UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Who leads? Who follows? Political representation and opinion formation Nuesser, Andrea 2017

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata

Download

Media
24-ubc_2017_september_nuesser_andrea.pdf [ 3.97MB ]
Metadata
JSON: 24-1.0348395.json
JSON-LD: 24-1.0348395-ld.json
RDF/XML (Pretty): 24-1.0348395-rdf.xml
RDF/JSON: 24-1.0348395-rdf.json
Turtle: 24-1.0348395-turtle.txt
N-Triples: 24-1.0348395-rdf-ntriples.txt
Original Record: 24-1.0348395-source.json
Full Text
24-1.0348395-fulltext.txt
Citation
24-1.0348395.ris

Full Text

	  WHO	  LEADS?	  WHO	  FOLLOWS?	  	  POLITICAL	  REPRESENTATION	  AND	  OPINION	  FORMATION	  	  by	  Andrea	  Nuesser	  	  B.A.,	  Heinrich	  Heine	  Universitaet	  Duesseldorf,	  2008	  M.A.,	  Simon	  Fraser	  University,	  2010	  	  A	  THESIS	  SUBMITTED	  IN	  PARTIAL	  FULFILLMENT	  OF	  THE	  REQUIREMENTS	  FOR	  THE	  DEGREE	  OF	  DOCTOR	  OF	  PHILOSOPHY	  in	  THE	  FACULTY	  OF	  GRADUATE	  AND	  POSTDOCTORAL	  STUDIES	  (Political	  Science)	  	  THE	  UNIVERSITY	  OF	  BRITISH	  COLUMBIA	  (Vancouver)	  	  June	  2017	  	  ©	  Andrea	  Nuesser,	  2017	     ii	  Abstract	  To understand political representation and opinion formation, we need to comprehend the dynamics between political parties and voters: Who leads? Who follows? The comparative representation literature assumes that voters lead and parties follow. Representation is understood as a principal-agent relationship in which citizens elect parties to act on their behalf. Most studies assume that voters have fixed issue opinions and regularly engage in policy voting. Conversely, the increasingly dominant view within the American public opinion literature is that parties lead and voters follow. Focussing on cognitive processes, these works suggest that the correspondence between policy preferences and party choice is not the product of policy-oriented evaluation, but of other psychological forces—mainly persuasion and projection—and conditional on partisanship. As almost all the evidence comes from the US, we know little about the impact of parties in multiparty systems, where voters are naturally pressed to think of governing coalitions. In the US, both processes have become more prevalent during the current era of polarization. To the extent that polarization animates the last twenty years of American scholarship, what is the story in Europe? A handful of single-country studies claim the opposite trend: depolarization. What is missing is systematic evidence from multiple countries and longer periods.  This dissertation bridges the gap between European and American scholarship and makes important contributions to the literature. First, it fills a Europe-wide gap on polarization and depolarization, suggesting that both movements occur and that both are functionally linked. While depolarization is the dominant trend on the general Left-Right dimension, polarization best describes party movement on European unification and multiculturalism. Second, the dissertation demonstrates that an increase in sophistication is required to deal with aggregate    iii	  notions of leading and following. It shows that depolarization is an under-theorized concept that should not be mistaken for simply the opposite of polarization. Third, using advanced estimation techniques, this thesis provides realistic assessments of leading and following. The results suggest that leading is much less prevalent in Europe than commonly assumed. Instead, there is solid evidence that European voters follow, conditional on partisanship.      iv	  Lay	  Summary	  To understand political representation and opinion formation, we need to unpack the dynamics between parties and voters: Who leads? Who follows? European scholars commonly assume that voters lead and parties follow. Conversely, the dominant view within the American literature is that parties lead and voters follow. This dissertation bridges the gap between European and American scholarship. First, it fills a Europe-wide gap on polarization and depolarization, suggesting that both movements occur: depolarization is the dominant trend on the general Left-Right dimension, while polarization best describes party movement on European unification and multiculturalism. Second, the dissertation demonstrates that depolarization is an under-theorized concept that should not be mistaken for simply the opposite of polarization. Third, this thesis provides realistic assessments of leading and following. The results suggest that leading is much less prevalent in Europe than commonly assumed. Instead, there is solid evidence that European voters follow the parties they like.    v	  Preface	  This dissertation is an original intellectual product of the author, Andrea Nuesser.   Chapter 2 is a co-authored project with Richard Johnston and Marc A. Bodet. An earlier version was presented at the 2014 meeting of the American Political Science Association and received the 2015 Frank L. Wilson APSA Paper Award. I was the lead author, and carried out the majority of the work. Specifically, I was responsible for all major areas of concept formation, data preparation and analysis, as well as manuscript composition. Marc A. Bodet was involved in the early stages of concept formation and drafted the section on L-kurtosis. Richard Johnston was the supervisory author on this project and was involved throughout the project in concept formation and manuscript composition.  All data sources used in this dissertation are publicly available.     vi	  Table	  of	  Contents	   Abstract .......................................................................................................................................... ii Lay Summary ............................................................................................................................... iv Preface .............................................................................................................................................v Table of Contents ......................................................................................................................... vi List of Tables .................................................................................................................................. x List of Figures ............................................................................................................................... xi List of Equations ........................................................................................................................ xiii Acknowledgements .................................................................................................................... xiv Chapter 1: Political representation and opinion formation .......................................................1 1.1 Important concepts ............................................................................................................. 5 1.1.1 Leading and following ................................................................................................ 5 1.1.2 Polarization and depolarization ................................................................................... 9 1.1.3 Proximity versus directional models ......................................................................... 12 1.2 Overview of the dissertation ............................................................................................ 14 1.2.1 Polarization and depolarization ................................................................................. 14 1.2.2 Parties: Leading and following ................................................................................. 16 1.2.3 Voters: Leading and following ................................................................................. 17 1.2.4 Implications for democracy ...................................................................................... 18 Chapter 2: Polarization and depolarization ..............................................................................20 2.1 Polarization: Concepts and measures ............................................................................... 22 2.1.1 Radicalization ........................................................................................................... 23    vii	  2.1.2 Alignment ................................................................................................................. 28 2.2 The Evidence: Left-Right and sub-dimensions ................................................................ 29 2.2.1 Party systems: Distributional indicators (SD and Range) ......................................... 29 2.2.2 Electorates: Distributional indicators (SD and L-kurtosis) ....................................... 32 2.2.3 Party systems: Association between dimensions ...................................................... 34 2.2.4 Electorates: Association between dimensions .......................................................... 35 2.3 Further Evidence: Multiculturalism and European Union ............................................... 36 2.3.1 Party systems: Distributional indicators (SD and range) .......................................... 36 2.3.2 Party systems: Association between dimensions ...................................................... 37 2.4 Conclusion ....................................................................................................................... 38 Chapter 3: Parties: Leading and following ...............................................................................52 3.1 The causal mechanism ..................................................................................................... 53 3.1.1 From polarization to sorting ...................................................................................... 53 3.1.2 From depolarization to unsorting? ............................................................................ 57 3.1.3 Measuring sorting and unsorting .............................................................................. 60 3.1.3.1 Observable implications for sorting ................................................................... 62 3.1.3.2 Observable implications for unsorting ............................................................... 63 3.2 Results from the UK (1992–2001) ................................................................................... 64 3.2.1 Unemployment and inflation .................................................................................... 65 3.2.2 Welfare services ........................................................................................................ 67 3.2.3 Redistribution ............................................................................................................ 68 3.2.4 Nationalization .......................................................................................................... 69 3.2.5 European unification ................................................................................................. 70    viii	  3.3 Results from Germany (1998–2013) ............................................................................... 71 3.3.1 Left-Right .................................................................................................................. 73 3.3.2 Nuclear power ........................................................................................................... 74 3.3.3 European unification ................................................................................................. 75 3.4 Discussion and conclusion ............................................................................................... 76 Chapter 4: Voters: Leading and following ................................................................................99 4.1 Dynamics of opinion and preference formation ............................................................ 100 4.1.1 Voters as leaders ..................................................................................................... 100 4.1.2 Voters as followers ................................................................................................. 101 4.1.3 Reciprocal effects .................................................................................................... 102 4.2 Methods: Challenges and estimation techniques ........................................................... 104 4.2.1 Modeling leading .................................................................................................... 108 4.2.1.1 The estimation model for the UK case ............................................................ 110 4.2.1.2 The estimation model for the German case ..................................................... 112 4.2.2 Modeling following ................................................................................................ 113 4.3 Leading and following in the UK .................................................................................. 115 4.3.1 Unemployment ........................................................................................................ 116 4.3.2 Welfare .................................................................................................................... 118 4.3.3 Redistribution .......................................................................................................... 119 4.3.4 European unification ............................................................................................... 120 4.3.5 Nationalization ........................................................................................................ 120 4.4 Leading and following in Germany ............................................................................... 121 4.4.1 Left-Right ................................................................................................................ 121    ix	  4.4.2 Nuclear power ......................................................................................................... 123 4.4.3 European unification ............................................................................................... 124 4.5 Discussion ...................................................................................................................... 124 4.6 Conclusion ..................................................................................................................... 128 Chapter 5: Conclusion ...............................................................................................................148 5.1 Contributions ................................................................................................................. 149 5.1.1 Polarization and depolarization ............................................................................... 150 5.1.2 Parties: Leading and following ............................................................................... 152 5.1.3 Voters: Leading and following ............................................................................... 153 5.2 Unanswered questions and future research .................................................................... 154 References ...................................................................................................................................156 Appendix 1: Data sources ..........................................................................................................171 British Election Study (BES) .............................................................................................. 171 Eurobarometer (EB) ............................................................................................................ 172 German Longitudinal Election Study (GLES) .................................................................... 172 Manifesto Project Dataset (CMP) ....................................................................................... 173 Appendix 2: Additional material ..............................................................................................174 Chapter 2 ............................................................................................................................. 174 Chapter 4 ............................................................................................................................. 177 	     x	  List	  of	  Tables	  	  Table 2.1 Party systems: Polarization and depolarization on Left-Right ..................................... 41 Table 2.2 Electorates: Polarization and depolarization on Left-Right .......................................... 42 Table 2.3 Party systems: Polarization on European integration and multiculturalism ................. 43 Table 4.1 Test of leading (“policy oriented evaluation”): UK, Unemployment ......................... 133 Table 4.2 Test of leading (“policy oriented evaluation”): UK, Welfare ..................................... 134 Table 4.3 Test of leading (“policy oriented evaluation”): UK, Redistribution ........................... 135 Table 4.4 Test of leading (“policy oriented evaluation”): UK, European unification ................ 136 Table 4.5 Test of leading (“policy oriented evaluation”): UK, Privatization ............................. 137 Table 4.6 Test of following (“persuasion”): UK, all issues ........................................................ 138 Table 4.7 Test of following (“projection”): UK, all issues ......................................................... 139 Table 4.8 Test of leading (“policy oriented evaluation”): Germany, Left-Right ........................ 140 Table 4.9 Test of leading (“policy oriented evaluation”): Germany, Nuclear power ................. 141 Table 4.10 Test of leading (“policy oriented evaluation”): Germany, EU ................................. 142 Table 4.11 Test of following (”persuasion”): Germany, all issues ............................................. 143 Table 4.12 Test of following (”projection”): Germany, all issues .............................................. 144 Table 4.13 Summary: Net leading and following effects in the UK .......................................... 146 Table 4.14 Summary: Net leading and following effects in Germany ....................................... 147 Table A.1 Party systems, summary of all dimensions ................................................................ 174 	     xi	  List	  of	  Figures	  	  Figure 2.1 Party systems: Depolarization on Left-Right and sub-dimensions ............................. 44 Figure 2.2 Electorates: Polarization and depolarization on Left-Right and postmaterialism ....... 45 Figure 2.3 Parties: Association between economic and social dimensions .................................. 46 Figure 2.4 Voters: Association between Left-Right and postmaterialism .................................... 47 Figure 2.5 Party systems: Polarization on European integration and multiculturalism ................ 48 Figure 2.6 Party systems: Association between Left-Right and European integration ................ 49 Figure 2.7 Party systems: Association between Left-Right and multiculturalism ........................ 50 Figure 2.8 Party systems: Association between multiculturalism and European integration ....... 51 Figure 3.1 Partisan sorting ............................................................................................................ 81 Figure 3.2 The difference between sorting and polarization ........................................................ 82 Figure 3.3 Unemployment: Comparing positions of British parties and voters ........................... 83 Figure 3.4 Unemployment: Comparing distributions (SDs) of British parties and voters ........... 84 Figure 3.5 Welfare: Comparing positions of British parties and voters ....................................... 85 Figure 3.6 Welfare: Comparing distributions (SDs) of British parties and voters ....................... 86 Figure 3.7 Redistribution: Comparing positions of British parties and voters ............................. 87 Figure 3.8 Redistribution: Comparing distributions (SDs) of British parties and voters ............. 88 Figure 3.9 Nationalization: Comparing positions of British parties and voters ........................... 89 Figure 3.10 Nationalization: Comparing distributions (SDs) of British parties and voters .......... 90 Figure 3.11 European unification: Comparing positions of British parties and voters ................ 91 Figure 3.12 European unification: Comparing distributions (SDs) of British parties and voters . 92 Figure 3.13 Left-Right: Comparing positions of German parties and voters ............................... 93    xii	  Figure 3.14 Left-Right: Comparing distributions (SDs) of German parties and voters ............... 94 Figure 3.15 Nuclear energy: Comparing positions of German parties and voters ........................ 95 Figure 3.16 Nuclear energy: Comparing distributions (SDs) of German parties and voters ........ 96 Figure 3.17 European unification: Comparing positions of German parties and voters .............. 97 Figure 3.18 European unification: Comparing distributions (SDs) of German parties and voters98 Figure 4.1 Party-voter relations from a positional perspective ................................................... 130 Figure 4.2 Leading model for the UK: Four-wave dynamic panel model (SEM) ...................... 131 Figure 4.3 Leading model for Germany: Three-wave dynamic panel model (SEM) ................. 132 Figure A.1 Party system depolarization: Polarization index ...................................................... 175 Figure A.2 Party system polarization: Polarization index .......................................................... 176 Figure A.3 Test of leading: Three-wave dynamic panel model (SEM) ...................................... 180 	  	   	     xiii	  List	  of	  Equations	  	  Equation 4.1 Policy oriented evaluation (issue voting) .............................................................. 101 Equation 4.2 Persuasion .............................................................................................................. 102 Equation 4.3 Projection ............................................................................................................... 102 Equation 4.4 Modeling leading: Policy oriented evaluation ....................................................... 108 Equation 4.5 Modeling leading: Reverse effects ........................................................................ 108 Equation 4.6 Modeling following: Persuasion ............................................................................ 113 Equation 4.7 Modeling following: Projection ............................................................................ 114     xiv	  Acknowledgements	  I was fortunate to work and study in a department full of wonderful and sharp individuals, who contributed to the completion of this dissertation in so many ways.   I am deeply indebted to the members of my supervisory committee, Richard Johnston, Fred Cutler, and Andrew Owen. Their support and guidance was invaluable for my success in the PhD program. I greatly appreciate that they pushed me to think harder about the concepts behind political representation and opinion formation and encouraged me to tackle ever more demanding questions. Their advice made this project so much better, and I cannot thank them enough for it. Richard Johnston is a treasure trove of knowledge and I will be forever thankful for his expertise and for encouraging me to go back to the origins of opinion research.  I owe a huge thank you to Mark Pickup, who stepped in at a crucial time and provided invaluable help with structural equation modeling. I would like to thank Steven Weldon for motivating me to pursue this degree and for his support throughout the years. I wish to extend my thanks to Kurt Huebner and Alan Jacobs, who generously provided office space and organized monthly workshops at the Institute for European Studies. I benefited greatly from the comments of workshop participants.  My fellow students helped me over some difficult hurdles. Edana Beauvais, David Moscrop and Daniel Westlake were excellent study and discussion partners during the Comprehensive Exams phase. My roomie, Sule Yaylaci made methods camp a highly enjoyable experience. Jen Allan, Jan Luedert, and Yana Gorokhovskaia were always there when I needed them.  Finally, I am grateful to my family for their love, support and patience.   1 Chapter 1: Political representation and opinion formation Political parties and voters are central actors in representative democracies, and their behaviours shape democratic processes. So does the relationship between them. To understand political representation and public opinion formation, we need to comprehend the dynamics between parties and voters: Who leads? Who follows? The comparative representation literature assumes that voters lead and parties follow. In democratic theory, representation is understood as a principal-agent relationship in which citizens (the “principals”) elect parties or party members (“agents”) to act on their behalf (e.g., Dahl 1956, 1971; Pitkin 1967; Urbinati and Warren 2008). Empirical studies commonly focus either on procedural or substantive representation. Work on procedural representation is concerned with the mechanisms that translate vote into seats. It takes citizens’ votes as the starting point and assumes that these reflect citizens’ policy preferences, without questioning the causal origin of these preferences (e.g., Cox 1997; Lijphart 1994; Rae 1967; Shugart and Carey 1992; Taagepera and Shugart 1989). Studies of substantive representation pay more attention to citizens’ preferences, but also treat them as predetermined. Regarding the link between citizens’ policy preferences and their vote choices, most studies assume that voters hold fixed and meaningful opinions on a range of issues. Come election time, voters choose the party that best represents their views on substantive matters. In other words, these accounts assume that voters engage in issue voting or policy oriented party evaluation (Brody and Page 1972; Downs 1957; Riker and Ordeshook 1973). Consequently, they focus on the level of ideological or issue congruence between the mass public and elected representatives to assess the quality of    2 representation (e.g., Kitschelt 2000; Miller and Stokes 1963; Miller et al. 1999; Powell 2000).1 To the extent that the level of congruence between citizens and elected representatives is imperfect, it is not because voters do not have real, predetermined preferences but because of aggregation barriers.2 This is not only the normatively appealing but also the dominant account—at least in Europe. Conversely, the increasingly dominant view within the American public opinion literature is that parties lead and voters follow. A long tradition of public opinion research has investigated the origins of citizens’ political preferences and challenges the view that voters’ opinions are unaffected by party positions, mainly on the grounds of voters’ cognitive abilities and motivations to engage in politics (e.g., Lazarsfeld et al. 1944; Berelson et al. 1954; Campbell et al. 1960; Stokes 1966). In order to make informed choices and hold parties accountable, voters need to pay attention to politics and understand where each party stands vis-à-vis their own positions. Yet, scholars have long documented low average levels of political interest and the absence of informed and stable positions within the electorate (e.g., Converse 1964; Delli Carpini and Keeter 1996; Gidengil et al. 2004). When navigating the political world, voters—especially those with low levels of political knowledge—rely on information shortcuts (e.g.,                                                  1 For a review of “Political Representation in Comparative Politics”, see Powell (2004). 2 Congruence captures the proximity between voters and elected representatives (or parties) on an issue dimension. Institutional factors that affect the level of congruence include the type of electoral system (Blais and Bodet 2006; Budge and McDonald 2007; Golder and Stramski 2010; Huber and Powell 1994; Kim 2009; McDonald and Budge 2005; McDonald, Mendes and Budge 2004; Powell 2000, 2006; Powell and Vanberg 2000), and party system structure (Powell 1989; 2010).    3 Cutler 2002; Lupia 1994; Sniderman 2000; Sniderman et al. 1986).3 So, rather than becoming informed about an issue, voters use other pieces of information more readily available to them to make their decisions. One shortcut that many less informed citizens rely on is their party identification (PID). When faced with an issue unfamiliar to them, they simply—and perhaps, blindly—adopt the views of their preferred party (e.g., Bartels 2002; Lenz 2009; 2012).  These findings suggest that the correspondence between policy preferences and party choice is not the product of policy driven considerations, but of other psychological forces—mainly persuasion and projection—and conditional on prior identification with parties. Persuasion describes a process in which voters change their own policy views to be closer to their preferred party. So, if voters take cues from parties, parties lead and voters follow. Projection is a form of motivated cognition that is driven by partisanship and, at least potentially, has consequences for voters’ behaviour. Partisans perceive the world in a distorted way and decrease the distance to a favoured party by placing it closer to their own positions (Brody and Page 1972; Page and Brody 1972; Markus and Converse 1979). This perceptual error prevents voters from weighing party options objectively. In other words, it keeps voters from leading. These processes were noted right at the beginning of survey research and have been influential over decades of social psychological research (e.g., Berelson et al. 1954; Festinger 1957; Bartels 2002). Most notably for political science, they played a central role in the                                                  3 These findings have ignited lively debates about the consequences of low information publics, especially for the quality of representation. Some scholars emphasize the existence of systematic knowledge gaps within the electorate, which ultimately lead to unequal representation for certain demographics (Bartels 2008; Gidengil et al. 2004). Others defend the so-called low-information rationality and argue that citizens can use heuristics or information shortcuts to make “as-if” informed decisions (Lupia 1994). The aggregationist account also takes the position that not all voters need to be informed about politics for a society to produce good, representative outcomes (Page and Shapiro 1992).     4 formulation of the concept of party identification (PID) (Campbell et al. 1960). To what extent persuasion and projection define party-voter relations in other contexts is unclear, as almost all the evidence comes from the US. The few studies that examine these dynamics in other contexts focus on party systems with a small number of parties or concentrate exclusively on the two biggest ones (e.g., Evans and Andersen 2004; Lenz 2012; Milazzo, Adams and Green 2012). As a result, we know little about the dynamic and directive impact of parties in multiparty systems, where voters are naturally pressed to think of governing coalitions.  What is more, both processes may have become more important over time—at least in the US—where parties are leading their followers in opposite directions during the current era of polarization (e.g., Hetherington 2009; Levendusky 2009). American research suggests that party system polarization alters the relationship between parties and voters and shapes how voters form opinions and make decisions (Druckman et al. 2013).   To the extent that polarization animates the last twenty years of American scholarship, what is the story in Europe? A handful of studies claim the opposite trend: depolarization. However, the evidence comes from single-country studies—the UK (Adams, Green and Milazzo 2012; Milazzo, Adams and Green 2012) and the Netherlands (Adams, DeVries and Leiter 2012). What is missing is systematic evidence from multiple countries and longer time periods. Yet, this is needed to understand variation in the balance of leading and following in Europe. This thesis fills both gaps by systematically addressing two central research questions: What characterizes the dynamic relations between parties and voters in contemporary democracies? How and why do they vary across and within political contexts? I argue that the standard European account of party-voter relations, characterized by unidirectional dynamics, is incomplete. More realistic is a reciprocal relationship between citizens and elites, featuring a mix    5 of different processes that include issue voting, persuasion, and projection. Furthermore, I propose that the structure of the party system can help explain variation in the balance of these dynamics—across and within political contexts. In addressing these research questions, this thesis draws from both American and European scholarship on political representation and public opinion. So far, the heavy lifting in terms of conceptual and measurement development has been done in the US, where decades of behavioural research have produced a sophisticated toolkit. Yet, because of the uniqueness of the American party system, not all concepts travel well. Studies that apply American frameworks elsewhere tend to oversimplify the European context by concentrating on the two largest parties, thereby ignoring the more complex nature of party competition in multiparty systems. This thesis extends the analysis to all relevant parties, which requires adjustments to existing concepts and measures. 1.1 Important concepts 1.1.1 Leading and following I adopt the language of leading and following from Lenz (2012), who investigates how voters respond to politicians’ policies and performance. I examine these dynamics from a positional perspective, focusing exclusively on the relationship between voters’ positions on issues, voters’ perceptions of party positions, and voters’ party preferences. I use the terms leading and following to describe attitudes and behaviour of political parties and voters.  If voters lead and parties follow, voters hold fixed opinions on issue, which inform their party preferences. By engaging in policy oriented party evaluation or issue voting voters provide incentives for parties to come to voters to win votes. Conversely, if parties lead and voters follow, parties steer the political course and voters move towards them. That is, voters follow if    6 they can be persuaded by a favoured party to change their own positions to bring them in line with the party’s. Likewise, voters follow if they project their own positions onto a party they like, and consequently vote for this party because they perceive it to be the closest to their own position.4  As illustrated above, the normatively appealing perspective is that parties follow to ensure that voters’ policy preferences are represented in the political process. As strategic actors, parties also have an incentive to follow, either the median voter or their supporters (Adams 2001; Strom 1990). That is, depending on their incentives, parties may follow movement in the median voter’s position or changes in the positions of their core supporters. A number of studies on the conditions of party responsiveness find that both party and voter characteristics play a role in patterns of party following. A limited European literature claims that parties of the center and right adjust their ideologies to public opinion shifts, while parties of the left do not systematically respond to public opinion (Adams et al. 2009). Similarly, mainstream parties display consistent tendencies to react to shifts in public opinion, while niche parties, such as Communist, Green, and extreme nationalist parties, do not. In fact, niche parties are punished at the polls when they moderate their policy positions, while mainstream parties do not pay the same price (Adams et al. 2006).  As far as voter characteristics are concerned, party responsiveness varies across sub-constituencies. Parties respond mainly to opinion leaders, not the general electorate (Adams and Ezrow 2009). Furthermore, the interaction between party and voter characteristics is critical: mainstream parties tend to adjust their Left-Right positions in response to shifts in the median                                                  4 See section on dynamics of opinion and preference formation in Chapter 4 for a more detailed discussion of these processes.     7 voter position, but appear unresponsive to the policy shifts of their supporters. Niche parties, on the other hand, are highly sensitive to shifts in the position of their mean supporter, while they do not respond systematically to the median voter in the general electorate (Ezrow et al. 2011).  Research that looks beyond the Left-Right positioning of parties and voters reveals that the type of issue is also relevant. Parties are more responsive to public opinion in some policy areas than in others. Dalton (1985), for example, found that in some cases there is close correspondence between opinions (e.g. economic and security issues), but in other instances the evidence of voter-party agreement is substantially weaker (e.g. foreign policy). As mentioned earlier, there is a gap between European and American scholarship. The widespread assumption in the European literature is that voters lead, but basic identifying conditions are rarely tested. Models of issue voting take voters’ policy opinions for granted and test their “effects” on party choice (e.g., Adams et al. 2005; Alvarez and Nagler 1995; 1998; 2004). Likewise, studies of ideological or policy congruence interpret associational evidence in the normatively appealing way that voter lead. However, because they do not take into account the possibility of voters following, they (most likely) overestimate the importance of issue voting (Brody and Page 1972). American public opinion studies pay more attention to party-driven dynamics. Most studies stress the importance of partisanship—or party identification (PID)—as the key mechanism linking parties and their supporters. Partisanship is commonly conceptualized as a long-standing commitment acquired early in life that changes only through extraordinary events (Campbell et al. 1960) or as a form of social identification (Green, Palmquist and Schickler 2002). Party identification is the best predictor of vote choice; at least in the US. Party identifiers are loyal and reliable voters who rarely defect or abstain. A perceptual screen is the mechanism    8 that produces durable partisan ties: it filters political information, thereby reinforcing existing views (Campbell et al. 1960). It is also the mechanism behind projection, as a form of partisan bias. It is now well established in the American context that party identification shapes judgments about a variety of issues, such as the economy, presidential approval, candidate evaluations, core political values, policy preferences, and even perceptions of objective facts such as the unemployment rate (Bartels 2002; Green and Palmquist 1990; Goren 2005; Layman and Carsey 2002; Goren et al. 2009; Evans and Pickup 2010). In short, conventional wisdom in the US suggests that there is a lot of following—conditional on party identification. Outside of the US partisanship is more fluid and often tied to vote choice, and may not be such a long-standing psychologically-rooted commitment (Butler and Stokes 1969; Dalton and Wattenberg 2000; Clemens and Bartle 2009). Clarke et al. (2004; 2009) further claim that PID as a social identity does not apply to British voters (see, also Dalton 2008). Johnston (2006), on the other hand, suggests that these differences between the US and Europe are measurement artefacts. Regardless, considerable cross-national variation exists in the proportion of respondents who express party attachments (e.g., Huber et al. 2007). Studies that test the effect of partisanship on opinion formation outside of the US find that party cues affect European partisans in a similar way as American party identifiers (Brader and Tucker 2012).   Whether or not a voter follows her preferred party depends not only her partisanship, but also on how firm her opinions are on an issue. There is systematic variation between “easy” and “hard” issues  (Carmines and Stimson 1980: 80). Easy issues are (1) symbolic or normative, rather than technical, in nature; (2) are concerned with ends, as opposed to means; and (3) have been on the political agenda for a long time (e.g., race or abortion). Conversely, hard issues are complex, technical issues that require detailed knowledge, such as policies regarding policy    9 means (e.g., European integration). Voters are more likely to hold stable opinions on easy issues and are, thus, less likely to take cues from parties (Converse 1964; Hillygus and Shields 2008; Mendelberg 2001; Valentino, Hutchinson and White 2002). Lenz (2012: 213-14) further suggests that “people may follow their party […] only because, even on the most prominent issues of the day, they rarely come across information about the issue in their everyday lives and lack incentives to seek out more information.” In other words, they use party positions as information shortcuts, when they are unfamiliar with an issue, especially if it is a complex or technical one.5 1.1.2 Polarization and depolarization Polarization and depolarization are central concepts in Chapters 2 and 3 and directly related to the question of leading and following. Longitudinal research on American elections has documented a process of elite polarization, which has fundamentally changed the character of both political parties and the party system. Until the late 1970s, the Republican and Democratic Parties were internally heterogeneous and the distance between mean Republican and mean Democratic members of Congress was rather small. Since the 1980s this distance has increased dramatically, and both parties have become more internally coherent (see, e.g., McCarty et al. 2006; Hetherington 2009). Thus, in the American case, political parties have taken the lead and moved towards the extremes.                                                   5 Studies of European integration lend support to this idea (see, e.g., Wessels 1995; Steenbergen and Jones 2002). Because of the complex and technical nature of the issue, citizens are likely to take cues from political parties when forming their opinions about European Union policy (Ray 2003; see also, Feld and Wildgen 1976; Van der Eijk and Franklin 1996; Gabel 1998; Anderson 1998).     10 The American electorate has not polarized to the same extent, in that the average voter has not taken on more extreme positions (Abramowitz 2008). Instead, voters have sorted along partisan lines: “Liberals became Democrats and Conservatives became Republicans” (Levendusky 2009). The result is that the polarized party system leaves the political middle, where most voters are located, vacant. In today’s American political system a sizeable section of the electorate is not represented by either of the two political parties. As far as the dynamics of mass-elite linkages are concerned, party polarization was not a response to voters moving. Rather, parties took the lead and voters responded—not by polarizing, but by sorting into distinct partisan camps. In addition, elite polarization has altered the relationship between political parties and the American electorate. The results are changes in political behaviour and opinion formation, such as “the growth of party voting, the decline of split-ticket voting, and the growth in attitudinal consistency observed over time” (Levendusky 2009: 3). Today’s American parties have more influence over voters’ opinions and behaviour than their older, non-polarized versions—they lead more. At the same time, polarization has revived American democracy. Elite polarization has stimulated political participation across the electorate and, perhaps perversely, improved perceptions of government responsiveness (Hetherington 2008). Comparative longitudinal research on the collective linkages between party systems and electorates is practically non-existent, but two polarization studies have applied the American framework to single European countries—the UK (Adams, Green and Milazzo 2010) and the Netherlands (Adams, DeVries and Leiter 2012). They find the opposite trend: depolarization of party systems, and weakening ties between parties and voters. Both cases show that, despite depolarization trends, the mechanisms identified in the US also operate in the UK and the    11 Netherlands. Changes in the level of party system polarization alter the relationship between parties and voters, and partisanship plays a central role. Both studies claim that, as in the American case, these processes are elite-driven. However, closer examination reveals that parties did not lead the depolarization process, at least not uniformly.6   With respect to leading and following, depolarization is not simply the opposite of polarization. To recap, if parties lead, they move away from their supporters. In case of polarization, they take on more extreme positions than their voters. In case of depolarization, they take on more centrist positions than their supporters. However, most electorates are, more or less, normally distributed with the majority of voters located in and around the centre. Thus, depolarization is more likely to be an instance of parties’ following, moving closer to their respective median party voters. Accordingly, polarization and depolarization create different dynamics of electoral competition, with consequences for opinion and preference formation. Comparative insights into the party-voter relations at the level of party systems and electorates focus on short-term dynamics and stress the importance of distributional representation. It reflects the idea that representative democracy should provide representation for the citizenry, including minority groups and their interests (e.g., Lijphart 1999). In countries where political parties contest elections a party system that represents the preferences of the people is a prerequisite. Only if the party system offers a full menu of choices that includes both majority and minority positions can the public elect representatives that truly represent its views (Pitkin 1967).                                                   6 See Chapter 3 for a detailed theoretical discussion on the effects of polarization and depolarization on party-voter relations, as well as an empirical investigation of the British case.    12  Ezrow (2007) examines the extent to which the party system as a whole responds to changes in the diversity of voter preferences, and whether this link is mediated by the electoral system. He finds that the twelve European party systems included in the analysis “display dynamic distributional representation. That is, the degree of policy diversity in these party systems systematically changes in response to changes in the diversity of mass publics’ policy preferences” (Ezrow 2007: 182). Ezrow’s interpretation of his findings suggests a voter-driven process: parties follow voters. The evidence, however, is associational and does not (necessarily) support this conclusion.  Again, this presumption that the dynamic relationship is one where parties respond to voters is typical for the European representation literature. Most studies assume a model of democracy akin to the responsible party model, according to which parties move their positions in response to voters, not vice-versa (see, e.g., Adams 2001; Alvarez and Nagler 2004; Abney et al. 2007). Consequently, they test the association between party and voter positions, and assume that voters lead. 1.1.3 Proximity versus directional models The spatial proximity model informs much of the research on voter behaviour and is the foundational claim in issue-voting theories (e.g., Downs 1957; Enelow and Hinich 1984; Adams et al. 2005). It predicts that voters derive the most utility from the party (or candidate) closest to them on some ideological or policy continuum. Accordingly, a voter is most likely to choose the party nearest to her own position.  A prominent alterative is the directional model of party evaluation (e.g., Macdonald, Listhaug, and Rabinowitz 1991; and Rabinowitz and Macdonald 1989; Rabinowitz, Macdonald and Listhaug 2001). Proponents of this model suggest that voters do not necessarily vote for the    13 party that is closest to their own position, but for the party that takes the strongest stand on their side of the issue. “In directional theory, issues for the mass public are bipolar, and people are differentiated according to which side they favour and how strongly they favour it” (Rabinowitz et al. 1991: 149). Thus, the directional model has two components—direction and intensity. Insights about direction go back to Matthews (1979), and Rabinowitz and his colleagues add the intensity component to the story. The key assumption is that policy dimensions have a neutral “zero-point” that marks the boundary between sides. In comparison, proximity models require less by way of assumptions. Self-consciously, or otherwise, students of US polarization accept this assumption, whatever they may think about the rest of the directional story.  Studies that have directly tested the performance of both models show that proximity models outperform directional ones (e.g., Blais et al. 2001; Johnston et al. 2000). However, Kedar (2005) shows that both models are useful in explaining how citizens vote, depending on the institutional environment. Her account of “policy balancing” builds on the notion that voters are outcome or policy oriented. In multiparty systems voters need to take into account the possibility of government coalitions. Hence, they tend to support parties whose positions differ from their own views insofar as these parties pull policy in a desired direction. In other words, citizens engage in directional voting and overshoot because they anticipate compromises and a watering down of policy. Similarly, Zakharova and Warwick (2014) show that both policy distance and being on the same side as the party on a Left-Right continuum influence party evaluation.  I argue that in the polarized American party system, directional considerations also trump proximity motivations. However, the causal mechanism is different than in multiparty systems. Scholars agree that polarization in the US is elite-led. That means parties have moved away from    14 the median voter and their own supporters. As a result, policy distances between parties and voters have widened, at least on average, because voters have not polarized to the same degree. Instead, voters have responded to elite polarization by sorting, which is defined as being on the same side of an issue as one’s preferred party (Levendusky 2009). As the distance between both parties increases, the stakes for voters rise. Although both parties are further away from their supporters—thus, representing them less—it matters more which party gets into power. In other words, being on the correct side becomes increasingly important. In short, polarization changes the calculus of voting. Directional considerations can help explain why party-voter ties strengthen with party system polarization (Lupu 2015), even though policy distances widen. They also demonstrate why depolarization is not just the opposite of polarization. In a depolarized party system, directional considerations are irrelevant, as all parties represent similar positions.  1.2 Overview of the dissertation This dissertation presents a comprehensive analysis of party-voter relations, guided by the questions: Who leads? Who follows? Each of the three empirical chapters addresses a different aspect of this relationship and illuminates a different dimension of representation. In doing so, it merges insights from different research traditions.  1.2.1 Polarization and depolarization The starting point is an analysis of party systems and electorates that focuses on long-term trends of polarization and depolarization. Polarization in the US has shaped party-voter relations and altered the balance of leading and following. The European situation is not well documented. Thus, the goal of this chapter is to fill a gap in the literature by providing a comprehensive assessment of the European landscape that includes both parties and voters over an extended    15 period. It is an exercise in descriptive analysis as, at this high level of aggregation, a causal analysis of leading and following is difficult, if not impossible. Long-term trends and parallel movement can potentially hint at a (causal) link between parties and voters, but the evidence is, by design, not conclusive.   The concepts and measures used in longitudinal polarization studies have been developed in and for the American context. Some are less meaningful in multiparty systems and require innovation and adaptation. Chapter 2 begins with a discussion of suitable polarization measures, and the analysis employs primarily distributional and correlational measures that include the positions of all parties. What is more, the chapter introduces a new indicator of polarization: L-kurtosis. It provides additional information about the shape of a distribution—how “fat” the tails are—and can help explain the rise of extreme parties during a period of depolarization.   The data come fro the Comparative Manifestos Project (CMP), for party positions, and the Eurobarometer (EB) surveys, for public opinion. Longitudinal data at the country level are available from a number of sources. However, only the CMP covers party positions from multiple European democracies over a period spanning multiple decades. What is more, these data include information on numerous issue dimensions. Likewise, the Eurobarometer (EB) surveys allow for a long-term tracking of voter positions. The number of available issue dimensions is, however, more limited. The comparison of party and voter trajectories in fifteen European democracies over a 40-year period reveals a depolarization trend along the general Left-Right dimension. This is true for both party systems and electorates. In contrast, we find that party systems have polarized on two newer dimensions—European integration and multiculturalism.     16 1.2.2 Parties: Leading and following The second empirical chapter explores the dynamics of leading and following at the level of political parties and their supporters. It focuses on the effects of polarization and depolarization on party-voter relations in the UK and Germany. Again, inspired by research from the US, I examine the causal mechanism behind sorting, and formulate more general expectations for a link between party system structure and party-voter relations.  As in the previous chapter, this requires careful theorizing, as multiparty systems present different incentive structures to both parties and voters than two-party systems. I apply these insights to my two cases and analyze the incentives for parties and voters through the lens of proximity versus directional motivations and their relevance conditional on party system structure. For the empirical analysis, I trace voter and party positions along multiple issue dimensions and across multiple electoral cycles. This longitudinal approach helps me identify when and where parties lead and follow. Furthermore, I track the dispersion of voter positions to detect patterns of sorting or unsorting within the electorate. The data come from the British Election Study (BES), for the UK, and the German Longitudinal Election Study (GLES), for Germany. Both studies asked respondents to place themselves and all relevant parties on a range of issue dimensions.   I find differences in correspondence between party and voter positions across both cases, as well as within—across parties and issues. Regarding the causal mechanism through which the party system structure shapes party-voter relations, I discover parallels to the US, conditional on parties leading. For the UK, I find unsorting as a response to depolarization, only in instances where parties move away from their supporters toward more central positions. Likewise, I find    17 sorting in reaction to party-led polarization. However, my results do not confirm a general unsorting of the British electorate. In the German case, I see much less movement over time. On average, party and voter positions are stable, and correspondence is closer than in the UK.  1.2.3 Voters: Leading and following The third empirical chapter addresses the question of leading and following from the perspective of individual voters in multiparty systems. The focus is on cognitive processes behind opinion and preference formation. It explores the dynamics within the triad of individual-level cognitive-motivational states and seeks to disentangle three processes: policy-oriented evaluation (or issue voting), persuasion and projection. Moreover, this chapter tests my argument about the reciprocal nature of party-voter relations. Methodologically, this chapter is the most sophisticated of the three.  Identifying whether voters lead or follow is a tricky task, as the processes above are not directly observable. What is more, their outcomes are observationally equivalent at any given point in time (Lenz 2012). Overcoming this challenge of observational equivalence requires panel data, which include multiple observations for each unit in the dataset and enable the researcher to directly observe change over time. To allow for causal inferences, panel data require careful analysis and correct model specifications. Both the econometrics approach and the structural equation modeling (SEM) tradition are well suited for the analysis of panel data. For a proper test of leading and following I need models that can estimate reciprocal effects between variables of interest. For this purpose, SEMs have advantages over economic (OLS) models, especially when it comes to estimating non-recursive models—models with simultaneous reciprocal effects. Yet, they place less weight on controlling for unobserved heterogeneity, which is a must when one tries to identify the direction of causality and rule out    18 spuriousness (e.g., Finkel 1995; Allison 2009). The econometric tradition has developed different types of fixed effects models, which control for stable unobserved factors. All panel models in this thesis are multi-wave SEM panel models that control for unobservables, either by first-differencing the data or including latent error variables.     As in the previous chapter, I focus on two cases: the UK and Germany. Data for the former come from the British Election Study (BES), where two suitable multi-wave panel studies were conducted between 1992 and 2001. For Germany, I rely on panel data from the German Longitudinal Election Study (GLES). I pool different three-wave panel studies that cover the period between 1998 and 2013. The findings from both cases support my argument that a mix of leading and following characterizes the relationship between parties and voters. More importantly, the results for leading suggest that most studies—in particular those using cross-sectional data—overestimate the importance of issue voting, as they do not account for effects in the opposite direction. In addition, the tests of persuasion and projection suggest that both processes are operational in Europe. These findings confirm that—similar to the US—British and German voters’ opinions and perceptions are conditional on party preference.  1.2.4 Implications for democracy Chapter 5 concludes the dissertation by synthesizing the findings across levels and contexts and by discussing their meaning for the health of democratic representation. To preview, this dissertation makes three major contributions to the literature, hence advancing our understanding of representation and opinion formation. First, it fills a Europe-wide gap on polarization and depolarization and suggests that both movements occur and that both are functionally linked. While depolarization is the dominant trend on the general Left-Right dimension, polarization    19 best described party movement on European unification and multiculturalism. Second, through the in-depth analysis of two cases—the UK and Germany—the dissertation demonstrates that a massive increase in sophistication is required to deal with aggregate notions of leading and following. It shows that depolarization is an under-theorized concept that should not be mistaken for simply the opposite of polarization. Depolarization creates distinct incentives for parties and voters, especially in multiparty systems, and correlational measures are inadequate to capture its effects. Third, through the use of advanced estimation techniques, this thesis provides a realistic assessment of issue voting by isolating the effects of policy positions on party preference and controlling for other psychological forces. The results suggest that leading—in the form of issue voting—is much less prevalent in Europe than commonly assumed. Instead, the analysis presents solid evidence that European voters follow, conditional on partisanship.   Although the dissertation provides answers to several important questions, it leaves many unanswered. The most obvious one is the question of party system effects on leading and following. Causal analysis of opinion and attitude formation requires specific types of data—multi-wave panel data at the level of individual citizens. This requirement limits the scope of my analysis and pre-selects my cases, as multi-wave panel studies spanning multiple elections are not nearly as widely available as cross-sectional data. Put simply, I cannot conduct a systematic comparative analysis that allows me to test the effects of party system structure on the relationship between parties and voters.   Still, the in-depth analysis of two cases provides novel insights about how voters in multiparty systems form opinions and preferences. These, in turn, help us assess the quality of political representation. Unfortunately, the news is not all positive.     20 Chapter 2: Polarization and depolarization Polarization is a central preoccupation in the United States, where elite division and policy deadlock have become the norm (McCarty, Poole, and Rosenthal 2006; Hetherington 2001; 2008; 2009). Over the last four decades, political party elites have drifted apart, significantly altering the dynamics of electoral competition and the relationship between parties and voters. Today, divergence between the two major parties has potentially serious implications for the representation of American voters, particularly as voters seem not to have moved as far towards extremes as elites have (Fiorina et al. 2005; Baldassarri and Gelman 2008; Levendusky 2009). By abandoning the political center and offering only extreme options to voters, the American party system leaves the moderate part of the electorate unrepresented. At the same time, voters have sorted and partisanship and ideology are now tightly coupled: “sorted Democrats are liberals, and sorted Republicans are conservatives” (Levendusky 2009: 3). Furthermore, the dimensionality of opinion in the mass public has dropped, as formerly separate evaluative axes become more predictive of each other. Trends here are weaker than for sorting, but are greatest among the most politically engaged (Baldassarri and Gelman 2008).  Despite disagreements over the degree of voter polarization, there is little debate among scholars about the origins and dynamics of these developments. Polarization is an elite-driven phenomenon, and these trends stem from voters following party cues (Levendusky 2009). What is more, both observational and experimental studies have documented the effects of polarization on processes of opinion formation. A polarized party system shapes how voters see the world and make decisions. In a polarized environment, parties have greater influence over voters than in non-polarized contexts. “[P]olarization intensifies the impact of party endorsements on opinions, decreases the impact of substantive information, and perhaps ironically, stimulates    21 greater confidence in those—less substantively grounded—opinions” (Druckman et al. 2013: 57). Thus, factual information becomes less relevant to voters’ opinions when polarized parties dominate the electoral arena. What is the record elsewhere? The emergence and growing popularity of far-right parties in various European countries raise concerns about extremism and radicalization, yet polarization has received little attention in these contexts. One recent, notable exception is Ezrow, Tavits and Homola (2014), which investigates the conditioning effect of partisanship on the relationship between voter polarization and support for extreme parties.  Party system polarization serves as a key explanatory variable in many areas of research, but longitudinal studies that track the relationship between political elites and the mass public are rare. In contrast to closer partisan ties in the US, a common European finding is a general decline of partisanship, expressed in loosening ties between parties and voters (see, e.g., Dalton and Wattenberg 2000; Franklin et al. 2009).  The American case demonstrates how polarization and partisanship are linked. Is this link unique to the US, or do similar mechanisms operate in Europe? Is it possible that a process of depolarization is (partially) responsible for partisan dealignment? The handful of existing studies that apply the designs and measures developed in the American context indeed suggest that the dominant aggregate pattern in Europe is depolarization. So far, however, the evidence comes predominantly from single-country studies—the UK (Adams, Green, and Milazzo 2012), and the Netherlands (Adams, DeVries, and Leiter, 2012). Following the American example very closely, all three studies concentrate exclusively on the two major parties and cover only limited periods. Two recent comparative studies provide mixed results: Focusing on economic and cultural values, Bartels (2013) finds a polarization trend within electorates, while Jansen et al. (2013)    22 conclude that there is no clear pattern in the Left-Right movement of European political parties. Munzert and Bauer (2013) study public opinion polarization in Germany and find a decrease over time. Consequently, whether depolarization is a Europe-wide trend and how it affects the relationship between voters and parties is unclear. This chapter investigates party-voter relations in fifteen European democracies at the level of party systems and electorates. Using data from party manifestos and survey data, the chapter tracks party and voter positions along multiple issue dimensions over a 40-year period. It begins with a discussion of the meanings of polarization and the measures appropriate to each. These measures are then applied to pooled public opinion and party policy data from 1973 to 2014 for consolidated European democracies. Opinion data are from the Eurobarometer surveys, while party data are from the Manifesto Research Group/Comparative Manifestos Project (MRG/CMP). Despite their shortcomings (see, e.g., Benoit and Laver 2007), only the CMP data track all developed democracies for each election in the full period. Both data sets enable the examination of relationships among parties, voters, and the social and economic sub-dimensions of Left and Right. For “hot-button” issues that have exploded in salience in recent years, multiculturalism and support for the EU, only party data are available.   2.1 Polarization: Concepts and measures Although polarization is a widely used term, it carries more than one meaning. Scholars distinguish between two broad concepts: radicalization along individual dimensions and alignment across dimensions. The former is distributional, movement by voters or parties toward or away from extreme values on some axis of political evaluation. The latter is correlational, the extent to which attitudes on one dimension predict attitudes on other dimensions.    23 2.1.1 Radicalization Understood as radicalization, polarization means that opinions or attitudes become more extreme over time and policy distances among individuals or parties increase. Giovanni Sartori (1976) was the first scholar to systematically include the distance between parties in a typology of party systems.7 For him, the emergence of extreme parties on either end of the ideological spectrum is of concern for two reasons. First, large and increasing distances signal high and heightening levels of conflict within a society that may ultimately lead to unbridgeable differences. Second, extreme parties may not only reflect the level of disagreement but also contribute to it by influencing disenchanted members of the electorate with populist ideas to provoke violent change. Thus, Sartori associates radicalization with fractionalization and political instability. Since then, studies have demonstrated that polarization and fractionalization are not necessarily linked (Dalton 2008). Still, party system polarization is commonly associated with a number of undesired outcomes, and negative consequences seem to outweigh positive ones. For example, greater party system polarization decreases government congruence by increasing a government’s distance from the median voter (Powell 2010). Likewise, citizens in centrist party systems tend to be more satisfied with the way democracy works in their country, at least on average (Ezrow and Xezonakis 2011). However, polarized party systems, especially those offering numerous options, lower the level of discontent among electoral “losers” by enabling diverse political voices to be articulated (Anderson 2012). What remains unclear is whether Sartori’s second concern is warranted: Is there a link between party system polarization and a polarized electorate?                                                  7 Previous to that, the number of parties was considered the central distinguishing party system feature.     24 In the US since the 1970s, political elites have drifted away from the center of the liberal-conservative spectrum (see, e.g., McCarty, Poole, and Rosenthal 2006). Students of the American context ask to what extent parties and voters have radicalized in tandem. Evidence is mounting that the American public has not radicalized along with the two major parties. Instead, most Americans occupy moderate positions on most issues and aggregate distributions have changed little (Fiorina et al. 2005; Layman 2001). Party elites aside, radicalization is a story about activists. Applying the American framework to Europe requires adjustments. As the US is a two-party system, the simple range between Republicans and Democrats suffices to index the level of radicalization. This is because Republicans and Democrats are both the extremes and the sole parties of government. For multiparty systems that typify the European landscape, the range between extremes is also important. It provides information about the breadth of positions that are being expressed in the political discourse. But it does not suffice. In these contexts, extreme parties do not commonly enter governing coalitions. They may be important in the final stages of coalition-building, but they are usually small and there are often incentives to exclude them. Initiative for government formation lies with bigger, usually moderate parties (Blais and Bodet 2006; Golder and Stramski 2010). To assess the level of polarization in multiparty systems we need to pay attention to both, as focusing exclusively on extremes misses movement toward or away from the center among the parties between the extremes. This requires a measurement that captures them all. The obvious additional indicator is the standard deviation (SD), as a generalization of the range. Like the range, the SD delivers information about distances. Unlike the range, it incorporates information about all the units.      25 In the literature, two versions of the SD are used to capture radicalization—unweighted or weighted for party size as in Dalton (2008). Weighting in this way aligns the SD with the logic of Laakso and Taagepera’s (1979) Effective Number of Parties (ENP). It ensures that small parties have a relatively smaller impact on the indicator than large parties, thereby presenting a more accurate picture of the level of party competition. Most studies of polarization in Europe use the weighted index, and appropriately so. For our focus, on the relationship between party systems and electorates, however, weighting by party size inserts the voter distribution into both measures. There is no consensus as to which is best suited to capture polarization in multiparty systems. A compromise appears to be the application of both measures (see, e.g., Ezrow 2007). We follow the same strategy and run our analysis with weighted and unweighted measures, where appropriate. We report the results of the unweighted measure in the text, and those of the weighted measure in the appendix. In our case, the results are almost identical, and different measures do not change the overall results. This may be due to the fact that the CMP is self-censoring: if the party is sufficiently obscure that the CMP research group cannot find it or chose not to code its manifesto, the party perforce disappears from the analysis.8  For electorates, the range makes no sense at all. In a survey sample, it is always likely that some respondents occupy the extremes. All the range indicator does is notice their existence. To put such extreme individuals in context requires information on overall dispersion. As with parties, an obvious summary indicator is the SD. Point by point comparison of party and electorate SDs is not very informative as the underlying variables are qualitatively different. For parties, the SD is based on a subset of discrete choices. For voters, the SD summarizes a                                                  8 For an in-depth discussion of how the type of measure affects empirical findings, see Best and Dow (2014).    26 continuous distribution. But we can look for trends in each and for parallels in those trends, as an extension of what Ezrow (2007) calls distributional representation (see, also, Pitkin 1967). What studies of European countries have found so far is the opposite of the US: a depolarization trend. In the UK, the two major parties converged after 1990—the end of the Thatcher era—on economic and social welfare policy—the traditional Left-Right dimension. Over the same period, the distribution for voters did not change (Adams, Green, and Milazzo 2012). In the Netherlands, mass and elite depolarization tracked each other: between 1986 and 1998 the CDA and PvdA converged, as did their partisans (Adams, DeVries, and Leiter 2012).9  The empirical situation is not very satisfying, however. First, the findings are from only two single-country studies and span a limited period. Serious generalization requires more cases and a longer time span. This, of course, is the central point of this chapter. Second, the picture of depolarization seems at odds with the growth of extremist parties, especially on the far right (Kitschelt 1997; Meguid 2008). We propose two extensions to address this shortfall. One is to consider an additional indicator of dispersion, which we can do only for voters, as it requires more data points than are available for parties. The other is to expand the issue space, to look at two policy dimensions that have become hot buttons. This we can do only for parties. The additional indicator is sensitive—but not over-sensitive—to extremes. The conceptual starting point is the distribution’s kurtosis. Where SD is the second moment of a distribution, kurtosis is the fourth moment. Although the SD incorporates information from all                                                  9 In a study not focused on polarization as such but on elite cues and class voting, Jansen, Evans and De Graaf (2013) touch on left-right differences in several party systems (including some non-European ones) as a possible source of the decline in class voting. They find that elite polarization is indeed positively related to class differences in the vote, but the effect is essentially cross-sectional. They find few instances of left-right depolarization, although more such instances than of increased polarization.      27 units, it does not portray all that is useful to know about the shape of the distribution. The openness of the system to disruption by extremist or single-issue parties requires us to look also at the shape of the distribution, in particular to see if the tails of the distribution are “fat.” As with SD, the calculation of kurtosis uses all available data points, in this case yielding a systematic and comparable measure of  “the location- and scale-free movement of probability mass from the shoulders of a distribution into its center and tails” (Balanda and MacGillivray 1988: 116). Most importantly, high kurtosis values indicate the heaviness of the tails, which for us means the weight of the political extremes. SD glosses over this weight, in that a necessary concomitant of heavy tails is massing around the mean.10 SD, in effect, averages across these logically linked, but contrary movements. Depolarization as indexed by SD may actually mask growth on the extremes. Kurtosis captures such growth directly. But all measures that use moments as classically constructed are over-sensitive to outliers. This is an especially acute issue when the number of points on the distribution is small. To hand, fortunately, is a more robust alternative, L-kurtosis, so designated because (like all L-moments) it is a linear combination of order statistics (Hosking 1992). In other words, “L-kurtosis is a standardized measure of kurtosis that adjusts for overresponsiveness of kurtosis to extremes” (Jones et al. 2009: 860).11 Our expansion of the issue space takes in multiculturalism, which is also a conduit for anti-immigrant sentiment in the EU, and orientation to the EU itself. Both seem to have become more charged in recent years, explosively so as we write. The CMP data enable over time                                                  10 “Any movement of mass from the shoulders to the tails must be accompanied by the movement of mass into the centre if the scale is to be left unchanged.” (Ibid., 116).  11 The fundamental definition of an L-moment is Hosking (1990), Equation 2.1, at 106. For a pointed comparison between the representational properties of kurtosis and L-kurtosis see Hosking (1992), Figures 1 and 2. For an extended application of L-kurtosis as a political indicator, see Jones et al. (2009).    28 comparisons fully comparable to those for Left-Right orientations. But as these are manifesto data, the extension is possible only for parties and party systems. 2.1.2 Alignment Although radicalization is the dominant conception of polarization, it is not the only one. The other is issue alignment, either to bring issues and party preference closer together or to agglomerate different issues into distinct “brands” or “ideologies.” Which is to say that parties or voters can polarize even if there is no overall massing at extreme positions. As Baldassarri and Gelman (2008: 419) put it, “… if people align along multiple, potentially divisive issues, even if they do not take extreme positions on single issues, the end result is a polarized society.” Tightly coupled issue dimensions may create dividing lines and lead to a polarized debate, even when positions are moderate. Likewise, it reduces the options for bargaining. Understood this way, the idea of alignment builds on Converse’s (1964) notion of attitude constraint, and is an expression in the domain of attitudes of older propositions about party systems. Whether groups in society were cumulative or cross-cutting in their support for parties was a major preoccupation in the 1960s and early 1970s (Lipset 1960; Lipset and Rokkan 1967; Rae and Taylor 1970) and seems to be undergoing a revival (Selway 2011). And it is this conception that dominates US analyses. The two major parties have created ideological brands that are internally coherent and sharply distinct, with little (in recent years, no) overlap. Although elite polarization combines radicalization and alignment, American voters have become only modestly radicalized, but rather more aligned with their respective parties (Baldassari and Gelman 2008; Hetherington 2008; 2009; Levendusky 2009). Again, European studies find a different pattern. In the post-Thatcher era, UK parties became less aligned across political dimensions, while among voters little changed (Adams,    29 Green, and Milazzo 2012). In the Netherlands, issue dimensions have decoupled for parties and voters alike, leading to less attitude constraint than before (Adams, de Vries, Leiter 2012). Munzert and Bauer (2013) find the same pattern for voters in Germany. 2.2 The Evidence: Left-Right and sub-dimensions For parties, Left-Right is measured by the CMP’s summary “rile” scale—an index comprising twenty-six variables that captures a variety of economic and non-economic policy issues. Left and Right do not anchor a single dimension, however, and a central challenge to European parties is to manage the increased importance of cultural politics (see, e.g., Kitschelt 1997). Accordingly, we follow Benoit and Laver (2006) and divide Left-Right into economic and social/cultural sub-dimensions.12  For voters, a clean economic versus social-cultural distinction is not available. The Eurobarometer includes a single Left-Right scale, ranging from 1 (Left) to 10 (Right). No distinction between economic and cultural sub-dimensions in hinted at. But the Eurobarometer does have a cultural index, derived from Inglehart (1977), that combines two value-priority variables and also has a 10-point range (1=post-materialist, 10=materialist). 2.2.1 Party systems: Distributional indicators (SD and Range) For parties, depolarization appears to be the dominant Europe-wide pattern, and the pattern holds for both traditional distributional indicators, standard deviation and range, and for all three representations of Left and Right.                                                   12 On the “rile” indicator the economic elements are: market regulation, economic planning, protectionism, controlled economy, nationalization, welfare state expansion/limitation, education expansion, labour groups, free enterprise, incentives, and economic orthodoxy. The social-cultural ones are: anti imperialism/anti colonialism, military, peace, internationalism, freedom and human rights, democracy, constitutionalism, political authority, national way of life, traditional morality, law and order, social harmony.    30 The top three panels in Figure 2.1 track trends in the standard deviation. All lines plot the mean of the pooled data with further smoothing by lowess (Cleveland 1979). We separate systems that were consolidated before the most recent waves of democratization from those that democratized only after the mid-1970s. Old systems appear with solid lines and new systems, with dashed ones. Increasing values signal polarization, while decreasing values point to depolarization. For comparability with the Eurobarometer, the original -100 to +100 CMP range has been transformed to a 1-10 scale. Depolarization is most evident for the summary Left-Right dimension depicted in the top left panel. The downward slopes of both lines indicate that the dispersion of party positions has narrowed since the 1970s. While older democracies (solid line) depolarize at a fairly constant rate, new democracies (dashed line) follow a more uneven pattern, with steep declines between 1985 and 1995. Regardless, party systems in both old and new democracies appear to have depolarized along the Left-Right dimension over the last four decades—at least until 2005. Since then, both lines go up, suggesting a recurrence of polarization, although not to former levels.  Looking at the panels for the social and economic sub-dimensions reveals that the non-monotonicity in the summary scale is driven by the economic sub-dimension. Convergence between 1970 and 2005 was entirely the result of developments in old democracies. In the new ones, parties were less polarized from the start and exhibited no net gains or losses. After 2005, however, both old and news systems polarized on roughly the same timetable and with roughly the same endpoint values. For all systems, the social policy trend is negative, toward depolarization. In old democracies, party positions were stable for most of the period, with downtrends early (1970-1975) and late (2005-2010). New democracies, on the other hand,    31 display a much clearer trend: although their parties were highly polarized in the 1970s, party convergence was dramatic, such that these systems are now less polarized than the old ones. For the range between the two most extreme parties, as shown by the three bottom panels in Figure 2.1, the overall pattern looks almost identical to that for SD. The trends seem shallower than for SD, but this is an artefact of scaling. As the indicator for extremes, the range necessarily has larger values than the SD. The essential fact is that depolarization in the economic element of Left and Right occurred across the board, as parties converged from extremes as well as around the center. Table 2.1 shows how individual countries fared during this period. The values reported are directions of coefficients, estimated in a series of OLS regressions in which the distributional indicator (SD or range, as indicated) is regressed on time. The results confirm the general depolarization trend on all three renderings of Left and Right, irrespective of distributional indicator. Yet Table 2.1 also reveals differences among countries. For SD, the depolarization trend along the Left-Right dimension is statistically significant in four countries: Finland, Ireland, Norway and Spain. All other countries, except Denmark and Italy, also show hints of depolarization, though their regression coefficients do not reach statistical significance. On the economic dimension, Italy and Portugal, and to some extent Germany, are the only countries with increasing polarization; all other countries depolarize over time. For the social dimension, the picture is mixed. Half of the systems seem to polarize and half, to depolarize. Trends are more robust on the depolarization side, including the only four systems in which the trend is statistically significant: Finland, Great Britain, Portugal, and Spain.  The depolarization trends are clearer when we use a simple range measure. For most systems, SD and range produce the same results. The clearest exception is Germany, where the    32 range has widened on all three representations of Left-Right. For SD there is no such trend. For Germany, we infer that gaps have closed around the center but widened at the extremes. 2.2.2 Electorates: Distributional indicators (SD and L-kurtosis) How do European electorates compare to their party systems? Figure 2.2 tracks the time path of dispersion of citizens’ positions along two dimensions: the summary Left-Right axis and a cultural dimension. The graphs follow the same design as the ones in Figure 2.1: both lines represent the smoothed means across all systems. Solid lines symbolize old democracies and dashed lines, new ones. Again looking first at the summary Left-Right axis, the solid line for old democracies shows a sharp decline before 1980. Thereafter it continues to drop, but at a much slower pace. The dashed line for newer democracies follows a longer downtrend, lasting until 2000, when it gets close to the level of older democracies. After 2004, the line moves upward, suggesting (temporary?) polarization. Regardless, even for new democracies the dispersion in the 2000s remains below that of 1980s and early 1990s. Thus, over the long-term the evidence suggests the same direction as for parties, that is, toward depolarization.  The trend is equally clear for the cultural (postmaterialism) dimension. Both smoothed-mean lines slope downward, especially between 1985 and 2005. As with parties in new democracies, electorates began as quite polarized but then depolarized sharply and ended up less polarized than their counterparts in old democracies. As mentioned earlier, different variance-based indicators capture different facets of dispersion. The SD says nothing about the shape of the distribution, and apparent stability in its values may mask politically consequential shifts at specific points in the distribution. It is    33 possible, for example, that as major parties converge, gaps widen on the flanks—possibly in reaction to that very convergence on the center.  For such developments the more informative indicator is a kurtosis-based one, as a warning signal of how vulnerable the system is to punctuations. In fact, L-kurtosis values increased in the survey data, according to the bottom panels in Figure 2.2. For reference, an L-kurtosis value of 0.12 corresponds to a normal distribution. A value less than 0.12 indicates thin tails and comparatively thick “shoulders.” A value greater than 0.12 indicates the opposite, thick tails and depleted “shoulders.”  For the Left-Right dimension, both smoothed lines show a slight upward trend and eventually cross the 0.12 threshold. This polarizing trend is even more obvious for the cultural dimension, at least until 2000. Both lines show a steep upward move, but at different levels. The solid line for old democracies is clearly above the 0.12 threshold after 1985 and remains there notwithstanding a downtrend between 2000 and 2005. The dashed line for new democracies follows a parallel trend, although always at a lower level. Indeed, the new democracy line never crosses the 0.12 threshold. The trend simply moved the distribution of voters’ opinion toward the normal. As among old democracies, the tails of the distribution apparently thinned out dramatically after 2005.  Table 2.2 confirms that the continent-wide depolarization trend found in Figure 2.2 is also present in the majority of European electorates. Again, the reported values are directions of coefficients, estimated in a series of OLS regressions in which the distributional indicator (SD or L-kurtosis, as indicated) is regressed on time. Along the Left-Right dimension, nine of the fifteen countries show clear depolarization trends, while eight countries depolarize along the social (postmaterialism) dimension. Here too, we see differences between countries. Electorates in    34 three countries—Denmark, Italy and Sweden—polarize on the Left-Right dimension. In one country—France—the electorate polarizes on the cultural dimension.  The alternative dispersion indicator—L-kurtosis—paints a very different picture. It suggests a polarization trend for most countries (eight out of fifteen) on both dimensions. Taken together, these results suggest that the overall distribution has narrowed over time, hence the decreasing values for the standard deviation. Simultaneously, the shape of the distribution has changed to one with wider flanks.  2.2.3 Party systems: Association between dimensions Although Baldassarri and Gelman (2008, Figure 2.4) do not find increased constraint issue by issue in the American public, they do find that respondents’ self-placement on a general liberal-conservative scale and their positions on issues with moral content have become more aligned. By implication, economic and moral dimensions are collapsing into each other. Has something like this happened in Europe as well? History suggests otherwise. European multipartism in tandem with the Proportional Representation electoral formula reflect and reinforce a long-standing multidimensionality (Sartori 1976). Catholic parties, for instance, combine anti-capitalist tendencies with strong social conservatism. More generally, European parties have strong incentives to keep the rank of the issue space high. On trends in the relationship between the dimensions the starting position must be an agnostic one.  For political parties, we examine the correlation between the economic and social scales introduced above. The scatterplots in Figure 2.3 illustrate this relationship and how it has developed over time at the level of national political parties. Because the number of parties in most systems is small, data are pooled by decade. The coordinates for each point are the party’s position on each scale for each election the party contested in that decade. Parties in older    35 democracies are represented by grey circles, those in newer democracies by black ones. The summary statement about the link between dimensions for a decade is an OLS regression line. By analogy to earlier figures, older democracies are represented by a solid line and newer democracies, by a dashed one.  For new democracies, little change appears between the 1980s and 2000s. Indeed the relationship may even have weakened. For older democracies, the steepening slope of the solid regression line suggests that the correlation between dimensions has increased. Political parties evidently are taking positions that are increasingly consistent across the two Left-Right dimensions. In the American literature, this would be critical evidence for polarization, for the forcing of both economic and social liberals into one elite camp and political and social conservatives into another. For multiparty systems, this is better read as a decrease in dimensionality. European choices are simplifying but the narrowing of the distribution around the centre indicates that this simplification accompanies a net convergence among parties. In newer democracies, there appears to have been no simplification of the choices. 2.2.4 Electorates: Association between dimensions Figure 2.4 illustrates the relationship for electorates. We employ the same scales as above, Left-Right and postmaterialism. The graph follows the same design as the one for parties with the exception that individual data points are suppressed for readability. Again, OLS regression lines present the summary statements about the link between dimensions for each decade. By analogy to earlier figures, older democracies are represented by a solid line and newer democracies, by a dashed one. The pattern is the opposite of that for parties: for voters, the dimensions have    36 become less coupled, especially in the last two decades (1990-2010). This is true for both older and newer democracies.13  Parties and voters appear to diverge on the bundling between economic and social dimensions: parties bundle the two dimensions more than before, voters less than before. The divergence could be an artefact: our measure of the social dimension diverges between parties and voters and may not be measure the same thing. But the picture is not one of either parties or voters assembling issues in a way that raises political tension. Parties bundle more than before but the positions that go into the bundles are more centrist than ever. For voters, links between the dimensions are loosening, not tightening.  2.3 Further Evidence: Multiculturalism and European Union  In recent European political discourse issues other than those captured by Left-Right ideology have gained prominence. Chief among these are multiculturalism and the European Union itself. As the Eurobarometer caught up with these issues only after they became salient, we do not have a usable time series for electorates. But the CMP/MRG is comprehensive in its coverage of manifestos. That manifestos said less about multiculturalism or about Europe in earlier decades is not fatal to the analysis; indeed, it speaks to the very point. 2.3.1 Party systems: Distributional indicators (SD and range) As before, we apply two dispersion indicators, standard deviation and range, to the party system data. The top panels in Figure 2.5 display the graphs for the standard deviation. Again, solid lines                                                  13 Data for newer democracies is not available for the 1970s.         37 represent old democracies, dashed lines new ones. The difference to the previous figures is quite striking in that we see clear polarization trends, along both dimensions. Gains in polarization are particularly striking for positions on the EU. The lines in the multiculturalism panel also show an upward trend, though the degree of polarization on the issue appears smaller than for the European dimension. The pictures in the bottom panels look almost identical to those immediately above: on both dimensions the range between extremes has expanded and done so for both for old and new democracies. Table 2.3 confirms the ubiquity of polarization at the level of individual countries. As in the previous tables, the entries are directions of coefficients, estimated in a series of OLS regressions in which the distributional indicator (SD or range, as indicated) is regressed on time. For the European dimension, fourteen of the fifteen coefficients for the standard deviation have a positive sign and four show statistically significant increases over time. The results are even clearer for the multiculturalism dimension. For the range between party extremes this pattern is even stronger.  2.3.2 Party systems: Association between dimensions In contrast to the tightening of links between the economic and social/cultural sub-dimensions of Left and Right, the newly polarized dimensions of multiculturalism and the EU remain quite separate from Left and Right even as they have become more tightly linked to each other. The evidence appears in Figure 2.6, Figure 2.7, and Figure 2.8, which are constructed along the same lines as Figure 2.3. Here too, party manifesto data are pooled by decade.  Links between parties’ Left-Right positions and their positions on new issues appear in Figure 2.6 and Figure 2.7. The lines for both old and new democracies are essentially flat and show very little change over time. That is, parties’ positioning on Europe and on    38 multiculturalism and their traditional ideological ordering are essentially unrelated. After 2000, parties of the right in old democracies made a modest turn. On Europe (Figure 2.6), this turn reversed a weak pattern facing the other way, as relative to the Left, the Right earlier mildly favoured Europe. After 2000, right-wing positions were associated with anti-Europe ones. On multiculturalism (Figure 2.7), the trend was broadly similar. Before 2000, there is the merest hint of a link between Left-Right and resistance to multiculturalism. After 2000, the link clearly strengthens.  Even so, the association is weak by comparison with the picture in Figure 2.3 for links within the Left-Right complex. It is also weak when compared to the tightened link, at least in the old democracies, between the EU and multiculturalism (Figure 2.8). Before 1990, the line is essentially flat. The scatterplot shows why: the outstanding fact is the vertical parade of circles on the neutral point along the multicultural axis. Where parties were already staking out positions on Europe, few bothered to take any position on the cultural axis. The few that did—mainly in favour of multiculturalism—were basically neutral on Europe. The 1990s signal the beginning of a shift, although many parties still remained indifferent on multiculturalism. After 2000, parties moved to positions on both dimensions, with a clear positive association between the two: support for Europe and embrace of multiculturalism now proceed in tandem.  2.4 Conclusion For the social and economic components of Left and Right, the dominant European theme of the last 40 years is depolarization, or more specifically, de-radicalization. In this, Europe stands in contrast to the US. For parties and party systems, this claim is basically unqualified: distances along the entire dimension have shrunk, including those between extremes. For voters, the claim is stronger for the central parts of the distribution than for the tails, which became “fatter”    39 (although this polarization at the extremes may have retreated after 2005). For parties, the two subdomains of Left-Right ideology, economic and cultural, became more aligned with each other, somewhat echoing the trend in the US. But, as implied by the de-radicalization claim, this alignment between dimensions accompanied a general clustering of parties around the centre. For voters, the opposite happened: social and economic opinions became detached from each other, most strikingly in the newer democracies. If the traditional ideological core of elections became less divisive, a new dimension (or dimensions) gained traction. The idea of Europe has been a matter of contestation for years and became only more so as the decades passed, especially after 2000. Before the 1990s, in contrast, parties barely took notice of multicultural policy. In that decade, parties started to take positions on the matter, and as they did so they became divided against each other. As with European integration, the divisions exploded after 2000. Accompanying the radicalization on each of the new dimensions was increased alignment between the two. Although there has also been a modest alignment between party positions within the traditional sphere of Left and Right and their positions on the newer issues—with the Right becoming modestly more anti-multicultural and more anti-Europe than the Left—the new issues are more tightly bound with each other than either is the old issues. The old parties seem threatened on both flanks. The contrast with the US is striking. In the US, the system may have been ripe for polarization. If nothing else, there was ample room to move in this direction. The strength of the Democratic Party in the South coupled with the growing liberalism of the party’s adherents in other regions was probably unsustainable. As long as it lasted, however, it made the Democrats a peculiarly incoherent body. It was also awkward that the mainsprings of electoral division carried resonances from the 19th century—Catholics versus Protestants, for instance. As the    40 system realigned and conflict extended (Layman and Carsey 2002), the US simply became more like the rest of the world. In Europe, Left and Right in both kinds captured differences of long standing. This was no less true of cultural differences—Red versus Black, for instance—than of economic ones. For the mechanisms underlying these trends, there is no lack of speculation. The logic of “catch-all” parties (Kirchheimer 1966) points toward de-radicalization, at least around the political center. So do changes in political economy. That Europe as such is on the agenda reflects the changes in the community’s boundaries, internal as well as external: the simultaneous extension and deepening of the European Union have primed the dimension. And as Europe and most of its constituent nations became more diverse, diversity itself became a major issue, linked to the Union.  Given the findings from the US that party system polarization has altered the relationship between parties and voters (e.g., Levendusky 2009), the expectations for Europe are mixed. If the mechanisms found in the American context also operate in European multiparty systems, we should expect to see different dynamics along different issue dimensions, conditional on the level of polarization.     41 Table 2.1 Party systems: Polarization and depolarization on Left-Right   Standard deviation (SD) Range  Country L-R Econ. Soc. L-R Econ. Soc. All - ** -** -** -** -** -** Austria - - - -** - + Denmark + - + + -** + Finland - ** -** -** -** -** -** France - -** + + -** +** Germany - + - +* +** + Great Britain - -** -* -** -** + Greece - - - -** -** - Ireland - * - - -** - -** Italy + +** + +** +** + Luxembourg - - + -** -** +* Netherlands - -* + -** -** +** Norway - * -** + -** -** +** Portugal - +** -** -** +** -** Spain - ** -** -** -** -** -** Sweden - - + - - + Polarization 0 (2) 2 (3) 0 (7) 2 (4) 3 (3) 4 (10) Depolarization 4 (13) 6 (12) 4 (8) 10 (11) 9 (12) 4 (5) Source: Comparative Manifesto Data Notes: This Table reports the direction of coefficients, estimated in a series of linear regression models in which the dependent variable is either the SD or range of party positions on different issue dimensions (as indicated), as a function of time.  * p < 0.1, ** p < 0.05      42 Table 2.2 Electorates: Polarization and depolarization on Left-Right  Standard deviation (SD) L-kurtosis Country L-R PM L-R PM All - ** -** +** +** Austria - ** - + - Denmark + ** -** -** +** Finland + + + - France - ** +** +** +** Germany - ** -** +** +** Great Britain - ** + +** +** Greece - * -** +** + Ireland - ** + +** +** Italy + ** -** -** +** Luxembourg - ** -** +** +** Netherlands - ** -** +** +** Norway + -* + + Portugal + - + - Spain - ** -** +** + Sweden + * - + - Polarization 3 (6) 1 (4) 8 (13) 8 (11) Depolarization 9 (9) 8 (11) 2 (2) 0 (4) Source: Eurobarometer Trend File Notes: This Table reports the direction of coefficients, estimated in a series of linear regression models in which the dependent variable is either the SD or L-kurtosis of voter positions on different issue dimensions (as indicated), as a function of time.  * p < 0.1, ** p < 0.05      43 Table 2.3 Party systems: Polarization on European integration and multiculturalism  Standard deviation (SD) Range Country EU MC EU MC All +** +** +** +** Austria +** +** +** +** Denmark + +** +** +** Finland +** + +** + France + - +** -** Germany + +** +** +** Great Britain + + +* +** Greece + +* +** +** Ireland + +** + +** Italy + + +** - Luxembourg - +** -** +** Netherlands +** +** +** +** Norway + + + + Portugal + +* + +** Spain + + +** + Sweden +* - +** + Polarization 4 (14) 8 (13) 11 (14) 9 (13) Depolarization 0 (1) 0 (2) 1 (1) 1 (2) Source: Comparative Manifesto Data Notes: This Table reports the direction of coefficients, estimated in a series of linear regression models in which the dependent variable is either the SD or range of party positions on different issue dimensions (as indicated), as a function of time.  * p < 0.1, ** p < 0.05     44  Figure 2.1 Party systems: Depolarization on Left-Right and sub-dimensions  Left-Right Economic Social Standard deviation    Range    Source: Comparative Manifesto Data Notes: This Figure reports values of distributional indicators of polarization (SD or range, as indicated) over time and across different issue dimensions. Solid lines represent the average trend for old democracies (Austria, Denmark, Finland, France, Germany, GB, Ireland, Italy, Luxembourg, Netherlands, Norway and Sweden). Dashed lines illustrate the average movement for new democracies (Greece, Portugal and Spain). All means are smoothed by lowess, bw=0.35.      45 Figure 2.2 Electorates: Polarization and depolarization on Left-Right and postmaterialism  Left-Right Postmaterialism Standard deviation   L-kurtosis   Source: Eurobarometer Trend File Notes: This Figure reports values of distributional indicators of polarization (SD or L-kurtosis, as indicated) over time and across different issue dimensions. Solid lines represent the average trend for old democracies (Austria, Denmark, Finland, France, Germany, GB, Ireland, Italy, Luxembourg, Netherlands, Norway and Sweden). Dashed lines illustrate the average movement for new democracies (Greece, Portugal and Spain). All means are smoothed by lowess, bw=0.35.      46 Figure 2.3 Parties: Association between economic and social dimensions     Source: Comparative Manifesto Data Notes: This Figure shows a series of scatterplots with OLS regression lines to illustrate the relationship between positions on the economy and positions on social/cultural issues, estimated for each decade, using political parties as the unit of analysis. Steeper slopes indicate tighter issue alignment across dimensions. Grey circles and solid lines represent old democracies (Austria, Denmark, Finland, France, Germany, GB, Ireland, Italy, Luxembourg, Netherlands, Norway and Sweden). Black circles and dashed lines represent new democracies (Greece, Portugal and Spain).      47 Figure 2.4 Voters: Association between Left-Right and postmaterialism     Source: Eurobarometer Trend File Notes: This Figure shows a series of OLS regression lines to illustrate the relationship between voters’ positions on Left-Right and positions on postmaterialism, estimated for each decade, using individual survey respondents as the unit of analysis. Steeper slopes indicate tighter issue alignment across dimensions. Solid lines represent old democracies (Austria, Denmark, Finland, France, Germany, GB, Ireland, Italy, Luxembourg, Netherlands, Norway and Sweden). Dashed lines represent new democracies (Greece, Portugal and Spain).       48 Figure 2.5 Party systems: Polarization on European integration and multiculturalism  EU Multiculturalism Standard deviation   Range   Source: Comparative Manifesto Data Notes: This Figure reports values of distributional indicators of polarization (SD or range, as indicated) over time and across different issue dimensions. Solid lines represent the average trend for old democracies (Austria, Denmark, Finland, France, Germany, GB, Ireland, Italy, Luxembourg, Netherlands, Norway and Sweden). Dashed lines illustrate the average movement for new democracies (Greece, Portugal and Spain). All means are smoothed by lowess, bw=0.35.      49  Figure 2.6 Party systems: Association between Left-Right and European integration     Source: Comparative Manifesto Data Notes: This Figure shows a series of scatterplots with OLS regression lines to illustrate the relationship between positions on the traditional Left-Right dimension and positions on European integration, estimated for each decade, using political parties as the unit of analysis. Steeper slopes indicate tighter issue alignment across dimensions. Grey circles and solid lines represent old democracies (Austria, Denmark, Finland, France, Germany, GB, Ireland, Italy, Luxembourg, Netherlands, Norway and Sweden). Black circles and dashed lines represent new democracies (Greece, Portugal and Spain).    50  Figure 2.7 Party systems: Association between Left-Right and multiculturalism     Source: Comparative Manifesto Data Notes: This Figure shows a series of scatterplots with OLS regression lines to illustrate the relationship between positions on the traditional Left-Right dimension and positions on multiculturalism, estimated for each decade, using political parties as the unit of analysis. Steeper slopes indicate tighter issue alignment across dimensions. Grey circles and solid lines represent old democracies (Austria, Denmark, Finland, France, Germany, GB, Ireland, Italy, Luxembourg, Netherlands, Norway and Sweden). Black circles and dashed lines represent new democracies (Greece, Portugal and Spain).    51  Figure 2.8 Party systems: Association between multiculturalism and European integration     Source: Comparative Manifesto Data Notes: This Figure shows a series of scatterplots with OLS regression lines to illustrate the relationship between positions on multiculturalism and positions on European integration, estimated for each decade, using political parties as the unit of analysis. Steeper slopes indicate tighter issue alignment across dimensions. Grey circles and solid lines represent old democracies (Austria, Denmark, Finland, France, Germany, GB, Ireland, Italy, Luxembourg, Netherlands, Norway and Sweden). Black circles and dashed lines represent new democracies (Greece, Portugal and Spain).    52 Chapter 3: Parties: Leading and following This chapter continues the analysis of polarization and depolarization at the level of parties and their supporters in two European democracies—the UK and Germany. The focus is on the behaviour of political parties, and I directly address the questions: Who leads? Who follows?  The inspiration for this investigation comes, again, from the US, where research on voting behaviour has uncovered links between party system polarization and the nature of the link between parties and voters. In particular, American studies have shown that voter sorting, rather than voter polarization, is an important consequence of elite polarization (Levendusky 2009). This chapter concentrates on the causal mechanism through which polarization inspires sorting, and the goal is to explore whether the same links operate in different contexts. This requires an unpacking of the causal process proposed for the American case. In doing so, I pay attention to the incentives inherent in proximity and directional models and formulate specific observable implications of the causal chain.  The two cases that I analyze differ from the American in the structure of the party system. Both the UK and Germany have more relevant parties than the US, and the distances between the parties are smaller. Moreover, the UK underwent a process of party system depolarization in the 1990s, when the two major parties converged on Left-Right issues (Budge 1999; Norris 1999; Webb and Farrell 1999). This makes it an ideal case for comparison. A longitudinal analysis can tackle the question whether changes in party system structure lead to changes in party-voter relations. Specifically, does elite depolarization cause the electorate to unsort?  To address this question, I perform a longitudinal analysis that tracks party and voter positions over time and along a number of issue dimensions, using graphical displays (similar to    53 the analysis in Chapter 2). The data come from the British Election Study (BES) and the German Longitudinal Election Study (GLES). Both studies include repeated questions that ask respondents to locate themselves and all parties on multiple issue dimensions. Unlike previous works, I do not use a correlational measure of sorting (Adams, Green and Milazzo 2012; Baldassarri and Gelman 2008; see also Levendusky 2009). Instead, I analyze changes in the distribution of party supporters’ positions over time. I argue this is a better approach to capturing the sorting (or unsorting) process. To support my argument, I demonstrate the flaws of correlational measures. 3.1 The causal mechanism Numerous studies have documented the polarization of the American party system, caused by party elites taking on more extreme positions (for a review, see Hetherington 2009). This change in party system structure has altered the relationship between parties and voters. As Levendusky (2009: 1) puts it, “[p]arty and ideology today are much more tightly aligned than they were a generation ago.” This is, however, not the result of voter polarization but of partisan sorting.  The following two sections illustrate what sorting means and how it is linked to elite polarization. Moreover, they unpack the causal mechanism proposed for the US and formulate expectations for electorate change in response to party system depolarization. The third section discusses measures of sorting and unsorting and introduces observable implications for each process. 3.1.1 From polarization to sorting Sorting is a response to elite polarization and describes movement by voters relative to parties. Levendusky (2009: 4-5) describes sorting as “a changing relation between partisanship and ideology, so that, in a sorted electorate, party and ideology are more closely related (more    54 correlated) than in an unsorted electorate. … Knowing someone’s ideology makes it more likely that one can correctly identify his partisanship.”   In the US case, the starting point for this process is the clarification of party options (Fiorina et al. 2005; Levendusky 2009; for a graphical illustration, see Figure 3.1). Over time, both parties have taken on more extreme positions, thereby moving away from the median voter and increasing the gap between Democrats and Republicans. At the same time, elites in both parties have become more homogenous, sending out clearer and more coherent messages. Party labels now function as brand names, signalling and representing a clear ideology. This simplification of a complex political space into two camps helps voters understand who the parties are and what they stand for. This, in turn, makes it easier for them to align their partisanship with their ideology. This alignment can occur in two ways: a voter can change her partisanship to match her ideology, or she can adjust her ideology to fit her party identification. The latter is the dominant pattern in the US, where “elite polarization has caused voters to adopt the ideological outlook of their same-party elites” (Levendusky 2009: 3). So, sorting is a kind of voter following. Parties persuade voters to adopt their positions—at least versions thereof. Yet it is different from mass polarization. Polarization implies notable shifts in voter positions, as voters move towards the extremes of the political spectrum, and moderates disappear (Fiorina and Levendusky 2006). Sorting, on the other hand, organizes citizens into teams of like-minded voters without necessarily changing the distribution of positions (see, Figure 3.2 for a visual illustration). All that is required for sorting is for voters to take the same side of an issue as their preferred party, not to take on more extreme positions. In the aggregate, the effects of sorting are subtle. What changes is the make-up of each voter group. In a sorted electorate the supporters of each party are a more homogenous group in terms of their    55 ideology than in an unsorted electorate. In short, in the US “most liberals are Democrats; most conservatives are Republicans” (Levendusky 2009: 5). Although sorting causes less of a direct shift in the distribution of voter opinions than mass polarization, it has important behavioural and attitudinal consequences. Sorted voters are more loyal to their party, more likely to adjust their own positions to the party’s, and to evaluate the party more positively (Levendusky 2009: 126-127). Thus, a sorted electorate allows parties to take the lead. Sorting and mass polarization are also based on different motivations. As parties polarize, they move away from the median voter and away from their median party supporter, thereby increasing the policy distance to voters. In that sense, they become less representative of their own supporters, as well as of the median voter’s position. Mass polarization is an attempt by voters to minimize the distance to their favoured party by moving their own positions towards the party’s. Thus, polarization of voters is driven by a proximity calculus.  Sorting is a different type of response to the same situation. Rather than moving with their party towards more extreme positions, voters sort by choosing the same side of an issue as their preferred party (Levendusky 2009: 44). Thus, directional considerations are behind sorting. Put simply, what matters in a polarized party system is which side the parties are on, not how far away they are from a voter. Because both parties move in opposite directions, they raise the stakes for voters, as the parties stand for increasingly different sets of policies. From the perspective of a voter, these might represent more extreme positions than her own. However, because they are so different, one is clearly preferable to the other. Hence, although each party represents its voters less, it matters more which one gets into office. Therefore, as parties drift towards more extreme positions and become less representative of their median party voter, the    56 proximity between parties and voters carries less weight for party evaluation. Instead, what counts is the general direction of policy. To synthesize, in the American case, party system polarization—via sorting—has altered the calculus of voting and the dynamics of leading and following. The importance of proximity considerations for party preferences has decreased, as party system polarization increased; instead directional orientations have taken their place. The distance that has gained in relevance is between the parties, not between the voter and the closest party to her. This relationship is positive: larger gaps between Democrats and Republicans have led to stronger party-voter ties, irrespective of the distance between parties and voters. Thus, stronger partisan ties are not the function of better representation, but of an interaction between a voter’s policy positions and the distance between parties.  To illustrate this last point, suppose a voter takes a moderate pro-choice stance on abortion. As both parties adopt more extreme positions, the voter realizes that the Democratic Party shares her general pro-choice outlook, but to a more extreme degree. The Republican Party, on the other hand, represents a stern pro-life position. If abortion is an important issue for this voter, she will likely feel much closer to the Democratic Party than to the Republican Party. Moreover, she may feel more attached to the Democrats than before, even though her own position remains unchanged and the party position is now farther away from her. In this scenario party-voter ties strengthen with the polarization of the party system, while policy distances increase. In the aggregate and over time, correlations between voters’ issue positions and party preferences would increase as a result of growing distances between the parties.    57 3.1.2 From depolarization to unsorting? The comparative analysis of polarization and depolarization (Chapter 2) has shown that the latter is, perhaps, the more prevalent phenomenon in Europe. So, given the findings in the US, should we expect to witness processes of unsorting in Europe? In addressing this question, it is useful to look at the American case through a comparative lens. From this perspective, the American polarization literature investigates a case of party system change and its effects on mass-elite linkages.  In the comparative literature, party systems are commonly categorized along two dimensions: the number of parties competing in elections and their ideological distances to one another (Sartori 1976). Over the last four decades, America’s moderate two-party system has changed into a polarized two-party system. For Sartori, it transformed from his ideal type, which is conducive to the development of stable political systems and a higher quality of democracy, to a much less desirable version. He considered party systems with large ideological distances to be prone for abuse by radical elites who use populist tactics to mobilize disenchanted members of the electorate, and thus endanger democratic performance and regime stability (Sartori 1976: 131-216). In other words, he was concerned that parties would take on the role of leaders. This rings true for the American case, where “party is once again the driving force behind political behaviour” (Levendusky 2009: 1).    58 More generally, different party system structures provide different incentives and opportunities (Sartori 1976; Mair 1999; 2002; 2006; Bardi and Mair 2008; Wolinetz 2006).14 As institutions, party systems facilitate certain actions, while prohibiting or punishing others (Cox 1997; Duverger 1954; Ostrom 1990; Hall and Taylor 1996; Strom 1990; Uslaner and Zittel 2006; North and Weingast 1989). They structure electoral competition, thereby influencing the policy positions and the success of political parties (Meguid 2008). At the same time, party systems determine how many choices voters have on election day and how meaningful (different) these choices are.  The American party system is rather unique and differs from the typical European system in both key dimensions: the number of parties and the level of polarization. Most European party systems are multiparty systems and less polarized. Both factors contribute to a less transparent environment of party competition. Put simply, compared to the US, party choices are more numerous and less distinguishable in European party systems. Thus, depolarization makes it harder for voters to identify, what the parties stand for and where they are located on different issue dimensions. This, in turn, complicates the decision-making process for voters, come election day. They face greater difficulties when trying to choose the party that best represents their own positions.                                                   14 The party system is heavily influenced by the country’s electoral rules, which shape the incentives for parties and the dynamics of party competition (see, e.g. Taagepera and Shugart 1989; Cox 1997). However, sufficient variation suggests that they do not fully determine party system structure. The example of the US party system is a case in point: while the electoral system remained steady, the party system has developed from a centralized to a polarized one, thereby changing the nature of electoral competition. Similarly, in several Westminster systems (e.g., the UK and Canada) the number of parties has increased over time, without a change in the electoral system.    59 How depolarization affects the importance of proximity and directional motivations depends on whether parties lead or follow. When I discuss whether parties lead or follow in this chapter, I refer to observable behaviour. The meaning of leading and following is intuitive and refers to the direction of movement. Parties lead if they move away from their voters. They follow, if they move closer toward their supporters. That is, if parties lead, they increase the policy distance to their supporters; if they follow, they converge on their supporters’ positions. Both sorts of movements can be towards or away from the median voter (or the centre of the political system). In regards to depolarization, parties lead if they move away from their voters, toward more centrist positions. That means parties move to positions that are less extreme than those occupied by their mean party supporter. As a result, the distances to their respective median party voters increase, which will disappoint proximity-oriented voters. At the same time, the intensity component of directional models becomes irrelevant. If parties are more moderate than their voters, there is no option for voters to choose a party that is on the same side but more extreme. Likewise, as parties move towards the centre, they become less distinguishable from each other. Hence, the importance of policy criteria decreases, and the weight of other factors increases (e.g., past performance, leader personality and competence). In this situation, voters have clear incentives to re-evaluate their party preferences and to unsort.  The picture looks different if parties follow their supporters to the centre. In this case, parties move toward their respective bases, closing the distance to their supporters. This satisfies proximity-oriented voters. At the same time, if parties follow, they do not become more centrist than their bases. In other words, they occupy locations that also satisfy directional voters,    60 because the party is still more extreme than the mean supporter. Consequently, voters have fewer incentives to unsort. I analyze the cases of the UK, between 1992 and 2001, and Germany, from 1998 to 2013, to learn more about party-voter relations in depolarized party systems. The UK case is best suited to test the applicability of the causal mechanism, as the party system underwent a change in the 1990s. Furthermore, it is similar to the US in that the two major parties are also the most extreme ones. Given the small number of cases, a systematic test of party system effects is impossible. A controlled comparison across cases that seeks to establish the causal effects of party system structure on voter-party relations would require me to hold everything else constant. My analysis does not hold up to rigorous methodological standards and, thus, cannot confirm my argument that the party system serves as an important independent variable that can help explain cross-national differences in party-voter relations. Still the analysis can help refute it: If I do not see differences where I expect them, the party system is probably less important for the relationship between parties and voters than I propose. What is more, the thorough analysis of two cases is a first step that can help us begin to understand how party system structure affects the dynamic relationship between parties and voters. 3.1.3 Measuring sorting and unsorting The American literature on sorting emphasizes the alignment of partisanship and ideology as the key characteristic of sorting, without explaining the mechanism behind it (e.g., Baldassarri and Gelman 2008; Hetherington 2001; 2009). It is Levendusky (2009) who defines sorting as a process and systematically links it to elite polarization. Accordingly, I measure sorting based on observable implications derived from the causal mechanism proposed by Levendusky (2009). Others have relied on a correlational indicator. In the following sections, I will first discuss the    61 correlational measure used by Adams et al. (2012) and point out its flaws. Thereafter, I will present my strategy.   Adams et al. (2012) investigate the effects of elite depolarization in the UK on the British electorate. They focus exclusively on the two major parties—Labour and Conservatives—and measure sorting using the changing correlation between “party and ideology.” For their party variable, Adams and his colleagues calculate the absolute difference between Conservative and Labour party evaluations (as a proxy for partisanship). Building on Achen’s (1977; see also, 1978) critique of correlational measures to assess changes in strengths of relationships, I argue this is problematic.  Achen makes a strong point against the use of correlational measures in studies of representation: “The larger the correlation between representatives’ and constituents’ views, the stronger the bonds between them are thought to be. Unfortunately, correlations incorporate not only the strength of a relationship, but also the diversity of the sampled constituencies. When constituencies are very different from each other on an issue dimension, large correlations will result even when voters are not particularly sensitive to that dimension. Large correlations can occur when representatives are distant from their constituents; small correlations can happen when they are near. Correlations should be abandoned in the study of representation” (Achen 1977: 805, my italics). Although the underlying variables are different, the same reservations also apply to a correlational measure of sorting, as “two correlations can differ because the variances in the sample differ, not because the underlying relationship has changed” (Achen 1977: 807). To illustrate: the distribution of Adams et al.’s (2012) party variable is likely to become tighter over time, simply as a function of depolarization. As parties move closer together, and voters notice    62 this depolarization trend, they may adjust their evaluations, especially of the other party. That is, if they notice that the differences between the two parties become less stark, they evaluate the other party less negatively. In this scenario, correlations between party and ideology would decrease without any behavioural or opinion changes by voters. Thus, these correlations tell us little about possible electoral consequences, as preferences may be unaffected. Even if the evaluation gap between Conservative and Labour shrinks, voters’ first preferences may remain unchanged.  I choose a different strategy and look for evidence of the causal mechanism described above, based on clearly stated observable implications. I use these observable implications to test whether electorates sort or unsort in response to party system changes. Thus, I focus on longitudinal changes in voter and party positions, and changes in their distributions. 3.1.3.1 Observable implications for sorting Sorting, understood as a process initiated by elite polarization, has a number of measurable effects at the aggregate level. First, if elite polarization is the cause of sorting, parties should move first and move away from their respective median party voters, towards more extreme positions (parties lead). Second, if the clarification of party options initiates the sorting process, voters should have a clearer understanding of where parties stand. They should, therefore, not only notice this move towards extreme positions, but also be able to place parties on a number of policy scales with greater precision. In more technical terms, party positions as perceived by voters should have a tighter distribution, as reflected in smaller standard deviations. Third, as voters sort along partisan lines, the distance between the mean policy preferences of supporters of different parties should increase, at least slightly. This is not because voters change their opinions, but because each party’s supporters should form a more coherent group (e.g., the mean    63 position of Democrats is now more liberal than before, because liberals represent the majority of this group). Likewise, we should see that supporter positions have tighter distributions, also expressed in smaller standard deviations. Fourth, sorting implies that ideology (or policy positions) and partisanship become more aligned. However, as explained above, this is not the result of closer proximity between party and voter positions. In fact, if parties polarize and voters sort, party-voter distances should increase over time.15  3.1.3.2 Observable implications for unsorting If voters in depolarizing party systems respond by unsorting, we can also expect to find evidence of this process in the aggregate. First, if elite depolarization is the cause of unsorting, parties move first and away from their supporters, towards more moderate positions (parties lead). Second, if voters unsort because party positions become less distinguishable to them, they should register parties’ moving. At the same time, they should have more difficulty locating each party on a policy scale, resulting in larger standard deviations of party positions. Third, as voters unsort, the gap between the average position of the groups of supporters of each of the parties should close. This is mainly the result of voter groups becoming more internally heterogeneous (e.g., a greater ideological mix is found among supporters of each party, pushing each supporter group’s mean positions toward the electorate mean). Similarly, the standard deviations for supporter positions of the same party should increase. Fourth, a voter’s position should become less predictive of her party preference. What this means for the distance between parties and voters, is difficult to say, as it depends on a number of factors: whether parties lead or follow and                                                  15 In the same setting, closing gaps between parties and voters would signal a polarizing electorate.     64 the extent of party movement relative to the degree of sorting. I refrain from interpreting these distances as clues for or against unsorting. 3.2 Results from the UK (1992–2001) The period of investigation in the UK immediately follows the Thatcher era and thirteen years of Conservative governments. Despite an ongoing recession and personnel shuffle within the Conservative party, 1992 brought about yet another Conservative victory. The Conservatives were reelected and John Major continued to serve as Prime Minister, a position he had held since taking over from Margaret Thatcher, following her resignation as Prime Minister and Conservative party leader in November 1990. However, the long run of Conservative governments came to an end five years later, in 1997, when the first of a series of “New Labour” governments under Tony Blair’s leadership was elected. The 1997 election, as well as the following one in 2001, brought landslide victories for Blair’s newly reformed Labour Party. New Labour represented a shift from its predecessor—away from a party of the working classes towards a more centrist “catch-all” party. At the same time, the Conservative Party moved towards more moderate positions. The post-Thatcher era has, hence, been described as a period of “dramatic elite depolarization on economic and social welfare policy” (Adams et al. 2012: 520; see also, Budge 1999; Norris 1999; Webb and Farrell 1999).   What about the British electorate? Did elite depolarization cause voters to unsort? To answer this question, I trace the positions of British parties and their supporters on a number of issues over the course of three elections (1992-1997-2001), using a series of graphs. In addition to absolute positions, I track their dispersions over time. The latter is important for two reasons: first, changes in the dispersion of perceived party positions reveal information about the clarity    65 of the environment; and second, changes in the dispersion of party supporters’ positions are necessary to identify shifting patterns within the electorate.  I present two figures for each of the five issues. The graphs comparing British parties’ and voters’ positions are made up of three panels—one for each relevant party. Within each panel, solid lines represent imputed party positions and dashed lines, supporters’ own positions. The grey dotted line shows the electorate mean. The indicator of party position is the mean party placement as estimated by all survey respondents. I distinguish between two types of supporters: party identifiers and all of that party’s voters. Thick dashed lines symbolize the mean positions of party identifiers; thin dashed lines stand for the mean positions of all party voters. Thus, the latter include identifiers and non-identifiers. To measure the clarity of party positions, I calculate the standard deviations (SDs) of all party positions. These are included in the second set of graphs, comparing British parties’ and voters’ dispersions. Each graph consists of three panels—party, voter, and identifier positions—and compares the three relevant parties. I calculate the SD of supporter positions to assess how homogenous these groups are in terms of their policy views.  The data for this analysis come from the British Election Study (BES), and survey respondents were asked to place themselves and the three major parties on five policy issues: (1) unemployment and inflation, (2) taxation and government services, (3) income redistribution, (4) nationalization and privatization, and (5) European unification.  3.2.1 Unemployment and inflation On the topic of unemployment and inflation, BES respondents were asked the questions: Some people feel that getting people back to work should be the government’s top priority. These people should put themselves in Box A. Other people feel that keeping prices down should be the    66 government’s top priority. These people should put themselves in Box K. And other people have views somewhere in-between. In the first row of boxes, please tick whichever box comes closest to your own views about unemployment and inflation. Now where do you think the parties stand? I recoded the answers to range from 0 to 1, with 1 meaning “getting people back to work should be the government’s top priority.”  Throughout the entire period, parties and their supporters demonstrate the expected Left-Right division: Conservatives are below the median voter (less supportive), Labour is above (more supportive), and Liberal Democrats are in the centre (Figure 3.3). Despite small changes in positions, they stay sorted in this formation, providing little evidence of unsorting. The results also show that, at least on this issue, parties are not the ones leading the depolarization trend. They do not move away from their supporters towards the centre. Instead, parties follow their voters towards more moderate positions. This is certainly true for the Conservative party, which is mostly responsible for depolarization on this issue. Conservatives move the most, closing the gap to their own supporters, as well as to Labour. The Labour party also moves, but to a much smaller extent, and with its supporters—at least until the late 1990s.  So, voters notice that parties move towards the centre, but this does not seem to cause confusion about the parties’ positions. In fact, the standard deviations of perceived party positions go down, overall, and most notably for the Conservative party (Figure 3.4). Still the dispersion of Conservative’s positions remains the highest of all three parties. This may be due to the fact that it is the party that moves the most.  For all three parties, supporter positions remain fairly stable—more so than party positions—and only minor differences exist between voters and identifiers (Figure 3.3). The slightly downward sloping line for the electorate mean indicates declining support for measures    67 to address unemployment. The gap between Labour and Conservative supporters closes minimally. The make-up of different supporters groups is also fairly stable (Figure 3.4). The SDs of supporters’ positions either go down (Conservative and Labour) or stay at the same level (Lib-Dems). However, there are differences between parties. Conservative supporters are a more heterogeneous group in terms of their position on unemployment than supporters of the other parties. Taken together, there is no evidence of parties leading depolarization, or voters unsorting in response—at least not on the issue of unemployment.  3.2.2 Welfare services On the second economic dimension—welfare services—respondents were asked to position themselves and the parties between “put up taxes a lot and spend much more on health and social services” and “cut taxes a lot and spend much less in health and social services.” The format of the question was the same as the previous, and I recoded the answers so that higher values (closer to 1) mean more support for welfare services.  On this dimension, it is less obvious where the initiative for depolarization comes from (Figure 3.5). Conservatives follow their supporters towards the centre, while Labour appears to be leading in this direction. Regardless, both parties more or less evenly share the responsibility for party system polarization. The Left-Right ordering is also less clear. Conservative party and supporters are below the electorate mean the entire time. Labour (party and supporters) start off above the electorate mean, but the party ends up just below after 1997. At this point, Labour and Liberal Democrats become almost indistinguishable.  Similar to the previous issue, the SDs of perceived party positions go down for Labour and Conservatives, and stay the same for Liberal-Democrats (Figure 3.6). This suggests that the clarity of positions does not decrease as a function of depolarization. Again, we see differences    68 across parties. SDs are highest for the Conservative party. Supporters of Conservative and Labour also show a very slight movement towards the centre, closing the gap between them (Figure 3.5). However, supporter groups do not become more heterogeneous. Their SDs remain stable (Conservatives and Liberal-Democrats) or go down (Figure 3.6). Even though there appears to be some leading by the Labour party, we see no convincing evidence for unsorting on this issue.  3.2.3 Redistribution The third economic issue—income redistribution—shares close resemblance with the previous two. Respondents were asked to take positions between “the government should make much greater efforts to make people’s incomes more equal” (coded 0) and “the government should be much less concerned about how equal people’s incomes are” (coded 1).   Again, who leads and who follows is not obvious (Figure 3.7). The Conservative party appears to be following its supporters towards a more centrist position. Labour, on the other hand, leads the depolarization trend after 1996. Similar to the previous issue, both parties contribute to party system polarization by moving towards the centre. All three parties stay in the typical Left-Right ordering through the entire period: Conservatives to the right of (below) the median voter, Labour to the left (above), and Lib-Dems in the centre. The Labour party comes close to crossing the line for the electorate mean after 1999.  Shortly before, the dispersion of the Labour party’s position begins to widen (Figure 3.8), suggesting less coherent signals about the party’s stance on the issue. Likewise, the SDs for the Conservative party increases after 1996, when the party starts moving towards the centre. So, on this issue we see the expected increased confusion about the parties’ positions, as a result of party depolarization.     69 Conservative supporters’ positions remain remarkably stable, while Labour’s move towards the centre. As a result, the gap between both groups closes over time. However, the change is much smaller than for parties. The dispersion of Conservative and Liberal-Democrat voters becomes narrower, whereas Labour supporters become a more heterogeneous group. However, the SD for Conservative supporters is highest, suggesting that this is the most heterogeneous group, in comparison. Taken together, this is evidence of unsorting for Labour voters in response to the party leading the depolarization movement.  3.2.4 Nationalization British voters were also asked to locate themselves and all parties on a scale with the following endpoints: “the government should nationalize many more private companies” (coded 1) and “the government should sell off many more nationalized industries” (coded 0).16   On this issue, the parties are clearly distinct, with Conservatives being least in favour of nationalization and clearly below the electorate mean, Labour being most supportive and below, and Lib-Dems occupying a more or less neutral position in the centre (Figure 3.9). Depolarization trends in terms of closing gaps between Conservative and Labour are weak and a function of Labour moving towards a more moderate position.  Especially the Conservative party’s position is much more stable than on the previous issues. The party occupies a more extreme position than its median supporter, and the gap remains substantial throughout the entire period. The Labour party also behaves differently than on the previous issues. Here, it follows its supporters towards the centre. Perhaps because they relatively stable, party positions become clearer to survey respondents, which is reflected in decreasing SDs for all parties (Figure 3.10).                                                   16 This question was only included until 1997.    70 Party supporters move slightly to the centre, closing the distance between Conservative and Labour (Figure 3.9). However, voter groups become a little more homogeneous over time (Figure 3.10). Similar to previous issues, the group of Conservative supporters is the most heterogeneous. In sum, there is little evidence of party depolarization and no evidence of unsorting within the electorate. Labour, the only party that moves, follows its supporters toward the centre. 3.2.5 European unification Lastly, British respondents were asked to place themselves and the parties on the issue of European unification. Here, the question prompted them to pick a location between “the government should do all it can to protect its independence from the European Community” (coded 0) and “the government should do all it can to unite fully with the European Community” (coded 1). Unlike the others, this issue is not part of the traditional Left-Right dimension. Do we see different patterns than for the other issues?  Yes, we do. First, the parties do not converge on this issue; instead, they polarize (Figure 3.11). Second, all three parties start off above the electorate mean (more supportive of European unification). Over time, the Conservative party moves to a less supportive position, while Labour becomes slightly more supportive. At the same time, all voters become less enthusiastic about European unification. Yet again, there is no unified dynamic of leading and following. The Conservative party follows its voters towards the centre, while Labour leads towards a more extreme position and away from its supporters. Voters seem to notice the clarification of party positions: dispersions shrink throughout this period (Figure 3.12).   Although Conservative and Labour move in the same direction—towards less supportive positions—the gap between them widens over time. This is the results of Conservative    71 supporters moving more quickly Figure 3.11). Voters and identifiers of all three parties become more internally homogenous groups over time (Figure 3.12). This is evidence for partisan sorting. What is more, it confirms the operation of the same mechanism as in the US: party system polarization leads to voter sorting—conditional on parties leading. 3.3 Results from Germany (1998–2013) I track the corresponding developments for Germany’s five main parties between 1998 and 2013. The 1998 election marked the beginning of a new era. After a sixteen-year tenure—by far the longest of any democratically elected German chancellor to date—Helmut Kohl’s CDU/FDP government was voted out of office in 1998. For the first time in a federal election, the entire government—rather than just one coalition partner—was replaced. This was made possible by the rise of the Green party. Since its first entry into parliament in 1983 it had established itself as a potential coalition partner for the Social Democratic Party (SPD). Before, coalitions between one of the two major parties (SPD or CDU) and the Free Democrats (FDP) as the junior partner were the norm. The newly elected government was a “red-green” coalition of SPD and Greens under the leadership of Gerhard Schroeder as German Chancellor. For the Green party, it was the first participation in a federal government. Noteworthy also is the rise of the Party of Democratic Socialism (PDS), later re-named The Left, the legal successor of the former governing party of the German Democratic Republic (GDR). After being seen as a party of the “East,” the PDS surpassed the critical five-percent electoral threshold for the first time since German reunification and gained representation and full parliamentary status in the Bundestag. Four years later, in 2002, the Schroeder government was re-elected and the red-green coalition continued for another three years. During these years, the coalition implemented the highly divisive reform package knows as “Agenda 2010.” Aimed at improving the economy, it    72 included the “Hartz reform”—a series of measures providing significant cutbacks in social welfare benefits for long-term unemployed persons. For a core segment of SPD voters, this was a slap in the face, provoking weekly protest marches (so-called Monday demonstrations), and costing the SPD a series of state elections—and ultimately the 2005 federal election.  2005 can be considered a historic election for at least two reasons. First, Angela Merkel (CDU) emerged as the first-ever female German chancellor. Second, it brought about a (rare) grand coalition of Christian Democrats (CDU/CSU) and Social Democrats (SPD), with the former being the stronger partner. While coalition governments have been the norm in Germany since the beginning, grand coalitions between the two major parties (Volksparteien) CDU/CSU and SPD have been historically rare events. In fact, until 2005, only one grand coalition formed the German government in the post-war era, from 1966 to 1969 under chancellor Kiesinger. The 2009 election put an end to the grand coalition and replaced it with a more traditional centre-right coalition of CDU/CSU and FDP, again under Merkel’s helm. However, by 2013 the support for the smaller coalition partner had plummeted and the FDP’s vote share dropped below the five-percent electoral threshold, leaving the party unrepresented in the Bundestag for the first time since 1949. Compared to 2009, CDU/CSU and SPD were both able to increase their vote share and, as a result, the 2013 elections produced another grand coalition of CDU/CSU and SPD and confirmed Merkel as German chancellor for the third consecutive term. Despite their increasing frequency, grand coalitions are still more of a “Zweckbuendnis” (alliance of convenience) than anyone’s first choice. The two big parties SPD and CDU have lost their strong hold on the electorate. On the left, The Left has emerged, splitting left voters between SPD and The Left. On the right, the Alternative fuer Deutschland (AfD) emerged in a vacuum left by the other parties, and serves as a venue for alienated and disgruntled voters.    73 All five Bundestag elections between 1998 and 2013 were carried out under the same mixed-member proportional representation (MMP) electoral system, while the party system remained relatively stable. The same five parties—Christian Democrats (CDU/CSU), Social Democrats (SPD), Free Democrats/ Libertarians (FDP), Green party (Greens) and Socialists (Left, formerly PDS)—competed in all five national elections. Four of the five parties were members of governing coalitions at some point between 1998 and 2013; only the PDS/Left party remained in the opposition at all times.  The data for the German case come from the German Longitudinal Election Study (GLES). Survey respondents of were asked to place themselves and the five parties described above on a number of policy scales. I choose three dimensions that were includes in three consecutive waves: Left-Right, European unification, and nuclear energy. For the analysis, I use the same types of graphs as above for the British case.  3.3.1 Left-Right Starting with the general Left-Right dimensions, I track the positions of parties and their supporters throughout the 1998–2013 period (Figure 3.13).17 As expected, CDU and FDP are to the right of the electorate mean, SPD, Greens and Left to the left. On the right, the CDU is the most extreme party (located between 0.6 and 0.7), on the left, the Left Party occupies the most extreme position (located between 0.1 and 0.2). Having a major party on the one end, and a minor party at the other creates an asymmetry, also leaving the right flank open to far-right parties. As expected, there is very little movement by parties or voters over time. The Left Party shows the only big movement, further toward the extreme. SPD and Greens move slightly closer                                                  17 For voters, Left-Right orientations are also available for 1994 and included in the graph.    74 to the centre. Regarding the clarity of party positions, the distributions are stable over time with only minor variation across parties. The only exception is the Left. In early years, its position on Left-Right is less clear than for other parties, but becomes clearer over time (Figure 3.14).  Party voters and identifiers are almost indistinguishable in their positions—especially for the “old” parties—CDU, SPD and FDP. Green and Left identifiers are more extreme than voters for the respective parties and slightly closer to the party. Overall, the dispersion of different voter groups does not vary markedly between parties or over time (Figure 3.14). Only Green party identifiers appear slightly more homogenous than other groups of party identifiers. As far as the distances between parties and their supporters are concerned, noticeable gaps exist only for CDU and Left—the two most extreme parties. Both parties are more extreme than their mean supporters.  3.3.2 Nuclear power Survey respondents were also asked to locate themselves and the parties on specific policy issues. One is nuclear power, where the continuum ranges from “immediate shut-down” (0) to “further expansion” (1) of nuclear power plants.18 Figure 3.15 shows the typical Left-Right division: CDU and FDP are more supportive of nuclear energy than the mean voter, all other parties less so. Overall, party positions look remarkably stable between 1998 and 2005. Thereafter, the range of positions narrows slightly, as CDU and Greens move towards the centre and towards their own supporters. That is, we see a slight depolarization trend, which also includes voters. Party-voter distances are largest for the CDU and Greens—the two most extreme                                                  18 Party positions are not available for 2013, as this item has not been included in the 2013 survey.    75 parties. Both parties occupy more extreme positions than their supporters. This is the same pattern as for Left-Right.  The distributions of perceived party positions show some fluctuations over time, but no clear trend (Figure 3.16). Not surprisingly, the position of the Green party is the clearest, especially between 1998 and 2005, when the party was in government. There appears to be confusion about the party’s standpoint in 2009. This coincides with the party moving toward the centre. As expected, Green party supporters are the most homogenous group on this issue (Figure 3.16). Despite some fluctuation, especially for the SPD, the graphs indicate no trend. Only Left party supporters appear to be becoming more heterogeneous over time. 3.3.3 European unification Another specific policy issue on which respondents were asked to place themselves and the parties is European unification—clearly a topic with increasing significance for voters and parties. Here the continuum ranges from “European unification already goes much too far” (0) to “promote European unification in order to establish a joint government” (1).  The overall picture shows a much more fluid situation for parties and voters than on the Left-Right dimension and nuclear power (Figure 3.17). In 1998, all parties start above the electorate mean, expressing clear support for European unification. Only the Left party is an exception. Initially, CDU and FDP are the biggest supporters of the European idea (located between 0.7 and 0.8). They are also the two parties showing the largest movements over time, towards the centre and less supportive positions. By taking on more centrist positions, both parties follow their supporters. SPD and Greens remain fairly stable in their support for Europe. The Left party is stable in their opposition to European integration.    76 Overall, and contrary to the general European trend described in the previous chapter, the range of positions occupied by German parties narrows over time, between 1998 and 2009, suggesting depolarization. However, this trend is not unanimous, as it is mainly driven by CDU and FDP.  Looking more specifically at party-voter pairs, the panels for CDU and SPD show notable gaps. The CDU’s distance to its median party supporter is extremely large in 1998 but closes over time—at least until 2005—when it starts opening up again. The gap between the SPD and its supporters closes slowly over time, as each moves toward the other. With four of the five parties located above the electorate mean, this space is rather crowded. The Left is the only party representing voters below the electorate mean on this issue. This appears to cause some confusion for voters as to where parties stand. The distributions of parties’ positions are much wider than for Left-Right, reflecting a lack of clarity (Figure 3.18). Moreover, standard deviations are rising for all parties.  Similarly, the positions of party supporters are also more dispersed. On European unification voters and identifiers of a given party are more heterogeneous groups than on Left-Right. They become even more so over time, providing evidence for unsorting (Figure 3.18). This is interesting and somewhat surprising. German parties depolarize on European integration by following their voters. Despite the lack of leading, there is evidence for unsorting. 3.4 Discussion and conclusion The analysis in this chapter has addressed this dissertation’s central question of leading and following. At this level of aggregation, the longitudinal setup allows me to draw causal inferences about whether parties lead or follow. The direction of movement by parties over time provides the answer. Parties lead if they move away from the median supporter; they follow if    77 they move toward it. This is true even if movement is recorded by survey respondents’ perceptions. Individuals’ perceptions may be biased (see next chapter) but averages seem broadly correct, and there is no reason to see movement in them as some kind of projection.   The extent of leading and following varies across party systems, parties, and issues. My results for the UK confirm existing studies that find a depolarization trend between the two dominant parties on issues that fall into the general Left-Right dimension (Adams et al. 2012; see also, Milazzo et al. 2012). In contrast, parties polarized on an issue that has not been included in previous studies: European unification.  However, parties do not always lead these movements. The predominant dynamic for the Conservative party is following. Whenever the party moves, it moves toward its supporters. Labour, on the other hand, tends to lead. This tendency to lead is most visible for welfare, redistribution, and European unification. The party follows its supporters on nationalization and moves with them on unemployment. I do not find broad evidence of a British electorate that is unsorting in response to party system depolarization. What I discover instead is one instance of unsorting among Labour supporters on redistribution. On the other issues, unemployment, welfare and nationalization, the level of heterogeneity of Labour supporters does not increase. Likewise, the groups of Conservative and Liberal-Democrat supporters do not change in their composition or orientation.19 What do these findings mean with regard to the general causal mechanism linking party system structure to party-voter dynamics? One explanation for why I do not find an unsorting                                                  19 These findings seem to contradict the conclusions drawn by Adams et al. (2012), who attest a sharp decline in partisan sorting. However, they can be explained by the choice of measurement.     78 trend across the board could be that depolarization is not led by parties—at least not universally. For the Conservative party moving towards the centre means following. Thus, the party decreases the distance to its supporters and becomes more representative of the median party voter. This should satisfy proximity-oriented voters. At the same time, the party always stays on the same side of the electorate mean and occupies a more extreme position than the mean party supporter. This should satisfy directional voters. Thus, Conservative voters have no motivation to unsort, and to change their opinions or party preference, despite a significant shift in policy.  The situation is different for Labour. Labour leads depolarization on welfare and redistribution, moving the party away from its supporters. The consequences are larger party-voter distances and, by moving towards the centre, elimination of a directional option to the left. Therefore, both proximity and directional voters have reasons for reorientation. As it happens, redistribution is the one dimension on which we, in fact, see an unsorting of Labour voters. On the other issues, Labour either moves with its supporters or follows them, and we see no unsorting. Thus, it appears that the causal mechanism has limited application. A condition that appears to be necessary, but not sufficient, is that parties lead. The analysis of European unification reveals the same pattern as for the US: party system polarization leads to sorting. This trend is party-driven to the extent that Labour moves away from its supporters, toward a more supportive position. The consequence is a widening gap between Labour and Conservative, as the latter moves in the opposite direction, along with its supporters. Decreasing standard deviations of party positions suggest that this clarification of positions helps voters see where the parties stand. Moreover, it helps them align their party preference with their issue position, or vice versa. On this issue, voters sort into more homogenous groups with smaller standard deviations of positions.    79 The British case confirms the general European trend found in the previous chapter—at least for the party system: depolarization on classic Left-Right questions and polarization on European unification. Furthermore, it demonstrates how the level of party system polarization affects party-voter relations beyond the US. So far, however, I have focused on a party system that has undergone longitudinal changes to test my argument. Does it apply more generally, independent of polarization and depolarization?  Germany serves as a test case to identify if the same mechanisms operate in a more stable five-party system. On the question of leading and following, there is less party movement overall. Especially the Left-Right dimension sees stable party positions over time. The Left party is the only one moving on this dimension—away from its supporters towards a more extreme position. Similarly, there is little party movement on nuclear power. The only exception is the Green party, which follows its supporters to a more moderate position. The only dimension on which we see party system change is European unification. Here, the German party system depolarizes. Responsible for depolarization are CDU and FDP who follow their median supporters to more central locations. European unification is also the only issue on which the electorate unsorts in response to depolarization.  For the other issues I do not find sorting or unsorting processes over times, which is not surprising, given the lack of party movement. However, there appears to be a link between the perceived clarity of options and the alignment of party preference and issue position. On Left-Right and nuclear power, the parties are distributed across a wide range of the 0 to1 scale and fall into the expected Left-right division: CDU and FDP on the right (above the electorate mean) and SPD, Greens and Left on the left (below the electorate mean). Comparing the standard deviations of party positions across issues reveals that voters place the parties with greater precision on    80 Left-Right than on the other issues. Similarly, the distribution of voter positions is tightest on Left-Right. This suggests that the German electorate is sorted along the general Left-Right dimension, but has not aligned all issue positions with party choice. Especially on European integration party-voter links are looser and loosening.  On average, across all issues and parties, the distances between parties and their mean supporters are smaller in Germany than in the UK. This is a function of more options. At the same time, it may suggest that proximity is more important than direction in party systems with a greater number of parties, as long as the party system as a whole reflects the distribution of voter positions.      81 Figure 3.1 Partisan sorting Elites polarize   Party/ideology mapping is clarified   Ordinary voters see the clarification   Voters sort Notes: This Figure illustrates the causal mechanism behind sorting. It is a reproduction of Figure 2.1 in Levendusky (2009, 13).       82 Figure 3.2 The difference between sorting and polarization  Example of an unsorted electorate L  L  L M  M  M C  C  C L  L  C  C M  M  M  M C  C  L  L C  C  C M  M  M L  L  L Democrats Independents Republicans    Example of a sorted electorate L  L  L M  M  M C  C  C L  L  L  L M  M  M  M C  C  C  C L  C  C M  M  M C  L  L Democrats Independents Republicans  Example of a polarized electorate L  L  L  L  L  C  C  C  C  C L  L  L  L  L  C  C  C  C  C L  L  L  L  L  C  C  C  C  C Democrats Independents Republicans Notes: This Figure illustrates the differences between unsorted, sorted, and polarized electorates using the US as an example. It is based on Table 1.1 in Levendusky (2009: 5).  L = liberal, M = moderate, C = conservative       83 Figure 3.3 Unemployment: Comparing positions of British parties and voters Reduce unemployment (0-1)     Source: BES Panel Studies Notes: This Figure reports where survey respondents place themselves and the three main British parties on unemployment between 1992 and 2001. The distance between the lines reflects the level of congruence between parties and their supporters.     84 Figure 3.4 Unemployment: Comparing distributions (SDs) of British parties and voters Reduce unemployment (0-1)     Source: BES Panel Studies Notes: This Figure reports the dispersion (SD) of positions on unemployment for parties, party voters, and party identifiers between 1992 and 2001. Smaller values reflect tighter, more homogenous distributions.      85 Figure 3.5 Welfare: Comparing positions of British parties and voters Welfare services (0-1)     Source: BES Panel Studies Notes: This Figure reports where survey respondents place themselves and the three main British parties on welfare between 1992 and 2001. The distance between the lines reflects the level of congruence between parties and their supporters.     86 Figure 3.6 Welfare: Comparing distributions (SDs) of British parties and voters Welfare services (0-1)     Source: BES Panel Studies Notes: This Figure reports the dispersion (SD) of positions on welfare for parties, party voters, and party identifiers between 1992 and 2001. Smaller values reflect tighter, more homogenous distributions.      87 Figure 3.7 Redistribution: Comparing positions of British parties and voters Redistribution (0-1)     Source: BES Panel Studies Notes: This Figure reports where survey respondents place themselves and the three main British parties on redistribution between 1992 and 2001. The distance between the lines reflects the level of congruence between parties and their supporters.     88 Figure 3.8 Redistribution: Comparing distributions (SDs) of British parties and voters Redistribution (0-1)     Source: BES Panel Studies Notes: This Figure reports the dispersion (SD) of positions on redistribution for parties, party voters, and party identifiers between 1992 and 2001. Smaller values reflect tighter, more homogenous distributions.      89 Figure 3.9 Nationalization: Comparing positions of British parties and voters Nationalization (0-1)     Source: BES Panel Studies Notes: This Figure reports where survey respondents place themselves and the three main British parties on nationalization between 1992 and 2001. The distance between the lines reflects the level of congruence between parties and their supporters.     90 Figure 3.10 Nationalization: Comparing distributions (SDs) of British parties and voters Nationalization (0-1)     Source: BES Panel Studies Notes: This Figure reports the dispersion (SD) of positions on nationalization for parties, party voters, and party identifiers between 1992 and 2001. Smaller values reflect tighter, more homogenous distributions.      91 Figure 3.11 European unification: Comparing positions of British parties and voters European unification (0-1)     Source: BES Panel Studies Notes: This Figure reports where survey respondents place themselves and the three main British parties on European unification between 1992 and 2001. The distance between the lines reflects the level of congruence between parties and their supporters.     92 Figure 3.12 European unification: Comparing distributions (SDs) of British parties and voters European unification (0-1)     Source: BES Panel Studies Notes: This Figure reports the dispersion (SD) of positions on European unification for parties, party voters, and party identifiers between 1992 and 2001. Smaller values reflect tighter, more homogenous distributions.      93 Figure 3.13 Left-Right: Comparing positions of German parties and voters Left-Right (0-1)        Source: GLES Panel Datasets (ZA4301, ZA4662, ZA5320, ZA5321, ZA5322) Notes: This Figure reports where survey respondents place themselves and the five main German parties on Left-Right between 1994 and 2013. The distance between the lines reflects the level of congruence between parties and their supporters.       94 Figure 3.14 Left-Right: Comparing distributions (SDs) of German parties and voters Left-Right (0-1)     Source: GLES Panel Datasets (ZA4301, ZA4662, ZA5320, ZA5321, ZA5322) Notes: This Figure reports the dispersion (SD) of positions on Left-Right for German parties, party voters, and party identifiers between 1994 and 2013. Smaller values reflect tighter, more homogenous distributions.      95 Figure 3.15 Nuclear energy: Comparing positions of German parties and voters Nuclear energy (0-1)        Source: GLES Panel Datasets (ZA4301, ZA4662, ZA5320, ZA5321, ZA5322) Notes: This Figure reports where survey respondents place themselves and the five main German parties on nuclear energy between 1998 and 2013. The distance between the lines reflects the level of congruence between parties and their supporters.       96 Figure 3.16 Nuclear energy: Comparing distributions (SDs) of German parties and voters Nuclear energy (0-1)     Source: GLES Panel Datasets (ZA4301, ZA4662, ZA5320, ZA5321, ZA5322) Notes: This Figure reports the dispersion (SD) of positions on nuclear energy for German parties, party voters, and party identifiers between 1998 and 2013. Smaller values reflect tighter, more homogenous distributions.      97 Figure 3.17 European unification: Comparing positions of German parties and voters European unification (0-1)        Source: GLES Panel Datasets (ZA4301, ZA4662, ZA5320, ZA5321, ZA5322) Notes: This Figure reports where survey respondents place themselves and the five main German parties on European unification between 1998 and 2009. The distance between the lines reflects the level of congruence between parties and their supporters.       98 Figure 3.18 European unification: Comparing distributions (SDs) of German parties and voters European unification (0-1)     Source: GLES Panel Datasets (ZA4301, ZA4662, ZA5320, ZA5321, ZA5322) Notes: This Figure reports the dispersion (SD) of positions on European unification for German parties, party voters, and party identifiers between 1998 and 2009. Smaller values reflect tighter, more homogenous distributions.    99 Chapter 4: Voters: Leading and following This chapter investigates party-voter relations from the perspective of individual voters in multiparty systems. The analytical focus is on cognitive processes of opinion and preference formation and driven by the question: Do policy opinions inform party preferences, or vice versa? Put differently, do voters lead, or do they follow?  In addressing this question, I investigate the triad of voters’ positions on issues, voters’ perceptions of party positions, and voters’ party preferences to identify the dynamics connecting all three. I focus on three processes: policy oriented party evaluation or issue voting—considered a form of leading—and persuasion and projection—both considered forms of following.20  I perform three types of analyses to demonstrate how each of these processes operates, starting with a test of leading that is based on policy oriented evaluation. It accounts for the possibility of reciprocal effects between policy distance and party evaluation. To the extent that the effects of distance on evaluation hold up when I control for effects in the opposite direction, voters lead. However, I also expect to find effects from party evaluation to policy distance. These could be the result of persuasion, projection, or both. To better understand this side of the ledger—if voters move their own opinions or their perceptions of parties’ positions—I perform separate tests for persuasion and projection.  Methodologically, identifying the causal dynamics between issue positions and party preferences is no easy task and demands suitable data and advanced estimation techniques.                                                  20 Since, in this chapter, the focus is on voters’ opinion and preference formation, the terms leading and following have different meanings than in Chapter 3. There, the notions are defined by aggregate shifts in direction and by (a) whether parties move first and (b) whether the move is toward or away from their followers.      100 Particularly, testing my argument that the relationship between parties and voters is reciprocal, featuring a mix of all three processes mentioned above, requires models that can estimate reciprocal effects. To overcome these challenges, the analysis employs panel data that include multiple observations of individual respondents and span multiple elections. Appropriate multi-wave panel data are available for the UK and Germany. Combined with correctly specified models, panel data offer the best solution to causal inference with observational data (Finkel 1995). I estimate a series of dynamic panel models using structural equation modeling (SEM) techniques.  The remainder of this chapter is structured as follows. The first part provides definitions of leading and following and illustrates the cognitive processes that shape voters’ opinion and preference formation. As methodologically, this chapter is the most sophisticated of the three, the second part is devoted to methodological challenges, estimation techniques and model specifications. The third and fourth sections present the results of the analysis, separately for the UK and Germany. The last two sections discuss the findings and conclude.   4.1 Dynamics of opinion and preference formation 4.1.1 Voters as leaders Classic accounts of representative democracy and models of issue voting, which dominate the European literature, approach party-voter relations from a positional perspective. They posit that voter and party positions combine to determine how voters evaluate a party and make their voting decisions. This evaluation process requires a comparison of their own position with the position of each party. Spatial proximity models predict that a voter derives the highest utility from the party (or candidate) closest to her on some ideological or policy continuum (e.g., Enelow and Hinich 1984; Merrill and Grofman 1999). The assumption is that voters have fixed    101 policy preferences and routinely engage in policy oriented evaluation of parties and candidates (Brody and Page 1972: 457): the distance between a voter and a party determines how a voter evaluates a party (Equation 4.1). Similarly, a voter updates her party preferences when the distance to a party changes.  Equation 4.1 Policy oriented evaluation (issue voting) Party evaluation = f (|Party position – Voter position|) Figure 4.1 presents a graphical illustration of this process. The bottom vertices are voter position and perceived party position on a policy issue, and the top vertex party evaluation. Policy oriented party evaluation (or issue voting) is represented by the bold grey arrow. I consider this leading.   4.1.2 Voters as followers American studies of voting behaviour have found little evidence of issue voting and stress the importance of two rationalization processes when citizens make decisions (represented by the diagonal dashed black arrows in Figure 4.1) (e.g., Berelson et al. 1954; Page and Brody 1972; Markus and Converse 1979; Bartels 2002; Lenz 2012). The first one is persuasion. For the American context, Brody and Page (1972: 457) define the concept as follows:  “If perceptions and evaluations were invariant and the respondent’s own position yielded to or formed around the position he believed was held by the candidate he favoured (or intended to vote for), we would say that the voter has been ‘persuaded’ by the candidate.”  Thus, persuasion suggests that perceived party position and party evaluation are fixed, and voter position is the moving part. Simply put, voter position is the function of an interaction between perceived party position and party evaluation (Equation 4.2). A persuaded voter moves her own positions toward a party she likes, and away from a party she dislikes (Markus and Converse 1979).    102 Equation 4.2 Persuasion Voter position = f (Party position * Party evaluation)  The second rationalization process is projection, where the perceived position of the candidate is a function of the voter’s own position and her evaluation of the candidate (Equation 4.3).  “Under this model, a voter could see a candidate as close to himself on an issue because he otherwise felt positive about the candidate, and he could see other candidates as distant from him on the issue(s) because he felt negative about them for other reasons. In other words, a voter would ‘project’ his own (or a nearby) stand onto candidates he favoured” (Brody and Page 1972: 457).  Equation 4.3 Projection Party position = f (Voter position * Party evaluation) I consider both rationalization processes forms of following, because policy positions are conditional on party pre-commitment. That is, either one’s own views or one’s perceptions are the results of party evaluation. Moreover, both have behavioural consequences. A persuaded voter changes her opinions. A voter who projects her own views onto a party she prefers for other reasons does not make an objective or rational voting decision. However, she might think that she does, because to her the party she prefers looks to be the closest to her own position. A perceptual screen is the mechanism behind projection. It distorts voters’ perceptions of parties’ positions and reinforces pre-existing party loyalties (Campbell et al. 1960).   4.1.3 Reciprocal effects Most of the comparative literature reduces complex party-voter relations to a principal-agent relationship, where voters lead and parties follow. American scholars have less faith in voters’ cognitive abilities or motivations to engage in policy oriented evaluation and routinely find evidence of following (e.g., Bartels 2002; Brady and Johnston 2006; Lenz 2009; 2012). Drawing from both traditions, I argue that European accounts of representation and issue voting, which    103 assume unidirectional leading dynamics, are incomplete. Instead, a mix of leading and following, including all three processes described above (and depicted in Figure 4.1), defines party-voter dynamics in Europe. In more technical terms, I propose that the causal effects between of voters’ positions on issues, voters’ perceptions of party positions, and voters’ party preferences are reciprocal.  European research that explicitly tests the causal direction of the effects between policy positions and party evaluations is rare. Most studies interpret the correspondence between policy distance and party evaluation (or vote) choice as evidence of leading. This is problematic, because what looks like policy oriented evaluation or issue voting could also be the result of persuasion, projection, or both. The outcomes of all three processes are “observationally equivalent” (Lenz 2012, Chapter 1) at any given point in time. Consequently, we know little about the causal dynamics of opinion and preference formation in Europe.  Even in the US, studies that test the operation of reciprocal causal effects are the exception. Those who have looked into the reciprocal nature of party-voter relations tend to focus on the link between party identification and voter positions (Carsey and Layman 2006; Dancey and Goren 2010; Milazzo, Adams and Green 2012), or party preference and performance evaluation (Evans and Pickup 2010; Chzhen, Evans and Pickup 2014). Studies that incorporate party positions or policy distances are even less common (see, Evans and Andersen 2004 for a rare example). Regardless, the shared finding is that the causal arrow points in both directions, often with stronger effects from party to voter positions.21                                                  21 Milazzo et al. (2012) is an exception. They find no persuasion effects from party to Left-Right ideology.    104 4.2 Methods: Challenges and estimation techniques Methodologically, tests of reciprocal dynamics are demanding, and the analysis in this chapter faces a number of difficult challenges. Page and Brody (1972: 981) illustrate these for issue voting, my test of leading.  “It is quite difficult to assess precisely how much effect policy preferences have upon voting. In order to do so, it is probably necessary to calculate citizens’ perceptions of similarity or dissimilarity between their own opinions and candidates’ stands on an issue; to ascertain the relationship between these ‘perceived issue differences’ and the vote; to estimate the extent of issue-oriented evaluation, disentangling that process from the confounding effects of rationalization (perceiving a favored candidate’s position as being the same as one’s own, or shifting one’s opinion to conform with the position of a favored candidate); and finally, to calculate net impact on the electoral outcome in terms of the average amount of issue difference from the candidate which is perceived by the population, and the average effect of a single unit of difference. It is probably correct to say, therefore, that the impact of issue voting has never been adequately measured.”  In other words, I need to overcome three important hurdles to identify the extent to which voters lead. First, on a given issue, I need to calculate the perceived distance between an individual voter and each party. This requires data that include respondents’ own positions and their estimates of all party positions, ideally on multiple issue dimensions. Second, I need to estimate the causal effect of perceived policy distance on vote (for a test of issue voting) or party evaluation (for a more general test of policy oriented evaluation).22 To identify the causal direction, it is insufficient to simply measure the association by regressing vote or party evaluation on policy distance, because it could also be the result of reverse or reciprocal causation. Thus, I need to specifically estimate the strength and direction of effects running in                                                  22 Issue voting and policy-oriented evaluation are based on the same cognitive process. The main difference is in the choice of the dependent variable. Vote choice, the dependent variable for issue voting, forces voters to select one option. A test of issue voting is, therefore, limited to voters of a given party. A test of policy-oriented evaluation uses party evaluation as the dependent variable and allows for an analysis that includes all respondents who have a valid answer.     105 both directions. Third, to estimate the effect of policy distance on party evaluation, I need to separate this effect from other possible determinants of party evaluation.  Panel data including multiple observations at the level of individual voters can help with this. Two features of panel data make them suitable for drawing causal inferences: The ability (1) to determine the direction of causal relationships, and (2) to control for unobserved, time-invariant confounders, (see, e.g., Finkel 1995; 2008; Allison 2009).  However, panel data are not, per se, solutions to causal inference. They require correctly specified models. In my case, I need models that can handle reciprocal effects between policy distance and party evaluation. Reciprocal effects mean that two variables influence one another, possibly with feedback. Cross-lagged panel models are appropriate if X and Y have lagged effects on each other. That is if Y is caused by prior values of X (and vice versa). In cross-lagged models, the temporal order is used to make inferences about the causal order. However, if X and Y (also) have synchronous effects on each other, nonrecursive models that can separate the direction and strength of both effects are required (Finkel 1995, Chapter 3). I suspect that leading and following are dynamic processes based on both past and current values of X and Y. Especially if the lags between two panel waves are long, as in most election studies, past values are, most likely, insufficient to capture the full reciprocity between policy positions and party evaluations. I therefore need non-recursive models that can handle cross-lagged and synchronous effects. Furthermore, these models must be able to separate the causal effects between both variables from other potential factors to rule out spuriousness. To be sure that X and Y cause each other—and that the correlation between both is not caused by a third unobserved variable—I need to control for unobserved heterogeneity. Dynamic panel models that include a version of    106 the lagged dependent variable and so-called “fixed effects” satisfy this condition (Finkel 1995, Chapter 5; Allison 2009).  Separate solutions are readily available, for each of these requirements. However, each depends on a set of assumptions, and finding models that combine both is no easy task. It gets worse once I factor in additional restrictions brought on by data availability. Generally speaking, panel models with more waves of data can be estimated with fewer restrictions. Yet, most panel studies have limited numbers of waves—in my cases, five waves for the UK and three for Germany.  To overcome these challenges, I draw from the two most prominent approaches to panel analysis: the econometric (OLS) approach, and the structural equation modeling (SEM) framework. The econometrics tradition takes a counterfactual approach to causal inference and focuses “most explicitly on the problem of unobservables in the causal system” (Finkel 2008: 477; see also, Baltagi 2005; Frees 2004; Hsiao 2002; Woolridge 2002). Economic panel models deal with the problem of unobserved heterogeneity by including so-called “fixed effects” methods, which use individual units as their own controls (see, e.g., Allison 2009). These models can hold everything else constant and, thus, allow researchers to draw causal inferences similar to those in experimental research. The structural equation modeling (SEM) framework approaches longitudinal panel analysis from a different angle. It is a method for the analysis of systems of equations that represent the causal linkages between variables. Here longitudinal processes are embedded within an overall “causal model”—a series of equations specifying how variables influence one another over time. More specifically, SEMs include a separate equation for each endogenous (dependent) variable at each panel wave of observations. Within this framework, the resulting    107 system of equations is estimated simultaneously—commonly through maximum likelihood (ML) or related methods. The SEM approach is, therefore, particularly useful for the estimation of models with reciprocal effects between variables (especially non-recursive models) and the identification of the causal direction and strength of effects (Finkel 2008).  Although it is less established in Political Science, I choose the SEM approach because of its advantages for estimating reciprocal causal relationships. Furthermore, within the SEM framework I can explore different time lags by including both cross-lagged and synchronous causal effects in the same models. This helps me to better understand the dynamic processes that characterize opinion formation in different electoral contexts and across a range of issues.  One weakness of SEM is that it does not traditionally account for unobservables. Yet, controlling for unobserved heterogeneity is key for determining causality (and ruling out spuriousness). It is possible to include insights from the econometrics tradition and estimate SEMs with fixed effects.23 This approach combines the best of both worlds, but it can be problematic. Vaisey and Miles (2017) urge researchers to use caution when including fixed effects in dynamic (three-wave) panel models. They demonstrate that it is extremely important to get the time lags right, otherwise these models can produce coefficients that are of opposite sign and half the magnitude of the true values. They also advise that “the problem with using lagged variables to establish temporal ordering is that the lags in our data rarely correspond to the lags                                                  23 Despite the ability to account for unobserved heterogeneity in SEMs, it does not seem to be common practice in Political Science. Recent publications in top tier journals suggest that SEMs without fixed effects are (still) acceptable in attitudinal research. For articles in the AJPS, see, e.g., Dancey and Goren (2010), Smith et al. (2016); for a JOP article, see, e.g. Milazzo et al. (2012).    108 present in real-world processes (Vaisey and Miles 2017, 24).24 Accordingly, I exercise caution when estimating dynamic panel models with fixed effects.  4.2.1 Modeling leading I model leading—or, more precisely, policy-oriented evaluation—as a process that is driven by past and present values of the two key variables: party evaluation (DV), and policy distance between a voter and a party on a specific issue (IV). To estimate this process, I use a dynamic panel model with cross-lagged and synchronous effects. To account for the presence of reciprocal dynamics—from distance to evaluation, but also from evaluation to distance—I include effects pointing in both directions. Thus, the model combines two equations: Equation 4.4 Modeling leading: Policy oriented evaluation Yt = α + β1Y(t-1) + β2 X(t) + β3 X(t-1) + U1 Equation 4.5 Modeling leading: Reverse effects Xt = α + β4 X(t-1) + β5 Y(t) + β6 Y(t-1) + U2 Suppose Equation 4.4 describes policy-oriented evaluation, then the dependent variable, Yt, is the score between 0 and 1 that a respondent assigns to a party on a likeability scale (party evaluation). This party evaluation score at time t is a function of how the same respondent evaluated the same party at a previous time t-1. This inclusion of the lagged dependent variable is theoretically important to model change. The lagged dependent (or endogenous) variable represents the stability coefficient. It also addresses one methodological challenge of panel data analysis: regression to the mean. Regression to the mean (also called regression effects) is “not always present in panel data, but it can be shown that a negative correlation between a variable’s initial value and                                                  24 Although their argument and proof only extend to three-wave models, Finkel (2015) finds that this problem can persist when four waves are available.    109 subsequent change can be expected whenever (a) the variable is not perfectly correlated over time and (b) its variance is relatively constant (Bohrnstedt 1969; Kessler and Greenberg 1981; Nesselroade, Stigler, and Baltes 1980). Under these circumstances, including Yt-1 in the regression model is a way of controlling for this phenomenon, and frames the analysis in the following fashion: Do the independent X variables influence changes in Y for fixed levels of Yt-1, that is, taking into account the negative effect of initial values of Y on subsequent change?” (Finkel 1995: 8-9). Party evaluation is further determined by X, the policy distance between a voter and a party at time t and t-1, also coded 0-1. I measure policy distance using the absolute linear subjective distance between a respondent and a party.25 Negative and statistically significant coefficients for β2 and β3 would suggest causal effects of policy distance on party evaluation in the hypothesized direction: leading. To ensure that the coefficients for β2 and β3 capture the causal effects of policy distance on party evaluation—and rule out that this correspondence is also shaped by persuasion and projection—I need to control for effects in the opposite direction. I therefore estimate Equation 4.5 in the same model, which describes the reverse dynamic: from party evaluation to policy distance. Here, policy distance (Xt) is a function of policy distance measured in the previous wave (Xt-1), and party evaluation previously (Yt-1) and currently (Yt). All variables are measured as in Equation 4.1 above. Negative and statistically significant coefficients for β5 and β6 would indicate the presence of reciprocal dynamics. Because both equations are estimated in the same model, a comparison of coefficient sizes can assess the causal predominance between both                                                  25 The absolute linear distance (or a linear loss function) has been shown to outperform the quadratic loss function in models of voting behaviour (Singh 2014).    110 directions. It is important to note that Equation 4.5 is not the estimation of persuasion or projection, but of some amalgam of the two. This is the case, because changing policy distances can be a function of changes in voter positions, changes in perceived party positions, or both. To identify the strength of each effect, I perform separate tests (see section on “Modeling following”). 4.2.1.1 The estimation model for the UK case For reasons of data availability, I estimate different models for the UK and Germany. Two multi-wave panel studies are available from the British Elections Study (BES). The interviews for these studies were conducted on an annual basis, between two national elections (1992-1997 and 1997-2001).26 Thus, most waves do not fall into an election year. I merge both datasets to examine the longest possible time period and draw more general conclusions.  To account for unobserved factors, I include fixed effects by first-differencing the data. That is, I subtract each variable from its previous value, thereby differencing out all stable, unmeasured factors not included in the model. The resulting first-differences (FD) model only uses within-respondent variation and discards all between-respondent variation. Unfortunately, first-differencing the data means losing one wave of data. I, therefore, estimate the four-wave structural equation model depicted in Figure 4.2.  The model is fully dynamic, including stability effects between all Xs and all Ys, as well as reciprocal cross-lagged and synchronous effects between X and Y (represented by solid arrows). All cross-lagged and synchronous effects are constrained to be equal over time, but                                                  26 Policy questions were not included in 1993. I therefore exclude this wave, leaving five waves per study.    111 stability effects are free to vary. That means, all “causal” effects between X and Y are the average effects across all waves. A synchronous covariation between disturbances between the time 2, time 3, and time 4 variables is included to capture omitted factors that may influence responses to both X and Y on one wave but not the next (Finkel 1995: 83). This is depicted in Figure 4.1 as the dashed link between ε1 and ε4, between ε2 and ε5, and between ε3 and ε6. In addition, I include over-time disturbances (between ε1 and ε2, ε2 and ε3, as well as between ε4 and ε4, and ε5 and ε6). The other dashed arrow between X1 and Y1 represents the covariance between the two exogenous variables in the model. All (error) covariances missing from the model are assumed to be 0.  This is a parsimonious model with only two variables. All other potential explanatory factors are excluded (or differenced out). I choose this setup to isolate the effect of policy positions on party evaluation. Realistic expectations are therefore for weak to moderate effects. This is also true for the stability effects in the model. Weak stability effects are a function of first-differencing the variables, and essentially removing all stabilizing factors.  I run this model separately for each party and issue, totalling fifteen models for the UK. To make the results comparable across parties and issues, I need to choose the model with the best average fit. Consequently, the model in Figure 4.2 presents a better fit for some parties and issue than others. All models report unstandardized maximum likelihood (ML) coefficients and regular standard errors. Because all variables are coded on the same scale (0-1 originally; -1/+1 after first-differencing), coefficient sizes can be compared to determine relative effect strength and causal predominance.    112 4.2.1.2 The estimation model for the German case For Germany, I pool different three-wave panel studies from the German Longitudinal Election Study (GLES). Each study spans three elections, and combined they cover the period from 1998 to 2013. Time lags between waves are three to four years long—one complete electoral cycle. Unfortunately, with only three waves of data available, a first-differences (FD) model with synchronous and cross-lagged reciprocal effects, identical to the UK model, is unidentified. I therefore need to choose a different strategy.   For reasons explained at length in the Appendix, I opt for a first-differences (FD) model with synchronous effects (Figure 4.3).27 Again, because I first-difference the variables I lose one wave, resulting in only two waves of data to run the model. For the model to be identified, I need to add more information. I therefore include the second lags of X and Y as instrumental variables (X1 and Y1).28   Otherwise, the model is as similar as possible to the UK model. It is fully dynamic with stability effects between all Xs and all Ys, as well as reciprocal synchronous effects between X and Y in the last wave (represented by solid arrows). Stability effects are free to vary over time. A synchronous covariation between disturbances between the time 2 variables (symbolized by the dashed link between ε1 and ε2) is included to capture omitted factors that may influence responses to both X and Y on one wave but not the next (Finkel 1995: 83). Covariance between the two exogenous variables and over-time disturbances (between ε1 and                                                  27 The model is a structural equation modeling (SEM) extension of the Anderson and Hsiao (1981; 1982) estimator proposed in Evans and Pickup (2010) and Chzhen, Evans and Pickup (2014). 28 For a detailed discussion of this approach, see Finkel (1995, Chapter 3), Evans and Pickup (2010), and Chzhen, Evans, and Pickup (2014).    113 ε3, and between ε2 and ε4) are also included and represented by dashed arrows. All (error) covariances missing from the model are assumed to be 0.  As for the UK, all coefficients are unstandardized ML estimates with regular standard errors in parentheses. Again, due to the parsimonious nature of the model, I expect weak to moderate coefficient sizes at best. Likewise, first-differences models “difference out” all stable factors (e.g. demographics), thereby eliminating a lot of explanatory power. 4.2.2 Modeling following The models above include effects between policy distance and party evaluation in both directions. Effects from policy distance to party evaluation capture the concept of policy-oriented evaluation. But how exactly can we interpret the reverse effects, from evaluation to distance? Distance changes can be due to (1) changes in voter positions (persuasion), (2) changes in perceived party positions (projection), or (3) a combination of both.   To better understand these effects, I follow Markus and Converse (1979) who propose empirical tests of persuasion and projection. The former captures “the idea that a voter’s issue orientations prior to the campaign may be modified as a function of the candidate’s issue stands, as the voter sees them. That is, the voter will move toward a favored candidate’s perceived position and/or away from that of a negatively evaluated candidate” (Markus and Converse 1979: 1058).  Equation 4.6 Modeling following: Persuasion Y(t) = α + β1 Y(t-1) + β2 (XA(t) - Y(t-1)) * ZA(t) + β3 (XB(t) - Y(t-1)) * ZB(t) + U  Equation 4.6 describes the persuasion hypothesis in a two-candidate (or two-party) race. Y(t) represents a voter’s position at time t, and Y(t-1) the same voter’s position at a previous time.    114 XA(t) stands for Party A’s perceived position at t, ZA(t) symbolizes a voter’s evaluation of party A at t.  I adopt this equation and include three parties for the UK (Conservative, Labour, and Liberal-Democrats) and five parties for Germany (CDU, SPD, FDP, Green and Left). Voter and party positions are coded 0-1, and party-voter distances range from -1 to +1. I do not use the absolute distance between parties and voters (as in the leading model), because the direction of opinion change matters. Negative values mean that the party is less supportive of an issue than the voter. Positive values signal the opposite order. Similarly, the party evaluations variables are coded differently than in previous models. I created a categorical measure to reflect three attitudes: dislike (-1), neutral (0), like (1). Positive coefficients for β2, β3, etc. support the persuasion hypothesis, suggesting that voter positions are malleable and conditional on party pre-commitment.    Similarly, my test of projection is inspired by Markus and Converse (1979: 1058) who hypothesize that a voter “will cognize a candidate’s policy position in such a way as to decrease the issue distance between the voter and a positively evaluated candidate or to increase the distance from a negatively evaluated candidate.”  Equation 4.7 Modeling following: Projection Y(t) = α + β1Ybar(t) + β2(X(t) - Ybar(t))*Z(t) + U Equation 4.7 describes the projection hypothesis. Y(t) represents a voter’s perception of a party on a given issue. Ybar(t) reflects the party’s actual position as estimated by all survey respondents. X(t) stands for the voter’s position on the issue, and Z(t) for her evaluation of the party. All variables are coded as above. Positive coefficients for β2 support the projection    115 hypothesis, suggesting that voters’ perceptions of party positions are conditional on party evaluation.  I estimate both equations within the SEM framework. Both models have a unidirectional setup, without reciprocal effects. I control for unobserved heterogeneity, not by first-differencing the variables, but by including a latent error variable. This has the advantage that I do not lose one wave of data. For persuasion, I run one model per issue but combine all parties. For projection, I run separate models—one for each party and issue. 4.3 Leading and following in the UK Starting with the British case, I estimate the models for leading and following described above for the three major parties and five policy issues introduced in Chapter 4. I run all models with all respondents, but separate by party and issue.  The leading model is the most complex of the three and combines different effect types: cross-lagged, synchronous, and stability effects. Of key interest are the coefficients for the first two. They indicate the extent to which policy distance and party evaluation influence one another—either with a lagged effect or instantaneously—and how strong the effect is in either direction. Negative and statistically significant effects of policy distance on party evaluation provide support for the “voters lead” hypothesis. Stability effects show how much both variables are determined by their own lags. However, because I first-differenced the variables, it is not unusual for the stability coefficients to be small and not statistically significant.  The models for persuasion and projection are a great deal simpler. Since I am mainly interested in understanding the reverse effects in the issue voting models (dynamics running from evaluation to distance), they assume unidirectional dynamics. Of key interest are the    116 coefficients for the interaction terms. Positive and statistically coefficients lend support to the respective hypotheses.   All models were estimated via maximum likelihood (ML) estimation, and all results are reported as unstandardized coefficients with regular standard errors in parentheses. Where available, model fit is assessed using four common goodness-of-fit (gof) measures: χ2 test of global fit, comparative fit index (CFI), standardized mean squared residuals (SRMR), and root mean square error of approximation (RMSEA).29 Variances, covariances and error covariances are omitted from these tables for readability. 4.3.1 Unemployment Table 4.1 presents the first results for leading (modeled after Figure 4.2). All three party models display excellent goodness-of-fit values, and policy-oriented evaluation is evident for all parties—either through lagged or synchronous effects (shaded grey). For the Conservative party, both past (-0.05) and present policy distances (-0.06) affect party evaluation. In the Labour model, policy distances have a lagged effect on party evaluation (-.04); in the model for the Liberal-Democrats the effect is synchronous (-0.04). These effects are correctly signed (negative) and statistically significant, after controlling for cross-lagged and synchronous effects in the opposite direction. As expected, coefficient sizes are relatively small. However, the combined effect for the Conservative party is more substantial and almost three times the size of the other parties’ effects (-0.05 + -0.06 = -0.11). Still, if the distance between a voter and the                                                  29 The Chi-square test is a statistical test of the null hypothesis that the specified model holds in the population. Smaller values (ideally <3 per df) indicate higher consistency. For the CFI, values of 0.95 or more reflect a good fit. SRMR values of 0.08 or less denote a good fit, and RMSEA with values of 0.06 or less are desirable.     117 Conservative party changes from 0 (identical positions) to 1 (largest possible distance), a voter adjusts her evaluation of the Conservative party by -0.11.  The models for Conservatives and Liberal-Democrats also show statistically significant effects pointing in the opposite direction (from evaluation to distance), confirming the presence of reciprocal dynamics. Again, both lagged (-0.17) and synchronous (-0.14) effects reach statistical significance in the Conservative party model; for Liberal-Democrats evaluation has only a lagged effect on distance (-0.06). Especially for the Conservative party the total effect from evaluation to distance (-0.31) is considerably larger than in the other direction. Coefficient sizes are much smaller for the Liberal-Democrats, but the pattern is similar: effects from evaluation to distance are stronger than for the other direction. What do these results tell us? First, they confirm the existence of reciprocal dynamics—effects from policy distance to party evaluation, and vice versa. A direct comparison of both effects shows that the net leading effect is negative for Conservatives and Liberal-Democrats, and positive for Labour. Second, by controlling for unobserved heterogeneity, it is possible to identify the causal effects between distance and evaluation. Their relatively small magnitudes show that distance considerations play a minor role for party evaluation. Other factors seem to carry more weight, but were differenced-out of the model.  To learn more about following, I perform additional tests. Starting with persuasion, the first column in Table 4.6 presents the results for unemployment. Positive and statistically coefficients for the interaction term (shaded grey) lend support to the persuasion hypothesis, suggesting that voters move their own positions closer to their preferred party’s positions and further away from parties they dislike. This is the case for all three parties. The persuasion effect is slightly bigger for Conservative (0.11) than for Labour (0.09) or Liberal-Democrats (0.07).    118 Table 4.7 presents the results for projection. Again, positive and statistically coefficients for the interaction term (shaded grey) lend support to the projection hypothesis, suggesting that voters’ perceptions of party positions are conditional on their own positions and their party pre-commitment. Projection effects are stronger for Conservative and Labour (0.14) than for Liberal-Democrats (0.08).  4.3.2 Welfare The results for welfare (Table 4.2) demonstrate that there is variation in leading and following not only across parties, but also across issues. Here, I find evidence of policy-orientated evaluation for the Conservative Party and for Liberal-Democrats, but not for Labour. Again, the effects are strongest in the Conservative party model, where both lagged (-0.04) and synchronous effects (-0.04) are statistically significant. For the Liberal Democrats, only past distances influence voters’ party evaluations (-0.03). The cross-lagged coefficient in the Labour model is correctly signed but only statistically significant at the 0.1-level (-0.03).   Conservative and Labour models also show statistically significant effects in the opposite direction (from evaluation to distance). For the Conservatives, these effects are lagged (-0.15) and synchronous (-0.11), for Labour only lagged effects exist (-0.06). So, for the Conservative party the pattern is similar for welfare and unemployment. On both issue dimensions, all effects are larger than for other parties, and net leading effects are negative. For Labour and Liberal Democrats, there are no similarities between both issues.  The results in the second column of Table 4.6 provide information about the persuasion effects on welfare. The magnitude of the persuasion effects for the two bigger parties is very similar (0.10 for Conservative and 0.11 for Labour) to those on unemployment. In comparison, the persuasion effect is much smaller for Liberal-Democrats (0.05).    119  Table 4.7 shows projection effects for all three parties. These are bigger for the Conservative party (0.19) than for Labour (0.15) or Liberal-Democrats (0.07). This is interesting, considering that the Conservative party takes a position that is far away from its supporter basis (see, Figure 3.5 in the previous chapter). 4.3.3 Redistribution The question on redistribution was not included in the 1994 panel wave, and the results in Table 4.3 cover a shorter period (1997-2001). They are notably different from previous issues. I do not find effects between policy distance and party evaluation for the Conservative Party. Yet, we see evidence of policy-oriented evaluation for Labour and Liberal-Democrats. In the Labour model, party-voter distances have both lagged (-0.04) and synchronous (-0.05) effects on party evaluation. In the Liberal-Democrats model, there is only a synchronous effect (-0.05). The total effect of distance on evaluation in the Labour model is about twice as large (-.09) as in the Liberal-Democrats model.   The evidence for reciprocal effects is weaker. In the Labour and Liberal-Democrats models lagged effects from policy distance to party evaluation (both -0.05) are only significant at the 0.1-level. Consequently, leading is the causally predominant effect for Labour; for Liberal-Democrats both effects are of equal strength.     For persuasion, the results in the third column of Table 4.6 show an almost identical pattern to the one for welfare. The effects for Conservative and Labour are of equal strength (0.11) and twice the size as for Liberal-Democrats (0.05). Conservative party evaluation also has an independent effect on voter position (0.03).  The projection effects (Table 4.7) are very similar to the previous issue. Again, I find the biggest effect for Conservative (0.16), followed by Labour (0.12) and Liberal-Democrats (0.09).    120 Recall hat, here too, the gap between the Conservative party and its supporters is significantly larger than for the other two parties.  4.3.4 European unification On the issue of European unification, the links between policy distance and party evaluation appear looser. Table 4.4 shows no effects in the Liberal-Democrats model, and fewer statistically significant coefficients in the other models. For the Conservative party, there is weak evidence of policy-oriented evaluation based on lagged distances (-.02). For Labour, the effect is synchronous and slightly stronger (-.05).   Evidence of effects running in the opposite direction (from evaluation to distance) is only present in the Conservative model. However, the coefficients for both lagged (-.06) and synchronous effects (-.12) are much larger in magnitude than for leading. Conversely, the Labour model shows no signs of reciprocal effects; leading is the predominant direction.    The results in the fourth column of Table 4.6 do not reflect these differences. The persuasion effects for European unification are strikingly similar to the other issues. The coefficients for Conservative and Labour are the same size (.10) and clearly lager than for Liberal-Democrats (.02).   Likewise, projection effects (Table 4.7) follow the same pattern as for previous issues. The effect is bigger for Conservative (0.19) than Labour (0.16) and Lib-Dems (0.07). Interestingly, and despite these similarities, the aggregate pattern (Figure 3.11) is different on this issue, and gaps between parties and supporters are similar in size for all three parties.  4.3.5 Nationalization Like for redistribution, the results in Table 4.5 cover a shorter period (1992-1997). On the question of nationalization, the links between parties and voters appear even weaker than on the    121 issue of European unification. In fact, only the Labour model shows statistically significant coefficients. Policy distances have both lagged (-.07) and synchronous effects (-.13) on party evaluation. There are no effects in the opposite direction. For the other parties, there are no effects at all.   I still look for persuasion effects on this issue, and the results in the last column of Table 4.6 suggest that, overall, these are stronger than for the other issues. The effects for Conservative and Labour (both .17) are larger than for Liberal-Democrats (.11). Conservative party evaluation has an independent negative effect on voter position (-.02), suggesting that the Conservative party has a slightly smaller persuasion effect than Labour.30 4.4 Leading and following in Germany The analysis for Germany is basically identical to the one performed for the UK. It includes the same parties and issues as in Chapter 3. Despite the different model setup for leading, the expectations are the same: negative and statistically significant coefficients from distance to evaluation (shaded grey in all tables) lend support to the leading hypothesis; effects in the other direction (from evaluation to distance) confirm the presence of reciprocal effects. The more explicit tests of following—persuasion and projection—are designed in the same way as for the UK. Here, positive and statistically significant coefficients for the interaction term support the hypothesis that voters follow.  4.4.1 Left-Right Table 4.8 presents the results for the leading test on the general Left-Right dimension. Overall, the results provide little support for leading (or policy-oriented evaluation) on this dimension.                                                  30 I cannot present projection results on this issue, as models would not converge.    122 Only the Green party model shows a correctly signed and statistically significant coefficient (-0.14), suggesting that smaller party-voter distances lead to higher evaluations of the Green party. It is worth noting that this is the average effect for all survey respondents, as I do not distinguish between different types of voters here. For the other parties, proximity considerations do not seem to matter for evaluation.  The lack of effects here is surprising, as Left-Right is the “workhorse” dimension for almost all comparative research on voting and opinion formation. Recall, however, that all variables on the model are first-differenced to control for unobserved factors. That means that all stable factors (e.g., demographics) are differenced out of the model. Furthermore, and unlike most analyses, the model controls for effects in the other direction. Thus, it distills the pure effect of distance on evaluation, which is (surprisingly) weak.  The coefficients for the opposite direction (from evaluation to distance) suggest that proximity links exist also for CDU, SPD and the Left, but that the dynamic is reversed. How respondents evaluate a party determines how they perceive the distance to this party. All effects are negative and statistically significant, indicating that higher party evaluations lead to smaller perceived distances. Effect sizes are similar for CDU (-.08) and SPD (-.07), and twice the size for the Left (-.16).   However, these coefficients are difficult to interpret, since they do not tell us whether voters change their own positions (persuasion) or their perceived party positions (projection). The results in the first column of Table 4.11 can shed light in this question. The positive and statistically significant coefficients for the interaction term (shaded grey) indicate persuasion effects for all parties, except for the FDP. These effects differ in size and are strongest for the CDU (.10). The effects for the other parties are similar in size (.05 for SPD, and .06 for Greens    123 and Left). Likewise, Table 4.12 shows projection effects for all five parties. They are very similar in size and range between 0.18 (for the Green party) and 0.21 (for CDU and SPD). 4.4.2 Nuclear power Table 4.9 presents the results for a more specific policy issue, nuclear power. Again, they do not show overwhelming support for issue voting. Only the coefficients for the SPD and Greens are in the expected direction (negative) and statistically significant. The effect is substantial for the Green Party (-.20) and clearly stronger than for the SPD (-.14). For the other parties, distances on nuclear power do not affect party evaluations.  Effects in the reverse direction (from evaluation to distance) are present in the CDU, SPD and Left models. However, the coefficients are positive for CDU (.10) and SPD (.12), suggesting that higher evaluations lead to larger distances on nuclear power. For the Left, the effect is negative and substantial in size (-.28), indicating that following is the predominant dynamic for the Left.  The test of persuasion (Table 4.11) reveals a pattern different from the one for Left-Right. Persuasion effects are only present for the CDU (.09). None of the other coefficients reach statistical significance. Interestingly, CDU party evaluation also has an independent effect on voter positions, and this effect is negative (-.06). Combined, these results suggest that, on the issue of nuclear power, no party has a big persuasion effect on voters. The picture looks different for projection (Table 4.12).31 Here, I find effects for all four parties. As for Left-Right, there is little variation across parties. All coefficients range between 0.8 (Left) and 0.11 (FPD). So, the magnitude of the projection effects is much smaller for nuclear power than on Left-Right.                                                  31 No results for the Green party because of convergence problems.    124 4.4.3 European unification The analysis of leading on European unification (Table 4.10) produces even fewer statistically significant coefficients than for the other issues. This suggests that, overall party-voter links are looser on this issue. The two coefficients that do reach statistical significance—for SPD and Greens—provide support for policy-oriented evaluation. The magnitude of the effect for the Green party (-.17) is about twice the size as for the SPD (-.08). The results provide no evidence of dynamics running from evaluation to distance. Still, the persuasion model (Table 4.11) shows some evidence of following. Again, the only discernible persuasion effect is for the CDU (.08). In addition, FDP evaluation has an independent negative effect (-.04) on voter positions. That is, voters who favour the FDP lower their support for nuclear power, while voters with a negative opinion of the party become more supportive of European unification. Again, I find projection effects for all parties (Table 4.12). Here, the variation across parties is greater than for previous issues. Coefficients range from 0.05 (for the FDP) to 0.13 (for the Left). So, they are similar in size compared to nuclear power, but much smaller than on Left-Right.  4.5 Discussion  What have we learned, and how do these findings compare to previous ones? Overall, these results confirm that party-voter relations are reciprocal. Leading—in the form of issue voting or policy oriented evaluation—does occur, but it is not the only dynamic, or even the dominant one. Effects in the opposite direction—from party evaluation to policy distance—are also part of the story. Taken together, these findings suggest that standard accounts of issue voting that do not    125 control for effects in the opposite grossly overestimate the impact of policy distance on party evaluation and vote choice.  The analysis also reveals differences across countries, parties and issues. Research in the US suggests that, at the country level, the degree of party system polarization affects the dynamic relationship between parties and voters (e.g., Baldassarri and Gelman 2008; Druckman et a1. 2013; Levendusky 2009). Likewise, comparative works have shown that polarization affects political representation (e.g., Powell 2010) and citizens’ satisfaction with democracy (Ezrow and Xezonakis 2011). Furthermore, Alvarez and Nagler (2004: 60) have shown that “the level of issue voting increases” as parties’ issue positions become more distinct. However, they do not control for party-driven influence on issue positions, which means that their results cannot be counted as evidence of leading. The analysis in this chapter is better suited to capturing the extent of leading, and thus makes important methodological improvements. Yet, whether polarization—or party system structure more generally—affects the extent to which voters lead is a question that this analysis cannot address. With only two cases that differ from each other in too many respects to conduct a systematic comparison, any claims about causal effects would be purely speculative and not scientifically sound.  Likewise, causal claims about effects of party or issue type on leading need to be made with caution, given the small number of cases. However, the summary tables for the UK (Table 4.13) and Germany (Table 4.14) may hint at a pattern. Across all issues, the predominant direction for the British Conservatives is from evaluation to distance (following). For Labour, leading is the predominant dynamic, with one exception: welfare. Effects for Liberal-Democrats are weaker overall and a mix of leading and following. Thus, it seems that the Conservative party has a stronger influence on voters than the other parties. Conversely, Labour is judged to a    126 greater extent on policy positions. Across issues, the results for the UK show mixed dynamics for unemployment, welfare, and European unification. For all, effects from evaluation to distance are stronger, on average. On the other two issues—redistribution and nationalization—leading is the dominant dynamic. The German case shares some similarities with the British (Table 4.14). Leading is stronger for the SPD (equivalent to Labour) than the CDU (equivalent to Conservative). There are no effects in any of the FDP models (closest equivalent to Liberal-Democrats). The Green party models show the strongest effects for leading on all dimensions. For the Left, following is the predominant dynamic. Put differently, policy considerations affect the two centre-left parties’—SPD and Green—evaluations more than the other parties’ evaluations. This particularly true for the Left party, for which policy distances appear to matter the least. Across issues, party-voter links are strongest on Left-Right, where net leading effects are negative. They are positive on nuclear power and European unification.  What about following, more specifically? Do party evaluations affect voters’ positions? My tests of persuasion suggest that voters’ issue positions are fairly stable over time. However, voters do look for cues from parties they support and adjust their views accordingly. These persuasion effects are stronger in the UK than in Germany. Persuasion effects are present for all three parties on all issues (Table 4.13). While the effect strengths for Conservative and Labour are similar, they are clearly weaker for Liberal-Democrats. Comparing across issues, all parties have the strongest persuasion effects on nationalization. Interestingly, this is the issue with the least amount of party movement. It suggests that parties can lead even when they do not move. These findings are in line with Brader and Tucker (2012: 416) who address “questions about whether partisanship means anything outside of the U.S., and particularly in multiparty    127 systems.” Their experimental results from Great Britain, Hungary, and Poland show that party cues affect the policy positions of partisans. In contrast, Milazzo et al. (2012) conclude that “British citizens […] typically update their party attachments to match their policy preferences, rather than vice versa” (Milazzo et al. 2012: 273). In their analysis of the links between voter positions and partisanship in the UK (1987-2001) they find strong evidence from Left-Right orientations to party attachments, but not vice versa. Because of their different setups, these analyses are not directly comparable, but the differences may inspire further investigations.  In comparison, German parties exert less influence over their supporters, as persuasion is mostly limited to Left-Right. The CDU is the only party with persuasion effects on the two other issues.  Are perceived party positions also conditional on party pre-commitment? I do find projection effects in both countries. For the UK, projection effects are present for all parties and issues (Table 4.13). While there are no obvious differences in effect strength across issues, variation exists across parties. Projection effects are strongest in the Conservative Party models. They are slightly weaker for Labour, and clearly smaller for Liberal-Democrats. Overall these findings are in line with Evans and Andersen (2004: 31), who demonstrate the existence of partisan contamination in the UK (1997-2001), where “pre-existing partisan orientations strongly bias the perception of proximity between self and party issue positions.”  Similarly, I discover projection effects for all German parties, and across all issues (Table 4.14). These effects are notably stronger on Left-Right—about twice the size compared to nuclear power and European unification.    128 4.6 Conclusion The analysis in this chapter extends existing works on reciprocal relations between parties and voters (see, e.g. Carruba 2001; Carsey and Layman 2006; Dancey and Goren 2010; Evans and Andersen 2004; Gabel and Scheve 2007; Goren 2005; Highton and Kam 2011; Layman and Carsey 2002; Milazzo, Adams and Green 2012). Most of these studies originate in the US or use American data and focus primarily on the reciprocal relationship between voter position and partisanship. Given the American two-party system and the importance of partisanship to explain voter behaviour in this context, this comes as no surprise. Whether partisanship has the same meaning outside of the US is an intensely debated, but unsettled, question (e.g., Butler and Stokes 1969; Dalton 2008; Clarke et al. 2004; 2009; Clements and Bartle 2009).   Because this dissertation focuses on party-voter dynamics in European multiparty systems, the analysis deviates from American models in two ways. First, instead of concentrating on partisanship and vote choice, I use a measure of party evaluation to capture voters’ party preferences. This allows me to open up the analysis to examine the dynamic relationship between all voters and parties—not just between pairs of parties and supporters. Furthermore, using party evaluation accounts for the possibility that voters like (and dislike) more than one party in a system that offers numerous options. To test whether voters respond differently to parties they support—versus those they oppose or hold a neutral opinion towards—I calculate a measure of party support that is based on party evaluation and divides respondents into three groups: support, neutral, and opposition. This captures the basic idea of partisanship, while recognizing that electoral competition in multiparty systems produces different party-voter ties than in two-party systems. At the same time, it reduces measurement error based on different question wording in different contexts (Johnston 2006).    129 Second, I directly incorporate party positions into all models and examine the reciprocal dynamics between party-voter distances and party evaluation. This allows me to perform direct tests of leading as captured by proximity models. Again, this is a sensible approach for multiparty system where Left-Right divisions matter, but where the distribution of parties along an issue continuum are equally—if not more—important (Zakharova and Warwick 2014). Again, using policy distances enables me to expand the analysis and include all relevant parties in a party system, thus capturing the dynamics of electoral competition more fully. These adaptations increase the complexity of the empirical analysis and present additional demands for estimation models and techniques. I perform all empirical tests within the SEM framework with a focus on the causal dynamics between parties and voters. To rule out reverse causation, I estimate non-recursive models with reciprocal effects; to avoid spurious conclusions, I control for unobserved heterogeneity by incorporating fixed effects in the form of first-differences (FD) models. The latter presents a methodological advancement and distinguishes this analysis from existing ones (e.g., Dancey and Goren 2010; Milazzo et al. 2012).  Moreover, my tests of leading perform proper tests of issue voting that account for partisan-motivated psychological forces. The results are, perhaps, surprising and certainly disappointing to proponents of issue voting. In the grand scheme, policy distances play a minor role for party evaluation and vote choice.  Conversely, the results present solid evidence for following. Voters’ own positions on issues, as well as voters’ perceptions of parties’ positions are conditional on party evaluation. Moreover, the results of the persuasion and projection tests suggest that partisanship (positive party evaluation) has the same effects as in the US. Voters tends follow their preferred party.     130	  Figure	  4.1	  Party-­‐voter	  relations	  from	  a	  positional	  perspective	   Notes:	  This	  Figure	  illustrates	  the	  relationship	  between	  a	  voter’s	  own	  position	  on	  an	  issue,	  where	  she	  perceives	  the	  party	  to	  stand	  on	  the	  same	  issue,	  and	  how	  she	  evaluates	  this	  party.	  Arrows	  represent	  the	  cognitive	  processes	  linking	  all	  three.	  The	  Figure	  is	  based	  on	  Brody	  and	  Page’s	  (1972:	  456)	  model	  of	  policy	  voting.	    ! Party&&evaluation&Voter&&position&Party&&position&Issue%voting%Persuasion%Projection%   131	  Figure	  4.2	  Leading	  model	  for	  the	  UK:	  Four-­‐wave	  dynamic	  panel	  model	  (SEM)	  	  	  Notes:	  This	  Figure	  illustrates	  the	  relationships	  between	  the	  variables	  estimated	  in	  the	  structural	  equation	  model	  for	  the	  UK.	  Solid	  arrows	  represent	  the	  direction	  of	  estimated	  causal	  effects.	  Dashed	  double-­‐headed	  arrows	  represent	  (error)	  covariances	  included	  in	  the	  model.	  Coefficient	  names	  are	  omitted	  for	  presentation	  purposes.	    !!!!!!X3!–!X2!!ε5!!Y2!–!Y1!X4!–!X3! X5!–!X4!Y3!–!Y2! Y4!–!Y3! Y5!–!Y4!ε2! ε3!!ε4!!! !! !ε1!ε6!X2!–!X1!   132	  Figure	  4.3	  Leading	  model	  for	  Germany:	  Three-­‐wave	  dynamic	  panel	  model	  (SEM)	  	  	  Notes:	  This	  Figure	  illustrates	  the	  relationships	  between	  the	  variables	  estimated	  in	  the	  structural	  equation	  model	  for	  the	  UK.	  Solid	  arrows	  represent	  the	  direction	  of	  estimated	  causal	  effects.	  Dashed	  double-­‐headed	  arrows	  represent	  (error)	  covariances	  included	  in	  the	  model.	  	  !!!ε2!X2!–!X1! X3!–!X2!Y2!–!Y1!Y1!X1!Y3!–!Y2!ε4!ε1! ε3!β1! β3!β2! β4!!!! β5! β6!   133  Table 4.1 Test of leading (“policy oriented evaluation”): UK, Unemployment  Conservative  Labour  Lib-Dem  Coef. SE  Coef. SE  Coef. SE Cross-lagged effects          Distance  Evaluation -.05* (.01)  -.04* (.01)  -.02 (.01)  Evaluation  Distance -.17* (.02)  -.03 (.02)  -.06* (.03) Synchronous effects          Distance  Evaluation -.06* (.02)  -.03 (.02)  -.04* (.02)  Evaluation  Distance -.14* (.05)  .01 (.06)  -.02 (.05) Stability effects          Party evaluation          1-2 -.23* (.01)  -.25* (.02)  -.38* (.02)  2-3 .24* (.07)  .09 (.06)  .02 (.03)  3-4 .09+ (.05)  .04 (.05)  .07* (.04)  Policy distance          1-2 -.42* (.02)  -.42* (.02)  -.49* (.02)  2-3 -.02 (.04)  .01 (.04)  .09+ (.05)  3-4 -.03 (.03)  -.06+ (.03)  .03 (.05) Model fit          N 2569  2598  2267  Model χ2 29.47  14.97  19.58  Degrees of freedom 10  10  10  RMSEA 0.03  0.01  0.02  CFI 1.00  1.00  1.00  SRMR 0.02  0.01  0.02 Notes: This Table reports unstandardized maximum likelihood (ML) coefficients, based on raw data and estimated in a series of structural equation models, which capture the reciprocal relationship between policy distance and party evaluation. Standard errors are reported in parentheses. All factors are coded -1/1. +p < .10, *p < .05.        134  Table 4.2 Test of leading (“policy oriented evaluation”): UK, Welfare  Conservative  Labour  Lib-Dem  Coef. SE  Coef. SE  Coef. SE Cross-lagged effects          Distance  Evaluation -.04* (.01)  -.03+ (.01)  -.03* (.01)  Evaluation  Distance -.15* (.02)  -.06* (.02)  .00 (.02) Synchronous effects          Distance  Evaluation -.04* (.02)  -.04 (.02)  -.03 (.02)  Evaluation  Distance -.11* (.04)  -.01 (.05)  .02 (.04) Stability effects          Party evaluation          1-2 -.22* (.01)  -.25* (.02)  -.38* (.02)  2-3 .24* (.07)  .07 (.06)  .06 (.04)  3-4 .10* (.05)  .06 (.06)  .05 (.05)  Policy distance          1-2 -.35* (.02)  -.44* (.02)  -.48* (.02)  2-3 .11* (.04)  .09* (.04)  .03 (.04)  3-4 .14* (.05)  -.02 (.03)  .00 (.04) Model fit          N 2571  2579  2274  Model χ2 39.81  29.46  19.66  Degrees of freedom 10  10  10  RMSEA 0.03  0.03  0.02  CFI 0.99  1.00  1.00  SRMR 0.02  0.02  0.02 Notes: This Table reports unstandardized maximum likelihood (ML) coefficients, based on raw data and estimated in a series of structural equation models, which capture the reciprocal relationship between policy distance and party evaluation. Standard errors are reported in parentheses. All factors are coded -1/1. +p < .10, *p < .05.       135  Table 4.3 Test of leading (“policy oriented evaluation”): UK, Redistribution  Conservative  Labour  Lib-Dem  Coef. SE  Coef. SE  Coef. SE Cross-lagged effects          Distance  Evaluation -.01 (.01)  -.04* (.01)  -.01 (.01)  Evaluation  Distance -.02 (.03)  -.05+ (.03)  -.05+ (.03) Synchronous effects          Distance  Evaluation -.02 (.02)  -.05* (.02)  -.05* (.02)  Evaluation  Distance -.00 (.07)  -.05 (.05)  .01 (.05) Stability effects          Party evaluation          1-2 -.32* (.02)  -.30* (.02)  -.37* (.02)  2-3 .02 (.05)  .20* (.06)  .10+ (.06)  3-4 .05 (.06)  .16* (.06)  .06 (.06)  Policy distance          1-2 -.46* (.02)  -.50* (.02)  -.47* (.02)  2-3 -.09* (.04)  -.09* (.04)  -.02 (.04)  3-4 .01 (.04)  .01 (.04)  .05 (.05) Model fit          N 1958  1946  1700  Model χ2 25.72  31.71  17.03  Degrees of freedom 10  10  10  RMSEA 0.03  0.03  0.02  CFI 1.00  0.99  1.00  SRMR 0.02  0.02  0.02 Notes: This Table reports unstandardized maximum likelihood (ML) coefficients, based on raw data and estimated in a series of structural equation models, which capture the reciprocal relationship between policy distance and party evaluation. Standard errors are reported in parentheses. All factors are coded -1/1. +p < .10, *p < .05.       136  Table 4.4 Test of leading (“policy oriented evaluation”): UK, European unification  Conservative  Labour  Lib-Dem  Coef. SE  Coef. SE  Coef. SE Cross-lagged effects          Distance  Evaluation -.02* (.01)  -.01 (.01)  -.00 (.01)  Evaluation  Distance -.06* (.02)  -.02 (.03)  -.00 (.03) Synchronous effects          Distance  Evaluation -.00 (.02)  -.05* (.02)  -.01 (.02)  Evaluation  Distance -.12* (.05)  .11 (.07)  .06 (.05) Stability effects          Party evaluation          1-2 -.23* (.02)  -.24* (.02)  -.37* (.02)  2-3 .21* (.07)  .10 (.06)  .10* (.05)  3-4 .09+ (.05)  .04 (.05)  .07 (.05)  Policy distance          1-2 -.44* (.02)  -.40* (.02)  -.43* (.02)  2-3 -.02 (.03)  .03 (.04)  .06 (.04)  3-4 .04 (.04)  -.04 (.04)  -.03 (.04) Model fit          N 2501  2458  2145  Model χ2 28.14  17.6  18.67  Degrees of freedom 10  10  10  RMSEA 0.03  0.02  0.02  CFI 1.00  1.00  1.00  SRMR 0.02  0.02  0.02 Notes: This Table reports unstandardized maximum likelihood (ML) coefficients, based on raw data and estimated in a series of structural equation models, which capture the reciprocal relationship between policy distance and party evaluation. Standard errors are reported in parentheses. All factors are coded -1/1. +p < .10, *p < .05.      137  Table 4.5 Test of leading (“policy oriented evaluation”): UK, Privatization  Conservative  Labour  Lib-Dem  Coef. SE  Coef. SE  Coef. SE Cross-lagged effects          Distance  Evaluation -.02 (.02)  -.07* (.03)  -.03 (.04)  Evaluation  Distance -.04 (.04)  -.06 (.05)  .04 (.05) Synchronous effects          Distance  Evaluation -.00 (.04)  -.13* (.05)  .01 (.07)  Evaluation  Distance -.12 (.08)  .13 (.10)  .01 (.09) Stability effects          Party evaluation          1-2 -.25* (.03)  -.31* (.03)  -.42* (.04)  2-3 .27* (.13)  .12 (.10)  .06 (.08)  3-4 .06 (.07)  -.03 (.09)  -.02 (.09)  Policy distance          1-2 -.46* (.03)  -.39* (.03)  -.46* (.03)  2-3 -.02 (.06)  -.06 (.05)  -.13* (.06)  3-4 -.02 (.07)  -.17* (.07)  -.03 (.09) Model fit          N 636  619  509  Model χ2 4.90  22.03  21.34  Degrees of freedom 10  10  10  RMSEA 0.00  0.04  0.05  CFI 1.00  0.99  0.99  SRMR 0.02  0.03  0.03 Notes: This Table reports unstandardized maximum likelihood (ML) coefficients, based on raw data and estimated in a series of structural equation models, which capture the reciprocal relationship between policy distance and party evaluation. Standard errors are reported in parentheses. All factors are coded -1/1. +p < .10, *p < .05.       138  Table 4.6 Test of following (“persuasion”): UK, all issues  Unemployment  Welfare  Redistribution  European Unification  Privatization  Coef. SE  Coef. SE  Coef. SE  Coef. SE  Coef. SE DV: Voter position (t)                Voter position (t-1) .98* (.02)  .84* (.02)  .91* (.03)  .89* (.02)  .82* (.04)  Conservative Dist. * Eval.  .11* (.01)  .10* (.01)  .11* (.01)  .10* (.01)  .17* (.02)  Conservative Dist. .25* (.01)  .17* (.01)  .32* (.02)  .23* (.01)  .21* (.02)  Conservative Eval. .01+ (.01)  .01* (.01)  .03* (.01)  -.01 (.01)  -.02* (.01)  Labour Dist. * Eval. .09* (.01)  .11* (.01)  .11* (.01)  .10* (.01)  .17* (.02)  Labour Dist. .40* (.01)  .28* (.01)  .30* (.02)  .29* (.01)  .27* (.03)  Labour Eval. .00 (.01)  .01 (.00)  .00 (.01)  -.01 (.01)  .02 (.03)  Lib-Dems Dist. * Eval. .07* (.01)  .05* (.01)  .05* (.02)  .02* (.01)  .11* (.02)  Lib-Dems Dist. .30* (.01)  .32* (.01)  .37* (.02)  .30* (.02)  .33* (.03)  Lib-Dems Eval. .00 (.01)  .01 (.00)  -.01 (.01)  .01 (.01)  -.01 (.01) Model fit                N 2327  2326  3211  2230  545  Model χ2 4098.96  3731.73  1028.82  4257.09  1280.30  Degrees of freedom 109  109  10  109  109  RMSEA 0.13  0.12  0.18  0.13  0.14  CFI 0.67  0.68  0.88  0.72  0.65  SRMR 0.05  0.04  0.02  0.03  0.04 Notes: This Table reports unstandardized maximum likelihood (ML) coefficients, based on raw data and estimated in a series of structural equation models in which the dependent variable is a voter’s issue position at time t, as a function of prior issue position, distance to all parties, evaluation of all parties, and their interactions. Standard errors are reported in parentheses. The redistribution model uses only three waves of data. +p < .10, *p < .05.     139  Table 4.7 Test of following (“projection”): UK, all issues  Conservative  Labour  Lib-Dem DV: Perceived party position (t) Coef. SE  Coef. SE  Coef. SE Unemployment          Actual party position (t) .82* (.09)  .94* (.17)  .87* (.28)  Distance * Evaluation (t) .14* (.01)  .14* (.01)  .08* (.01)  Distance (t) .21* (.01)  .25* (.01)  .25* (.01)  Evaluation (t) -.01 (.01)  .03* (.00)  .01* (.00) Model fit          N 2569  2598  2267  CFI 1.00  1.00  1.00  SRMR 0.02  0.02  0.02        Welfare          Actual party position (t) .98* (.07)  1.21* (.40)  .52+ (.30)  Actual distance * Evaluation (t) .19* (.01)  .15* (.01)  .07* (.01)  Distance (t) .13* (.01)  .26* (.01)  .29* (.01)  Evaluation (t) -.03* (.01)  .01+ (.00)  .01* (.00) Model fit          N 2571  2579  2274  CFI 1.00  1.00  1.00  SRMR 0.02  0.02  0.01          Redistribution          Actual party position (t) .92* (.22)  1.40* (.31)  1.33* (.55)  Actual distance * Evaluation (t) .16* (.01)  .12* (.01)  .09* (.01)  Distance (t) .24* (.01)  .21* (.01)  .24* (.01)  Evaluation (t) -.02* (.01)  .02* (.01)  .02* (.00) Model fit          N 3448  3442  3123  CFI 1.00  1.00  1.00  SRMR 0.01  0.01  0.01          European Unification          Actual party position (t) .97* (.17)  1.21* (.27)  1.00* (.25)  Actual distance * Evaluation (t) .19* (.01)  .16* (.01)  .07* (.01)  Distance (t) .20* (.01)  .22* (.01)  .21* (.01)  Evaluation (t) -.00 (.01)  .03* (.01)  .01* (.00) Model fit          N 2501  2458  2145  CFI 1.00  1.00  1.00  SRMR 0.01  0.02  0.02 Notes: This Table reports unstandardized maximum likelihood (ML) coefficients, based on raw data and estimated in a series of structural equation models in which the dependent variable is a voter’s perceived party position on an issue at time t, as a function of a party’s actual position on this issue, voter-party distance on this issue, party evaluation, and the interaction between distance and evaluation. Standard errors are reported in parentheses. The redistribution model uses only three waves of data. Data on privatization is only available for 1992-1997. +p < .10, *p < .05.       140   Table 4.8 Test of leading (“policy oriented evaluation”): Germany, Left-Right  CDU  SPD  FDP  Green  Left  Coef. SE  Coef. SE  Coef. SE  Coef. SE  Coef. SE Synchronous effects                Distance  Evaluation -.02 (.05)  -.01 (.04)  -.08 (.05)  -.14* (.05)  .02 (.06)  Evaluation  Distance -.08* (.03)  -.07* (.02)  -.03 (.03)  -.03 (.03)  -.16* (.04) Stability effects                Party evaluation                1-2 -.56* (.02)  -.60* (.03)  -.63* (.03)  -.55* (.02)  -.43* (.02)  2-3 .05 (.05)  .10+ (.05)  .05 (.06)  .11* (.06)  -.36* (.03)  Policy distance                1-2 -.62* (.02)  -.68* (.03)  -.66* (.03)  -.61* (.03)  -.59* (.03)  2-3 .01 (.05)  .04 (.05)  -.05 (.05)  .02 (.05)  -.41* (.02) Model fit                N 1250  1259  1196  1207  1136  Model χ2 218.9  138.4  150.8  139.4  486.0  Degrees of freedom 5  5  5  5  7  RMSEA 0.19  0.15  0.16  0.15  0.25  CFI 0.90  0.94  0.93  0.93  0.71  SRMR 0.07  0.05  0.06  0.06  0.11 Notes: This Table reports unstandardized maximum likelihood (ML) coefficients, based on raw data and estimated in a series of structural equation models, which capture the reciprocal relationship between policy distance and party evaluation. Standard errors are reported in parentheses. The model for the Left Party is computed without stability error covariances. All factors are coded -1/1. +p < .10, *p < .05.       141 Table 4.9 Test of leading (“policy oriented evaluation”): Germany, Nuclear power  CDU  SPD  FDP  Green  Left  Coef. SE  Coef. SE  Coef. SE  Coef. SE  Coef. SE Synchronous effects                Distance  Evaluation -.05 (0.4)  -.14* (.04)  -.09 (.06)  -.20* (.04)  .11 (.07)  Evaluation  Distance .10* (0.4)  .12* (.05)  .01 (.06)  .09+ (.05)  -.28* (.10) Stability effects                Party evaluation                1-2 -.51* (.03)  -.52* (.03)  -.60* (.04)  -.56* (.03)  -.45* (.05)  2-3 .27* (.08)  .29* (.08)  .05 (.08)  .20* (.07)  -.32* (.06)  Policy distance                1-2 -.59* (.03)  -.82* (.03)  -.71* (.04)  -.64* (.03)  -.61* (.06)  2-3 .10 (.07)  .02 (.06)  .01 (.06)  -.06 (.06)  .08 (.14) Model fit                N 909  905  658  938  272  Model χ2 107.7  64.24  53.95  148.3  61.33  Degrees of freedom 5  5  5  5  6  RMSEA 0.15  0.11  0.12  0.18  0.18  CFI 0.92  0.96  0.95  0.90  0.85  SRMR 0.06  0.04  0.05  0.07  0.09 Notes: This Table reports unstandardized maximum likelihood (ML) coefficients, based on raw data and estimated in a series of structural equation models, which capture the reciprocal relationship between policy distance and party evaluation. Standard errors are reported in parentheses. The model for the Left Party is computed without stability error covariances. All factors are coded -1/1. +p < .10, *p < .05.       142 Table 4.10 Test of leading (“policy oriented evaluation”): Germany, EU  CDU  SPD  FDP  Green  Left  Coef. SE  Coef. SE  Coef. SE  Coef. SE  Coef. SE Synchronous effects                Distance  Evaluation -.04 (.03)  -.08* (.03)  -.06 (.04)  -.17* (.04)  -.10 (.08)  Evaluation  Distance .02 (.04)  .08 (.05)  -.00 (.04)  .11 (.05)  .01 (.11) Stability effects                Party evaluation                1-2 -.45* (.03)  -.55* (.03)  -.61* (.04)  -.49* (.03)  -.40* (.04)  2-3 .26* (.08)  .18* (.08)  -.02 (.07)  .14 (.09)  -.37* (.05)  Policy distance                1-2 -.89* (.03)  -.90* (.03)  -.91* (.03)  -.91* (.04)  -.79* (.05)  2-3 .01 (.05)  -.12* (.05)  .02 (.05)  -.04 (.06)  -.47* (.05) Model fit                N 863  857  722  701  358  Model χ2 35.89  21.52  13.63  50.62  123.7  Degrees of freedom 5  5  5  5  7  RMSEA 0.09  0.06  0.05  0.11  0.22  CFI 0.98  0.99  0.99  0.96  0.78  SRMR 0.03  0.02  0.02  0.03  0.08 Notes: This Table reports unstandardized maximum likelihood (ML) coefficients, based on raw data and estimated in a series of structural equation models, which capture the reciprocal relationship between policy distance and party evaluation. Standard errors are reported in parentheses. The model for the Left Party is computed without stability error covariances. All factors are coded -1/1. +p < .10, *p < .05.     143  Table 4.11 Test of following (”persuasion”): Germany, all issues  Left-Right  Nuclear power  European unification  Coef. SE  Coef. SE  Coef. SE DV: Voter position (t)          Voter position (t-1) .78* (.05)  .79* (.10)  .70* (.09)  CDU Dist. * Eval. .10* (.02)  .09* (.04)  .08* (.04)  CDU Dist. .29* (.03)  .19* (.06)  .20* (.05)  CDU Eval. .00 (.01)  -.06* (.02)  .02 (.02)  SPD Dist. * Eval. .05* (.02)  -.03 (.04)  .00 (.04)  SPD Dist. .26* (.03)  .24* (.06)  .24* (.06)  SPD Eval. -.01+ (.01)  -.01 (.02)  .02 (.02)  FDP Dist. * Eval. .01 (.02)  .00 (.04)  .03 (.04)  FDP Dist. .06* (.03)  .17* (.06)  .30* (.07)  FDP Eval. .00 (.01)  .01 (.02)  -.04* (.02)  Green Dist. * Eval. .06* (.02)  .05 (.05)  .02 (.04)  Green Dist. .18* (.03)  .22* (.07)  .04 (.05)  Green Eval. .01 (.01)  -.01 (.03)  .04* (.02)  Left Dist. * Eval. .06* (.02)  .05 (.04)  .05 (.04)  Left Dist. .15* (.03)  .21* (.05)  .12* (.05)  Left Eval. -.02 (.01)  .00 (.02)  .01 (.02) Model fit          N 1139  372  470  Model χ2 707.56  374.61  275.68  Degrees of freedom 16  16  16  RMSEA 0.20  0.25  0.19  CFI 0.80  0.68  0.73  SRMR 0.02  0.02  0.02 Notes: This Table reports unstandardized maximum likelihood (ML) coefficients, based on raw data and estimated in a series of structural equation models in which the dependent variable is a voter’s issue position at time t, as a function of prior issue position, distance to all parties, evaluation of all parties, and their interactions. Standard errors are reported in parentheses.  +p < .10, *p < .05.    144  Table 4.12 Test of following (”projection”): Germany, all issues  CDU  SPD  FDP  Green  Left  Coef. SE  Coef. SE  Coef. SE  Coef. SE  Coef. SE Left-Right                Actual party position (t) 1.03* (.08)  .72* (.27)  2.37* (.23)  .66* (.23)  .85* (.23)  Distance * Evaluation (t) .21* (.01)  .21* (.02)  .20* (.02)  .18* (.02)  .19* (.03)  Distance (t) .14* (.02)  .12* (.02)  .11* (.02)  .15* (.02)  .00 (.03)  Evaluation (t) .04* (.02)  -.05* (.01)  .00 (.01)  -.01 (.01)  -.10* (.02) Model fit                N 3452  3474  3294  3356  3107  Model χ2 26.79  14.39  14.44  3.60  185.06  Degrees of freedom 5  5  5  5  5  RMSEA 0.04  0.02  0.02  0.00  0.11  CFI 0.98  0.99  0.99  1.00  0.65  SRMR 0.01  0.01  0.01  0.00  0.02                Nuclear power                Actual party position (t) .47* (.13)  .93* (.18)  .73* (.10)     1.42* (.30)  Distance * Evaluation (t) .09* (.02)  .09* (.02)  .11* (.02)     .08* (.03)  Distance (t) .09* (.02)  .10* (.02)  .11* (.02)     .16* (.03)  Evaluation (t) .01 (.02)  -.02 (.02)  -.01 (.02)     -.03 (.03) Model fit                N 3029  3040  2333    1248  Model χ2 11.61  25.09  5.55    10.43  Degrees of freedom 5  5  5    5  RMSEA 0.02  0.04  0.01    0.03  CFI 0.96  0.91  1.00    0.98  SRMR 0.01  0.01  0.01    0.01      145 Table 4.12 continued Test of following (”projection”): Germany, all issues  CDU  SPD  FDP  Green  Left  Coef. SE  Coef. SE  Coef. SE  Coef. SE  Coef. SE European Unification                Actual party position (t) 1.27* (.08)  .95* (.12)  1.22* (.14)  .99* (.14)  .44+ (.25)  Distance * Evaluation (t) .11* (.02)  .07* (.02)  .05* (.02)  .12* (.02)  .13* (.03)  Distance (t) .22* (.02)  .14* (.02)  .19* (.02)  .12* (.02)  .09* (.03)  Evaluation (t) .06* (.02)  .03 (.02)  .04+ (.02)  .02 (.02)  .14* (.03) Model fit                N 2958  2951  2528  2518  1529  Model χ2 5.59  7.88  5.74  8.94  12.32  Degrees of freedom 5  5  5  5  5  RMSEA 0.01  0.01  0.01  0.02  0.03  CFI 1.00  0.99  1.00  0.99  0.98  SRMR 0.01  0.01  0.01  0.01  0.01 Notes: This Table reports unstandardized maximum likelihood (ML) coefficients, based on raw data and estimated in a series of structural equation models in which the dependent variable is a voter’s perceived party position on an issue at time t, as a function of a party’s actual position on this issue, voter-party distance on this issue, party evaluation, and the interaction between distance and evaluation. Standard errors are reported in parentheses.  +p < .10, *p < .05.    146  Table 4.13 Summary: Net leading and following effects in the UK  Conservative Labour Liberal-Democrats Leading    Unemployment — + — Welfare — — + Redistribution  + + Nationalization  +  EU — +      Following Pers. Proj. Pers. Proj. Pers. Proj. Unemployment + + + + + + Welfare + + + + + + Redistribution + + + + + + Nationalization +  +  +  EU + + + + + + Notes: This Table summarizes the results reported in the previous tables. The reported effects for leading are net effects, with plus signs indicating that leading is stronger than following. Plus signs for persuasion and projection represent evidence for these effects.         147  Table 4.14 Summary: Net leading and following effects in Germany  CDU SPD FDP Green Left Leading      Left-Right — —  + — Nuclear power  +  + — EU  +  +        Following Pers. Proj. Pers. Proj. Pers. Proj. Pers. Proj. Pers. Proj. Left-Right + + + +  + + + + + Nuclear power + +  +  +    + EU + +  +  +  +  + Notes: This Table summarizes the results reported in the previous tables. The reported effects for leading are net effects, with plus signs indicating that leading is stronger than following. Plus signs for persuasion and projection represent evidence for these effects.       148 Chapter 5: Conclusion Representative democracy relies on a set of relations between citizens and political parties, and classic accounts of democracy maintain that competitive elections provide the key mechanism for political representation: they authorize politicians, ensure their responsiveness to the people, and hold them accountable for their actions (Pitkin 1967). In other words, “representation is understood as a principal-agent relationship, in which the principals—constituencies formed on a territorial basis—elect agents to stand for and act on their interests and opinions, thus separating the sources of legitimate power from those who exercise that power” (Urbinati and Warren 2008: 389).  The ideal is clear, but is this how democratic representation works in practice? The basic premise for the “delegate model of representation” (Dahl 1956; 1971) is that citizens hold meaningful policy preferences. They are necessary for voters to make rational voting decisions and elect parties or candidates that represent them. Likewise, firm opinions are necessary to hold parties accountable for their actions and promises. Only if voters have independently formed policy view do parties have incentives to be responsive. If voters do not hold informed opinions, the model collapses. Voters who are slavish followers of their party upset the principle of “separating the sources of legitimate power from those who exercise that power” (Urbinati and Warren 2008: 389). That is, if voters simply adopt party views, political party elites become both the source of power and its executors.   Hence, if we seek to understand the reality of political representation we need to investigate how voters form policy opinions and party preferences. In particular, we need to understand the dynamic relationship between parties and voters: Who leads? Who follows?      149  This is, of course, the guiding question of this dissertation. Answering it is no easy task. One of the key challenges that attitudinal studies interested in causal relationships face is the problem of observational equivalence (Lenz 2012). The cognitive processes behind voters’ opinion and preference formation are not directly observable, and their outcomes are observationally equivalent at any given point in time. Empirically identifying these processes in the real world requires panel data and suitable estimation techniques. Through careful theorizing and methodological sophistication, this dissertation overcomes some of the hurdles and makes a number of important contributions to the literatures on political representation and opinion formation. 5.1 Contributions Conceptually, this dissertation bridges some of the gaps between European and American scholarship on political representation and opinion formation. Primarily, it applies American concepts and ideas to entrenched multiparty systems—a task that requires conceptual innovation and adaptation to account for the more complex dynamics in multiparty systems.   Methodologically, the dissertation makes important contributions in three areas: First, the analyses have a longitudinal—rather than the standard cross-sectional—design. For the comparative analysis of polarization and depolarization the advantage is that I can trace party and voter positions over time, thus detecting longitudinal trends and patterns. At lower levels of aggregation, a longitudinal setup allows me to identify the temporal order of party and voter movements, thus enabling me to address the causal question of leading and following. Lastly, at the level of individual voters, the use of panel data, containing repeated observations of the same individuals, lets me trace opinion and attitude change over time, thus allowing me to overcome problems of observational equivalence.     150 Second, Chapter 2 introduces a new indicator (L-kurtosis) to the existing catalogue of polarization measures. Unlike the SD—the standard indicator used in multi-party contexts—the L-kurtosis reveals information about the shape of a distribution, especially the thickness of the tails. Thus, it captures the presence of voters or parties on the distant flanks—a potential threat to which the SD is insensitive.  Third, the analysis of panel data (Chapter 4) represents an improvement to existing works. The use of dynamic SEM panel models with reciprocal effects allows me to test the direction and strength of different effects. Furthermore, these models are better suited than most previously employed ones to draw causal inferences, because they control for unobserved heterogeneity.  Empirically, each of the three chapters fills a gap in the literature and advances our understanding of party-voter relations in European multiparty systems. Each one inspects a different dimension of representation and opinion formation, and the next sections summarize the key findings in chapter order. 5.1.1 Polarization and depolarization Chapter 2 contributes to our knowledge on political representation and opinion formation in two ways. First, it investigates political representation at the level of party system and electorates. Tracking party and voter positions over the course of four decades sheds brighter light on dynamic distributional representation (Ezrow 2007; Pitkin 1967). Second, it fills a gap in the literature by providing a comprehensive Europe-wide account of polarization and depolarization for parties and voters. Since party system polarization has been shown to affect party-voter relations—at least in the US—the results are relevant for our understanding of leading and following in later chapters.     151  We find a Europe-wide depolarization (or de-radicalization) trend for parties and voters, and along both social and economic sub-dimensions of Left-Right. Applying the L-kurtosis measure shows that changes within the electorate occur in two areas: the centre of the opinion distribution becomes more crowded, while the tails of the distribution become fatter. This finding can help explain the success of extreme (right-wing) parties despite this being a period of depolarization. As existing parties converge in the centre, they leave voters located in the flanks unrepresented. Furthermore, they create empty spaces that invite the appearance of new parties on the wings. Likewise, party convergence on one dimension creates opportunities for party competition along other dimensions. In the European case, convergence on social and economic Left-Right questions heats up contestation over Europe and multiculturalism. This goes to show that depolarization has a more complex meaning when there are many parties.  Polarization can also refer to the alignment of issue dimensions (or constraint). When looking at the correlations between social and economic issue positions, we find increasing alignment over time for parties, but not for voters. Combined with the de-radicalization finding, this further strengthens the evidence for a depolarization trend among parties, but not voters. For the other two issue dimensions—European unification and multiculturalism—we find the opposite trend: polarization. Especially after 2000, party positions have become more dispersed, and both dimensions have become increasingly aligned. Yet, we see little evidence for either dimension becoming more aligned with Left-Right.   These results are interesting in their own rights, as they provide the first Europe-wide assessment of positional party and voter movements over such a long time period. They also shed light on the patterns of electoral competition that seem to be shifting away from Left-Right    152 and towards European immigration and multiculturalism. This increasing salience of other dimensions can help explain the growth of right-wing populist party success. 5.1.2 Parties: Leading and following Chapter 3 examines how party system polarization and depolarization affect the relationship between parties and their supporters. The analysis tracks party and voter positions over multiple elections and issue dimensions, focusing on the causal mechanism that links party system movement to changes within the electorate. The results suggest that the mechanism proposed for the US has limited applicability to Europe. For the UK, I find that, if parties lead by polarizing, voters follow by sorting. Likewise, if parties lead by depolarizing, voters sometimes follow by unsorting. However, instances of parties leading occur only for the Labour party, on three issue dimensions—welfare, redistribution, and European unification. The Conservative party tends to follow its voters.   Thinking through the causal mechanism that links polarization to sorting from the perspective of proximity and directional models reveals that sorting is most likely to occur in polarizing systems, where the two dominant parties are also the most extreme ones. This is the case for the US in general and for the UK on European unification. The opposite—depolarization and unsorting—occurs only under certain conditions. These insights suggest that a massive increase in conceptual sophistication is required to deal with aggregate notions of leading and following and to understand the consequences of polarization and depolarization in multiparty systems. The US is a rare example of a party system with only two parties, and directional assumptions are buried in the simple two-party case. To apply these concepts to contexts with more parties, these assumptions need to be dealt with explicitly. The discussion in Chapter 3 takes a first stab at it.    153  In addition, relating polarization and sorting to classic models of representation produces interesting insights about the meaning of sorting and suitable measures. In the American literature, sorting is broadly defined as a tighter alignment of party and ideology. This definition is misleading, as it suggests that the level of representation increases as parties polarize and voters respond by sorting. However, the opposite is the case: the quality of representation is poorer when parties lead towards extreme positions and voters sort around the centre. Perversely, voters may feel more attached to a party, but they are less well represented in the aggregate. Measuring sorting is even more problematic. The most common indicator of sorting is the correlation between partisanship and ideology. This is despite their being proof that correlational indicators are inadequate to capture the strength of this relationship. These insights are not new but have long been ignored. This dissertation puts them back on the table. 5.1.3 Voters: Leading and following  Chapter 4 investigates party-voter relations from the perspective of individual voters in multiparty systems. The focus is on cognitive processes of opinion and preference formation. The results confirm that the relationship between parties and voters is reciprocal in nature. In other words, I find reciprocal effects between voters’ issue positions, voters’ perceptions of parties’ issue positions, and voters’ party preferences. Separate tests of leading—policy oriented party evaluation or issue voting—and following—persuasion and projection—allow me to estimate the direction and strength of these effects.  The tests of leading present a vast improvement to standard models of issue voting, as they allow me to isolate the causal impact of policy distances on party preference. My models control for effects in the opposite direction (from party preference to policy distance) and “difference-out” all confounding factors. According to my findings, the impact of policy    154 considerations on party preference is rather small. This challenges conventional wisdom in Europe but points to parallels in the US. Issue voting, which has been contested in the US since the dawn of polling, is somehow just assumed to capture the European landscape. It turns out, that is not true.  The results of the persuasion and projections tests provide solid evidence for voters as followers in European multiparty systems. These findings support the claim that the American concept of partisanship applies in Europe to a greater extent than scholars have acknowledged. Voters’ own issue positions and voters’ perceptions of the parties’ positions are conditional on party preference.   5.2 Unanswered questions and future research One of the most obvious questions concerns the effects of party system structure on the dynamics of leading and following. The analyses in Chapters 2 and 3 suggest that there might be links, but testing this proposition is difficult, given the availability of suitable data. Due to the shortage of multi-wave panel data I face the challenge of too many variables and too few cases (countries) to draw meaningful causal inferences about the effects of party system structure. There is, of course, a trade-off between the in-depth exploration that is possible with the BES and GLES panels and the opportunity to generalize across systems. Cross-national datasets like those generated by the Comparative Study of Electoral Systems (CSES) allow for Large-N tests of country-level effects. I opted for the small-n alternative to better understand dynamic processes that occur at the level of individuals and unfold over time. As additional panel datasets become available, it would be interesting to not only analyze more cases, but also to explore different time lags to learn more about the temporal dimension of opinion and preference formation (e.g., within-campaign vs. between campaigns).      155 Moreover, I do not engage in analyses of different subgroups within the electorate—apart from distinguishing between party voters and party identifiers. Thus, my results cannot speak to differences in opinion and preference formation by gender, education, or class. Previous research has shown systematic representation gaps for women, and citizens with lower levels of education and income (e.g., Bartels 2008; Gidengil et al. 2004). An extension of this research to explore variation in leading and following seems a promising avenue to improve our understanding of representation.  This thesis focuses on positional issues and, to that extent, buys into a quite intellectualized account of elections. Considering that issue voting, as suggested by proximity models, is a rather demanding task that requires a lot of information and somewhat abstract calculations, it may not be surprising that voters take cues from parties. What if voters evaluate parties on other criteria, such as issue priorities or past performance? Research has shown that, here too, the relationship is reciprocal (see Neundorf and Adams (2016) on issue priorities; Lenz (2012) on past performance). Lastly, valence issues, more generally, affect party evaluation and vote choice. These include the economy and other managerial matters, but also candidate and leader traits. It is clear that persuasion and perceptual bias matter a lot for these issues (e.g., Bartels 2002; Czhen et al. 2014; Evans and Pickup 2010).   Taken together, these findings point to a major democratic deficit. If voters mainly follow, they give up their power to democratic governance and hand a wild card to political elites. Since following means that parties shape voters’ attitudes and perceptions, voters may not even notice if parties do not represent their true interests. Then the separation of “the sources of legitimate power from those who exercise that power” vanishes (Urbinati and Warren 2008: 389).    156 References Abramowitz, Alan I. 2008. “Is Polarization a Myth?” Journal of Politics 70: 542-55.  Abney, Romni, Andrea Morrison, and Gary Stradiotto. 2007. “The Stability of Representation: A Cross-National Analysis of Party Policy Dispersion.” Representation 43(3): 151-65.  Achen, Christopher H. 1977. “Measuring Representation: Perils of the Correlation Coefficient.” American Journal of Political Science 21: 805-815. Achen, Christopher, H. 1978. “Measuring Representation.” American Journal of Political Science 22(3): 475-510.  Adams, James. 2001. Party Competition and Responsible Party Government: A Theory of Spatial Competition Based upon Insights from Behavioral Voting Research. Ann Arbor: The University of Michigan Press.  Adams, James, Michael Clark, Lawrence Ezrow, and Garrett Glasgow. 2004. “Understanding Change and Stability in Party Ideologies: Do Parties Respond to Public Opinion or to Past Election Results?” British Journal of Political Science 34(4): 589-610.  Adams, James, Michael Clark, Lawrence Ezrow, and Garrett Glasgow. 2006. “Are Niche Parties Fundamentally Different from Mainstream Parties? The Causes and Electoral Consequences of Western European Parties' Policy Shifts, 1976-1998.” American Journal of Political Science 50(3): 513-29.  Adams, James, Catherine DeVries, and Debra Leiter. 2012. “Subconstituency Reactions to Elite Depolarization in the Netherlands: An Analysis of the Dutch Public's Policy Beliefs and Partisan Loyalties, 1986-98.” British Journal of Political Science 42: 81-105.  Adams, James, and Lawrence Ezrow. 2009. “Who Do European Parties Represent? How Western European Parties Represent the Policy Preferences of Opinion Leaders.” Journal of Politics 71(1): 206-23.  Adams, James, Jane Green, and Caitlin Milazzo. 2012. “Has the British Public Depolarized along with Political Elites? An American Perspective on British Public Opinion.” Comparative Political Studies 45: 507-30.  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 42(5): 611-39.  Adams, James, and Samuel Merrill. 2006. “Why Small, Centrist Parties Motivate Policy Divergence by Major Parties.” American Political Science Review 100(3): 403-17.     157 Adams, James, Samuel Merrill, and Bernard Grofman. 2005. A Unified Theory of Party Competition: A Cross-National Analysis Integrating Spatial and Behavioral Factors. Cambridge, UK: Cambridge University Press. Allison, Paul. 2009. Fixed Effects Regression Models. Sage Publications. Alvarez, R. Michael, and Jonathan Nagler. 1995. “Economics, Issues and the Perot Candidacy: Voter Choice in the 1992 Presidential Election.” American Journal of Political Science 39: 714-744. Alvarez, R. Michael, and Jonathan Nagler. 1998. “Economics, Entitlements and Social Issues: Voter Choice in the 1996 Presidential Election.” American Journal of Political Science 42: 1349-1363. Alvarez, R. Michael, and Jonathan Nagler. 2004. “Party System Compactness: Consequences and Measures.” Political Analysis 12: 46-62.  Anderson, Christopher J. 1998. “Parties, Party Systems, and Satisfaction with Democratic Performance in The New Europe.” Political Studies XlVI: 572-588. Anderson, Christopher J. 2009. “The Interaction of Structures and Voter Behaviour.” In The Oxford Handbook of Political Behavior, eds. H.-D. Klingemann and R. J. Dalton. Online ed., Oxford University Press.  Anderson, Christopher J. 2012. “How Electoral Systems Shape What Voters Think About Democracy.” Presented at the conference on “Duty and Choice: Participation and Preferences in Democratic Elections” on the occasion of Andre Blais’ 65th birthday. 20-21 January, 2012. University de Montreal. Anderson, T. W., and C. Hsiao. 1981. “Estimation of Dynamic Models with Error Components.” Journal of the American Statistical Association 76: 598-606. Anderson, T. W., and C. Hsiao. 1982. “Formulation and Estimation of Dynamic Models Using Panel Data.” Journal of Econometrics 18: 47-82. Balanda, K. P., and H. L. MacGillivray. 1988. “Kurtosis: A Critical Review.” The American Statistician 42: 111–119. Baldassarri, Delia, and Andrew Gelman. 2008. “Partisans without Constraint: Political Polarization and Trends in American Public Opinion.” American Journal of Sociology 114: 408-46.  Baltagi, B. H. 2005. Econometric Analysis of Panel Data. London: Wiley.    158 Bardi, Luciano, and Peter Mair. 2008. “The Parameters of Party Systems.” Party Politics 14: 147-166. Bartels, Larry M. 2002. “Beyond the Running Tally: Partisan Bias in Political Perceptions.” Political Behavior 24: 117-50.  Bartels, Larry M. 2008. Unequal Democracy: The Political Economy of the New Gilded Age. Princeton and Oxford: Princeton University Press.  Bartels, Larry M. 2013. “Party Systems and Political Change in Europe.” Presented at the Annual Meeting of the American Political Science Association, Chicago. Benoit, Kenneth, and Michael Laver. 2006. Party Policy in Modern Democracies. London: Routledge. Benoit, Kenneth, and Michael Laver. 2007. “Estimating Party Policy Positions: Comparing Expert Surveys and Hand-Coded Content Analysis.” Electoral Studies 26: 90–107.  Berelson, Bernard, Paul Lazarsfeld, and William McPhee. 1954. Voting: A Study of Opinion Formation in a Presidential Campaign. Chicago: University of Chicago Press.  Best, Robin E., and Jay Dow. 2014. “Measuring Party System Polarization: Strategies and Performance.” Presented at the Annual Meeting of the Midwest Political Science Association, Chicago.  Blais, Andre, and Marc A. Bodet. 2006. “Does Proportional Representation Foster Closer Congruence between Citizens and Policy Makers?” Comparative Political Studies 39: 1243-1262. Blais, Andre, Elisabeth Gidengil, Richard Nadeau, et al. 2001. “The Formation of Party Preferences: Testing the Proximity and Directional Models.” European Journal of Political Research 40(1): 81-91. Bohrnstedt, G. W. 1969. “Observations on the Measurement of Change.” In Sociological Methodology 1969, ed. E. F. Borgatta. San Francisco: Jossey-Bass, 112-136. Brader, Ted, and Joshua A. Tucker. 2012. “Following the Party’s Lead: Party Cues, Policy Opinion, and the Power of Partisanship in Three Multiparty Systems.” Comparative Politics 44(4): 403-420. Brady, Henry E., and Richard Johnston, eds. 2006. Capturing Campaign Effects. Ann Arbor: The University of Michigan Press. Brody, Richard A., and Benjamin I. Page. 1972. “Comment: The Assessment of Policy Voting.” The American Political Science Review 66(2): 450-458.    159 Budge, Ian. 1999. “Party Policy and Ideology: Reversing the 1950s?” In Critical Elections: British Parties and Voters in Long-term Perspective, eds. G. Evans and P. Norris. London: Sage, 1-21. Budge, Ian, and Michael D. McDonald. 2007. “Election and Party System Effects on Policy Representation: Bringing Time into Comparative Perspective.” Electoral Studies 26(1): 168-179. Butler, David, and Donald Stokes. 1969. Political Change in Britain. London: Macmillan. Campbell, Angus, Philip Converse, Warren Miller, and Donald Stokes. 1960. The American Voter. New York: John Wiley & Sons, Inc.  Carmines, Edward G., and James A. Stimson. 1989. Issue Evolution: Race and the Transformation of American Politics. Princeton, NJ: Princeton University Press. Carruba, Clifford. 2001. “The Electoral Connection in European Union Politics.” Journal of Politics 63(1): 141-158. Carsey, Thomas, and Geoffrey Layman. 2006. “Changing Sides or Changing Minds? Party Identification and Policy Preferences in the American Electorate.” American Journal of Political Science 50(2): 464-477. Chong, Dennis, and James N. Druckman. 2007. “Framing Public Opinion in Competitive Democracies.” The American Political Science Review 101(4): 637-655. Chong, Dennis, and James N. Druckman. 2010. “Dynamic Public Opinion.” American Political Science Review 104(4): 663-680. Chong, Dennis, and James N. Druckman. 2011. “Identifying Frames in Political News.” In Sourcebook for Political Communication Research, eds. Erik P. Bucy and R. Lance Holbert. New York: Routledge, 238-267. Chzhen, Kat, Geoffrey Evans, and Mark Pickup. 2014. “When do Economic Perceptions Matter for Party Approval? Examining the Endogeneity of Economic Perceptions Before and During the Economic Downturn.” Political Behavior 36: 291-313. Clarke, Harold D., David Sanders, Marianne C. Stewart, and Paul F. Whiteley. 2004. Political Choice in Britain. Oxford: Oxford University Press. Clarke, Harold D., David Sanders, Marianne C. Stewart, and Paul F. Whiteley. 2009. Performance Politics and the British Voter. New York: Cambridge University Press.  Clements, Ben, and John Bartle. 2009. “The European Issue and Party Choice at British General Elections, 1974-2005.” Journal of Elections, Public Opinion and Parties 19(4): 377-411.    160 Cleveland, W. S. 1979. “Robust Locally Weighted Regression and Smoothing Scatterplots.” Journal of the American Statistical Association 74: 829-36. Converse, Philip E. 1964. “The Nature of Belief Systems in Mass Publics.” In Ideology and Discontent, ed. D. E. Apter. New York: Free Press, 206-261. Converse, Philip E., and R. Pierce. 1986. Political Representation in France. Cambridge, MA: Harvard University Press. Cox, Gary W. 1997. Making votes count: Strategic coordination in the world's electoral systems. Cambridge, UK: Cambridge University Press.  Cutler, Fred. 2002. “The Simplest Shortcut of All: Sociodemographic Characteristics and Electoral Choice.” The Journal of Politics 64(2): 466-90.  Dahl, Robert. 1956. A Preface to Democratic Theory. Chicago, IL: University of Chicago Press.  Dahl, Robert. 1971. Polyarchy: Participation and Opposition. New Haven, CT: Yale University Press.  Dalton, Russell J. 1985. “Political Parties and Political Representation.” Comparative Political Studies 17: 267-99.  Dalton, Russell J. 1996. Citizen Politics: Public Opinion and Political Parties in Advanced Western Democracies. Chatham, NJ: Chatham House. Dalton, Russell J. 2008. “The Quantity and the Quality of Party Systems: Party System Polarization, Its Measure, and Its Consequences.” Comparative Political Studies 41: 899-920. Dalton, Russell J., and Martin P. Wattenberg, eds. 2000. Parties without Partisans: Political Change in Advanced Industrial Democracies. Oxford: Oxford University Press.  Dancey, Logan, and Paul Goren. 2010. “Party Identification, Issue Attitudes, and the Dynamics of Political Debate.” American Journal of Political Science 54(3): 686-699. Davis, O. A., M. J. Hinich, and P. Ordeshook. 1970. “An Expository Development of a Mathematical Model of the Electoral Process.” American Political Science Review 64: 426-448. Delli Carpini, Michael X., and Scott Keeter. 1996. What Americans Know About Politics and Why It Matters. New Haven, CT: Yale University Press.  Downs, Anthony. 1957. An Economic Theory of Democracy. Harper.     161 Druckman, James N., Erik Peterson, and Rune Slothuus. 2013. “How Elite Partisan Polarization Affects Public Opinion Formation.” American Political Science Review 107(1): 57-79. Duverger, Maurice. 1963[1954]. Political Parties. Taylor and Francis.  Enelow, J. M., and M. J. Hinich. 1984. The Spatial Theory of Voting: An Introduction. CUP Archive. Evans, Geoffrey, and Robert Andersen. 2004. “Do Issues Decide? Partisan Conditioning and Perceptions of Party Issue Positions Across the Electoral Cycle.” British Elections and Parties Review 14(1): 18-39. Evans, Geoffrey, and Mark Pickup. 2010. “Reversing the Causal Arrow: The Political Conditioning of Economic Perception in the 2000-2004 U.S. Presidential Election Cycle.” The Journal of Politics 72 (4): 1236-51.  Ezrow, Lawrence. 2007. “The Variance Matters: How Party Systems Represent the Preferences of Voters.” The Journal of Politics 69: 182-92.  Ezrow, Lawrence, Catherine De Vries, Marco Steenbergen, and Erica Edwards. 2011. “Mean Voter Representation and Partisan Constituency Representation: Do Parties Respond to the Mean Voter Position or Their Supporters?” Party Politics 17(3): 275-301.  Ezrow, Lawrence, Margit Tavits, and Jonathan Homola. 2014. “Voter Polarization, Strength of Partisanship, and Support for Extremist Parties.” Comparative Political Studies 47: 1558-83. Ezrow, Lawrence, and Georgios Xezonakis. 2011. “Citizen Satisfaction With Democracy and Parties’ Policy Offerings.” Comparative Political Studies 44: 1152-1178. Falter, Juergen, Oskar W. Gabriel, Hans Rattinger, Klaus Schmitt. 2012. Political Attitudes, Political Participation and Voting Behaviour in Reunified Germany (Panel 1994-2002). GESIS Data Archive, Cologne. ZA4301 Data file Version 1.1.0, doi:10.4232/1.11408.  Feld, Werner, and John Wildgen. 1976. Domestic Political Realities and European Unification: A Study of Mass Publics and Elites on the European Community Countries. Boulder, CO: Westview Press. Festinger, Leon. 1957. A Theory of Cognitive Dissonance. Stanford, CA: Stanford University Press.  Finkel, Steven E. 1995. Causal Analysis With Panel Data. Sage University Paper Series on Quantitative Applications in the Social Sciences. Beverly Hills: Sage Publications.    162 Finkel, Steven E. 2008. “Linear Panel Analysis.” In Handbook of Longitudinal Research, ed. Scott Menard. New York: Elsevier Press.  Finkel, Steven E. 2015. Course material prepared for “Analysis of Panel Data”, University of Gothenburg, June 15-18, 2015. Retrieved from: https://gul.gu.se/public/courseId/70156/lang-en/publicPage.do?item=29873131 Fiorina, Morris P., Abrams, S., and Pope, J. 2005. Culture War? The Myth of a Polarized America. New York, IL: Pearson Education. Fiorina, Morris P., and Matthew S. Levendusky. 2006. “Disconnected: The Political Class versus the People.” In Red and Blue Nation? Characteristics and Causes of America’s Polarized Politics, Vol. 1, eds. P. S. Nivola and D. W. Brady. Washington, DC: Brookings Institute Press, 49-57. Franklin, Mark N., Thomas T. Mackie, and Henry Valen. 2009. Electoral Change. Responses to Evolving Social and Attitudinal Structures in Western Countries. Colchester: ECPR Press. Frees, E. W. 2004. Longitudinal and Panel Data: Analysis and Applications in the Social Sciences. Cambridge, UK/New York: Cambridge University Press.  Gabel, Matthew. 1998. “Public Support for European Integration: An Empirical Test of Five Theories.” Journal of Politics 60(2): 333-354. Gabel, Matthew, and Kenneth Scheve. 2007. “Estimating the Effect of Elite Communications on Public Opinion Using Instrumental Variables.” American Journal of Political Science 51(4): 1013-1028. Gidengil, Elisabeth, Andre Blais, Neil Nevitte, and Richard Nadeau. 2004. Citizens. Vancouver: UBC Press. Golder, Matt, and Jacek Stramski. 2010. “Ideological Congruence and Electoral Institutions.” American Journal of Political Science 54: 90-106. Goren, Paul. 2005. “Party Identification and Core Political Values.” American Journal of Political Science 49(4): 882-97.  Goren, Paul, Christopher M. Federico, and Miki Caul Kittilson. 2009. “Source Cues, Partisan Identities, and Political Value Expression.” American Journal of Political Science 53(4): 805-20.  Green, Donald, Bradley Palmquist, and Eric Schickler. 2002. Partisan Hearts and Minds: Political Parties and the Social Identities of Voters. New Haven: Yale University Press.     163 Green, Donald, and Bradley Palmquist. 1990. “Of Artifacts and Partisan Stability.” American Journal of Political Science 34: 872-902.  Heath, A., R. Jowell, and J.K. Curtice. 1998. British Election Panel Study, 1992-1997 [computer file]. Colchester, Essex: UK Data Archive [distributor]. Heath, A., R. Jowell, and J.K. Curtice. 2002. British Election Panel Study, 1997-2001; Waves 1 to 8  [computer file]. 4th Edition. Colchester, Essex: UK Data Archive [distributor]. Hetherington, Marc J. 2001. “Resurgent Mass Partisanship: The Role of Elite Polarization.” American Political Science Review 95: 619-631. Hetherington, Marc J. 2008. “Turned Off or Turned On: The Effects of Polarization on Political Participation, Engagement, and Representation.” In Red and Blue Nation? Vol. II, eds. Nivola and Brady, 1-33. Hetherington, Marc J. 2009. “Review Article: Putting Polarization in Perspective.” British Journal of Political Science 39: 413-448. Highton, Benjamin, and Cindy Kam. 2011. “The Long-term Dynamics of Partisanship and Issue Orientations.” Journal of Politics 73(1): 202-215. Hillygus, D. Sunshine, and Todd G. Shields. 2008. The Persuadable Voter: Wedge Issues in Presidential Campaigns. Princeton, NJ: Princeton University Press. Hosking, J. R. M. 1990. “L-Moments: Analysis and Estimation of Distributions Using Linear Combinations of Order Statistics.” Journal of the Royal Statistical Society. Series B (Methodological) 52: 105-124. Hosking, J. R. M. 1992. “Moments or L Moments? An Example Comparing Two Measures of Distributional Shape.” The American Statistician 46: 186-189.  Hsiao, Cheng. 2002. Analysis of Panel Data. Cambridge, UK: Cambridge University Press. Huber, John. D., and G. Bingham Powell. 1994. “Congruence between Citizens and Policy Makers in Two Versions of Liberal Democracy.” World Politics 46: 291-26.  Huber, John D., Georgia Kernell, and Eduardo Leoni. 2007. “Institutional Context, Cognitive Resources, and Party Attachments Across Democracies.” Political Analysis 13: 365-86.  Inglehart, Ronald. 1977. The Silent Revolution: Changing Values and Political Styles among Western Publics. Princeton, NJ: Princeton University Press.    164 Jansen, Giedo, Geoffrey Evans, and Nan Dirk de Graaf. 2013. “Class Voting and Left–Right Party Positions: A Comparative Study of 15 Western Democracies, 1960–2005.” Social Science Research 42: 376-400. Johnston, Richard. 2006. “Party Identification: Unmoved Mover or Sum of Preferences?” Annual Review of Political Science 9: 329-51.  Johnston, Richard, Patrick Fournier, and Richard Jenkins. 2000. “Party Location and Party Support: Unpacking Competing Models.” Journal of Politics 62(4): 1145-1160. Jones, B. D., Baumgartner, F. R., Breunig, C., Wlezien, C., Soroka, S., Foucault, M., François, A., Green-Pedersen, C., Koski, C., John, P., Mortensen, P. B., Varone, F. & Walgrave, S. 2009. “A General Empirical Law of Public Budgets: A Comparative Analysis.” American Journal of Political Science 53: 855–873. Kedar, Orit. 2005. “When Moderate Voters Prefer Extreme Parties: Policy Balancing in Parliamentary Elections.” American Political Science Review 99(2): 185-199. Kedar, Orit. 2009. Voting for Policy, Not Parties: How Voters Compensate for Power Sharing. Cambridge: Cambridge University Press. Kessler, R. C., and D. F. Greenberg. 1981. Linear Panel Analysis: Models of Quantitative Change. New York: Academic Press. Kim, Myunghee. 2009. “Cross-national Analyses of Satisfaction With Democracy and Ideological Congruence.” Journal of Elections, Public Opinion and Parties 19(1): 49-72. Kirchheimer, Otto. 1966. “The Transformation of the Western European Party Systems.” In Political Parties and Political Development, ed. M. Weiner and J. LaPalombara. Princeton, NJ: Princeton University Press. Kitschelt, Herbert, with Anthony J. McGann. 1997. The Radical Right in Western Europe: A Comparative Analysis. Ann Arbor: University of Michigan Press. Kitschelt, Herbert. 2000. “Linkages between Citizens and Politicians in Democratic Polities.” Comparative Political Studies 33: 845-879. Laakso, Markkus, and Rein Taagepera. 1979. “Effective Number of Parties: A Measure With Application to West Europe. Comparative Political Studies 12: 3-27.   Layman, Geoffrey C. 2001. The Great Divide: Religious and Cultural Conflict in American Party Politics. New York, NY: Columbia University Press. Layman, Geoffrey C., and Thomas M. Carsey. 2002. “Party Polarization and ‘Conflict Extension’ in the American Electorate.” American Journal of Political Science 46: 786-802.    165 Lazarsfeld, Paul F., Bernard Berelson, and Hazel Gaudet. 1944. The People’s Choice: How the Voter Makes Up His Mind in a Presidential Campaign. New York: Columbia University Press. Lenz, Gabriel S. 2012. Follow the Leader? How Voters Respond to Politicians' Policies and Performances. Chicago, IL: University of Chicago Press.  Lenz, Gabriel S. 2009. “Learning and Opinion Change, Not Priming: Reconsidering the Priming Hypothesis.” American Journal of Political Science 53: 821-37.  Levendusky, Matthew. 2009. The partisan sort: How liberals became democrats and conservatives became republicans. Chicago, IL: University of Chicago Press.  Lijphart, Arend, 1994. Electoral Systems and Party Systems: A Study of Twenty-Seven Democracies, 1945-1990. New York: Oxford University Press. Lijphart, Arend. 1999. Government forms and performance in thirty-six countries. New Haven, CT: Yale University Press.  Lipset, Seymour, M. 1960. Political Man: The Social Bases of Politics. New York, NY: Doubleday. Lipset, Seymour, and Stein Rokkan. 1967. Party systems and voter alignments: Cross-national perspectives. New York: Free Press. Lupia, Arthur. 1994. “Shortcuts Versus Encyclopedias: Information and Voting Behaviour in California Insurance Reform Elections.” American Political Science Review 88(1): 63-76.  Lupu, Noam. 2015. “Party Polarization and Mass Partisanship: A Comparative Perspective.” Political Behavior 37: 331-356. MacDonald, Stuart Elaine, Ola Listhaug, and Georg Rabinowitz. 1991. “Issues and Party Support in Multiparty Systems.” American Political Science Review 85(4): 1107-1131. Mair, Peter. 1998. Party System Change: Approaches and Interpretations. Oxford: The Claredon Press. Mair, Peter. 2002. “Comparing Party Systems.” In Comparing Democracies 2: New Challenges in the Study of Elections and Voting, eds. L. LeDuc. R. G. Niemi, P. Norris. London: Sage, 88-107. Mair, Peter. 2006. “Party System Change.” In Handbook of Political Parties, eds. R. S. Katz, W. J. Crotty. London: Sage, 63-73.    166 Markus, Gregory B., and Philip E. Converse. 1979. “A Dynamic Simultaneous Equation Model of Electoral Choice.” American Political Science Review 73(4): 1055-1070. Matthews, Steven A. 1979. “A Simple Directional Model of Electoral Competition.” Public Choice 34(2): 141-156. McCarty, Nolan, Keith T. Poole, and Howard Rosenthal. 2006. Polarized America: The Dance of Ideology and Unequal Riches. Cambridge, Massachusetts: The MIT Press.  McDonald, Michael D., and Ian Budge. 2005. Elections, Parties, and Democracy: Conferring the Median Mandate. Oxford, UK: Oxford University Press.  McDonald, Michael D., Silvia M. Mendes, and Ian Budge. 2004. “What Are Elections For? Conferring the Median Mandate.” British Journal of Political Science 34(1): 1-26. Meguid, Bonnie M. 2008. Party Competition between Unequals: Strategies and Electoral Fortunes in Western Europe. Cambridge, UK: Cambridge University Press. Mendelberg, Tali. 2001. The Race Card: Campaign Strategy, Implicit Messages, and the Norm of Equality. Princeton, N.J.: Princeton University Press. Merrill, Samuel, and Bernard Grofman.1999. A Unified Theory of Voting: Directional and Proximity Spatial Models. Cambridge University Press. Milazzo, Caitlin, James Adams, Jane Green. 2012.” Are Voter Decision Rules Endogeneous to Parties' Policy Strategies? A Model with Applications to Elite Depolarization in Post-Thatcher Britain.” Journal of Politics 74: 262-76.  Miller, Warren E., R. Pierce, J. Thomassen, R. Herrera, S. Holmberg, et al. 1999. Policy Representation in Western Democracies. Oxford: Oxford University Press. Miller, Warren E., and Donald E. Stokes. 1963. “Constituency Influence in Congress.” American Political Science Review 57: 165-177. Munzert, Simon, and Paul C. Bauer. 2013. “Political depolarization in German public opinion, 1980-2010.” Political Science Research and Methods 1: 67-89.  Nesselroade, J. R., S. M. Stigler, and P. B. Baltes. 1980. “Regression Toward the Mean and the Study of Change.” Psychological Bulletin 88: 622-637. Neundorf, Anja, and James Adams. 2016. “The Micro-Foundations of Party Competition and Issue Ownership: The Reciprocal Effects of Citizens’ Issue Salience and Party Attachments.” British Journal of Political Science: 1-22. doi:10.1017/S0007123415000642.    167 Norris, Pippa. 1999. “Institutional Explanations for Political Support.” In Critical Citizens: Global Support for Democratic Governance, ed. Pippa Norris. Oxford: Oxford University Press, 217-235. North, Douglass C. 1990. Institutions, Institutional Change, and Economic Performance. New York: Cambridge University Press. North, Douglass C., and Barry R. Weingast. 1989. “Constitutions and Commitment: The Evolution of Institutions Governing Public Choice in Seventeenth-Century England.” Journal of Economic History 49(4): 803-832. Ostrom, Elinor. 1990. Governing the Commons: The Evolution of Institutions for Collective Action. New York: Cambridge University Press. Page, Benjamin I., and Richard A. Brody. 1972. “Policy Voting and the Electoral Process: The Vietnam War Issue.” American Political Science Review 66(3): 979-995. Page, Benjamin I., and Robert Shapiro. 1992. The Rational Public: Fifty Years of Trends in Americans' Policy Preferences. Chicago: University of Chicago Press.  Pitkin, Hanna F. 1967. The Concept of Representation. Berkeley and Los Angeles: University of California Press.  Powell, G. Bingham. 1989. “Constitutional Design and Citizen Electoral Control.” Journal of Theoretical Politics 1: 107-30.  Powell, G. Bingham. 2000. Elections As Instruments of Democracy: Majoritarian and Proportional Views. New Haven, CT: Yale University Press.  Powell, G. Bingham. 2004. “Political Representation in Comparative Politics.” Annual Review of Political Science 7: 273-96.  Powell, G. Bingham. 2006. “Election Laws and Representative Governments: Beyond Votes and Seats.” British Journal of Political Science 36(2): 291-315.  Powell, G. Bingham. 2010. “Party Polarization and the Ideological Congruence of Governments.” In Citizens, Contexts, and Choice: How Context Shapes Citizens’ Electoral Choices, eds. R. J. Dalton and C. J. Anderson. Oxford; New York: Oxford University Press. Powell, G. Bingham, and Georg Vanberg. 2000. “Election Laws, Disproportionality and the Left-Right Dimension.” British Journal of Political Science 30(3): 383-411.  Rabinowitz, George. 1978. “On the Nature of Political Issues: Insights from a Spatial Analysis.” American Journal of Political Science 22(4): 793-817.    168 Rabinowitz, George, and Stuart E. MacDonald. 1989. “A Directional Theory of Issue Voting.” American Political Science Review 83: 93-121. Rabinowitz, George, Stuart E. MacDonald, and Ola Listhaug. 1991. “New Players in an Old Game: Party Strategy in Multiparty Systems.” Comparative Political Studies 24: 147-185. Rae, Douglas W. 1967. The Political Consequences of Electoral Laws. New Haven, CT: Yale University Press. Rae, Douglas W., and Michael Taylor. 1970. The Analysis of Political Cleavages. New Haven, CT: Yale University Press. Rattinger, Hans. 2012. Political Attitudes, Political Participation and Voting Behavior in Reunified Germany (Panel 1998-2005). GESIS Data Archive, Cologne. ZA4662 Data file Version 1.1.0, doi:10.4232/1.11409. Rattinger, Hans, Sigrid Rossteutscher, Ruediger Schmitt-Beck, Bernhard Wessels, Juergen Falter, Oskar Niedermeyer, Tatjana Rudi. 2012. Long-Term Panel 2002-2005-2009 (GLES). GESIS Data Archive, Cologne. ZA5320 Data file Version 2.0.0, doi:10.4232/1.11350. Rattinger, Hans, Sigrid Rossteutscher, Ruediger Schmitt-Beck, Bernhard Wessels, Steffen Kuehnel, Oskar Niedermeyer, Bettina Westle, Tatjana Rudi, and Jan Eric Blumenstiel. 2016. Long-Term Panel 2005-2009-2013 (GLES). GESIS Data Archive, Cologne. ZA5321 Data file Version 2.1.0, doi:10.4232/1.12626. Rattinger, Hans, Sigrid Rossteutscher, Ruediger Schmitt-Beck, Bernhard Wessels, Christof Wolf, Tatjana Rudi, and Jan Eric Blumenstiel. 2016. Long-Term Panel 2009-2013-2017 (GLES 2013). GESIS Data Archive, Cologne. ZA5322 Data file Version 1.1.0, doi:10.4232/1.12627. Ray, Leonard. 2003. “When Parties Matter: The Conditional Influence of Party Positions on Voter Opinions about European Integration.” Journal of Politics 65(4): 978-994. Riker, William H., and Peter C. Ordeshook. 1973. An Introduction to Positive Political Theory. Prentice Hall. Sartori, Giovanni. 1976. Parties and Party Systems: A Framework for Analysis, Volume I. Cambridge: Cambridge University Press.  Sartori, Giovanni. 1968. “Representational Systems.” International Encyclopedia of the Social Sciences 13: 470-5.  Schmitt, Herrmann, and E. Scholz. The Mannheim Eurobarometer trend file, 1970-2002. Prepared by Zentralarchiv fur Empirische Sozialforschung. ICPSR04357-v1. Mannheim, Germany: Mannheimer Zentrum fur Europaeische Sozialforschung and Zentrum fur Umfragen, Methoden und Analysen [producers], 2005. Cologne, Germany: Zentralarchiv    169 fur Empirische Sozialforschung/Ann Arbor, MI: Inter-university Consortium for Political and Social Research [distributors], 2005-12-06. http://doi.org/10.3886/ICPSR04357.v1 Selway, Joel S. 2011. “The Measurement of Cross-cutting Cleavages and Other Multidimensional Cleavage Structures.” Political Analysis 19: 48-65. Shugart, Matthew S., and J. M. Carey. 1992. Presidents and Assemblies: Constitutional Design and Electoral Dynamics. New York: Cambridge University Press. Singh, Shane P. 2014. “Linear and Quadratic Utility Loss Functions in Voting Behavior Research.” Journal of Theoretical Politics 26(1): 35-58. Smith, Kevin B., John R. Alford, John R. Hibbing, Nicholas G. Martin, ad Peter K. Hatemi. 2016. “Intuitive Ethics and Political Orientations: Testing Moral Foundations as a Theory of Political Ideology.” American Journal of Political Science. doi:10.1111/ajps.12255.  Sniderman, Paul M. 2000. “Taking sides: A fixed choice theory of political reasoning.” In Elements of reason: Cognition, choice, and the bounds of rationality., eds. Arthur Lupia, M. D. McCubbins and Samuel L. Popkin. Cambridge, UK: Cambridge University Press, 67-84.  Sniderman, Paul M., Michael G. Hagen, Philip E. Tetlock, and Henry E. Brady. 1986. Reasoning chains: Causal models of policy reasoning in mass publics. British Journal of Political Science 16(4): 405-30.  Steenbergen, Marco R., Erica E. Edwards, and Catherine E. De Vries. 2007. Who's Cueing Whom? Mass--elite Linkages and the Future of European Integration. European Union Politics 8(1): 13-35.  Steenbergen, Marco R., and Bradford S. Jones. 2002. “Modeling Multilevel Data Structures.” American Journal of Political Science 46(1): 218-237. Stimson, James, Michael Mackuen, and Robert Erikson. 1995. “Dynamic Representation.” American Political Science Review 89: 543-65.  Stokes, Donald E. 1966. “Some Dynamic Elements of Contests for the Presidency.” American Political Science Review 60: 19-28. Strom, Kaare. 1990. “A Behavioral Theory of Competitive Political Parties.” American Journal of Political Science 34(2): 565-598. Taagepera, Rein, and Matthew Shugart. 1989. Seats and Votes: The Effects and Determinants of Electoral Systems. New Haven: Yale University Press.  Hall, Peter A., and Rosemary C. R. Taylor. 1996. “Political Science and the Three New Institutionalisms.” Political Studies XLIV: 936-957.    170 Urbinati, Nadia, and Mark Warren. 2008. “The Concept of Representation in Contemporary Democratic Theory.” Annual Review of Political Science 11: 387-412.  Uslaner, Eric M., and T. Zittel. 2006. “Comparative Legislative Behavior.” In Oxford Handbook of Political Institutions, eds. Rod A. W. Rhodes, Sarah Binder and Bert Rockman. Oxford: Oxford University Press, 455-473. Vaisey, Stephen, and Andrew Miles. 2017. “What You Can—and Can’t—Do with Three-Wave Panel Data.” Sociological Methods and Research 46(1): 44-67. Valentino, Nicholas A., Vincent L. Hutchings, and Ismail K. White. 2002. “Cues That Matter: How Political Ads Prime Racial Attitudes During Campaigns.” American Political Science Review 96: 75-90. Van der Eijk, Cees, and Mark N. Franklin, eds. 1996. Choosing Europe? The European Electorate and National Politics in the Face of Union. Ann Arbor: University of Michigan Press. Volkens, Andrea, Pola Lehmann, Nicolas Merz, Sven Regel, Annika Werner, with Lacewell Onawa Promise and Henrike Schultze. 2014. The Manifesto Data Collection. Manifesto Projet (MRG / CMP / MARPOR). Version 2014b. Berlin: Wissenschaftszentrum fuer Sozialforschung (WZB).  Webb, Paul, and David M. Farrell. 1999. “Party Members and Ideological Change.” In Critical Elections: British Parties and Voters in Long-term Perspective, ed. Geoffrey Evans and Pippa Norris. London: Sage Publications. Weissberg, Robert. 1978. “Collective Vs. Dyadic Representation in Congress. American Political Science Review 72(2): 535-47.  Wessels, Bernhard. 1995. “Evaluation of the EC: Elite or Mass Driven.” In Public Opinion and Internationalized Governance, ed. Oskar Niedermeyer and Richard Sinott. Oxford: Oxford University Press, 105-136. Wolinetz, Steven B. 2006. “Party Systems and Party System Types.” In Handbook of Political Parties, eds. Richard Katz and William Crotty. London: Sage, 51-62. Woolridge, Jeffrey M. 2002. Econometric Analysis of Cross-Sectional and Panel Data. Cambridge, MA: MIT Press. Zakharova, Maria, and Paul Warwick. 2014. “The Sources of Valence Judgments: The Role of Policy Distance ad the Structure of the Left-Right Spectrum.” Comparative Political Studies 47(14): 2000-2025.  Zaller, John. 1992. The nature and origins of mass opinion. Cambridge, UK: Cambridge University Press.     171 Appendix 1: Data sources British Election Study (BES) The data for the UK in Chapters 3 and 4 come from two BES panel studies: the 1992-1997 BES Panel Study, and the 1997-2001 BES Panel Study. Both surveys were designed to produce representative samples of the adult population living in private households in Britain. The 1992-1997 panel study includes six waves of interviews, conducted on an annual basis. The policy questions relevant for my analysis were not asked in 1993. The analysis is therefore restricted to five waves. For the 1997-2001 panel study, I analyze all five waves. To cover the longest possible period, I merge both datasets. Question wording Unemployment and inflation: • Some people feel that getting people back to work should be the government’s top priority. These people should put themselves in Box A. Other people feel that keeping prices down should be the government’s top priority. These people should put themselves in Box K. And other people have views somewhere in-between. • In the first row of boxes, please tick whichever box comes closest to your own views about unemployment and inflation.  • Now where do you think the parties stand?  Taxation and government services: • … put up taxes a lot and spend much more on health and social services (A) • … cut taxes a lot and spend much less in health and social services (K)  Nationalization and privatization: • … nationalize many more private companies (A) • … sell off many more nationalized industries (K)  Redistribution: • … make much greater efforts to make people’s incomes more equal (A) • … much less concerned about how equal people’s incomes are (K)  European Community: • … do all it can to unite fully with the European Community (A)    172 • … do all it can to protect its independence from the European Community (K)   Party ID: • Generally speaking, du you think of yourself as Conservative, Labour, Liberal Democrat, or what?  Party evaluation: • Please choose a phrase from this card to say how you feel about [party]: (1) Strongly in favour, (2) In favour, (3) Neither in favour nor against, (4) Against, (5) Strongly against   Eurobarometer (EB) Public opinion data in Chapter 2 come from the Eurobarometer surveys—regular public opinion polls conducted on behalf of the European Commission in all member countries of the European Union. The key data source for the analysis is The Mannheim Eurobarometer Trend File, 1970 -2002, prepared by Hermann Schmitt and Evi Scholz, and updated by Iris Leim and Meinhard Moschner. I extended the time series for questions on Left-Right (until 2007) and post-materialism (until 2005) by amending the trend file with relevant variables form other EB studies.  German Longitudinal Election Study (GLES) The data for Germany in Chapters 3 and 4 come from GLES panel studies. I merged five panel studies (ZA4301, ZA4662, ZA5320, ZA5321, and ZA5322) to create a dataset that spans the period between 1994 and 2013. The earlier studies (ZA4301 and ZA4662) each include three waves of interviews, conducted in election years. The later ones (ZA5320, ZA5321, and ZA5322) also include interviews in some non-election years. For consistency, I only use election year waves for the analysis of the German case.    173 Question wording Left-Right:  • In politics people often talk about “left” and “right”. Using this scale from 1 to 11, how would you assess yourself, provided, “1” means left and “11” means right? • Parties are often assessed as being “left” or “right”. Please tell me how you assess the following parties by using this scale. [1 – left, 11 – right]  Nuclear power: • Let us first turn to the issue of nuclear power. “1” means that more nuclear power stations should be built, whereas “7” means that all nuclear power stations should be closed down immediately. Should more nuclear power stations be built or should all nuclear power stations be closed down immediately? • In your opinion, what are the positions of the political parties on this issue? • What is your personal opinion on the issue of nuclear power?  European unification:  • Should the European unification be further promoted in order to establish a joint government or does European unification already go much too far? • In your opinion, what are the positions of the political parties on this issue? • What is your personal opinion on the issue of European unification?  Party ID: • Do you usually think of yourself as close to any particular party?  Party evaluation: • Scalometer, political parties: I’d like to know what you think about each of our political parties. Please rate it on a scale from -5 to +5, where -5 means that you strongly dislike that party and +5 means that you strongly like that party.  Manifesto Project Dataset (CMP) Party data in Chapter 2 come from the Manifesto Project Dataset, originally created by the Manifesto Research Group (MRG) and later continued under the name Comparative Manifestos Project (CMP). Since 2009 the responsibility for the dataset lies with the project Manifesto Research on Political Representation (MARPOR). The project tracks parties at national elections in 56 countries, between 1920 and the present day. I use version 2014b, published in December 2014.     174 Appendix 2: Additional material Chapter 2 Additional tables and figures, using weighted polarization index Table A.1 Party systems, summary of all dimensions  Polarization index (weighted) Country L-R Econ. Soc. EU MC All -** +** +** +** +** Austria - - - +** +** Denmark - - + + +** Finland -** -** -** +** + France -* -** - + - Germany - - + + +** Great Britain -* -* -** - + Greece - - - + +* Ireland -* - - + +** Italy +* +** + + - Luxembourg - - - - +** Netherlands -* -** - +** +** Norway + - + + + Portugal - +* -** + +** Spain + - - + + Sweden + + - +* - Polarization 1 (4) 2 (3) 0 (4) 4 (13) 8 (12) Depolarization 5 (11) 4 (12) 3 (11) 0 (2) 0 (3) Source: Comparative Manifesto Data Notes: This Table reports the direction of coefficients, estimated in a series of linear regression models in which the dependent variable is a weighted polarization index of party positions on different issue dimensions (as indicated), as a function of time.  * p < 0.1, ** p < 0.05    175  Figure A.1 Party system depolarization: Polarization index  Left-Right Economic Social Dispersion (weighted)    Source: Comparative Manifesto Data Notes: This Figure reports values of weigthed distributional indicators of polarization over time and across different issue dimensions. Solid lines represent the average trend for old democracies (Austria, Denmark, Finland, France, Germany, GB, Ireland, Italy, Luxembourg, Netherlands, Norway and Sweden). Dashed lines illustrate the average movement for new democracies (Greece, Portugal and Spain). All means are smoothed by lowess, bw=0.35.       176 Figure A.2 Party system polarization: Polarization index EU Multiculturalism   Notes: This Figure reports values of weighted distributional indicators of polarization over time and across different issue dimensions. Solid lines represent the average trend for old democracies (Austria, Denmark, Finland, France, Germany, GB, Ireland, Italy, Luxembourg, Netherlands, Norway and Sweden). Dashed lines illustrate the average movement for new democracies (Greece, Portugal and Spain). All means are smoothed by lowess, bw=0.35.      177 Chapter 4  Justification of the estimation model for the German case For Germany, available panel studies from the German Longitudinal Election Study (GLES) include only three waves of data. Unfortunately, a three-wave model with cross-lagged and synchronous effects that controls for unobserved heterogeneity is unidentified (see, for example, Finkel 1995, Chapter 5). I, therefore, cannot use the same model as for the UK case.   Since my focus is on the causal dynamics between parties and voters, I (need to) put a premium on controlling for unobserved heterogeneity. The consequence of this decision is that I can only run separate models for cross-lagged and synchronous effects (no combined model). Ideally in this situation, I estimate two models that are identical in their setup, except for their lag structure. If the lag structure (synchronous or cross-lagged effects) is the only difference between these models, I can compare effect strength and direction. However, there is one caveat: both models make problematic assumptions. The synchronous effects model assumes that there are no cross-lagged effects between X and Y. Similarly, the cross-lagged effects model assumes that synchronous effects between X and Y are 0. Therefore, coefficient sizes (effect magnitudes) in both models may be biased and need to be interpreted with caution.  With that in mind, I first run a synchronous effects model that is as similar as possible to the UK model (see, Figure 4.3). Overall, the results are plausible (see, Table 4.8 to Table 4.10). When I estimate the equivalent model with cross-lagged effects, I encounter the problem described by Vaisey and Miles (2017). The “causal” effects between policy distance and party evaluation switch signs and become implausible. These results suggest that the lag structure in the data does not match the temporal structure of the process I (seek to) model. Given the long lags in the German data (3-4 years), this does not come as a surprise.      178 There is an alternative approach to controlling for unobserved heterogeneity within the SEM framework. It is possible to control for unobservables by including a latent variable with no indicators. This latent error variable (or U term) represents stable omitted unit-level variables and thus stands in for all time-invariant factors that have not been included in the observed data. In practice, U is modeled as influencing all the X and Y over time to capture the possible spuriousness induced by the unobservables (Finkel 1995, Chapter 5; see also Finkel 2015, Slide set 5). According to Finkel (1995: 83), “the main advantage of this approach is that is allows for the direct specification of the assumed interrelationship between measured and unmeasured variables in the causal system and, given enough degrees of freedom with which to identify the parameters, provides the most detailed estimates of the magnitude and direction of all causal effects.” The appropriate dynamic panel model for three waves of data, including cross-lagged effects is depicted in Figure A.3. This model setup does, indeed, work well with cross-lagged effects. The synchronous effects version, however, produces implausible results. To check if this is a problem specific to the GLES data, I run an adapted synchronous effects model with the BES data, but the problem persists. Despite its advantages, this method is not widely used, and examples of nonrecursive model with a latent error variable to control for unobserved factors seem to be nonexistent.  That leaves me with two models—one synchronous and one cross-lagged effects model—producing plausible results. The catch: because of their different specifications, the results are not directly comparable. This forces me to choose one model. For a number of reasons, the synchronous effects model is the more suitable one. First, the model specification is as similar as possible to the UK model, making it a better fit for comparisons. Second, the    179 implausible results produced by the FD model with cross-lagged effects indicate that the lag structure in the data does not match the process as it unfolds in reality. This problem persists in the cross-lagged model with the latent variable. Although it does not blow up, the results are more likely to be biased than in the synchronous effects model. Third, given the long time lags between waves, it is not unreasonable to expect that past values are irrelevant.    180	   Figure	  A.3	  Test	  of	  leading:	  Three-­‐wave	  dynamic	  panel	  model	  (SEM)	  	  	  Notes:	  This	  Figure	  illustrates	  the	  relationships	  between	  the	  variables	  estimated	  in	  the	  structural	  equation	  model	  for	  the	  UK.	  Solid	  arrows	  represent	  the	  direction	  of	  estimated	  causal	  effects.	  Dashed	  double-­‐headed	  arrows	  represent	  (error)	  covariances	  included	  in	  the	  model.	  The	  Figure	  was	  adapted	  from	  Finkel	  (1995,	  84).	   !!!!!ε3!X2! X3!Y2!Y1!X1!Y3!ε4!ε1! ε2!β1! β3!β2! β4!! β5!β6!β7!β8!!!U!!!!!

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
http://iiif.library.ubc.ca/presentation/dsp.24.1-0348395/manifest

Comment

Related Items