- IPSA World Congress 2016, Poznan - RC Playing the Multilevel Game: Federalism and the Articulation of Power

Transcription

1 - IPSA World Congress 2016, Poznan - RC Playing the Multilevel Game: Federalism and the Articulation of Power Simon WILLOCQ PhD Candidate F.R.S.-F.N.R.S. Fellow Researcher Cevipol / Université libre de Bruxelles simon.willocq@ulb.ac.be Camille KELBEL PhD Candidate IAP PartiRep Cevipol / Université libre de Bruxelles camille.kelbel@ulb.ac.be Why do voters make distinct party choices at the different levels of government? An analysis of the causes of ballot-splitting at the simultaneous federal, regional and European elections of May 2014 in Belgium - Abstract - In multi-level political systems, citizens are sometimes given the opportunity to participate in different elections that are held on the same day at different levels of government - e.g. simultaneous national and regional elections, concurrent European and local elections. On these occasions, an important proportion of voters decide to split their ticket - that is, to choose different parties in the different electoral contests. Up to now, there have been very few comprehensive studies on the causes of split-ticket voting. The present paper aims to fill this gap by investigating the characteristics and motivations of ticket-splitters in the simultaneous federal, regional and European elections of May 2014 in Belgium. Using the data from the 2014 PartiRep Belgian Voter Survey, we test a model of ballot splitting, which incorporates various types of potential explanations. The results of our multivariate analyses indicate that party identification reduces the probability of split-ticket voting. By contrast, voters indecision proves to be a catalyst of ballot-splitting: timing of the voting choice emerges as a strong predictor of ballot splitting, with late deciders being more prone than early deciders to split their vote. Moreover, we point at the existence of a complex and ambivalent relationship between the use of preferential voting and ballot splitting, which partially contradicts the findings of previous studies. Finally, in line with the second-order elections theory, it appears that some Belgian voters have split their ticket in the concurrent federal, regional and EP elections, because they perceive a differential in importance between these three types of elections. 1

2 Introduction In the field of electoral research, split-ticket voting is usually depicted as an interesting and intriguing type of electoral behaviour which often proves to be extremely difficult to account for. Ballot splitting occurs when an elector simultaneously votes for at least two different political parties in different elections that take place concurrently (Sanz, 2008). Ticket splitting may be either horizontal (when the simultaneous elections are held to fill different offices at the same level of government) or vertical (when the concurrent elections are held at different tiers of government). Since the 1950s, the phenomenon of ticket splitting has been widely investigated in the United States where every four years, voters are invited to participate in simultaneous presidential and congressional elections (Alesina & Rosenthal, 1995; Campbell & Miller, 1957; De Vries & Tarrance, 1972; Fiorina, 1996; Jacobson, 1990). Similarly, ballot splitting has been intensively studied in countries with Mixed Member electoral systems such as Germany, Japan and New Zealand, where voters are given the opportunity to cast two ballots for the same election: the first vote for a candidate competing in a SMD contest and the second one for a party list competing in a PR contest (Burden, 2009; Jesse, 1988; Karp, 2009; Pappi & Thurner, 2002; Schoen, 1999). Until the late 1990s, scholars have devoted much less attention to vertical ballot splitting, with the exception being Campbell and Miller study on ticket splitters in US simultaneous national (presidential and congressional), state and local elections (Campbell & Miller, 1957). More recently, however, research on vertical split-ticket voting has developed due to the appearance of new multilevel political systems in Europe. Many European states have engaged in both a process of decentralization (or even federalization) and the process of European integration, which have led to a multiplication of the tiers of government and as a corollary, a multiplication of elections. As a consequence, the coincidence in time of electoral contests held at different levels of government has become more frequent in those countries, thereby allowing vertical ticket splitting. Several studies have then examined vertical ballot splitting in these European democracies and brought some interesting insights into why 2

3 some voters make distinct party choices at the different tiers of government (Elklit & Kjær, 2005; Heath et al., 1999; Maddens & Hajnal, 2002; Rallings & Thrasher, 2003; Sanz, 2008; Van Aelst & Lefevere, 2011). Yet, they often focus on specific types of explanations and thus lack an encompassing perspective. The present article precisely aims to provide a more comprehensive understanding of vertical split-ticket voting. To this end, we seek to identify the main determinants of ballot splitting in the 2014 Belgian simultaneous federal, regional and European elections, by testing a wide range of potential explanations suggested in the literature. Generally speaking, the phenomenon of ticket splitting clearly challenges classic theories of electoral behaviour which emphasize the image of a stable and loyal voter whose party choice is largely determined by long-term variables such as sociodemographic characteristics (Lazarsfeld et al., 1968) or party identification (Campbell et al., 1960). These theories fail to explain why a voter decides to endorse different parties at simultaneous elections, while his/her socio-structural characteristics and the strength of his/her partisan attachment obviously remain constant across the different electoral contests. Faced with this challenge, students of voting behaviour have developed specific theories to account for split-ticket voting and have empirically analysed the characteristics and motivations of ballot splitters. In this rich scholarly debate, ticket splitting is argued to be triggered by a wide range of factors such as the absence of partisan attachment (Campbell & Miller, 1957), voter s indecision (McAllister, 2002), the willingness to support particular candidates (Beck et al., 1992) or the desire to obtain moderate policy outputs (Fiorina, 1996). Regarding vertical ballot splitting more specifically, the second-order elections theory put forward by Reif and Schmitt (1980) holds that a large proportion of the electorate votes for different parties at different tiers of government, because they tend to behave more strategically in national elections (that they perceive as being the most important ones) and more sincerely in subnational and supranational elections (that they perceive as being much less important). 3

4 Far from being mutually exclusive, these various types of explanations for ballot splitting seem to be quite complementary. Using individual-level data from the 2014 PartiRep Belgian Voter Survey, we test these various hypotheses in the Belgian context, while controlling for some relevant sociodemographic and attitudinal characteristics of the voters. In many respects, the Belgian multilevel elections of May 2014 seem to provide an ideal setting for studying vertical split-ticket voting. First, these multilevel elections were considered as being an exceptional political event; for the first time since 1999 and only for the second time in the Belgian electoral history, citizens were invited to vote simultaneously for the three highest levels of government: the European, federal and regional levels 1. They could decide either to cast a straight-ticket vote in favour of a given party or to divide their votes between two or even three different parties across the different levels. Second, the 2014 Belgian elections took place in a quasi-experimental context in which both the political supply and the electoral system were the same for the three elections. Variations in the partisan supply or in the electoral regulation across the different electoral contests often make the study of split-ticket voting extremely difficult. In some countries, a large number of voters cannot cast a straight-ticket vote because of variations in the running parties across the different electoral arenas. They have no choice but to split their ballot, since their favourite party does not field candidates in each of the concurrent elections (Burden & Helmke, 2009). In other cases, split-ticket voting is induced by variations in the electoral system across the different contests. In the German Bundestag elections, for instance, many voters split their ticket because the Mixed Member electoral system leads them to cast a more strategic vote in the 1 Two small categories of voters were even invited to cast a fourth vote in addition to their federal, regional and European votes on 25 May These two categories are citizens from the small German-speaking Community (about 50,000 voters) and Dutch-speaking citizens from the Brussels Region (about 55,000 voters). Voters from the German-speaking Community had to take part in the elections that were held to choose the members of the Parliament of the German-speaking Community. Dutch-speaking voters from the Brussels Region had to participate in the electoral contest held to elect their representatives in the Parliament of the Flemish Community. 4

5 SMD contest and a more sincere vote in the PR contest (Pappi & Thurner, 2002). In the 2014 Belgian elections, none of these two problems occurred. Virtually all main parties fielded candidate lists at the three levels, with the sole exception being the Flemish right-wing populist party Libertair, Direct, Democratisch (LDD) that only competed in federal elections. Furthermore, in all three elections, the same electoral system was applied, namely a PR system with semi-open lists 2. It is also worth reminding that in Belgium, voting is compulsory for each type of election. Finally, a third interesting aspect of our case study lies in the fact that the Belgian political landscape is split into two separate party systems: the Dutch-speaking one and the Frenchspeaking one (Deschouwer, 2012). Indeed, Dutch-speaking parties only field candidate lists in Flanders and in Brussels, whereas French-speaking parties only field candidate lists in Wallonia and in Brussels 3. None of the main parties competes in both Flanders and Wallonia 4. In fact, Belgium contains two distinct political spectrums which deserve separate investigation. In the present study, we first examine the determinants of ticket splitting among all respondents included in the overall sample and then perform the same analysis separately for Flemish and Walloon respondents in order to compare the dynamics at work in both regions. Nevertheless, as far as our hypotheses are concerned, we do not expect to observe different results for Flanders and Wallonia. 2 The same electoral formula was used for each of the three elections, but it should be pointed out that the number of constituencies varied depending on the electoral contest. The Belgian territory was divided into 4 constituencies for the European elections, into 11 constituencies for the federal elections and into 19 constituencies for the regional elections. For more details about the Belgian electoral system, see: Deschouwer, Brussels is the only region in which parties of both language groups field candidate lists. Yet, this region is left out of our analysis, since Brussels citizens were not interviewed in the 2014 PartiRep Belgian Voter Survey. This panel survey was only conducted in the two other regions: Flanders and Wallonia. 4 More precisely, 12 out of the 13 political parties represented in the federal Parliament compete in only two of the three regions (Brussels and Wallonia for the French-speaking parties, Brussels and Flanders for the Dutchspeaking parties). The small radical left party PTB-PVDA is the only political force that fields candidate lists in all three regions of the country. 5

6 This article is divided into four sections. The first part is dedicated to our theoretical framework and reviews different types of potential explanations for ballot splitting. The second section presents the data and describes the operationalization of our dependent and independent variables. The third part deals with the results of our empirical study on the determinants of split-ticket voting at the 2014 concurrent federal, regional and EP elections. The final section contains some concluding remarks on the implications of our findings and some suggestions for further research. Theoretical framework: the determinants of split-ticket voting In this section, we consider five potential explanations of split-ticket voting: the absence of partisan attachments, voter s indecision, candidate-oriented voting, the policy balancing theory and the second-order national elections theory. In the electoral research literature, party identification is deemed as a key factor in explaining political attitudes and voting behaviours in advanced democracies. According to the researchers of the Michigan school (Campbell et al., 1960), partisanship provides the average citizen with an invaluable cue which enables him or her to deal with the complexity of politics, to interpret political stimuli and to make voting choices with a minimal cognitive effort. Campbell and his colleagues also stress the stabilizing effect of partisan affiliation on electoral behaviour, by showing that party identifiers almost invariably remain party loyalists voting in line with their partisanship. Party identification is thus thought to form a barrier against electoral volatility in general and ballot splitting in particular (Campbell et al., 1960). In their seminal study on the sources of split-ticket voting at the 1952 and 1956 American elections, Campbell and Miller (1957) found the strength of partisan attachment to be a good predictor of ballot splitting. The level of straight-ticket voting was the highest among strong partisans and somewhat lower 6

7 among weak party identifiers. By contrast, independents were much more inclined to split their ticket (Campbell & Miller, 1957). In line with these early findings, a large number of recent studies conducted in the United States as well as in other Western countries clearly demonstrate that partisanship increases the chance of casting a straight-ticket vote in concurrent elections (Beck et al., 1992; Crow, 2005; Dalton, 2013; Lachat, 2007; Rallings & Thrasher, 2003). Hypothesis 1: Party identification decreases the likelihood of voting for different parties at concurrent elections. Indecision is a second factor that may stimulate split-ticket voting. Over the last decades, most Western democracies have witnessed a dramatic growth in the proportion of voters who postpone their vote decision until the election campaign is under way (Cautrès & Jadot, 2009; Dalton, 2013; Dalton et al.; 2000; McAllister, 2002). This tendency to delay one s voting decision can be seen as a consequence of the partisan dealignment process. Since the percentage of voters who report partisan attachment is gradually decreasing, a growing group of citizens enter the campaign without a firm vote intention. This phenomenon of late decision making is claimed to induce split-ticket voting (Chaffee & Rimal, 1996; McAllister, 2002). While most pre-campaign deciders mainly rely on their long-term partisan affiliation to guide their electoral choice, late deciding voters are argued to be highly susceptible to short-term campaign stimuli such as issues, candidates traits or retrospective evaluations of government performances. Given that their voting choices are largely determined by short-term factors which are, by definition, variable and erratic, these last minute deciders usually display a high level of uncertainty and instability in their party preferences during the weeks preceding the elections (McAllister, 2002). When different elections are held simultaneously, late deciders tend to make their voting choice for each electoral contest separately on the basis of specific issues or on the basis of a candidate by candidate evaluation. Hence, they should be more prone than early deciders to split their ballot. In this respect, Chaffee and Rimal (1996) have shown 7

8 that last minute deciders were much more likely than pre-campaign deciders to divide their votes between different parties at simultaneous elections. Split-ticket voting becomes more probable the later the voter makes a decision for any one office (Chaffee & Rimal, 1996: 278). McAllister (2002) has also pointed at the existence of a strong relationship between timing of the voting choice and ticket splitting. Late deciders were found to be more inclined than early deciders to distribute their votes among different parties in concurrent elections (McAllister, 2002). Hypothesis 2: The later a voters takes his/her vote decision, the higher will be his/her probability of splitting his/her ticket at concurrent elections. Third, the willingness to support particular candidates has also been shown to be a key determinant of ballot splitting. When two elections take place simultaneously, voters can cast a vote for their favourite party in one election and for their favourite candidate(s) of another party in the other contest or even cast a ballot for their favourite candidate(s) in each contest regardless of the partisan label of candidates (Burden, 2009; Pappi & Thurner, 2002). Previous research has convincingly demonstrated that candidates characteristics could overwhelm partisan attachments and lead voters to split their ticket (Beck et al.; 1992, Burden, 2009; Campbell & Miller, 1957; Karp, 2009; Pappi & Thurner, 2002; Wattenberg, 1991). Interestingly, this relationship between candidate-oriented voting and ballot splitting has not only been observed in candidate-centred electoral systems such as the American SMD system or the mixed member systems used in Germany, Japan and New-Zealand (Beck et al., 1992; Burden, 2009; Karp, 2009; Pappi & Thurner, 2002). Candidates traits have also been shown to trigger ticket splitting in party-centred systems such as the Spanish PR system with closed party lists (Sanz, 2008). In this respect, it should be noted that the Belgian electoral system is located at the midpoint between candidate-centred systems and party-centred systems. Indeed, the electoral system used in Belgium at each level of government is a PR system with semi-open 8

9 lists. It allows voters either to cast a party list vote or to cast preference votes in favour of one or several individual candidate(s) within a single party list. In the Belgian context, the use of preference votes has previously been interpreted as a sign of candidate-centred voting (André et al., 2012). As candidate-oriented voting is assumed to induce ballot splitting, we hypothesize that voters who cast preference votes are more likely to split their ticket than are voters who endorse a party list. Hypothesis 3: Casting one or multiple preference vote(s) increases the likelihood of split-ticket voting. Voter s ideological profile is a fourth potential predictor of ticket splitting. In his influential book Divided Government, Fiorina (1996) proposes the policy balancing theory which holds that voters location in the ideological space is a key factor to understand split-ticket voting in the United States. According to this model, ticket splitters in the US presidential and congressional elections are centrist voters who seek to favour a divided electoral outcome, in such a manner that Democrats and Republicans would have to share the power (with one party holding the presidency and the other holding the congress), thereby producing a situation of policy balancing and mutual control (Alesina & Rosenthal, 1995; Fiorina, 1996). In other words, ballot splitters in the US would be moderate citizens who perceive Democrats as being too liberal and Republicans as being too conservative and who therefore vote simultaneously for candidates of both parties in order to obtain centrist policies associated with divided government. Following the policy balancing model, moderate voters who are located within the ideological space defined by the positions of the two main parties on the liberal-conservative axis would be more likely to split their ballot than would be voters who display a radical ideological profile and who are thus located outside the space between the two parties (Alesina & Rosenthal, 1995; Fiorina, 1996). As Sanz (2008) has shown, the model can be easily adapted to European countries. It only requires to replace the liberal-conservative dimension by the left- 9

10 right ideological axis which is more suitable to the classic structure of the European electoral competition. In the Belgian fragmented multiparty system, ideological spaces between the different political alternatives are relatively small, since a large majority of the main parties (i.e. those represented in the federal Parliament) display a quite moderate ideological profile and are located close to the centre (see: Deschouwer, 2012). Socialists and Greens position themselves to the centreleft, Christian-democrats to the centre and liberals to the centre-right. Belgian electors with moderate ideological preferences (i.e. those who place themselves close to the centre on the left-right scale) are therefore located within the small ideological spaces defined by the positions of the main political forces. Hence, they have a plethora of possible party choices and they should be inclined to split their votes between different parties at the concurrent federal, regional and European elections. By contrast, voters with a radical ideological profile (i.e. those who place themselves to the far left or to the far right) are located outside the ideological spaces between the main parties and most of the time, there is only one single party that is close to their position. Consequently, they should be less prone to split their ticket. Hypothesis 4: The more radical the ideological profile of a voter, the lower will be his/her probability of voting for different parties at concurrent elections. Finally, the second-order elections model developed by Reif and Schmitt (1980) might be highly valuable for explaining split-ticket voting in simultaneous elections across different tiers of government. This model primarily aims to account for differences observed between national elections outcomes on the one hand, and subnational and European elections results on the other hand. Reif and Schmitt distinguish between two kinds of electoral contests: first-order elections and second-order elections. First-order elections are the electoral contests whose outcomes determine the partisan composition of the national executive, namely the legislative elections in parliamentary regimes and the presidential elections in presidential regimes. The 10

11 category of second-order national elections covers all the electoral contests held at the other tiers of government, namely the European elections as well as the various types of subnational elections (Reif, 1985, Reif & Schmitt, 1980). According to the second-order elections hypothesis, the main characteristic of these subnational and supranational electoral contests lies in the fact that there is less at stake than in first-order national elections. This less-at-stake aspect is claimed to induce major consequences on participation and electoral results in second-order elections. Turnout would be lower than in first-order national elections; a large number of citizens would abstain from voting in subnational and supranational elections, since they consider these contests as not sufficiently important to go to the polls. Regarding election results, parties in the national government would lose electoral support compared to previous first-order national elections, whereas small opposition parties would garner higher vote shares in comparison to previous first-order elections (Reif & Schmitt, 1980). Reif and Schmitt suggest that these characteristics of second-order elections outcomes can be explained by two mechanisms operating at the individual level. The first explanation, which accounts for government parties losses, is that voters use the subnational and supranational elections to voice their dissatisfaction with their national government (Reif, 1985; Reif & Schmitt, 1980; see also Hix & Marsh, 2007; Marsh, 1998). They try to send a signal to parties in the national executive in order to encourage the government to change its political line. In the 2014 Belgian context, however, the simultaneity of the three elections has largely neutralized the impact of this mechanism on ballot splitting. Indeed, it was not worth using the regional and European votes to send a message to the federal government, since the federal elections took place on the same day. The second mechanism accounting for differences between national, subnational and supranational elections posits that voters behave more strategically in first-order elections and more sincerely in second-order elections (Reif & Schmitt, 1980; see: also Hix & Marsh, 2007; Marsh, 1998). In first-order national contests, 11

12 most citizens tend to vote strategically in order to avoid wasting their votes. When their favourite party seems to have no real chance of winning, they may refrain to vote for that small party and decide to cast a vote for a bigger party which corresponds to their second preference. To the extent that there is less at stake in second-order elections, electors care less about what party will win. As a consequence, their voting choice is less constrained by strategic considerations and they may cast a more sincere vote, which favours small parties. In sum, the second-order elections hypothesis is based on the assumption that voters vote differently in national, subnational and supranational elections, because they perceive a differential in importance between the levels of government. In our analysis, we will test the validity of this assumption by investigating whether or not voters who perceive a large differential in importance between the levels are more likely to split their ticket in the 2014 Belgian concurrent multilevel elections. Hypothesis 5: The larger the perceived differential in importance between the tiers of government, the higher will be the probability of split-ticket voting. Data and operationalization This research draws on the data from the 2014 PartiRep Belgian Voter Survey which was conducted among a geographically stratified random sample of eligible voters in Flanders and Wallonia. This survey had a panel design; it consisted of a pre-electoral wave and a postelectoral wave. In the pre-electoral wave, respondents were interviewed face-to-face about their personal background, opinions, interests, political activities and vote intentions. The postelectoral wave, by contrast, consisted in phone interviews whereby respondents were asked about their voting behaviours, their vote motives and the election campaign. While the pre- 12

13 electoral interview phase was completed by 2,019 respondents, only 1,528 respondents participated in the post-electoral wave. In the present study, the dependent variable is split-ticket voting. This variable is operationalized by means of respondents reported voting behaviours in the three concurrent elections. For each respondent, we compare party choices at the three levels. In the 2014 Belgian multilevel elections, voters could decide either to cast a straight-ticket vote in favour of a given party or to split their ballot in one of the following four ways: to vote for a party A in federal and regional elections and for a party B in European elections (F = R E); to support a party A in the federal contest and a party B in the regional and European contests (F R = E); to endorse a party A at the federal and European levels and a party B at the regional level (F = E R); to vote for a party A in federal elections, a party B in regional elections and a party C in European elections (split between three parties). The dependent variable is dichotomous; straight-ticket voters are coded 0, whereas those respondents who opted for one of the four types of ballot splitting are coded 1. Table 1 shows that 34.5 % of the respondents report having divided their votes between different parties at the 2014 simultaneous elections. Interestingly, the largest group of ballot splitters (14.8 % of the overall sample) consists of those electors who voted for a party A in both federal and regional elections and for another party in European elections. It is also worth noting that only a very small proportion of respondents (4.5 %) indicate having made three distinct party choices at the three tiers of government. As evident from the table, the level of split-ticket voting proves to be slightly higher in Flanders than in Wallonia, but generally speaking, differences between both regions are relatively small. 13

14 Table 1. Levels of split-ticket voting per modality Total (Full sample) Straightticket F = R E F = E R F R = E Split between three parties Total 65.5% 14.8% 8.4% 6.8% 4.5% 100% Wallonia 66.3% 14.7% 9.0% 7.2% 2.8% 100% Flanders 64.9% 14.9% 7.9% 6.5% 5.8% 100% The overall amount of ticket splitting in the 2014 elections appears to be very high, especially when we compare it to figures observed in the 2009 simultaneous regional and European elections. In 2009, about one fifth (21%) of the PartiRep 2009 survey respondents voted differently in the regional and European contests (Van Aelst & Lefevere, 2011), but this lower level of ballot splitting can be partially explained by the fact that there were only two (and not three) concurrent elections. In fact, the 2014 level of split-ticket voting should be rather compared to that observed during the 1999 elections which were also exceptional elections, with federal, regional and EP contests taking place simultaneously. In 1999, the proportion of ticket splitters among the then respondents was 29% (Maddens & Hajnal, 2002), which is quite comparable to the 2014 numbers. From a cross-national comparative perspective, Belgian figures seem to be somewhat higher than those observed in other European countries such as Germany or the United Kingdom (Heath et al., 1999; Pappi & Thurner, 2002; Rallings & Thrasher, 2003). Nevertheless, such a comparison is far from being straightforward, since the level of ticket splitting depends on a series of factors such as the number of concurrent elections, the types of elections or the number of parties competing in the different contests. In the remaining of this section, we describe the operationalization of the independent variables. Hypothesis 1 holds that party identification decreases the chance of voting for different parties 14

15 at concurrent elections. The variable party identification takes the value 1 for respondents who indicate feeling close to a particular party and the value 0 for other voters. According to hypothesis 2, late decision making could also be one of the possible causes of ballot splitting. The survey questionnaire enables us to measure timing of the voting choice, as it included the following question: When did you decide to vote for that party at the federal elections? Three options were offered to respondents: before the start of the campaign (coded 0); during the campaign (coded 1); on Election Day (coded 2). Another type of explanation for split-ticket voting lies in the willingness to vote for (a) particular candidate(s) rather than for a party. Hypothesis 3 holds that some voters may decide to cast a vote in one of the three concurrent elections for a party other than the one they usually vote for because they are attracted to (a) particular candidate(s). In order to test this hypothesis, we examine whether casting (a) preference vote(s) for one or several individual candidate(s) increases the probability of splitting ones ballot. For each of the three elections, we thus introduce a dichotomous variable recording whether the voter cast a party list vote (coded 0) or whether he/she cast (a) preference vote(s) in favour of one or several candidate(s) (coded 1). Voter s degree of ideological extremeness is another potential predictor of ballot splitting. According to the policy-balancing hypothesis (H4), electors with a radical ideological profile should be less likely than moderate voters to divide their votes between different parties in simultaneous elections. The degree of ideological extremeness is measured on the basis of the respondent s left-right self-placement on an 11-point scale where 0 means The left and 10 means The right. For each respondent, we calculate the distance between his/her self-reported position on the left-right axis and the ideological centre (i.e. the value 5 on that axis). Hence, the variable Degree of ideological extremeness takes values ranging from 0 (when 5 was reported) to 5 (when 0 or 10 was reported). 15

16 The last hypothesis (H5) suggests that some voters split their ballot at the 2014 simultaneous federal, regional and EP elections, because they perceived a differential in importance between these three electoral contests. The PartiRep survey does not allow us to directly measure this differential since respondents were not asked to indicate how important each of these electoral contests was for them. Consequently, we use two proxies to assess the perceived differential in importance between the elections. The first proxy is the variable differential in the intention to participate in the various types of elections. A major indicator of the importance given by a voter to a particular electoral contest is of course his/her participation in this election, but in the Belgian context marked by compulsory voting, this indicator has little significance. In the 2014 elections, turnout was very high and quite consistent across the three electoral contests 5. Nevertheless, the PartiRep survey contained the following question: If voting was not compulsory anymore, how often would you take part in the various types of elections? For each type of election, respondents were asked to give a value on a 4-point scale (from 0 to 3) where 0 means Never, 1 Sometimes, 2 Most of the time and 3 Always. For each respondent, we calculate the mean value for the three types of elections (i.e. the average hypothetical frequency of participation in the various types of elections) as well as the standard deviation. We interpret this standard deviation as an indicator measuring the differential in the hypothetical intention to participate in the three types of elections. Respondents with a high differential are those who indicate that they would take part in one or two type(s) of elections much more often than they would participate in the other type(s) of elections. The second proxy for assessing the perceived differential in importance between the three types of elections is the variable perceived differential in political influence between the three levels 5 Turnout was 89.4% for the federal elections, 89.7% for the regional elections, and 89.6% for the European elections. 16

17 of government. This variable is constructed on the basis of the answers to the following question: For each of the following institutions (the regional authority, the federal authority and the European Union), could you tell us to what extent it exerts an influence on your daily life? For each of the three tiers of government, respondents gave a value on a 4-point scale where 0 means No influence at all, 1 A not very strong influence, 2 A fairly strong influence and 3 A very strong influence. For each respondent, we calculate the mean value for the three levels of government, which represents the average perceived degree of influence on daily life exerted by these three levels. On the basis of this mean value, we then calculate the standard deviation which can be taken as a measure of the perceived differential in political influence on daily life between the three tiers of government. Furthermore, we include in our models a series of control variables 6. First, we control for gender. Although Campbell and Miller (1957) found men to be more prone than women to split their ticket, most studies have demonstrated that gender usually had no significant impact on the probability of split-ticket voting (e.g. Beck et al., 1992; Rallings & Thrasher, 2003). Our second control variable, age, is claimed to be a strong predictor of ballot splitting. Young voters would be considerably more inclined than their elders to divide their votes between different parties in simultaneous elections (Campbell & Miller, 1957; De Vries & Tarrance, 1972). Third, we include the level of education. Most scholars consider the level of education as a major determinant of split-ticket voting. Yet, they disagree on the question of whether a high level of education increases or reduces the chance of splitting one s ballot. In their seminal study, Campbell and Miller (1957) showed that ticket-splitters were more likely to be drawn from the ranks of less educated citizens. By contrast, more recent studies have pointed at the existence of a positive relationship between the level of education and ballot splitting, with highly educated voters displaying the highest probability of voting for different parties in concurrent 6 For coding details of the control variables, see: appendix. 17

18 elections (Beck et al., 1992; Dalton, 2013; De Vries & Tarrance, 1972). The reason therefor would be that well-educated citizens would not have to rely on parties as a heuristic shortcut to guide their electoral decisions and that they would thus tend to make their voting choice for each election separately on the basis of a candidate by candidate evaluation or on the basis of specific issues. Besides these sociodemographic characteristics, we control for the level of political sophistication which is often argued to be related to ticket splitting (Dalton, 2013; Lachat, 2007). We include two distinct indicators of political sophistication: the level of political interest and the degree of campaign attention. Previous research has shown that the level of ballot splitting was higher among voters with a high degree of political interest (Dalton, 2013; Maddox & Nimmo, 1981). Results To the extent that our dependent variable is dichotomous, we perform binary logistic regressions in order to test our hypotheses. First, we regress ballot splitting on the abovementioned independent and control variables among all respondents included in our overall sample. Table 2 presents the results of this multivariate analysis. 18

19 Table 2. The determinants of split-ticket voting (binary logistic regression) for the overall sample DV: split-ticket voting B S.E. Exp(B) Gender Age Level of education 0.343*** Political interest Campaign attention Party identification ** Late decision-making 0.849*** Preferential voting in federal elections * Preferential voting in European elections 0.602*** Preferential voting in regional elections Perceived differential in political influence on daily life between the levels 0.367* Differential in the intention to participate in the various types of elections Ideological extremeness Constant *** Nagelkerke R² = Sign.: # p<0.1 * p < 0.05, ** p < 0.01, *** p < N= 1,383 At first glance, it appears that most coefficients are in the expected direction. Our results provide strong empirical evidence in support of the hypothesis that party identification decreases the probability of split-ticket voting (H1). Voters who report no partisan attachment are more inclined than party identifiers to distribute their votes among different parties at concurrent elections. Additionally, regression results demonstrate the presence of a statistically significant relationship between timing of the voting choice and ticket splitting, thereby confirming the validity of hypothesis 2. The later a voter takes his/her electoral decision, the higher will be his/her probability of voting for different parties at simultaneous elections. This finding clearly indicates that voter s indecision can be considered as one of the key determinants of split-ticket voting. The hypothesis that casting preference votes increases the likelihood of ballot splitting (H3) can only be partially confirmed, because our three indicators of preferential voting have different 19

20 types of effects that somewhat contradict each other. In line with our expectations, the use of preferential voting in European elections is significantly and positively related to ballot splitting. Electors who cast one or multiple preference vote(s) at the European level are more likely to divide their votes between different parties than are those who cast a party list vote at that level. It seems that the popularity of several high-profile politicians who were candidates for the EP elections (e.g. Guy Verhofstadt, Marianne Thyssen and Louis Michel) has led some Belgian voters to cross party lines and to split their ticket at the 2014 concurrent multilevel elections. Interestingly, casting preference votes in the regional elections also has a positive effect on split-ticket voting, but this effect does not reach the conventional level of statistical significance. By contrast, the use of preferential voting in the federal elections is found to be significantly and negatively associated with ticket splitting, which is contrary to our hypothesis. Voters who cast one or several preference vote(s) in favour of individual candidate(s) at the federal level are less likely to distribute their votes among different parties than are those who endorse a party list at that level. On the basis of our analysis, we cannot provide a plausible explanation for why the effect of preferential voting is positive in the European elections, but negative in the federal contest. Because of these paradoxical findings, we may not conclude that candidate-oriented voting always stimulates ticket splitting. The complex relationship between preferential voting and ballot splitting in the Belgian context undoubtedly deserves further investigation. As can be read from table 2, our results lend no support for the policy-balancing hypothesis (H4) which suggests that voters with a radical ideological profile are less prone than moderate voters to split their ballot between different parties. Admittedly, the effect of ideological extremeness is in the expected negative direction, but it is far from achieving statistical significance. 20

21 Besides, the analysis partially confirms the second-order elections hypothesis (H5) which argues that some electors support different parties in concurrent national, subnational and supranational elections because they perceive a differential in importance between the tiers of government. We observe that voters who perceive a large differential in political influence (on their daily life) between the levels of government are more likely to vote for different parties in the concurrent federal, regional and European elections. It must be pointed out that the indicator perceived differential in political influence between the levels does not enable us to determine which tier of government is considered as the most influential one and which tier is seen as the least influential one. This indicator only allows us to investigate whether electors perceive a large differential in influence between the levels. Yet, for the purpose of this study, the mere observation that those voters who do perceive such a differential are more likely to split their ticket is sufficient to give credit to the second-order elections hypothesis (H5). This hypothesis, however, cannot be totally confirmed, since the other indicator designed to measure the perceived differential in importance between the levels, namely the differential in the intention to participate in the various types of elections, is not significantly associated with split-ticket voting. Regarding the control variables, our results show that gender has no significant impact on ballot splitting. More surprisingly, we find no statistically significant relationship between age and ticket splitting, while a large number of studies have demonstrated that young voters were more prone than their elders to support different parties in concurrent elections. In fact, the level of education proves to be the only sociodemographic characteristic that significantly influences the probability of split-ticket voting. As the positive coefficient indicates, well-educated citizens are more inclined than less educated ones to divide their votes between different parties in concurrent elections. This finding does contradict the conclusions of pioneer work (e.g. Campbell & Miller, 1957), but is consonant with the results of more recent studies (e.g. Beck 21

22 et al., 1992). Finally, it is worth noting that none of the two indicators of political sophistication, namely political interest and campaign attention, has a significant effect on ticket splitting. After having examined the determinants of ballot splitting among all electors interviewed in the 2014 PartiRep survey, we then perform the same multivariate analysis by dividing the overall sample into two groups: Flemish respondents and Walloon respondents. Since the Belgian political landscape is totally split into two separate party systems (the Dutch-speaking one and the French-speaking one) whose respective electoral dynamics differ fundamentally from each other, it is worth analysing the sources of ticket splitting separately in Flanders and in Wallonia. Table 3 reports the results of our analyses for Flemish and Walloon respondents. Table 3. The determinants of split-ticket voting (binary logistic regressions) for Flanders and Wallonia Flanders Wallonia DV : split-ticket voting B S.E. Exp(B) B S.E. Exp(B) Gender Age Level of education 0.377** * Political interest Campaign attention Party identification *** Late decision-making 0.836*** *** Preferential voting in federal elections # Preferential voting in European elections 0.900*** ,247 Preferential voting in regional elections Perceived differential in political influence on daily life between the levels 0.567* Differential in the intention to participate in the various types of elections Ideological extremeness Constant *** Nagelkerke R² N Sign.: # p<0.1 * p < 0.05, ** p < 0.01, *** p <

23 Regression results clearly show that the mechanisms accounting for ticket splitting in the Flemish subsample display many similarities with the mechanisms underlying ballot splitting in the overall sample. In line with our hypothesis 1, Flemish ballot-splitters are more likely to be drawn from the ranks of voters who report no partisan attachment. Our results also lend support for hypothesis 2, as they indicate that timing of the voting choice significantly influences ticket splitting. Flemish late deciding voters are more inclined than pre-campaign deciders to distribute their votes among different parties at concurrent elections. By contrast, hypothesis 3 can only be partially confirmed, because the effects of the three indicators of preferential voting differ from each other. As expected, casting preference votes for individual candidates in the European elections significantly increases the chance of voting for different parties at the 2014 simultaneous multilevel elections. At the same time, the use of preferential voting in the Flemish regional elections negatively correlates with ballot splitting, but this relationship does not reach the conventional level of statistical significance. Just as we have noticed for the overall sample (see above), preferential voting in the federal elections has a negative impact on ballot splitting among Flemish respondents, but contrary to what holds for the overall sample, this impact is not statistically significant. As evident from the table, the policy-balancing hypothesis (H4) must be rejected, since the degree of ideological extremeness is not significantly related to split-ticket voting. Our last hypothesis (H5), which is drawn from the second-order elections theory, can be partially confirmed; while the indicator differential in the intention to participate in the various types of elections does not achieve statistical significance, the variable perceived differential in political influence between the levels proves to be a reliable predictor of ticket splitting in Flanders. Flemish voters who perceive a large differential in political influence between the three tiers of government are more prone than other voters to divide their votes between different parties at the concurrent federal, regional and EP elections. 23

24 As far as the control variables are concerned, our results show that only the level of education significantly correlates with ticket splitting in Flanders. The higher the level of education, the higher will be the probability of endorsing different parties at simultaneous elections. Just like in the overall sample, the other control variables (gender, age, political interest and campaign attention) have no significant effect on split-ticket voting in the Flemish subsample. Finally, we examine the determinants of split-ticket voting in Wallonia. To put it bluntly, regression results reveal virtually no statistically significant relationship between our independent variables and ticket splitting in the Walloon subsample. Hypothesis 2 is the only hypothesis that receives empirical support among Walloon respondents; results demonstrate the existence of a strong link between timing of the electoral decision and ballot splitting. The later a voter makes up his/her mind, the higher will be his/her likelihood of voting for different parties at the concurrent elections. None of the other hypotheses can be confirmed in the Walloon case, since the various independent variables (party identification, preferential voting, perceived differential in political influence on daily life between the levels and differential in the intention to participate in the various types of elections) do not reach an acceptable level of statistical significance. Nevertheless, it should be noted that the effect of preferential voting in federal elections almost achieves statistical significance. But unlike what we expected, it is voting for a party list rather than for individual candidate(s) that triggers ballot splitting. Concerning the control variables, it appears that the level of education has a significant impact on ballot splitting. As in Flanders, in Wallonia as well, highly educated voters are more inclined than less educated citizens to divide their votes between different parties at simultaneous elections. The four other control variables (age, gender, political interest and campaign attention) are not significantly related to ballot splitting. 24