Motivating the European Voter: Parties, Issues, and Campaigns in European Parliament Elections

Transcription

1 Motivating the European Voter: Parties, Issues, and Campaigns in European Parliament Elections SARA B. HOBOLT University of Oxford Department of Politics and International Relations Manor Road, Oxford, OX1 3UQ United Kingdom JAE-JAE SPOON University of Iowa Department of Political Science 341 Schaeffer Hall Iowa City, IA United States Paper prepared for presentation at the PIREDEU Final Conference, Brussels, Belgium, November 2010 * We would like to thank Catherine de Vries, Marc Hooghe, and Michael Malecki for their helpful comments and suggestions and Traci Wilson for her research assistance.

2 Abstract Voters behave differently in European Parliament (EP) elections compared to national elections because less is at stake in these 'second-order' elections. While this explains the primary characteristic of EP elections, it has often led to a conflation of three distinct motivations for changing behavior, namely sincere, strategic protest and arena-specific voting. By distinguishing these motivations, we address the question of when and why voters are more likely to rely on each consideration. We argue that the mediating factors are located at two levels. At the individual-level, voters behavior depends on which type of party they normally support. At the context-level, the degree of politicization of the EU in the domestic debate shapes the amount of arena-specific voting. We test these propositions in a multi-level analysis in 27 countries in the 2009 EP elections. Our findings have important implications for understanding why voters change their behavior between different types of elections.

3 1 Introduction In the European Parliament (EP) elections in June 2009, citizens across all 27 member states voted to elect representatives to this law-making body of the European Union (EU). The choice of parties was almost identical to that in national parliamentary elections, yet almost half of voters abandoned their national party choice by either switching to another party or abstaining. This resulted in many smaller parties doing better than they had in recent national elections, and governing parties generally performing worse. In Britain, for example, a party without national representation - the United Kingdom Independence Party (UKIP) - won 13 of 72 UK seats in the EP. This phenomenon raises the question of why voters change their behavior between different types of elections, such as national and EP elections? The most prominent explanation in the literature on EP elections is the so-called 'secondorder national election' theory, pioneered by Reif and Schmitt (1980), which has its roots in theories of midterm elections in the United States (Campbell 1960; Tufte 1975; Fiorina 1981). At the heart of this theory is the proposition that people vote differently in 'second-order' EP elections because less is at stake compared to 'first-order' elections for national office. Numerous empirical studies of EP elections over the past decades have lent support to this model of secondorder elections by showing that turnout is lower than in national elections, smaller parties perform better, and parties in national government are punished, particularly during the midterm (see e.g. Reif 1984; van der Eijk and Franklin 1996; Marsh 1998; Hix and Marsh 2007, 2010). These patterns of behavior are generally interpreted as voters responding to the low salience context of EP elections. Yet, few studies have unpacked the individual-level motivations that lead to different types of behavior. This paper seeks to do just this by developing a framework of voting behavior

4 2 in second-order elections. To understand the micro-foundations of vote choices in these elections, our model distinguishes between observed behavior and motivations. We identify two primary ways in which voters may alter their behavior between first- and second-order elections: by changing the party they vote for (switching) or by not voting at all (abstaining). These observed behaviors, we argue, are driven by three distinct types of motivations: sincere voting (changing behavior due to ideological differences), strategic protest voting (changing behavior due to dissatisfaction with government performance) and finally, arena-specific voting (changing behavior due to European concerns). While these motivations have been discussed in the extant literature on second-order elections, going back to Reif and Schmitt (1980), the first contribution of this paper is to explicitly distinguish among the different types of motivations. It is not sufficient, however, to simply identify that voters can be driven by different motivations. Importantly, we need to understand when and why voters are more likely to base their vote choice on sincere, strategic or arena-specific motivations. The second contribution of this paper is thus to present a framework for understanding the factors that mediate the importance of different motivations. We propose mediating factors at two levels. At the individual-level, we distinguish between voters according to which type of party they voted for in the national election, arguing that niche party voters are more likely to be sincere voters, whereas mainstream party voters are more likely to be strategic protest voters. At the contextual level, we argue that the politicization of the EU issue conditions the extent to which voters rely on arena-specific considerations in EP elections. Specifically, we posit that higher levels of party polarization on the EU issue and negative EU campaign coverage magnify the importance of EUspecific considerations. We test these propositions in a multi-level analysis of voting behavior in EP elections across 27 countries, using European Election Studies (EES) data and Media Study

5 3 data from the 2009 EP elections (EES 2009; van Egmond et al. 2010; Schuck et al. 2010). These data allow us to examine individual-level motivations and also explore how the specific campaign context conditions these motivations. The paper proceeds as follows. First, we briefly review the literature of second-order elections. Second, we present our theoretical framework of behaviors and motivations in secondorder elections, as well as the factors that mediate their importance. Third, we test the hypotheses derived from the theoretical model. Our findings show that sincere, strategic and arena-specific motivations all matter in EP elections, but that both individual voter-types and system-level politicization of the EU issue condition the degree to which voters are driven by specific motivations. The findings thus have important implications for understanding when and why voters change their behavior between different types of elections. Voting Behavior in European Parliament Elections There is plenty of evidence that Europeans vote differently in national and European elections. Notwithstanding the increasing importance of the European Parliament since the first direct elections in 1979, empirical studies have demonstrated three broad patterns of empirical regularities repeated in all seven EP elections (see e.g. Reif and Schmitt 1980; Marsh 1998; Hix and Marsh 2007, 2010). First, levels of turnout are lower than in national elections. Second, citizens favor smaller parties over larger parties compared to national elections. Third, parties in national governments do worse in EP elections than in national elections, especially when the EP elections take place during the midterm of the national election cycle. The classic explanation is that changing patterns of voting behavior between national and European elections are due to the fact that less is at stake in second-order elections, and as a

6 4 consequence, voters cast their votes not only as a result of conditions present within the specific context of the second-order arena, but also on the basis of factors in the main political arena of the nation (Reif and Schmitt 1980, 9; see also van der Eijk and Franklin 1996; Ferrara and Weishaupt 2004; Hix and Marsh 2007). Because less is at stake voters are expected to vote more sincerely in EP elections, compared with first order elections. Reif and Schmitt (1980, 10), however, recognize that some, who generally support the government, vote for the opposition in secondary elections in order to apply pressure to the government although not fundamentally changing their party allegiance. While the central intuitions of the second-order election model are consistent with the patterns of behavior that we observe in EP elections, recent studies have raised two interrelated concerns about the absence of an individual vote choice model and the use and interpretation of aggregate-level data. The first concern is primarily theoretical and focuses on the lack of an explicit individuallevel model of vote choice to explain the aggregate-level regularities (Bakker and Polk 2010; Hobolt, Spoon and Tilley 2009; Hobolt and Wittrock 2010; Marsh 2007; Marsh and Mikhaylov N.d.; Tiemann 2009). Recent contributions to the literature have gone some way towards developing models of individual vote choice. Hobolt, Spoon and Tilley (2009), for example, present a model of why some voters defect from governing parties in EP elections, which builds on the spatial voting literature. According to this model, defection is a function of distance to the governing party on the left-right and EU dimensions and non-policy motivations, conditioned by the campaign context. Similarly Tiemann (2009) draws on the 'unified' model of issue voting, incorporating both proximity and directional components, to develop a spatial model of vote choice in EP elections. Recent work has also made progress on categorizing different types of voters rather than treating all voters as in second-order elections as equal (Weber 2009; Bakker and Polk 2010). In our theoretical model below, we build on these recent innovations in the

7 5 literature to present a more comprehensive framework of second-order voting that incorporates both observed behaviors and motivations. The second concern about the classic second-order election literature relates to the interpretation of the aggregate-level regularities in EP election outcomes. Scholars have found that the same patterns are consistent with an alternative explanation, namely arena-specific (or 'Europe matters') motivations for voting (Marsh 2007; Hobolt, Spoon and Tilley 2009). According to this perspective, low turnout and defection from governing parties may not be entirely due to the general decline in government support at midterm, it may also be caused by dissatisfaction with the position of these parties on Europe. Hobolt, Spoon and Tilley (2009) demonstrate that voters are more likely to defect from a governing party if they have more Euroskeptic preferences than the party (see also Clark and Rohrschneider 2009). Similarly, de Vries et al. (2010) have shown that attitudes towards Europe are an important factor deciding individual-level vote choice in the 2009 EP elections, especially among those who are more politically sophisticated and when media attention is greater. In a systematic comparison of the second-order and the Europe matters theses, Hix and Marsh (2007), using aggregate-level data from six EP elections, show that large parties tend to lose votes in EP elections regardless of their position on European integration. However, they also find some electoral gains for anti-eu parties and parties that emphasize the European issue (see also Hix and Marsh 2010). These alternative explanations of changes in behavior between first- and second-order elections raise the methodological problem of observational equivalence. That is, the differences in levels of support for parties in national and European elections may be due to the fact that voters base their choices on preferences relating to the performance of governing parties in the first-order (national) arena, but it may be equally a result of voters changing behaviors due to arena-specific (European) issues.

8 6 Importantly, this also highlights the need to more systematically distinguish among the types of individual-level motivations that drive changing behavior between first- and second-order elections. In the next section, we develop a theoretical framework that unpacks the different types of motivations and behaviors. Motivating the European Voter The classic second-order model implicitly identifies two separate, and not easily compatible, motivations for a voter to change behavior between first- and second-order elections: sincere voting and strategic protest voting. Whereas sincere voting implies that voters choose their preferred party, strategic voting suggests the very opposite, namely that voters rationally decide to vote for a party other than their overall favorite (McKelvey and Ordeshook 1972; Cox 1997). Accordingly, a single voter will not be acting sincerely and strategically in the same election. It is possible, of course, that some voters vote sincerely and others vote strategically in these elections, yet the second-order model provides little guidance as to when (or why) voters are likely to vote with their heart or when they use their head to strategically signal dissatisfaction with the government. As discussed above, more recent work on EP elections has also suggested a third motivation for changing patterns of behavior, namely arena-specific voting. According to this explanation of changes in behavior, attitudes relating to the European arena become more salient to voters in EP elections, and consequently voters may decide to switch their party or abstain because they disagree with the party they support in national elections on European issues (see Hix and Marsh 2007; Hobolt, Spoon and Tilley 2009; de Vries et al. 2010). Although the classic second-order literature has tended to emphasize that such considerations play a minimal role to

9 7 voters in second-order elections, even Reif and Schmitt (1980, 10) acknowledged that arenaspecific motivations are not entirely irrelevant. Hence, to understand what drives changing behavior among first- and second-order elections, our theoretical framework explicitly distinguishes among three types of motivations: sincere voting, strategic protest voting and arena-specific voting. Treating these motivations separately is important for two reasons. First, few voters will be motivated by all three motivations simultaneously. For example, it is unlikely although not impossible - that a voter will use the EP ballot to signal dissatisfaction with the government while also switching to a party that has more compatible views on both domestic and European issues. It is thus critical to understand which individual-level motivations are more important for driving changes in behavior between first- and second-order elections. Second, by delineating these types of motivations we can go one step further and examine when and why these motivations are likely to matter more. That is, what are the primary factors conditioning the importance of sincere, strategic and arenaspecific motivations for vote choice? [FIGURE 1 ABOUT HERE] Figure 1 outlines our theoretical framework. It shows the three primary types of motivations - sincere, strategic protest and arena-specific - that drive second-order electoral behavior. To distinguish among different types of behaviors, we build on recent studies of EP elections that have categorized types of voters. Weber (2009), for example, draws on Hirschman's (1979) Exit, Voice and Loyalty framework to construct a model of four types of second-order voters: those who exit from their party or from the system, those who express 'voice', and core voters. In another recent study, Bakker and Polk (2010) create a typology of eight types of voters based on whether their motivation is strategic or sincere and if their vote in the EP election is

10 8 consistent with their national election vote. Our model explicitly seeks to distinguish between observed behaviors and the motivations for these behaviors. We identify three broad types of behaviors in second-order elections: partisanship (voting for the same party as in the national elections), switching (switching to different between in the second-order election) and abstaining (voting in the national election, but abstaining in the EP election). Moreover, we propose two sets of mediating factors that condition the degree to which different motivations drive behavior, namely voters' party type at the individual level and the politicization of the EU at the country level. Before we discuss these mediating factors in greater detail, we first derive our hypotheses concerning the direct effect of motivations on behavior. As discussed above, the proposition of sincere voting lies at the heart of the second-order model. From a spatial voting perspective (Downs 1957; Enelow and Hinich 1984), sincere voters will vote for the candidate or party which is closest to them on a given issue. Conversely, when national office is not at stake, voters are less likely to support a party which is further away from them, and some may be more likely to abstain because of disagreements over policy positions. We capture sincere motivations in terms of distance to the parties on the dominant dimension of contestation in Europe, namely the left-right dimension (McDonald and Budge 2005; Mair 2008). This generates our hypothesis concerning sincere voting: H1: The larger the distance between voters and the position of their national party on the left-right dimension, the more likely they are to switch or abstain in EP elections. The second motivation identified in the second-order election literature is that some voters tend to use EP elections strategically to punish governments for poor performance. Importantly, this effect depends on the incumbent s performance, conditioned by when the European election falls in the national election cycle (Reif 1984; van der Eijk et al 1996; Marsh 1998; Kousser

11 9 2004). We would thus expect that the government s record will be more important to voters who are motivated by strategic protest considerations. If a voter is dissatisfied with the government's performance, she will be more likely to express her protest by either switching to another party or by abstaining. This leads to our strategic protest voting hypothesis: H2: The greater the dissatisfaction with government performance, the more likely voters are to switch or abstain in EP elections. The final motivation has been highlighted in more recent studies of EP elections, and suggests that changing patterns of voting between national and European elections are driven at least partly by concerns specific to the European arena (Hix and Marsh 2007; Marsh 2007; Hobolt, Spoon, and Tilley 2009; de Vries et al. 2010). Similar to the sincere voting hypothesis, voters are expected to support the party which adopts policy positions closest to their own on a given dimension. But the saliency of different policy dimensions does not necessarily stay constant across elections and arenas. Voters motivated by arena-specific considerations will place greater emphasis on the European integration issue - notably preferences concerning 'more' or 'less' integration (Hooghe et al. 2002) - in EP elections than in national elections, and they may therefore decide to switch parties or abstain if the party they support in national elections adopts a different position on the European integration dimension. 1 This brings us to our arena-specific voting hypothesis: 1 We recognize that European integration can also be considered a national issue in some contexts (and left-right issues are equally relevant at the European level). However, as we seek to unpack voters motivations in EP elections, we are interested in explaining under what conditions vote choice is determined by area-specific or European issues. Furthermore, while we acknowledge that operationalizing voters attitudes towards Europe as a unidimensional preference for more or less integration may be seen as an over-simplification, this question does appear to capture latent generic attitudes towards European integration, and the item is highly correlated with other questions focused on attitudes towards more or less EU involvement in specific types policy (see Gabel 1998).

12 10 H3: The larger the distance between voters and the position of their national party on the European integration dimension, the more likely they are to switch or abstain in EP elections. We now turn to the factors that mediate the importance of these motivations to secondorder vote choice. As shown in Figure 1, we identify mediating factors at two levels. At the individual-level, we argue that the type of party a voter supported in the previous national election will influence her motivations in the EP election. At the context-level, we posit that the politicization of the EU issue will condition the extent to which a voter relies on arena-specific considerations in EP elections. If a voter is a governing party supporter in national elections, it is reasonable to expect that she is more likely to use her ballot in second-order elections to signal protest or praise of the government's performance. In contrast, an opposition party supporter has few incentives to change her behavior simply because she is unhappy with the government s record, thus:: H4: Government satisfaction has a greater impact on the likelihood of switching or abstaining in EP elections for government party voters. We can categorize voters not only according to whether they voted for governing or opposition parties, but also depending on whether they support niche or mainstream parties in national elections. Niche parties have been identified as a group of parties that do not follow the traditional class-based orientation of politics, whose issues do not coincide with existing lines of political division, and that limit their party manifestos to focus on a narrow set of policies (Meguid 2005, 347-8). Meguid (2005, 2008) includes far right, green, and ethnoterritorial, or regional, parties in her categorization of niche parties, such the Hungarian Jobbik, French Greens,

13 11 and Scottish National Party. 2 As many of these parties formed in reaction to the larger parties non-response to new issue demands by certain cohorts of voters (Kitschelt 1988; Hug 2001), ideology is very important to them (see Adams et al 2006; Spoon 2009, 2011). Recent work on niche parties has demonstrated that in various situations, these parties behave differently because of their focus on ideology (see, for example, Heller 2000; Jensen and Spoon 2010). Although little work has been done on niche party voters as such, we know that they tend to be more sincere or expressive in their voting decisions. Expressive voters are often motivated by ideology (Schuessler 2000, 54). They may support a niche party, for example, because they identify themselves as party x voters and seek to attach themselves to a collective of similarly-minded individuals. Thus, just as the parties are more focused on policy distinctions, we also expect niche party voters to behave more sincerely because of their focus on ideology: H5: Distance to the national party on the left-right and EU dimensions has a greater impact on the likelihood of switching or abstaining in EP elections for niche party voters. Next, we turn to the way in which the political context mediates the importance of different motivations in EP elections. Specifically, we focus on how the politicization of the European issue conditions the extent of arena-specific voting. Recent work on political behavior in Europe has argued that the issue of European integration is becoming increasingly politicized as we are witnessing public contention over European matters in referendums, party competition and media reporting (de Vreese 2003; Tillman 2004; de Vries 2007; Kriesi et al. 2008; Hobolt 2009; Hooghe and Marks 2009). EU politicization refers to the increasing contentiousness of decision making in the process of European integration (Schmitter 1969). Hooghe and Marks 2 Other scholars' have differing definitions of niche parties. Adams et al. (2006, 513), for example, consider all parties which have either an "extreme ideology" or a "noncentrist niche ideology" and thus include communist parties in their typology of niche parties.

14 12 posit that this politicization has changed both the content and the process of decision making (2009, 8). Importantly, however, the level of politicization of European issues varies considerably across countries and specific electoral contests. We know from studies of vote choice in referendums and elections that arena-specific voting - so-called EU issue voting - is more pronounced when the European issue is politicized in the domestic sphere (Tillman 2004; de Vries 2007; Hobolt 2009). Equally, we would expect that in member states where the issue of European integration is politicized in the period leading up the EP elections, more voters will be motivated by arena-specific concerns. Two key indicators of the level of politicization are the degree of party polarization on the issue and the contentiousness of European integration in the campaign coverage. Party polarization is a central determinant of the politicization of political issues. When parties are divided on an issue, their positions become clearer, and voters are able to make real choices on the issue (Carmines and Stimson 1989). Greater party polarization on the European issue not only offers voters with greater choice on that dimension, but also increases the salience of European issues to voters, and thus makes it more likely that they will vote on this basis in EP elections (van der Eijk and Franklin 2004; de Vries 2007; Hobolt 2009). We thus expect arenaspecific voting to be more pronounced when parties issue positions regarding the EU diverge: H6: Distance to the national party on the EU dimension has a greater impact on the likelihood of switching or abstaining in EP elections when national parties are more polarized on the EU dimension. A second indicator of the level of politicization is the contentiousness of the EU issue in the campaign coverage. Since voters receive information and cues from the mass media, we know that the media has an important agenda-setting role in shaping voting behavior (McCombs

15 13 and Shaw 1972; Popkin 1991). There is also ample evidence that the tone of the media coverage can affect voters perceptions, preferences, and decision to vote (see, for example, Norris et al 1999; Wattenberg and Brians 1999; Druckman and Parkin 2005). In the context of EP elections, we expect that the more critical the media coverage of the European integration process is during the campaign, the more voters will be primed to consider arena-specific concerns when casting their ballot. In other words, we expect that disagreements between voters and parties over EU issues will play a greater role in voters second-order vote choices when the problems associated with European integration are highlighted in the media during the campaign: H7: Distance to the national party on the EU dimension has a greater impact on the likelihood of switching or abstaining in EP elections when EU campaign coverage is negative. In the following section, we discuss how we evaluate these propositions empirically. Data and Methods To test our models of voting behavior in second-order elections, we use the 2009 European Election Study (EES 2009; van Egmond et al. 2009) and Media Study (Schuck et al. 2010). The voter study was fielded during the four weeks immediately following the EP elections of June 2009, with independently-drawn samples of over 1,000 respondents in each of the EU s 27 member states. The Media Study contains content analysis of news stories from the three weeks before the elections from the main national evening news broadcast, two broadsheets, and one tabloid newspaper in each of the 27 member states. A total of 52,009 television and newspaper stories were coded.

16 14 Our dependent variable is whether a voter voted for the same party as she did in the national election, switched to a different party, or abstained in the 2009 EP elections, based on question items on EP and national vote choice. These three categories of voter behavior are partisans, switchers, and abstainers, respectively (see Figure 1). Across the 28 political contexts, 3 there is a great deal of variance in these three categories. The mean percentage of partisans is 54.1%, ranging from under 30% in Latvia and the United Kingdom to over 70% in Cyprus, Belgium (Flanders), Italy, and Malta. Among the 20.6% of respondents who switched their vote, there is also a large range across countries, from under 10% in Cyprus, the Czech Republic, Malta, and Slovakia to over 35% in Ireland and Latvia. Finally, 25.4% of voters abstained, varying from under 10% in two of the countries with enforced compulsory voting (Belgium and Luxembourg) to over 40% in Lithuania, Poland, and Slovakia. Table 1A in the Appendix reports the values of the dependent variable by country. Our individual-level models estimate the impact of the three motivations on these behaviors. To assess the extent of sincere and arena-specific voting we include two variables capturing the distance between a voter s position and that of the party she voted for in the last national election: one for the left-right dimension and one for the European integration dimension. We calculated these variables by subtracting the respondent s self-placement on the left-right and EU dimensions from the mean placement of her party, both of which are 11 point scales, using absolute distances. To test the degree of strategic protest voting, we use a dichotomous government approval variable. We also include a series of control variables. As we are interested in what motivates abstaining or switching, we employ a dichotomous partisanship variable. Following the extant literature (Campbell 1960; Verba Nie and Kim 1987; Lewis-Beck et al 3 As Belgium s two regions of Flanders and Wallonia have two separate party systems, we consider them as two different political contexts. Thus, we have 27 countries and 28 contexts in our dataset.

17 ), we expect that those who are more partisan are less likely to abstain or switch. A voter s level of political awareness is also likely to affect behavior, by making her less likely to abstain and more likely to switch in EP elections, and we therefore created an additive seven-point scale based on questions asked about both national and European political knowledge (Cronbach s alpha=0.74). We also include three demographic control variables. Education is a five-category variable based on the age when the respondent stopped full time education. Gender is a dichotomous variable, where 0 is male and 1 is female. Finally, our measure of age is a continuous variable. Table 2A in the Appendix lists the questions used and their exact wording. To test our hypotheses concerning the factors mediating the importance of motivations, we include additional individual- and context-level variables. At the individual level, we use two variables that describe the type of party the voter supported in the national election. To measure whether a voter supported a government party or niche party, we include two dichotomous variables if the voter voted for a government party or niche party. 4 Importantly, these two categories are not mutually exclusive. A voter who supported the Irish Green Party in the 2007 election, for example, would be both a government and niche party voter, as the Greens entered into a governing coalition with Fianna Fáil. We also control for party size, measured as the seat share of the voter s party in the last national election. 5 At the contextual level, we include two variables in our model that measure EU politicization across the different political contexts: the polarization of the party system on the issue of EU integration and the tone of the campaign. EU polarization is measured as the standard 4 Following Meguid s (2005, 2008) typology, we coded a party as a niche party if it is green, radical right, or ethnoterritorial/regional. We also estimated our models with separate dummies for green and radical right parties (rather than a niche party dummy) and the results were substantively identical. 5 We also ran our models with vote share and a large party dummy variable for those parties which won 5% or more of the vote in the national election. Both variables yielded the same results.

18 16 deviation of the mean party positions in each system using the voters placement of the parties in the EES survey. 6 To operationalize the tone of the overall campaign, we use a variable that measures the tone of the coverage in television and newspaper stories on the EU. 7 We also have several control variables at the contextual level. First, following the secondorder election argument about electoral cycles, we include a dichotomous variable for midterm election, coded as 1 if the EP election was held more than 12 months since the last national election and 0 otherwise. Second, we use a dichotomous variable for newly established democracy, coded as 1 for the EU member states which transitioned to democracy in the 1990s. Malta and Cyprus are thus are coded as 0. 8 Third, we include a dummy variable for whether a country has compulsory voting 9. See Table 3A in the Appendix for the descriptive statistics of all of the independent variables. To explain the variation in second-order vote choice both within and across countries, we use multi-level analysis, since neglecting the hierarchical structure of the EES data could lead to an underestimation of standard errors and spurious inferences (Steenbergen and Jones 2002). A multi-level approach corrects for dependence of observations within countries (intra-class correlation) and makes adjustments to both within and between parameter estimates for the clustered nature of the data (Snijders and Bosker 1999). Moreover, a random-effects approach 6 Our findings are robust when using the range of the parties positions in each system with the EES data as well as the standard deviation and range using expert survey placement of parties (Benoit and Laver 2006; Hooghe et al 2010). 7 The visibility of the EU in the media is another way of operationalizing the context of the campaign. One could argue that the more stories there are about the EU, the more likely a voter is to be motivated by EU-specific concerns (see de Vries et al. 2010). However, since we are interested in the type of information that voters are getting (and not necessarily the amount), we chose to use media tone as our proxy of the tone of the overall campaign. 8 We chose to use this operationalization instead of new versus old EU member states as this more accurately reflects the substantive difference between the two sets of countries. Importantly, the results do not change in any substantive way if we use the alternative operationalization. 9 These data are from the International Institute for Democracy and Electoral Assistance (IDEA) database.

19 17 allows us to explicitly model differences in voting behavior according to the specific political context, which would not be possible with a fixed effects model. Since our dependent variable is multinomial, we estimate a hierarchical generalized linear model - with voters nested within political contexts - with a logit link function and a Bernoulli sampling model. 10 We discuss the results in the ensuing section. Results To test our hypotheses, we estimate three sets of models for both abstaining and switching. Our first model tests the strength of sincere, strategic and agenda-specific motivations (H1-H3). Our second model evaluates the mediating effect of voters party type (H4-H5) by including individual-level interactions with party type. Finally, the last model tests the mediating effect of EU politicization (H6-H7) and thus includes relevant context-level variables and crosslevel interactions. [TABLE 1 ABOUT HERE] Table 1 shows the results from our individual level model of voting behavior. We find support for our first three hypotheses. The larger the distance between the voter and her party on the left-right dimension, the more likely she is to abstain and switch in the EP election. On the EU dimension, the more Euroskeptic a voter is compared to her party, the more likely she is to switch or abstain. Finally, when voters approve of their national government s performance, they are less likely to abstain or switch. All of these results are significant at the 99% level. The control variables also behave as expected: partisans are less likely to switch and abstain, more 10 We also estimated the models using a multinomial logit (and probit) link and as a three-level model with voters nested within parties and political systems. The results of these estimation strategies are very similar, with none of the substantive results (significance or magnitude) differing. For a discussion of binary versus multinomial logit models, see Alvarez and Nagler (1998).

20 18 knowledgeable and educated voters are less likely to abstain, but more likely to switch in EP elections, and older voters are generally more likely to stick to the same party in both first- and second order elections. Gender has no significant effects. [TABLE 2 ABOUT HERE] Given that we are interested not only in the significance, but also in the relative magnitude of the effect of the three motivations, we calculate the change in the predicted probability of abstaining or switching for each of the motivation variables. To compare the size of the marginal effects, we increase the values of left-right distance, EU distance and government approval from the minimum to the maximum, holding all other variables at their means. Table 2 shows the resulting effect on the predicted probability of abstaining or switching (with 95% confidence intervals in brackets). For distance on the EU dimension, moving from the minimum to the maximum values increases the likelihood of switching by 5% and abstaining by 7%. On the leftright dimension, a similar change in values increases the probability of abstaining by 8% and switching by 7%. For voters who approve of their government s performance, the likelihood of switching is 24%, but this increases to 29% for those who are not satisfied with their national government, while the predicted probability of abstaining increases by 6% when moving from approval to disapproval. These results illustrate that sincere, strategic, and arena-specific motivations all have about the same magnitude of effect on the likelihood of changing behavior between first- and second-order elections. In other words, vote choice in second-order elections are no more likely to be driven by domestic concerns (whether sincere or strategic), as posited by the classic second-order election literature, than arena-specific EU concerns. Next, we turn to the question of what conditions the importance of each of the three motivations. Starting at the individual-level, Table 3 presents the mediating effect of the type of

21 19 party that voters' supported in the national election. To test H4, that strategic protest voting is more pronounced among voters who supported the government party in the national election, the model includes an interaction between government party vote and government approval. The results lend strong support to our hypothesis, as they demonstrate that government disapproval only increases the likelihood of switching and abstaining for those voters who supported the government in the national election. The evidence for H5 is more mixed. As expected the results show that niche party voters are more likely to be driven by sincere considerations (left-right distance) when deciding whether to change parties in second-order elections. However, they do not suggest that niche party supporters are more likely to be motivated by arena-specific concerns. It is worth noting, that single issue anti-eu parties, such as the Swedish Junilistan, were not included in our niche party category, since such parties rarely win seats in national legislative elections. We illustrate the interaction between niche party voters and sincere motivations in Figure 2, which plots how the marginal effect of being a niche party voter on the likelihood of abstention and switching varies across distances on the left-right dimension (Kam and Franzese 2007; Brambor, Clark and Golder 2008). The plot clearly demonstrates the importance of ideology as a motivation for this group of voters. 11 [TABLE 3 ABOUT HERE] [FIGURE 2 ABOUT HERE] Our final models shown in Table 4 estimate the conditioning impact of EU politicization on the degree of arena-specific voting. To do this, we include interactions between the 11 One of Meguid s (2005) niche party conditions is that the issues on which the party focuses does not coincide with existing lines of political division (347-8). However, we find that their supporters vote choice in second-order elections is driven by just these divisions. Identifying differences in behavior between niche parties and their voters (and the causes of these) would be a fruitful line of future research.

22 20 conditioning variables (EU party polarization and EU campaign tone) and EU distance. We also include an additional cross-level interaction between midterm elections and government approval to control for the national electoral cycle effect of strategic protest voting hypothesized in the second-order election literature (see van der Eijk and Franklin 1996; Marsh 1998; Hix and Marsh 2007). The results corroborate the proposition that switching due to government dissatisfaction is more common when elections are held during the midterm. 12 We also find strong support for H6: the higher the level of party polarization, the greater the effect of arena-specific considerations (EU distance) on the likelihood of switching and abstaining. Equally, a more negative campaign tone increases the importance of EU distance on switching, as hypothesized (H7). To interpret both the magnitude and significance of these conditioning effects, we graphically illustrate how the marginal effects of EU distance on switching and abstention vary across the range of values of EU party polarization and EU campaign tone in Figures 3 and 4. Figure 3 shows that the marginal effect of the distance between voters and their national party on the EU dimension on switching and abstention increases, the more polarized the national party system. For both types of behaviors, it is only when the level of polarization is above the mean that EU distance has a significant effect. Figure 4 demonstrates that the marginal effect of moving from the minimum to the maximum EU distance on switching and abstention across different levels of the tone of the campaign. It shows that EU distance only matters when the campaign coverage is neutral or negative. [TABLE 4 ABOUT HERE] [FIGURE 3 & 4 ABOUT HERE] 12 When we re-run the models controlling for concurrent election, our overall results are similar, and this additional variable is insignificant.

23 21 Figures 3 and 4 thus clearly illustrate that arena-specific voting in EP elections occurs primarily in political systems where the issue of European integration is politicized. We can look to Austria as an illustration of the influence of politicization. In the campaign leading up to the 2009 EP elections, the issue of the future of European integration was fiercely debated. The Austrian party system is one of the most polarized on the EU dimension (polarization score of 1.66) with the two large mainstream parties, the People s Party (ÖVP) and Socialist Party (SPÖ), broadly supportive of further European integration whereas the populist far-right parties - the Freedom Party (FPÖ) and Alliance for the Future of Austria (BZÖ) - and the single-issue Hans Peter Martin list campaigning on a Euroskeptic platform. The coverage of the EU in the Austrian media was the most negative across the entire EU (EU tone score of -0.16). Considering this highly contentious campaign context, it is not surprising that arena-specific voting was very pronounced in Austria. A shift from the minimum to the maximum distance to the party on the EU dimension would make the average Austrian voter 27% more likely to abstain and 18% more likely to switch. 13 Not least due to arena-specific voting, Hans-Peter Martin's List "For genuine control in Brussels" won 18% of the vote and 3 out of 17 seats, whereas his party won no seats in the previous Austrian legislative elections of In contrast, EU distance has no significant effect on second-order election behavior for the average voter in countries where the EU issue is far less politicized, such as Bulgaria (polarization score=0.51, EU tone score=0.02) and Lithuania (polarization score=0.56, EU tone score=0.01). Conclusion At first glance, the outcome of the seventh direct elections to the European Parliament in 2009 seems to lend further support to the second-order election model, which has guided our 13 We calculate the marginal effects based on the models shown in Table 1, holding all other variables at their mean.

24 22 interpretation of EP elections for three decades: at 43%, turnout was not only significantly below that in national elections, but also lower than in any previous EP elections. Large governing parties generally performed poorly compared to national elections, while small opposition parties enjoyed substantial victories. Not only did green and far right parties win larger seat and vote shares than in recent national elections But, parties such as the Swedish Pirate Party and the Danish 'People's Movement against the EU' also gained representation. We can interpret these patterns as a sign that voters behave differently in EP elections because less is at stake and they are more concerned with domestic matters. Yet, other aspects of these elections may not fit so neatly with the second-order model. Anti-European parties generally performed much better than in national elections and socialist parties were punished across Europe, regardless of whether they were in government or in opposition (Hix and Marsh 2010). These outcomes suggest that concerns over European integration were important to voters and that pan-european issues mattered in the campaigns. The aim of this article has been to unpack the individual-level motivations that drive changes in voting behavior between first- and second-order elections. This contributes to our understanding of these aggregate patterns in EP elections. These findings also advance our comprehension of what motivates voters decisions in second-order elections more generally. The article proposes a theoretical framework for understanding second-order election behavior, identifying three distinct types of motivation, namely sincere, strategic protest and agenda-specific. By explicitly distinguishing among different types of motivations, we are able to not only to evaluate their relative impact on observed behaviors, but also to examine what factors condition their impact. Our results clearly show that all three motivations play a significant role in driving changes in electoral choices between first- and second-order elections. In fact, the

25 23 magnitude of the effect of sincere, strategic protest and agenda-setting motivations on the likelihood of switching party or abstaining in the EP election is remarkably similar. This suggests that the longstanding debate about whether 'Europe matters' in EP elections may now have been supplanted by the more pertinent question of when and why arena-specific considerations matter to European voters. Hence, rather than assuming that all voters are motivated by similar concerns (be they 'domestic' or 'European') in second-order elections, we find that these are mediated by both individual and contextual variables. At the individual-level, we show that the type of party a voter normally supports influences the motivations that drive her behavior. Perhaps unsurprisingly, we find that governing party voters are more likely to use EP elections as a ballot to signal their dissatisfaction with ruling parties. More interestingly, we also demonstrate that niche party voters place greater emphasis on party positions on the left-right dimension when deciding. The importance of arenaspecific considerations - i.e. EU issue voting - also varies considerably across individual and political contexts. Our findings also demonstrate that the higher the level of politicization of the European issue, the more European concerns drive vote choice in EP elections. While this finding is fairly intuitive, it has important implications for our understanding of electoral behavior and democracy in the European Union. Examining the conditioning effect of party polarization on the EU dimension shows that voters only take EU-specific considerations into account when political parties provide them with clear choices. In other words, the behavior of parties is crucial in shaping the nature of electoral choices. As an example, party polarization on the EU dimension is in general significantly lower in the newly established democracies in the East compared to the established democracies in the West. When comparing agenda-specific voting in these two sets

26 24 of countries, we also find that distances to parties on the EU dimension have a large and significant effect on second-order behavior in the West but no effect in the East. This naturally raises further questions about why the politicization of the EU varies so considerably across the member states and how the motivations of voters may encourage parties to adopt more polarized positions and the media to provide more critical coverage of the EU. The relationship among party strategies, media coverage, and voter motivation is likely to be reciprocal in ways that cannot we fully explore in a cross-national study, such as this one, but should be examined in future research.