(very draft version comments most welcome)

CONTEXT WITHIN A CONTEXT: ON THE DIFFERENT IMPACT OF 'CONTEXTUAL FACTORS IN NATIONAL AND SECOND ORDER ELECTIONS (very draft version comments most welcome) Radoslaw Markowski, Michal Kotnarowski and Mikolaj Czesnik Warsaw School of Social Sciences and Humanities and Polish Academy of Sciences PAPER PREPARED FOR PRESENTATION AT THE FINAL USER CONFERENCE OF THE PIREDEU DESIGN STUDY BRUSSELS, NOVEMBER 18-19, 2010 PANEL SESSION IV.2: LINKING NATIONAL AND EUROPEAN ELECTORAL ARENAS

Objective The main objective of this paper is to compare the effects of interaction between two types of contextual variables. We distinguish the 'traditional' institutional and social contextual factors (federalism vs. unitarism, electoral rules, party system format, disproportionality and the like) at the first MACRO level and we add to it a sort of 'metacontextual' variable at the second MACRO level, which is the type of election (national vs. European). We investigate its' impact on voter turnout, which is one of the major problems in studies on EP elections. The research is only partly conclusive about the causes of this phenomenon, but in almost all European countries the turnout in EP elections is significantly lower than in national elections. Studies in electoral research (most often implicitly) assume that macro-level determinants of voter behaviour operate in the same way on both levels (i.e. in national and in EP elections level). This assumption might be misleading and thus it needs in-depths scrutiny. Briefly: it remains an open empirical question whether contextual (macrolevel) factors have a similar effect on voter turnout in both national and EP elections. Our scholarly curiosity is focused on checking whether and if yes, to what extent we can claim that the same 'traditional' contextual variables exert a different influence on voter participation depending on the 'meta-context' of the type of an election. The relationship controls for the impact of socio-demographics. Our paper is a comparison of those countries which feature in both CSES and PIREDEU projects. Theoretical background The broadest conceivable understanding of a context refers to an environment in which individuals are located, in which they live and decide to behave in a certain way. Sprague (1982: 100) insists that in terms of the structure of the impact, contextual analysis seeks to answer how these 'environmental properties determine variation in a given behaviour Context within a context 2

of interest'. The most obvious set of institutions for behaviourally oriented political scientists is, of course, the institutional design of countries' political institutions (Klingemann 2007). The other way to conceptualize 'context' derives from macro-sociology, it is about the social settings in which individuals live and is founded on an assumption of behavioural interdependence, and empirically manifests itself as aggregate of individual traits, say average educational attainment in a country (Hukfeldt & Sprague 1993). In case of this analysis, we concentrate predominantly on the impact of political institutions (sometimes their consequences) at the first macro level, but we ask ourselves additional question whether this institutions perform similarly in two 'meta-contexts' national and European. According to the well-established scholarly research and literature voter turnout is a function of both individual traits and contextual factors. It is assumed that people vote (or abstain) because of two major reasons. Firstly, they have certain characteristics (traits, attitudes, resources etc.), which have an impact on their decision whether to vote or to abstain. Some people are better equipped with these resources, some are poorer in this regard. Secondly, they live in particular (cross-nationally distinct) socio-political milieus. Those milieus are different as far as ease of voter decision is concerned: some are voter-friendly, other are more demanding. Accordingly, in empirical research into voter turnout two key approaches can be distinguished. If students of politics are about to explain voter turnout, they usually either refer to institutions, or analyze citizens. While first approach addresses macro-level variables, the second is first of all focused on individual-level characteristics. Models of the first approach for the most part address question of cross-national variance in voter turnout, whereas second approach models are principally used to answer the question why some individuals vote while other do not. Context within a context 3

There is a vast scholarly literature, based on rich empirical evidence, showing that voter turnout is in fact a function of both micro-level and macro-level characteristics, and the interplay between them (c.f. Anduiza-Perea 2003, Franklin et al. 2004). There is however another important scholarly thread, on the so-called second-order elections [SOEs], that covers the other part of our story presented in this paper, namely the impact of the type of elections on voting behaviour (Reif and Schmitt 1980; Carrubba and Timpone 2000; Koepke and Ringe 2006). The theory, proposed and developed by Reif and Schmitt three decades ago in its trivialized form and focused on the issues of interest to us in this paper goes as follows: The SOEs (de facto are and so are perceived by citizens) have lower stakes they do not decide about power distribution among the most important political institutions, they do not determine who governs in the country. Because they are considered less important, several consequences follow: (a) the turnout is lower than in the national elections, (b) voters overproportionally vote for opposition parties, (c) they tend to vote more sincerely and less strategically (Blais et al 2001; Alvarez & Nagler 2000), (d) they vote expressively, i.e. wishing to convey a critical message to the incumbents that they are not performing as good as expected (Oppenhuis et al. 1996). In addition, SOEs main campaign theme is more related to national that EU politics (Marsh 1998; Hix 1999). As a consequence, small parties receive relatively higher support, whereas governmental and big parties lose support. The above, deliberately short and simplified, description of the ontology of EP elections is an important starting point for the design of our analysis. Its' design relates the voting behaviour directly to the type of election it takes place (in our terminology to the 'meta-context'), without accounting for particular 'traditional macro-institutional context'. In our analysis both contexts are taken into account. In this particular paper we are first of all interested in investigating effects of macrolevel variables. Their impact on voter turnout is twofold. Firstly, they have a direct impact on Context within a context 4

decision whether to vote or to abstain. Secondly, they also have an indirect impact on decision whether to vote or to abstain: context (macro-level variables) structures the relationship between micro-level variables and voter turnout. Research questions Our research questions are de facto focuses on different nature of contexts. We are interested in investigating the impact of macro-level variables on voter turnout in different contexts. In other words, we aim to study context within a context : whether the impact of contextual factors on voter turnout the same in the national and EP elections? Contemporary political research does not provide relevant empirical evidence on this matter. Indeed, the effect of macro-level variables on political phenomena in different contexts has been very rarely a subject of scholarly research. Let us stress at this point that in the usual (prior to PIREDEU initiative) research on elections to the European Parliament (EP), the hypotheses derived from the SOEs literature assume there exists a direct link between the type of elections and what happens next at the individual level. In other (our) words, it is about the impact of the meta-context alone on individual behaviour. In national election studies, comparative ones in particular, as in CSES, there is no meta-context. The overall causal linkage assumes political institutions (federalism, presidentialism, electoral rules) impact, both directly and indirectly, the micro-level preferences and behaviour of individuals. Because of the novelty of our approach, aimed at testing how one context operates within a higher level context, rather than detailed hypotheses, we put forward several research questions and hypothetical expectations. First group pertains to the impact of the institutional contextual variables, the second, to the classical political-attitudinal ones (like PID, LR and the like), and, the third refers to classical socio-demographic ones. Needless to say that the Context within a context 5

final composition of each of the baskets is selected taking into account their theoretical relevance as well as methodological and statistical applicability (collnearities, tautologies, irrelevance). The simple hypothetical expectations (HE) concerning the impact of institutional variables in both 'meta-contexts' submit the following: HE1.1: Semipresidential design of a political system is anticipated to impede voter turnout in both types of elections, as in parliamentary-cabinet democracies double executive enhances legislative stalemate, uncertainty, disorients voters and in addition gives another opportunity to express citizens will, which in turn (and paradoxically) means it 'consumes' part of the otherwise limited reservoir of citizens' political activity. 1 Yet, this factor is expected to have less of an impact on participation in the EP elections (EP meta-context), because of lower relevance of presidents for EU politics, in which governments, and their heads have clearly more prominent position. HE1.2: In more consensual multi-party systems, the ones unveiling higher proportionality of the seat distribution vis a vis votes, we expect higher turnout, mainly because votes are not wasted, because there is more programmatic choice, no experience with discouraging effects of 'safe seats' environment and encouragement for minorities to seek representation. This effect, however, is expected to be less pronounced in EP election megacontext, as in these elections the PR rules are universally applied. HE1.3: Decentralization and federalism are political tools by which people should execute more specific, regional, and local interest. In national elections this factor is expected to enhance electoral participation, moreover we anticipate that its' impact is even stronger in the EP meta-context, because of the widespread incentives to represent local/regional idiosyncratic interests that are recognized and thus can be directly address at EU level. 1 This argument has to do with the simple logic the more elections and other political choice acts (referenda, plebiscites) citizens are exposed to the more likely in each of them to see less voters participating. Switzerland being the best known example of the mechanism at work. Context within a context 6

HE1.4: Communist legacies of a country are envisaged to hinder electoral participation, mainly because of the newness of the democratic institutions, high institutional instability, voter volatility, no time to develop intergenerational party identification and consolidate stable relationships between social groups and labour market positions with the party system supply. This factor is presumed to impact electoral behaviour even stronger in EP setting, as the newness of the EP elections and uncertainty about its general sense is more of an obstacle to participation in EP that in the more 'familiarized' (several electoral cycles compared to one for EP) by now. HE 1.5 Closely related expectation pertains to the overall affluence of a given country expressed in terms of GDP per capita. The factor is strongly related with the communist past, so we are interested in both how the affluence affects turnout alone and once we control for communist past. The prediction is that the more affluent the country the more likely its citizens to participate, however underdeveloped and poorer countries of EU have very strong incentive to influence the composition of their representatives in the pan-european institution where distribution of huge amounts of funds are being decided. HE 1.6 Is similar in logic and structure to HE1.5, except that it refers to the overall scholarization level of a given country. Briefly, we expect high aggregated educational attainment to enhance participation, however less 'educated' societies of the poorer countries might exert a direct and positive impact on turnout.. At the individual level we ought to hypothesize mostly about the political attitudinal factors satisfaction with the system, party identification and ideological radicalism. HE2.1 The expectation here is that in both 'meta-contexts' the effect of these three political variables will exert a positive effect on electoral participation, i.e. the participation is envisaged to be higher among party identifiers, satisfied with the system and more Context within a context 7

ideologically radically oriented. We expect however that they will be less significant (or even irrelevant) in EP context, as far as party identification is concerned in particular. RQ2.2: We are interested as well how the classical political-institutional macro-factors operate, once we control for this set of political-attitudinal variables. Research design and data We model voter turnout as a function of several micro-level and macro-level variables. In order to compare the effects of macro-level variables on voter turnout we run two identical models (on national data and EP data). Our dependent variable is voter turnout in the last parliamentary election (CSES module 2) and voter turnout in the last 2009 EP election (PIREDEU). To be more precise, the dependent variable is self-reported voter turnout in both elections. The data are not validated, due to the fact that validation of voter turnout data is illegal (or at least impossible) in many European countries. The dependent variable is binomial (0-abstained, 1-voted). In order to test our hypothesis we employ binomial logistic regression (due to the dichotomous character of the dependent variable). Our independent variables include socio-demographics (gender, age, education, place of residence), political variables (party identification, political radicalism 2, satisfaction with democracy in the country) and macro-level 'contextual' variables (proportion of citizens with higher education, disproportionality Gallagher index, direct election of a head of the state, fiscal decentralisation, communist legacy, GDP per capita). In the analyses below CSES (module 2) and EES/PIREDEU (2009) data are used. Altogether 17 countries, which feature in both CSES and PIREDEU projects, are included in the analysis. 2 Operationalised as squared left-right self-placement (recoded from 0 to 10 scale into -5 to 5 scale). Context within a context 8

Empirical analyses Tables 1-4 present the main findings. They contain the binomial logistic regression results (coefficients and standard errors are reported); models are designed separately for the CSES and PIREDEU data. Firstly (table 1 model 1 and table 2 model 1), voter turnout is modelled as a function of socio-demographic variables (gender, age, education, place of residence). Secondly (table 1 model 2 and table 2 model 2), it is modelled as a function of political variables (party identification, political extremism, satisfaction with democracy in the country). Thirdly (table 1 model 3 and table 2 model 3), it is modelled as a function of macro-variables (proportion of citizens with higher education, disproportionality Gallagher index, direct election of a head of the state, fiscal decentralisation, communist legacy, GDP per capita 3 ). Finally (tables 3 and 4), voter turnout is modelled in a 'full model' as a function of all these variables. Both complete models show a moderate fit to the data, with a pseudo- R² values of 0,09 (CSES) and 0,04 (PIREDEU) respectively. Let us start with the less important variables and their relationships with voter turnout. Overall, we note many predictable results, consistent with the previous scholarly literature on voter turnout. Firstly, we find statistically significant, positive effects of two sociodemographics, namely age and education. Mature and better educated respondents are more likely to vote. Also place of residence has a statistically significant impact on voter turnout, but this effect is entirely different in each meta-contexts : in the EP elections it is positive and in the national elections - negative. Gender is significantly related to voter turnout only in the EP election: female are less likely to vote. Secondly, as presumed in HE2.1, we find statistically significant, positive effects of all three political variables included in the analysis: party identification, political extremism and satisfaction with democracy in the country. Each of these variables significantly increases 3 Initially, more macro-variables were included in the analysis, but several have been omitted due to collinearity problems. Context within a context 9

probability of voting: respondents, who are more partisan, unveil more radical LR positioning and are more satisfied with democracy, are at the same time more likely to vote. If satisfaction with democracy is replaced with the satisfaction with government 4, the results are different: statistically significant impact of satisfaction with government on voter turnout is observed only in national elections. Table 1. Determinants of voter turnout in national elections (CSES module 2). Estimate Std. Error z value Pr(> z ) Model 1 Female -0.03 0.04-0.70 0.482 Age 0.03 0.00 16.74 0.001 *** Educsecondary 0.42 0.05 7.91 0.001 *** eductertiary incl. Incompleted 0.87 0.07 12.28 0.001 *** residencesmall or middle sized town -0.00 0.06-0.02 0.980 residencesuburbs of large town or city -0.03 0.07-0.47 0.638 residencelarge town or city -0.26 0.06-4.15 0.001 *** Model 2 pid01 0.94 0.04 19.05 0.001 *** lr2 0.01 0.00 4.21 0.001 *** Satdem 0.23 0.03 7.49 0.001 *** Model 3 high_edu_percent 1.88 0.54 3.47 0.001 *** elec_97p -7.26 0.61-12.07 0.001 *** dec_97p 1.87 0.24 0.76 0.441 comm_leg -3.59 0.11-3.19 0.001 ** Gdp 2.65 0.47 0.55 0.579 direct_head_bin -4.19 0.56-7.39 0.001 *** Source: CSES. 4 Due to collinearity this variables cannot be included in the same model; results of this analysis are not reported in the paper; they are however available from the authors upon request. Context within a context 10

However, these micro-level findings are of lower relevance to us. The independent variables we are most interested in are the institutional macro-level, contextual variables. Our major aim in this paper is to assess their impact on voter turnout in two different metacontexts : in national elections and in EP elections. More precisely, we are interested in whether or not these relationships are identical, similar or different. According to the results of the empirical analyses only one out of six macro-variables included in our models is significantly (and similarly) related to voter turnout in both contexts (national and European), even when we control for micro-level variables and other macrolevel variables effects. In both meta-contexts we find statistically significant, negative effect of communist legacy on voter turnout: post-communist citizens are less likely to vote. The remaining macro-variables operate differently in different meta-contexts. Table 2. Determinants of voter turnout in the 2009 EP elections. Estimate Std. Error z value Pr(> z ) Model 1 Female -0.35 0.03-9.61 0.001 *** Age 0.01 0.00 12.26 0.001 *** edu_age 0.02 0.00 7.76 0.001 *** residencesmall or middle sized town 0.09 0.04 1.97 0.048 * residencesuburbs of large town or city 0.09 0.06 1.61 0.107 residencelarge town or city 0.13 0.05 2.56 0.014 * Model 2 pid01 0.57 0.04 14.12 0.001 *** lr2 0.01 0.00 8.71 0.001 *** Satdem 0.12 0.02 5.35 0.001 *** Model 3 high_edu_percent -1.11 0.46-2.41 0.015 * elec_97p -1.45 0.62-2.97 0.002 ** dec_97p 2.43 2.08 1.17 0.241 comm_leg -7.77 0.96-8.02 0.001 *** Gdp -1.22 0.45-2.68 0.007 ** direct_head_bin -9.01 4.98-1.81 0.070 Context within a context 11

Source: PIREDEU. Firstly, our prediction expressed in HE1.6 is confirmed -- in both meta-contexts we find statistically significant effect of proportion of citizens with higher education on voter turnout. However, the effect is entirely different in diverse electoral meta-contexts. It is positive in the national elections (in countries, where there are more educated people, citizens are more likely to vote) and negative in the EP elections. This effect is more statistically significant in EP elections (where 'full' model effects are accounted for). Table 3. Determinants of voter turnout in national elections (CSES module 2). Estimate Std. Error z value Pr(> z ) (Intercept) 0.33 0.21 1.60 0.11 Female 0.02 0.05 0.35 0.73 Age 0.02 0.00 15.23 0.00 *** Educsecondary 0.42 0.05 8.02 0.00 *** eductertiary incl. Incompleted 0.81 0.07 11.24 0.00 *** residencesmall or middle sized town -0.16 0.06-2.68 0.01 ** residencesuburbs of large town or city -0.17 0.07-2.38 0.02 * residencelarge town or city -0.32 0.06-5.14 0.00 *** pid01 0.81 0.05 16.45 0.00 *** lr2 0.01 0.00 3.70 0.00 *** Satdem 0.22 0.03 7.36 0.00 *** high_edu_percent 0.02 0.01 3.20 0.00 ** elec_97p -0.08 0.01-12.56 0.00 *** dec_97p 0.39 0.25 1.54 0.12 comm_leg -0.54 0.12-4.54 0.00 *** Gdp 0.00 0.00-1.21 0.22 direct_head_bin -0.34 0.06-5.85 0.00 *** Source: CSES. Context within a context 12

Secondly, supposition expressed in HE 1.2 turned out fairly correct, i.e. with certain caveats. Only in national elections we find clear, statistically significant, negative effect of disproportionality of electoral system (operationalised with the Gallagher index) on voter turnout: the higher the value of index of disproportionality (for the country), the lower the probability of voting (of a citizen of this country). In the EP elections the picture is more ambiguous; here this effect is still significant and direct, though less pronounced than in national Elections when we do not control for micro-level variables, but disappears once they are added to the model. Table 4. Determinants of voter turnout in the 2009 EP elections. Estimate Std. Error z value Pr(> z ) (Intercept) -0.74 0.17-4.28 0.00 *** Female -0.32 0.04-8.88 0.00 *** Age 0.01 0.00 8.33 0.00 *** edu_age 0.03 0.00 7.33 0.00 *** residencesmall or middle sized town 0.11 0.05 2.28 0.02 * residencesuburbs of large town or city 0.15 0.06 2.48 0.01 * residencelarge town or city 0.13 0.05 2.64 0.01 ** pid01 0.48 0.04 11.89 0.00 *** lr2 0.02 0.00 9.84 0.00 *** satdem_cntry 0.10 0.02 4.14 0.00 *** high_edu_percent -0.02 0.00-3.38 0.00 *** elec_9706 0.00 0.00-0.58 0.57 dec_9706 0.18 0.21 0.86 0.39 comm_leg -0.83 0.10-8.28 0.00 *** Gdp -0.00 0.00-3.09 0.00 ** direct_head_bindirect election -0.07 0.05-1.35 0.18 Source: PIREDEU. Thirdly, as predicted in HE1.1, the effect of direct election of a head of the state on voter turnout is registered only in one of our meta-contexts -- only in the national elections it Context within a context 13

unveils statistical significance. In countries, where people elect the head of the state directly, citizens are less likely to vote, all other things being equal. This variable does not count in the EP elections context. Fourthly, only in the EP elections we find statistically significant, negative effect of GDP per capita on voter turnout: voters from richer countries are on average less likely to vote, all other things being equal (HE 1.5). In the national elections this effect is absent. And finally, we do not observe any significant impact of the fiscal decentralisation variable in neither of the meta-contexts, although the signs are in the predicted (HE 1.3) and in the national context it borders significance (z value =1.54, in the 'full' model). Quite intriguing is the fact that in both simple models (including only macro-level, contextual variables) and complete models (in which we control for the effect of micro-level variables) we observe very similar patterns. In other words, addition of micro-level variables into the model does not change the effects of macro-level variables on voter turnout in most cases they remain unchanged. The results of our analyses and empirical evidence gathered support our hypothetical expectations only to a certain extent. Our first hypothetical expectation (HE1.1) is confirmed partially. Semipresidential design of a political system impedes voter turnout only in national elections, while in EP elections it does not have an impact on voter turnout. This is however somewhat in line with our hypothetical expectations, as this variables has been expected to have less of an impact on voter turnout in the EP meta-context (because of lower relevance of presidents for EU politics, in which government, and their heads have clearly more prominent position). Our second hypothetical expectation (HE1.2) is rather confirmed. Proportionality results in higher (probability of) voter turnout. This effect is more visible and stronger in national elections context. In the EP meta-context this effect is less pronounced, probably due Context within a context 14

to the very fact that in these elections the PR rules are universally applied (because of EU regulations). Third of our hypothetical expectations (HE1.3) must be rejected. We do not find any empirical support for the thesis that decentralization and federalism enhance voter turnout. Even in the EP meta-context, where we have expected more incentives to represent local/regional idiosyncratic interests that are recognized and thus can be directly address at EU level, decentralization is not significantly related to voter turnout. Our fourth hypothetical expectation (HE1.4) is fully confirmed. Communist legacies of a country do diminish voter turnout regardless the electoral context. Somewhat contrary to our presuppositions, this variable operates quite similarly in both meta-contexts. Fifth hypothetical expectation (HE1.5) is confirmed only in minor part. Our prediction is that the more affluent the country the more likely its citizens to participate. In national elections there is no such effect whatsoever. In the EP elections, on the other hand, we find something quite contrary to our expectations (statistically significant, negative effect of GDP per capita on voter turnout). This fact is confirming another of our expectations: citizenries of underdeveloped and poorer countries of EU have very strong incentive to influence the composition of their representatives in the institution where distribution of huge amounts of money are being decided. Finally, empirical results referring to our sixth hypothetical expectation (HE1.6) are puzzling. It is confirmed only to some degree. Scholarisation level of a given country in both meta-contexts exerts statistically significant effect on voter turnout, but these relationships are completely distinct: positive in the national elections and negative in the EP elections. Our micro-level expectations (HE2.1) are fully confirmed. In both 'meta-contexts' the effects of all three political variables included in the analysis are similar: the more partisan, extremist and satisfied with democracy the voters, the more they are likely to vote. Moreover, Context within a context 15

the additional expectation that the relative magnitudes of the effects differ between the two 'mega-contexts' also turned out to be correct as predicted in HE2.1. Conclusions Our preliminary findings suggest that the impact of macro-level (contextual) variables on voter turnout is not the same in the national and EP electoral contexts. Although in both contexts the patterns of voter turnout are parallel, and voter turnout in national and EP elections seems to be moved by similar factors, several noteworthy differences can be put forward. In national elections both proportionality of the electoral system and (possibility of) direct election of a head of the state are important, at least as far as voter turnout is concerned. On the contrary, in EP elections these factors do not play a role, but economy (or wealth) seems to matter GDP per capita does have an significant impact on voter turnout. Thus, somewhat in line with the SOE theory, we find support for the thesis that metacontext of election matters. Though the general (individual-level) pattern of voter turnout is similar in national and EP elections, the ultimate results (the level of voter turnout in each country, which is a function of millions of individual decisions) are different due to different electoral meta-context. In other words, contextual variables operate (i.e. influence individual behaviours) within a meta-context of elections. Further analyses are needed, most notably concentrated on precise interactions between the three levels of analyses presented in this paper. What definitely helps understand the theme we introduce 'context within a context' is, on the one hand, need for more cases (be it in time or in space) and, on the other, more detailed separate analyses of country casestudies. Context within a context 16