Issue Voting and Party Systems in Central and Eastern Europe

Issue Voting and Party Systems in Central and Eastern Europe Gábor Tóka Central European University, Budapest, Hungary tokag@ceu.hu Final draft of “Issue Voting and Party Systems in Central and Eastern Europe.” in Bürger und Demokratie in Ost und West. Studien zur politischen Kultur und zum politischen Prozess - Citizens and Democracy in East and West. Studies in Political Culture and the Political Process, ed. by Dieter Fuchs, Edeltraud Roller, and Bernhard Wessels. Opladen: Westdeutscher Verlag, pp. 169-85. Since the 1970s, partly in reaction to the somewhat bleak assessment of the individual citizens' political competence in the works of the Columbia school and Campbell et al. (1960), a whole generation of specialists in political behaviour has tried to show that ordinary people can make good use of their vote. Their studies suggested that citizens, after all, intelligently interpret programmatic differences between parties and candidates, and vote on the basis of their concern with serious, substantive political questions. V.O. Key and his echo metaphor concisely anticipated their verdict about earlier reports on the dearth of issue voting in the electorate: 'Even the most discriminating popular judgment can reflect only ambiguity, uncertainty, or even foolishness if those are the qualities of the input into the echo chamber.' (Key 1966: 2). Hans-Dieter Klingemann's works have offered a particularly upbeat version of this line of reasoning. A recurrent theme of them has been that political parties tend to offer meaningful alternatives, that their programmatic differences are not vanishing under the pressures of electoral competition (Klingemann 1995), and these differences - rather than a convergence of party positions - give citizens control of public policies (Klingemann/Budge/Hofferbert 1994). It is in this context that his critical remarks about formal modelling of party competition (cf. Klingemann 1995: 185; Klingemann/Budge/Hofferbert 1994: 23) can be best understood. Their probably best justification lays in his insistence on three points that, together, attribute party ideologies an unusually central place in democratic representation. First, he sees (most) political parties as fundamentally and inevitably – if not exclusively policy-oriented actors. Sometimes they may even be called ‘visionaries’, but sometimes they seem anything but interested in principles and ideology. Thus, the scholar studying them has to reckon that the actors’ motives are mixed and subject to change – a situation that rational choice analysis is naturally at odds with. Second, against the natural predilection of formal models to treat the issues of partisan conflict as endogenous to party competition, Hans-Dieter Klingemann traces their roots to intense, pre-existing societal conflicts and the distinct ideological baggage that the parties, given their past record and the demands of hard-core supporters, just can’t help carrying. Last but not least, he considers the presence of programmatically differentiated political parties a must for democratic representation. Ideological vision and the conflict of 1 principles, rather than opportunism and convergence, seem to signal the health of democracy to him. This stands in sharp contract with much formal modelling of party competition, where convergence born out of competition is the ultimate guarantee of popular representation, and multiple dimensions of ideological conflict always raise the superstitiously feared spectre of cycling, chaotic policy outcomes. A crucial reason for his work attributing such a central role to party ideologies is that he sees the performance of citizens in the democratic process fairly unproblematic. Albeit their political sophistication depends on their education and the like (Klingemann 1973, 1979b), they tend to make good sense of the political choices they are facing (Fuchs/Klingemann 1989). If meaningful alternatives obtain, they develop party preferences on the basis of substantive policy- and ideology-oriented considerations, in proportion to their preceding opportunities to learn about relevant differences between the parties and candidates (Klingemann 1979a; Klingemann/Wattenberg 1992). For instance, whatever the difference between voters in old and new democracies may be, its fundamental cause should be the different political experience they were exposed to (Kaase/Klingemann 1994). In other words, citizens’ skills, orientations and behaviour remain important determinant of the emergence and quality of democracy (Welzen/Inglehart/Klingemann 2000), but political behaviour itself depends on variables exogenous to political culture, like the party system. Where the latter shows programmatic structuring, citizens’ behaviour will not be an obstacle to the emergence of policy congruence between parties and their voters. The offering of the parties, and not some kind of ethno-religiously defined culture in itself are the critical variables that directly affect democratic performance. Social structures and culture constrain parties, but only influence outcomes through political intermediaries. 1. The model Below I offer an empirical test of some of the above propositions as I can operationalize them in the context of issue voting. The notion of issue voting refers to the impact of policy-, rather than party- or personality-related (or essentially non-political consideration) on citizens’ electoral choices (cf. Klingemann/Taylor 1977). Much of the literature on the 2 topic more or less implicitly assumes that the bigger the influence of issue concerns, the more reasoning, rational, and politically efficacious voters’ behaviour is (cf. Berelson/Lazarsfeld/McPhee 1954; Franklin 1985; Granberg/Holmberg 1988; Key 1966; Pomper 1972; RePass 1971; Rose/McAllister 1986; Nie/Verba/Petrocik 1976). It remains debated whether issue voting really is a quintessentially rational aspect of electoral behaviour, and whether it has increased in Western democracies since the 1950s. But the proposition that issue voting is facilitated by the clarity of policy differences between the major parties and citizens’ political competence has certainly remained popular over the decades (cf. Carmines/Stimson 1980; Alvarez 1997). Below I try to investigate whether the level of issue voting is indeed responding to the clarity of programmatic alternatives. More concretely, I will use survey data to investigate how the correlation between vote choice and issue attitudes is related to the clarity and degree of programmatic differences between the parties, the fractionalization of the party system, the age of democracy, and the pool of relevant knowledge in the voting population. While the main bone of contention is simple enough, there is surprisingly little in the way of systematic comparative evidence that could be marshalled to support or refute it. If anything, Hans-Dieter Klingemann’s works teach us that however logical something appears to be, we’d better check if it is true before believing. My model tries to test an underlying substantive message of his works: if only parties staked out clear positions, the voters would react. In other words, the party system need not be particularly simple, the party alternatives do not have to be particularly old, and party positions can be fairly moderate to allow for issue voting in the electorate. The clarity of party policies is the single most important party system characteristic that determines the extent to which policy congruence between voters and parties can emerge. The truth of this intuition is not obvious. The age of the political system could be expected to positively influence voters’ information level regarding party positions as much as the clarity of those positions. Party fractionalization can also be expected to impact issue voting by affecting all voters' information costs - and consequently knowledge - regarding party policies. Its likely effect through knowledge is negative, of course: the more parties there are, the bigger the voters’ information costs and the less their knowledge: thus issue voting must be less abundant. It is not immediately clear why these 3 effects would be less pronounced than the likely positive effect of the clarity of party positions on voters’ information level and issue voting. Furthermore, the same three system characteristics may have a direct effect on issue voting even when we hold voters’ information level regarding party positions constant. First, the aging of democracy may alter political culture. Citizens’ may increasingly adapt to norms that are widely – wisely or not is another matter - cherished in democracies, for instance that they should care ‘about the issues’, and not seemingly more superficial cues when it comes to voting in elections. Secondly, the more parties there are to choose from, the more refined differences in citizens’ attitudes can be expressed through vote choice. Finally, a greater clarity of party positions may result from bigger policy distances between them. This may not only reduce citizens’ information costs but may also increase the policy stakes in competition and hence citizen’s motivation to engage in issue-oriented voting. Thus, clarity of party positions may appear to have a direct effect on issue voting over and above its indirect effect through citizens’ knowledge. Hence, two regression models will be estimated below. The first assesses how the age of democracy, the number of parties and the clarity of party positions shape citizens’ knowledge of party positions. The second explores the direct impact of both the dependent and independent variables of the first equation on the degree of issue voting in the electorate. The expectation suggested by V.O. Key’s echo chamber metaphor as well as many of Hans-Dieter Klingemann’s work is that the clarity of party positions is a predominant influence on citizens’ knowledge, and the latter has far bigger impact on issue voting than the three system characteristics. Note that this argument does not refer to just about any possible understanding of what ‘issue voting’ amounts to. The ultimate dependent variable in the theoretical argument is the collective capacity of citizens to hold politicians accountable to popular preferences. Thus, what matters is not the number of individual citizens who sincerely believe that they voted for the party that matches their issue positions best. If citizens totally lack a shared understanding of which party stands for what, then election results cannot intelligibly express voters’ preferences. Rather, what empowers citizens, according to the above argument, is if voters with similar policy preferences support similar parties. This collective behaviour can assure – and make politicians expect - that parties will win 4 and lose votes according to the popularity of the positions that they, according to the perception of most ordinary voters, advocate. It follows from this that the cases in the empirical analysis must not be individual citizens, but entire electorates. Due to the scarcity of comparable data on the clarity of party positions, the cases in the analysis were selected solely on the ground of data availability. Two separate analyses were carried out, which allows checking the robustness of the findings in the face of slight differences in the measurement of some key variables, and a very substantial change in the sampled universe. 2. Data and measures in the first analysis The first test covers the Czech Republic, Hungary and Poland in the mid-1990s. The chief data sources are a cross-national survey of East Central European middle-level party elites directed by Herbert Kitschelt of Duke University, and a longitudinal comparative study on party images that was initiated and sponsored by the Central European University, Budapest (see CEU 1992-). For the latter, national probability samples of the voting age population were interviewed 6-8 times in each country covered between September 1992 and January 1997. For the former, face-to-face structured interviews were conducted with around 120-130 regional party leaders, mayors, city councillors and similar party politicians in each country in spring 1994. In the computations for this paper the data were weighted so that the relevant parties – numbering between six and ten per country – were equally represented in the weighted sample. The degree of issue voting is likely to vary considerably across issues, time and countries. Thus, in order to test propositions about what influences its extent, it is desirable to construct a database in which issues, time-points and countries all vary while the measures remain constant. This requirement has some undesirable implications for the precision with which issue voting can be estimated. As a rough indicator I will use the strength of statistical association, measured with the eta coefficients, between party preference (on the measurement see Appendix B) and issue attitudes. The original issue attitude questions asked the respondents to rate some ten political goals on a nine-point scale according to how strongly they were in favour or against them, and five of these items were included in the present analysis. These items were identically phrased to those 5 for which respondents had to reveal their perception of party positions, and are used to measure citizens’ knowledge below (see Appendix A). Respondents without either a party preference or a valid response to the issue question were excluded from the calculation of the eta coefficient. The higher the value of the coefficient, the more the variation across citizens in issue attitudes is concentrated between, rather than within the supporters of individual parties. In Granberg and Holmberg’s (1988) words, this indicates the ‘rational democratic component’ of the congruence between voters’ own issue attitude and what issue position they attribute to their preferred party. Unlike the simple correlation between the last two factors, the rational democratic component cannot result from just wishful thinking about party positions. It is fed by two factors only: that people with similar issue attitudes tend to vote for the same party, and that some people accept the issue position associated with the party that they – for whatever reason – support. In the first case, issue position is the cause that is followed by vote choice, while in the second case the issue position is the result of party choice. But in either case, a party-mediated and policy-based linkage obtains between the representatives and the represented. However, only the first of the two elements of the ‘rational democratic component’ has something to do with issue voting. Hence, it is unfortunate that the present data do not allow their separate treatment. This may result in statistically insignificant findings where the theoretically anticipated effects on issue voting are obscured by a zero relationship between the independent variables on the one hand, and the second ingredient of the rational democratic component, namely that some people may adopt their preferred party’s issue positions, on the other. This is particularly likely when, as in the second analysis reported below, issue positions are replaced with party- and self-placements on an abstract left-right scale (cf. Inglehart/Klingemann 1976). However, it is improbable that this measurement error leads to false positive findings. A priori, the independent variables of my model do not seem likely to influence either the parties’ ability to persuade voters about their issue position, or the voters’ eagerness to be persuaded in the same direction as the system characteristics, according to my expectations, impact the degree of issue voting. Hence, the measurement error may 6 make my test somewhat conservative, but are unlikely to inflate any one of the effects estimated in the statistical models below. Note that the measure of issue voting was computed separately for five issues and several time points in each of three countries. Altogether, there are 105 cases: five issues in each of eight surveys in the Czech Republic, seven in Hungary, and six in Poland. Obviously, the same country at time t is not an independent event from the same country at time t-1, therefore an important assumption in the estimation of statistical errors would be violated if they were treated as separate cases. Therefore, in order to avoid the underestimation of the statistical error of the parameter estimates, the cases in the analysis were weighted. The weight factor was one divided by the number of surveys in the respective country, yielding a weighted sample size of 15 (i.e. the number of issues times the number of countries). In this first analysis, the age of democracy was measured with the number of months passed since the first free legislative election after the fall of communism, i.e. March 1990 in Hungary, June 1990 in Czechoslovakia, and October 1991 in Poland. For each time point covered by the analysis, the Effective Number of Parties variable was computed from the distribution of party list votes in the last lower house election in the respective country. The Laakso-Taagepera index was used, which equals 1/(1-F), where F is Rae’s index of fractionalization - one minus the sum of the squared proportion of the vote won by each party. For instance, if there are only five parties winning noticeable electoral support, each taking one-fifth of all valid votes, then F = 1- 5(.20)² = .80, and the effective number of political parties is five, irrespectively of the number of parties winning (nearly) no votes at all. Vote choice can only be motivated by citizens’ issue and ideological concerns if the voters knew, rather than just believed, something about party positions. This knowledge may not be correct, but must be more than mere fantasy. In other words, it has to be intersubjective – otherwise it is not exogenous to the voting decision itself, and cannot be a cause of the latter. A visible form of such knowledge is consensus among citizens about where particular parties stand on an issue dimension. Its degree must depend partly on citizens’ motivations and cognitive skills, and partly on how clear party positions are. Thus, the measurement of clarity needs to be separated from voters’ knowledge about party 7 positions, since that knowledge may be influenced by the skills and motivation of ordinary citizens too. These are, in a nutshell, the basic ideas behind the procedure developed for measuring the clarity of party positions and how knowledgeable citizens are about them. It suffice to sketch it here since the details are discussed elsewhere (Tóka 1998). The data come from Herbert Kitschelt’s survey of party activists. All respondents were asked to locate all relevant parties in their country on several 20-point issue scales (e.g. more vs. less progressive income tax). The items selected for the present analysis were close parallels of the ones that were used to calculate issue voting and the index of citizens’ knowledge (see Appendix A). I assume that middle-level party activists are far more knowledgeable about party positions than individual citizens, and that – to the extent that there really are predictable party positions on the given issue - their knowledge is equally close to (or equally far from) perfect across issues, countries and parties. On this basis, one could measure the "diffuseness" of party positions with the standard deviation of the placements of each party j on issue k across all respondents on a given issue (see Kitschelt et al. 1999). But this solution raises several problems. First, some of the variation in the placement of a party on an issue may merely reflect the diversity of partisan viewpoints in the jury, rather than a genuine lack of identifiable party positions. Secondly, standard deviations – as well as distances between party positions – depend on how the endpoints of the scale were defined. Had, for instance, point 20 of the income tax scale meant a poll tax (instead of a less progressive tax than the existing one), the same respondents might (indeed should) have placed the same parties in a narrower range. Then, the standard deviation of the judgements on any party j's position would have been smaller too. Last but not least, distances and variances are not comparable across issues as long as we cannot define an explicit exchange rate between a unit difference on one issue (say income taxation) and another (say abortion rights or NATO-membership). To avoid these problems, a percentage-based measure was developed that has a naturally standardized metric, with the maximum value of 100 indicating that (1) not all parties were attributed exactly the same position by everyone, and (2) there was a perfect consensus among politicians/supporters of each party about the position of all the parties they were asked about. The first step to achieve this was to organize the responses into a 8 data matrix where each issue k was a column, and each unique combination of respondents i and parties j was a separate row. Thus, the number of cells in each column was equal to the number of respondents times the number of parties they were asked to place on the issue scales. For each cell the entry was calculated as the ith respondents rating of party j’s position on issue k minus the average of all relevant parties’ placement on issue k by respondent i. In other words, party positions were expressed in terms of the perceived direction and magnitude of their deviations from the all-party average on the given issue. The Clarity of Party Positions variable shows, for each issue separately, the percentage of the total variance in party placements that was explained, in a variance analysis, by the two independent variables and their interaction: namely which party a response referred to, and which party the judge belonged to. A strong interaction between the two variables could obtain if, for instance, members of left-wing parties placed their own parties at centre-left and right-wing parties at far-right positions, while members of the right-wing parties placed their own parties just a bit right of the centre, and the leftwing parties at far-left positions. Such a response pattern would not imply cross-party disagreement, but a high degree of consensus: not only about the relative distance between the parties on the issue, but also about which positions (namely the centrist ones) are the most desirable. The maximum value of the index (100) can also be reached if the interaction effect is nil, but everyone is at perfect agreement about the placement of each party, and these party locations at least minimally differ from each other. At the opposite extreme, the clarity of party positions is at its minimum when the variance in the placement of every single party on the issue shows as much variance within the members of each party as in the total sample. In this case, the parties’ true positions do not differ in an intelligible way, and they do not even dispute in a coherent way each other’s claims to represent a particular position. In this situation, no party-based representation of citizens’ issue concerns can occur, and the Clarity of Party Positions assumes its theoretical minimum value (0). Note that within nearly every single issue domain the Clarity of Party Positions has an almost perfect correlation with the spread of party position – i.e. the standard deviation of each party’s mean placement in the sample. This was determined the following way. Both the clarity and spread measures were computed for all four countries and each of the 9 20 issue and ideological scales included in the Kitschelt-survey, i.e. not only the five issues and three countries included in the rest of my analysis. The correlation between spread and clarity was then computed for each of the 20 scales separately. Each correlation is based on just four cases - i.e. the four countries -, therefore it is only their mean value (.94) that should interest us. The impressively high value is not a methodological artefact: even where the mean distance between the placement of two or more parties is high, there could be an even bigger variance in where exactly the same party is located by different respondents, hence producing a relatively high spread but low clarity of party positions. Nor we can explain away the stunningly high mean correlation by pointing to the small number of cases (just four countries for each of the 20 correlation coefficients), since the standard deviation of the 20 correlation coefficients is a modest .12. Rather, the high correlation seems to signal that a higher clarity of party positions can only be expected from greater policy distances between the parties. However, since no meaningful quantitative comparison of policy distances is possible across issue domains, this analysis has to rely on a measure of clarity. The spread of party position can only replace it where, as in the second analysis reported below, the comparison is restricted to a single scale measured identically across a number of cases. The Citizens' Knowledge of Party Positions variable closely parallels the Clarity of Party Positions index, but is derived from surveys of citizens. Unfortunately, the citizen survey relied on a different question format and probed different issue scales than the party-activist survey. The respondents were asked which (up to three) parties were most, and which were least likely to pursue certain goals. These goals included five that seem to tap nearly the same issue dimensions as the five picked from the elite survey - i.e. privatization, market economy, churches, former communists, and nationalism (see Appendix A on wording). Note that the citizen sample was also asked, at a separate question, to tell how important they considered these five goals, and the correlation of their ratings of the goals with their party preference was used to measure issue voting (see above). Hence, I obtain independent measures of three, presumably connected but distinct elements in a causal chain: how much consensus there is among party elites about the position of the parties; how much consensus there is about the same matter among voters; 10 and how much issue voting emerges on the same issues among citizens. From the citizen responses to which parties follow particular goals a close equivalent of the elite-data was created. Here too, the unit of observation was the combination of respondent i and party j. The same parties were considered as those rated by the party activists (for the list see Kitschelt et al. 1999), except that the Polish UD and KLD, which merged between the time of the Spring 1994 elite and the Autumn 1994 mass survey, were replaced with their successor organization (UW), and the mass responses about the Czech Socialist Party (CSS) and Agrarian Party (CSZ) were considered equivalent to the activists’ responses about their electoral alliance, the Liberal Social Union (LSU). First, each party j attributed the goal in question by respondent i was coded plus one, the party "least likely" to pursue that goal minus one, and all other responses zero. Then, exactly as in the Kitschelt-data, the average ‘placement’ of all parties by respondent i were computed for each issue k, and subtracted from the respondent’s placement of each party j on that issue. Finally, the same variance analyses were carried out for each issue as for the party activist survey, with the position of parties j as perceived by respondents i as the dependent, and the identity of the rated party and the party preference of respondent i as the two independent variables. In order to be included in the variance analyses the party preference data had to identify the respondent as a supporter of one of the parties rated in the survey of party activists. This step completes the construction of the variables needed for the first empirical analysis. Note again that the unit of observation is a particular political issue domain in a particular country at a particular point in time, but the data are weighted so that each of the 15 issue-country pairs is counted as just one case in the analysis. Recall too that some effects of the party system characteristics on issue voting may well be mediated by citizens’ knowledge. Therefore, the empirical tests require two regression equations, shown in Tables 1 and 2, respectively. The results of the first analysis, using the East Central European data described above, are displayed in the upper panel of the tables. Only the Clarity of Party Positions appears to have a statistically significant effect on Citizens’ Knowledge about Party Positions: the Age of Democracy and the Effective Number of Parties do not. Issue Voting, in its turn, is only effected significantly by Citizens’ 11 Knowledge – and thus, indirectly, also by the Clarity of Party Positions. All significant effects are in the expected – i.e. positive – direction. The theoretical implications will be discussed after presenting the results of the second analysis. Table 1 and 2 about here 3. Data and measures in the second analysis The survey data in the second analysis come from the Comparative Study of Electoral Systems (see CSES 2000). In each country covered by the study, national probability samples of the adult population were interviewed shortly after a national election held some time between 1996 and 1999. The data sets and study documentation were downloaded from the website of the project, the participants of which are certainly not responsible for possible errors in my use of the data. In all analyses reported here the CSES-data are weighted with the variables provided through the CSES Supplementary Weight File. Scotland and East Germany were treated as separate cases not only because of their peculiar party systems but also because of their substantial overrepresentation in the German and British samples – a condition that did not obtain in the case of Wales and the regions of Spain, for instance. Of the countries covered by CSES, Japan, Lithuania and the United States were excluded from the analysis because one or more variables utilized here were missing. Some of the relevant Israeli data were kindly provided by Asher Arian, whose prompt help is gratefully acknowledged. All in all, my second analysis covers 18 party systems and one issue dimension in each. The 18 data points enter the analysis unweighted. The only issue variables available in the CSES-survey are the respondents’ ideological self-placement and their placement of up to six relevant parties on the same 11point left-right scale. It is assumed here that the left-right discourse absorbs whatever the major divisive partisan issues in the given country are (cf. Inglehart/Klingemann 1976; Fuchs/Klingemann 1989). Therefore, party- and self-placements on the scale reflect some kind of weighted average of relevant issue positions. Starting from this assumption the Issue Voting and Citizens’ Knowledge of Party Positions variables were computed from the left-right placements, using exactly the same procedures as in the first analysis. The 12 coding of the party preference variable is described in the Appendix; the list of parties placed by the respondents is available from the study codebook (see CSES 2000). Only the responses about Plaid Cymru by the few respondents in the Welsh counties of Britain were excluded from my computations. Apart from citizens’ knowledge, no information on the clarity of party positions is available in the CSES data. The latter variable was therefore substituted with a proxy that I call the Spread of Party Positions. As shown in the previous section, the spread and the clarity of party positions are very strongly correlated in the East Central European elite data, presumably reflecting a close causal relationship between the two variables. The numerical value of spread is equal to the standard deviation of the mean leftright placement of each relevant party in a given country. This measure has been extensively used in the previous literature for cross-national comparisons of party polarization. I assume that the relative cross-national differences are not influenced by whether the calculus is based on data from surveys of citizens, party elites or other expert judges. The Effective Number of Parties was measured exactly as in the first analysis, on the basis of vote fractionalization in the last legislative election. The Age of Democracy, however, was not counted in months, since further aging beyond a certain point seems unlikely to cause a notable change in voters’ information or motivation. Instead, a crude distinction was used to capture what seems to be the most relevant variation in political system age within the second sample. Thus, the ‘old’ democracies of Australia, Britain, Western Germany, Israel, The Netherlands, New Zealand, and Norway were coded 2; the ‘almost old’ democracies of Argentina and Spain 1.67; the new democracies of the Czech Republic, East Germany, Hungary, Poland, Romania and Taiwan 1.33, while Mexico and Ukraine were coded 1 as they were still dubiously democratic countries at the time when the CSES-module was administered. The results of the second analysis are displayed in the bottom panel of Tables 1 and 2. Citizens’ Knowledge of Party Positions is almost perfectly explained by the model – see how close the adjusted R-square gets to one. The only statistically significant effect belongs to the Spread of Party Positions, which is far stronger than the parallel impact of the Clarity of Party Positions in the first analysis. The difference of the effects cannot be 13 explained with the difference between the two variables: as I indicated above, the spread and clarity of party positions are so closely correlated that, in appropriate comparative contexts, they must be interchangeable. Rather, the strikingly stronger effect of spread in the second than of clarity in the first analysis must be due to other factors. The first is provided by the differences in the wording of the issue scales that were used, in the first analysis, to calculate the Clarity of Party Positions variable on the one hand, and Citizens’ Knowledge of Party Positions on the other (see Appendix A). The second reason must be that in the second analysis both the dependent and the independent variable are based on citizens’ responses, which was not the case in the first analysis. When all four independent variables are simultaneously entered in the second equation, none appears to register a statistically significant effect in the second analysis (see Table 2). However, the explained variance is respectably high – in fact it is even higher than in the first analysis, where citizens’ knowledge recorded a significant effect. This puzzle is explained by the extremely close correlation between Citizens’ Knowledge and the Spread of Party Positions in the second analysis. When a stepwise regression procedure is employed to tackle their colinearity, then only Citizens Knowledge of Party Positions enters the equation, recording an effect significant well below the .001 level and explaining a healthy 65 percent of the cross-national variance in the degree of issue voting (data not shown, but cf. Table 3 about the pairwise correlations of the variables). 4. Discussion The results neatly support V.O. Key’s proposition about what ‘the most discriminating popular judgement’ can achieve when party positions are not sufficiently clear. Citizens’ knowledge of party positions responds to the clarity/spread of party position, and if the same issue scales are used to measure both, the latter alone explains almost all crossnational variance in the former. The degree of issue voting, in its turn, is strongly dependent on consensus in the electorate about where each relevant party stands on the issue. Citizens’ Knowledge of Party Positions alone explains between one and two-thirds of the variance in issue voting across political contexts. By and large, then, citizens live up to the role assigned to them in normative democratic theory, provided that the party system offers them an opportunity to do so. Minor additions to the costs of information – like 14 those implied by more parties or the context of a new democracy – do not have a big impact on their behaviour. To be sure, I could not refuse the proposition that the age of democracy and the effective number of parties may directly influence the degree of issue voting – after all, the respective effects are positive, although clearly insignificant, in both analyses. Yet, even if their small direct effects were real, they surely dwarf the large indirect effect of the clarity (and spread) of party positions. At least in this respect, then, citizens’ behaviour really does not differ much between old and new, two-party and multiparty democracies – at least not in comparison with the large difference between political systems with greatly varying clarity of party positions. It is also noteworthy that the spread of party position does not seem to have (much) direct effect on the degree of issue voting either. This implies that high party polarization does not increase issue voting directly, i.e. through raising the stakes of competition: its only effect on issue voting is through the clarity of party positions. This would all seem to confirm the central importance of clear, though not necessarily extreme party stances for the emergence of policy congruence between voters and parties. Yet, we saw that the clarity of party positions is almost perfectly explained by their spread – i.e. the policy distance between the parties. This suggests that ideological polarization has at least one very positive contribution to the democratic process. So do extremist parties, whenever they are called extremist because of their commitment to an unusual minority position on issues, and not for a widely condemned behavioural pattern. Whatever the merits of their issue positions are otherwise, they help to clarify the relative place of parties in the issue space. This last conclusion may not be entirely to the personal liking of Hans-Dieter Klingemann. But if there really is a trade-off between the positive and negative contributions of extremist parties to the health of democracy, it was the work of scholars like him, and their demonstration of the virtues of principled disagreement in the democratic process, which helps us discover it. 15 Appendix A: Issue scales in the first analysis As explained in the main text, in the first analysis the Clarity of Party Positions and the Citizens’ Knowledge of Party Positions were calculated from parallel – but not strictly comparable – survey data coming from the Kitschelt et al. survey of party activists and the CEU (1992-) surveys of citizens. Five issue questions were chosen from both surveys, and these were paired with each other so as to produce comparable data on the clarity of party positions and how much citizens know about them in five different issue domains in three countries. The respondents in the elite-survey were asked to tell how important some ten potentially controversial issues were for their party, and to locate all parties on 20-point issue (e.g. more or less progressive income taxation) and ideological (Left vs. Right, clerical vs. secular, and so on) scales. The preambulum of Q.17 (the question on perceived party positions) in the CEU-surveys read as follows: "I am going to read some political goals. Please, tell me after each, which party or parties in ... [COUNTRY] you think really wish to reach that objective. You can name maximum three parties in each case. Then I am going to ask you which party you think is the least likely to pursue that goal. Please, consider every parties operating in our country, not only those which we talked about earlier." The five issue domains and the phrasing of the paired items from the two surveys were as follows. Privatization (Kitschelt-survey): "According to some politicians the privatization of the state owned companies and the selection of the new owners should be directed by the goals of economic efficiency and fast privatization. According to other politicians, also the aspects of social and political justice must be taken into account even if this leads to a slow down of the privatization process." (CEU-surveys): "Speed up the privatization of stateowned companies." Market economy (Kitschelt-survey): "Please place each party on a scale where supporters of state intervention into the economy are on the one end, and supporters of free market economy on the other." (CEU-surveys): "Help the development of private enterprises and a free market economy in ... [COUNTRY]" Churches (Kitschelt-survey): "According to some politicians religion has to provide the moral guidelines for post-communist ... [COUNTRY]. Therefore, it is mandatory for 16 the state to help promoting religious faith, and the churches must have a significant saying on the content of public education. According to other politicians religion belongs to the private sphere and it is not the responsibility of the state to help promoting religious faith. Thus, churches should not exercise a significant influence on the curicula of state-run schools." (CEU-surveys): "Increase the influence of religion and the Church(es)." Former communists (Kitschelt-survey): "According to some politicians the former upper and intermediate level leaders of the ... [ruling party of communist period], because of their past sins, must be excluded from political life and from the privatization of state property by legal, administrative and political means. According to other politicians former communists must be guaranteed the same opportunities to exercise political and economic rights as anybody else. They think that any law, administrative or political rule that aims at excluding former communists from economic or political life is unjustifiable." (CEUsurveys): "Removing former communist party members from positions of influence." Nationalism (Kitschelt-survey): "Please place each party on a scale where supporters of the values of liberal individualism are on one end, and supporters of traditional ... [Polish, Czech, Hungarian] culture and national solidarity are located on the other end." (CEU-surveys): "Strengthen national feelings." Appendix B: Coding of citizens’ party preference In the computation of the Issue Voting and the Citizens’ Knowledge of Party Positions variables for the first analysis, the respondents’ party preference was derived from their responses to a question about which party they would vote for if there were a parliamentary election next weekend. In the second analysis, respondents’ recalled vote in the last legislative election served as the indicator of party preference. To increase the number of cases in the analysis, some closely related small parties had to be collapsed in one category. In the first analysis, the variable was coded as follows. Czech Republic: 1=ODA; 2=CSSD; 3=KDU-CSL; 4=OH, SD, SD-LSNS; 5=SPR-RSC; 6=KSCM; 7= ODS; Hungary: 1=FIDESZ (from 1995: Fidesz-MPP); 2=FKGP; 3=KDNP; 4=MDF; 5=MSZP; 6=SZDSZ; Poland: 1=PC; 2=ZChN, KKWO or just ‘Christian party'; 3=PSL; 4=SLD, SDRP; 5=UD, KLD, UW; 6=KPN; 7=UP; 8=Solidarnosc; 9=BBWR. 17 In the second analysis, the party preference variable was coded as follows. Argentina: 1=Alianza UCR-Frepaso; 2=PJ; 3=ApR; 0=other parties; Australia: 1=Liberal; 2=Australian Labor; 3=National; 4=Australian Democrats; 5=Greens; 0=other parties; Czech Republic: 1=CSSD; 2=KDU-CSL; 3=KSCM; 4=ODA; 5=ODS; 6=SPR-RSC; 0=other parties; East and West Germany: 1=CDU, CSU; 2=SPD; 3=FDP; 4=B90-Gruene; 5=DS; 0=other parties; Hungary: 1=Fidesz-MPP; 2=FKGP; 3=MDF; 4=MIEP: 4=MSZP; 6=SZDSZ; 0=other parties; Israel: 1=Likud; 2=Avoda, Meretz; 3=Shas, Mafdal; 4=Tzomet; 0=other parties; Mexico: 1=PAN; 2=PRI; 3=PRD; 4=PT; 0=other parties; The Netherlands: 1=PVDA; 2=CDA; 3=VVD; 4=D66; 5=GroenLinks; 6=SGP, GPV, RPF; 7=SP; 0=other parties; New Zealand: 1=Labour; 2=National; 3=New Zealand First; 4=Alliance; 5=ACT; 6=Christian Coalition; 0=other parties; Norway: 1=Red Electoral Alliance; 2=Socialist Left Party; 3=Labor Party; 4=Liberal Party; 5=Christian People's Party; 6=Center Party; 7=Conservative Party; 8=Progress Party; 0=other parties; Poland: 1=UP; 2=UW; 3=AWS; 4=SLD; 5=PSL; 6=ROP; 0=other parties; Romania: 1=USD and components; 2=PDSR; 3=UDMR/RMDSZ; 4=PNR; 5=CDR and components; 0=other parties; Spain: 1=PP; 2=PSOE; 3=IU; 4=CiU; 0=other parties; Taiwan: 1=KMT; 2=DPP; 3=New Party; 0=other parties; Ukraine: 1=All-Ukraine Association Gromada; 2=Party of Greens of Ukraine; 3=Communist Party of Ukraine; 4=Peoples Rukh of Ukraine; 5=Peoples-Democratic Party; 6=Electoral block Socialist Party of Ukraine; 7=SocialDemocratic Party of Ukraine; 8=Progressive Socialist Party; 0=other parties; England and Wales: 1=Conservative; 2=Labour; 3=Liberal Democrats; 0=other parties; Scotland: 1=Conservative; 2=Labour; 3=Liberal Democrats; 4=SNP; 0=other parties. 18 References: Alvarez, R. Michael (1997): Information and Elections. Ann Arbor, MI: University of Michigan Press. Berelson, Bernard R./Lazarsfeld, Paul F./McPhee, William N. (1954): Voting. A Study of Public Opinion Formation in a Presidential Campaign. Chicago, IL: The University of Chicago Press. Campbell, Angus/Converse, Philip E./Miller, Warren E./Stokes, Donald (1960): The American Voter. New York: John Wiley & Sons. Carmines, Edward G./Stimson, James A. (1980): The Two Faces of Issue Voting. In: American Political Science Review 74, pp. 78-91. CEU (Central European University) (1992-97): The Development of Party Systems and Electoral Alignments in East Central Europe. Machine readable data files. Budapest: Department of Political Science, Central European University. Distributor: Tárki, Budapest. CSES (Comparative Study of Electoral Systems and Center for Political Studies, University of Michigan) (2000): Module 1 Micro-Data, June 2000 Version. Ann Arbor, MI: University of Michigan, Center for Political Studies [producer and distributor]. URL: http:/www.umich.edu/~nes/cses/ Franklin, Mark (1985): The Decline of Class Voting in Britain. Oxford: Oxford University Press. Fuchs, Dieter/Klingemann, Hans-Dieter (1989): The Left-Right Schema. In: Jennings, M. Kent/van Deth, Jan W. (Eds.) (1989): Continuities in Political Action: A Longitudinal Study of Political Orientations in Three Western Democracies. Berlin: Walter de Gruyter, pp. 203-34. Granberg, Donald/Holmberg, Sören (1988): The Political System Matters: Social Psychology and Voting Behavior in Sweden and the United States. Cambridge: Cambridge University Press. Inglehart, Ronald/Klingemann, Hans-Dieter (1976): Party Identification, Ideological Preference, and Left-Right Dimension among Western Mass Publics. In: Budge, Ian/Crewe, Ivor /Farlie, Dennis (Eds.) (1976): Party Identification and Beyond: Representations of Voting and Party Competition. London: John Wiley & Sons, pp. 19 243-75. Kaase, Max/ Klingemann, Hans-Dieter (1994): The Cumbersome Way to Partisan Orientations in a "New" Democracy: The Case of the Former GDR. In: Jennings, M. Kent/Mann, Thomas E. (Eds.) (1994): Elections at Home and Abroad. Essays in Honor of Warren E. Miller. Ann Arbor, MI: The University of Michigan Press, pp. 123-56. Key, V. O., Jr. (1966): The Responsible Electorate. Cambridge, MA: The Belknap Press of Harvard University Press. Kitschelt, Herbert/Mansfeldova, Zdenka/Markowski, Radoslaw/Tóka, Gábor (1999): PostCommunist Party Systems: Competition, Representation, and Inter-Party Cooperation. Cambridge: Cambridge University Press. Klingemann, Hans D. (1973): Dimensions of Political Belief Systems: 'Levels of Conceptualization' as a Variable. Some Results for the USA and FRG 1968/69. In: Comparative Political Studies 5, pp. 93-106. Klingemann, Hans-Dieter (1979a): Measuring Ideological Conceptualizations. In: Barnes, Samuel H./Kaase, Max (Eds.) (1979): Political Action: Mass Participation in Five Western Democracies. Beverly Hills, CA: Sage, pp. 215-54. Klingemann, Hans-Dieter (1979b): The Background of Ideological Conceptualizations. In: Barnes, Samuel H./Kaase, Max (Eds.) (1979): Political Action: Mass Participation in Five Western Democracies. Beverly Hills, CA: Sage, pp. 255-77. Klingemann, Hans-Dieter (1995): Party Positions and Voter Orientations. In: Klingemann, Hans-Dieter/Fuchs, Dieter (Eds.) (1995): Citizens and the State. Oxford: Oxford University Press, pp. 183-205. Klingemann, Hans-Dieter/Hofferbert, Richard I./Budge, Ian (1994): Parties, Policy and Democracy. Boulder, CO: Westview Press. Klingemann, Hans-Dieter/Taylor, Charles L. (1977): Affektive Parteiorientierung, Kanzlerkandidaten und Issues. In: Kaase, Max (Ed.) (1977): Wahlsoziologie heute. Opladen: Westdeutscher Verlag, pp. 301-47. Klingemann, Hans-Dieter/Wattenberg, Martin P. (1992): Decaying Versus Developing Party Systems: A Comparison of Party Images in the United States and West Germany. In: British Journal of Political Science 22, pp. 131-49. 20 Nie, Norman H./Verba, Sidney/Petrocik, John R. (1976): The Changing American Voter. Cambridge, MA: Harvard University Press. Pomper, Gerald M. (1972): From Confusion to Clarity: Issues and American Voters 19561968. In: American Political Science Review 66, pp. 415-28. RePass, David (1971): Issue Salience and Party Choice. In: American Political Science Review 65, pp. 389-400. Rose, Richard/McAllister, Ian (1986): Voters Begin to Choose. London: Sage. Tóka, Gábor (1998): "Party Positions." In: van Deth, Jan W. (Ed.) (1998): Comparative Politics: The Problem of Equivalence. London: Routledge, pp. 180-204. Welzel, Christian/Inglehart, Ronald/Klingemann, Hans-Dieter (2000): "Human Development as a Universal Theory of Social Change: Cross-Level and CrossCultural Evidence from 63 Societies." Paper presented at the 28th Joint Sessions and Workshops of the European Consortium for Political Research, Copenhagen, Denmark, 14-19 April 2000. 21 Table 1: Determinants of Citizens' Knowledge of Party Positions First analysis (weighted N=15 issue-party system pairs, each observed at 6-8 points in time) Independent variables: b (s.e.) beta Age of Democracy (Version 1) Effective Number of Parties Clarity of Party Positions -.052 (.358) -.050 -2.427 (2.757) -.303 .780 (.371) .533* Adjusted R2 .125 Second analysis (N=18 party systems, each observed at a single point in time) Independent variables: b (s.e.) Age of Democracy (Version 2) Effective Number of Parties Spread of Party Positions 6.474 (4.424) 2.191 (1.260) 27.038 (2.729) Adjusted R2 .871 beta .129 .155 .881** *: The coefficient is significant at the .10 or lower level **: The coefficient is significant at the .01 or lower level Note: Table entries were derived from OLS-regression analyses with Citizens' Knowledge of Party Positions as the dependent variable. On data sources and the construction of the variables see the main text. 22 Table 2: Determinants of Issue Voting First analysis (weighted N=15 issue-party system pairs, each observed at 5-7 points in time) Independent variables: Age of Democracy (Version 1) Effective Number of Parties Clarity of Party Positions Citizens' Knowledge of Party Positions Adjusted R2 b (s.e.) beta .001 .009 .000 .004 (.002) (.012) (.002) (.001) .161 .222 .016 .720* .333 Second analysis (N=18 party systems, each observed at a single point in time) Independent variables: b Age of Democracy (Version 2) -.010 Effective Number of Parties .007 Spread of Party Positions .003 Citizens' Knowledge of Party Positions .007 Adjusted R2 (s.e.) beta (.080) (.023) (.130) (.005) -.023 .051 .012 .793 .566 *: The coefficient is significant at the .10 or lower level **: The coefficient is significant at the .01 or lower level Note: Table entries were derived from OLS-regression analyses with Issue Voting as the dependent variable. On data sources and the construction of the variables see the main text. 23 Table 3: Pairwise correlations between the variables in the analysis First analysis (weighted N=15 issue-party system pairs, each observed at 5-7 points in time) Age of Effect. Clarity Citizen Democr. N of of Party Know(Vers. 1) Parties Positions ledge Effective Number of Parties -.68** 1.00 Clarity of Party Positions -.09 .16 Citizens' Knowledge of Party Positions .11 -.19 Issue Voting .09 -.02 1.00 .49* .39 1.00 .70** Second analysis (N=18 party systems, each observed at a single point in time) Age of Effect. Spread Citizen Democr. N of of Party Know(Vers. 2) Parties Positions ledge Effective Number of Parties Spread of Party Positions Citizens' Knowledge of Party Positions Issue Voting .14 .09 .23 .17 1.00 .18 .33 .32 1.00 .92** 1.00 .75** .82** *: The coefficient is significant at the .10 or lower level (two-tailed) **: The coefficient is significant at the .01 or lower level Note: On data sources and the construction of the variables see the main text. 24