Stefanie Beyens, Vrije Universiteit Brussel Tom Verthé, Vrije Universiteit Brussel

The Impact of Perceived Coalition Potential on Vote Intentions for Parties in Proportional Electoral Systems. Belgian Voters at the 2014 Federal Elections. Stefanie Beyens, Vrije Universiteit Brussel (sbeyens@vub.ac.be) Tom Verthé, Vrije Universiteit Brussel (tverthe@vub.ac.be) Paper prepared for presentation at the ECPR General Conference, Glasgow, 3-6 September 2014 (This research paper is in a very preliminary stage, please contact the authors for a more recent version before citing) Abstract The decline of strong linkages in advanced industrial democracies between voters and parties has resulted in increased levels of volatility. Voters are less loyal to one party and party switching from one election to another has become more common. One of the ways in which this party switching can be explained is through strategic voting. But what reasons do voters have to act strategically in a proportional electoral system like the Belgian one? There are currently 6 parties that together make up the Belgian federal government. Yet, no Belgian voter is ever in a position to vote strategically in the conventional sense of the word (keeping a less-preferred party out): coalition formation is a party prerogative. However, voters do have expectations about the chances parties have to end up in the governing coalition and might behave accordingly in the voting booth. In this paper we use the 2014 PartiRep voter survey data to explore the part of strategic coalition voting in insincere voting and to determine the effect of perceived coalition potential on vote choice. We show that a substantial part of voters with one distinct party preference intends to vote insincerely. The majority of this group reports a higher coalition potential for the party they intend to vote for with respect to the party they generally prefer. This points towards coalition potential as a culprit for a large part of insincere voting behaviour. We then tried to determine whether coalition potential, as part of a vote utility variable, helps in better predicting voting behaviour than models that include general preference. Even though this is not true for all parties, the general assumption that coalition potential plays a significant part in determining vote choice does seem to hold. 1

Introduction The current literature on strategic voting has a strong bias towards plurality electoral systems, which is not entirely justified (see, a.o. Cox 1997: 31). Scholars have started to shed light on strategic considerations in proportional multiparty systems (a.o. Alvarez and Nagler 2000; Alvarez, Boehmke, Nagler 2004; Blais, Young, Rucotte 2005), but quite some work remains to be done. We believe that overlooking proportional systems has been a mistake that needs to be rectified, since the main reason for doing so has been based on the grounds of the difficulty of modelling voting behaviour in these systems and not the absence of incentives for strategic voting per se. In fact, one could argue that there are more incentives for strategic voting in a proportional multiparty electoral context, as long as some of the voters believe that they can exert influence on the election outcome in such a complex environment. We put forward that the strongest underlying dynamic for strategic voting remains the wasted- vote logic, even in a proportional context. Indeed, Abramson et al (2010) have found that there is more strategic voting under proportional representation (PR) rules than under first- past- the- post (FPTP) rules, especially because there are more (small) parties in PR settings. However, this wasted- vote logic does not limit itself to the electoral threshold and the predicament of small parties trying to get represented in the legislature. In most proportional multiparty systems coalition formation is the norm rather than the exception. This means that being in or out of the government coalition is a question that shapes the behaviour of parties and voters alike and potentially affects a much larger share of the voting population than those voters whose preferred party faces a struggle for survival. Casting a vote for a party that you perceive to have little chance of ending up in government, even though it has ample opportunities to conquer seats, might also be interpreted as a wasted vote (see also Bargsted and Kedar 2009). In this paper we try to determine to which extent coalition voting as a subset of strategic voting can be used to explain vote choice. Theory The fact that strategic voting is being portrayed as a subset of insincere voting follows from the definitions of the concepts themselves. Insincere voting is commonly defined as voting in a way that does not reflect one s true preference order (Merrill, 1984), while strategic voting is (a) a vote for a party/candidate which is not the preferred one (b) with the purpose of influencing the election outcome (Blais, Nadeau, Gidengil, Nevitte 2001). In order to achieve this, the voter should not vote for the party/candidate of his first preference and, additionally, the voter will have made this choice based on his perception of the anticipated election results. These results are not mere numbers. Party size is a tool for parties to exert influence, to translate votes into policy. This holds both for parties that 2

struggle to overcome the electoral threshold as well as for parties that want to be part of a coalition government. In this paper, we will deal with coalition potential. Our main goal is to ascertain the impact of perceived coalition potential on voting behaviour, but that perception is totally up to the individual voter. If electoral thresholds do enter into the argument, this will happen at the voter level. We do not look at party size per se, but we cannot exclude (and we even hypothesize the contrary) that voters do equal expected party size to coalition potential in the case of small parties. We argue here that strategic voters take account of the perceived election outcome in the sense that they envisage a viable coalition of parties, which does not always coincide with their assumed goal of wishing as many votes as possible for their preferred party. The strategic coalition voters are thus faced with a highly complex task. Now, this does not mean we should conclude that only political scientists with an intricate understanding of electoral systems and government formation could exhibit strategic coalition voting. After all, voters do not have to be able to predict the election outcome in order to act strategically; they just have to believe their predictions to be somewhat accurate (Gschwend 2007). Using the rather moralising term insincere voting creates the false picture of incongruence. However, the term incongruence is chosen deliberately here because it refers to how voters preferences are translated into those of parties- in- power. In proportional electoral systems, the next government s policy is not decided in the voting booth by voters; rather, it is decided by political parties negotiating the next governing coalition, both in terms of who gets a seat at the table and in terms of how much clout they have. Simply put, one needs to control the executive in order to control policy setting. Voters are aware of this and, if they are motivated first and foremost by issues and policy, they will take this into account when in the voting booth. One could argue whether it is really rational to do so, but strategic voting (in all of its facets) reflects the desire of voters to (maximally) influence the election outcome and not waste their votes. It is this wasted- vote logic that drives both those voters that do not want to waste it on parties that might be too small as those that do not want to waste it on an opposition party. It is Realpolitik in action. These voters may indeed be behaving insincerely, but not incongruently. They are staying true to their policy preferences, but also take into account that their vote needs to be translated into policy and that the most secure way of doing this is to bet on the horse that can actually cross the finish line, i.e. partake in the executive. Strategic voting is at least partly an expression of risk- aversion on the part of the voter (Slinko & White, 2010). The fact that voters do vote strategically in proportional multiparty systems does not, however, deny the reality that the complexity of determining when and to what extent a party has coalition potential is higher than guessing the odds of a party crossing the electoral threshold. This is especially the case in Belgium, where parties from both sides of 3

the linguistic divide have to come together on a coalition agreement. This has resulted in the last government being composed of no less than six parties. It can also be argued that this exercise has become even more difficult ever since a new Flemish- nationalist party became a major player (2010): this party does not have a logical mirror image in its party family on the other side of the language border. This complexity is what makes Belgium such an interesting case to study strategic coalition voting. This paper will continue the literature s discussion on strategic coalition voting by taking a first look at the PartiRep II voter survey for the very recent federal Belgian elections (25 May 2014). First, we draw a picture of insincere voting (intentions) and coalition potential. Then we briefly consider an alternative conceptualisation of strategic coalition voting. We finish this exploratory journey with some logistic models to look at how (perceived) coalition potential affects (intended) vote choice. Data: the 2014 PartiRep Voter Survey The data we will be using in this paper is the pre- electoral wave (CAPI) of the 2014 PartiRep Voter Survey. PartiRep 1 is an interuniversity consortium that consists of the Vrije Universiteit Brussel, Université Libre de Bruxelles, KU Leuven, Université Catholique de Louvain, Universiteit Antwerpen, Universiteit Leiden and Universität Mannheim (MZES). The dataset we are using is the earliest release (beta version), has not completely been cleaned and does not yet contain any weights. Therefore any results we discuss have to be seen as a first rough attempt which will have to be completed and refined when better data and the data of the post- electoral wave (CATI) become available. The survey has been conducted by TNS- Dimarso on a representative sample of the Belgian population (drawn from the national registry) of roughly 4000. The response rate was about 50%, which has resulted in a sample of 2000 voters, about 1000 in both Flanders and Wallonia. The Brussels region was not part of the sample. At the moment we lack more detailed meta- data on the sample. This information will, of course, be made available as soon as possible. Insincere voting - a first look We asked 976 Flemish and 1005 Walloon respondents for their assessments on Flemish parties - the only parties to actually appear on the Flemish ballots, even in elections for the federal parliament. We first asked respondents to say, on a 11- point like- dislike scale, how much sympathy they have for each party. 2 From these answers we determined which party 1 www.partirep.eu 2 There are many political parties in our country. I would like to hear from you how much you like these parties. You can give these parties a score out of 10; 0 means you don t like the party at all; 10 means you like the party very much. The better you like a party, the higher the score you give it. If you don t know a party 4

was a respondent s preferred party. 644 Flemish and 674 Walloon respondents indicated to have a preferred party by giving one party a higher score than any of the others; 189 Flemish and 151 Walloon respondents gave two parties the same highest score; 141 Flemings and 180 Walloons gave more than two parties the same highest score. In this exploratory part of the paper we only concern ourselves with those 644 Flemish and 674 Walloon respondents who indicate they have a preferred party. The reason we disregard voters who have the same levels of sympathy for more than one party is to be found in the definition of strategic voting. Foregoing a preferred party/candidate vote in order to influence an election outcome implies that strategic voters in fact have a clear preference for one party (or candidate). A next stage of the research project for which this paper is the first step, we will pay special attention to those voters with tied preference scores. It would be very interesting to ascertain the motives behind the ultimate choice these voters make. Table 1. Party preference(s) and sincere voting - Flanders Voters with one party preference 644 of which intend to vote for that party 495 of which do not intend to vote for that party 149 Voters with a tied party preference 189 Voters with more than 2 preferences 141 TOTAL 974 Table 2. Party preference(s) and sincere voting - Wallonia Voters with one party preference 674 of which intend to vote for that party 507 of which do not intend to vote for that party 167 Voters with a tied party preference 151 Voters with more than 2 preferences 180 TOTAL 1005 well enough to answer, feel free to mention this. Er zijn veel politieke partijen in ons land. Ik zou graag van u weten hoe sympathiek u deze partijen vindt. U kunt iedere partij hiervoor punten geven tussen de 0 en de 10, waarbij 0 helemaal niet sympathiek is, en 10 heel sympathiek. Hoe sympathieker u een partij vindt, hoe meer punten u die partij geeft. Als u een partij niet of onvoldoende kent, zegt u het dan gerust. 2 5

In a separate question we asked respondents which party they intended to vote for. After all, the assumption that vote intention and preferred party are one and the same completely ignores the possibility of voters making a strategic choice. As we mentioned earlier, one of the conditions for strategic voting is the occurrence of an insincere vote, indicated by a discrepancy between the respondent s stated voting intention and the preferred party we assigned to her on the basis of her like/dislike party scores. From the 644 Flemish respondents with a preferred party, 149 intended to vote insincerely, meaning that they would choose a party which they did not give the highest score to on the like/dislike scale (23.1%). For the Walloons, 167 respondents intended to vote insincerely (24,8%). We do not want to interpret this as strategic voting yet, because we have no way of determining the motives behind this move. However, as we have indicated, we have reasons to expect that some voters go down this road because they want to influence the outcome of the election in terms of government formation. We explore whether these insincere voters are actually casting a strategic coalition vote. Respondents were asked to assign each party a score on coalition potential (0 none 10 a lot). 3 For each respondent we then coupled both the preferred party and the party of voting intention with this perceived coalition potential and compared the two. Table 3. Coalition potential and sincere voting - Flanders Coal. Pot. vote intention > Coal. Pot. preferred party 53 Coal. Pot. vote intention < Coal. Pot. preferred party 26 Coal. Pot. vote intention = Coal. Pot. preferred party 16 TOTAL 95 Table 4. Coalition potential and sincere voting - Wallonia Coal. Pot. vote intention > Coal. Pot. preferred party 46 Coal. Pot. vote intention < Coal. Pot. preferred party 19 Coal. Pot. vote intention = Coal. Pot. preferred party 11 TOTAL 76 3 This question asks the voter to state how likely they think it is that each party will be part of the next federal government coalition on a 0-10 scale (0 being very unlikely and 10 being very likely): Could you for each of these parties indicate on a scale of 0 to 10 how likely you think it is they will be a part of the next federal government? 0 means you think it is highly unlikely; 10 means you think it is highly likely they well be part of the government. You can make your answer more nuance by indicating the scores in- between 0 an 10. Kan u voor elk van de volgende partijen zeggen op een schaal van 0 tot 10 hoe waarschijnlijk u het acht dat ze deel zullen uitmaken van de volgende Belgische federale regering, waarbij 0 totaal niet waarschijnlijk betekent en 10 heel waarschijnlijk? Met de tussenliggende scores kan u uw antwoord nuanceren. 6

A first look at the combined occurrence of insincere voting and a chosen party s coalition potential does indeed suggest we are seeing strategic coalition voting in action. More than half of respondents, both in Flanders and Wallonia, who intend not to vote for their preferred party assigned their voting intention party a higher coalition potential than their preferred party. This is an interesting find and we analyse the role of coalition potential in vote choice more in detail later in the paper. While it would be nice to delve deeper into this trend here, we have to face an all too familiar problem for researchers interested in strategic voting, and that is low numbers. This issue prompts us to try and operationalize our concept of insincere voting in another way in order to explore the impact of perceived coalition potential on vote choice. Insincere voting versus having two vote intentions In previous research on strategic voting (Beyens & Verthé 2013), we have operationalized the concept differently and asked voters if they could imagine voting for another party, in other words, do they have a second vote intention. The 2014 PartiRep voter survey asks the same thing, which enables us to compare the occurrence of a hypothetical second vote intention with insincere voting, which we defined as not voting for a preferred party. Table 5. Preferences and vote intentions - Flanders Stating 2 vote intentions Insincere voters no yes TOTAL no 184 311 495 yes 73 76 149 TOTAL 257 387 644 Table 6. Preferences and vote intentions - Wallonia Stating 2 vote intentions Insincere voters no yes TOTAL no 219 288 507 yes 96 71 167 TOTAL 315 359 674 7

Remember that we conceptualised insincere voters as those respondents who have a preferred party (based on highest score in like/dislike scale), but who do not vote for that party. We would expect insincere voters to belong to the group who state that they have a second vote intention. Yet, the table shows a not insignificant number of voters with only one vote intention while they report to like another specific party more than the one they intend to vote for. 73 Flemish voters do this, which is 48.99% of all Flemish insincere voters who answered the second vote intention question. 96 Walloon voters do the same, which is a staggering 57,49% of all Walloon insincere voters. This calls for a more in- depth analysis of vote choice, in general, and the role played by like/dislike scores and coalition potential, in particular. Explaining vote choice: comparing logit models Our ultimate goal is to ascertain to what extent voters coalition perceptions can account for strategic - and thus insincere - voting. One way of doing this, without only focusing on the group of insincere voters, is to determine the extent to which coalition perception influences vote choice in the overall voting population. If we compare vote choice models (per party) without coalition perceptions and models with coalition perceptions, we should be able to determine whether a model that takes the voter s perceived coalition potential for a party into account is indeed better able to predict intended vote choice 4 for that party. We use intended vote choice because the available data comes from the pre- electoral wave of the voter survey and we do not yet know the actual (reported) vote choice. We first need to explain how we will include this perceived coalition potential. Merely including it as an independent variable would be useless, since there is no reason why we would assume that coalition potential by itself would be an added independent predictor of intended vote choice. Theoretically strategic voters are utility maximizers. This means that it is a function of how much you like a party and how probable you think it is that this party will achieve representation or membership of the government coalition. In this paper we only look at coalition voting, and thus only the latter interpretation of utility will be explicitly considered. We define this utility as the multiplication of the score on the like/dislike question 5 and the score on the coalition potential question 6 for each party 4 If the federal elections were today, which party would you vote for? Als er vandaag al verkiezingen voor de Kamer zouden zijn, op welke van de volgende partijen zou u dan stemmen? 5 See note 2 for an English translation of the question we asked in the Dutch- language en French- language questionnaires. 8

separately. Both of these questions have been conceived as 11- point scales. Our hypothesis is that the vote choice model with our utility variable will predict intended vote choice better than the model with the like/dislike variable (both as independent variables). We also expect that liking a party very much in combination with thinking it has a high coalition potential will exponentially increase the chances of voting for that party. In order to run our logistic regression models, we could only include those respondents that answered the questions that make up our independent variables. Our sample was thus reduced from 976 to 679 Flemish voters and from 1005 to 567 Walloon voters. This reduction is put in perspective if we look at the questions: the respondents were asked two questions, but had to give each individual party a score. We believe this goes some way in explaining the non- response for these items. For the federal elections in Flanders, there were 8 main competing parties. These were: Christian democrats (CD&V), Greens (GROEN), New- Flemish Alliance (N- VA, regionalists), social democrats (SP.A), Flemish Interest (Vlaams Belang, extreme- right), liberal democrats (Open VLD), libertarians (LDD) and Labour Party (PVDA+, extreme- left). The last two parties, LDD and PVDA+, were unlikely to get seats in the Chamber. In Wallonia there were 7 main parties contending the elections: Christian democrats (CDH), Greens (ECOLO), social democrats (PS), liberals (MR), French Democratic Front (FDF), National Front (FN, extreme- right) and Labour Party (PTB- GO!, extreme- left). Only the extreme- right FN was not expected to get any seats. Starting from this election, there is no more direct election of the Senate. The federal elections therefore only encompass the Lower Chamber election. The fact that we consider Flemish and Walloon voters and Dutch- and French- speaking parties separately does not render an analysis of the federal elections void. Because of the particular institutional and electoral organization of Belgium, the federal elections are de facto regionalized (see Russo & Deschouwer, 2014 for a more detailed explanation). Only in the Brussels region are voters still able to vote for French- and Dutch- speaking parties. In the rest of the country this is not the case. 7 We have conducted logistic regression analyses for each party with four models. The tables can be found in the Annex. The First model only includes socio- demographics and other independent variables that could influence vote choice. For model 2 we also included the score for the like/dislike question. We have divided the respondents into three categories 6. See note 3 for an English translation of the question we asked in the Dutch- language en French- language questionnaires. 7 We also excluded the voters from the 6 municipalities with special status (faciliteiten/facilités) around Brussels. In these municipalities voters could choose to vote as a member of the Flemish (Vlaams- Brabant province) or Brussels constituency. This is important because this means they were also able to vote for French- speaking parties, which would complicate our model. 25 respondents were excluded for this reason. 9

on the basis of their responses to the like/dislike question. Basically we divided the respondents into the three tertiles, which means that the respondents with the 33% highest scores on the 11- point scale end up in group high, the next 33% in group medium and the remaining 33% into to group low. According to us it makes more sense to divide these groups on the basis of the distribution instead of using fixed categories based on the points on the scale (e.g. 8-9- 10 as high ) because what is high and low might differ from respondent to respondent. In Model 3 we include the coalition potential variable. The division into three categories has been conducted in the same way as the like/dislike variable. We do not expect model 3 to show any interesting results, since we believe coalition potential only has an added value when it is combined with liking or disliking a party. We do this in model 4. To be able to see the effect of coalition potential we could simply construct an interaction term that multiplies the score on the like/dislike question with the coalition potential score. However, this way we would end up with the issue that a respondent with a 7 on the like/dislike scale * a 3 on the coalition potential scale (for a particular party) would end up with the same value (21) as someone who mentions a 3 on the like/dislike scale and a 7 on the coalition potential scale. This would undoubtedly blur the results. Furthermore, we do not expect that any score on the coalition potential scale would influence the vote utility for a given party. In order to explain voting intentions we believe it is a high coalition potential that will boast the most effect. When using a simple interaction term the other two categories (low and medium) would drown out or even counteract the effect of a high coalition potential. If we want to look at the specific contribution of a high coalition potential, we need to turn the combination of coalition potential and like/dislike scores into a categorical variable. This also closely reflects our theoretical underpinnings: certain, but not all, combinations of coalition potential and like/dislike scores will influence voting behaviour. Since the categorical variable combines the coalition potential and like/dislike variables, we end up with 9 categories: Coalition Potential Like/Dislike Low Low 1 2 3 4 5 6 7 8 9 These 9 categories can be found in the tables (both numbered and in words). This operationalization will allow to distinguish the effect of each combination of values and to 10

ascertain to what extent these effects are driven by the most powerful predictor (like/dislike variable) and whether the coalition potential variable actually adds anything to the explanatory power of the model. We would like to stress once again that these models are based on a preliminary, non- weighted dataset and that some other interesting and crucial independent variables can only be included when the post- electoral wave of the Voter Survey will have been concluded. These data should thus be seen as a first attempt to investigate our hypotheses and not as fully- fledged final tests. Discussion of the results The first thing we should notice is that the number of voters for certain parties makes them unfit for a decent analysis. We can see that the models for VLAAMS BELANG, LDD, PVDA+, FDF and FN do not function properly because of a too low N. As for the other parties, we believe that we can see two interesting results. First of all we can see that the pseudo R^2 values increase with each model except for GROEN, SP.A and PTB- GO!. However, for each of these parties we would contend that the lesser fit of the model is caused by the fact that each time 1 of the 9 cells of our categorical vote utility variable is empty, which can occur when working with real- life data. When we look at model 3 (which includes both coalition potential and like/dislike separately and doesn t suffer from empty cells) we can actually see that the pseudo R^2 values indeed increase compared to model 2, except for GROEN for which the values remain the same. This would lead us to the first tentative conclusion that the coalition potential variable does add some explanatory power. We do acknowledge that these increases are not always large, but we believe that these results are encouraging since these are first attempts of modelling the impact of coalition potential. Secondly, we can see that our expectations are at least partially confirmed when it comes to the effects of coalition potential. When we look a the categories from our vote utility variable we can see that the like/dislike variable is the driving force behind a large part of the effects, which was something we should already have suspected when looking at the performance and pseudo R^2 values of model 2. However, if it were only the like/dislike variable that drives the effect of vote utility, then categories 1/4/7 (low score like/dislike); 2/5/8 (medium score like/dislike); 3/6/9 (high score like/dislike) should show the same linear effect. But, that is not what the data tells us. We can see that the effects of categories 1/4/7 and 2/5/8 are relatively linear, yet for all parties except GROEN, N- VA, OPEN VLD and MR we see that the odds ratios for category 9 (high coalition potential and high like for 11

the given party) are always highly significant and exponentially larger compared to categories 3 and 6. This means that it is the high coalition potential that drives this exponentially large effect and turns the positive relationship into a curvilinear one. This is exactly what we expected: liking a party and thinking it has a high chance of ending up in government is a stronger predictor for vote choice than just liking that party. We should also note that for N- VA and Open VLD the odds ratios of category 9 are always significantly larger than those of categories 3 and 6. This shows that also for them the effect is not a linear one and that it is the group of respondents that report a medium coalition potential that turns it into a U- shaped curve. This does not violate our expectations, since we foresaw that it was mainly a perceived high coalition potential that would help in better explaining vote choice in combination with a high score on the like/dislike scale. The two remaining parties, GROEN and MR, reveal a quite linear effect when combining coalition potential and the like/dislike score, which means that for them coalition potential does not add any real explanatory value. At least for the Flemish greens this makes sense, since they are notorious for having a voting population that is less concerned with (or in some cases even opposed to) taking part in government. In 2003 they actually lost all of their federal MPs after being part of the federal government for the first time in history. In 2004 GROEN managed to safeguard their presence in the regional parliament and started rebuilding the party. Looking at their election results (8.5% of the Dutch- speaking voting population and an increase of one seat, from 5 to 6) show they have digested this difficult episode pretty well, but it remains to be seen how quickly they will take the risk of governing again. Conclusion We set out to explore strategic coalition voting and vote choice for the most recent Belgian federal elections. The PartiRep voter survey, or at least a preliminary version of the dataset, was used to achieve this goal. Results from the first explorative analyses show that 23.1% of Flemish respondents and 24.8% of Walloon respondents with a distinct party preference intended to vote insincerely. By this we mean that these respondents indicated to like one particular party best, yet they refrained from naming it as their intended vote choice. In more than 50% of these cases respondents reported a higher coalition potential for the party they intended to vote for than for their preferred party. This points towards the fact that perceived coalition potential does play a role in determining (intended) vote choice. We then moved beyond the group of insincere voters and tried to determine the impact of coalition potential on vote choice in general. Our logistic regression analyses show that, overall, the models that include perceived coalition potential as a part of a new utility variable predict intended vote choice better. This does not mean that all models perform better for all parties, but the results do point towards the added value of including voters perceived coalition potential in vote choice analyses. We readily admit that these results 12

are preliminary, but we hope more complete data and additional analyses will confirm these expectations. 13

Bibliography Abramson, P.R., J.H. Aldrich, A. Blais, M. Diamond, A. Disking, I.H. Indridason, D.J. Lee and R. Levine (2010), Comparing Strategic Voting Under FPTP and PR. Comparative Political Studies, 43(1), pp. 61-90 Alvarez, R.M., Boehmke, F.J. and J. Nagler (2006), Strategic Voting in British Elections. Electoral Studies, 25(1), pp. 1-19 Alvarez, R.M. and J. Nagler (2000), A New Approach for Modelling Strategic Voting in Multiparty Elections. British Journal of Political Science, 30(1), pp. 57-75 Bergstad, M.A. and O. Kedar (2009), Coalition- Targeted Duvergian Voting: How Expectation Affect Voter Choice under Proportional Representation. Americal Journal of Political Science, 53(2), pp. 307-323 Beyens, S. and T. Verthé (2013), De strategische kiezer. Partijvoorkeuren en complexe stemmotieven bij de Belgische gemeenteraadsverkiezingen van 2012, in Dassonneville, R., M. Hooghe, S. Marien and J.- B. Pilet (eds.) De lokale kiezer. Het kiesgedrag bij de Belgische gemeenteraadsverkiezingen van oktober 2012. Brussel: ASP Blais, A., Nadeau, R., Gidengil, E., and N. Nevitte (2001), Measuring Strategic Voting in Multiparty Plurality Elections. Electoral Studies, 20(3), pp. 343-52 Blais A., Young, R., and M. Rucotte (2005), Direct or indirect? Assessing Two Approaches to the Measurement of Strategic Voting. Electoral Studies, 24(2), pp. 163-176 Cox, G.W. (1997), Making Votes Count. Strategic Coordination in the World s Electoral Systems. Cambridge: Cambridge University Press Gschwend, T. (2007), Ticket- splitting and Strategic Voting under Mixed Electoral Rules: Evidence from Germany. European Journal of Political Research, 46(1), pp. 1-23 Merrill III, S. (1984), A Comparison of Efficiency of Multicandidate Electoral Systems. American Journal of Political Science, 28(1), pp. 23-48 Russo, L. and K. Deschouwer (2014), Electoral Nationalization in a Denationalized Party System: The Case(s) of Belgium. Paper presented at the 3 rd European Conference on Comparative Electoral Research Slinko, A. and S. White (2010), Proportional Representation and Strategic Voters. Journal of Theoretical Politics, 22(3), pp. 301-332 14

ANNEX Tables Logistic Regression Vote Choice per Party Data: PartiRep Voter Survey 2014 (Beta version). Coefficients are odds ratios. Significant * p<0,05; ** p<0,01; ***p<0,001 CD&V Flanders CD&V%(%1%=%147) 1,03 *** 1,04 *** 1,04 *** 1,04 *** 1,48 * 1,32 1,35 1,33 1,03 1,05 1,06 1,06 0,96 0,99 0,98 0,98 1,10 2,10 ** 2,17 ** 2,20 ** 0,98 0,88 * 0,87 ** 0,87 ** 7,98 *** 2,53 *** 129,84 *** 41,63 *** Coalition$Potential$$(Ref.$Cat.$=$low) 0,49 1,19 ** Like/Dislike Coalition2Potential 2 Low 12,46 *** 3 Low 173,11 *** 4 Low 2,41 5 14,42 *** 6 212,73 *** 7 Low 7,43 * 8 26,95 *** 9 395,13 *** N 679 679 679 679 Pseudo$R^2 0,0339 0,3825 0,3932 0,4152 15

GROEN Groen&(&1&=&63) 0,98 ** 0,99 0,99 0,99 0,85 0,85 0,85 0,87 1,15 * 1,11 1,11 1,11 0,93 0,89 0,89 0,89 1,22 1,35 1,36 1,36 1,02 1,00 1,00 1,01 2,05 2,16 30,74 *** 32,68 *** Coalition$Potential$ 0,88 0,85 Like/Dislike Coalition2Potential 2 Low 1,93 3 Low 26,01 *** 4 Low 1$(empty) 5 1,61 6 26,30 *** 7 Low 2,56 8 1,45 9 22,83 *** N 679 679 679 679 Pseudo$R^2 0,0374 0,2682 0,2686 0,2436 16

N-VA N"VA%(%1%=%225) 1,00 1,00 1,00 1,00 0,84 0,71 0,69 0,72 0,94 0,89 0,90 0,89 1,11 * 1,08 * 1,08 * 1,10 * 0,99 1,00 1,06 1,04 1,05 1,00 1,00 1,01 4,23 *** 4,22 *** 43,54 *** 39,42 *** Coalition$Potential (Ref.$Cat.$=$low) 1,44 1,76 * Like/Dislike Coalition2Potential 2 Low 7,33 ** 3 Low 130,55 *** 4 Low 5,83 ** 5 15,69 *** 6 83,88 *** 7 Low 1,32 8 14,89 *** 9 170,26 *** N 679 679 679 679 Pseudo$R^2 0,0154 0,3316 0,3366 0,3520 17

SP.A SP.A%(%1%=%104) 0,99 1,01 1,01 1,01 0,77 0,81 0,80 0,79 0,93 0,94 0,95 0,95 0,96 0,97 0,97 0,97 0,81 0,87 0,89 0,88 0,93 0,91 0,91 0,91 7,43 ** 7,13 * 142,82 *** 132,60 *** Coalition$Potential (Ref.$Cat.$=$low) 0,93 1,74 Like/Dislike Coalition2Potential 2 Low 10,87 * 3 Low 164,96 *** 4 Low 1$(empty) 5 7,90 * 6 168,98 *** 7 Low 8,07 8 14,46 * 9 293,55 *** N 679 679 679 619 Pseudo$R^2 0,0210 0,3753 0,3821 0,3628 18

VLAAMS BELANG VB#(#1#=#19) 0,98 0,99 0,99 0,99 1,09 1,07 0,96 0,94 1,00 1,03 1,05 1,05 0,87 0,88 0,88 0,88 0,23 * 0,17 ** 0,16 ** 0,16 ** 0,79 ** 0,79 * 0,79 * 0,79 * 1$(empty) 1$(empty) 10,58 ** 7,15 * Coalition$Potential (Ref.$Cat.$=$low) 1,43 4,08 Like/Dislike Coalition1Potential 2 Low 1$(empty) 3 Low 0,39 4 Low 0,12 * 5 1$(empty) 6 0,29 7 Low 0,16 8 1$(empty) 9 1$(omitted) N 679 459 459 299 Pseudo$R^2 0,1124 0,2126 0,2452 0,1725 19

OPEN VLD Open%VLD%(%1%=%101) 1,00 1,01 1,01 1,01 1,09 0,91 0,85 0,87 1,03 0,99 0,99 0,99 1,06 1,08 1,09 1,10 1,30 1,46 1,45 1,42 1,07 0,99 0,98 0,98 1,85 2,07 28,34 *** 29,63 *** Coalition$Potential (Ref.$Cat.$=$low) 0,64 1,40 Like/Dislike Coalition1Potential 2 Low 2,73 3 Low 45,93 *** 4 Low 1,14 5 1,98 6 22,92 *** 7 Low 2,71 8 2,39 9 55,80 *** N 679 679 679 679 Pseudo$R^2 0,0133 0,2734 0,2846 0,2874 20

LDD LDD#(#1#=#3) 1,00 1,01 1,03 1,03 1,54 1,46 1,48 1,48 0,84 0,88 0,86 0,93 1,01 1,04 1,04 1,14 1,84 1,79 2,06 1,62 0,60 0,62 0,59 0,53 1$(empty) 1$(empty) 2,18 0,72 Coalition$Potential (Ref.$Cat.$=$low) 0,41 1$(omitted) Like/Dislike Coalition1Potential 2 Low 1$(empty) 3 Low 1$(empty) 4 Low 0,90 5 1$(empty) 6 1$(empty) 7 Low 1$(empty) 8 1$(empty) 9 1$(omitted) N 679 436 198 126 Pseudo$R^2 0,1791 0,3500 0,1958 0,1914 21

PVDA+ PVDA+&(&1&=&17) 1,01 1,03 1,03 1,03 1,53 1,31 1,22 1,21 1,28 1,28 1,29 1,32 * 0,83 0,83 0,84 0,84 0,87 0,82 0,72 0,72 1,15 1,05 1,03 1,03 0,05 *** 0,05 *** 1$(omitted) 1$(omitted) Coalition$Potential (Ref.$Cat.$=$low) 2,13 1,17 Like/Dislike Coalition1Potential 2 Low 1$(empty) 3 Low 1,54 4 Low 1$(empty) 5 1$(empty) 6 3,58 7 Low 1$(empty) 8 0,39 9 1$(omitted) N 679 440 440 233 Pseudo$R^2 0,0589 0,2350 0,2455 0,1694 22

Wallonia PS PS#(#1#=#206) 1,00 1,01 1,01 1,01 1,24 0,95 0,97 0,98 0,87 *** 0,90 * 0,90 * 0,90 * 0,91 * 0,87 ** 0,87 ** 0,87 ** 0,56 ** 0,95 0,92 0,96 1,01 1,00 0,99 0,99 7,51 *** 7,71 *** 48,84 *** 47,17 *** Coalition$Potential$ 0,88 1,62 Like/Dislike Coalition1Potential 2 Low 12,53 *** 3 Low 43,19 *** 4 Low 1,88 5 8,45 *** 6 36,60 *** 7 Low 2,09 8 10,89 *** 9 244,00 *** N 567 567 567 567 Pseudo$R^2 0,0597 0,3339 0,3404 0,3522 23

MR MR#(#1#=#130) 1,01 1,01 1,01 1,01 0,46 *** 0,50 ** 0,48 ** 0,49 ** 1,16 ** 1,14 * 1,14 * 1,15 * 1,11 * 1,07 1,06 1,06 1,54 1,96 * 1,91 * 1,83 * 1,06 1,03 1,03 1,02 1,99 1,90 22,66 *** 21,94 *** Coalition$Potential$ 1,67 1,06 Like/Dislike Coalition1Potential 2 Low 1,46 3 Low 45,13 *** 4 Low 1,97 5 4,23 * 6 39,25 *** 7 Low 1,47 8 3,27 9 23,71 *** N 567 567 567 567 Pseudo$R^2 0,0719 0,3312 0,3362 0,3410 24

CDH CDH$($1$=$84) 1,02 * 1,02 1,02 * 1,02 * 1,48 1,13 1,17 1,15 1,01 1,02 1,02 1,01 1,07 1,02 1,02 1,03 1,03 1,12 1,14 1,10 0,98 0,97 0,97 0,96 3,44 ** 3,10 ** 22,81 *** 19,15 *** Coalition$Potential (Ref.$Cat.$=$low) 1,25 2,03 Like/Dislike Coalition1Potential 2 Low 6,12 * 3 Low 13,27 *** 4 Low 0,98 5 2,46 6 27,08 *** 7 Low 2,71 8 6,10 * 9 35,91 *** N 567 567 567 567 Pseudo$R^2 0,0178 0,2485 0,2564 0,2639 25

ECOLO ECOLO%(%1%=%80) 0,97 ** 0,98 * 0,98 * 0,98 * 1,36 0,94 0,90 0,92 1,17 ** 1,08 1,08 1,08 1,00 1,02 1,03 1,03 0,63 0,84 0,82 0,82 0,94 0,90 0,90 0,90 4,10 ** 3,81 ** 20,47 *** 17,80 *** Coalition$Potential (Ref.$Cat.$=$low) 1,23 1,87 Like/Dislike Coalition1Potential 2 Low 5,36 3 Low 27,42 *** 4 Low 1,68 5 5,82 ** 6 28,88 *** 7 Low 3,36 8 9,81 ** 9 40,58 *** N 567 567 567 567 Pseudo$R^2 0,0613 0,2663 0,2722 0,2737 26

FDF FDF#(#1#=#5) 1,00 1,00 1,00 1,00 0,72 0,65 0,64 0,59 1,04 1,01 1,01 0,99 0,77 0,74 0,73 0,73 4,05 5,16 5,42 5,58 1,10 1,11 1,11 1,14 1,40 1,31 5,61 5,18 Coalition$Potential (Ref.$Cat.$=$low) 2,64 2,00 Like/Dislike Coalition1Potential 2 Low 1$(empty) 3 Low 1$(empty) 4 Low 1$(empty) 5 1,17 6 5,34 7 Low 1$(empty) 8 0,92 9 2,80 N 567 567 567 374 Pseudo$R^2 0,0777 0,1163 0,1225 0,1222 27

FN FN#(#1#=#13) 0,98 0,99 1,00 1,00 1,37 1,33 1,02 1,00 0,90 0,93 0,97 0,98 0,96 1,02 1,03 1,03 2,03 1,88 2,24 2,54 0,99 1,01 1,00 0,99 1$(empty) 1$(empty) 25,35 ** 15,34 ** Coalition$Potential (Ref.$Cat.$=$low) 3,77 8,76 Like/Dislike Coalition1Potential 2 Low 1$(empty) 3 Low 1$(empty) 4 Low 1$(empty) 5 1$(empty) 6 14,70 ** 7 Low 1$(empty) 8 1$(empty) 9 32,53 ** N 567 522 522 352 Pseudo$R^2 0,0346 0,1982 0,2464 0,2182 28

PTB- GO! PTB$GO!(((1(=(49) 1,00 1,01 1,01 1,01 0,83 0,74 0,71 0,71 0,91 0,91 0,93 0,93 0,98 1,01 1,03 1,04 2,21 ** 2,56 ** 2,45 ** 2,43 ** 0,98 0,96 0,97 0,97 9,42 ** 7,70 ** 54,20 *** 39,16 *** Coalition$Potential (Ref.$Cat.$=$low) 1,54 2,20 Like/Dislike Coalition1Potential 2 Low 8,27 3 Low 35,88 ** 4 Low 2,41 5 8,33 6 69,50 *** 7 Low 1$(empty) 8 20,27 ** 9 82,43 *** N 567 567 567 543 Pseudo$R^2 0,0296 0,2428 0,2507 0,2455 29