughly three units to the right of the Coalition parties, which themselves are separated by about 1.5 units. Dem Gm Lab Lib Nat ON | l l . | l I | I | ‘ l l l -6_3 -o.t “1.4 O 3.6 5.0 7.8 RSQ=.809 Figure 7.]: Party Locations on the Underlying Dimension Preference orderings can be determined from voters” ideal points. Based on the Down- sian proximity voting assumption, voters will rank the parties on their ballots in accordance with these preference orders. For example, consider a voter with ideal point —5.0. This point corresponds to the preference ordering Labor—Green—Democrat—Lil)eral-National—One 140 Nation, indicating that this voter will order the parties as such on her ballot. A voter with ideal point 1.0, alternatively, will order the parties Liberal—National—Labor-One Nation— Green-Democrat. Figure 7.2 displays a smoothed histogram of the voter ideal points, showing that the majority of Australian voters in 2004 have ideal points near the center of the distribution. At the mean (0.0) of this distribution the corresponding preference ordering places the Liberals over the ALP. This result is expected, given the victory of the Coalition in the 2004 election. 0.! .. m —: A? e w . Q) :3 LO Q _ o d Dem Gm ALP Lib Nat ON —10 Is 6 t if) Ideal Point Figure 7.2: Density Plot of Individual Ideal Points 7 .3 What is the Underlying Dimension? The parties are aligned from left to right along the dimension in accordance with their social and economic policy stances. On the far left are the Democrats and Greens, both of which are relatively progressive in both their social and economic policies. On the far right is One Nation, which maintains socially conservative stances, such as swift and harsh punishment of criminals and a return to family values. as well as economically libertarian 141 stances, such as the elimination of taxes on profits. capital gains. income. and savings. The ALP, which moderates its progressive socioeconomic stances to appeal to a greater proportion of voters, is situated at the center-left of the continuum. The Coalition parties, which govern from a conservative standpoint, but temper their platforms so as to remain viable, are located on the right, between the zero point and One Nation. Thus, I contend that the underlying continuum recovered in this analysis is the common “socioeconomic” political dimension. Note that Ganghof and Brauninger (2006), using expert survey data from Laver and Hunt ( 1992) to examine Australian parties along the socioeconomic dimension, obtain party locations that reflect the findings in this study. Though they only examine four parties, they locate the Democrats on the far left, the ALP to their immediate right, and the Coalition parties nearby each other on the far right. Furthermore, Huber and Inglehart (1995), placing the same four parties using a different expert survey, also obtain party locations of the same order found in this study. The assertion that Australian politics are well-described by a single, socioeconomic continuum runs counter to previous studies which emphasize the postmaterialist dimen- sion (Blount 1998; Western and Tranter 2001). However, to develop their measures of postmaterialism, these studies rely on a survey question based on Inglehart’s (1977. 28) 4-item index.‘1 Davis, Dowley, and Silver (1999), based on analysis of this same question and others from the World Values Survey, give evidence that rejects its usefulness as a gauge of a “materialist-postmaterialist” dimension. They instead find that answers are “related to conditions in society.” Furthermore, they expect stability in these positions due. to both “the relationship between broad economic and social conditions and people’s concern with specific material issues” (960). 4The question states, “Here is a list of four aims that different people would give priority: 1. Maintain order in the nation. 2. Give people more say in important. government decisions. 3. Fight rising prices. 4. Protect freedom of speech. If you had to choose among these four aims, which would be your first choice? And which would be your second choice?” Those who choose any combination of responses 2 and 4 are labeled “Postmaterialists,” while those choosing any combina- tion of responses 1 and 3 are called “lVlaterialists.” All other responses are assigned to a mixed category. 142 Therefore, while Blount (1998) and Western and Tranter (2001) are likely accurate in their statistical conclusions that minor party voting is associated with postmaterialism, their measure of the concept may be skewed. These authors rely on a question that gauges opinions based on economic and social societal conditions rather than postmaterialism, and can therefore be captured with the socioeconomic dimension. However, note that. this glitch in the survey mechanism does not affect the findings of McAllister and Studlar (1995), who show minor party voting to be associated with a postmaterialist, new politics dimension through an examination of issue attitudes. Thus, to further test whether a postmaterialist dimension exists in Australia, I corre— lated the postmaterialist variables envirtmmentalism and religiosity, defined in 'I‘ahle 7.2, with the residuals5 from the unfolding analysis. The absolute value of every correlation coefficient is below .30, with most falling in the .10 to .20 range. To check for a dimension based on innnigration or race, I correlated the residuals with the immigration variable, defined in Table 7.2. In this case every correlation coefficient had an absolute value below .10. The information left unexplained by the single dimension does not arise from either postmaterialist or immigration-based dimensions. 7.4 Determinants of Party Preferences Past work on Australian voting examines vote choice either as a dichotomous decision between Labor and the Coalition parties (for example, Gow 1990) or arbitrarily assigns values to each party.6 Other research, making use of multinomial regression models, does not assign an order to the parties (for example, Western and Tranter 2001). To its credit, this approach avoids arbitrarily classifying parties along a. single continuum. However, it does not give insight into how the parties are dimensionally situated. Furthermore, as 5The residuals are defined for each party and measure the difference between an intjhwdual’s observed rating of a given party and their distance from that party on the recovered continuum. (Bean (1994) assigns values of 0, .5, and 1 to Labor, the Democrats, and the, Coalition respectively. Blount (1998) assigns a 1 to the Coalition, .5 to Labor, and (l to any minor party. 143 Table 7.2: Variable Definitions and Descriptive Statistics Variable Description n Mean S.D. Ideal Point Ideal point from the unfolding analysis 1627 0.00 2.21 Demographic Income Income quintile 1513 2.92 1.42 Age Age in years 1554 49.29 16.74 Female Dummy: 1 2 Female 1640 0.51 0.50 University Education Dummy: 1 = R has attended university 1606 0.33 0.47 Unemployment Dummy; 1 = Unemployed 1607 0.02 0.16 Birth Nation Dummy: 1 = Born in Australia 1637 0.77 0.42 Religiosity Attendance of religious services 1615 2.40 1.70 1 = “Never” thru 6 = “Once a Week” Rural Dummy: 1 2 Rural 1668 0.22 0.41 Union Member Dummy: 1 -— Union Member 1527 0.25 0.43 Blue Collar Dummy: 1 2 Blue Collar 1473 0.21 0.41 Fhite Collar Dummy: 1 2: White Collar 1473 0.77 0.42 farmer Dummy: 1 = Farmer 1473 0.03 0.16 Partisanship id Democrat Dummy: 1 = Party ID is Democrat 1362 0.01 0.09 id Green Dummy: 1 2 Party ID is Green 1362 0.06 0.24 id Labor Dummy: 1 2 Party ID is ALP 1362 0.38 0.49 id Liberal Dummy: 1 = Party ID is Liberal 1362 0.49 0.50 id National Dummy: 1 :- Party ID is National 1362 0.04 0.19 id One Nation Dummy: 1 = Party ID is One Nation 1362 0.01 0.09 Opinion and Info. Ideology 0 2 “Left”, 10 2: “Right” 1399 5.32 2.10 Protest Dummy: 1 :- R has protested or 1552 0.14 0.35 demonstrated in past 5 years Pol. Information Dummy: 1 = R answered three political 1067 0.45 0.50 information questions correctly Gov. Performance 1 2 “Very Bad”, 4 = “Very Good” 1611 2.83 0.70 Deni. Satisfaction 1 : “Not Satisfied”, 4 :- “Very Satisfied” 1634 2.98 0.71 Terrorism Dummy: 1 = R feels terrorism 1603 0.05 0.21 is the most important election issue Defense Dummy: 1 = R feels defense 1603 0.06 0.23 is the most important election issue Environment Dummy: 1 = R feels the environment 1603 0.06 0.23 is the most important election issue Iraq Dummy: 1 = R feels the Iraq war 1603 0.04 0.20 is the most important election issue. Iunnigration Dummy: 1 : R feels immigration 1603 0.02 0.13 is the most important. election issue 144 the unfolding results indicate that Australian political space is well-captured by a single dimension, multinomial regression methods are unnecessarily complex. To determine preferences over all parties, I instead examine individuals’ ideal points, recovered by the unfolding analysis. I model the ideal points as a linear function of sev- eral individual-level variables, derived from previous theory. Table 7.2 summarizes the independent variables used in the analysis and Table 7.3 displays the results of the. regres— sions. Because there are missing data on each of the covariates, the number of observations decreases from 1627 to 711 in Model 1. I include dummy variables for partisanship7 in Model 2, in which missing data drops the number of observations to 600. The effects of most of the explanatory variables do not change across the models and there is only slight variation in the significance levels of the coefficients across the models. Because it is the more complete model, I refer solely to the results in Model 2 from now on. In line with previous theory, I find that partisanship plays an important role in Aus- tralian voter behavior. The differences between the conditional intercepts for all of the included partisan categories and the reference category (Liberal) are significant at. the .01 level and in the expected direction, with the exception of Orie Nation. The magnitude of the differences between partisan groups is quite large. For example, all else constant, a Labor identifier is expected to be about 3.5 units to the left of a National identifer. Of the variables in the “opinion and information” category, ideology, which captures an individual’s values, beliefs, and attitudes about government, is significant at the .01 level and large in magnitude. In addition, those who protested recently or listed the Iraq War as the most important election issue are considerably to the left on the underlying continuum. As the 2004 Election took place soon after the U.S.-led invasion of Iraq, which was supported by the Coalition parties, it is likely that those who claimed to have recently 7I exclude these regressors from Model 1 due to the possibility that an individual will assign a higher value to the party he or she identifies with. In this case including partisanship on the right-hand side of the model will introduce simultaneity prob- lems. However, the literature on partisanship, starting with Campbell et a1. (1960), establishes that psychological attachments to parties do not necessarily correspond with agreement on issue positions. Thus, a voter may assign a high rating to a party which she currently agrees with on the issues, whether or not she has a psychological attaclm‘ient to the party. Table 7.3: Detm‘minants of Ideal Point Model 1 Model 2 Variable Coef. (p—value) Coef. (p-value) Demographic Income 0.007 (.881) -0.052 (257) Age 0.004 (.413) 0.004 (.372) Female -0.984 (.013) -1.216 (.001) Female >< Age 0.016 (.037) 0.021 (.004) University Education -0.127 (.345) -0.234 (.076) Unemployment -0.879 (.074) —0.973 (.042) Birth Nation 0.561 (.000) 0.559 (.000) Religiosity 0.032 (.362) 0.017 (.625) Rural 0.390 ( .009) 0.131 (.347) Union Member -0.326 (.013) -0.125 (.323) Blue Collar 0.058 (.736) -0.054 (742) Farmer 0.612 (.101) 0.153 (.668) Partisanship id Democrat -1.432 (.010) id Green -2.366 (.000) id Labor -2.314 (.000) id National 1.094 (.001) id One Nation 1.232 (.104) Opinion and Information Ideology 0.363 (.000) 0.174 (.000) Protest -0.786 (.000) -0.405 (.011) Pol. Information 0.085 (.488) 0.005 (.965) Gov. Performance 1.099 (.000) 0.634 (.000) Dem. Satisfaction 0.090 (.355) -0.011 (.908) Terrorism 0.193 (.529) 0.242 (.400) Defense 0.278 (.272) 0.130 (.593) Environment -0 . 704 ( .002) -0.382 (.096) Iraq —0.700 (.013) —0.739 (.012) Immigration -0.l23 (.775) 0.321 (.447) constant -5.889 ( .000) -2.071 (.000) n? 0504 0.735 Prob > F 0.000 0.000 'n 711 600 146 protested were protesting the war. The fact that these individuals have ideal points to the left of their counterparts is of little surprise, as the three parties on the left of the continuum, Labor, the Greens, and the Democrats, all ran on anti-war platforms. Also in the “opinion and information” category, the government performance variable is significant. A one point increase in ratings of the government’s performance correspmids with a .63 unit shift to the right on the underlying continuum, all else held constant. As the “government” being evaluated at the time of the survey consisted of the Liberal and National parties, this result is sensible; both parties fall on the right side of the continuum. Finally. the coefficient. indicating that those who see the environment as the most important. election issue are to the left of their counterparts (and more toward the Greens) on the continuum is significant at the .10 level. Regarding the demographic variables, the unemployed have ideal points significantly to the left of the employed and those born in Australia have ideal points to the right of the fmeign-born. In addition, at the .10 significance level, the results indicate that individuals with a university education lie to the left of those without on the underlying continuum. In addition, there is an interactive relationship between gender and age. Gender has a significant effect, with females tending to have ideal points to the left. of males, a finding in conflict with some previous research (for example, Jaensch 1995; Charnoek 1997). In addition, confirming Leigh’s (2005) finding that older women are less likely to vote Labor than Coalition, the interaction term shows that as women age, their ideal points shift to the right. 7 .4.1 Shifting Preferences How do party preferences diverge due to differences in individual—level variable values? First, consider a leftist voter, voter l, with an ideal point at -4.0. As shown in Figure 7.3, under the proximity voting assumption, this voter will order the parties Labor—Green— Deinocrat-Liberal-National-One Nation. Now imagine an individual, voter 2, who is essen- tially identical to this voter, except he is unemployed, and unlike his counterpart he sees the Iraq war as very important and thus joins some protests. The OLS results indicate 147 that his ideal point will be 973+.739+.405 2 2.12 units to the left of his counterpart’s. He thus will list the Greens first on his ballot, followed by the Democrats, Labor, the Liberals, the Nationals, and One Nation. These two voters will order the parties differently on their ballots based on differences in their values on the individual-level variables. Dem Gm Lab Lib Nat ON -6. -6.1 -4.4 O 3.6 5.0 7.8 I | . I I I l l I i l I Figure 7.3: T we Hypothetical Voters in Relation to the Six Parties 7.5 Ideal Point and Vote Choice: A Corroborat- ing Test Do the conclusions reached from a sole examination of the spatial proximity model, esti- mated with unfolding analysis, hold when actual vote choice data is introduced? Previous research that uses unfolding has found a direct link between actual vote choice and the distance between individuals and parties or candidates. For example, Lin, Chu, and Hinich (1996) find vote choice in Taiwan to relate to a voter’s proximity to a party. This conclu- sion is also reached by Hinich, Khmelko, and Ordeshook (1999) in a study of Ukrainian voting, and Dow (1998), also using unfolding analysis, shows the proximity of voter and party ideal points to relate to vote choice in Chile. Based on the insights from the cross-national analyses conducted in Chapter 5. prox— imity voting behavior should be prevalent in Australia; because its political variation arises from a single dimension, voters are likely to correctly identify the party closest to them in political space. To test whether the distance from a voter’s ideal point to a party’s ideal point truly corresponds to vote choice in Australia, I use multinomial logistic regression. 148 While. data on how the voters s1.1rveyed in the CSES actually ranked the parties is not included, information on the party each voter planned to assign their first prefercnce to is available. This variable is observed across 1474 respondents. The independent variable is individual ideal point from the unfolding results. Table 7.4 sunnnarizes the results of the regression. Table 7.4: Ideal Point. and Vote Choice Variable Coefficient (p—value) Democrats / Liberal Ideal Point ~1.054 (.000) constant -3.702 (.000) Greens / Liberal Ideal Point -1.395 (.000) constant -2.196 (.000) Labor / Liberal Ideal Point -1.185 (.000) constant -0.341 ( .000) National / Liberal Ideal Point 0.349 (.000) constant -2.985 (.000) One Nation/ Liberal Ideal Point 0.557 (.002) constant —5 .296 ( .000) n. 1474 Log Likelihood -1206.082 Psuedo—R2 0.294 The coefficients on ideal point for each equation in the multinomial model are in the expected directimi and significant; as a voter’s ideal point. moves rightward, she is more likely to select a party of the right as her first preference. As these coefficients in their raw form are not easilv interpretable, I display the predicted probabilities of voting Labor and voting Liberal across the range of ideal points in Figure 7.4.8 Preferences for Labor and Liberal peak once on either side of the continuum and then decline quite rapidly. This supports the proximity voting assumption; the propensity of voters to select a. given party decreases as the distance. between their ideal points and the. 8The figure was produced using Spost, with reference to Long and Freese (2006). 149 Q J E“? “ .5 (U .0 9 o. ‘O V: ‘ £2 .2 “U 2 n. o _ I I I I I -10 -5 0 5 1O Ideal Point ----- Labor —— Liberal Figure 7.4: Predicted Probability of Voting ALP or Liberal across Ideal Points 150 party’s location on the underlying dimension increases. The point of indifference between Liberal and Labor predicted by the multinomial logit model is at ideal point -.29. This is a striking result. as the point of indifference predicted by the spatial proximity model is at ideal point —.40. The indifference points between Labor and Liberal predicted by the proximity model and the multinomial logit model differ by only .11 units, less than 1% of the entire range of the ideal points. Clearly the proximity voting assumption is met and voter positions on the underlying continuum do predict vote choice. The four smaller parties are not included in Figure 7.4. To be sure that voter ideal points along the recovered socioeconomic dimension also correspond to vote choice for mi- nor parties, I produced a similar plot (Figure 7.5) to examine the relationship between ideal point and vote choice for the Democrats, the Greens, the National Party, and One Nation. The figure is again produced using predicted probabilities derived from the multinomial legit estimation depicted in Table 7.4. Regarding One N ation, the figure shows that as a voter‘s ideal point moves rightward, the probability of assigning a first preference to One Nation increases. The same is true for the National Party, and the probability of picking the Nationals is consistently higher than the. probability selecting One Nation. On the left, the figure shows that the proba- bility of voting Green systematically increases as ideal point moves leftward. Finally, the probability of assigning a first preference to the Democrats is low for all voters sampled. irrespective of ideal point. Nevertheless, there is a slight boost in the probability of voting Democrat for individuals with left-of-center ideal points. Thus, voter locations along the socioeconomic continuum predict vote choice not only for Labor and Liberal, but also for the. smaller Australian political parties. While this result does not eliminate the possibil- ity that postmaterialism helps to explain minor party voting, it further demonstrates the importance of the socioeconomic dimension in Australia for vote choice over all political parties. Predicted Probability 2 .3 .1 1 ....-.-. o, ./' o — :T;._._...-;:.>-. r T I I I ‘10 ‘5 0 5 1o Ideal Point -""-" Democrat —— Green --------- National —.—-- O.N. Figure 7.5: Predicted Probability of Voting Democrat, Green, National, or One Nation across Ideal Points 7.6 Conclusion In elections to the Australian House of Representatives, the use of alternative vote requires that voters rank order all of the competing parties on their ballots according to their preferences. Thus, studies of voting in Australia, or other multiparty systems that employ preferential ballots, should take into account all competing parties and the unique nature of the balloting formulae employed. To do so, this chapter uses the unfolding method developed in Chapter 2 to examine voting in Australia. I find Australian party and voter positions to be well-described by a single dimension — the common socioeconomic continuum. Parties are aligned from left to right on this dinwnsion according to their social, and economic policy stances. In addition, voters’ positions along the continuum are shown to vary with a host of individual-level factors. Thus, this study enhances the literature on Australian voting behavior in three ways. First. it confirms that the socioeconomic dimension is paramount to vote choice in Aus— tralia. Second, it identifies a host of variables that are empirically related to voter ideal points along the socioeconomic dimension, and these ideal points are shown to predict vote. choice for both major and minor parties in 2004. Lastly, it uncovers a clear and simple picture of party and voter locations in Australian political space using an empirical spatial analysis. Finally, this study adds to the comparative political science literature by introducing a new way to study voting behavior in preferential systems. By recovering voter ideal points, preferences over each competing party can be obtained. As simply considering first preferences of voters in preferential systems does not fully accoimt for the mechanics of these. institutions, future studies of voting behavior in such nations may benefit. from the approach employed in this project. 153 Chapter 8 The Dimensionality of Politics and Voter Behavior under Proportional Representation: The Case of Peru Alberto Fujimori won the presidency of Peru in 1990 in part due to his ability to moderate his rightist credentials and appeal to a wide spectrum of voters. In fact, many voters normally supportive of leftist candidates cast their ballots for the political newcomer. Thus the explanatory power of traditional link from socioeconomic class to vote choice was diminished. This hints that vote choice in Peru is very complex and may not conform to traditional theories of voter behavior. Can vote choice in Peru be predicted with quantitative models, or does its nascent party system, unique candidates, and numerous other idiosyncracies preclude such generalizations of voter behavior? Peru is a developing country situated on the western coast of South America. Hav- ing proclaimed independence from Spain 1821, Peru has since remained sovereign. With roughly 29 million people and a per capita GDP of U.S.$7,800(PPP), it falls in the middle of Latin American countries in terms of both population and wealth. In 1979, democratic. 154 government returned after 30 years of military rule. Peru’s unicameral legislature contains 120 members and is elected with open-list pro- portional representation via the d’Hondt electoral formula. Legislators were elected from 25 different districts in 2001 and 2006, meaning district magnitude was 4.8, and in 2000 Peru used a single national district to elect its lower house. Thus, the Peruvian electoral system is very permissive to small party entry. The effective number of parties in congress, using Laakso and Taagepera’s (1979) measure, was roughly four in 2000, 2001, and 2006,1 and in each election at least seven parties gained one or more seats in congress. As shown in Chapter 4, political variation in countries with permissive electoral systems generally does not conform well to a single dimension. And, as expected. Peruvian political space. is not unidimensional. In fact, the R2 values from the unfolding results produced in Chapter 3 are .63 in 2000, 0.62 in 2001, and 0.72 in 2006, each below the mean value of 0.7:"). This means that a single dimension explained less than two-thirds of variation in party and voter positions in Peru in 2000 and 2001, and less than three-fourths of this variation in 2006. Using data from the Comparative Study of Electoral Systems, which conducted com- }n'ehensive surveys of Peruvian voters during the presidential and congressional elections of 2001 and 2006, I conduct an in depth analysis of voter behavior. Because the theory advanced in this project does not consider presidential elections, I only examine voter behavior in the congressional contests. The CSES also surveyed Peruvian voters in 2000, but did not record congressional vote choice. Based on the insights from the cross-national analyses conducted in Chapter 5, prox- imity voting behavior should not be observed in Peru; because political variation does not arise from a single dimension, voters are less likely to correctly identify the party closest to them in political space. In addition, Peru’s multiparty elections and compulsory voting rules should negatively affect the likelihood of casting a. proximity vote. I test these expectations with an alternative-specific multinomial probit. model. which is well-suited for the study of multiparty elections. As anticipated, I find that proximity 1111 2000 it was 3.84, in 2001, it was 4.37, and in 2006 it was 3.95. 155 voting did not occur in Peru in 2001. In 2006, proximity considerations did enter the voting calculus, but had a minimal role relative to the effects of other factors, including both partisanship and positive affect for political parties. In addition, education, gender, income, and evaluations of government performance each affected voter behavior in 2006, while faith in the political process and education levels shaped vote choice in 2001. The links from the independent variables vote choice are not uniform across the election years. demonstrating a lack of systematic empirical relationships between external factors and vote choice in Peru. 8.1 The 2001 and 2006 Congressional Elections in Peru The decade leading up to Peru’s 2001 elections was colored by the presidency of Alberto Fujimori. Much of Fujimori’s rise to power in 1990 was due to the ineptness his main competitor and a deterioration of the party system throughout the 19808, which left a hole in the middle of the political spectrum. Peru’s use of open-list proportional representation with a high district magnitude also weakened party loyalties and set up rivalries within right-wing parties. Though Fujimori was a candidate of the right, he appealed to the center by speaking in broad generalities (Schmidt 1996). Fujimori’s presidency, however, was riddled with scandal and nondemocratic reforms, and his 2000 reelection was widely condemned as being neither free or fair. And, by late that year, Fujimori fled to Japan facing allegations of corruption (Schmidt 2002). Peru’s 2001 elections were conducted in the wake of the 2000 debacle. However, un- like the 2000 election, the 2001 contests were internationally praised. The winner of the presidential race was Alejandro Toledo, who ran as a candidate of the Possible Peru (PP) party that he founded. In terms of votes received, the main competing parties in the 2001 congressional elec— tions were Possible Peru, the American Popular Revolutionary Alliance (APRA), National Unity (UN). and the Independent Moralizing Front (FIM). The PP finished in first place in 15 of Peru’s 25 districts and won 26.5% of the vote and 37.5% of congressional seats. APR A benefited from a split of the center—right. which was caused by Fujirnori’s exit from Peruvian politics (Schmidt 2003), and received 19.7% of the vote and 23.3% of seats. UN received 13.8% of the vote and 14.2% of seats, while FIM received 11.0% of the vote and 9.2% of the 120 elected seats. Several smaller parties also competed, but none received more than 6% of the vote. Table 8.1: Main Peruvian Parties in the 2001 and 2006 National Electirms Party General Description Alliance for the Future Right-wing ideology. (AF) Contested 2006 elections. Associated with Alberto Fujimori. American Popular Social democratic ideology. Revolutionary Alliance. Won presidency in 2006. (APRA) Strong showing in both 2001 and 2006. Has existed in Peru since 1920s. Independent Moralizing Centrist ideology. Front (FIM) Contested 2001 elections. Aligned with Possible Peru. Disbanded after poor showing in 2006 elections. National Unity (UN) Christian democratic ideology. Contested 2001 and 2006 elections. Possible Peru (PP) Centrist. ideology. Won presidency and congressional majority in 2001. Won only two seats in 2006. Union for Peru (UPP) Social democratic ideology. Won congressional majority in 2006. President Toledo and his PP led Peru through a half-decade of robust economic growth. Ilr‘mrever, voters became disillusioned with the president’s extravagant lifestyle and the corruption within his government, and the PP (lid not put forth a presidential candidate in 2006 (Schmidt 2007). Instead, APRA won the presidency behind the leadership of Alan 157 Garcia and enjoyed another strong showing in congress, winning 20.6% of the votes and 30.0% of the seats. The Union for Peru (UPP) won the largest share of the congressional vote at 21.2%, which translated to 37.5% of seats. UN had another solid showing with 15.3% of the vote and 14.2% of seats. Finally, the Alliance for the Future (AF), formed through an alliance of pro—Fujimori parties, came in fourth place in the congressional elections with 13.1% of the vote and 10.8% of seats. Several smaller parties also contested the. election, all receiving less than 8% of the vote. In this research I only examine parties that received 10% or more of the vote congressional elections. Each party examined in this chapter is described in Table 8.1. 8.2 Previous Research, Variables, and Expecta- tions There is an extensive subfield of research on voting behavior and elections dating back several decades. "Using previous comparative and American voting research as a guide. I identify several variables related to vote choice. I also make some specific predictions about vote choice in Peru based on the limited body of research on Peruvian voting behavior. First. positive affect for a party will influence one‘s vote. Put simply, individuals will vote for parties that they like. As expected. Echegaray (2005, 138-140) finds that, in the 1995 presidential election, voters were much more likely to vote for the incumbent if they held a favorable view of him. In 2001 and 2006 the CSES asks voters to rate most or all of the competing political parties in their home nation.2 I use this question to gauge each individual’s feelings over each party. The “Michigan model” of Campbell, Converse, Miller, and Stokes (1960) connects psychological attachments to political parties with voter behavior. These attaclmients, 2 Question wording: “I’d like to know what you think about each of our political parties. After I read the name of a political party, please rate it on a scale from 0 to '10, where 0 means you strongly dislike that party and 10 means that you strongly like that party. If I come to a party you haven’t heard of or you feel you do not know enough about, just say so.” 158 instilled at a young age, can affect voting behavior throughout an individual‘s life. Scores of studies have reaffirmed the notion that partisanship is related to vote choice, both in the U.S. and cross-nationally (Bartels 2000; Green, Palmquist, and Schickler 2002; Miller and Shanks 2001; Nadeau and Lewis-Beck 2001). Thus, although partisanship is comparatively weak in Peru (Norris 2004, 139), voters there do rely on partisan attachments when casting their ballots (Echegaray 2005, 138-139). In the 2001 and 2006 election cycles the CSES asked respondents which major parties competing in the election they most. identified with, if any. Using this question, I create a dummy variable which equals 1 if an individual most identifies with a given party and 0 otherwise. Regarding socioeconomic status and class, Roberts and Arce (1998), who examine the patterns of support for Alberto Fujimori in the mid-1990s, find that the rightist candidate enjoyed significant support from the lower-class. Fujimori was a unique candidate with numerous idiosyncracies voters found pleasing and an ability to appeal to a broad range of voters. While this achievement may not have been possible for most rightist candidates, it demonstrates a breakdown of Peru’s once-polarized class-based voting patterns (Schmidt 1996). In addition. Echegaray (2005, 138) finds that class is only weakly related to vote choice in Peru. As such, there may not be a clear relationship between socioeconomic status and voting patterns. To gauge household income, the CSES separates respondents into quintiles. I use this measure to capture respondents’ socioeconomic class. The retrospective theory of economic voting3 posits that voters reward or punish gov- ernments and politicians for recent economic outcomes, whereas prospective theory argues that voters choose based on future expectations of performance.4 Perceptions of economic outcomes may be either egotropic, based on the voter’s own well-being, or sociotropic, based on the well-being of society as a whole. Thus, if the economy is functioning well. incumbents should enjoy an electoral bump, and in periods of recession, incumbents should 3In the American context, see Key (1966) and Fiorina (1981). In the comparative context, see Lewis-Beck and Stegmaier (2000). 4For example, MacKuen, Erikson, and, Stimson (1992) claim that long-term expectations of the economy affect presidential approval, which is tied to voting, rather than past performance, rejecting long-standing claims that the electorate views parties and leaders retrospectively. expect to be punished. Echegaray (2005) surveyed 519 voters in Peru shortly before the 1995 election to ex- amine what variables affected incumbent support. He finds strong evidence that Peruvian voters base their decisions on both retrospective and prospective evaluations of the econ— omy. Other studies examine presidential approval in relation to economic concerns. This is a useful approach, as presidential approval is strongly tied to an individual’s vote de— cision (Abramson, Aldrich, and Rohde 2007; Fiorina 1981). First, Morgan Kelly (2003), examining approval in Peru throughout the 19908, concludes that Peruvians hold politi— cians accountable for both the economic past and “future implications of past actions” (864). W'eyland (2000) finds that approval of President Fujimori depended on approval of his economic policies, and individuals were more likely to vote for him when the econ- omy was performing well. Finally, Singer (2007) finds that support for President Toledo‘s government from 2001-2006 was positively related to prospective sociotropic evaluations of the economy (82, 197), and approval of President Fujimori in 2000 was related to both prospective and retrospective evaluations (131, 204). The CSES in 2001 asks voters only their retrospective sociotropic opinions of the economy,5 limiting my ability to compare the predictive power of the prospective and retrospective theories of economic voting. The variable is coded from 1 to 5, with higher values indicating positive feelings about the economy. For the 2006 election the CSES does not ask respondents to evaluate the economy specifically, but instead asks their opinions of government performance on the issue they saw as most important.6 The variable is coded from 1 to 4, with higher values indicating positive feelings about govermnental pt‘irformance. Note that 55% of voters saw economic issues as most important. Seeing the political process as valid may also affect onc‘s vote choice; those who are upset. by the current state of affairs may cast a protest vote for an “outsider" party, if 5Question wording: “What do you think about the state of the economy these days in Peru? Would you say that the state of the economy is very good, good, neither good nor bad, bad, or very bad?” bQuestion wording: “And thinking about [the issue you see as most important], how good or bad a job do you think the government/ president in Lima has done over the past five years. Has it / he done a. very good job? A good job? A bad job? A very bad job?” 160 they choose to vote at all. I gauge political efficacy with a CSES question which inquires whether a respondent. feels that his or her vote makes a difference in the political process.7 The variable is split into five categories, with higher values corresponding to more political efficacy. I also include variables for age, education, and gender. In 2001, age is measured in quartiles.8 and in 2006 it is measured simply as a respondent’s age in years. To gauge education, I create a dummy variable differentiating university graduates, coded 1. from others, coded 0. Gender is coded as a dummy variable, 1 for females and 0 for males. Each variable is obtained from the CSES and is summarized in Table 8.2. Table 8.2: Summary Statistics —— Variable Mean Std. Dev. Min. Max. n 2001 Elections Distance 2.317 1.853 0.057 7.404 4072 Affect 4.559 3.281 0 10 4339 Party ID 0.062 0.241 0 1 4472 Age 2.394 1.149 1 4 1118 Female 0.498 0.500 0 1 1118 Income 3.021 1.196 1 5 1118 Education 0.216 0.412 0 1 1115 Efficacy 4.229 1.203 1 5 1102 Econ. Performance 2.107 0.852 1 5 1117 2006 Elections Distance 2.494 1.937 0.009 7.268 6716 Affect 3.333 3. 252 0 10 7439 Party ID 0.082 0.274 0 1 8128 Age 37.947 14.438 18 95 2032 Female 0.501 0.500 0 I 2032 Income 2.985 1.422 1 5 1907 Education 0.113 0.317 0 1 2031 Efficacy 3.761 1.380 1 5 1924 Gov. Perfm'mance 2.208 0.767 1 4 1958 ’ Question wording: “Some people say that no matter who people vote for, it won‘t make any difference to what happens. Others say that who people vote for can make a. difference to what happens. Using the scale on this card, where would you place yourself?” RF ‘ . - ‘ 3 t l f ' ‘" F he categorles are: 1, 18—25 years; 2, 26-35 years; 3, 36-45 ; 4, 40-65 years 161 8.2.1 Proximity Voting in Peru The proximity model of voting (Downs 1957; Hotelling 1929) is generally accepted as an accurate portrayal of voter behavior in the political science literature and has weathered several theoretical and empirical tests throughout the past several decades (see, for exam- ple, Blais, Nadeau, Gidengil, and Nevitte 2001; Westholm 1997). The model predicts that. voters choose the candidate or party closest to them on some ideological continuum in any given election. If the theoretical predictions of the proximity model hold, an increase in the distance from a particular party should decrease the likelihood of voting for that party. However, as shown in Chapter 5, proximity behavior is less likely in countries where political variation is not. captured by a single dimension. The logic behind this relationship is straightforward: as political space becomes more complex, it becomes more difficult. for voters to locate the most, proximate party and cast their vote accordingly. In addition, multipartism and compulsory voting9 in Peru should negatively affect. the likelihood of casting a proximity vote. Chapter 5 shows that multiparty elections lessen the likelihood of a proximity vote, as increasing the number of parties makes it harder to “correctly” discern which party is most proximate (Lau and Redlawsk 1997). Addi- tionally, compulsory voting increases turnout among disinterested and uniformed citizens (Jackman 2001). Thus, proximity considerations will be less likely when voting is coerced, as disinterested voters may choose essentially at random when in the polling both. Thus, l do not. expect. to find strong evidence of proximity voting in Peru. Party and voter positions are available from the unfolded placements produced in Chapter 3. 1 do not, however, use these positions to obtain the distance of each party from each voter. Political variation in Peru did not arise from a single dimension in either 2001 or 2006. As mentioned, the R2 value from the unfolding results in Peru is 0.62 in 2001 and 0.72 in 2006, both below the mean value of 0.75. Thus, the recovered unidimensional party and voter placements do not necessarily correspond to a logical ordering of political players. ( . . . . . JVotmg is mandatory for c1tlzens under 70 years of age. 162 PL Instead, I obtain party positions from aggregated individual perceptions of the politi- cal parties’ positions, as the CSES asks respondents to locate parties along the left-right continuum with a number between 0 and 10. Measuring party positions with such re- sponses can be problematic. Respondents may place their most-preferred party closer to their own position, regardless of that party’s true position (Adams, Merrill, and Grofman 2005, 170). Additionally, they may shape their responses to meet the proximity voting criterion (Boatright 2008). Such rationalization of perceptions lessens the reliability of individual placements (Macdonald, Rabinowitz, and Listhaug 2007). While I would have preferred to use expert party placements, these were available only for the 2006 survey. Thus, to maintain consistency, I use averaged individual-level placements for each year. To minimize the potential problems with these placements, I consider only the perceptions of college-educated respondents. The CSES asks voters to locate themselves along the same left—right continuum on which they place the parties.10 I compare each voter’s self-reported location to each party’s position to gauge proximity. The resulting variable, distance, is summarized in Table 8.2. Figure 8.1 displays party locations and the distribution of the self-reported voter lo- cations. UN is further to the right than expected, based on its moderate dispositions (Schmidt. 2002, 345). In addition, in 2006, the pro-Fujimori AF is located at the center of the spectrum, which is surprising in light of Fujimori’s right-wing policies. However, this may reflect. the fact. the Fujimori was particularly adept at reaching out to centrist voters. 8.3 Methodology To assess vote choice in Peru I use an alternative-specific multinomial probit model (ASMN P). This model is commonly employed to study the decisions of individual voters over several political parties (Alvarez and Nagler 1995; 1998). Like case—specific multino- 10Question wording: “In politics people sometimes talk of left and right. Where would you place yourself on a scale from 0 to 10 where 0 means the left and 10 means the right?” 163 r..- v...2.! F. .il..'~‘l MIN '3 '_ ' | u 2001 2006 r2 0.000 168 Table 8.4: Vote Choice in the 2006 Peruvian Congressional Elections Variable Coefficient (p-value) Alternative-Specific Variables Distance -0.094 (.007) Affect 0.240 (.000) Party ID 0.986 (.00) APRA/UPP Age -0.005 (.499) Female 0.277 (.199) Income 0.176 (.024) Education -0.878 (.012) Efficacy 0.004 (.953) Gov. Performance -().082 (.524) UN / UPP Age -0.001 (.844) Female 0.356 (121) Income 0.229 (.007) Education ~0.086 (.798) Eflicacy -0.125 (.108) Gov. Performance -0.119 ( 383) AF / UPP Age -0.021 (.195) Female 1.197 (.012) Income -0.026 (.874) Education —0.538 (.453) Efficacy 0.010 (.945) Gov. Performance -0.806 (.015) '7) (case-specific) 686 n (alternative-specific) 2744 Log Likelihood -553.346 Prob > 3,2 0.000 '1 69 alum @531 E mEoBEooo Emowfimfi 50c vapflsofimo m2_.EEmnoE “53.555 mode- mood- fiend- god Sod owed Eod- m4. mmod- maod mid wmod nwmd wmmd owed- ZD owed made- owed mmod- ommd Pond mmod- ommoommofiofizofix‘. F 8cm .sEEEOE was; a mango es same «mm 2an E 358580 Emowfiwfi 89a wmofizofimo mmflzfienoa @8214an owed- hmod- wmmd mafia 32m mmod mmod- End wumd 75 ammo- mood oomd mmmd «dag. mmfio some oomd owmd mm 392 on 4:2 .33 3 .25 oz mEQH on min: .52 3+3 3 1 d homoEm— comawosvm n: 3.3m newtd. 3.8m meE§E> oEooamemmO 4 exacts.) oEoonézpeEepZ _ _ flametzfio.” m_.:3. Emowcfi; no. man Sam .._.__L ../. 5 Lw:.b 170 As expected, the coefficient on the distance variable is insignificant in the 2001 election. However, in the 2006 election, the coefficient on the distance variable is negative and significant; supporting Downsian predictions, as the distance between a given party and voter on an underlying ideological continuum increases, the probability of voting for that. party decreases. Nevertheless, Table 8.6 shows that the substantive effect of this variable is minimal in 2006', a shift. from the mean of the distance variable to one standard deviation above the mean decreases the probability of voting for a particular party by only a very small margin. relative to the effects of the other variables. 8.5 Conclusion Peru is an independent Latin American nation with 30 years of continuous democracy. The country elects its unicameral legislature with open-list proportional representation, which fosters a robust multiparty system. Chapter 4 shows that political variation in countries with permissive electoral systems generally does not conform well to a single dimension. And. as expected, Peruvian politics are not unidimensional; a single dimension explained less than two-thirds of variation in party and voter positions in Peru in 2000 and 2001, and less than three-fourths of this variation in 2006. Using data from the Comparative Study of Electoral Systems, in this chapter I con- duct an in depth analysis of Peruvian voter behavior in the 2001 and 2006 congressional elections. Based on the conclusions of the cross-national analyses conducted in Chapter 5, I expect that proximity voting should be minimal in Peru. That is, because political variation in the country does not arise from a single dimension, voters are less likely to cor- rectly identify the party closest to them in political space. In addition, Peru’s umltiparty elections and compulsory voting rules should negatively affect the likelil'mod of casting a. proximity vote. I use a sophisticated alternative-specific multinomial probit model, which is well suited to the study of multiparty elections, to test these expectations. As anticipated. 1 find that proximity voting did not occur in Peru in 2001. It'loreover, in 2006, while proximity 171 considerations did enter the voting calculus, they played only a minor role as compared to the effects of other farmers. In sum, this case study of Peru confirms the expectations derived from the cross-national analyses, while providing an in depth analysis of Peruvian elections in 2001 and 2006. Moreover, it. demonstrates that the links from the independent. variables to vote choice are not uniform across the two elections; there are few systematic empirical relationships between external factors and vote choice in Peru. 172 nl-Im-In—r-‘I‘ .M'W. .1 . Chapter 9 Conclusion Previous research identifies a clear relationship between political institutions and electoral behavior. However, the avenues through which institutions affect behavior are frequently muddled. The causal chain I put forth in this dissertation connects party and voter behav- ior to the underlying dimensional configuration of a country’s political space. I introduce new measures of dimensionality and party and voter locations, and I show that dimensional configurations are themselves affected by electoral institutions. As such, dimensionality acts as a catalyst in the well-established relationship between institutions and political be— havior. In short. this research identifies and measures a previously unconsidered mediary between institutions and behavior. In researching this causal chain, this work examines the subfields of voter behavior, party behavior. and electoral institutions, connecting each chapter to the next with the common theme of dimensionality. The overall contribution of this dissertation, then, is twofold: First, I add a new variable, “political dimensionality,” and empirically derived party and voter locations to the existing cross-national electoral research. Second, I un— cover a theoretical and empirical link between a nation‘s dimensionality. its institutimial setup, and its political outcomes. 1. 73 9.1 So What? The conclusions of this dissertation are of value to politically interested academics, jour- nalists, and casual observers. While it provides a foundation for terms commonly used by journalists and commentators, it also introduces unique explanations of empirical phe- nomena. and original data to academic researchers. 9. 1 . 1 Real—World Importance Journalists and commentators often couch their speech and writing in dimensional terms. When a politician is described as “left-wing,” there is an implication that he is on the left side of a single continuum that accounts for most of the variance in the political outcomes of a given country. However, we do not know what differentiates “left” from “right” in this context, how much variance this continuum accounts for, or whether there is another dimension which differentiates this politician from others. This work explicitly examines the oft-referred to left-right continuum, testing its validity and examining whether the construct varies systematically across nations. In the end a clearer understanding of “left” and “right” across countries emerges. Thus, this work adds empirical and theoretical substance to these everyday concepts. For example, politics in the United Kingdom and Australia are well—captured by a single dimension, which is defined by common socioeconomic political positions. Thus, describing a voter as “left” in either of these nations should conjure up thoughts such as support for workers, fair trade, and decreased governmental involvement in personal mm'al decisions. In addition. this simple one-word label accurately places a voter in space in either of these. countries as other dimensions either do not exist or are much less salient than the socioeconomic continuum. Alternatively, in countries such as Hong Kong or Taiwan, describing a voter as “left.” invokes entirely different political positions. In these nations, though politics are quite well- caj.)tured by a single dimension, it does not describe socioeconomic positions. Instead, the dimension generally captures orientations toward the People’s Republic of China. Again. 174 this one-word label accurately places a voter in space in either of these countries, as other dimensions do not play an important role in capturing political variation. Finally. in countries such as Peru and Mexico describing a voter as “left” does little to convey information about this individual. Because political variation in these nations is not. well-captured by a single dimension, unipolar descriptions of voter locations are not sufficient for distinguishing their political inclinations. Instead. to convey information about an individual, one must also consider her position along a second, third, or even fourth dimension. Thus, this work shows that the implications of the terms “left” and “right” vary in both substance and explanatory power cross-nationally. While it. is well understood in- dividual countries have unique political variation due to their idiosyncratic histories, cul- tures, religions, economies, and traditions, no previous work has systematically examined the underlying political space which captures such variation. This work takes a step in this direction, providing a rigorous treatment. of the differences in the dimensionality of politics across countries. 9.1.2 Academic Importance Most institutionalists do not consider the dimensions of conflict in the nations they study. And, if they do, these dimensions are hypothesized by formal models and rarely extracted empirically. Thus. this project’s first contribution is new empirical measures of dimension— ality and 1')arty and voter positions. Previous work measures party positions using various data sources and methods. For example, expert surveys, which rely on well-informed indi- viduals sub jective judgements of party locations, usually along one dimension, are. often used to gauge party positions. Recent work by McDonald and Mendes (2001) notes that the party positions found in three previous studies that used eXpert scales all correlate highly (Huber and Inglehart 1995; Laver and Hunt 1992; Castles and Mair 1984). Other researchers, such as Laver, Benoit, and Garry (2003) and Budge, Klingemann, Volkens, Bara, and Tanenbaum (2001) gauge positions based on party manifestos. inter- preting them either with human coders or computer-assisted analysis. This approach has 175 an advantage over expert surveys in that it provides a means of estimating party positions over time (Gabel and Huber 2000). However, it is heavily reliant on the subjective. evalua- tions of those who code the manifestos (though computer-aided analysis helps amelim'ate this problem). Another emerging method uses the words spoken during parliamentary deliberation to estimate policy positions (see, for example, Monroe, Colaresi, and Quinn 2008). As noted by Marks (2007, 3), “There are two ways to increase the volume of infor— mation: one can repeat an observation that one has already made, trying to keep all relevant. conditions the same, or one can observe from a different angle, using a different method.” In this work I do the latter, determining party and voter positions using an empirical method. I surmise that the method used to locate parties and voters, empirical estimation of a spatial proximity model with unfolding analysis, is more. objective than the methods employed by previous measures; as it extracts party positions from actual data on voter opinions, it creates a picture of political space based solely on individuals’ party preferences within a given nation. In addition, the technique employed provides an overall gauge of how well a single dimension represents these preferences. More specifically, this measure quantifies the pro- portion of variation in party and voter locations captured by a single dimension. Thus, I provide. much new data to comparative researchers, including a measure of dimensionality, voter positions, and party positions, all extracted from information supplied by voters. No previous research explains the link between electoral institutions and sociopolit— ical outcomes with dimensionality. Thus, the second contributirm of this work is the introduction of dimensionality as an intermediate step in this link. This project uses the dimensionality of politics to tie together several political science. subtopics. including party behavior, voter behavior, and electoral systems. Linking these areas with a unifying concept provides a parsimonious explanation of their relationships. I show that the (li- mensionality of politics is dependent on a nation's electoral system. and, in turn. it affects voting behavior and party behavior. As such, a clearer picture of the link from electm'al institutions to parties and voters is drawn, and several interesting and important. academic observations about political behavior are made. 9.2 Summary of the Project This dissertation is broken into three parts. The first part of the project, put forth in Chap— ters 2 and 3, details the methodology used to measure dimensionality and displays and explores this new measure. The second part, put forth in Chapters 4—6, cross—nationally examines the interplay between the dimensionality of politics, electoral institutions, voter behavior, and party behavior. The last part, put forth in Chapters 7 and 8, employs in depth country-level analyses to substantiate and test the. cross-national findings. Below I provide a short summary of each substantive chapter. Chapter 2: Methodology and Measures In this chapter I introduce a new way to conceptualize and measure the dimensionality of politics across countries. While previous measures are concerned with the number of issues or ideological divides in a country, the measure introduced here explicitly gauges the space in which political parties compete. The measure is created with unfolding analy- sis. The method, which is based on an underlying geometric model of spatial proximity, recovers the dimensionality of the space it is applied to and locates stimuli (parties) and individuals (survey respondents) along these dimensions with interval-level values. Associ— ated with these results are statistics indicating the “goodness of fit” of the. model, or how much variance in voter preferences the recovered dimensional construct explains. Fiom these statistics I derive the new measure introduced here, “political dimensionality.” Chapter 3: Political Dimensionality across Nations In this chapter I introduce the new measure of political dimensionality derived from the unfolding analysis. The measure covers several countries across the years 1996-2006. 177 Comparing the measure to other variables indicates that, for most countries, the most salient political dimension is the common left-right, socioeconomic continuum. However, in nations where national politics are defined by atypical forces, the substance of the di- mension is different. I also report the party and voter locations associated with the new measure in this chapter. In many nations, especially those without fringe parties. the placement of the political parties is intuitive. In other nations party placements along a single dimension do not correspond to expectations. For example, parties that embody se[_)aratist issues are often placed at nonintuitive locations. This is likely because such parties base their existence on a second dimension that is highly unrelated to, or even orthogonal with, the primary dimension in their home country. Chapter 4: Electoral Systems and the Dimensionality of Politics Electoral systems are known to shape numerous political, social, and economic out— comes. However, their relationship with the dimensionality of politics in a country is scarcely explored. Previous theory posits that entrenched parties compete over fewer issues when electoral systems are restrictive to party entry. thus lowering political di- mensionality. Conversely, in permissive systems, parties are inclined to adopt emerging issues out. of the fear of losing parliamentary seats, thus increasing the dimensionality of underlying political space. In this chapter I examine these expectations with the new measure of dimensionality. I find that electoral rules do systematically affect the character of a nation’s underlying political space, even when other potentially salient factors and endogeneity are accounted for. Majoritarian electoral institutions lead to unidimensional political space, while the politics of countries with proportional systems do not conform to a single dimension. This provides an important contribution to the understanding of dimmsiouality across nations. Chapter 5: Electoral Behavior and the Dimensionality of Politics: A Cross- National Examination of Proximity Voting 178 It is generally held that individuals vote for the party that most closely aligns with their preferences, yet previous research identifies numerous factors which lead individu- als to stray from the proximity logic. To shed light on this phenomenon, in this chapter I examine proximity voting from a comparative perspective. Results from a multilevel model indicate that several individual- and election-level factors affect the likelihood of a proximity vote. I also find proximity voting to occur less in countries where political variation is not well-captured by a single dimension. These findings shed light on the bases of proximity voting and add to the general understanding of the nature of voting behavior. Chapter 6: Electoral Rules, the Dimensionality of Politics, and Party-Voter Correspondence across Nations In democracies the relationship between the constituent and the representative is of fundamental importance. Yet the nature of representation is not uniform throughout the world; political institutions are known to place constraints on leaders and citizens that shape their behavior, and thereby the character of representation. In this chapter I ex- pand upon the cross—national examination of representation, examining how it varies with the dimensionality of politics in nations. I expect that party-voter correspondence will be high in nations with simple dimensional constructs. Alternatively, in countries where political space is not well-captured by a single dimension, representatives are less likely to accurately reflect the desires of constituents. To test these expectations, I examine how well party positions mirror both the median and spread of voter preferences, conditional on the electoral institutions and political dimensionality of nations. Using data from a wide sample of nations and the new measure of dimensionality, I find that the positions of parties correspond more closely to those of voters in countries with low-dimensional polit— ical space, whereas electoral systems play a smaller role in the nature of representation. Chapter 7: The Dimensionality of Politics and Voter Behavior in Preferential 179 Systems: The Case of Australia This chapter examines the dimensionality of politics and voter preferences in Aus- tralia. As the theory in Chapter 4 predicts, parties and voters are well-organized along a unidimensitmal socioeconomic continuum in Australia’s majoritarian electoral system. Individual-level variables, derived from previous theory, are used to predict voter ideal points on this continuum. Hem the ideal points, voter preferences over each party are ascertained. Thus, this analysis adds credence to the cross-national portions of this dis— sertation while providing an in depth analysis Australian electoral behavior. In addition, it introduces a new way to study voter behavior in preferential electoral systems; because this approach allows for a full examination of voter preference orderings, it is important to the study of voting behavior and representation under such electoral arrangements. Chapter 8: The Dimensionality of Politics and Voter Behavior under Propor- tional Representation: The Case of Peru In this chapter I conduct an in depth analysis of Peruvian voter behavior in the 2001 and 2006 congressional elections. As shown in Chapter 5, because political variation in Peru does not arise from a single dimension, voters are less likely to correctly identify the party closest to them in political space. As such, I expect that proximity voting should be minimal in Peru. Using an alternative-specific. multinomial probit model, I find that proximity voting did not occur in Peru in 2001. Moreover, in 2006, while proximity ctuisideraticms did enter the voting calculus, they played only a minor role as compared to the effects of other factors. This analysis confirms the expectations derived from the cross—national portions of this dissertation, while providing an in depth analysis of Peruvian elections in 2001 and 2006. 180 9.3 Shortcomings Though this project makes several important contributions to con'tparative political sci- ence, it also has its share of drawbacks. First, while this project introduces new measures of voter and party positions across countries, the recovered unidimensional party and voter placements do not necessarily correspond to a logical ordering of political players. This is because political space is poorly captured by a single dimension in some countries, making voter and party placements nonintuitive or sometimes downright nonsensical. To combat this problem, I substantiate all quantitative analyses with party locations derived from either expert opinions or aggregate perceptions of voters. In addition, I consider self—reported voter positions in addition to those derived from the unfolding model. In addition, in nations where unidimensional space does not sufficiently capture po- litical variation, this project ignores the potentially-salient extra dimensions. While mul- tidimensional unfolding models do exist and could be used to explicitly model complex political space, the focus of this project is on the strength of the first dimension. Thus, this project. leaves something to be desired. Future work should aim to explore the politics of nations with complex political space and determine the substantive character of each dimension in these countries. Lastly, the measure of dimensionality introduced in this work spans only 42 countries. As some nations are surveyed across multiple elections, there are a total of 81 cases. While this expands the coverage of previous measures of dimensionality, an increase in sample size is desired to better-facilitate the estimation of advanced statistical models and increase the generalizability of findings. Fortunately, as the CSES continues to survey more and more. countries and elections, new measures of “political dimensionality” can be produced for each election. 181 9.4 Final Thoughts: Flatland and the Dimension- ality of Politics In his 1884 novel, F latland, Edwin Abbott describes the life of a humble square living in two dimensions (2006). One night, the square has a dream about a journey to a unidimensional world called Lineland. While in Lineland, he tries to convince the domain’s ruler of the existence a second dimension. However, as the rulers perspective is myopic due to a life of 11nidin'iensionality, the square is unable to convince him to envision life beyond the single dimension in which he lives. Similarly, the square himself knows nothing of a third dimension until he is visited by a sphere from Spaceland. Initially hesitant to believe in the existence of extra dimensions, the sphere eventually provides enough evidence to convince the square of their reality. The square then tries to spread this knowledge to the other inhabitants of Flatland. However. he is unsuccessful in doing so, as other F latlanders are hesitant to accept his outlandish ideas. In addition, he upsets the sphere when he suggests the existence of dimensions beyond the third. This dissertation. written in a three-dimensional universe on a two-dimensional screen, examines the salience of the first dimension to political outcomes across countries. In other words, this work examines whether politics across countries live in Lineland. In many instances, political variation is indeed unidimensional. In other cases, Lineland is insufficient for explaining politics. This dissertation develops and tests theory about why this variation exists - why the dimensionality of politics varies across countries. In doing so. it finds many interesting patterns across nations and makes important contributions to comparative and quantitative political science. Nevertl‘ieless, just. as the square in F latiand saw the world more clearly once he learned of Spaceland. in many countries we may see politics more clearly if we go beyond Lineland. However, this work does not venture into multidimensional spaces. What remains to he done, then, is a systematic cross-national study of the second, third, fourth, ..., and nth political dimensions. Hopefully in the end, just. as the eyes of the two-dimensional 182 protagonist in F latland were opened to a new world when he was visited by a sphere, a cross-national visit to n—space may well open the eyes of political scientists to new and important measures, empirical patterns, and theories. 183 n! J-l .I‘Ifi'i :3 u” BIBLIOGRAPHY Abbott, Edwin A. 2006. F latland: A Romance of Many Dimensions. Oxford: Oxford University Press. Abramson, Paul R., John H. Aldrich and David W. Rohde. 2007. Change and Continuity in the 2004 and 2006 Elections. Washington: CQ Press. Achen, Christopher H. 1977. “Measuring Representation: Perils of the Correlation Coellieient." American Journal of Political Science 21(4):805~15. Adams, James, Andrea B. Haupt and Heather Stoll. 2009. “What Moves Parties? The Role of Public Opinion and Global Economic Conditions in Western Europe.” Comparative Political Studies forthcoming. Adams, James F., Samuel Merrill III and Bernard Grofman. 2005. A l.»'nified Theory of Party Competition: A Cross-National Analysis Integrating Spatial and Behav- ioral Factors. Cambridge: Cambridge University Press. Adams, James, Michael Clark, Lawrence Ezrow and Garrett Glasgow. 2004. “Un- derstanding Change and Stability in Party Ideologies: Do Parties Respond to Public Opinion or Past Election Results?” British Journal of Political Science 34(4):.589—610. " Adams, James and Samuel Merrill III. 2009. “Policy-Seeking Parties in a Parlia- mentary Democracy with Proportional Representation: A Valence—Uncertainty Model." British Journal of PoliticalScience forthcoming. Aitkin, Don and Donald Stokes. 1977. Stability and Change in Australian Politics. Canberra: Australian National University Press. Alesina, Alberto, Arnaud Devleeschauwer, William Easterly, Sergio Kurlat and Romain T. W’acziarg. 2003 “Fractionalization.” Journal of Economic Growth 8(2):155~‘J—l. Alvarez, R. Michael and Jonathan Nagler. 1995. “Economics, Issues and the Perot. Candidacy: Voter Choice in the 1992 Presidential Election.” American. Joumal of Political Science 39(3):714— 44. 184 Alvarez, R. Michael and Jonathan Nagler. 1998. “When Politics and Models Col- lide: Estimating Models of Multiparty Elections.” American Journal of Political Science 42(1):55-96. Alvarez, R. Michael and Jonathan Nagler. 2004. “Party System Compactness: Mea- surement and Consequences.” Political Analysis 12(1):46~62. Amorim Neto, Octavio and Gary W. Cox. 1997. “Electoral Institutions, Cleavage Structures, and the Number of Parties.” American Journal of Political Science 41(1):l49r-74. Austen-Smith, David and Jeffrey Banks. 1988. “Elections. Coalitions, and Legisla- tive Outcomes.” American Political Science Review 82(2):405~22. Bargsted, Matias A. and Orit Kedar. 2008. “Voting for Coalitions: Strategic Voting under Proportional Representation.” Presented at the Annual Meetings of the Midwest Political Science Association, Chicago, Illinois. Bartels, Larry M. 2000. “Partisanship and Voting Behavior.” American Journal of Political Science 44(1):35—50. Baum. Christopher, Mark Schaffer and Steven Stillman. 2003. “Instrlunental Vari- ables and GMM: Estimation and Testing.” Stata Journal 3(1):1—-31. Bawn, Kathleen. 1993. “The Logic of Institutional Preferences: German Elec- toral Law as a Social Choice Outcome.” American Journal of Political Science 37(4):965-89. Bean. Clive. 1994. “Issues in the 1993 Election.” Australian Journal of Political Science 29(2):134-57. Belanger, Eric. and Bonnie M. Meguid. 2008. “Issue Salience, Issue Ownership, and Issue—Based Vote Choice.” Electoral Studies 27(3):477~91. Bennett, Joseph F. and William L. Hays. 1960. “Multidimensional Unfolding: Deter— mining the Dimensionality of Ranked Preference Data.” Psychometrika 25(1):27—— 4:5. Benoit, Kenneth. 2004. “Models of Electoral System Change.” Electoral Studies 23(3):363~89. Benoit, Kenneth. 2007. “Electoral Laws as Political Consequences: Explaining the Origins and Change of Electoral Institutions.” Annual Review of Political Science 10:363-90. Benoit, Kenneth and Jacqueline Hayden. 2004. “Institutional Change and Per— sistence: The Evolution of Poland’s Electoral System, 1989 - 2001.” Journal of Politics 66(2):396-—427. Blais. André and Louis Massicotte. 1997. “Electoral Formulas: A Macroscopic Per- spective.” European Journal of Political Research 32(1):107-—29. Blais. André and Marc Andre Bodet. 2006. “Does Proportional Representation Foster Closer Congruence Between Citizens and Policymakers?” Comparative Political Studies 39(10):1243—~62. Blais. Andre, Richard Nadeau, Elisabeth Gidengil and Neil Nevitte. 2001. "The Formation of Party Preferences: Testing the Proximity and Directional Models.” European Journal of Political Research 40(1):81~-91. Blount, Simon. 1998. “Postmaterialism and the Vote for the Senate in Australia.” Australian Journal of Political Science 33(3):441~49. Boatriglit, Robert Guy. 2008. “Who are the Spatial Voting Violators?” Electoral Studies 27(1):116-25. Bobbio, Norberto. 1996. Left 55 Right: The. Significance of a Political Distinction. Cambridge: Polity Press. Bowler, Shaun, Todd Donovan and Jeffrey A. Karp. 2008. “Incentives for Strate- gic Voting in a PR System.” Presented at the Annual Meetings of the l\’li(lwest Political Science Association, Chicago, Illinois. Brambor, Thomas, William Roberts Clark and Matt Golder. 2006. “Understanding Interaction Models: Improving Empirical Analyses.” Political Analysis 14(1):63—~ 8‘2. Braumoellcr, Bear F. 2004. “Hypothesis Testing and N‘Iultiplicative Interaction Terms.” International Organization 58(4):807—-20. Budge, Ian, David Robertson and Derek Hearl. 1987. Ideology, Strategy, and Party Change: Spatial Analyses of Post- War Election Programmes in 19 Democracies. Cambridge: Cambridge University Press. Budge, Ian and Dennis Farlie. 1978. “The Potentiality of Dimensional Analyses for Explaining Voting and Party Competition.” European Journal of Political Re- .S'ea.7‘(.'h 6(2):?03-32. Budge, Ian and Dennis Farlie. 1983. Party Competition: Selective Emphasis or Direct Confrontation? An Alternative View with Data. In Western European Party Systems: Continuity and Change, ed. H. Daalder and P. Mair. London: Sage Publications pp. 267— 305. Budge, Ian, Hans-Dieter Klingemann, Andrea Volkens, Judith Bara and Eric Tanen- baum. 2001. Mapping Policy Preferences: Estimates for Parties, Electors, and Governments, 1945-1998. Oxford: Oxford University Press. Butler, David and Donald Stokes. 1969. Political Change in Britain: Forces Shaping Electoral Choice. New York: St. Martin’s Press. 186 Cameron, Lisa and Mark Crosby. 2000. “It’s the Economy Stupid: h’lacroeconomics and Federal Elections in Australia.” Economic Record 76(235):354—64. Campbell, Angus, Philip E. Converse, Warren E. Miller and Donald E. Stokes. 1960. The American Voter. New York: John Wiley. Cantillon, Estelle. 2001. “Electoral Rules and the Emergence of New Issue Dimen- sions.” Discussion paper, Cowles Foundation 1291. Castles. Francis Geoffrey and Peter Mair. 1984. “Left—Right. Political Scales: Some ‘Expert’ Judgements.” European Journal of Political Research 12(1):73-~88. Carkoglu, Ali C. and Melvin J. Hinich. 2006. “A Spatial Analysis of Turkish Party Preferences.” Electoral Studies 25(2):369~92. Charnock, David. 1997. “Class and Voting in the 1996 Australian Federal Election.” Electoral Studies 16(3):281—300. Cliarnock, David and Peter Ellis. 2004. “Postmaterialism and Postmodernization in Australian Electoral Politics.” Electoral Studies 23(1):45-72. Chhibber, Pradecp and Ken Kollman. 1998. “Party Aggregation and the Number of Parties in India and the United States.” American Political Science Review 92(2):329—-42. Colomer, Josep M. 2005. “It’s Parties That Choose Electoral Systems (or, Duverger’s Laws Upside Down).” Political Studies 53(1):1—21. Converse, Philip E. 1964. The Nature of Belief Systems in Mass Publics. In Ideology and Discontent, ed. D. E. Apter. New York: Bree Press. Coombs, Clyde H. 1950. “Psychological Scaling Without a Unit of Measurement.” Psychological Review 57(3):145—58. Coombs, Clyde H. 1964. A Theory of Data. New York: John Wiley and Sons. Cox, Gary W. 1990. “Centripetal and Centrifugal Incentives in Electoral Systems.” American Journal of Political Science 34(4):903~-35. Cox, Gary W. 1997. Making Votes Count: Strategic Coordination in the World ’s Electoral Systems. Cambridge: Cambridge University Press. Cusack. Thomas, Torben Iversen and David Soskice. 2007. “Economic Interests and the Origins of Electoral Institutions.” American Political Science Review 101(3):373 91. Dalton, Russel J. 2000. “Citizen Attitudes and Political Behavior.” Comparative Political Studies 33(6/7):912-—40. 187 Davis, Darren W., Kathleen M. Dowley and Brian D. Silver. 1999. “Postmaterialism in World Societies: Is It Really a Value Dimension?” American Journal of Political Science 43(3):935—62. Dodd, Lawrence C. 1976. Coalitions in Parliamentary Government. Princeton: Princeton University Press. Dow, Jay K. 1998. “A Spatial Analysis of Candidate Competition in Dual Member Districts: The 1989 Chilean Senatorial Elections.” Public Choice 97(3):451-—74. Dow, Jay K. 2001. “A Comparative Spatial Analysis of Majoritarian and Proper- tional Elections.” Electoral Studies 20(1):109_25. Downs, Anthony. 1957. An Economic Theory of Democracy. New York: Harper Collins. Duverger, Maurice. 1954. Political Parties, Their Organization and Activity in the Modern State. London: Methuen. Echegaray. F abian. 2005. Economic Crises and Electoral Responses in Latin Amer- ica. Lanham: University Press of America. Erikson, Robert S., Michael B. MacKuen and James‘A. Stimson. 2002. The Macro Polity. Cambridge: Cambridge University Press. Ezrow, Lawrence. 2007. “The Variance Matters: How Party Systems Represent the Preferences of Voters.” Journal of Politics 69(1):182~92. Ezrow, Lawrence. 2008. “Parties Policy Programmes and the Dog that Didn’t Bark: No Evidence that Proportional Systems Promote Extreme Party Positioning.” British Journal of Political Science 38(3):47€%97. Fiorina, Morris P. 1981. Retrospective Voting in American National Elections. New Haven: Yale University Press. Gabel, Matthew J. and John D. Huber. 2000. “Putting Parties in Their Place: Infer— ring Party Left-Right Ideological Positions from Party Manifestos Data.” Ameri- can Journal of Political Science 44(1):94-103. Gallagher, Michael. 1991. “Proportionality, DiSproportionality, and Electoral Sys- tems.” Electoral Studies 10(1):33-—51. Gallagher, Michael. 1998. “The Political Impact of Electoral System Change in Japan and New Zealand, 1996.” Party Politics 4(2):203—28. Ganghof, Steffen and Thomas Brauninger. 2006. “Government Status and Legisla- tive Behavior: Partisan Veto Players in Australia, Denmark, Finland and Ger- many.” Party Politics 12(4):521~39. 188 Geys, Benny. 2006. “District Magnitude, Social Heterogeneity and Local Party System Fragmentation.” Party Politics 12(2):281~97. Gow, David. 1990. Economic Voting and Postmaterialist Values. In The Greening of Australian Politics, ed. C. Bean, I. McAllister and J. Warhurst. h/Ielbourne: Longman Cheshire. Green, Donald, Bradley Palmquist and Eric Schickler. 2002. Partisan Hearts and Minds. New Haven: Yale University Press. Green, Donald P. and Ian Shapiro. 1994. Pathologies of Rational Choice Theory: A Critique of Applications in Political Science. New Haven: Yale University Press. Green, Jane and Sara B. Hobolt. 2008. “Owning the Issue Agenda: Party Strategies and Vote Choices in British Elections.” Electoral Studies 27(3):460—76. Gujarati, Damodar N. 2003. Basic Econometrics. Fourth ed. New York: McGraw— Hill. Hinich, Melvin J. and Michael C. Munger. 1998. “Empirical Studies in Comparative Politics.” Public Choice 97(3):219-27. Hinich, Melvin J ., Valeri Khmelko and Peter C. Ordeshook. 1999. “Ukraine’s 1998 Parliamentary Elections: A Spatial Analysis.” Post-Soviet A fiairs 15(2):149~~85. Hoffman, Ross J. and Paul Levack. 1949. Burke’s Politics. New York: Alfred A. Knopf. Hoskin, Gary and Gerald Swanson. 1973. “Inter-Party Competition in Colombia: A Return to La Violencia'?” American Journal of Political Science 17(2):316~~50. Hoskin, Gary and Gerald Swanson. 1974. “Political Party Leadership in Colombia: A Spatial Analysis.” Comparative Politics 6(3):395~423. Hotelling, Harold. 1929. “Stability in Competition.” The Economic Journal 39(153):41—57. Huber, John D. 1989. “Values and Partisanship in Left-Right Orientations: Mea— suring Ideology.” European Journal of Political Research 17(5):599~621. Huber, John D. and Jr. Powell, G. Bingham. 1994. “Congruence Between Cit.- izens and Policymakers in Two Visions of Liberal Democracy.” World Politics 46(3):291—326. Huber, John and Ronald Inglehart. 1995. “Expert Interpretations of Party Space and Party Locations in 42 Societies.” Party Politics 1(1):73-111. Inglehart, Ronald. 1977. The Silent Revolution : Changing Values and Political Styles Among Western Publics. Princeton: Princeton University Press. 189 Inglel'iart, Ronald. 1984. The Changing Structure of Political Cleavages in Western Society. In Electoral Change in Advanced Societies: Realignment 0r Dealignment, ed. R. Dalton, S. C. Flanagan and P. A. Beck. Princeton: Princeton University Press pp. 25 69. J ackman, Simon. 1998. “Pauline Hanson, the Mainstream, and Political Elites: The Place of Race in Australian Political Ideology.” Australian Journal of Political Science 33(2):167—86. Jackman, Simon. 2001. Compulsory Voting. In International Encyclopedia of the So- cial and Behavioral Sciences, ed. N. J. Smelser and P. B. Baltes. Oxford: Elesvier. Jackman, Simon. 2003. Political Parties and Electoral Behavior. In The Cambridge Handbook of Social Sciences in Australia, ed. I. McAllister, S. Dowwrick and R. Hassan. Cambridge: Cambridge University Press. J acoby, William G. 1982. “Unfolding the Party Identification Scale: Improving the l\leasurement of an Important Concept.” Political Methodology 8(1):33~59. Jacoby. William G. 1991. Data Theory and Dimensional Analysis. Newbury Park: Sage. Jacoby, William G. 1999. “Levels of Measurement and Political Research: An Op- timistic View.” American Journal of Political Science 43(1):271~301. Jacoby, William G. 2000. “Loess: A Nonparametric, Graphical Tool for Depicting Relationships Between Variables.” Electoral Studies 19(4):577—613. Jacoby. William G. and Saundra K. Schneider. 2009. “A New Measure of Policy Spending Priorities in the American States.” Political Analysis forthcoming. sziensch, Dean. 1995. Election! How and Why Australia Votes. St. Leonards: Allen and Unwin. Karp, Jeffrey A. and Susan A. Banducci. 2002. “Issues and Party Competition Under Alternative Electoral Systems.” Party Politics 8(1):123~41. Kedar, Orit. 2005. “How Diffusion of Power in Parliaments Affects Voter Choice.” Political Analysis 13(4):410—29. Kennedy, Peter. 1998. A Guide to Eritonornetrics. Feurth ed. Cambridge: The MIT Press. Key .lr., V.O. 1966. The Responsible Electorate. Cambridge: Harvard University Press. Klingemann, Hans-Dieter. 1995. Party Positions and Voter Orientations. In Citizens and the State, ed. H.-D. Klingemann and D. Fuch. Oxford: Oxford University Press. 190 Kok Kheng. Yeoh. 2001. “Towards an Index of Ethnic Fractionalization.” FEA W orlting Paper No. 2001-3 . Kollman, Ken, John H. Miller and Scott E. Page. 1992. “Adaptive Parties in Spatial Elections.” American Political Science Review 86(4):929—37. Kramer, Jorgcn and Jorgen Rattinger. 1997. “The Proximity and the Directional Theories of Issue Voting: Comparative Results for the U.S. and Germany.” Euro- pean Journal of Political Research 32(1):1~29. Laakso, Markuu and Rein Taagepera. 1979. “Effective Number of Parties: A Mea- sure with Application to Western Europe.” Comparative Political Studies 12(1):3 — 27. Lachat, Romain. 2008. “The Impact of Party Polarization on Ideological Voting.” Electoral Studies 27(4):687-98. Lau, Richard R., David J. Andersen and David P. Redlawsk. 2008a. “An Exploration of Correct Voting in Recent U.S. Presidential Elections.” American Journal of Political Science 52(2):395—411. Lau, Richard R. and David P. Redlawsk. 1997. “Voting Correctly.” American Politi- cal Science Review 91(3):585—98. 00030554 American Political Science Association. Lau. Richard R. and David P. Redlawsk. 2008. “Older but. Wiser? Effects of Age on Political Cognition.” Journal of Politics 70(1):168--85. Lau, Richard R., Parina Patel, Dalia F. Fahmy and Robert R. Kaufman. 2008b. “Correct Voting Across 32 Democracies (and 51 Elections).” Presented at the Annual Meetings of the Midwest Political Science Association, Chicago, Illinois. Laver, Michael. 1998. “Models of Government Formation.” Annual Review of Polit- ical Science 1:1-25. Laver. Michael. 2005. “Policy and the Dynamics of Political Competition.” American Political Science Review 99(2):263—81. Laver, Michael, Kenneth Benoit and John Garry. 2003. “Extracting Policy Positions from Political Texts Using Words as Data.” American Political Science Review 97(2):311~31. Laver, Michael and W. Ben Hunt. 1992. Policy and Party Competition. New York: Routledge. Leigh, Andrew. 2005. “Economic Voting and Electoral Behavior: How Do Individ- ual. Local, and National Factors Affect Partisan Choice?” Economics 8i Politics 17(2):265w96. Lewis—Beck, Michael S. and Mary Stegmaier. 2000. “Economic Determinants of I‘llectoral Outcomes.” Annual Review of Political Science 3:183 219. 191 Lijphart, Arend. 1984. Democracies: Patterns of Majoritarian and Consensus Gov- ernment in Twenty-One Countries. New Haven: Yale University Press. Lijphart, Arend. 1999. Patterns of Democracy: Government Forms and Performance in Thirty-Sis: Countries. New Haven: Yale University Press. Lin. 'l‘se-Min. Yun—Han Chu and Melvin J. Hinich. 1996. “Conflict Displacement and Regime Transition in Taiwan: A Spatial Analysis.” World Politics 48(4):453~81. Lipset, Seymour M. and Stein Rokkan. 1967. Party Systems and Voter Alignments. New York: Free Press. Listhaug, Ola, George Rabinowitz and Stuart Elaine Macdonald. 1990. “A Compar- ative Spatial Analysis of European Party Systems.” Scandinavian Political Studies 13(3):227-54. Long. J. Scott and Jeremy Freese. 2006. Regression Models for Categorical Dependent variables Using Stata. Second ed. College Station: Stata Press. l\lacdonald, Stuart Elaine, George Rabinowitz and Ola Listhaug. 1998. “On At— tempting to Rehabilitate the Proximity Model: Sometimes the Patient J ust Can’t Be Helped.” Journal of Politics 60(3):653—90. Macdonald, Stuart Elaine, George Rabinowitz and Ola Listhaug. 2007. “Simulating Models of Issue Voting.” Political Analysis 15(4):406—27. :\r‘IacKuen, Michael B., Robert S. Erikson and James A. Stimson. 1992. “Peasants or Bankers? The American Electorate and the U.S. Economy.” American Political Science Review 86(3):597—611. ;\flansfield, Harvey C. 1971. “Hobbes and the. Science of Indirect Government." American Political Science Review 65(1):97—110. Marks, Gary. 2007. “Introduction: Triangulation and the Square-Root Law.” Elec- toral Studies 26(1):1—10. hrlarks, Gary N. 1993. “Partisanship and the Vote in Australia: Changes over Time 1907—1990.” Political Behavior 15(2):137~66. McAllister, Ian and Clive Bean. 2000. “The Electoral Politics of Economic Reform in Australia: The 1998 Election.” Australian Journal of Political Science 35(3):383— 99. l\rchllister, Ian and Donley T. Studlar. 1995. “New Politics and Partisan Alignment: Values, Ideology, and Elites in Australia.” Party Politics 1(2):197~220. 1\~IcDonald, Michael D., Silvia M. Mendes and Ian Budge. 2004. “\Nhat Are Elec~ tions For? Conferring the Median Mandate.” British Journal of Political Science 34(1):1-26. 192 McDonald, Michael and Silvia Mendes. 2001. The Policy Space of Party Manifestos. In Estimating the Policy Positions of Political Actors, ed. M. Laver. London: Routledge pp. 90—114. McIver, John P. and Edward G. Carmines. 1981. Unidimensional Scaling. Beverly Hills: Sage Publications. Meguid, Bonnie M. 2005. “Competition between Unequals: The Role of Main- stream Party Strategy and Niche Party Success.” American Political Science Re- view 99(3):347~60. Merrill, Samuel and Bernard Grofman. 1999. A Unified Theory of Voting: Direc- tional and Proximity Spatial Models. Cambridge: Cambridge University Press. Miller, \Narren E. and Donald E. Stokes. 1963. “Constituency Influence in Congress.” American Political Science Review 57(1):45-—56. Miller, Warren E. and J. Merrill Shanks. 2001. Multiple—Stage Explanation of Polit- ical Preferences. In Controversies in Voting Behavior, ed. R. G. Niemi and H. F. Weisberg. 4th ed. Washington, DC: CQ Press pp. 221—39. l\*lolinar, Juan. 1991. “Counting the Number of Parties: An Alternative Index.” American Political Science Review 85(4):1383~91. i\lonroe, Burt, Michael P. Colaresi and Kevin M. Quinn. 2008. “Fightin’ Words: Lexical Feature Selection and Evaluation for Identifying the Content of Political Conflict.” Political Analysis 16(4):372—403. Moreno. Alejandro. 1999. Political Cleavages: Issues, Parties, and the Consolidation of Democracy. Boulder: Westview Press. Morgan Kelly, Jana. 2003. “Counting on the Past or Investing in the Future? Economic and Political Accountability in Fujimori’s Peru.” Journal of Politics 65(3):864~80. l\'ly(;>rson, Roger B. 1993. “Incentives to Cultivate Favored Minorities Under Alter- native Electoral Systems.” American Political Science Review 87(4):856~69. Nadeau, Richard and Michael S. Lewis-Beck. 2001. “National Economic Voting in U.S. Presidential Elections.” Journal of Politics 63(1):159—~81. Narud, Hanne Marthe and Henrik Oscarsson. 1999. “Mass-Elite Perceptions of the Policy Space: a Comparison between Norway and Sweden.” Presented at the ECPR Joint Sessions of Workshops, Mannheim. Niemi, Richard G. and Herbert F. Weisberg. 1974. “Single-Peakedness and Guttman Scales: Concept. and Measurement.” Public Choice 20(1):33—45. 193 Norpoth, Helmut. 1979. “The Parties Come to Order! Dimensions of Preferential Choice in the West German Electorate, 1961-1976.” American Political Science Review 73(3):724—36. Norris, Pippa. 2004. Electoral Engineering: Voting Rules and Political Behavior. Cambridge: Cambridge University Press. Nyblade, Benjamin. 2004. “The “Effective” Number of Issue Dimensions: A Measure with Application to West Europe.” Presented at the Midwest Political Science Association Annual Meeting, Chicago, Illinois. Ordeshook, Peter C. 1997. The Spatial Analysis of Elections and Committees: Four Decades of Research. In Perspectives on Public Choice, ed. D. C. Mueller. Cam- bridge: Cambridge University Press. Ordeshook, Peter C. and Olga V. Shvetsova. 1994. “Ethnic Heterogeneity, District Magnitude, and the Number of Parties.” American Journal of Political Science 38(1):100~23. Papadakis, Elim. 1990. Minor Parties, The Environment and the New Politics. In The Greening of Australian Politics, ed. C. Bean, I. McAllister and J. Warhurst. h-Ielbourne: Longman Cheshire. Persson, Torsten and Guido Tabellini. 2003. The Economic Effects of Constitutions. Cambridge: MIT Press. Petrocik, John R., William L. Benoit and Glenn J. Hansen. 2003. “Issue Own- ership and Presidential Campaigning, 1952-2000.” Political Science Quarterly 118(4):599626. Pierce, Roy. 1997. “The Directional Theory of Issue Voting: III: Directional versus Proximity Models: Verisimilitude as the Criterion.” Journal of Theoretical Politics 9(1):61~74. Pitkin, Hanna Fenichel. 1967. The Concept of Representation. Berkeley: University of California Press. Poole, Keith and Howard Rosenthal. 1997. Congress: A Political-Ecmiomic History of Roll Call Voting. Oxford: Oxford University Press. Poole, Keith T. 1984. “Least Squares lV-Ietric, Unidimensional Unfolding.” Psy- cho'rnetrika 49(3):311—23. Pmner, Daniel N. 2004. “The Political Salience of Cultural Difference: Why Chewas and Tumbukas Are Allies in Zambia and Adversaries in Malawi.” American P0- lilical Science Review 98(4):529—45. Powell, G. Bingham. 2000. Elections as Instruments of Democracy : Majoritarian and Proportional Visions. New Haven: Yale University Press. 194 Powell, G. Bingham and Georg S. Vanberg. 2000. “Election Laws, Disproportionality and Median Correspondence: Implications for Two Visions of Democracy.” British Journal of Political Science 30(3):383~411. Powell, G. Bingham Jr. 2004. “Political Representation in Comparative Politics.” Annual Review of Political Science 7:273~96. Powell Jr., G. Bingham. 2009. “The Ideological Congruence Controversy: The Im- pact of Alternative Measures, Data, and Time Periods on the Effects of Election Rules.” Comparative Political Studies forthcoming. Przcworski, Adam and John Sprague. 1986. Paper Stones: A History of Electoral Socialism. Chicago: University of Chicago Press. R abe-Hesketh, Sophia and Anders Skrondal. 2005. Multilevel and Longitudinal M od- eling Using Stata. College Station: Stata Press. Rabinowitz, George and Stuart Elaine Macdonald. 1989. “A Directional Theory of Issue Voting.” The American Political Science Review 83(1):93-121. Rabinowitz, George, Stuart Elaine Macdonald and Ola Listhaug. 1991. “New Players in an Old Game: Party Strategy in Multiparty Systems.” Comparative Political Studies 24(2):147~85. Ray, Leonard and Hanne Marthe Narud. 2000. “Mapping the Norwegian Political Space: Some Findings from an Expert Survey.” Party Politics 6(2):225—39. Renfrow, Patty. 2003. Gender Politics. In The Cambridge Handbook of Social Sci- ences in Australia, ed. 1. McAllister, S. Dowwrick and R. Hassan. Cambridge: Cambridge University Press. Richman, Jesse Travis. 2005. “Why the Issue Dimensionality of Political Systems Varies, and What. Difference it Makes: Policy Space. Ideological Space and Insti- tutions.” Ph.D. Dissertation, Carnegie Mellon University. Rikcr, William. 1962. The Theory of Political Coalitions. New Haven: Yale Univer- sity Press. Riker, \Nilliam H. 1982. Liberalism Against Populism: A Confrontation Between the Theory of Democracy and the Theory of Social Choice. Prospect Heights: \‘VflVOlalld Press, Inc. Roberts. Kenneth M. and Moisés Arce. 1998. “Neoliberalism and Lower-Class Voting Behavior in Peru.” Comparative Political Studies 31(2):217 <46. Salthouse, Timothy A. 2004. “What and When of Cognitive Aging.” Current Di- rcctions in Psychological Science 13(4):140~44. Sartori. Giovanni. 1976. Parties and Party Systems. New York: Cambridge Univer- sity Press. 195 Schmidt, G. D. 2003. “The 2001 Presidential and Congressional Elections in Peru.” Electoral Studies 22(2):344-51. Schmidt, Gregory D. 1996. “Fujimori’s 1990 Upset Victory in Peru: Electoral Rules, Contingencies, and Adaptive Strategies.” Comparative Politics 28(3):32l~54. Schmidt, Gregory D. 2002. “The Presidential Election in Peru, April 2000.” Electoral Studies 21(2):346-~57. Schmidt, Gregory D. 2007. “Back to the Future? The 2006 Peruvian General Election.” Electoral Studies 26(4):813—19. Schofield, Norman, Andrew Martin, Kevin. Quinn and David Nixon. 1998. “Mul— tiparty Electoral Competition in the Netherlands and Germany: A l\-*'Iodel Based on Multinomial Probit.” Public Choice 97(3):257~93. Sclmfield, Norman and Itai Sened. 2005. “Modeling the Interaction of Parties, Ac- tivists and Voters: Why Is the Political Centre so Empty?” EurOpean Journal of Political Research 44(3):355—90. Singer. Matthew McMinn. 2007. “The Electoral Politics of Vulnerability and the Incentives to Cast an Economic Vote.” Ph.D. Dissertation, Duke University. Stimson, James A., Michael B. Mackuen and Robert S. Erikson. 1995. “Dynamic Representation.” American Political Science Review 89(3):543—65. Stoll, Heather. 2005. “Social Cleavages, Political Institutions and Party Systems: Putting Preferences Back into the Fundamental Equation of Politics.” Ph.D. Dis- sertation, Stanford University. Stoll, Heather. 2008. “Social Cleavages and the Number of Parties: How the Mea- sures You Choose Affect the Answers You Get.” Comparative Political Studies 41(11):1439w65. Stoll. Heather. 2009. “Dimensionality and the Number of Parties in Legislative Elections.” Party Politics forthcoming. Studlar, Donley T., Ian McAllister and Bernadette C. Hayes. 1998. “Explaining the Gender Gap in Voting: A Cross-National Analysis.” Social Science Quarterly 79(-4):779--98. Ti—i.agcpera, Rein. 1999. “The Number of Parties as a Function of Heterogeneity and Electoral System.” Comparative Political Studies 52(5):531~~48. Taagepera, Rein and Bernard Grofman. 1985. “Rethinking Duverger’s Law: Pre- dicting the Effective Number of Parties in Plurality and PR Systems - Parties Minus Issues Equals One.” European. Journal of Political Research 13(4):.‘341 53. Tavits, l\-’Ia.rgit. 2007. “Principle vs. Pragmatism.” American Journal of Political Science 51(1):151 ‘65. 196 Todosijevic, Bo jan. 2005. “Issues and Party Preferences in Hungary: A Comparison of Directior‘ial and Proximity Models.” Party Politics 11(1):109~~26. Tomz. Michael and Robert P. Van Houweling. 2008. “Candidate Positioning and Voter Choice.” American Political Science Review 102(3):303~18. Tsebelis, George. 2002. Veto Players: How Political Institutions Work. Princeton: Princeton University Press. \Varwick, Paul V. 2002. “Towards a Conunon Dimensionality in West European Policy Spaces.” Party Politics 8(1):101~-22. \\"arwick, Paul V. 2005. “Party Positions and Constraints in West European Policy Spaces: Estimates from a New Expert Survey.” Presented at the Joint Sessions of \\-’()1:ksliops, European Consortium for Political Research, Granada, Spain. \\'eakliern, David L. and Mark Western. 1999. “Class Voting, Social Change, and the Left in Australia, 1943-96.” British Journal of Sociology 50(4):609~30. \\'cisherg. Herbert F. 1972. “Scaling Models for Legislative Roll-Call Analysis.” American Political Science Review 66(4):1306“15. Wessels. Bernhard and Hermann Schmitt. 2008. “Meaningful Choices, Political Supply. and Institutional Effectiveness.” Electoral Studies 27(1):]9—30. \Vestern, Mark and Bruce Tranter. 2001. “Postmaterialist and Economic Voting in Australia, 1990-1998.” Australian Journal of Political Science 36(3):439 ’58. Westholm, Anders. 1997. “Distance versus Direction: The Illusory Defeat of the Proximity Theory of Electoral Choice.” American Political Science Re'zncw 91(4):865 83. \v\-'evlan(:l. Kurt. 2000. “A Paradox of Success? Determinants of Political Support for President Fujimori.” International Studies Quarterly 44(3):481H502. \\'olfers. Justin and Andrew Leigh. 2002. “Three Tools for Forecasting Federal Elections: Lessons from 2001.” Australian Journal of Political Science 37(2):223 - 10. Wood, David M. and William G. Jacoby. 1984. “Intraparty Cleavage in the British House of Commons: Evidence from the 1974-1979 Parliament.” American Journal of Political Science 28(1):203— 23. Wu, Tsimg-h’lin. 2004. “A Trade History of Taiwan.” Unpublished hilanuscript. Young. Forrest W. 1981. “The Quantitative Analysis of Qualitative Data.” Psy— chomctrika 46(4)::357- 88. 197 ”imityfilliitiliiriii‘t 083