Issue voting under different institutional contexts

Issue voting under different institutional contexts Political communication and effective representation in 37 countries Stefan Dahlberg Henrik Oscarsson Daniel Berlin Paper prepared for presentation at the XIV NOPSA Conference August 11-13, 2005 Reykjavik, Iceland Stefan Dahlberg, Ph. D-student. Göteborg University. stefan.dahlberg@pol.gu.se +46 (0)31 773 46 86 Henrik Oscarsson, Associate Professor. Göteborg University. henrik.oscarsson@pol.gu.se +46 (0)31 773 46 66 Daniel Berlin, Ph. D-student. Göteborg University. Daniel.berlin@pol.gu.se +46 (0)31 773 46 00 Abstract 1

The impact of electoral systems on party systems has long been regarded as one of the most important effects of electoral engineering (Duverger 1954; Lijphart 1994; Sartori 1997; 1976a). A common assumption among rational choice institutionalists is that electoral rules have a certain impact on the party systems and the patterns of competition, since the adoption of certain rules creates incentives for rational vote-seeking parties to either emphasize particularistic or programmatic benefits during the election campaigns (Downs 1957b; Sjöblom 1968). The strategies adopted by political elites are indirectly expected to influence the voting behavior among citizens. Yet, it is not quite clear to what extent electoral institutions affect voters on a micro level, only that they do. By combining contextual variables on electoral system features with cross-national surveys of voting behavior, we empirically explore how citizens are affected by the institutional context. A simultaneous modeling of macro and micro data from CSES shows a weak but significant contextual effect of type of electoral system on voting behavior. Proportional multiparty-systems seem to induce parties to communicate their policy positions by means of ideology. The parties left-right positions are more known to voters in proportional systems than in non-proportional systems and the effect of ideology on party choice is also stronger in proportional systems. The fact that the left-right dimension is not a valid operationalization of ideology in some countries does raise doubts over the validity of the findings. If only polities where the left-right dimension plays an important role are included, the support for this main finding becomes stronger. A more elaborated conclusion would be that among countries where the left-right dimension is an important distinction for voters, we find a significant effect of institutional context on the levels of ideological voting. Introduction There is a newly awakened interest among political scientists for empirically founded research on constitutional design and electoral outcome. The vast amount of research in this field have long focused on how politicians and parties are affected by different institutional factors, which in turn is believed to have a certain impact upon voters as well. According to Giovanni Sartori, electoral systems are the most specific and manipulative instrument of politics (Sartori 1997) and few would today oppose that institutions matter, both explicitly and implicitly (Reynolds 2002; Norris 2003b). Yet, it is not quite clear how electoral institutions affect voters on a micro level, only that they do. By combining contextual variables on electoral system features with cross-national surveys of voting behavior, we have the opportunity to empirically examine how citizens are affected by the institutional context. Our theoretical approach originates from rational choice 2

institutionalism where we are theorizing on how formal and informal rules 1 may affect the behavior of actors within a political system, which in turn is believed to influence the voting behavior among individuals. 2 Just as James Madison asserted two centuries ago, we believe that one of the keys to a well-functioning representative democracy lies in the architectural creation of political institutions (Ferejohn 1990: 6). More precisely, we are interested in how the institutional contexts affect voters perceptions of parties and how this may promote a higher degree of ideological issue voting, which we believe is normatively appealing to popular theories of effective representation. Our interest in political perceptions and ideological voting stems from the fact that these factors affect the extent to which voters are meaningfully represented in a political system, as they are important determinants of outcomes of electoral processes (Granberg 1988: 2; Brug 1997: 2). Voting behavior Several models of political representation have been developed over the years, where one of the most prominent is the responsible party model (RPM). The RPM has been the normative foundation in several studies concerning the functions of representative democratic systems, where it has been used for theorising about necessary conditions for citizens to select parties or candidates that represent their preferences, i.e. policy positions. The model presumes a prospective mandate where parties, during the election campaign, present coherent policy programs on which voters make their choices. An important prerequisite assessed by the model is that the parties should present divergent and stable policy positions so that voters are given meaningful electoral choices. Regarding voting-behavior, the RPM assumes that voters base their decisions on a comparative evaluation of the policy programmes among the competing parties. Thereafter, in order to directly influence government policies, they vote for the party or candidate whose policy-program are in best congruence with the voters own preferences (H. S. J. Thomassen 1999: 116; Adams 2001: ch. 1; Esaiasson & Holmberg 1996; Klingemann, Hofferbert & Budge 1994; Pennings 1998b). The importance of issue agreement between voters and their representatives in the RPM implies that the model easily can be incorporated in 1 Formal rules are here understood as the legislative framework embedded in official documents or constitutional conventions such as the degree of proportionality, effective thresholds etc. Among the informal rules or noninstitutional contextual factors that may influence the individual act of voting, we refer to variables such as the breadth of alternatives, the dimensionality of the ideological space and the polarization of the party system. 2 Our definition of an institution relies on B. G. Peters where: institutions are conceptualized as collections of rules and incentives that establish the conditions for bounded rationality, and therefore establish a political space within which many interdependent political actors can function Peters 1999: 44. 3

traditional spatial models of electoral competition, where smallest spatial distance is decisive for how voters choose to cast their ballots. An important determinant of the quality of the outcomes of electoral processes is the amount of political knowledge among voters, as it is more likely that well informed, rather that ill-informed voters make electoral decisions that reflect their attitudes toward public policies. If the process of democratic representation is to be acceptable from a normative point of view, a necessary condition that has to be fulfilled is that voters are informed about the different policy alternatives. Voters knowledge on matters of political or societal character has been the subject in several studies over the years and most of these studies have come to a similar conclusion: the amount of information among average voters is poor. The possibilities for reasoned choices, as acquired by most of the normative theories on representative democracy, may therefore be considered as being beyond the capability of the vast majority of the democratic citizenry (Delli Carpini & Keeter 1996; Campbell et al. 1960; Larry M. Bartels 1996; Page & Shapiro 1992; Converse 1964; Petersson et al. 1998; Holmberg & Oscarsson 2004; Gordon 1997; Lupia 1998). An argument why political representation in mass democracies can work at all in spite of the fact that citizens in general are not very knowledgeable about political matters is that citizen and elites often share the same belief systems (L. M. Bartels 1996; Luskin 2003). Rational voters use heuristics such as ideologies, cognitive schemas or belief systems as cost-reducing devices or shortcuts (see fc Downs 1957b; Zaller 1992; Lupia 1998; Popkin 1991). The role of belief systems in models of political representation has recently been highlighted by Jacques Thomassen (Holmberg 1999; Schmitt & Thomassen 1999; J. Thomassen 1999; van der Brug & van der Eijk 1999; van der Eijk, Franklin & van der Brug 1999). According to Thomassen, unidimensional belief systems is a desired characteristics of representative democracies as the effectiveness of the responsible party model depends on the extent to which the policy views of both the masses and elites are constrained by a one-dimensional conflict dimension, more specifically, a left-right dimension (H. S. J. Thomassen 1999: 34). His simple but strong argument is well founded. It rests upon social choice theory as well as the classic Downsian spatial theory of elections, where actors perceptual agreement of the political landscape is an important prerequisite (Downs 1957b). In short, Thomassen s statement is that simplicity in the political world is important for effective political representation. 3 According to 3 The hypothesis that simple, unidimensional party systems makes it easier to produce effective representation, leads to higher policy congruence between masses and elites and, eventually, also produce higher levels of political legitimacy is worth pursuing 4

this perspective, ideology provides the fundamental means of communication and constitutes a crucial linkage between citizens and elected representatives. In this perspective, it is the existence of a common political language, spoken and understood by both elites and masses that make modern mass democracies not break down. Since the ideological position of an actor can be used to trace the standpoints on a large number of underlying issues (Downs 1957b; Brug 1997), it could be argued that vote-decisions based on ideology, from a normative point of view, is desirable as it is both informative and demands some sort of issue orientation among the voters as required by the RPM. A range of studies, this included, focus on citizens abilities to perceive political messages from elites, i.e. how well policy alternatives and issue positions are being communicated to citizens. An electorate with clear and common perceptions of the party space has repeatedly been identified as one important prerequisite for successful political representation (Converse 1975; van der Brug 1996; Schmitt & Thomassen 1999). To maintain meaningful mass-elite communication, citizens need to have clear and correct perceptions of the main alternatives, or at least the most important differences between them, and what the ideological conflicts in the party system are all about (Berelson 1952). Perceptual agreement is considered to be a necessary, although not sufficient, condition for the electoral process to function as an effective channel of communication (van der Brug & van der Eijk 1999; van der Eijk, Franklin & van der Brug 1999). The more agreement over parties positions in the ideological space, the stronger the impact of ideology on voters party preferences (Oppenhuis 1995; Tillie 1995; van der Eijk & Franklin 1996). Strong agreement over party positions among voters is being interpreted as an indicator of successful communication of policy standpoints on behalf of the parties. The left-right position as a cognitive cue In this article, we will make use of the Left-Right dimension, since it has been considered as being one of the most universal and salient issue-dimensions within several political systems. It is also one of the few comparable measurements for belief systems available (Jones 2004; Budge 1987; Klingemann, Hofferbert & Budge 1994; H. S. J. Thomassen 1999). The strength with the left-right dimension is, as mentioned, that it summarizes positions on a large number of underlying issues and in the end, left-right ideology does not only structure voters preferences but may also be used as an information short cut on parties standpoints on several concrete a bit further. A European-wide analysis revealed a weak positive aggregate correlation between voters perceptual agreement over parties left-right positions and levels of political legitimacy (satisfied with democracy) (Oscarsson 2001). 5

issues (Downs 1957a: 98; Brug 1997: 38; 1999a; Fuchs 1990; Holmberg 2004; 1981: 192ff). This assumes that voters implicitly use a spatial image in their perceptions of political actors, issues or ideologies,(budge 1987; Oscarsson 1998; Brug 1997: 10) which favour the use of spatial models in the study of political communication. 4 The left-right distinction is probably most accurately described as a historically successful instrument of communication. Left and Right have proven to be powerful spatial metaphors that greatly help communicate political preferences and ideological standpoints between elites and masses in modern democracies one of the keystones in many popular models of political representation, such as the Responsible Party Model. Thus, the merit of the distinction is the simple distinction itself, not its content. More boldly put, the uni-dimensional mental map of the political landscape may have played a key role in the rise of well functioning representative democracies. Since its birth in the late 18 th century, the bipolar left-right construction has shown remarkable resistance, reproducing itself in new polities and in new historical and social contexts, constantly challenged and reshaped by new ideological dimensions. 5 Although the substantive meaning and interpretation of the left-right distinction have not remained the same for two hundred years albeit for the key element of equality (Bobbio 1996) the distinction has thus, played a crucial role as an information cost reducing device for political actors. Leaders have made extensive use of the left-right grammar for sending political messages that otherwise probably would have been incomprehensible for most voters. Left-Right have important orientational functions for individual citizens, and communication functions for political systems (Fuchs & Klingemann 1989; Inglehart & Klingemann 1976). Why do we expect electoral systems to have an impact on voting behavior? The impact of electoral systems on party systems has long been regarded as one of the most important effects of electoral engineering. Generally speaking, proportional systems tend to produce a greater number of parties than majoritarian systems (Duverger 1954; Sartori 1976a; 4 We know from a number of national studies that perceptual consensus over parties positions is dependent on the saliency of issue dimensions (Brug 1997; Oscarsson 1998. This means that the more intensively parties communicate their left-right positions to voters, the more common will voters perceptions of the parties left-right positions become. Therefore, aggregated measures of perceptual accuracy can be regarded as an indicator of how well the process of political representation works for left-right related issues in a political system. Working with this particular data, we have no other ideological dimensions for which parties and voters placements have been measured. 5 The absorptive power of the Left-Right semantics is impressive. For instance, religious, liberal economic and materialist values all contribute to rightist identification among citizens, while secular, economic leftist and postmaterialist values contribute to leftist identification (Knutsen 1995; Knutsen 1999.). 6

1997; Lijphart 1994; Golder 2005). 6 Rational choice institutionalism emphasizes that formal rules generate incentives that are shaping and constraining the rational goal-seeking behavior of the actors within a political system (Laver 1997; Peters 1999). A common assumption is then that electoral rules exerts an influence on the party systems and the patterns of competition, since the adoption of certain rules creates incentives for rational vote-seeking politicians to either emphasize particularistic or programmatic benefits during the election campaigns (Downs 1957b; Sjöblom 1968). The strategies adopted by political elites are then, indirectly expected to influence the voting behavior of citizens. 7 Consequently, proportional party-list formulas tend to produce incentives to shift the attention from individual politicians with particularistic policies to political parties with programmatic policies (Norris 2003b; Sjöblom 1968). In such a system, voters tend to vote for parties instead of single candidates. The candidates are also dependent on the parties for their election as the party leadership nominates them through the party-lists. A candidate s personal characteristics may then be considered as less important as they are acting more on behalf of the party. Party-ballots can thus, be expected to be effective in promoting party discipline, coherence and programmatic campaigning. Effective well-functioning parties may serve several important tasks in a political system by simplifying electoral choices, organize campaigns, aggregate interests, channelling the debate and select candidates (Dalton 2000) which in turn may affect voters perceptions of party positions and their voting behavior. The adoption of a proportional partylist formula may increase the number of parties in a political system and increase the incentives for the development of programmatic ideologically driven political parties and the presence of a shared belief system and a commonly spoken political language is an essential ingredient for a well functioning representative political system. 6 This division should thus not be thought of as a dichotomy since most electoral systems are a combination of functions from them both. An electoral system is therefore more or less proportional or majoritarian 7 The expectation that single member plurality systems will produce a two-party system has not gained fully empirical support. Those who want to predict the number of parties in a given political context must also consider the patterns of the social cleavages and not only the type of electoral system (Volwes 2004.) For example, Sartori argues that a proportional system do not multiply the number of parties in it self; instead, this is an indirect effect of the proportional formula. PR may in some cases remove pre-existing obstacles and thereby open up for new parties. For the same reasons, a plurality system cannot by itself produce a two party system but will help to maintain an existing one (Grofman & Lijphart 1986: 2-5.). The number of parties in a political system is also dependent on the level of the effective threshold that is applied in many proportional systems, in an attempt to decrease the party fragmentation and to promote stable governments (Lijphart 1998.) Nevertheless, it is clear that electoral formulas have a fundamental impact on the characteristics of the party systems (Sartori 1997; Lijphart 1994; Duverger 1954. In general, countries that are using any form of proportional representation have twice as many parliamentary parties than countries using any form of majoritarian electoral system (Norris 2003b: 84. 7

As stated earlier, electoral rules influence the patterns of competition and in line with spatial models of party competition, parties in a multiparty system will try to position themselves more ideologically and differentiate themselves as much as possible from the competing parties around in an attempt to mobilize their electoral support. While a two party system will make the parties to converge ideologically at the median voter position, parties in a multiparty system will maintain the clarity in their positions since they are not necessarily striving to attract the median voters (Downs 1957a: 115-27; Sjöblom 1968). When more parties are encouraged in the competition, they will face centrifugal incentives to take more dispersed positions in the policy space (Cox 1990; Sartori 1976b). Reasonably, more parties mean greater polarization, which should facilitate for voters to perceive the differences between them, which should increase the degree of ideological issue voting. 8 Reasonably, there is an upper limit where the effective number of parties within a system will exceed the benefits with breadth of alternatives. If too many parties are involved in the competition, voters may find it difficult to discern the policy positions between the parties, but this is an empirical question. When certain types of electoral formulas tend to affect the number of parties, which in turn is expected to be decisive for the degree of competition, i.e. centripetal or centrifugal competition, other historical factors account for the dimensionality in a political system. 9 The number of social cleavages in a political system is not necessarily affecting the number of parties in a political system (Jones 2004) and the parties are not by themselves capable of creating societal cleavages but they may induce or suppress already existing ones (Lipset & Rokkan 1967; Bartolini 1990; O. S. Knutsen, E. 1995). Nevertheless, the number of cleavages in a society may still be decisive for the effectiveness of the communication between parties and voters. Mainly because some parties may be more attached to alternative cleavages, such as environmentalist, religious- or linguistic parties, why the clarity in their positions will be depending on the saliency of the different cleavages. However, the initial adoption of certain rules such as a proportional formula may increase the incentives for bonding strategies among the competing parties, since proportional systems tend to produce multiparty systems that in turn may maintain or even reinforce existing cleavages. In 8 In contrast to the proximity theory, the advocators of the directional theory claims that voters in general are unable to recognise slighter adjustments in policy positions among parties. Instead, voters evaluate parties policies on their direction and intensity rather than in terms of their closeness to a voters own policy position. Given these assumptions, the perceptual agreement on a party s position may be expected to be higher for parties with extremer positions, as it will be more likely that the voters know the position of a party with such a position. See (Rabinowitz 1991: 157-66; Adams 2001: 97; MacDonald 1991; Rabinowitz 1989; Merill III 1999; Westholm 1997; Gilljam 1997.) 9 Our definition of a social cleavage rests on the work of Bartolini and Mair, where a cleavage exists if there is an interest conflict characterized by a) a social differentiation b) that is institutionalized and c) politically organized (Bartolini 1990. 8

plurality systems, on the other hand, parties may adopt bridging strategies that may suppress the group consciousness in the electoral arena and even downplay the importance of pre-existing cleavages (Norris 2003b). Something that can be expected to have far-reaching consequences, both for how voters perceive the parties but also on their voting behavior. If proportional systems imply that parties will adopt bridging strategies, then the voters within these systems should obtain higher levels of perceptual agreement and to a higher extent lean upon ideological issue voting than in plurality- or majoritarian systems. So far, we have mainly focused on effects of electoral systems in terms of the simple dichotomy between majoritarian and proportional systems. However, a constitutional order that bring large votes/seats disparities where the seat allocation is not congruent with the parties actual vote share may also reduce the incentives for individuals to be fully involved. A high degree of disproportionality may induce tactical voting where citizens are voting for their second choice instead of their first hand preferences. This may in turn decrease the incentives for gathering accurate information of the different policy alternatives (Gordon 1997). Reasonably, the more proportional a system is, the more likely it is that voters will vote on their first hand choice, which also should imply higher levels of ideological proximity. 10 This leads us toward our main hypothesis, which in a broader sense claims that it is all about simplicity. Perceptual agreement and the degree of ideological voting will be higher among voters in: a proportional, uni-dimensional political system, characterized by the occurrence of both party-list ballots and an electoral threshold and a polarized party system where several, but not too many, parties are involved in the competition, since: 1). Proportional systems tend to produce multiparty systems, and parties within such a system will adopt bridging strategies and try to differentiate and positioning them selves more ideologically in an attempt to maintain the clarity in their positions; and 2). with several parties involved in the competition, the parties will face centrifugal incentives to take more dispersed positions in an attempt to profile themselves from other competing parties and a high degree of polarization will, in turn, facilitate for the voters to apprehend the positions of the parties. 10 There is always some distortion embedded in the electoral systems due to various factors, why there are no completely proportional system. The lowest electoral threshold is to be found in the Netherlands where it is.67 percentages (Farrell 2001.) 9

In sum, the more proportional an electoral system is the more likely it is that the political representatives will communicate their policies in terms of ideology, which will increase the perceptual agreement and the degree of ideological voting among the voters. The purpose with these not too daring hypotheses is not as much an attempt to win new terrain, but rather to systematically examine traditional assumptions about the effect of electoral systems on voting behavior and effective representation. With access to a new cross-national dataset containing data on both voters and institutions, we now finally have the opportunity to test this empirically. Data, design and operationalizations This study uses data from the CSES (Comparative Study of Electoral Systems) which is a collaborative project of cross-national research among election studies conducted in over fifty states. The strengths with the dataset, besides that it is cross-national, is that it contains both micro-level data that include vote choice, candidate and party evaluations etc. and macro-level data on aggregate electoral returns, electoral rules and formulas. This allows us to conduct both cross-level, as well as cross-national analyses on the effects of electoral institutions on voting behavior. 11 Our analysis is based on data from CSES module 1 and 2, where respondents in some 37 countries were asked to place themselves and the parties on an eleven-point left-right scale 0-10. The CSES-question was: In politics people sometimes talk of left and right. Where would you place yourself on a scale from 0 to 10 where 0 means the left and 10 means the right. 12 A universal application of such a left-right scale is not uncontroversial. We know from earlier studies that the left-right distinction obviously makes a lot more sense in some countries than in others (Huber & Inglehart 1995; Klingemann 1995). We also know that the substantive meaning of the terms left and right differ much between polities and even between decades in the same polity. So, how can we justify the use of left-right division as a world-wide operationalization of ideology? 11 The data can be downloaded from: www.umich.edu/~cses 12 In Japan, a liberal-conservative scale was used. In the Netherlands, a ten-point scale 1-10 was used instead of an eleven-point scale. 10

A simple indicator of the importance of the left-right dimension in a polity is the strength of the bivariate relationship between party choice and voters self-reported ideological left-right positions (see table 1). In the CSES data, the highest correlations are found in Sweden (eta=.77), Czechoslovakia (eta=.76) and Poland (eta=.72). In these countries, we can explain large portions of the dependent variable party choice just by asking respondents about their own left-right position. Party choice is highly correlated with left-right ideology also in Spain, Iceland, Israel, Denmark, Portugal and Switzerland. In sum, 18 of 37 countries in the study have correlations higher than eta=.50. In conclusion, it is fairly unproblematic to use left-right ideology as a single indicator of ideology in these countries. 13 In general, the relationship between left-right ideology and party choice are high. The importance of left-right ideology is substantively higher on the European continent than outside Europe. Note that there are no non-european countries with high correlations, with the exception of New Zeeland (eta=.64), which undoubtedly also belong to the Western European culturalhistorical sphere. In polities as Romania, Korea, Belgian Wallonia, Taiwan and Peru, there are only weak bivariate correlations between citizens left-right orientations and party choice. Again, the left-right dimension s inability to perform well outside a Western-European cultural-political context is an important result in itself and will have consequences for the remaining analyses in this article. The rich variation of electoral systems in the world also makes it complicated for political comparativists. Many categorizations of electoral systems have been applied in contemporary research, and there is an on-going debate over what distinctions are most fruitful for empirical analysis (Powell 2000; Shugart & Wattenberg 2001; Anckar 2002). In previous analyses of CSES data, the operational classification of electoral systems has been into three large categories: proportional list systems, semi-proportional combined systems, and majoritarian systems (see Holmberg 2002; Norris 2003a). We have earlier argued that effective representation largely rests upon successful communication and that a prerequisite for ideological issue voting is that the parties actually are appealing to the voters in terms of ideology and that they have managed to communicate their ideological 13 Of course, if you dig deeper into the ideological space of national polities, you will find many reasons to expand the analysis with one, two or more important ideological dimensions, or even single issue dimensions. Even in Sweden, probably the most uni-dimensional party space in Europe (Oscarsson 2001.), more in-depth analysis must expand the number of dimensions to two (Oscarsson 1998.) 11

positions to the voters so that they have clear and consistent perceptions of the party positions. If ideology, i.e. left-right position, is going to work as a more or less accurate information shortcut for the different policy alternatives that a voter is facing during an election, the parties have to be able to communicate their left-right positions to the voters. Perceptual agreement among voters is not necessarily the same as perceptual accuracy. Nevertheless, if a majority of the voters are agreeing on the wrong position of a party, the communication has obviously failed. It is thus, quite unlikely that this actually will happen, and therefore we consider perceptual agreement to be a valid indicator on the degree of success in political communication (Brug 1997: 44). For measuring the degree of perceptual agreement among voters, we will use an estimate that originally was developed by Cees van der Eijk (van der Eijk 1998). The PA-measure has been employed a number of times in similar analyses (van der Eijk & Franklin 1996: 433-4; van der Brug & van der Eijk 1999: 137-9). PA is calculated for each individual party (se Appendix) and can also be aggregated into an overall measure of perceptual agreement in a political system. The measure of Perceptual Agreement is developed for ordered rating scales, such as the eleven-point left-right scale, and reflects the deviation from a unimodal distribution and ranges from 1 (perfect bimodality, where half of the voters place the party on the most leftist position and half of the voters place the party on the most rightist position) to +1 (perfect unimodality, where all voters place a party at the same position). Uniform distributions will get a PA-value of zero. The higher the PA-value gets, the higher perceptual agreement (van der Eijk 1998). 14 A high degree of agreement among voters on parties policy positions may then be interpreted as that the parties have managed to communicate their policy positions successfully. 15 In a second step, we will continue the analysis with ideological issue voting. The most straightforward way to operationalize ideological proximity is simply by calculating the absolute difference between the respondents self-reported left-right position and the parties positions along the eleven-point left-right scale. Ideological proximity will range from 0 (minimum voterparty distance) to 10 (maximum voter-party distance). To estimate the effects on probability to vote for a party, we will perform the procedure of stacking the data-set (see van der Eijk & 14 The calculations of Cees van der Eijks Measure of Perceptual Agreement were made in SAS. The syntax for calculating PA-measures for 7, 10 and 11-point ordered rating scales can be acquired by sending a mail to Henrik.Oscarsson@pol.gu.se. 15 In the CSES data set, the number of parties for which we have left-right placements are limited to six. This means that for many multi-party systems included in this study, the party systems have been truncated, in some cases dramatically so. Many multi-party systems, such as Sweden, the Netherlands, and Poland, have more than six relevant parties. In many analyses, we only include the respondents who voted for parties for which we have available data on left-right positions, which in some cases is a drastic limitation. Which parties to include in the final data set has been decided by the research teams behind the respective national election studies. A full list of what parties are included in the study can be found in the Appendix. 12

Franklin 1996) and thereby convert to a new unit of analyses, namely party-voter dyads. In each polity, there will be as many party-voter dyads as there are respondents times the number of parties. The party choice variable will become a binary response variable (0=did not vote for the party, 1=voted for the party), allowing to estimate the probability to vote for the party with logistic regression. Perceptual agreement and ideological voting under different institutional contexts How does the institutional context affect the political communication and the level of ideological issue voting in a polity? Do parties in proportional multiparty systems communicate in terms of ideology and are voters in these systems inclined to vote ideologically? In this analysis, we will estimate the effects of different institutional settings on the perceptual agreement among voters and the effects of ideological left-right proximity on the probability to vote for a party under different electoral systems. Consequently, we expect perceptual agreement and the effects of ideological proximity on party choice to be greater in proportional systems than in majoritarian systems. This hypothesis can be deduced from the simple fact that proportional list systems tend to offer more choices, i.e. effective number of parties (Duverger 1954), and that voters in proportional list systems more often choose between political programs of parties and not between candidates (Holmberg & Oscarsson 2003; Holmberg 2002). Thus, we expect ideology to play a more important role in multi-party list proportional systems than in two-party candidate-oriented majoritarian systems since parties in a multiparty system are prone to differentiate and positioning themselves in terms of ideology in an attempt to maintain the clarity in their positions(downs 1957b: 115-27). If all this is correct, we also expect that the ideological positions of parties will be more well known to voters in a proportional list system than in majoritarian. The perceptual accuracy and perceptual agreement will be higher in proportional election systems than in semi-proportional or majoritarian systems, since parties in proportional systems have higher incentives to carefully communicate ideological standpoints and mark the ideological distances to other alternatives (Downs 1957b). In turn, voters in proportional list systems need to develop skills and well-functioning cognitive schemas to be able to comprehend political messages from the party elites and orient themselves ideologically. In table 1, the 37 countries in the analysis have been sorted according to the correlation (eta) between left-right orientation and party choice. Included in the table are the measure of left-right 13

polarization (Wing Party Distance), and the Weighted Perceptual Agreement (WPA) for each polity. We also included the type of electoral system (CSES-categorizations), along with some characteristics of the electoral systems (effective threshold, degree of disproportionality, the effective number of parties (Anckar 2002). 14

Table 1: Type of Election System, Effective Threshold, Disproportionality, Effective Number of Parties, Wing Party Ideological Left-Right Distance (WPD), Weighted Perceptual Agreement (WPA) 16 and Correlation between Left-Right Self Placements and Party Choice (Eta, Cramer s V) Election Year Country Election System (CSES) Effective Threshold Dis-proportionality Effective # of parties Dimensionality 17 Government Stability Wing Party Distance (WPD) Perceptual Agreement (WPA) 1998 Sweden List PR 4.0 0.9 4.3.59.047 7.78 0.65.771 1996 Czechoslovakia List PR 5.0 8.9 4.1.56. 8.33 0.62.755 1997 Poland List PR 5.0 10. 13.6.51.195 6.49 0.58.718 2000 Spain List PR 9.7 6.0 2.5.43.038 4.86 0.59.701 1999 Iceland List PR 16. 1.1 3.5..025 6.01 0.65.698 1996 Israel List PR 1.5 1.8 5.6.68.063 5.95 0.50.688 1998 Denmark List PR 2.0 0.3 4.5.39.10 6.12 0.68.687 2002 Portugal List PR 6.6 5.0 2.6.52.063 6.00 0.59.675 1999 Switzerland List PR 8.6 3.0 5.2.. 4.58 0.53.662 1997 Lithuania Semi-Prop....57. 3.71 0.52.656 1997 Norway List PR 4.0 3.6 4.4.39.063 5.89 0.59.654 1998 Ukraine Semi-Prop....39. 5.52 0.42.648 1996 New Zeeland Quasi-Prop Mixed 5.0 4.5 3.8.44.033 5.35 0.53.639 1998 The Netherlands List PR 0.7 1.1 4.8.40.025 4.38 0.62.589 1999 Russia Semi-Prop....52. 6.21 0.41.568 1996 Spain List PR 9.7 5.5 2.7.43.038 5.56 0.62.521 1997 Great Britain Majority 37. 16. 2.1.54.02 3.20 0.45.516 1998 Hungary Qvasi-Maj Mixed. 22. 3.4.82.025 4.98 0.45.509 1996 Slovenia Semi-Prop....43. 3.13 0.35.494 1998 Germany Qvasi-Prop Mixed 5.0 3.1 3.3.51.01 4.23 0.51.485 1997 Canada Majority 37. 13. 3.0.37.04 2.60 0.37.406 1996 Japan Qvasi-Maj Mixed 25. 13. 2.9.54.125 5.19 0.49.402 1996 Australia Majority (Alt Vote) 37. 11. 2.6.45.08 2.66 0.42.399 1999 Belgium List PR 8.8 2.6 9.1.44.05 3.44 0.37.385 2000 Hong Kong Prop....45. 2.34..323 1996 USA Majority 37. 3.2 2.0.....312 1997 Mexico Quasi-Maj Mixed 22. 8.3 2.9.50 3.34 0.03.293 2000 Peru List PR....44. 3.14 0.17.265 1998 Hong Kong Prop....45. 1.93..262 1996 Romania List PR....36. 2.94 0.29.157 2000 Korea Qvasi-Maj Mixed 32...41. 2.35 0.42.112 1996 Taiwan Qvasi-Maj Mixed 11. 5.6 2.5.49. 2.12 0.38.090 1999 Belgium List PR 8.8 2.6 9.1.39.05 3.88 0.51.079 2001 Peru List PR....44. 2.35 0.11.071 2000 Mexico Qvasi-Maj Mixed 32...50. 3.97 0.09.011 2001 Belarus Majority 11. 5.6 2.5.60. 2.63 0.26 N a 2001 Thailand Majority 37. 3.3 4.3.38.111... Note: Data on Electoral Systems are from the CSES macro data set. Data on effective threshold, disproportionality, effective number of parties and government stability is from Anckar (2003). = The effective number of parties is calculated from Herfindahl s index of concentration, 2 H i = p 1 i, where pi is the population proportion for group i of votes, where 1/H then is the effective number of parties. This measure has the pleasant feature that Eta 16 The parties, for which the perceptual agreement has been calculated, have been weighted according to the election results (percentages). 17 This variable does only refer to the degree of uni-dimensionality as such and is yet not specifically related to the left-right dimension, which it will be in the near future by a Spearman correlation between the ordering of the parties along the j-scale obtained from the unfolding analysis and the left-right ordering of the parties. For this reason it is not included in the analysis. 15

it will equal n when there are exactly n groups of size 1/n (Lijphart 1984: 120; Pennings 1999a: 259-61) Polarization is a simple measure of the wing party distance, which is calculated as the difference between the means of the most leftist- and the most rightist parties 18 (see Gilljam & Oscarsson 1996). The left-right polarization is, thus, a first simple indicator of how successful an application of the left-right division is in a polity and an indicator on the degree of left-right competitiveness. The degree of proportionality has been calculated as the average deviation between the vote and the seat share of the two largest parties in each election (Pennings 1999a: 261). Dimensionality refers to the degree of uni-dimensionality and is computed with MUDFOLD 2.0, based on party evaluations where the respondents were asked to rate the parties on a scale from 0 to 10, where 0 means strongly dislike that party and 10 strongly like that party. Among the top-ten countries in table 1, nine have proportional systems, which must be regarded as a rather strong support for the hypothesis that proportional systems tend to induce ideological proximity voting. It is also evident that a high degree of perceptual agreement, with the exception of Spain and The Netherlands, seems to co-occur with a high amount of ideological issue voting. 19 A more elaborated analysis of the relationship between the institutional context and perceptual agreement confirms that voters under list proportional electoral systems tend to agree more over parties left-right positions (Average WPA=.51) than voters under majoritarian (Average WPA=.38) or mixed electoral systems (Average WPA=.38). The rather strong relationship between type of electoral system and perceptual agreement in table 2 (.48), is not very surprising since we expected that proportional electoral systems should facilitate the communication between parties and voters in terms of ideology. It was also expected that the left-right polarization should be highly correlated with perceptual agreement (r PAxWPD =.67). The greater left-right distance between the two flanking parties in a system, the higher are the levels of perceptual agreement over parties left-right positions. Obviously, polarized systems are promoting an effective representation since greater polarization seems to facilitate for voters to perceive parties ideological positions. 18 The highest levels of left-right polarization can be found in Czechoslovakia (WPD=8.33), Sweden (WPD=7.78) and Poland (WPD=6.49) (see table 3). In some other countries, the perceived left-right polarization is very low, indicating that the left-right distinction is not very important, alternatively that the terms left and right have no substantial meaning to voters, notably Hong- Kong (WPD=1.93), Taiwan (WPD=2.12), and Peru (WPD=2.35). Obviously, there are no parties in Peru that fit well into the categorization of party families used in the CSES, which indicates that the importance of the left-right distinction is negligible in the Peruan party system. Instead, religion seems to be the main ideological conflict. In Hong-Kong and Taiwan, parties compete mainly along a dimension pro-con mainland China. For these polities, left-right ideology is not a very appropriate measure of ideology, which may limit the scope of the following analyses. 19 A bivariate correlation between WPA and Eta yields a coefficient of.75, Pearson s r. 16

Table 2. Bivariate correlations between Weighted Perceptual Agreement and Type of Election System, Effective Threshold, Disproportionality, Effective Number of Parties, Wing Party Ideological Left-Right Distance (WPD) (Pearsons r and Eta). Perceptual Agreement Election System 1.48* Effective Threshold,-53** Disproportionality -.37 Effective # of Parties.11 Government Stablility.11 Wing Party Distance.67** Uni-dimensionality.045 1 Election systems were coded for the CSES-categorization as 1 if majoritarian, 2 for Qvasi-maj, 3 for semi.prop. 4 for qvasi-prop, 5 for prop and 6 for List PR. Coefficient for Election System is Eta value. ** Correlation is significant at the 0.01 level (2-tailed). * Correlation is significant at the 0.05 level (2-tailed). ** Correlation is significant at the 0.01 level (2- tailed). The negative and significant correlation between effective threshold and PA supports the relationship between type of electoral system and PA as a high electoral threshold may have a similar effect on a party system as the majoritarian electoral formula by reducing the number of parties. The weak impact of the effective number of parties is in line with the simplicityhypothesis, since a complex party-system with a great number parties involved in the competition, should make it even harder for the voters to perceive their positions clearly, most likely the relationship is curve-linear as. According to the results in table 3, perceptual agreement seems to be promoted by party systems containing of approximately 2,5 to 5 effective number of parties (4.7 in average among the list-proportional countries). Thereafter, the PA are decreasing as the number of parties grows bigger. The degree of uni-dimensionality is positively co-varying with PA but the relationship is weak and insignificant. Reasonably because the results from the mudfold analysis are connected to the degree uni-dimensionality in general and is not specifically related to the left-right dimension as the PA-measure is. In table 3, we list the results from the logistic regression analyses performed within each polity available in the data set in order to examine the effect of ideological distance on the probability to vote for a party. We have sorted the data according to the overall fit of the regression model (Pseudo R 2 ). 17

Table 3 Effects of ideological left-right proximity on voting within different electoral systems (Logistic effects, Pseudo R 2 ) Election System (CSES) Voterparty dyads Logistic effect (b) Effective Inter- Year Country n cept 2002 Portugal List PR 726 4202 +.13 -.81.36 1998 Ucraine Semi-Prop 241 1047.10* -.52.35 1998 Denmark List PR 1537 9010 -.01* -.94.33 2000 Spain List PR 682 2919 +.24 -.82.33 1996 Israel List PR 504 2407 -.12* -.53.26 1998 Sweden List PR 789 4700 -.17 -.67.26 1996 Tjeckien List PR 962 5612 -.21 -.60.26 1998 Hungary Qvasi-Maj 753 3641 -.11* -.56.25 1999 Iceland List PR 1121 5272.01* -.64.25 1996 New Zeeland Quasi-Prop 2827 15190 -.17 -.61.25 1997 Canada Majority 945 4431 -.09 -.57.22 1997 Norway List PR 1527 9113 -.32 -.66.22 1999 Switzerland List PR 993 4859 -.22 -.57.21 1998 The Netherlands List PR 1411 8146 -.36 -.73.21 1997 Great Britain Majority 1853 5864 +.14 -.45.19 1997 Poland List PR 945 5397 -.47 -.43.18 1999 Russia Semi-Prop 849 4703 -.51 -.38.17 1997 Lithuania Semi-Prop 165 477.15* -.31.16 1996 Spain List PR 756 3339 -.31 -.41.16 1996 Australia Majority (Alt 1312 6175 -.56 -.42.13 1998 Germany Qvasi-Prop 1461 8536 -.67 -.46.13 1996 Slovenia Semi-Prop 642 3141 -.56 -.34.12 1999 Belgium Flanders List PR 1584 9345 -.78 -.36.10 2001 Peru List PR 642 3482 -.75 -.29.10 2000 Mexico Quasi-Maj 1016 5412 -.80 -.22.09 1996 Taiwan Qvasi-Maj 356 1013 -.10* -.29.08 1996 Romania List PR 451 2104 -.71 -.23.07 2000 Korea Qvasi-Maj 574 2910-1.13 -.12.02 1999 Belgium List PR 1227 5444-1.29 +.02*.00* 1997 Mexico Quasi-Maj 247 1360-1.39 -.03*.00* All Countries 29119 149314 -.43 -.45.17 Pseudo R 2 Note: Belarus, Hong-Kong, Japan, Peru (2000), Thailand and USA is excluded from the analysis, due to lack of data on both left-right self-placements and party-placements. * means that the coefficient is not significant (p>.05). As expected, there are significant and large negative bivariate effects of ideological distance on party choice in many of the countries where the left-right dimension is strong. The probability to vote for a party decreases with ideological distance. In concord with previous findings, the best model fit (Nagelkirkes maximum rescaled pseudo R 2 ) is found in Portugal, Ukraine, Denmark and Spain, whereas in Korea, Taiwan, Romania, Belgian Wallonia, Peru and Mexico, there are very small or non-existing effects of ideological left-right proximity on party choice. What about the effects of institutional contexts? Seven of the top ten countries in table 3 do have proportional list electoral systems, but there is no clear-cut association. In some non-proportional systems, the level of ideological left-right voting is very high, such as in Ukraina, Hungary and 18

New Zeeland. And there are a number of proportional systems where the levels of ideological left-right voting is very low, such as in Romania, Belgium and Peru. A more formal test of the impact of institutional context is of course possible. For all countries included in the study, we can model a multiplicative interaction effect of electoral system (0=nonproportional; 1=proportional) on the relationship between ideological proximity and probability to vote. The results show that there is a small but significant institutional effect on ideological voting, i.e. there is a larger negative effect of ideological distance on voting probability in proportional list systems than in non-proportional list systems. That the institutional effect is very small can be illustrated by the estimated probability functions in figure 1. Figure 1: Ideological Proximity x Probability to Vote in Proportional List Systems vs Semi-Proportional/Majoritarian Systems.,5,4,3,2,1 0,0 0 1 2 3 4 5 6 7 8 9 10 Note: Data are from CSES Module 1 (unweighted). The probabilites to vote are estimated from a logistic regression analysis of a stacked dataset (voter-party dyads). Only respondents voting for parties that were located on the leftright dimension is included in the analysis. The filled line represents the probability-to-vote curve among voters in proportional list systems. The dotted line represents the probability-to-vote curve proximity among voters in nonproportional list systems, i.e. semi-proportional/majoritarian systems. Figure 3 show the estimated probabilities to vote for parties that are located at different left-right distances from a voter. The probability curve is somewhat steeper in polities that have proportional list systems, i.e. the effects of ideological proximity are larger in proportional systems than in non-proportional systems. This simple analysis and the findings may not convince us that proportional list systems encourages ideological voting, but nevertheless give at least some support to the hypothesis. 19