• 1 Introduction

  This study argues that it is the importance of issues that can help explain the outcome of the 2019 federal elections in Belgium, in both Flanders and Francophone Belgium. The policy preferences of voters are structured along two broader dimensions: economic left-right dimension, involving topics such as the intervention of the state in the economy and redistribution, and cultural left-right dimension, which deals with topics such as immigration, the environment, and law and order. These two dimensions are not entirely orthogonal, they are correlated; voters who tend to be economically left-wing often also take a culturally left-wing position and vice versa. Yet, among citizens, the correlation between both dimensions is only moderate with quite some citizens holding left-wing stances on one dimension and right-wing stances on the other, or the other way around. The party supply in Flanders and Francophone Belgium, though, is strongly and consistently organised along just one dimension: among the parties, the economic and cultural left-right cleavages are strongly overlapping and, in fact, almost concurring. More concretely, there are no parties that combine a left-wing position on one dimension with a right-wing position on the other. This has resulted in a difference between the demand by voters and supply by parties. The atypical voters, combining a left-wing position on one dimension with a right-wing position on the other, do not have a ‘natural’ party that offers them the best possible match on both the dimensions. Hence, these voters need to prioritise their position on one of both cleavages when trying to cast a vote that matches their policy preferences. In other words, the study demonstrates, in two different party systems, that it is the salience of the two dimensions that explains where these ‘unserved’ voters will end up and whether they vote for a right- or a left-wing party. The same mechanism also applies to voters who are not atypical but who are consistently left- or right-wing on both dimensions and who are ‘served’ by often several parties. These voters, our study shows, use the salience of the two dimensions to choose for a party that is most adjacent on the dimension most dear to them.

• 2 The Dimensionality of Voter Demand and Party Supply

  Party and election scholars have long since debated the so-called ‘dimensionality’ of citizen and party preferences. There are two subtly different ways of thinking about this. One school talks about political cleavages that are structural and hark back to the foundational work by Lipset and Rokkan (1967) on the coming about, and effects, of political cleavages (see also Bartolini & Mair, 2007; Rokkan, 1977). The other way of talking about the dimensionality considers issue dimensions as ideational without a ‘material’ base and focuses more on how parties and elites try to shift the debate by switching issues (e.g. Riker, 1982; Schattschneider, 1960). However, according to both takes, the basic idea is that citizen’s political preferences – their views about society, their ideological stances, their policy positions, their concerns … – are not random but are rather structured around a limited number of, or even one, dominating fault line(s). People’s preferences are, in other words, ‘constrained’ in the sense that their preference with regard to one policy matter is correlated with their preference with regard to another matter (Converse, 1964). If one knows what a certain voter thinks about issue A, one can predict what (s) he thinks about issue B, as the response to A and B depends on the same underlying, stable ideological attitude and/or interest. This structure of mass beliefs then ‘generates’ a party system that matches citizens’ convictions and that provides party programmatic ‘supply’ to voters on both sides of a dimension.
  Traditionally, the economic left-right conflict is considered as the main fault line in most countries (Huber & Inglehart, 1995). This conflict pitches the proponents of equality and redistributive state intervention against the champions of freedom pleading for as little state intervention as possible (e.g. Marks et al., 2006). The success of green and, a little later, radical-right parties sparked an intense debate about the potential rise of a new cleavage (e.g. Inglehart, 1990; Kitschelt, 2004). The literature is vast and broad. The new cleavage has received many different labels. Some speak, for instance, about the ‘GAL-TAN’ dimension (green-alternative-libertarian vs. traditional-authoritarian-nationalist) (Hooghe et al., 2002) as being the new cleavage complementing the existing economic left-right cleavage. Others simply talk about the ‘new politics’ dimension. Still others say that the rift is not by definition new but that the nature of the second dimension simply shifted. Where in the past this was about ethical questions, nowadays it rather positions the ‘winners’ against the ‘losers’ of globalisation (Kriesi et al., 2008). What these approaches have in common is that the new cleavage includes non-economic, often post-materialistic, issues such as the environment, crime, migration, multiculturalism, identity, lifestyle, democracy, and, according to some, also attitudes with regard to internationalism and, in Europe, European Union integration (see also Wheatley, 2015). As the issues ‘loading’ on this second cleavage are so heterogeneous, and as they all relate to a non-economic definition of what a ‘good society’ entails, we will use the term ‘cultural left-right’ in this study.
  The rise of this new dimension has shaken both mass beliefs and party systems. Party systems have adapted to the new situation with voters holding different, and more complex, combinations of relevant preferences. In some countries, though, this led to a mismatch between the demand of voters and the supply by parties, as the structuring force of the underlying ‘belief systems’ appears to differ across the demand and supply side (Thomassen, 1999). Thomassen (2012), for instance, based on Dutch data, shows that a large group of voters who hold economic left-wing opinions together with cultural right-wing opinions are not represented by any of the Dutch parties (or at least not by the preferences of the elected representatives) – he speaks of a ‘blind corner’ of representation. A study on Swiss voters and elites reaches similar conclusions (Rosset et al., 2016). It shows that, among the political candidates they interviewed, all dimensions basically converge into one encompassing left-right divide. Swiss citizens, in contrast, have more complex preferences and the different dimensions that structure their preferences are much less correlated. This may lead, the study says, to voters having trouble finding the right representative.
  The most influential study about a potential mismatch between voter demand and party supply has been authored by van Der Brug and van Spanje (2009). They draw on comparative cross-European evidence and argue that, looking at the party supply, there are no two left-right dimensions but only one. Indeed, their analysis of parties’ positions shows that European parties’ economic left-right and their cultural left-right positions largely coincide. Economically left parties are almost invariably also culturally left, and vice versa. However, among many European voters, they do find two more independent left-right dimensions. In particular, many voters are culturally right-wing, yet combine this with economically left-wing stances. The authors therefore conclude that there is a good deal of European voters whose demands are not adequately met by the party supply and that therefore there is a ‘representation gap’. In the conclusion of the study, these authors speculate that voters, lacking a party that represents them well on both dimensions, will swing to the right or to the left depending on the issues they are ‘mainly concerned’ about, but they do not offer any empirical test for that argument.
  A study by Lefkofridi et al. (2014) does offer some first empirical validation of the role of saliency in making unserved voters pivot towards the left or right. Focusing on, what they call, ‘left authoritarians’ – in our language, economic left combined with cultural rights – these scholars contend that this unserved audience mostly cares about economic issues and, as a consequence, tends to vote for left-wing parties. In other words, the study shows that salience is crucial to understand the vote decisions of cross-pressurised voters. In another study, but in a very different context, Feld et al. (2014) also provide evidence that, at least in the two-party system of the United States, the salience that citizens give to a dimension determines to what extent the ideological distance to a party on that dimension affects their vote choice. Yet, they predominantly study this at the aggregate level rather than at the individual level.
  In this study, we further develop these ideas and test, based on, we believe, superior measures of voters’ position on both cleavages, whether similar mechanisms have been at play during the 2019 elections in Flanders and Francophone Belgium.

• 3 Two-Dimensional Voter Demand Versus One-Dimensional Party Supply

  In Flanders, we used the University of Antwerp Citizen Panel and surveyed 16,908 respondents in July 2019. In Francophone Belgium, we used the UC Louvain Citizen Panel and surveyed 7,978 voters in March 2020. The last panel is a mix of Walloon and Brussels respondents. While the Antwerp panel is a long-lasting (2003) opt-in citizen panel, the respondents used for this study were all newly recruited through the vote advice application (VAA) Stemtest 19 that contained an opt-in option. The data from the VAA itself were not used, though, and all data presented here are based on a new survey. The Louvain panel is new and consists exclusively of opt-in respondents recruited through the VAA Test Electoral 2019. Due to missing data, the actual number of useful respondents in both panels is a little smaller. We consider the inclusion of two separate regions – each with its own party system, media landscape and voters – to offer a double, and independent, test of our expectations on two different cases.
  Both online panels we draw upon are diverse and contain respondents of all ages, levels of education and income. But the two samples are definitely not representative for the voters of Flemish and Francophone Belgium. A crucial skew is that the higher educated are overrepresented and that most participants have higher than average levels of political interest. In fact, the skew in the panels is of such a magnitude that weighing the two samples is not a feasible option but tests show that weighing the data does not change the results.¹x If we just weigh the Flemish sample based on age, sex and education, the weights soar and reach levels of 14 and more (lower educated women). This means that one person would get the weight of 14 ‘standard’ individuals. Such enormous weights are unwanted as too much depends on the answers of just a few people. Still, weighing the data with such large weights does not change the results. We tested it in Flanders and there hardly were differences, neither in the conditional logit model nor in the multinomial model. The coefficients of the socio-demographic control variables change a little but the substantive results are the same. That is why we opted to use unweighed data. That the respondents are not representative should make us very cautious to generalise to the entire Belgian population. We do come back to this in the conclusion section where we argue that the skewed sample is not very likely to be the reason for finding what we find.
  In each region, we asked the respondents for their position with regard to 40 policy proposals: 20 of those proposals were meant to tap into the classic economic left-right cleavage and 20 others were aimed to gauge the cultural left-right division. Note that we deliberately used different proposals in both regions and that the proposals also include policies at the regional, national and European levels. In fact, the 40 proposals were drawn from all policy proposals that were tested and included in the preparatory phase of Stemtest19/Test electoral 2019 and of which we had party answers (N = 275 in Flanders and 326 in Francophone Belgium). We sampled 20 + 20 proposals that best reproduced the party placements if we used all VAA proposals. As the sample of initial VAA proposals is different across the two regions, the selections of 20 + 20 proposals are also different. A full list of used items can be found in Appendix 1.
  The question wording was as follows: ‘Below you will find a series of policy proposals. Can you indicate for each of them whether you disagree or agree?’ The 20 economic items all relate to state intervention in economic and social life and to economic redistribution. The 20 cultural items are more diverse and cover a wider array of issues with regard to the environment (and climate), migration, crime, Europeanisation, and democracy. While there is still some scholarly debate about the exact meaning of the cultural left-right dimension and on whether, for instance, European integration and the environment fall under the cultural dimension (Kriesi et al., 2008), in general these are the issues that are considered by most scholars as the core of the cultural GALTAN dimension and around which the ‘new’ challenger parties have mobilised (Hooghe et al., 2002; Kitschelt, 2004; van der Brug & van Spanje, 2009). We therefore argue that the 20 cultural policies used in Flemish and Francophone Belgium cover the main constituting issues quite well: the issues of migration, and to a lesser extent democracy and environment, were important issues not only in the campaign, they were also considered important for many voters. As such, our measurement taps a good deal of relevant topics related to the cultural left-right dimension.
  To examine whether each batch of 20 items forms a reliable scale, we tested all four batches separately using Cronbach alpha estimations. With alpha scores of 0.75 (economic) and 0.79 (cultural) in Flanders and of 0.73 (economic) and 0.71 (cultural) in Francophone Belgium, all four scales perform quite well, with each of the 20 items correlating internally. Moreover, omitting items from the scale does not meaningfully improve them, so we opt to use the full set of items here. To test the validity, we also correlated the party positions on the resulting scales with the Chapel Hill Expert Survey (CHES) scores on the economic and cultural dimension (Bakker et al., 2020). For both dimensions, and in both regions, the correlations are all above 0.9, demonstrating a high convergent validity between our scales and the CHES scores.²x The CHES score for the cultural dimension is based on the average of the party scores on EU, immigration, multiculturalism, personal freedom and environment.
  We construct simple additive scales and then standardise results to a -5 to +5 scale. Although they are theoretically clearly different, in both regions, there is a substantial correlation between the positions occupied by citizens on the two dimensions (Pearson’s r in Flanders, r = 0.60, and in Wallonia, r = 0.57). This correlation is not exceedingly high but does show that, also among citizens, among both Flemings and Francophones, these dimensions are not entirely independent. A further validation of the validity of the two scales is the fact that both of them, as expected, correlate with citizens’ self-placement on a non-specified left-right continuum (question wording: “In politics the terms ‘left’ and ‘right’ are often used. Can you place your own convictions on a scale from 0 to 10 whereby 0 means ‘left’, 5 ‘centre’ and 10 ‘right’?”) There is a (Pearson) correlation of 0.63 (FL) and 0.66 (WAL) between respondents’ left-right self-placement and economic position and of 0.68 (FL) and 0.55 (WAL) between left-right self-placement and cultural position. To sum up, our two scales seem to form valid and reliable measures of two, partially distinct, dimensions in voter preferences.
  The question now is where the Flemish and Francophone voters (we use their most recent 2019 vote) are positioned in the two-dimensional space created by the two ideological dimensions. Figure 1 shows the position of all surveyed Flemish voters and the Flemish parties in the two-dimensional space; Figure 2 does the exact same for the surveyed Francophone voters and parties. Each voter is represented by one light-blue dot. Both figures underpin the correlation between both ideological dimensions as the observations are structured in an oval and not in a circle or an amorphous cloud (see also the correlation coefficients mentioned earlier). Yet, at the same time, a good deal of voters deviate from the predominant diagonal pattern and are positioned in the upper left and lower right quadrants.³x One could argue that our simple calculation of the four quadrants is too rude and that many voters are actually centre voters. This is true, of course. If we define all voters who, on our -5 to +5 scale, score between 0 and ±1, as centre voters, we see that 13.3% (FL) and 20.4% (WAL) can be considered to be left nor right on neither of the dimensions. This stricter operationalisation reduces the share of economic left + cultural right (FL: 8.8%, WAL: 9.9%) and of economic right + cultural left (FL: 11.2%, WAL: 7.2%) voters somewhat, but they remain considerable groups. Even more nuanced categorisations are possible whereby voters combine a left or right position on one dimension with a centre position on the other. These are the voters for whom a left or right position on one dimension is not predictive for their position on the other dimension. In both the Flemish and Francophone samples, the share of these ‘atypical citizens’ in the quadrants 1 and 4 is smaller than that of the group of mainstream voters, who are left or right on both dimensions, in the quadrants 2 and 3. The left-left voters (Q3) form 45% (FL) and 55% (WAL) of our sample, the right-right voters (Q2) 30% (FL) and 14% (WAL). The two atypical quadrants – Q1 left-right (FL: 12%, WAL: 21%) and Q4 right-left (FL: 13%, WAL: 10%) – encompass about one-fourth of the voters in our Flemish sample and about one-third in the Francophone sample.
  The size of the segments of atypical voters we find in our samples in Flemish and Francophone Belgium is smaller than what van der Brug and van Spanje (2009) found comparatively. In fact, according to their calculations, especially the left-right segment covers, in most European countries, the biggest group of voters (see also Lefkofridi et al., 2014, who estimate the share of ‘left-authoritarian’ voters in Belgium at 20%). Yet, these scholars’ results were based on heterogeneous data combining different types of measures and scales while our results are based on homogeneous and standardised measures. However, note that we do not claim that about 25% (Fl.) or 30% (Fr.) would be the exact size of these quadrants in the full population of Flemish or Francophone voters – the panels we work with do not constitute representative samples. The fact that we find considerable shares of ‘unconstrained’ citizens in a sample of politically interested citizens suggests that in the population as a whole we may even have bigger groups of atypical voters. Indeed, when we look separately at citizens with different education levels and levels of political interest, we see that the correlation between both dimensions is smaller for the low educated and the low politically interested. In other words, among our panellists, there is a substantive effect of education and political interest on the size of the correlation between people’s economic and cultural positions. The overall correlation among all respondents together between both sets of positions is 0.60. The lower educated have an r of 0.53, the middle educated 0.57 and the higher educated 0.62. For respondents with less interest in politics (lower than 7 on a 0-10 scale), the correlation between their sets of positions is 0.46, and for those with high levels of interest it is 0.60. In other words, the two dimensions are more correlated among the citizens we have in our sample (highly educated and with a high interest in politics) compared to the general population and we would probably have found a larger share of atypical voters had we had a representative sample.

  Position of Flemish voters (n = 14,678) and Flemish parties according to two ideological dimensions
  
  Position of Francophone voters (n = 5,799) and Francophone parties according to two ideological dimensions
  

  While a good deal of voters are positioned in the two atypical quadrants, the offer of the parties is almost entirely one dimensional. Indeed, in the figures we also added the position of the seven Flemish and the six Francophone parties. All parties provided us with their official position on the same 40 policy proposals in the framework of the Voting Advice Application Stemtest19/Test electoral 2019 that was made by some of the authors of this study. As we have the same data for voters and parties, we can locate parties in the ideological space of the same graph.
  The offer of the parties clearly follows a single left-right structure and both left-right dimensions are very highly correlated (Pearson’s r = 0.85 (Fl.) and r = 0.88 (Fr.)), which is a good deal higher than the correlation between these dimensions among citizens. In addition, none of the parties are situated in either of the two ‘deviant’ quadrants and the parties are almost perfectly situated on the diagonal. This means that two of the ideological quadrants among voters are basically ‘unserved’: there is not a single party that provides a set of positions that matches these voters’ preferences on both dimensions at the same time. For voters who hold left-wing positions both economically and culturally, the choice is relatively simple. In Flanders three parties (Groen, sp.a and PVDA) and in Wallonia even five parties (Ecolo, PS, PTB, cdH and Défi) largely cover their policy preferences. For voters having right-wing opinions consistently, life is similarly simple: three parties supply such a combination of stances in Flanders (N-VA, Open Vld and Vlaams Belang) and there are two such parties in Wallonia (MR and PP). But things are more complicated for voters holding left and right positions simultaneously. For whom do these voters of quadrants 1 and 4 vote? As we will explore in the next section, depending on the salience of the issues underlying the two dimensions, the voters turn towards the right-right or the left-left quadrant.

• 4 Connecting Demand and Supply and the Importance of Salience

  Table 1 shows the voting behaviour of the voters who voted for one of the major parties⁴x We dropped the voters who voted for other parties, who did not turn out, were not allowed to vote, voted blank, or forgot whom they voted for. This leaves us with 14,096 Flemish and 5,285 Francophone voters. We do not find that these unserved voters end up more in these non-party categories. split up by ideological quadrant.

  Table 1 Party vote of voters situated in four ideological quadrants

  	Economic and cultural left (Q3)	Economic left and cultural right (Q1)	Economic right and cultural left (Q4)	Economic and cultural right (Q2)
  Flanders
  PVDA (Q3)	21.3	9.4	2.6	0.7
  Groen (Q3)	39.4	3.6	21.2	0.9
  sp.a (Q3)	19.4	9.8	4.9	1.7
  CD&V (Q0)	9.2	12.8	14.9	7.0
  Open Vld (Q2)	4.1	6.7	30.5	13.1
  N-VA (Q2)	5.4	37.6	23.7	65.0
  Vlaams Belang (Q2)	1.2	20.0	2.4	11.7
  Total Flanders	100	100	100	100
  Wallonia
  PTB (Q3)	14.9	16.6	1.8	2.1
  Ecolo (Q3)	45.6	10.8	20.0	3.0
  PS (Q3)	23.6	25.3	5.9	3.3
  cdH (Q0)	7.0	14.4	13.6	9.3
  Défi (Q0)	4.7	8.6	7.1	5.3
  MR (Q2)	4.1	20.6	51.1	69.6
  PP (Q2)	0.1	3.8	0.5	7.4
  Total Wallonia	100	100	100	100

  For the consistently left- or right-leaning electorates, results are pretty straightforward: they overwhelmingly vote for parties in their ‘corner’ of the ideological space. ‘Left-left’ voters (Q3) in Flanders vote for PVDA and Groen (and sp.a) and in Wallonia for Ecolo and PS. The same applies to ‘right-right’ voters (Q2); they have an easy choice as well for N-VA, Open Vld (and Vlaams Belang) in Flanders and for MR in Wallonia. When we examine the voting behaviour of the two unserved quadrants (Q1 and Q4), an interesting pattern emerges. In Flanders, both these segments appear to have voted predominantly for right-wing parties. The economic left + cultural right voters (Q1) prefer N-VA and Vlaams Belang – had they followed their economic stances they would have ended up with one of the left parties. The economic right + cultural left voters (Q4) as well opt mostly for the right-wing parties Open Vld and N-VA, although Groen is not far behind – had these voters followed their cultural stances they would have voted Groen, sp.a or PVDA. In Wallonia, the unserved voters are more torn between both poles. PS and MR are most popular amongst voters with ‘left-right’ positions (Q1) while Ecolo and MR are most popular among ‘right-left’ voters (Q4). What explains the votes (sometime right and sometimes left) of the voters in the unserved quadrants Q1 and Q4? How come they sometimes turn to the left and other times turn to the right although their preferences would warrant both votes? Our answer is simple: the salience of the two dimensions explains the choices the unserved voters make. Voters in the unserved quadrants use salience as an arbiter to make a trade-off between their contradictory preferences. In the rest of this study, we present proof for that interpretation.
  As a first step, how do we go about measuring salience? We do not have a direct measure of how important voters find the two (abstract) dimensions but rather use the salience they attribute to issues. We asked the following question: ‘Can you indicate which two issues have been the most important when casting your vote at the recent elections?’ A list of 16 different issues was presented to respondents asking them to select two. We classified seven issues as typically being economic and five as typically being cultural. One can debate our classification of issues as belonging to the two dimensions, but our classification of issues is consistent with the classification of the 40 policy statements above (see Appendix 1) and allows to attribute weights to the two dimensions. Social security, economy, poverty, taxes, pensions, health and employment are considered socio-economic issues. Migration, climate, democracy, environment and crime are socio-cultural issues. Figure 3 shows the shares of people picking any issue; note that the total percentages amount to more than 100% as each respondent could select two issues.

  Salience of policy issues among citizens (Flanders n = 14,786; Francophone Belgium n = 5,859)
  

  The first seven issues in the graph refer to the economic dimension, the five subsequent issues to cultural dimension, and the final four issues do not belong to any of the two dimensions. The graph makes it clear that, among our Flemish respondents (dark bars), two cultural issues dominate: migration (26%) and climate (24%). The third and fourth issue in Flanders are economic: social security (21%) and economy (18%). In Francophone Belgium, two other cultural issues dominate, democracy (23%) and the environment (22%), but their dominance is less strong. They are followed by climate (19%), social security (17%) and economy (17%). Our panel populated by the highly educated displays altogether quite small salience differences among our Flemish and Francophone respondents.
  Based on the two most salient issues for each respondent, we can construct a variable that taps into the salience of the two dimensions going from -2 (cultural dimension salient) to +2 (economic dimension salient). Voters who picked two cultural or two economic issues got a score of -2 and +2, respectively; voters with mixed issues got a score of 0 and voters picking one cultural or one economic issue combined with one of the non-dimensional issues got a score or -1 or +1, respectively. In Table 2 we show the resulting salience distribution of both groups of voters.
  The data show a fair distribution of respondents over the salience scale with, at both sides of the language border, the largest groups situated in the middle, sharing economic with cultural concerns. Further, almost equally sized groups prefer economic or cultural issues. In Flanders, those who prioritise cultural issues are a bit more numerous than in Francophone Belgium, while a little more Francophones seem to care more about economic issues, but differences are small. Note that the fact that we used seven economic issues compared to only five cultural issues may have had an effect on this distribution. Had we had seven cultural issues, that dimension may have dominated.
  To test the idea that the salience of a dimension pushes voters to the left or to the right and that voters do not weigh distances on the two dimensions equally, we run three conditional logit models. Conditional logit models are useful in discrete choice situations where units (i.e. voters) can select one option out of a set of alternatives (i.e. parties). The advantage of this modelling strategy is that it enables to explain a respondent’s vote choice for a given party in one single model, essentially mapping the decision-making process of a respondent. To run these models the data are restructured and stacked in such a way that each case represents a respondent × party unit, i.e. each respondent is present as a case seven times, each time linked to one of the seven Flemish or Francophone parties. The aim of the models is to explain why a respondent does (DV = 1) or does not (DV = 0) cast a vote for that given party, combining individual characteristics of the voter – i.e. gender, education, age and political interest – with variables that vary on the respondent × party unit – i.e. the ideological distance between the respondent and the party on the economic and cultural dimension. Although technically the stacking makes the data hierarchical, multi-level models cannot be used here, as the dependent variable is not independent across the seven party observations for a respondent; a voter can vote for only one party and knowing that (s) he voted for that party immediately implies that the DV for all other party observations is zero. The conditional logit model takes this dependence into account. One drawback of this approach is that we cannot incorporate the direct effects of salience into the model, as these measures do not vary by alternative.⁵x One assumption of the conditional logit model is the assumption of independence of irrelevant alternatives (IIA). A Haussmann test shows that our main model actually violates this assumption. For that reason, we also ran a mixed logit choice model of the main model, which loosens this assumption but takes more computational time. This alternative model results in identical results, which is why we continue with the conditional logit models.
  In Table 3, the relevant coefficients for the conditional logit models are summarised, full models with all controls and model diagnostics can be found in Appendix 2. The first key variable is the distance between a voter and a party on both dimensions (which we expect to exert a negative effect on the vote). Based on the voter and party answers on the same 20 statements related to each dimension, we calculate the distance between each voter and each party on each dimension separately. The second key variable is the salience of both dimensions. In the conditional logit models, we do not use the explained salience scale but rather separate ordinal variables of salience for each dimension that gauge whether a voter, among his/her two most important issues, has selected one (moderately salient) or two (highly salient) issues of a dimension (compared to selecting no economic or cultural issues).

  Table 2 Distribution of the salience of the two dimensions (Flanders, n = 14,682; Francophone Belgium, n = 5,859)

  	Flanders (%)	Francophone Belgium (%)
  Only cultural (-2)	16.2	14.2
  Cultural combined (-1)	15.5	12.2
  Mixed (0)	38.6	37.9
  Economical combined (+1)	11.2	12.3
  Only economical (+2)	18.4	23.4
  Total	100	100

  Table 3 Conditional logit models explaining vote choice based on positional distance and salience of two dimensions

  	Flanders			Francophone Belgium
  	Model 1 (main effects)	Model 2 (economic salience interactions)	Model 3 (cultural salience interactions)	Model 1 (main effects)	Model 2 (economic salience interactions)	Model 3 (cultural salience interactions)
  Distance economic dimension	-0.393*** (0.009)	-0.260*** (0.016)	-0.447*** (0.015)	-0.540*** (0.018)	-0.396*** (0.034)	-0.601*** (0.028)
  Distance cultural dimension	-0.421*** (0.008)	-0.543*** (0.016)	-0.310*** (0.015)	-0.401*** (0.02)	-0.511*** (0.037)	-0.280*** (0.033)
  Economic dimension moderately salient (ref. not salient) × distance economic dimension	–	-0.146*** (0.018)	–	–	-0.162*** (0.038)	–
  Economic dimension highly salient (ref. not salient) × distance economic dimension	–	-0.240*** (0.026)	–	–	-0.236*** (0.053)	–
  Economic dimension moderately salient (ref. not salient) × distance cultural dimension	–	0.138*** (0.019)	–	–	0.158*** (0.041)	–
  Economic dimension highly salient (ref. not salient) × distance cultural dimension	–	0.277*** (0.025)	–	–	0.251*** (0.060)	–
  Cultural dimension moderately salient (ref. not salient) × distance economic dimension	–	–	0.073*** (0.018)	–	–	0.081*** (0.035)
  Cultural dimension highly salient (ref. not salient) × distance economic dimension	–	–	0.185*** (0.028)	–	–	0.237*** (0.061)
  Cultural dimension moderately salient (ref. not salient) × distance cultural dimension	–	–	-0.132*** (0.018)	–	–	-0.097*** (0.039)
  Cultural dimension highly salient (ref. not salient) × distance cultural dimension	–	–	-0.272*** (0.028)	–	–	-0.358*** (0.068)
  N	93,086	93,086	93,086	32,312	32,312	32,312

  Note: Standard errors in parentheses; *** p< 0.01; the models control for gender, age, education, political interest and the alternative specific constants. See Appendix 2 for full model.

  The main effects in model 1 show that, both in Flanders and in Francophone Belgium, the positional distance measures work as expected: as the distance between a voter and a party increases, the likelihood that this voter would vote for that party diminishes strongly; this logic applies similarly to both dimensions. Things become more complicated in models 2 and 3 that test the core of our argument that salience leads to attributing more weight to one dimension in comparison to the other. Model 2 shows that, in comparison to voters who do not consider the economic dimension to be salient (none of the two issues they marked as important was related to the economic dimension), the voters who do consider economic issues to be moderately (one economic issue marked) or highly salient (two economic issues marked) give economic distance a larger weight in their voting decision: the interaction coefficients are negative and for those who attribute the most importance to the economic dimension the size of the coefficient is larger than for those who only attribute moderate importance to the economic dimension. Interestingly, and in line with our argument, interacting the salience of the economic dimension with cultural distance strongly diminishes the negative effect. In other words, voters who consider the economic dimension to be (much) more important than the cultural one are much less bothered about their cultural differences with the parties. Model 3 presents the same logic but this time interacts the distance measures with the salience of the cultural dimension. Results are remarkably identical. Cultural salience has very similar effects on the vote compared to the effects of economic salience: it further increases the negative effect of cultural distance on the likelihood to vote for a party and strongly diminishes the negative effects of economic distance. Remarkably and reinforcing our argument, these results are almost exactly the same in Flanders and Francophone Belgium. To sum up, in Table 3, simply all coefficients are in line with our argument. This is a pretty strong result, we believe.
  The conditional logit models presented so far underpin the powerful mechanism at play here (see effect size via marginal effect plots in Appendix 3). The results show that the position on the two dimensions, combined with the salience of each dimension, exerts a strong effect on party vote. However, results so far do not demonstrate that variations in salience made the unserved voters in quadrants 1 (left-right) and 4 (right-left) end up voting for parties in quadrants 2 (right-right) or 3 (left-left). In other words, we did not yet establish the pivotal role of salience making unserved voters rotate to the left or to the right. Therefore, we need to focus especially on the voters in the unserved quadrants 1 and 4 and change our dependent variable to voting for a party in the consistently right quadrant 2 or the consistently left quadrant 3. Below, we present the results of two logistic regressions: one for the economic left and cultural right quadrant 1 voters and one for the economic right and cultural left quadrant 4 voters. The question is does salience determine in which quadrant these atypical voters do end up voting? Note that, in these models, we discarded the CD&V voters in Flanders and the Défi voters in Francophone Belgium as these voted for a party that is not in a quadrant but almost right in the middle (CD&V) or on the edge between two quadrants (Défi) (see Figure 1).⁶x Note that voters who experience cross pressures from their contradictory positioning on the two dimensions and from the absence of a party that fully answers their demand may also end up voting for a centre party (such as CD&V in Flanders). Instead of pivoting to the right or the left, another escape route out of the dilemma for cross-pressurised voters might be to ‘flee’ to the centre party (that does not match the voter on any of the two dimensions but that may be, overall, the party that is positionally least distant to these voters’ preferences). The quadrant 2 parties are N-VA, Open Vld and Vlaams Belang in Flanders and MR and PP in Francophone Belgium; the quadrant 3 parties are sp.a, Groen and PVDA in Flanders and PS, Ecolo, CdH, and PTB in Francophone Belgium. Further, it is important to mention that voters situated in the centre on one of the dimensions (between -1 and +1) were excluded from the analyses as well: they cannot be considered as clearly belonging to a specific quadrant and they should experience cross pressures much less due to their atypical combination of policy preferences. This reduces the number of observations and zooms in strongly on the crucial category of voters. As our initial sample is large, we still keep enough voters in the models below. Table 4 shows the results of two logistic regressions with voting (1) for quadrant 3 parties (consistently left-wing) compared to voting (0) for quadrant 2 parties (consistently right-wing) as a dependent variable.
  Results are quite convincing and confirm our expectations in both parts of the country. In all models, the salience variable is highly significant, explaining a good deal of the swap of these two groups of unserved voters to the left or to the right. Voters holding an economic left-wing position and who combine it with a right-wing position on the cultural dimension tend to switch to left-wing parties when they consider economic issues to be more important compared to cultural issues. These voters prioritise their economic positions over their cultural stances. A similar logic, but then reversed, applies to voters who combine economic right-wing with cultural left-wing ideas. They tend to vote for right-wing parties if they care more about economic than about cultural issues (see the negative coefficient). To sum up, we can confirm that salience acts as a pivot that makes unserved voters rotate to the left- or the right-wing party quadrant. To further demonstrate the strength of these effects Figure 4 shows the predicted values across the values of the salience scale for Flemish and Francophone voters separately. In the graphs, on the left (-2) are the voters who consider cultural issues most important, on the right (+2) those who consider economic issues most important. Again, results are consistent in Flemish and Francophone Belgium and the graphs suggest the effect of salience to be quite potent on every step of the scale.

  Table 4 Logistic regressions predicting unserved voters’ party vote for a left-wing (Q3) party instead of a right-wing (Q2) party

  	Flanders		Francophone Belgium
  	Quadrant 1 voters (economic left + cultural right)	Quadrant 4 voters (economic right + cultural left)	Quadrant 1 voters (economic left + cultural right)	Quadrant 4 voters (economic right + cultural left)
  Male	0.161	-0.228	0.424	-0.239
  	(0.205)	(0.142)	(0.303)	(0.321)
  Age	-0.022***	-0.015***	0.015	-0.015
  	(0.007)	(0.004)	(0.009)	(0.010)
  Middle education (ref. cat. = lower)	0.030	-0.028	-0.430	-1.063
  	(0.267)	(0.433)	(0.418)	(0.797)
  Higher education (ref. cat. = lower)	0.386	0.352	-0.231	-1.215
  	(0.256)	(0.409)	(0.402)	(0.760)
  Political interest	-0.101*	-0.020	0.020	-0.081
  	(0.059)	(0.049)	(0.071)	(0.088)
  Salience (from cultural to economic)	0.810***	-0.527***	0.232**	-0.523***
  	(0.0713)	(0.058)	(0.100)	(0.113)
  Constant	0.543	-0.011	-0.199	2.373
  	(0.655)	(0.590)	(0.786)	(1.131)
  N	904	1.119	359	273

  Standard errors in parentheses; ***p < 0.01, **p < 0.05, *p < 0.1.

  Plot of the effect of salience change on party vote for a left-wing (Q3) party instead of a right-wing (Q2) party
  

• 5 Conclusion

  When the structure of political demand by voters and political supply by parties is similar, voters are able to cast a vote for the most proximate party. But if demand and supply are structured according to (partially) different dimensions, voters are forced to prioritise one set of preferences above another set. Which set of preferences will exert a larger effect on a voter’s party choice depends on the salience of their preference sets. This was the argument we made and demonstrated empirically with regard to the 2019 federal elections in Flanders and Francophone Belgium. Our theoretical claim in itself may not be entirely new or surprising but we do believe that the measures we use in this study – especially the fact that we could position voters and parties in the same space using identical (and many) issues and that we had good measures of people’s dimension salience – are original and improve on what earlier work could do.
  Concretely, in both Belgian regions party supply is one dimensional as parties’ economic and cultural left-right offers largely converge. In contrast, although voter demand partially converges along the same unified dimension, a good deal of Flemish and Francophone voters hold preferences that are structured two dimensionally. They hold economic left-wing views and combine this with cultural right-wing views, or vice versa. In our study the share of these ‘deviant’ voters was estimated as being between one- third (Francophone Belgium) and one- fourth (Flanders). When casting a party vote, these voters use the salience of their economic or cultural preferences to break the tie and arbitrate between economic and cultural proximity. The effect of salience is quite potent. Our data offer further confirmation of the notion that salience is a key driver of electoral choice especially for cross-pressured voters.
  Our findings about the role played by salience have implications for election outcomes not just in Belgium, of course, but in other places as well. A key matter is that issue salience is more variable than issue positions are. Once people have adopted a certain position with regard to a policy issue, their position becomes rather unmalleable. Yet, in contrast, the salience of their considerations is quite variable (Dennison, 2019). For instance, hundreds of studies have shown that issue salience is, amongst others, affected by short-term variations in media coverage (they all follow in the footsteps of the initial agenda-setting study by McCombs & Shaw, 1972). Salience is, more than positional considerations, the dynamic force producing electoral change. This is also what van der Brug and van Spanje (2009) implicitly conclude when they say that the mismatch between voters and parties is a driver of electoral volatility (as the salience that basically generates a left- or a right-wing vote is variable).
  Of course, our study has several shortcomings. First, we only focused on the ‘big’ trade-offs voters have to make when they are situated in a quadrant that is unserved by any party. Salience weights have a dramatic effect on these voters’ behaviour as they arbitrate between party options that are very distant from one another. They explain why, for instance in Flanders, a socialist party voter (sp.a), may swap to voting for the radical right party (Vlaams Belang) – a party that is almost the furthest away from his/her initial socialist party choice. Yet, salience also plays a role for the ‘smaller’ decisions voters have to take within the quadrant of their choice. Consistently, left- or right-wing voters still have to pick one party from among the offer of several left- or right-wing parties. For instance, left-wing voters in Flanders have the choice between, among others, sp.a and Groen, and we expect dimension salience, in combination with the two left-right dimensions, to play a role in that decision as well. If the cultural dimension prevails left-wing, Flemish voters will tend to flock towards Groen; if the economic dimension is most salient then both sp.a and Groen are equally attractive as they have a similar position on that dimension. In these minor exchanges between adjacent parties, of course, next to positional proximity and salience, other considerations such as issue ownership and directional voting probably play an important role as well.
  Second, we did not tackle the question why the Flemish and Francophone party systems have blind spots and do not contain a party offer combining, for instance, a set of economic left with cultural right positions. In other words, why are there no parties in quadrants 1 and 4? How come the two- party systems we look here have not adapted? It cannot be a coincidence that in most countries this particular quadrant of the political market is empty (Lefkofridi et al., 2014; van der Brug & van Spanje, 2009). There could be ideological reasons. It seems as if economic left-wing positions are ‘naturally’ matching cultural left-wing positions, just like economic right-wing positions tend to be complemented mechanically by right-wing cultural preferences. This explains why, in many countries, socialist parties have a hard time dealing with the migration issue, for instance. A tough stance on migration appears ideologically hard to combine with a soft position on redistribution. Both migration and redistribution are inherently matters of solidarity and it appears tricky to ideologically combine more (with nationals) with less solidarity (with foreigners).
  Finally, and most importantly, although our evidence with no less than 20 policy items per dimension and, especially, identical and direct measures of the positions of both voters and parties is unique, a major drawback is the unrepresentative character of our citizen sample: our respondents are more educated and more interested in politics than the average Flemish/Francophone citizen. We need to be very careful in making claims about the general voter in both the Belgian regions. It would be great to be able to replicate the study on a better sample. We believe it to be likely that the underlying mechanism, the combination of two positional dimensions and their weighing based on salience, would be the same but that cannot be guaranteed. On the one hand, due to the fact that the correlation between both left-right dimensions is most likely smaller in the population as a whole compared to what we find in our sample, even a larger group of citizens in the population as a whole are unserved and the role of salience in the trade-off between the two dimensions may even be larger. Yet, on the other hand, the fact that our respondents tend to be more educated and interested in politics may make that they take policy considerations more into account than other voters which makes the mechanism highlighted here more prevalent in this group. In any case, the fact that we find as good as identical results in Flemish and Francophone Belgium, although on a clearly skewed sample, remember that these are entirely separate party systems, reinforces confidence that our results may have value outside of our sample and even outside of the borders of the beautiful kingdom of Belgium.

References

• Bartolini, S. & Mair, P. (2007). Identity, competition and electoral availability: The stabilisation of European electorates, 1885-1985. ECPR Press.

• Bakker, R., Hooghe, L., Jolly, S., Marks, G., Polk, J., Rovny, J., Steenbergen, M. & Vachudova, M. A. (2020). 2019 Chapel Hill expert survey. Version 2019.1. www.chesdata.eu. University of North Carolina.

• Converse, P. E. (1964). The nature of belief systems in mass publics. In D. E. Apter (Ed.), Ideology and discontent (pp. 206-261). Macmillan.

• Dennison, J. (2019). A review of public issue salience: Concepts, determinants and effects on voting. Political Studies Review, 17(4), 436-446.

• Feld, S., Merrill III, S. & Grofman, B. (2014). Modelling the effects of changing issue salience in two-party competition. Public Choice, 158, 465-482.

• Hooghe, L., Marks, G. & Wilson, C. J. (2002). Does left/right structure party positions on European integration? Comparative Political Studies, 35(8), 965-989.

• Huber, J. & Inglehart, R. (1995). Expert interpretations of party space and party locations in 42 societies. Party Politics, 1(1), 73-111.

• Inglehart, R. (1990). Culture shift in advanced industrial society. Princeton University Press.

• Kitschelt, H. (2004). Diversification and reconfiguration of party systems in postindustrial democracies. Bonn: Friedrich Ebert Stiftung.

• Kriesi, H., Grande, E., Lachat, R., Dolezal, M., Bornschier, S. & Frey, T. (2008). West European politics in the age of globalization. Cambridge University Press.

• Lefkofridi, Z., Wagner, M. & Willmann, J. E. (2014). Left-authoritarians and policy representation in Western Europe: Electoral choice across ideological dimensions. West European Politics, 37(1), 65-90.

• Lipset, S. M. & Rokkan, S. (1967). Party systems and voter alignments: Cross-national perspectives. MacMillan.

• Marks, G., Hooghe, L., Nelson, M. & Edwards, E. (2006). Party competition and European integration in the east and west: Different structure, same causality. Comparative Political Studies, 39(2), 155-175.

• McCombs, M. E. & Shaw, D. (1972). The agenda-setting function of mass media. Public Opinion Quarterly, 69(4), 813-824.

• Riker, W. H. (1982). Liberalism against populism: A confrontation between the theory of democracy and the theory of social choice. Freeman.

• Rokkan, S. (1977). Towards a generalized concept of verzuiling: A preliminary note. Political Studies, 25(4), 563-570.

• Rosset, J., Lutz, G. & Kissau, K. (2016). Representation of political opinions: Is the structuring pattern of policy preferences the same for citizens and elites? In M. Bühlman & J. Fivaz (Eds.), Political representation. roles, representatives and the represented (pp. 117-134). Routledge.

• Schattschneider, E. E. (1960). The semi-sovereign people. Holt.

• Thomassen, J. (1999). Political communication between political elites and mass publics. The role of belief systems. In W. E. Miller, R. Pierce & J. Thomassen (Eds.), Policy representation in western democracies (pp. 33-58). Oxford University Press.

• Thomassen, J. (2012). The blind corner of political representation. Representation, 48(1), 13-27.

• Van Der Brug, W. & Van Spanje, J. (2009). Immigration, Europe and the ‘new’ cultural dimension. European Journal of Political Research, 48(3), 309-334.

• Wheatley, J. (2015). Restructuring the policy space in England: The end of the left-right paradigm? British Politics, 10(3), 268-285.

Appendix 1: Full list of policy proposals

Flanders

Policy proposals related to the economic cleavage

1. Eigenaars van meerdere huizen moeten meer belast worden

2. Eerder dan zelf sociale woningen te bouwen, moet de overheid mensen helpen om op de privémarkt te huren.

3. In elke gemeente zou een vast aandeel van de woningen sociale woningen moeten zijn.

4. Europa moet hard optreden tegen lidstaten die zich niet houden aan de Europese begrotingsregels.

5. Europa moet hard optreden tegen lidstaten die grote belastingvoordelen toekennen aan sommige bedrijven.

6. In elke lidstaat van de EU moet er een kinderbijslag ingevoerd worden.

7. In heel Europa moet er een minimumloon ingevoerd worden.

8. De belasting op de winst van bedrijven moet omlaag.

9. Het vaderschapsverlof moet uitgebreid worden naar meer dan 10 dagen.

10. Naar de huisarts gaan, moet volledig gratis worden.

11. De mogelijkheid om in de horeca bij te verdienen tegen een verlaagd belastingtarief (flexi-jobs) moet afgeschaft worden.

12. Winkels mogen zelf kiezen wanneer ze solden doen.

13. Voor pendelaars moet de trein gratis zijn.

14. De fiscale voordelen voor pensioensparen moeten vervangen worden door een verhoging van de wettelijke pensioenen.

15. Omdat leerkrachten een zwaar beroep uitoefenen, moeten ze vervroegd met pensioen kunnen gaan.

16. Leefloners moeten gemeenschapswerk doen.

17. Werklozen moeten hun uitkering na een tijd verliezen.

18. De VRT mag bedrijven niet langer laten betalen om hun producten in programma’s te tonen.

19. Ook in het middelbaar onderwijs moet er een maximumfactuur voor de ouders komen.

20. De inkomensgrens voor huursubsidies moet omhoog, zodat meer mensen er recht op hebben.

Policy proposals related to the cultural cleavage

21. Europa moet meer genetisch gewijzigde gewassen toelaten.

22. Er moet een Europese taks op vliegtuigbrandstof komen.

23. Europa mag enkel handelsakkoorden afsluiten met landen die de mensenrechten respecteren.

24. Er moet een Europese identiteitskaart komen die de nationale vervangt.

25. De EU moet beslissen of het zomeruur of winteruur wordt ingevoerd, niet de lidstaten.

26. De overheid moet praktijktesten doen om te controleren op discriminatie bij aanwervingen.

27. De verkoop van nieuwe benzine- en dieselwagens moet verboden worden vanaf 2030.

28. Ook na 2025 moeten we kerncentrales openhouden.

29. De aankoop van de nieuwe gevechtsvliegtuigen moet worden teruggedraaid.

30. Belangrijke politieke beslissingen moeten via een referendum aan burgers kunnen worden overgelaten.

31. Burgers moeten zelf kunnen kiezen of ze gaan stemmen.

32. Scholen moeten kinderen verplichten om ook op de speelplaats Nederlands te praten.

33. De meest vervuilende auto’s moeten verboden worden in heel Vlaanderen (lage-emissiezone).

34. Over de besteding van een deel van de Vlaamse begroting moeten de burgers rechtstreeks zelf kunnen beslissen.

35. Boerkini’s moeten toegelaten worden in openbare zwembaden.

36. De Vlaamse overheid moet praktijktesten doen om discriminatie bij het verhuren van woningen op te sporen.

37. De Vlaamse overheid mag geen nieuwe moskeeën meer erkennen.

38. In het gemeenschapsonderwijs mogen leerlingen op school geen hoofddoek dragen.

39. Nieuwkomers moeten maximaal hun oorspronkelijke cultuur en gewoonten vervangen door de Vlaamse cultuur en gewoonten.

40. De betonstop moet sneller uitgevoerd worden.

Wallonia

Policy proposals related to the economic cleavage

1. La Wallonie doit moins investir dans des logements sociaux et aider financièrement les citoyens à louer sur le marché privé

2. Il faut une nouvelle taxe sur les bénéfices des banques dans l’Union européenne

3. L’Europe doit agir fermement contre les États membres qui ne respectent pas les règles budgétaires européennes

4. Le profit obtenu lors de la vente d’actions doit toujours être taxé

5. Les impôts sur les bénéfices des sociétés doivent diminuer

6. Plus de 8 consultations chez un psychologue doivent pouvoir être remboursées par an

7. Il faut supprimer la possibilité, dans l’HORECA, d’offrir des ‘flexi-jobs’ moins taxés

8. Il doit être plus facile de licencier des travailleurs

9. Une semaine de 4 jours doit être introduite, en conservant le même salaire

10. Les allocations de chômage doivent être versées par l’administration, et non par les syndicats

11. Les autorités doivent contrôler plus sévèrement si les personnes qui reçoivent un revenu d’intégration sociale possèdent une propriété à l’étranger

12. Les propriétaires qui mettent un logement en location via AirBnB doivent pouvoir le faire sans limitation de nombre de jours par an

13. Les droits de succession doivent être diminués

14. Les propriétaires de plusieurs biens immobiliers doivent être plus taxés

15. La Wallonie doit arrêter d’investir dans des industries en difficulté

16. Les TEC doivent être privatisés

17. Les règles d’implantion des centres commerciaux doivent être plus strictes

18. Le prix des maisons de repos publiques doit être proportionnel au montant de l’allocation de retraite

19. Les allocations de rentrée scolaire doivent être augmentées pour les revenus les plus faibles

20. Les parents qui travaillent doivent avoir la priorité pour l’accès aux crèches

Policy proposals related to the cultural cleavage

21. La Wallonie doit soutenir fiscalement l’installation de bornes de recharge pour voitures électriques

22. Le port du voile islamique doit être interdit pour le personnel au guichet des administrations

23. Le gouvernement wallon doit compter autant d’hommes que de femmes

24. Une partie du budget wallon doit être décidée via une consultation citoyenne

25. Il faut une taxe européenne sur le kérosène

26. Il faut créer une armée européenne commune pour remplacer les armées nationales

27. Les contrôles aux frontières entre les États membres doivent être rétablis

28. Héberger des migrants ‘en transit’ doit être un délit punissable

29. Il faut taxer les billets d’avion pour qu’ils soient plus chers

30. Les centrales nucléaires doivent rester opérationnelles après 2025

31. Une troisième catégorie de sexe ‘neutre’ doit être prévue sur la carte d’identité

32. L’avortement doit aussi être autorisé au-delà de 12 semaines de grossesse

33. Le gouvernement doit s’excuser pour les atrocités pendant la colonisation du Congo

34. Les empreintes digitales de tous les citoyens doivent être conservées dans une base de données centrale

35. Tous les personnes condamnées à des peines de prison doivent purger leur peine jusqu’au bout

36. Les décisions politiques importantes doivent pouvoir être laissées aux citoyens par le biais d’un référendum

37. Un nombre minimum de personnes d’origine immigrée doit être engagé dans l’administration wallonne

38. Le parcours d’intégration doit être obligatoire pour tous les nouveaux immigrés

39. Les étrangers en situation irrégulière en Wallonie doivent bénéficier d’une assistance publique minimale

40. La vitesse maximale sur les routes nationales (90 km/h) doit être diminuée

Appendix 2: Full conditional logit models

Flanders

	Model 1 (main effects)	Model 2 (economic salience interactions)	Model 3 (cultural salience interactions)
eco_dis_	-0.393***	-0.26***	-0.447***
	(0.009)	(0.016)	(0.015)
cult_dis_	-0.421***	-0.543***	-0.31***
	(0.008)	(0.016)	(0.015)
Salience economic 1 *		-0.146***
distance economic dimension		(0.018)
Salience economic 2 *		-0.24***
distance economic dimension		(0.026)
Salience economic 1 *		0.138***
distance cultural dimension		(0.019)
Salience economic 2 *		0.277***
distance cultural dimension		(0.026)
Salience cultural 1 *			0.073***
distance economic dimension			(0.018)
Salience cultural 2 *			0.185***
distance economic dimension			(0.028)
Salience cultural 1 *			-0.132***
distance cultural dimension			(0.018)
Salience cultural 2 *			-0.272***
distance cultural dimension			(0.029)
Man_groen	-0.134	-0.119	-0.115
	(0.082)	(0.085)	(0.085)
Man_nva	-0.204***	-0.195**	-0.189**
	(0.079)	(0.08)	(0.08)
Man_vld	-0.341***	-0.36***	-0.355***
	(0.087)	(0.087)	(0.087)
Man_pvda	-0.177*	-0.158*	-0.15
	(0.092)	(0.094)	(0.094)
Man_spa	-0.155*	-0.145	-0.143
	(0.09)	(0.092)	(0.091)
Man_vb	-0.085	-0.101	-0.098
	(0.105)	(0.108)	(0.108)
Age_groen	-0.026***	-0.024***	-0.024***
	(0.003)	(0.003)	(0.003)
Age_nva	0.016***	0.016***	0.016***
	(0.002)	(0.002)	(0.002)
Age_vld	-0.015***	-0.015***	-0.015***
	(0.003)	(0.003)	(0.003)
Age_pvda	-0.009***	-0.013***	-0.013***
	(0.003)	(0.003)	(0.003)
Age_spa	0.006*	0.002	0.002
	(0.003)	(0.003)	(0.003)
Age_vb	-0.015***	-0.014***	-0.014***
	(0.003)	(0.003)	(0.003)
Edu_mid_groen	0.023	0.013	0.015
	(0.179)	(0.186)	(0.186)
Edu_mid_nva	0.132	0.079	0.083
	(0.146)	(0.148)	(0.148)
Edu_mid_vld	0.219	0.21	0.215
	(0.194)	(0.195)	(0.195)
Edu_mid_pvda	-0.258	-0.151	-0.152
	(0.173)	(0.175)	(0.175)
Edu_mid_spa	-0.213	-0.101	-0.101
	(0.17)	(0.172)	(0.172)
Edu_mid_vb	-0.394**	-0.376**	-0.373**
	(0.159)	(0.164)	(0.164)
Edu_high_groen	0.242	0.186	0.181
	(0.164)	(0.172)	(0.172)
Edu_high_nva	-0.421***	-0.521***	-0.515***
	(0.135)	(0.138)	(0.138)
Edu_high_vld	0.264	0.233	0.244
	(0.181)	(0.182)	(0.182)
Edu_high_pvda	-0.886***	-0.692***	-0.702***
	(0.16)	(0.163)	(0.163)
Edu_high_spa	-0.703***	-0.489***	-0.5***
	(0.156)	(0.159)	(0.159)
Edu_high_vb	-1.626***	-1.627***	-1.621***
	(0.152)	(0.157)	(0.157)
PI_groen	-0.029	-0.005	-0.007
	(0.027)	(0.028)	(0.028)
PI_nva	0.033	0.027	0.028
	(0.025)	(0.025)	(0.025)
PI_vld	0.004	0.007	0.006
	(0.028)	(0.029)	(0.029)
PI_pvda	-0.042	-0.046	-0.044
	(0.031)	(0.031)	(0.031)
PI_spa	0.067**	0.065**	0.064**
	(0.031)	(0.031)	(0.031)
PI_vb	-0.043	-0.033	-0.031
	(0.031)	(0.032)	(0.032)
asc_groen	3.32***	2.083***	1.987***
	(0.301)	(0.343)	(0.342)
asc_nva	1.631***	2.289***	2.248***
	(0.279)	(0.308)	(0.307)
asc_vld	1.53***	1.24***	1.488***
	(0.319)	(0.351)	(0.35)
asc_pvda	3.269***	1.45***	1.705***
	(0.33)	(0.385)	(0.385)
asc_spa	0.405	-0.793**	-0.669*
	(0.332)	(0.376)	(0.376)
asc_vb	2.777***	1.856***	1.311***
	(0.335)	(0.395)	(0.393)
Salience_Eco_groen		-0.281***	-0.215***
		(0.088)	(0.083)
Salience_Eco_nva		-0.576***	-0.619***
		(0.078)	(0.069)
Salience_Eco_vld		0.319***	0.139*
		(0.083)	(0.08)
Salience_Eco_pvda		1.248***	1.067***
		(0.114)	(0.106)
Salience_Eco_spa		0.874***	0.8***
		(0.097)	(0.095)
Salience_Eco_vb		-0.333***	-0.036
		(0.112)	(0.107)
Salience_Cult_groen		1.176***	1.278***
		(0.09)	(0.094)
Salience_Cult_nva		-0.023	0.011
		(0.075)	(0.082)
Salience_Cult_vld		-0.012	-0.127
		(0.087)	(0.091)
Salience_Cult_pvda		0.73***	0.625***
		(0.116)	(0.123)
Salience_Cult_spa		0.344***	0.304***
		(0.105)	(0.106)
Salience_Cult_vb		1.168***	1.468***
		(0.108)	(0.113)
N	93086	93086	93086
R2	0.28	0.325	0.324

Standard errors are in parentheses
*** p < 0.01, ** p < 0.05, * p < 0.1

Wallonia

	Model 1 (main effects)	Model 2 (economic salience interactions)	Model 3 (cultural salience interactions)
eco_dis_	-0.54***	-0.396***	-0.601***
	(0.018)	(0.034)	(0.028)
cult_dis_	-0.401***	-0.511***	-0.28***
	(0.02)	(0.037)	(0.033)
Salience economic 1 *		-0.162***
distance economic dimension		(0.038)
Salience economic 2 *		-0.236***
distance economic dimension		(0.053)
Salience economic 1 *		0.158***
distance cultural dimension		(0.041)
Salience economic 2 *		0.251***
distance cultural dimension		(0.06)
Salience cultural 1 *			0.081**
distance economic dimension			(0.035)
Salience cultural 2 *			0.237***
distance economic dimension			(0.061)
Salience cultural 1 *			-0.097**
distance cultural dimension			(0.039)
Salience cultural 2 *			-0.358***
distance cultural dimension			(0.068)
Man_defi	-0.152	-0.172	-0.166
	(0.181)	(0.181)	(0.181)
Man_ecolo	-0.316**	-0.339**	-0.343**
	(0.132)	(0.137)	(0.137)
Man_mr	-0.334**	-0.321**	-0.319**
	(0.142)	(0.143)	(0.142)
Man_pp	-0.302	-0.383	-0.385
	(0.279)	(0.286)	(0.288)
Man_ps	-0.15	-0.101	-0.107
	(0.143)	(0.145)	(0.145)
Man_ptb	0.004	0.04	0.035
	(0.158)	(0.16)	(0.16)
Age_defi	-0.025***	-0.024***	-0.024***
	(0.005)	(0.005)	(0.005)
Age_ecolo	-0.03***	-0.027***	-0.027***
	(0.004)	(0.004)	(0.004)
Age_mr	-0.02***	-0.019***	-0.019***
	(0.004)	(0.004)	(0.004)
Age_pp	-0.034***	-0.036***	-0.036***
	(0.008)	(0.008)	(0.008)
Age_ps	-0.001	-0.004	-0.004
	(0.004)	(0.005)	(0.005)
Age_ptb	-0.027***	-0.031***	-0.031***
	(0.005)	(0.005)	(0.005)
Edu_mid_defi	1.034***	1.072***	1.071***
	(0.351)	(0.352)	(0.352)
Edu_mid_ecolo	0.395	0.471	0.476
	(0.297)	(0.304)	(0.304)
Edu_mid_mr	0.414	0.45	0.44
	(0.303)	(0.304)	(0.303)
Edu_mid_pp	1.61***	1.517***	1.513***
	(0.463)	(0.476)	(0.479)
Edu_mid_ps	1.42***	1.321***	1.319***
	(0.284)	(0.287)	(0.287)
Edu_mid_ptb	1.677***	1.595***	1.595***
	(0.299)	(0.302)	(0.301)
Edu_high_defi	0.318*	0.357*	0.357*
	(0.189)	(0.19)	(0.19)
Edu_high_ecolo	0.05	0.136	0.137
	(0.143)	(0.149)	(0.149)
Edu_high_mr	0.126	0.149	0.141
	(0.147)	(0.148)	(0.148)
Edu_high_pp	1.291***	1.248***	1.25***
	(0.25)	(0.258)	(0.26)
Edu_high_ps	0.66***	0.516***	0.514***
	(0.147)	(0.15)	(0.15)
Edu_high_ptb	0.836***	0.736***	0.736***
	(0.159)	(0.162)	(0.162)
PI_defi	0.021	0.022	0.02
	(0.048)	(0.049)	(0.049)
PI_ecolo	-0.049	-0.047	-0.048
	(0.035)	(0.037)	(0.037)
PI_mr	0.081**	0.087**	0.085**
	(0.037)	(0.037)	(0.037)
PI_pp	-0.004	0.001	-0.002
	(0.066)	(0.067)	(0.068)
PI_ps	0.09**	0.078**	0.079**
	(0.04)	(0.04)	(0.04)
PI_ptb	0.008	0.005	0.005
	(0.042)	(0.042)	(0.042)
asc_defi	0.778	0.506	0.425
	(0.485)	(0.54)	(0.542)
asc_ecolo	5.284***	4***	3.818***
	(0.365)	(0.43)	(0.43)
asc_mr	1.585***	1.003**	1.149***
	(0.385)	(0.428)	(0.429)
asc_pp	0.785	-0.435	-0.377
	(0.703)	(0.844)	(0.842)
asc_ps	1.458***	0.356	0.603
	(0.412)	(0.472)	(0.471)
asc_ptb	3.063***	1.277**	1.44***
	(0.432)	(0.515)	(0.514)
Salience_Eco_defi		-0.099	-0.034
		(0.158)	(0.157)
Salience_Eco_ecolo		-0.127	-0.088
		(0.152)	(0.124)
Salience_Eco_mr		0.348***	0.268**
		(0.122)	(0.12)
Salience_Eco_pp		0.111	0.105
		(0.291)	(0.289)
Salience_Eco_ps		1.026***	0.849***
		(0.153)	(0.134)
Salience_Eco_ptb		1.253***	1.124***
		(0.182)	(0.164)
Salience_Cult_defi		0.39**	0.447***
		(0.168)	(0.169)
Salience_Cult_ecolo		1.233***	1.393***
		(0.133)	(0.165)
Salience_Cult_mr		0.262**	0.166
		(0.133)	(0.135)
Salience_Cult_pp		1.487***	1.436***
		(0.291)	(0.292)
Salience_Cult_ps		0.267*	0.147
		(0.149)	(0.169)
Salience_Cult_ptb		0.882***	0.815***
		(0.177)	(0.197)
N	32312	32312	32312
R2	0.281	0.328	0.329

Standard errors are in parentheses
*** p < 0.01, ** p < 0.05, * p < 0.1

Appendix 3: Effect sizes of conditional logit models

Flanders





Francophone Belgium

Noten

• 1 If we just weigh the Flemish sample based on age, sex and education, the weights soar and reach levels of 14 and more (lower educated women). This means that one person would get the weight of 14 ‘standard’ individuals. Such enormous weights are unwanted as too much depends on the answers of just a few people. Still, weighing the data with such large weights does not change the results. We tested it in Flanders and there hardly were differences, neither in the conditional logit model nor in the multinomial model. The coefficients of the socio-demographic control variables change a little but the substantive results are the same.

• 2 The CHES score for the cultural dimension is based on the average of the party scores on EU, immigration, multiculturalism, personal freedom and environment.

• 3 One could argue that our simple calculation of the four quadrants is too rude and that many voters are actually centre voters. This is true, of course. If we define all voters who, on our -5 to +5 scale, score between 0 and ±1, as centre voters, we see that 13.3% (FL) and 20.4% (WAL) can be considered to be left nor right on neither of the dimensions. This stricter operationalisation reduces the share of economic left + cultural right (FL: 8.8%, WAL: 9.9%) and of economic right + cultural left (FL: 11.2%, WAL: 7.2%) voters somewhat, but they remain considerable groups. Even more nuanced categorisations are possible whereby voters combine a left or right position on one dimension with a centre position on the other.

• 4 We dropped the voters who voted for other parties, who did not turn out, were not allowed to vote, voted blank, or forgot whom they voted for. This leaves us with 14,096 Flemish and 5,285 Francophone voters. We do not find that these unserved voters end up more in these non-party categories.

• 5 One assumption of the conditional logit model is the assumption of independence of irrelevant alternatives (IIA). A Haussmann test shows that our main model actually violates this assumption. For that reason, we also ran a mixed logit choice model of the main model, which loosens this assumption but takes more computational time. This alternative model results in identical results, which is why we continue with the conditional logit models.

• 6 Note that voters who experience cross pressures from their contradictory positioning on the two dimensions and from the absence of a party that fully answers their demand may also end up voting for a centre party (such as CD&V in Flanders). Instead of pivoting to the right or the left, another escape route out of the dilemma for cross-pressurised voters might be to ‘flee’ to the centre party (that does not match the voter on any of the two dimensions but that may be, overall, the party that is positionally least distant to these voters’ preferences).