Government coalition heterogeneity and the electoral rise of the Euroskeptic parties

Transcription

1 *Preliminary Draft, Please do not cite without authors' permission Government coalition heterogeneity and the electoral rise of the Euroskeptic parties Stefano Camatarri Francesco Zucchini DEPARTMENT OF SOCIAL AND POLITICAL SCIENCES Università degli Studi di Milano Abstract The last European elections were characterized by a growth in many countries of the percentage of votes in favor of Euroskeptic parties. However this growth has not been uniform. While for example Euroskeptic (EUSK) parties have obtained 30.9% of votes in Italy and 34.4% in the Netherlands, in Germany they have not exceeded 8% and they were essentially absent in Spain. Countries economic performances are not so clearly associated with this growth. In some political systems that experienced a relatively good economic performance these political forces won a very high percentage of votes (for instance Denmark and Netherlands), while in other countries that were substantially put under the tutelage of the European institutions and IFM for their financial and economic conditions, the same type of parties was less successful or even almost a marginal phenomenon (for instance Portugal and Spain). In sum, on the one side the great variance among European countries suggest the importance of the national contexts in explaining the success of EUSK parties; on the other side national economic conditions do not seem to be a very promising explanatory factor. In this paper we investigate about the role played by a country specific factor different from the economic performance, the heterogeneity of government coalitions in the policy space. We argue that in countries characterized by heterogeneous governments in the last years before EP elections, along the pro-anti EU integration dimension, citizens were more likely to vote for Euroskeptic parties. When the status quo cannot be changed then a certain level of government heterogeneity (polarization) on a salient policy dimension is going to feed further radicalization on the same dimension in the voting behavior without necessarily any further radicalization in the original preferences of voters. And on the other side an increase in the radicalization of political preferences of voters is more likely to be translated in a more extreme vote when the government coalition is heterogeneous. We take advantage of the 2014 EES Voter Study, of the Chapel Hill Expert Survey and of the ParlGov database in order to test these hypotheses. Introduction The last European elections have been characterized by a huge increase of votes for anti-

2 establishment parties that are also very critical with European institutions and European policies. 1 At first glance, such a phenomenon does not look like very surprising. Indeed, the recent economic and financial crisis has severely affected a considerable part of the European population. The European institutions and many national governments did not seem able to counter unemployment and loss of income. Therefore large sectors of the electorate could have punished the traditional pro-europe parties that have been also government parties during the crisis years and just before the EP elections (e.g. Hernandez and Kriesi 2016). However, a closer investigation reveals that countries economic performances are not so clearly associated with the success of the so-called euroskeptic parties (from now on EUSK parties). In other terms, although the success of EUSK parties during the last European elections is hardly undeniable (e.g. Treib 2014), in some countries that experienced a relatively good economic performance these political forces won a very high percentage of votes (for instance Denmark and Netherlands), while, on the contrary, in other countries that were substantially put under the tutelage of the European institutions and IFM for their financial and economic conditions (for instance Portugal and Spain), the same type of parties was less successful or even almost a marginal phenomenon (De Sio et al. 2014). In Fig.1a and 1b you can contrast the overall change in the electoral performance (percent of votes) of EUSK 2 parties in each EU country from 2009 to 2014 EP elections, with the percent variation of the employment rate and the percent variation of the median net equivalized income 3 between 2009 and As noticeable, there is no plausible association 4. 1 Since in this work we focus on voting behavior at the European elections of 2014, we will simply refer to these parties as Euroskeptic parties. Nevertheless, it has to be noted that their general opposition to the policies of the EU has been often considered as overlapped to their general opposition to the functioning of national political systems (e.g. Taggart 1998). A quite developed literature on populism and anti-establishment politics argues, in fact, that these parties try to emphasize issues that are away from the mainstream party competition so to attract voters that are deeply disenchanted with the processes and output of conventional politics, both at the national and at the EU level (e.g. Barr 2009; Mèny and Surel 2002; Mudde 2004, Deiwiks 2009). For an overview of the concept of anti-establishment party, see also: Schedler (1996); Abedi (2004). 2 A party is defined as EUSK party are according to the scores of 2014 Chapel Hill expert survey (see below) 3 For a definition see below and at explained/index.php/glossary:equivalised_disposable_income 4 We excluded Czech Republic that is an extreme outlier because of the impressive decline in consensus of the main EUSK party between last two EP elections. Nevertheless the correlation indexes are very low.

3 Fig.1 Percent Variation of Votes to EUSK parties by Percent Variation of Employment Rate and Percent Variation of Average Personal Net Income ( ) GRC Variation in Percent of votes to EUSK parties CY PRT ESPHRVSVN BGR Percent variation employment rate from 2009 to 2013 ITA DNK NLD SVK IRL FRA POL FIN BEL UK AUT ROU LUX SWE DEU LTU EST LVA HUN MLT GRC IRL LVA ESP BGR DEU FRA DNK AUT FIN ROU CY LTU SVN LUX EST MLT NLD HUN Percent Variation of Average Personal Net Income from 2009 to 2013 PRT ITA POL BEL UK SVK SWE In sum, on the one side the great variance of Euroskeptic vote shares among European countries suggest the importance of the national contexts in explaining the success of EUSK parties; on the other side national economic conditions do not seem to be a very promising explanatory factor. In this paper we investigate about the role played by a country specific factor different from the economic performance. It is the role of the heterogeneity of government coalitions in the policy space. Other authors have already focused on the explanatory importance of coalition outcomes for voting decisions (e.g. Cox 1997; Blais et al. 2006; Duch et al. 2010; Indridasson 2011). Nevertheless, in this contribution we argue that in countries characterized in the last years before EP elections by heterogeneous national governments along the pro-anti EU integration dimension, citizens have been more likely to vote for EUSK parties. On one side, we expect an increase of government heterogeneity to have a positive effect on the EUSK vote the more the voter has a negative attitude towards EU Integration. On the other side, we assume that a strengthening in Anti-EU voter s attitude is more likely to be translated in an increase of probability to vote EUSK parties when the heterogeneity of governments is high. These hypotheses are inferred from a theoretical account that shares with other well-known contributions (Kedar 2005, 2009) in the literature the idea that high levels of policy stability induce extremist voting. However, differently from these contributions we assume that: a) voting is mainly retrospective and it is not oriented by calculations about future policy outcomes b) decision making processes in parliamentary democracies are characterized by the existence of a variable number of effective veto players, usually government parties, that prevent an easy and complete translation of electoral changes in policy change. Therefore, we assume that on the

4 dimension they consider salient, voters evaluate comparatively the opposition platforms and the government outcome in order to choose the party to be voted. If the status quo is trapped in the government range and (according the veto players theory) cannot be changed (or can be only marginally changed), then the government party platforms far from the status quo will not be believed (or will be strongly resized ) as the voters will consider the government policy outcome, namely the preservation of the status quo, as a better approximation of the real party platform. It is somehow not very easy to believe to the promise of a change, at least in a specific policy dimension, of someone who allies with those who defend its status quo. If we consider the EU integration as a salient dimension of 2014 EP election (see below) then in all European countries the status quo on this dimension would be unchangeable and quite similar everywhere given the common country membership to the EU. Therefore, what really makes the difference is the participation in the government coalition of a moderately Euroskeptic party, that extends the ideological range of the government on EU integration. Its membership, since not effective in changing the status quo in that dimension, convinces moderately euroskeptic voters to support strongly euroskeptic parties. A preliminary country level investigation (see Fig.2) seems not to disconfirm this research hypothesis. If we plot the average government range on the EU integration dimension (see below) against the variation in percent of votes to EUSK parties (see below) between 2009 and 2014, the existence of a linear relationship is pretty evident, with a Pearson correlation index significant and equal to Fig.2 Percent Variation of Votes to EUSK Parties by Average Government Range on European Integration dimension ( ) ITA GRC SVK DEU FRA IRL DNK POL SWE UK PRT AUT FIN 0 MLT ESP SVNEST LUX CYHRV ROU LTU NLD LVA HUN BEL BGR Average Government Range in European Integration Dimension ( )

5 In order to test rigorously our hypotheses we take advantage of the 2014 European Election Voter Study (EES). This is a EU-wide survey that allows to conduct an analysis at individual level and to control for a number of variables already ascertained in the literature as explanatory factors of Euroskeptic voting. In the next section we try to sketch a theoretical account to explain why and how national government political heterogeneity should affect electoral behavior at the EP elections. In the third section we will present the data and the statistical analysis used to test our hypotheses. A final section dedicated to a discussion of the main results will follow. A theoretical account According to a canonical spatial voting model, the voter would compare the party platforms and he/she would vote for the party that proposes the platform closest to his/her ideal point (Davis, Hinich and Ordeshook 1970; Henelow and Hinich 1984). The status quo policy as well as the feasible policy change would not be taken in consideration by the citizens when they vote. The limits and lack of realism of this description has been noticed for quite long time. A departure from it implies implicitly or explicitly an explanatory role for features of the political system that affect the policy outcome. According to Grofman (1985), a voter could prefer a party that does not have the closest platform to his/her ideal point if the promised change of the status quo is in the right direction for the voter. The obstacles that this party will face in implementing its platform are the fundamental ingredient that allows to seemingly violate the downsian criterion of proximity (Downs 1957). If, because of these obstacles, the expected new status quo does not correspond to the party s platflorm but somehow to a more moderate version of this platform, then the voter could rationally vote for this party instead of voting for another party that proposes a platform closer to his/her ideal point but on the opposite side with respect to the status quo. Obviously, the existence of these obstacles depends on political and institutional circumstances. It is not difficult to imagine as obstacles to a full implementation of party platforms the divided government in US political system and the coalition governments in European democracies. More recently Kedar (2005) in a similar vein has argued (and empirically tested) that in consensual democracies, insofar as they pull policy in the desired direction, extremist parties can be attractive also for non extremist voters. In countries where policies are often the result of institutionalized multiparty bargaining, i.e. consensual democracies, votes are watered down by power sharing and an incumbent (government) party cannot fully keep its electoral promises. Therefore, many voters would

6 compensate for this watering down by supporting parties whose positions differ from their own and are more extreme. Kedar considers the compromise that attenuates the importance of the policy change as the main effect of a power sharing political system that voters perceive. By misrepresenting the real preferences many voters would try to shift the content of this compromise as near as possible to their ideal point (that is not necessarily very close to extremist party platform). Kedar s theory (and other similar approaches) is based upon two assumptions. First, voting behavior is assumed to be instrumental and prospective. Voters are indeed supposed to take in consideration general features of political system for their effects on the future policies. Secondly, voters are supposed to believe that the final policy outcome is an average of the policy positions of all parties in a parliament (Kedar 2005, pp. 188). In truth, both these assumptions are controversial. With reference to the second one, according to the most well established spatial theories of law making (Kreihbiel 1998; Cox and McCubbins 2005; Tsebelis 2002) it is precisely in power sharing systems that we expect a low correlation between changes in electoral preferences and policy change. In a parliamentary democracy, unless voting for extreme parties creates the condition for a complete change of government majority, the increase of consensus for an extremist party does not necessarily generate a policy change and in some circumstances it increases the probabilities of policy stability. What matters is the spatial position of the veto players (or pivotal players). If you hypothesize that voters behave according to spatial theories in order to link their electoral behavior to future policy outcomes, then no strong argument can support the Kedar s compensational vote. Even more important for our scope, Kedar s first assumption about the nature of voting behavior is quite at odds with the features of the European elections as second-order national elections (see Hobolt and Wittrock 2011; Hix and Høyland 2013). These latter, in fact, are clearly national contests, as they imply a choice among parties at a national level. Nevertheless, as they only affect the composition of the European Parliament, they are at most only weakly linked with the future national policies and policy makers, In other terms, EP elections are not really the best context for plausibly assuming a prospective and instrumental voting (see also Marsh and Mikhaylov 2010). Despite this, we do agree with Kedar that the features of a national political system affect voting behavior. However, first we consider a specific and changing element, the government coalition heterogeneity, that affects the policy outcomes, instead of very general and permanent classification e.g. the Lijphart s types of democracies (see Lijphart 1999); secondly, we do not assume that voters reason according to the next possible policy outcomes. Rather, we assume that they vote expressively (Hilmann 2010; Hamlin and Jennings 2011) for the closest policy positions, as they

7 perceive them given the policy outcomes. In other terms, we assume that voters adjust the policy positions of the incumbent parties by considering their policy outcomes and compare them with the platforms of the opposition parties. Therefore, even if a government party platform is the closest to a specific voter, such a voter can vote for an opposition party if the status quo preserved by the government is far enough from his/her ideal point. Voters (or at least an important portions of voters) are here assumed to vote with the aim to express their level of agreement with policies, both those kept in force by the government parties and those that were promised by the opposition parties. Differently from the causal mechanism proposed by Kedar, the expressive vote allows always the existence of an interaction between the features of the national political systems that influence policy change/policy inertia and electoral behavior, even when voting is not supposed to directly affect future policies. A simple spatial illustration can help to clarify the argument and to link it to voting behavior at the last EP elections (Fig.3). Imagine that at time t0 a new government is formed. Given the status quo SQ the government at time t1 will decide about the change (or preservation) of SQ. At time t2 the voter V has to vote for the European Parliament elections. Level of desired European Integration is the salient dimension along which we suppose voters choose the party. This hypothesis has been already supported empirically by previous analyses testing the so-called Europe Matters model (e.g. Bellucci, Garzia and Rubal Maseda 2012). More recently, it has also found concrete evidences among the first studies on 2014 EP elections, in particular those focused on Western European Countries (e.g. Trechsel, De Sio and Garzia 2014). In such a political system there are four parties, running from a strong pro-europe party PE on the right (high values) to a fully euroskeptic party NE on the left (low values). The status quo SQ is located on the pro-europe side of the ideological spectrum and coincide with PE position 5. Therefore SQ cannot be changed. Cutpoints cp help to identify the decision of the voters according to their position in the policy space. On the left or on the right of each cut point voters will vote for the closest perceived party platform, respectively on the right or on the left. We can figure out different scenarios. In the scenario a there is a single party government formed by PE. In the scenario b PE and the ME form a two party government quite homogenous along the European integration dimension. In the scenario c, on the contrary, the government is heterogeneous as ME1 is a moderately euroskeptic party. In scenario a the voters decide according to the canonical proximity model. In the scenario b, cp1 shifts on the right and voters that previously would have voted for ME will vote for ME1, as now SQ is considered the real 5 Obviously we can relax this assumption and imagine that PE is only close to SQ. The substantive implications of the model do not change.

8 policy position of ME 6. However we do not expect in this scenario any increase of consensus for the fully euroskeptic party NE. In the scenario c, when the government is pretty heterogeneous, cp2 shifts on the right and additional voters on the left of ME1 are available to vote for NE as SQ if farther than NE. Therefore if we compare scenario a or b with scenario c we will find out that voters with the same policy preferences will vote differently and in scenario c for a more extreme party. Fig. 3 Spatial Model with Government Heterogeneity And Expressive Voting The very simple idea is that policy inertia, by making less credible political forces that will (or affirm to will) a moderate change and are at the government, increases the propensity to vote for more extremist parties without necessarily any change in the voters preference distribution. The implications of this model are straight and can be summarized in the two following hypotheses: 1) An increase of government heterogeneity on the EU Integration dimension has a larger positive effect on the EUSK vote the more the voter is extremist on the same dimension; 2) An increase in Anti-EU attitude of voters is more likely to be translated in an increase of probability to vote EUSK parties the higher is the heterogeneity of governments on EU integration dimension. 6 The hypothesis that in the voter s perspective the policy position of a government party fully coincides with the government s policy outcome can be relaxed without substantial changes in the argument we propose. In general,, where MEp is the perception of ME policy position and k is a discount factor. In our example, k is assumed equal to 0. However, even for slightly higher values the results are qualitatively similar.

9 Data As mentioned earlier, in order to test the hypotheses outlined above we rely upon the most recent post-electoral study at the European level, the 2014 EES Voter Study 7. Such study gathers information about electoral behavior in all the EU countries in a single data set and it has been specifically thought and planned for the EP elections. The sample has approximately interviewees, and data collection has been done via CAPI 8. The original matrix has been integrated by political and economic variables at party and country level. Macroeconomic conditions are inferred from Eurostat data regarding employment rate and median net income for the time span ( while government political heterogeneity is calculated on the basis of party policy positions that are derived from the most recent Chapel Hill Expert Survey data set. Finally, information about government party compositions are derived from the ParlGov data set ( The dependent variable Similarly to some recent studies (e.g. Auel and Raunio 2014b; Tronconi and Valbruzzi 2014), in order to identify the Euroskeptic parties we referred to the mean of the Chapel Hill survey (CHES) experts responses about overall orientation of national party leaderships towards European integration in In particular, on a scale ranging from 1 (= strongly opposed to EU integration) to 7 (= strongly in favor of EU integration) we categorized as Euroskeptic parties those parties with a mean score inferior or equal to 3. The resulting dependent variable (EUSKVOTE) is thus equal to 1 when the respondent declares to have voted for an Euroskeptic party in the 2014 EP elections 10 and zero when he/she declares to have voted for another type of party. Since the causal mechanism we propose is based on the disrepute that affects the ruling parties that are unable to keep their promises of change, when the 7 Schmitt, Hermann; Hobolt, Sara B.; Popa, Sebastian A.; Teperoglou, Eftichia; European Parliament, Directorate-General for Communication, Public Monitoring Unit (2015): European Parliament Election Study 2014, Voter Study, First Post- Election Survey. GESIS Data Archive, Cologne. ZA5160 Data file Version 3.0.0, doi: / The sample is not a simple random sample of European voters but a disproportionate stratified random sample, with about respondents from each EU country, with the exceptions of Malta and Luxembourg, where the sample size is about 500. Given the huge variation in country size between United Kingdom at one extreme and Luxembourg at the other, this is very disproportionate. However, such strategy provides the necessary database for identifying national differences in voting behavior. A sample of the same size designed simply to maximize inference to the European electorate would have far too few individuals to assess the extent to which voters in Ireland, or Denmark, or Sweden fitted the European model as well as those in France, The Netherlands or Finland (Marsh 2002, 12). 9 Ryan Bakker, Erica Edwards, Liesbet Hooghe, Seth Jolly, Gary Marks, Jonathan Polk, Jan Rovny, Marco Steenbergen, and Milada Vachudova Chapel Hill Expert Survey. Version Available on chesdata.eu. Chapel Hill, NC: University of North Carolina, Chapel Hill) 10 In the 2014 ESS Voter Study respondents were asked, Which party did you vote for in these recent European Parliament elections? and Which party did you vote for in these last parliamentary elections? After these questions, they were presented with the names of the parties deemed as more relevant in their political systems.

10 respondents report having voted an Eurosceptic party that was government member during the period under consideration the variable EUSK is equal to A complete list of the categorized Euroskeptic parties, including those that were in government during the time span we considered, is reported in Appendix. Independent and control variables According to the theoretical model, an increase in the Government heterogeneity on EU integration dimension increases the propensity to vote for an Euroskeptic party the more the voter is located on the euroskeptic horn of the ideological spectrum. On the other side, an increase in the voter s euroskepticism is more likely to be translated in an euroskeptic vote the larger is the government range on the EU integration dimension. Therefore our main independent variable is in fact an interaction term between Government heterogeneity and the voter s attitude toward EU s integration. In tune with the theoretical model, our measure of heterogeneity is spatial. We considered the range of the government coalitions on EU integration issue as you can infer from the CHES expert survey (see above). We explicitly focused on the scores of those parties who have been part of a government coalition between (September) 2008 and (April) 2014 in their country, namely from the official beginning of the financial and economic crisis and the EP elections. First, we measured ideological heterogeneity of each government in each country as the distance between the two parties with the highest and the lowest score in a government coalition. The other government parties have been considered as absorbed, since they were located inside the core of the most extreme ones (Tsebelis 2002). For each country we consider as many government coalition ranges as the governments that were in office during the selected period. In order to get an overall measure of spatial heterogeneity over time, we calculated the weight in days of each government and computed the country average coalition range from 2008 to 2014 (AVEREURANGE) 12. The position of the voter along the EU integration dimension is not easy to infer. In this regard, we alternatively referred to two different questions in the 2014 EES questionnaire. First, we focused on the utilitarian dimension of EU integration (e.g. Lubbers and Scheepers 2005, 2010; see also van Spanjie and De Vreese 2011) that is well represented by the following question: Generally speaking, do you think 11 If we run the regression models (see below) by considering EUSK equal to 1 even when the euroskeptic parties are in government results are quite similar and models fit slightly worse. Analyses are available on request. 12 In case of a caretaker cabinet (for instance the Monti s Government in Italy) we considered as coalition government party every party that supported the government investiture. In preliminary statistical analyses we have also considered a dummy variable Minority government that has values between 0 and 1 according to the proportion of days minority governments have been in office in each country during the period we consider ( September 2008-April 2014) It is not ever significant and does not affect ever the coefficients of the other variables.

11 that our country s membership of the EU is a good thing, a bad thing or neither a good thing nor a bad thing?. The variable EUT is equal to 1 when the respondent answers that his/her country membership is a bad thing, while it assumes value of 0 in all other cases. The other variable (EUN) allows to identify the attitude s intensity towards the EU integration process in general, without a specific reference to his/her country. Such information is inferred from the following question: Some say European unification should be pushed further. Others say it already has gone too far. What is your opinion? Please indicate your views using a scale from 0 to 10, where '0' means unification has already gone too far" and '10' means it "should be pushed further". What number on this scale best describes your position?. The government heterogeneity on EU dimension as well as the vote for EUSK parties in 2014 EP elections could both depend on the previous vote for EUSK parties. One can hypothesize that if in a country EUSK parties have been very successful during the years before the last EP elections, then the same parties have become government parties or other government parties tried to locate themselves on positions increasingly euroskeptic in order to keep consensus. And on the other side who had voted for an EUSK party in the past is a priori very likely to confirm his/her choice in 2014 s EP elections. Therefore the relationship between Government range on the EU dimension and Euroskeptic voting might be spurious (see picture). In order to insulate the effect of government heterogeneity from the past Euroskeptic vote we introduce past vote recall for Euroskeptic parties (PASTEUSKVOTE) at the last National Parliamentary elections as control variable 13. The rest of the variables that a priori, also according to previous studies, can have an explanatory role 13 In the Appendix we also report an alternative analysis without PASTEUSKBOTE, in which the EUSK voting is substituted with vote switching towards an EUSK party. This is an alternative dependent variable, which takes value of 1 if the respondent declares to have switched his/her vote from a pro-eu party to an anti-eu party at the last EP Elections, and assumes value 0 in all other cases.

12 can be divided into socio- economic individual variables, political individual variables and contextual (or country) variables. Among these latter we considered alternatively two measures of country economic performance: the percent change of the national Median equivalised net income (VARMENI) and the percent change of the national Rate of employment (VARE), both from 2009 to The first one, in particular, indicates the median income of an household after the taxes and after having considered the household composition (in terms of household members age) 14. Both these measures summarize countries economic performances and are politically sensitive indicators (see also van der Brug, van der Eijk and Franklin 2007). As we are interested in controlling the effect of previous electoral behavior we have also to consider institutional and political factors that can attenuate or strengthen the propensity of voters to vote for the same party they voted in the last parliamentary elections before 2014 EP elections. Indeed, you can hypothesize that the explanatory power of past voting in national elections is weakened when the national electoral system encourages strategic voting. In this circumstance voters, at least according to an instrumental approach, should be more inclined to vote their second best option (i.e. not an EUSK party) in national elections and their best option in the European Elections (i.e. an EUSK party) as everywhere EP elections take place with a proportional system that discourages strategic voting. We took in consideration the Gallagher s disproportionality index (DISPRO) calculated on the bases of the outcomes of the last parliamentary elections (votes vs. seats) in interaction with the PASTEUSKVOTE in order to control for this effect. Similarly, the probability that a voter insists on voting the same party he/she voted in the last national parliamentary election seems a priori to be negatively related with the time span that divides this election from the European one. Therefore, we have also introduced an interaction between PASTEUSKVOTE and the variable SPANEL, that for each country is equal to the months that separate the two elections. The last country-level variable is a dummy variable called POSTCOM. This one is equal to one when the country was in the past a communist country. We take this information into account because post-communist environments have been often associated to limited electoral experiences and low participation (e.g. Schmitt 2005). Since also non- voting could be interpreted as an alternative way to signal Euroskepticism (Hobolt et al. 2008, see below), then we expect that in these countries the probability to vote for Euroskeptic parties is low as the same distaste towards the EU is often communicated by the abstention (in Appendix 1 we consider also regression models with only Western European Countries). 14 Using median, rather than mean income, results in a much more accurate picture of the average income of the middle class since the data will not be skewed by gains and abnormalities in the extreme ends.

13 The individual socio-economic variables can be classified in sociotropic evaluations, egotistical (pocketbook) evaluations, and objective socio-demographic conditions. As sociotropic evaluations we have considered predictors concerning retrospective (RETROECON) and prospective economic (PROSPECON) perceptions. The retrospective question wording in 2014 EES Voter Study is What do you think about the economy? Compared to 12 months ago, do you think that the general economic situation in our country is a lot better, a little better, has stayed the same, a little worse, a lot worse, while the prospective one reads: And over the next 12 months, how do you think the general economic situation in our country will be?. Possible answers to both questions are ranked in a fivelevel Likert scale ( A lot better / a little better / stay the same / get a little worse / get a lot worse ). As egotistical perceptions we considered the evaluation about the social position. SOCIAL STATUS regards individual self-placement on the social ladder, measured on a scale from 1 to 10, where 1 means the lowest level in society and 10 means the highest level in society. As socio-demographic variables we took in consideration AGE, GENDER, LIVING AREA and OCCUPATIONAL CONDITION. The latter is a dummy variable that is equal to one when the respondent is a unemployed, manual worker or housekeeper, while the place of residence is classified according to the following trichotomy: living in a rural area or village, in a small or middle sized town or in a large town. Several studies found that the level of interest in politics and the level of information affect vote choice at EP Elections to some extent (e.g. Hobolt et al. 2008). Therefore, we consider as individual cognitive variables a dummy measure of political information (POLINFO) that is equal to 1 when the respondent follows the news at least once a week (on internet or on TV or in the newspapers), a measure of education (EDU) that is equal to 1 when the respondent continued studying after the age of 15, and a measure of political interest (POLINT) that is equal to 1 when the respondent is at least moderately interested in politics. Finally, according to an important stream of the literature, EP Elections are second-order national elections where vote choice is prevalently driven by national concerns, i.e. they are supposed to work as mid-term referendums on government performance (Reif and Schmitt 1980; van der Eijk and Franklin 1996; Marsh 1998; Schmitt 2005; Hix and Marsh 2007; van der Brug and van der Eijk 2007). Therefore, we introduce as an additional control variable the predictor GOVDISAP, that is equal to 1 when the respondent disapproves the current government s record and assumes the value of 0 when he is satisfied with it. For a better understanding of the described variables, we report in the Appendix a synthetic table of their summary statistics.

14 Empirical analysis The nature of data we analyze implies that individual observations are not completely independent from each other, as they are nested into a superior level, the country, to which each respondent belongs. This inclusion could lead to residuals that are not independent within the same country. If we used ordinary regressions then the standard errors and the significance of coefficients would be respectively under and overestimated (Steenbergen and Jones 2002). In order to make our analyses reliable, we applied multi-level logistic regression analysis to EES data (see also Snijders and Bosker 1999). Specifically, we run two classes of regression models according to the different types of operationalization of individual attitudes toward EU (EUT or EUN) and contextual economic variables (VARMENI or VARE). We present here only the models with VARE. Results with VARMENI are very similar and reported in Appendix. TABLE1: Predictors of EUSK voting VARIABLES (1) (2) (3) (4) GENDER (Female=1) *** *** *** *** (0.0952) (0.0941) (0.0958) (0.0949) AGE ( ) ( ) ( ) ( ) LIVING AREA (Rural, benchmark level) Small or middle sized town (0.114) (0.112) (0.115) (0.113) Large town (0.126) (0.124) (0.127) (0.125) SOCIAL STATUS (0.0332) (0.0327) (0.0333) (0.0328) OCCUPATIONAL CONDITION 0.358*** 0.329*** 0.375*** 0.354*** (0.108) (0.107) (0.109) (0.107) RETROECON (0.0673) (0.0670) (0.0673) (0.0671) PROSPECON *** *** *** *** (0.0683) (0.0676) (0.0681) (0.0676) GOVDISAP 0.734*** 0.772*** 0.774*** 0.807*** (0.107) (0.106) (0.109) (0.108) POLINT (0.107) (0.105) (0.107) (0.106) POLINFO * * (0.102) (0.102) (0.103) (0.102) EDU (0.137) (0.136) (0.139) (0.138) VARE (4.405) (4.756) (4.209) (4.553) POSTCOM *** *** *** *** (0.541) (0.584) (0.531) (0.572)

15 PASTEUSKVOTE 4.402*** 4.434*** 5.608*** 5.546*** (0.123) (0.121) (0.352) (0.351) AVEREURANGE 0.618** 0.689** ** (0.293) (0.317) (0.327) (0.308) EUT 1.333*** 0.857*** (0.122) (0.220) EUN 0.178*** 0.119*** (0.0177) (0.0303) EUT*AVEREURANGE 0.353** (0.137) EUN*AVEREURANGE *** (0.0220) DISPRO (0.0461) (0.0500) SPANEL (0.0188) (0.0205) PASTEUSKVOTE*DISPRO (0.0228) (0.0230) PASTEUSKVOTE*SPANEL *** *** ( ) ( ) Constant *** *** *** *** (0.571) (0.581) (0.816) (0.836) Log likelihood Residual Intraclass Correlation AIC BIC Observations 8,996 9,442 8,996 9,442 Number of country_num ) Models without interaction terms The first two models, alternatively with EUN or EUT show that, once you control for vote recall at the last national elections, as well as for other well-known predictors of EUSK voting, government Heterogeneity (AVEREURANGE) along the EU integration dimension effectively has a marginally significant effect on the probability to vote for an EUSK party in 2014 EP elections. Indeed, an increase of one unit of Government coalition range implies, coeteris paribus, an average increase between 85% and 106% in the probability to vote for an EUSK party. The control variables that behave as expected in these cases are PROSPECON, POLCOM, GOVDISAP and OCCUPATIONAL CONDITION. Euroskeptic voters in 2014 EP elections are pessimist on the economic prospective, they are mainly manual workers or unemployed, they disapprove the government record and belong more frequently to Western European countries. Differences in employment rate at the country level (VARE) do not play any role, as well as all individual cognitive variables: POLINT, EDU and POLINFO are not ever significant. Finally, Euroskepticism seems to be a male affair. Being women, in fact, decreases the probability to vote

16 for an EUSK party of 42%. 2) Models with interaction terms Our theoretical account predicts that government heterogeneity on the EU integration dimension favors EUSK voting mainly among those voters with at least a moderate Euroskeptic attitude. Moreover, it assumes that a radicalization of the negative attitude towards EU integration is more likely to be translated in EUSK voting when such government heterogeneity increases. In other words, the theory suggest us to consider the effect of government heterogeneity as mediated by individual attitudes towards the EU (and the reverse). In order to test these hypotheses, we introduce in models 3 and 4 an interaction term composed of AVEREURANGE and EUN (or EUT). In addition, we also add two other interaction terms between PASTEUSKVOTE and DISPRO and between PASTEUSKVOTE and SPANEL. As explained in the previous section, by doing so we aim to take into account factors that may affect the persistence of the electoral choice over time. Fig.4 Average Marginal Effects Average Marginal Effects of AVEREURANGE with 90% CIs Effects on Pr(EUSKVOTE=1) Average Marginal Effects of AVEREURANGE with 90% CIs EUN 0 1 EUT

17 Average Marginal Effects of EUN with 90% CIs Effects on Pr(EUSKVOTE=1) Average Marginal Effects of EUT with 90% CIs AVEREURANGE AVEREURANGE Average Marginal Effects of PASTEUSKVOTE with 90% CIs (EUT) Effects on Pr(EUSKVOTE=1) SPANEL European Election - Last First Order Election. Time span in months ' If we consider as a relevant attitude towards the EU the individual level of accordance with the unification process, then an increase of government heterogeneity on the EU integration dimension has a positive and significant impact on EUSKVOTE as far as the voters have at least a moderately euroskeptic attitude towards EU unification (beyond score 6 on variable EUN). The marginal effect

18 is not significant below 6 and increases with the strength of this attitude. For voters who have the highest score (11) an increase of one unit in Government Heterogeneity implies on average an increase of 6% in the probability to vote for an EUSK party. If we consider as attitude towards the EU the dichotomous evaluation of one s country membership to the EU (EUT, 0 = neutral / a good thing; 1 = a bad thing) then when the respondent considers a bad thing his/her country membership of EU, the same increase of the government heterogeneity implies an increase of almost 8% of probability of EUSK vote. Also the reverse relationship, is confirmed, even if in the case of EUN the impact is very small. An increase of one unit of EUN (namely more negative evaluation of EU unification perspective) implies an increase in the probability to vote for an EUSK party of around 2,5% when the government range on EU integration dimension is equal to 3. When it is zero the effect is close to be null. On the contrary when we consider the variable EUT the effect of the radicalization of negative attitude towards European Union is stronger and increases with the increase of the government range on EU integration dimension. When the government range is equal to 0, then shifting from a non negative to a negative opinion about one s own country membership to European Union implies an increase in the probability of EUSKVOTE of around 3%. When the range is 3 such an increase reaches almost 23%.. Even if a direct comparison between the explanatory power EUN and EUT is not possible, given the different nature of the two variables, what would really matter for the voters seems not to be a general evaluation of the European Union project but the relationship between his/her country and the European Union in order to define the EU integration dimension. The effect of past voting has been evaluated in interaction with some contextual variables. As we expect the time span between the last national election and the 2014 EP election affects the voting persistence. Voters who voted in the past national elections EUSK parties tend to vote EUSK parties the closer the national elections are to the EP elections. Unexpectedly, we find a result opposite to our predictions about the interaction between past voting and electoral systems. Where the national electoral system was disproportional the voters who vote for EUSK parties, compared to the other voters, have more probability to replicate their vote in EP Elections than voters in the same condition but belonging to countries with electoral systems more proportional. In other terms, it does not seem that there are more crypto-euroskeptic voters who strategically renounce to vote euroskeptic parties during the national elections and vote Euroskeptic in EP election when the national electoral system is disproportional. Quite on the contrary, voters confirm more often their past national euroskeptic vote when they are from national political systems that give less chances to small parties (including

19 presumably EUSK parties) to be represented. In other terms, they seem even more strongly motivated when their demand of representation is somehow expressed but unmet at the national level. Conclusions An increasing number of scientific contributions insists on the importance of political and institutional constraints in explaining the different aspects of electoral behavior (Adams, Merrill III, and Grofman 2005; Fiorina 1992; Grofman 1985). However contributions that try to explain the nature of voting behavior (and more precisely extremist voting) seem to hypothesize that, according to voters, policy outcomes are the results of the relative importance in consensual democracies, in terms of seats weight, of all parties in parliament (Kedar 2005, 2009). This presumption is clearly at odds with the most celebrated spatial models of law making in the parliamentary democracies, in particular with the veto players approach and with the empirical evidence that supports this theory (Tsebelis 2002; Tsebelis and Chang 2004; Curini and Zucchini 2011). Voters are somehow supposed to have false beliefs about the way the national policy making works. Very recently other studies take seriously in consideration the veto players theory and the amount of (the promised) policy change as a crucial explanatory variable but while they are interested in explaining the level of turnout (Carlin and Love 2013) or the strength of partisanship (Hahm 2016), they are completely silent about the political content of the electoral choice. Finally, all of these explanations deal with first order elections and assume more or less explicitly that voting is instrumental and prospective. In other words, they assume that what would really matters for the voter are the future policies and how political system s features (namely the decision-making capacity) affect them. In this paper we investigated the 2014 European elections. This type of election is usually considered as second-order national elections (see Reif and Schmitt 1980). Indeed, their electoral campaign takes place at the national level. Nevertheless, their electoral stakes are not (at least not directly) the future national policies or the national governments composition. In such a context, it is difficult to imagine that prospective and instrumental voting is the main component of the electoral choice. Indeed, according to the prospective and instrumental view of voting behavior, we should not find any significant relationship between the lack of national policy change (more precisely the political constraints that prevent this change) and the content of voting behavior. We addressed euroskepticism as the specific content of vote choice and - consistently with this research interest and veto player

20 theory (see Tsebelis 2002) - we have considered as political constraint the heterogeneity of government coalitions on the European integration dimension. However, the casual mechanism we proposed is different, if compared to previous studies. According to our model, in fact, voters choose retrospectively and correct the official positions of the government parties by looking at the real achievements of the government. In other words, by voting they express their level of agreement with the current policies, both those in place and those which are promised by the opposition parties. Since the status quo, namely the country position in European Union, could not be changed, voters moderately euroskeptic can prefer fully euroskeptic parties when parties moderately euroskeptic are (ineffective) members of the government, and consequently the government range is relatively large. We argue that when the status quo cannot be changed then a certain level of government polarization (heterogeneity) is going to feed further radicalization, in euroskeptic terms, in the voting behavior without necessarily any further radicalization in the original preferences. And on the other side an increase in the radicalization of political preferences is more likely to be translated in a more extreme vote when the government coalition is heterogeneous. Our expectations are largely met and suggest further researches also on national elections. We can hypothesize that the possible new extremisms can grow more intensely on those dimensions of the political competitions where, on the one side, the status quo cannot be easily changed and, on the other side, the current government composition is more heterogeneous and political forces in favor of a moderate change are partially or fully discredited because of the preservation of the status quo. Only future research will be able to fully assess the explanatory power of this hypothesis.