Explaining Variation of EU Issue Voting at the Individual Level: the Role of Attribution of Responsibility

Transcription

1 Explaining Variation of EU Issue Voting at the Individual Level: the Role of Attribution of Responsibility Cal LE GALL University of Salzburg (SCEUS) Abstract: Voters take EU related issues into account when voting in national parliamentary elections (de Vries 2007, 2010, de Vries and Tillman 2011). To explain EU issue voting heterogeneity at the individual level, the literature emphasizes on political sophistication along contextual factors such as information supply and party polarization (De Vries et al., 2011). This study takes a fresh look at this phenomenon by exploring the impact of attribution of responsibility. In a multi-level system of governance where responsibilities are shared, I first expect EU issue voting to be affected by citizens perception of the EU as a ruler. Thus, I expect voters who deem the EU responsible for policy-making to be more prone to vote in national parliamentary elections based on EU issues. Conversely, EU issue voting can also be impacted by perceptions of constraints on national governments. The rationale here is different: the more a voter deems national governments to be constrained in their policy room-tomanoeuvre, the more she will be likely to vote on non-traditional issues. To explore the variation of EU issue voting at the individual level, I use a two-step hierarchical model. The estimations are carried out thanks to the use of the 2009 European Election Study (EES). The results show that attribution of responsibility to the EU has almost no effect on the variation of EU issue voting when controlled with other variables. Most importantly, attribution of responsibility to the national government is systematically the best predictor when compared with every other individual variable: both in terms of significance and coefficient. Paper presented at IPSA Panel on Leader Effects and Electoral Accountability on the 25 th of July

2 Introduction The processes of EU integration and globalization are increasingly reducing national governments room-to-manoeuver in the policy-making process. First, the integration of world markets is decreasing national governments abilities to control economic outcomes. As Hellwig puts it: policy-maker attempts to pursue strategies that diverge from neoliberal prescriptions [are now] undercut by impersonal market actors (Hellwig 2008: 1128). Second, EU institutions have gained policy competences over the years: the EU institutional triangle has exclusive competences over commercial policy (Common Commercial Policy), external trade, customs, competition rules and the conservation of marine biological resources (Common Fisheries Policy), while the ECB controls monetary policies. In these policy areas, only the EU can legislate and adopt legally binding acts, while member states need EU s approval to do so. Hence, voters should adapt to these evolutions when they are at the ballot box. Indeed, there are reasons to believe that governments fortunes in elections are linked with their respective functional responsibilities (Stein 1990): voters assess governments performance based on policies they think their governments are ought to deliver. Simply put, voters do not punish or reward mayors for the state of foreign affairs since mayors are not in charge of these particular policies. Then, if national governments (or, the EU) lose (conversely, gains) policy competences in a given policy area, voters should put more emphasis (conversely, less) on this issue in their calculus of voting. Empirically, studies confirm this zero-sum game pattern: economic-voting decreases in more opened economies (2014), while non-economic issues weight increases mechanically (Hellwig 2008). In addition, other studies also suggest that multi-level governance also decreases the extent of economic-voting (Anderson 2006). All in all, it seems that both issue and performance voting is changing with the evolution of the distribution of policy competences between and within levels of governments. In this paper, I investigate the impact of perceptions of room-to-manoeuvre constraints (hereafter RMCs) and of EU empowerment on EU issue voting, i.e. the extent to which voters take political parties stances on European integration when voting. Indeed, an increasing bulk of the literature demonstrates that EU-related issues affect voting behaviour in national electoral contests (De Vries 2007, 2010, De Vries and Tillman 2011). To explain EU issue voting heterogeneity at the individual level, the literature emphasizes on political sophistication, i.e. citizens awareness of the functioning of the EU, alongside with contextual factors such as information supply and party polarization (De Vries et al., 2011). This study takes a fresh look at this phenomenon by exploring the impact of attribution of responsibility on EU issue voting variation among citizens.

3 I focus on EU issue voting in national elections because it allows discriminating between two lines of arguments regarding attribution of responsibility. Indeed, the EU is whether a policy that has to be dealt with by national governments or it is a proper government responsible for delivering a multiple set of policies. In a multi-level system of governance setting, I first expect voters to link their ballot with the level of governance which they deem responsible of producing policies. Recently, studies have informed us that EU citizens are able to distinguish between EU policy domains and attribute responsibility accordingly (Hobolt and Tilley 2013); that EU popularity function is affected by the macroeconomic situation (Francois et al. 2012); and that they assign representational responsibilities separately to the national and European level of governance (Rohrschneider and Clark, 2008). In other words, EU citizens are able to properly distinguish between EU and national governments competences. The question here is whether this EU-based electoral behaviour is driven by the perception of European Union as being responsible for policy making. Hence, EU issue voting should be more pronounced among those voters who deem the EU to be in charge of policy making. Concurrently, I expect another kind of behaviour to happen: citizens who attribute less responsibility to the national governments should be more prone to vote on an EU-basis. The rationale here is different: voters who perceive national governments to be weakened by external institutions and/or processes should be more likely to care about other / non-traditional issues when voting (Hellwig 2008). This is in line with recent empirical findings in economic voting: in more open economies, citizens are less likely to punish or reward policy makers over the economy (Hellwig and Samuels, 2007, Hellwig 2008). In fact, globalization reduces voters perception of their governments leeway of manoeuvre which eventually weakens the connection between economic performance and incumbent performance. Following this argument, I hypothesize that the perception of government constraints may affect individual voting-behaviour. In turn, I expect those voters who deem their national governments to be constrained in their policy leeway in general to be more prone to vote on nontraditional issues such as the European integration. To explore the variation of EU issue voting at the individual level, I use a two-step hierarchical model. In the first stage of the modelling, I determine the extent to which EU congruence, i.e. the relationship between self-placement on the EU scale with the placement of political parties on the same scale, affects the propensity to vote for a party (PTV). I then estimate an ordered-logistic regression to assess the extent of EU issue voting. I do so because the dependent variable is ordinal (a 1-10 scale). In the second stage, I want to determine how EU issue voting is conditional upon different explanatory variables. Thus, I first compute residuals of the regressions which include EU congruence and I use the stored estimates as dependent variables. Stated differently, residuals thus provide a proxy-measure of EU issue voting. Then, I employ interaction effects for each of the important explanatory variable and compare their impact on the main independent variable, i.e. EU congruence. The estimations are

4 carried out thanks to the use of the 2009 European Election Study (EES) which is a cross-national survey of the 27 EU member states, with a representative sample of around 1,000 respondents in each country. It is an excellent data set since it provides us with all the individual-level variables needed, i.e. attribution of responsibility, identity-related factors, political sophistication and other socioeconomic variables. The results first show that attribution of responsibility to the EU has almost no effect on the variation of EU issue voting when controlled with other variables. Most notably, attribution of responsibility to the national government is systematically the best predictor when compared with every other individual variable: both in terms of significance and coefficient. Stated differently, the more the citizens attribute responsibility to the national government, the less they are likely to vote for a political party on the basis of EU related issues. The article proceeds as follows. I first take a look at the literature on EU issue voting and its determinants. Next, I explain hypotheses regarding attribution of responsibility and EU issue voting. In a fourth section, I discuss my modelling strategy. The two last sections aim at providing the results and discuss them. Determinants of EU Issue Voting at the Individual Level Since the mid-nineties and the referendums held in Denmark and France, European integration has become a more controversial issue within domestic politics of EU member states, with the notable rise of political formations resisting the EU process (Szczerbiak and Taggart 2008, Kriesi et al. 2008, Hooghe and Marks 2006, De Vries and Tillman 2011). According to the post-functionalist theory, this change has been made possible because the elite has had to make room for a more Eurosceptical public (Hooghe and Marks 2009: 9). The main assumption is that public opinion has become more polarized and more critical towards European integration because of increased visibility of the European integration process. The literature focusing on citizens attitudes towards European integration largely agrees upon the observation that support for EU integration has been shrinking in the last two decades, although indifference seems to be the most important opinion towards EU integration (Van Ingelgom 2012 ). Indeed, both mean and net support decreased since the mid-nineties and the Maastricht treaty (Dalton and Eichenberg 2007). As a result, Van der Eijk and Franklin show that European issues appear to be ripe for politicization and that it is only a matter of time before political entrepreneurs in some countries seize the opportunity, presented to their parties by these quite polarized opinions, to differentiate themselves from other parties in EU terms (Van Der Eijk and Franklin 2004: 47). They argue that it was mainly because of a lack of willingness to promote and to offer a variety of choice on European issues from the side of political elites that a sleeping giant of

5 public opinion has not yet altered or transformed the structure of domestic political dimension. Yet, EU attitudes matter to understand voting-behaviour, even in national competitions. Indeed, several researchers have shown that EU-related issues are able to influence national vote choice, noteworthily in countries such as Austria, Finland, Sweden, United Kingdom and Denmark (Gabel 2000, Tillman 2004, de Vries 2007, de Vries et al. 2011), others have pointed out the fact that there was evidence of a party-public cueing on EU-related issues between elites and public opinion whether assessing for a top-down or a bottom-up process or both (Carrubba 2001, Ray 2003, de Vries et al. 2007, Hooghe, 2007), and most significantly, de Vries and Tillman have shown that EU issue voting indicated by attitudes towards European integration which translate into national vote choices was consistent in both West European and Eastern and Central European countries (de Vries and Tillman 2011). Evidence of EU issue voting is now attested at the national level (de Vries 2007, de Vries and Tillman, 2011, de Vries et al. 2011, Hobolt et al. 2009), notably in the aftermath of referendums (Petithomme 2011), in contexts of accession, or when eurosceptic entrepreneurs voice their concerns about European integration in the public media (Van der Eijk and Franklin 2004). EU issue voting thus varies given contextual factors, i.e. the extent of salience (media attention) and the level of polarization among political parties (De Vries 2007, Hobolt, 2009, De Vries et al. 2011). First, the extent of media attention to an issue is decisive to understand electoral behaviour. Citizens tend to rely on the media agenda setting when deciding for which party to vote for (Della Vigna and Kaplan 2007) because of the difficulty for most of them to make the difference between parties stances on issues without mediators. Empirically, past studies confirm that media salience influences European attitudes (Semetko et al. 2003) and enhances EU issue voting (de Vries et al. 2011). Second, party conflict or polarization on an issue has also been depicted as an important variable to explain the extent of issue voting (Van der Eijk and Franklin 2004, de Vries 2007). The assumption is that the more political parties will be divided and, thus, will take clear (and strong) stances on an issue within the party system, the more voters will be able to distinguish between political offers because of the increasing clarity of parties positions. Furthermore, party conflict also influences the extent of issue voting because a larger political offer enables voters to voice their own political preferences more easily (Carmines and Stimson, 1986). There is thus greater evidence that European integration is able to influence electoral behaviour in national arenas. To sum up, past literature has intensively investigated dynamics of EU issue voting at the contextual level. But, we still know little about the variation in EU issue voting at the individual level in different contexts (de Vries et al., 2011). Yet, it is unlikely that Europe matters equally for all voters and within all contexts (Ibid: 18). At the individual level, studies have emphasized on political sophistication (de Vries et al. 2011). The assumption is that political sophistication is a crucial variable for citizens to intelligibly express their attitudes and opinions, notably on complex issues. This ability

6 to have a clear view of its own attitudes affects electoral behaviour because these opinions will act upon voting choice only if they are easily retrievable. Given the complex features of EU-related issues and the unclear division of power between supranational institutions and national governments, it is difficult for citizens to know political parties stances on European issues and, thus, to take this positioning into account when voting. Thus, voters who know the functioning of the EU are likely to vote more on an EU basis. Empirically, there is now evidence that the level of political sophistication matters to understand variation of EU issue voting among individuals (de Vries et al., 2011), but only in the context of European elections. In addition, many studies have demonstrated the important role of identity-related matters to apprehend attitudes towards European integration (Carey 2002, McLaren 2007, Marks and Hooghe 2005, Kriesi 2008) without taking this variable into account in an EU issue voting model. In a comparison between Germany, United-Kingdom and Spain, Diez-Medrano shows that nationality remains the most relevant factor to explain how individuals perceive the European integration process (Diez-Medrano, 2003). Moreover, Hooghe and Marks confirmed the importance of identity in a more recent comparative analysis of the determinants of support for the EU (Hooghe and Marks 2005). Their rationale is the following: European integration is perceived as a political institution contesting the authority of the nation-state while intensifying the opening up of the borders, thus it can be seen as a threat to the national community for a number of citizens. Other studies explain that group-serving biases are decisive variables to explain individuals political behaviour. It is argued that the in-group/out-group division, which is rooted in people s conception of their own (national) identity (Ibid: 7), is paramount to understand how citizens build their judgements on EU responsibility (Hobolt and Tilley 2014). Drawing on the partisanship literature, they show that EU enthusiasts are more likely to attribute responsibility to the EU when conditions are improving, whereas Eurosceptics tend to deny the EU any responsibility for positive outcomes (Ibid: 16). So, identity-related matters are important variables to explain the relationship between individuals and the EU (Kriesi et al. 2008). This study takes a fresh look at determinants of EU issue voting at the individual level by focusing on attribution of responsibility. Next part builds up the theoretical framework from which I derive hypotheses. Attribution of Responsibility and EU Issue Voting The central objective of this study aims to assess the extent to which EU issue voting is conditional upon attribution of responsibility. In this article, I start from the assumption that attribution of responsibility matters when explaining EU issue voting heterogeneity among individuals. Not only does a citizen need a high level of political sophistication so as to take EU issues into account when voting (de Vries et al., 2011), but, in a multi-layered system where responsibilities are shared voting

7 behaviour should be conditional upon citizens perception of their rulers ability to influence the policy making process. Attributing responsibility to the right level of governance in a multi-level system where competences are shared is not an easy task. Empirical evidence suggests that both federalism and decentralization processes hamper citizens capacity to differentiate between policy competences (Cutler 2009, Leon 2011). Indeed, a bulk of studies shows that clarity of responsibility is blurred in MLG settings such as Spain (Leon 2011, 2012), Canada (Cutler 2004, 2008) or the US (Brown 2010). In this setting, citizens face a daunting task when assigning correctly policy responsibility at each level of government. The fragmentation of competences diffuses the power and makes it more costly for citizens to gather information in order to hold the accurate level of governance responsible for policy outcomes. Furthermore, it provides political parties with incentives to engage in blame-shifting and credit taking which eventually lower individual perceptions of responsibility (Hobolt and Tilley, 2014). Finally, the mere fact that these systems provide citizens with multiple elections might increase the voter fatigue (Downs, 1999). In a nutshell, it is assumed that multilevel systems can undermine clear responsibility linkages for policy choices. Past studies on electoral accountability support these findings since it is commonly assumed that low clarity political systems alter voters ability to punish and sanction their governments on the basis of their economic record (Powell and Whitten 1993; Whitten and Palmer 1999). In an empirical study, Anderson (2006) demonstrated that economic voting is weaker in countries where multilevel governance is most prominent. Yet, other studies find opposite empirical evidence. Stein (1990) confirms that voters do not act similarly in elections at different levels: US citizens choose senators on the basis of the national economic performance, while governors are not punished for the state of the national economy (except when they are incumbent Republicans), suggesting that individuals are aware of the functional responsibilities that differentiate national and state-local government and act accordingly when casting their ballots. These findings are corroborated by Atkeson and Parkin (1995). The authors find that voters deem governors accountable for perceived state economic conditions, while they do not hold Senators accountable for state economy. Rather, they suggest that the former are blamed or rewarded for the state of the national economy. This implies that voting behaviour in state elections is linked to issues pertaining to governors. Vote functions are thus conditioned by the structural distribution of policy competences among governments in federal systems. In addition, Arceneaux shows that US citizens make accurate distinctions among policy competences at different levels of government, i.e. local, state and federal (2006). This study corroborates findings about attribution of responsibility within the EU context. Indeed, EU citizens seem to be able to get it right when attributing responsibilities to the EU even though policy competences are more intertwined than in the US context. Comparing experts surveys with citizens surveys, Hobolt and Tilley (2014) find that the

8 average person does not significantly deviate from his fellow expert when attributing responsibility to the EU. Moreover, they find that EU citizens outside the Eurozone attribute less responsibility to the EU for the economic situation in comparison with citizens inside a country belonging to the European Monetary Union (EMU), suggesting that the institutional structure in which citizens operate is taken into account when assigning responsibility. Moreover, evidence shows that citizens attribute responsibility to the EU for the macroeconomic situation since attitudes towards the EU are affected by the economic conditions only after EU accession (Francois et al. 2012). Finally, EU citizens are able to assign representational responsibilities separately to the national and European level of governance (Rohrschneider and Clark 2008). Some evidence thus supports the idea that citizens assign responsibility to the EU appropriately, at least to some extent. In turn, I expect that EU issue voting might be affected by responsibility attributions. After all, if a voter acts rationally and considers the EU level to have some impact on the decision-making process, her voting decision, even at the national level, will be conditional on the extent to which she attributes responsibility to the EU. Thus, I first expect that, all things being equal, voters who attribute more responsibility to the EU will take EU considerations more into account when casting their ballots. EU Responsibility Hypothesis (H1): EU issue voting is expected to be larger among citizens who deem the EU responsible for policy-making Concurrently, I expect another kind of behaviour to happen: citizens who attribute less responsibility to the national governments should be more prone to vote on an EU-basis. The rationale here is different: voters who perceive national governments to be weakened by external institutions and/or processes are more likely to care about other / non-national issues when voting. This is in line with recent empirical findings in economic voting: globalization reduces voters perception of their governments leeway of manoeuvre. In turn, it weakens the connection between economic performance and incumbent performance. For this reason, citizens are less likely to punish or reward policy-makers over the economy (Hellwig and Samuels 2007). Governmental room-to-manoeuvre constraints (hereafter RMCs) in the economic field reduce the weight assigned to economic performance evaluations and to party positions on economic issues (Hellwig 2008). Moreover, voters who perceive the most fiercely a loss of policy competences on the side of their governments might be prone to vote for parties on the basis of other issues than those that were typically held by governments. Thus, I hypothesize that the perception of national governments RMCs affects individual voting-behaviour: voters will ultimately care more about non-traditional issues such as EUrelated ones if they perceive more fiercely this governments RMCs. Thus, I expect EU issue voting to be conditional on attribution of responsibility to the national government. More precisely, I expect that, all things being equal, voters who attribute less responsibility to their national governments will take EU considerations more into account when casting their ballots:

9 RMCs Hypothesis (H2): EU issue voting is expected to be stronger among voters who attribute less responsibility to the national governments. The next section aims to test these two hypotheses and presents the empirical strategy which intends to respond to this challenge. Empirical Analysis Before presenting the empirical strategy, I first present the data I use to test my hypotheses. Data The individual-level data are derived from the 2009 European Election Study (EES). This is a crossnational survey which gathers information on electoral behaviour and political attitudes from the citizens of the 27 EU member states, with a representative sample of around 1,000 respondents in each country. This survey has the advantage of providing us with individuals propensity to vote for a given party (PTVs or electoral utilities), voters self-placement on a European integration scale and their perceptions of parties EU positioning. Moreover, it also asks respondents how they assign responsibility to both the EU and the national government for a particular policy domain, with five policy areas in total, i.e. economy, health care, immigration, interest rates and climate change. The EES 2009 also provides modules which allow us to measure political sophistication, identity-related matters, government approval, subjective assessments of the economic situation of the country and other socio-economic controls such as age, age when leaving school and gender. Empirical modelling: two-step hierarchical model To explore the variation of EU issue voting at the individual level and how it is conditional upon attribution of responsibility, I use a two-step hierarchical model. In the first stage of the modelling, I determine the extent to which EU congruence, i.e. the relationship between self-placement on the EU scale with the placement of political parties on the same scale, affects the propensity to vote for a party (PTV). The rationale is the following: issue-voting is a matter of distance between voters positioning and parties positioning on a given issue. In this setting, the closer a party A will be to the voter i on the issue of European integration, the more likely she will be to vote for party A. Thus, I expect EU congruence to be a decisive factor explaining the propensity to vote for a party. The advantages of PTVs over reported vote choice questions are numerous (for an

10 extensive review, see van der Eijk et al. 2006). First, unlike discrete-choice models, voters of small parties are not excluded from the statistical estimation in survey-based PTVs. Indeed, these surveys directly measure electoral utilities without inferring them like Conditional-Logit (CL) or Multinomial- Logit (MLN) models. Yet, the elimination of respondents who favour small parties can considerably alter coefficients of determination and the ordering of independent variables within estimations. More importantly, it is particularly problematical when one wants to analyse coefficients of voter-party distances on scales such as European integration or left-right positioning since their values can noticeably shift in both directions (Ibid: 438). To sum up, less information is lost with empirical observed electoral utilities and biases for estimated coefficients are considerably reduced with this technique. I then estimate an ordered-logistic regression because the dependent variable is ordinal and categorical: Propensity to vote for a party on a 1-10 scale. To perform regressions such as Ordinary Least Squares (OLS) or ordered logit with PTVs, a stacked data matrix derived from the original matrix needs to be created. This is due to the fact that the survey produces a number of variables for each individual: one for each party mentioned in the survey. Then, the unit of analysis becomes respondents*party scores on the transformed stacked data. So, the dependent variable is the observed PTV for voters and it displays relationships between respondents and political parties rather than respondents characteristics. It eventually allows to take into account both voters and political parties characteristics in a dynamic fashion. In the regression, the two main explanatory variables are EU congruence and Left-right congruence. EU congruence is constructed as the relationship between self-placement on the European integration scale with the placement of political parties on the same scale, where 0 indicates unification has already gone too far and 10 means it should be pushed further. Left-right congruence is constructed in the exact same fashion where left indicates 0 and 10 means right. Note that variables are recoded in positive terms and 0 thus indicates perfect match between party positioning and voter positioning on EU and left-right matters. A couple of other control variables are added in the regression. First, I add indicators of political performance, i.e. government approval, perception of democratic functioning and perception of economic performance. Political performance variables are added because they are common explanatory variables of voting-behaviour. The literature on economic voting particularly notably emphasizes on economic indicators, should they be sociotropic or egotropic, to explain why incumbents are punished or rewarded during elections (Lewis-Beck and Stegmaier 2013). Second, I add three items to control the effect of traditional cleavages, i.e. religious attendance, self-perception of social status and area of living. I add these historical cleavages because they have long been used as the main theoretical framework to approach voting choice in Western democracies (Lipset and Rokkan 1967). Finally, I also add socio-economic controls such as age, age when leaving school and gender. I have no particular expectations regarding these variables. Note that I also add fixed effects in the

11 ordered logit regression to control for variation between countries because of the non-randomness nature of the data. At this stage, I expect EU congruence to be significant and to display a negative sign. Indeed, the independent variables indicate the distance between party and voter positioning. In addition, I expect the effect of EU congruence to be lower than the effect left-right congruence because of the importance of this political dimension on voting choice (de Vries and Tillman 2011). In the second stage, I want to determine whether EU issue voting is conditional upon different explanatory variables. More precisely, I want to know whether EU issue voting is determined by the perception that EU is responsible for policy-making as H1 predicts, or whether perceptions of RMCs on national governments is more determinant as H2 predicts. To do so, I first compute residuals of the regressions which include EU congruence and I use the stored estimates as dependent variables. More precisely, I store six regressions residuals in total: the first regression only estimates the impact of EU congruence on PTVs; the second estimates the impact of EU congruence on PTVs while taking into account country fixed-effects; the third estimates the impact of EU congruence and control variables on PTVs; the fourth estimates the impact of EU congruence and control variables on PTVs while taking into account country fixed-effects; the fifth estimates the impact of EU congruence, Left-Right congruence and control variables on PTVs; the final regression estimates the impact of EU congruence, Left-Right congruence and control variables on PTVs while taking into account country fixed-effects. Then, the residuals are all computed so that their sign becomes positive. In turn, residuals are thus the positive sum of the distance between the outliers. Ultimately, it gives a proxy measure indicating the extent to which the EU issue voting model does not perform. Next, I estimate OLS regressions on these six dependent variables including the main independent variables of interest, i.e. perceptions of EU and national governments responsibility. I do so because the dependent variables are residuals which cannot be estimated via other estimations. First, to test my hypotheses, I need a measure of perceptions of RMCs and perceptions of EU responsibility. To do so, I use a module of questions on attribution of responsibility which is available as part of the 2009 European Election Studies. In this survey, there are questions asking respondents to which extent they assign responsibility to both the EU and the national government for a particular policy domain, with five policy areas in total, i.e. health care, economy, environment, interest rates and immigration, where 0 indicates no responsibility and 10 indicates full responsibility. To construct my two main independent variables, i.e. total of attribution of responsibility to the EU and total of attribution of responsibility to the national government, I compute the five questions together. Then, I have a proxy measure allowing me to estimate attribution of responsibility for the EU and national governments at the individual level.

12 In the regression, I also add other variables which can affect the variation of EU issue voting at the individual level. The literature on the subject emphasizes mainly on political sophistication (de Vries et al. 2011). To operationalize this variable, I use the same political knowledge scale as in de Vries et al. (2011). The scale is an addition of good answers on four different questions: whether Switzerland is a member of the EU; the number of EU member states; whether all countries elect the same number of members to the EP; and finally the number of parliamentary seats in the national parliament. Then, I change the variable into a binary one where 0 equals/or is lower than 3 and 1 indicates 4, suggesting that politically sophisticated are the individuals who score perfectly at the four questions. Moreover, I add a variable which aims to measure identity-related matters. Recently, some studies have argued that identity is critical in shaping contestation in Europe (Kriesi et al. 2008). A new cleavage separating people who benefit from the opening up of the borders and those who don t would have increased identity-related matters, notably regarding sovereignty. Moreover, studies on partisanship argue that individuals who identify the most with the European project to attribute more responsibility to the EU (Hobolt and Tilley 2013). Therefore, I assume that identity should affect EU issue voting at the individual level. More precisely, I expect those who feel more European to vote more on EU issues. To construct this variable, I take advantage of the question regarding respondents feeling of identity which is divided in four categories: only [nationality], [nationality] and European, European and [nationality], only European. I then change the variable into a binary one where 0 indicates the two first dimensions while 1 indicates the two last dimensions. Three other types of control variables are added in the OLS regressions. First, I add indicators of political performance, i.e. government approval, perception of democratic functioning and perception of economic performance. Second, I add three items to control the impact of traditional cleavages, i.e. religious attendance, self-perception of social status and area of living. Note that I also add fixed effects in the OLS regressions to control for a variation between countries. Finally, I also add socioeconomic controls such as age, age when leaving school and gender. The next section is dedicated to the presentation of the empirical findings. Empirical results The main independent variables measure the extent to which individuals attribute responsibility for policy-making to the EU and to the national governments. First, it must be noted that there is a strong correlation between the two variables since Spearman's coefficient is higher than 0.5. The mean of the total attribution of responsibility to the EU is around 28 out of 50 while the mean of the attribution of responsibility is around 35. Thus, individuals tend to attribute more responsibility, in general, to their national governments compared with the EU.

13 The data structure is stacked, so instead of having around 27,000 cases, the dataset gathers 406,035 observations. Consequently, the matrix also creates a high number of missing values. Thus, the structure of the data deeply alters the quality of the R-squared. The first ordered logit regressions performed to assess the impact of EU congruence on the Propensity to vote for a party indicates that the R-squared is indeed very low, i.e. around Nevertheless, the expectation that EU congruence affects the choice of a party is confirmed since it is the only significant explanatory variable alongside education and Left-right congruence. As expected, the sign is negative: the more the distance between parties stances and voter s positioning on EU issues, the less voters are likely to vote. Also note that the coefficient is clearly smaller than the left-right coefficient, indicating the predominance of the leftright dimension when voting (de Vries and Tillman 2011). Table 1 presents the results of the ordered logit regression in more details. Table 1 Determinants of propensity to vote for a party, with fixed effects (ordered logit regressions) Independent Variables Propensity to Vote for a Party EU congruence *** Left-right congruence *** Retrsopective evaluation of the economy Prospective evaluation of the economy Partisanhip Government disapproval State of democracy in the country State of democracy in the EU Interest in politics Religious attendance Gender *** Education *** Subjective class perception * Area of living R-squared 0.08 Observations 59,946 *Statistically significant at.1, two-tail. **Statistically significant at.05, two-tail. *** Statistically significant at.01 two-tail. Source: EES (2009) In the second stage, I want to determine whether attribution of responsibility to the EU and to the national government explains the variation of EU issue voting while taking into account other explanatory variables, i.e. political sophistication and identity-related matters. I have two different expectations regarding attribution of responsibility to the EU and to the national governments. In the case of attribution of responsibility to the EU, the results should be significant and should display a negative term. The sign should be negative because the dependent variable, i.e. residuals, is the distance between the outliers. In other words, the more the distance between the outliers, the more it

14 predicts that EU issue voting does not perform. Thus, residuals should be negatively impacted by the degree to which individuals attribute responsibility to the EU since my expectation is that attribution of responsibility to the EU has a positive impact on the effect of EU congruence on PTVs, i.e. EU issue voting. In the case of attribution of responsibility to the national government, the results should also be significant but I expect that the sign will be positive. The residuals should be positively impacted by the degree to which individuals attribute responsibility to the national government since my expectation is that attribution of responsibility to the EU has a negative impact on EU issue voting. First, I want to see the overall impact of each variable on all the OLS regressions. Table 2 and 3 present the results of OLS regressions on the six stored residuals. In these regressions, I only retain four variables of interest, namely attribution of responsibility, political sophistication and identityrelated matters, for a question of readability. Table 2 Determinants of EU issue voting with residuals (OLS regressions) Residuals EU Residuals EU & Residuals EU & Independent Variables Fixed Effects ctrl variables Responsibility attribution EU Responibility attribution National Govts *** *** *** Political sophistication *** *** *** EU Identity ** ** ** Constant *** *** *** R-squared Observations 29,403 29,403 29,403 *Statistically significant at.1, two-tail. **Statistically significant at.05, two-tail. *** Statistically significant at.01 two-tail. Source: EES (2009) Table 3 Determinants of EU issue voting with residuals (OLS regressions) Independent Variables Residuals EU & Fixed Effects & ctrl variables Residuals EU & Left- Right & ctrl variables Residuals EU & Left- Right & ctrl variables & Fixed Effects Resp. EU Resp. Nat. Govts *** *** *** Political sophistication *** *** *** EU identity ** *** *** Constant *** *** *** Observations 29,403 28,359 28,359 R-squared *Statistically significant at.1, two-tail. **Statistically significant at.05, two-tail. *** Statistically significant at.01 two-tail. Source: EES (2009)

15 The first striking result is that attribution of responsibility to the EU never displays any significant estimate on the residuals in any OLS regressions. When compared with every other determinants of EU issue voting, the fact of perceiving the EU to be responsible for policy making does not affect EU issue voting at the individual level. Yet, signs are in the expected direction. Indeed, in most of the regressions (five out of six), residuals are negatively correlated with the level of attribution of responsibility to the EU. Thus, these non-significant results indicate that individuals who attribute more responsibility to the EU are assumed to vote more on an EU basis. All in all, H1 is, nevertheless, rejected by the model. The second important result is that attribution of responsibility to the national government is always significant and displays a positive sign as expected, indicating that residuals, i.e. non EU issue voting, are explained by the fact of a low attribution of responsibility to the national government, i.e. RMCs. Hence, the more voters perceive their national governments to be constrained in their leeway, the more they will vote on EU issues. Furthermore, attribution of responsibility to the national government is almost systematically the best predictor in terms of coefficient compared with the other variables. As tables 2 and 3 show, the Beta coefficient of attribution of responsibility varies from 0.03 to In comparison, political sophistication which is the second best predictor is almost always less significant while its Beta coefficient revolves around to The second hypothesis is thus better supported by the model. Furthermore, both European identification and EU political sophistication are significant and have the expected signs. First, residuals are negatively correlated with the level of EU political sophistication. Stated differently, individuals who know the EU better tend to vote more on the basis of EU issues. This result has already been found in European elections (de Vries et al. 2011), but not in national elections. Second, and most importantly, these results show that EU identity-related matters play a role in understanding EU issue voting. Indeed, it is the second most important explanatory variable in the six models. More precisely, results show that results are systematically negatively correlated with EU identification. Stated differently, the more a voter identifies with the EU, the more he is likely to vote for a party based on EU issues. This relation has never been tested, but other studies point to the same direction (for example, see Hooghe and Marks 2005, Kriesi et al. 2008) Second, I do not include attribution of responsibility to the government to measure the effect of attribution of responsibility to the EU alone to avoid problems of collinearly. Table 4 presents the results of the OLS regressions on two of the six stored residuals. In these regressions, I only retain three variables of interest, namely attribution of responsibility to the EU, political sophistication and EU identification. Here, the empirical results show that attribution of responsibility to the EU is significant without attribution of responsibility to the national government. Nevertheless, the degree of

16 signification is systematically less important than the one displayed by attribution of responsibility to the national government and even to the one displayed by political sophistication. Moreover, the Beta coefficient is also clearly weaker and explains around Most importantly, the sign is reversed: it is as if the more one attributes responsibility to the EU, the less she is prone to vote on an EU basis. This finding is at odds with H1 and further contributes to its rejection. Table 4 Determinants of EU issue voting with residuals & Attribution of Responsibility to the EU only (OLS regressions) Independent Variables Residuals EU Residuals EU & Fixed effects Responsibility attribution EU *** ** Political sophistication *** *** EU identity ** ** Constant *** *** Observations 29,692 29,692 R-squared *Statistically significant at.1, two-tail. **Statistically significant at.05, two-tail. *** Statistically significant at.01 two-tail. Source: EES (2009) I continue this robustness check by estimating regressions without attribution of responsibility to the EU in order to measure the effect of attribution of responsibility to the government independently. Table 5 presents the results of the OLS regressions on two of the six stored residuals, i.e. the ones with only EU congruence and EU congruence and fixed-effects. In these regressions, I only retain three variables of interest, i.e. attribution of responsibility to the government, political sophistication and identity-related matters. The estimations clearly confirm the first results. As expected, attribution of responsibility to the national government is always significant and displays a positive sign. In addition, the Beta coefficients are still the most important ones in comparison with the other variables, that is around and These findings lend some support to the second hypothesis, i.e. EU issue voting increases when individuals attribute less responsibility to their government, while disproving the first one.

17 Table 5 Determinants of EU issue voting with residuals & Attribution of Responsibility to the national governments only (OLS regressions) Residuals EU Residuals EU & Fixed Independent Variables effects Political sophistication *** *** EU identity ** ** Responsibility attribution Nat *** ( ) *** ( ) Constant *** *** Observations 30,242 30,242 R-squared *Statistically significant at.1, two-tail. **Statistically significant at.05, two-tail. *** Statistically significant at.01 two-tail. Source: EES (2009) Conclusion This paper investigates the impact of globalization and multi-level governance on voting behaviour. More precisely, it sheds light on how the evolution of distribution of policy competences affects issue voting in national elections. So far, we know that economic issue-voting evolves along the evolution of national governments economic competences: voters adapt to the loss of national governments room-to-manoeuvre in the economic area by reducing the importance of economic issues in their calculus of voting (Hellwig and Samuels 2007, Hellwig 2008). In the meantime, other studies show that economic voting is weaker in multi-level systems of governance (Anderson 2006), eventually suggesting that diluting economic competences of national governments drive voters to judge their national governments on economic issues. In this paper, I take EU issue voting as an example to discriminate between these two lines of arguments. In a multi-level system of governance setting, voters should link their ballot with the level of governance which they deem responsible of producing policies. In this scenario, EU issue voting should thus be positively influenced by the extent to which individuals attribute responsibility to the EU. Conversely, voters might perceive national governments to be increasingly constrained in their leeway to produce policies due to both globalization and European integration. In that case, voters are assumed to vote more on non-traditional issues (Hellwig 2008). Thus, I argue that those voters should be more prone to choose a party based on EU issues. Results show that perceptions of RMCs are paramount to explain EU issue voting heterogeneity at the individual level. Indeed, voters who think that national governments have low responsibility in terms of policy making tend to be the ones who take EU issues more into account in their calculus of voting. This finding is coherent with previous studies showing that the opening of world markets reduces the

18 importance of performance and issue-voting in the economic area (Hellwig and Samuels 2007, Hellwig 2008, 2014). On the contrary, voters who perceive the EU to be responsible for policies do not choose political parties on EU issues. This result is puzzling since responsibility attribution for governments is at the core of processes of accountability. An explanation might be that citizens do not perceive the EU as a ruler. Rather, it might be that EU is merely seen as an issue that has to be dealt nationally as the liberal integovernmentalist theory suggests (Moravscik 1993, 2002) or as it is the case in European elections (Hix and Marsh 2007) Furthermore, results confirm that political sophistication influences issue voting: the more the voter knows about an issue, the more she can mobilize her opinions on this issue when voting. In the case of EU issues, we know that EU political sophistication improves EU issue voting in European Elections. In this paper, I show that EU political sophistication also enhances EU issue voting in national contests at the individual level. More importantly, I demonstrate that identity-related matters affect voting behaviour regarding the EU, as it has been already suggested in other studies (Hooghe and Marks 2005, Kreisi 2008). Indeed, the more voters identify to the EU, the more they tend to vote for political parties based on their proximity on EU issues. This is as if there were some sort of EU partisanship which would enhance the extent to which citizens take into account EU issues in their calculus of voting as in the case of determinants of attribution of responsibility to the EU (Hobolt and Tilley 2013). Bibliography Anderson, C.D. (2006) Economic voting and multilevel governance: a comparative individual level analysis, American Journal of Political Science 50: 2, pp Arceneaux, K. (2006) The Federal Face of Voting: Are Elected Officials Held Accountable for the Functions Relevant to Their Office?, Political Psychology 27: 5, pp Atkeson, L.R. and Partin, R.W. (1995) Economic and referendum voting: A comparison of gubernatorial and senatorial elections, American Political Science Review 89:1, pp Carey, S. (2002) Undivided Loyalties. Is National Identity an Obstacle to European Integration?, European Union Politics 3:4, pp Carrubba, C.J. (2001) The electoral connection in European Union politics, The Journal of Politics 63:1, pp Cutler, F. (2004) Government Responsibility and Electoral Accountability in Federations, Publius 34:2, pp Cutler, F. (2008) Whodunnit? Voters and Responsibility in Canadian Federalism, Canadian Journal of Political Science 41:3, pp DellaVigna, S. and Kaplan, E. (2007) The Fox News Effect: Media Bias and Voting, The Quarterly Journal of Economics 122:3, pp