Preference votes without preference? Institutional effects on preference voting: an experiment

ABSTRACT

A classical question of political science is to what extent electoral systems influence voting behaviour. Yet, many of these studies examine how different electoral systems affect the election results in terms of vote distribution across parties. Instead, we investigate how electoral rules affect intra party preference voting. Given the importance of the debate on the personalization of politics, insight into how electoral rules shape intra-party choice is a valuable contribution to the literature. In our study, we focus on the effect of two specific rules: the option to cast a list vote and on a single versus multiple preference votes. The results of experiments conducted in Belgium and the Netherlands show that electoral rules indeed influence voting behaviour with regard to intra party preference voting, although differences exist between the Netherlands and Belgium. Moreover, we find that the option to cast a list vote equally affects votes for the first candidate on the list, as well as lower positioned candidates. This suggests that preference votes might be less preferential than has often been assumed.

Introduction

Starting with Duverger (1951), the question how electoral rules influence electoral behaviour has been of central interest to political scientists (Blais and Carty 1991; Clark and Golder 2006; Cox 1997; Singer and Stephenson 2009). However, whereas most of these studies examine how different electoral systems influence election results in terms of vote distribution across parties, in this paper, we investigate how electoral rules influence voting for candidates within a party. Intra-party preference votes are an important feature in a number of European countries such as Belgium, Denmark, Latvia, The Netherlands, Slovenia, Switzerland, and Sweden (Katz 2003). For candidates, preference votes are important because they matter in determining which candidates take up the seats won by the party, influencing the composition of parliament. Furthermore, they reflect a candidate’s popularity and may influence a candidate’s ballot list position at the next election or help candidates to obtain more internal power within the party (André et al. 2015; Wauters, Verlet, and Ackaert 2012). The idea of intra-party preference voting also fits within the broader personalization thesis (McAllister 2007; Rahat and Sheafer 2007), which states that individual political actors are becoming more important at the expense of collective political actors. According to Karvonen (2010, 63) “in those countries where the possibility of choosing between individual candidates has existed for a long time, the relative importance of individual candidates seems to have increased”. For example, until the 1980s, in the Netherlands, the percentage of preference votes often stayed under 10%, whereas during the last elections, this has fluctuated around 20%, with a peak in 2002 of 27% (Van Holsteyn and Andeweg 2012). A similar trend can be found in Belgium, where compared to previous decades, much more voters cast a preference vote, peaking in 2003, when almost 70% of the voters casted a preference vote (André, Wauters, and Pilet 2012). At the same time, Karvonen (2010) did not find conclusive evidence for a trend towards more candidate-centred electoral systems. However, in an extensive study on electoral reforms, Renwick and Pilet (2016) came to a different conclusion. They do observe a trend towards more candidate-centred electoral systems, probably because they used a “more fine-grained classification of the intra-party dimension” (Renwick and Pilet 2016, 267). Examples are Bulgaria, where a closed list system was replaced by an open list system with preference voting in 2011 and Belgium, where rules were introduced in 2002, which increased the weight given to preference votes. Thus, intra-party preference voting is increasingly becoming important in many European countries and therefore, to fully understand voting behaviour, we also have to understand intra-party preference voting.

Previous research has examined which voters are more likely to cast preference votes and found that especially political interest is an important predictor (André, Wauters, and Pilet 2012; Marsh 1985; Van Holsteyn and Andeweg 2012). However, these studies focus mostly on individual characteristics of citizens to explain preference voting. Other studies look more specifically at institutions, classifying different electoral formulas and explaining how they might affect the vote-seeking behaviour of candidates (Carey and Shugart 1995). Yet, they do not examine how these institutions affect voters. With this paper, we aim to fill this gap. We test the influence not of an electoral system as such, but of variations of the same electoral system: flexible list systems (Carey and Shugart 1995). Flexible list proportional representation systems are amongst the most commonly used electoral systems in Europe. Yet, the system (and its effects) is understudied (André et al. 2015). In flexible list systems, both preference votes and the rank ordering of candidates by the party determine which candidates are elected. Therefore, the system should be distinguished from closed lists (where only the party rank ordering determines which candidates are elected) and open lists (where only preference votes determine the elected candidates) (Shugart 2005).

There is considerable variation between the different countries where flexible list systems are used with regard to how open or restrictive the system is. Probably, the most important distinction is whether casting a preference vote is optional or not. In some countries, such as Austria, Belgium, Luxemburg, and Sweden, voters can choose whether they vote for a candidate or cast a so-called list vote, a vote for the party list as a whole. In other countries, such as Finland and the Netherlands, voters cannot cast a list vote and preference voting is obligatory. Shugart (2005, 43) calls this subtype of the flexible list system a latent list. A second difference relates to the number of preference votes voters may cast (Karvonen 2011).¹ In some countries, voters may cast multiple preference votes. In other countries, voters are restricted to a single candidate. In this study, we focus on these two differences and therefore contribute to the understanding and effects of variations of preferential list systems of which “little is currently known” (Shugart 2005, 43).

Most studies on the effects of electoral systems answer their question by comparing election results between different systems with different electoral rules. Problematic with this approach, however, is that many factors might be involved which were not taken into account in the analysis. In response to this problem, a recent strand of literature has emerged which conducts experiments in order to study how voters react to changes in the electoral system (Blais et al. 2012; Blumenau et al. 2014; Laslier et al. 2015; Van der Straeten, Laslier, and Blais 2013). With this study, we follow this literature, and conduct an experiment to investigate the role of electoral rules on voting behaviour. An extra feature of our experiment is that it is executed in two countries: Belgium and the Netherlands. These two countries are the opposite of each other on the two rules we are interested in. The electoral system of Belgium allows voters to cast either a list vote or one or multiple preference votes. Belgian voters who support the party and have no preference for individual candidates (or agree with the party on the order of candidates) have the option to cast a list vote. Voters who have a preference for individual candidate(s) can cast multiple preference votes as long as these candidates belong to the same party (De Winter 2005). Dutch voters, on the other hand, are forced to vote for a single candidate (Andeweg 2005). Therefore, we should notice that in our experiment, the Dutch perspective is more interesting. Dutch voters – compared to Belgian voters – are more restricted in showing their actual preference. The experiment allows us to investigate what happens when these constrains are lifted. An extra benefit is that this gives us an indication of how preferential a preference vote is. Studies have often argued that due to the nature of the Dutch system, where voters are forced to cast a single preference vote, the vote for the list puller does not really reflect the preference for a candidate, but simply a choice for the party. With a survey, Van Holsteyn and Andeweg (2010) show that approximately only 30% of the voters who casted a vote for the list puller actually stated to have a preference for that candidate. Therefore, in general, only votes for candidates from the second position onwards are considered to be actual preference votes in the Netherlands (Andeweg 2005). With the use of our experiment, we can examine to what extent this is indeed the case by investigating how the introduction of a list vote affects the votes for list pullers and for other candidates. A second question our experiments enable us to answer is how voters with multiple preferences deal with the fact that they can only cast a single preference vote.

The benefit from conducting our experiment in two countries is that it enables us to test our expectations in two different contexts. Of course, we should stress that we still have to be careful when comparing the results of the experiments in both countries (we discuss this in more detail below). Our analysis will therefore focus on discussing the effects within a country first. However, conducting the results in two contexts and comparing their results allow bringing into account one limitation of the experimental design, namely that electoral system effects become real via a learning process. Thus, when introduced to new rules, in the beginning, respondents may still stick to their usual voting habits and behaviour, and only over time adopt the new rules. This may lead us to underestimate the effects of our treatments. Comparing the results between two countries, which currently stand opposite to each other on the rules we test, provides us an (tentative) insight into the extent of this underestimation.

Electoral rules and the influence on preference voting

Using an experiment to investigate the effects of electoral rules on the election outcome is in line with a recent trend. Blumenau et al. (2014), for example, study the effect of open versus closed list electoral systems on party choice. They demonstrate that voters under an open list system are likely to switch from niche parties to mainstream parties. In the study of Blumenau et al., this meant that some Eurosceptic voters voted for UKIP under the closed list rules and switch their vote to a Eurosceptic candidate of the conservative party under the open list rules. Other experiments were conducted as part of the “making electoral democracy work” project (Blais 2010). These experiments were conducted as quasi-experiments around real elections, inviting people to vote under different electoral systems. The general conclusion of these experiments is that a substantial number of voters vote differently in different systems, showing that voters are responsive to changes in electoral rules (Blais et al. 2012; Laslier et al. 2015; Van der Straeten, Laslier, and Blais 2013). We expect similar results with regard to preference voting. Especially for the Dutch case, where some respondents get “new” opportunities to express their vote, we expect that when these new rules become available, some Dutch voters will actively use the option to cast a list vote and others choose to cast multiple preference votes.

We are also interested in how rules shape the electoral outcome. As discussed earlier, usually in the Netherlands, only votes for candidates on the second place on the list or lower are regarded as actual preference votes, because the choice for the first candidate is often a choice for the party (Andeweg 2005, 494)². Therefore, it seems reasonable to expect that a substantial number of these voters would have casted a list vote if that option had been available to them. It is less likely that voters who voted for other candidates (i.e. candidates from the second position on the candidate list onwards) switch to a list vote. In general, these votes are considered as an expression of a real preference and, moreover, they are a more sophisticated type of voting behaviour (Marsh 1985), as casting a vote for a specific candidate who is not the list puller requires additional information about the candidates. The general assumption is that these voters made a conscious choice for a specific candidate within a party, so there is no reason to expect that a substantial number of them would switch to casting a list vote if that option were available. The consequence of this is that we can expect that in a situation where Dutch voters have the option to cast a list vote, this would reduce the votes cast for the list puller, but not so much for other candidates. The latter is seen as an actual preference vote, whereas the former is often a reflection of support for the party. These expectations are from a “Dutch” perspective, that is, from the perspective that the list vote becomes available.

So, would we expect different outcomes if the perspective is the other way around: if the option to cast a list vote disappears, that is, from the Belgian perspective? Voters who normally cast a list vote – accepting the order of the list – need to change their behaviour if the option to cast a list vote disappears. We expect that especially the list puller will profit from this. André et al. show that voters with less political interest and sophistication are more likely to cast list votes, because party labels “act as brand names from which rationally ignorant and risk-averse voters can readily infer information about the issue positions and policy commitment of all candidates a party endorsed” (André, Wauters, and Pilet 2012, 296). When these less sophisticated voters can no longer use the easy option of a list vote, they have to find an alternative (easy) solution. We expect that the most likely heuristic for these voters would be to rely on the ballot list position: they will cast a vote for the list puller, instead of a vote for another candidate. This bias towards the first option on a list can be interpreted as a primacy effect (see Brockington 2003; Miller and Krosnick 1998). Consequently, we expect that once voters can no longer cast a list vote, it is especially the list puller profiting from this, although in that case this vote does not reflect a real preference for the list puller but simply a preference for the party. To summarize, we expect that:

1. Voters who do not have the option to cast a list vote are more likely to vote for the list puller than voters who do have the option to cast a list vote.

2. Voters who do not have the option to cast a list vote and voters who do have the option to cast a list vote will behave similarly when it comes to casting votes for other candidates.

Another major difference between the Belgian and Dutch system is the number of preference votes citizens can cast. We expect that the lower the candidates are placed on the list, the more they benefit if voters have the option to cast multiple preference votes. List pullers will hardly be affected by the number of preference votes that may be cast. In Belgium, many voters combine a vote for the list puller with a vote for another candidate, often a more local candidate (Wauters, Weekers, and Pilet 2004). If these voters were forced to vote for only one candidate, we expect that more voters would drop the vote for the other candidate(s). First of all, previous studies have shown that candidates at the top of the list receive more preference votes (Marcinkiewicz 2014; Miller and Krosnick 1998). Since this proven tendency to vote for higher placed candidates, we expect that it is more likely that voters drop the preference for a lower placed candidate when they are forced to vote for only one candidate. Second, a bandwagon effect could occur if citizens vote for the candidate who is expected to receive more votes. This could be because they think that their vote gets wasted if they vote for a lower placed candidate, who is not likely to get enough votes in order to get elected. When voters get the opportunity to cast more preference votes, this will benefit lower ranked candidates, because voters who normally only vote for the list puller might decide to also cast a vote for another candidate who they wish to give additional support. The question is whether we should expect something different for the group of voters who also have the option the cast a list vote and the group of voters who do not have this option. Other than that, the option to cast a list vote will lead to a lower number of preference votes; we do not expect that the pattern will be different. These expectations can be summarized as:

1. Voters who can cast multiple preference votes are more likely to vote for lower positioned candidates than voters who can only cast a single preference vote.

2. Voters who can cast multiple preference votes and voters who can only cast a single preference vote will be equally likely to vote for higher positioned candidates.

Research design

We conducted an experiment simultaneously in Belgium (Flanders) and the Netherlands using a between-group design. The advantage of an experiment similarly in two countries is that if we find similar results, this strengthens our conclusions, whereas different trends show that the effect institutions exert on voting behaviour is influenced by the specific context. Since we test the effect of two institutional rules (stimuli), we need four groups in each country to which respondents are randomly assigned (see Table 1). Respondents in group 1 have the option to either cast a list vote or to cast a preference vote for one candidate. Group 2 reflects the Belgian system: respondents can cast a list vote or cast a preference vote for one or multiple candidates, as long as these candidates belong to the same party. Respondents in group 3 can vote according to Dutch rules: they are forced to cast a vote for a single candidate. Finally, respondents in group 4 do not have the option to cast a list vote, but can cast multiple preference votes.

Table 1. Experimental groups.

		Number of preference votes
		One	Multiple
Option to cast a list vote	Yes	Group 1	Group 2 (Belgian system)
	No	Group 3 (Dutch system)	Group 4

As part of the data collection, we created two almost identical surveys to conduct our experiment, one for Dutch and one for Belgian respondents. Both surveys were distributed by Survey Sampling International (SSI). This guaranteed that respondents in both countries received an identical survey with regard to layout and so on. SSI has its own panels from which respondents were drawn. These panels consist of people who registered themselves.³ We aimed to have 750 respondents for each country. Since we conduct an experiment, the prerequisite of having a representative sample is less important than in a normal survey, as long as respondents are randomly assigned to a treatment group. Nevertheless, in order to increase our external validity, we tried to aim for representative samples. Therefore, representative quotas on key socio-demographic characteristics (gender, age, region, and education) were used. These quotas were applied to each characteristic individually, and not combined. For this research especially, the quota on education level is important if we want to increase our external validity, as previous studies show that political interest and knowledge are important predictors of casting a preference votes (André, Wauters, and Pilet 2012). Although the relation is not one on one, education is strongly related to political knowledge (e.g. Rasmussen 2015) and therefore forms a good proxy for it.

The survey had four parts. In the first part, respondents were asked for which party they would vote if national elections were held that day. This question served as a filter to decide whether respondents would participate in the experiment or not: only respondents who said that they would vote (for a party) were included. We also excluded respondent for the smallest parties (see online Appendix A), because it would have been too difficult to design a candidate list for these parties with enough candidates that could be recognized by the voters. Eventually, we had 785 respondents in the Netherlands and 788 in Belgium who passed the filter questions and participated (see online Appendix E). Second, respondents received an explanation of the electoral rules under which they could vote, based on the group to which they were randomly assigned⁴ (see online Appendix B). After respondents had read these instructions, they preceded to the third part of the survey: a ballot paper. The ballot paper they received depended on the group they were assigned to and their party choice (see online Appendix C for examples). Finally, the fourth section contained some more general questions, such as a question on how sympathetic respondents find the list puller of their party.

Each ballot paper included 20 candidates of the chosen party and, for the relevant experimental groups, also allowed voters to cast a list vote. We designed candidate lists with real politicians, instead of fictional candidates.⁵ Although fictional candidates have the advantage that the researcher exerts more control over the experiment, it would nevertheless be problematic for the design of this study. Fictional candidates are not known by respondents and therefore are more likely to drive respondents in the direction of casting a list vote or a vote for the first candidate, since they are unlikely to have preferences for candidates they do not know. Hence, we opted for real candidates, to make sure that, next to the party leader, each ballot paper has a number of candidates that are also known by respondents. Thus, we also put ministers and members of parliament on the lists. Of course, there may still be respondents left who only know the party leader, but this is in line with reality.

To increase comparability not only between Belgium and the Netherlands, but also between political parties within each country, we first designed a general concept list that could be applied to each party using criteria that were held constant across political parties and the two countries. Therefore, differences between political parties with regard to the electoral strength of the ballot list would not be too large. For example, we made sure that for all parties, the sixth place was occupied by an incumbent parliamentarian. To design this concept list, we used standardized mixture of a few characteristics to guide us: incumbency, gender, ethnicity, and regionalism. Not only are these characteristics taken into account by real-life selectors (Put and Maddens 2013; Gallagher 1988), they are also predictors for the success of a candidate in terms of preference votes (e.g. Maddens et al. 2007; Van Holsteyn and Andeweg 2012). Subsequently, using this general concept list, we created candidate lists for each party with actual politicians from that party. We designed two concept lists: one for the larger (government) parties and another for the smaller parties. This made it possible to put more familiar candidates on the list of the larger parties. If we had made a draft list which should apply to all parties, we would have been restricted in, for example, the number of members of parliament we could include on the list since some smaller parties only have a few members of parliament.

The list we created is different from what Belgian voters are used to, because respondents normally vote in a district where each party presents a district-specific list. However, we kept the lists the same so all Belgian respondents, irrespective of their district, got the same ballot with candidates from all districts. This was done to guarantee comparability of the experiments since we wanted to avoid that the presented lists in the Netherlands were “stronger” than in Belgium. The draft lists and identification of parties as large/small are given in online Appendix D.

The way we set up the experiment has a potential problem. We gave voters only limited information on how their votes influence the (hypothetical) seat distribution, while we know from the literature on strategic voting that this could influence the decision voters take (e.g. Blais et al. 2012). The choice, to give respondents no detailed information on what the rules meant and what their (potential) purpose is, was deliberate. We did not want to prime voters in a certain direction and thereby artificially increase the use of specific options. After all, in a real ballot box, voters also only receive a ballot paper. Consequently, it is possible that voters turned to the knowledge of their own electoral system. Dutch voters and Belgian voters thus could have something different in mind while participating in our experiment. This is especially true for the introduction of the list-vote in the Netherlands, since we only told participants that with a list-vote they “support the candidate list as a whole”. This complicates the results of our experiment. However, we do not think it is too problematic. In our analysis, the main focus will be on the results of both countries individually. In that part of the analysis, this is no problem, because we can assume that since voters are used to the same rules in real life, there is no difference between the experimental groups in their starting point. With regard to the list vote for Dutch voters, we cannot be sure what voters have in mind. In the conclusion, we reflect on this aspect of our experiment design. This could, however, be problematic when we compare the results between Belgium and the Netherlands. Therefore, we should be careful when we say something about differences between results in both countries.

In order to test our expectations, we run two logistic regressions for each country with respectively voting for the list puller and voting for other candidates as dependent variables. Our independent variables are the two treatments: the option to cast a list vote and the option to cast multiple preference votes. Moreover, we control for a number of factors that the literature has identified as key predictors for preference voting: political interest, education and being a party member (André, Wauters, and Pilet 2012). We also expect that the type of (preference) vote may be a result of the evaluation of the list puller. We therefore control for this evaluation, by asking voters on a 10-point scale the sympathy for the party they voted for, and the sympathy for its leader. We include two dummy variables: one for lower list puller evaluations than party evaluations and another for higher list puller evaluations than party evaluations. Thus, the reference category here is voters who evaluated the party and its list puller at the same level.

Results

Respondents were randomly assigned to one of the experimental groups (see online Appendix E for the distribution and party choice of respondents per group). Before we start comparing between our experimental groups, we should note that when we look at the control groups of respondents who keep the same system as they are used to, these respondents vote very much in line with the actual electoral results. In group 2 (list vote or multiple preference vote), which represents the Belgian electoral system, we notice that 35% of the Belgian respondents cast a list vote (see Figure 1). This is more or less consistent with actual election results which normally fluctuate around 40%. In group 3 (single preference vote), 33% of the Dutch respondents cast a vote for another candidate than the list puller (see Figure 2); this is also more or less in line with the real outcome, although slightly higher than in real elections. This could be a result of the way we designed the candidate lists. Familiar candidates tend to be overrepresented on our candidate lists and even more concentrated at the top of the list. Thus, it is not surprising that more voters cast a vote for another candidate than we observe in real-life elections.

Figure 1. Types of votes cast.

Figure 2. Types of preference votes cast.

The first thing we notice when comparing between groups is that voters respond to new rules. In the Netherlands, approximately 20% of the respondents in groups 1 (list vote or single preference vote) and 2 (list vote or multiple preference votes) – respondents who in real life do not have this option – casted a list vote. The same is true for the option to cast multiple preference votes. In the Netherlands, 15% of the Dutch respondents casted a preference vote for multiple candidates. Thus, a substantial part of Dutch voters made use of the new options, suggesting that the configuration of electoral rules matters.

However, when we look at how many voters used these new rules, we have to conclude that Dutch voters made less use of the option to cast a list vote and also were more likely to stick to a single preference vote compared to Belgian voters, where approximately 40% of the respondents casted a list vote or used the option to cast multiple preference votes. Additionally, when we look at how many preference votes were cast by the respondents in groups 2 (list vote or multiple preference votes) and 4 (multiple preference votes), we find that in Belgium, voters who did cast a preference vote in one of these two groups, on average, voted for 2.7⁶ candidates. In the Netherlands, this was 1.3. In the discussion section, we reflect on the differences between Belgium and the Netherlands.

Having shown that electoral rules influence voting behaviour, we can now turn to our expectations. We expect that the option to cast a list vote affects votes for the list puller. When we look at the model in Table 2, we indeed see a significant negative relationship between the option to cast a list vote and voting for the list puller. In both Belgium and the Netherlands, the list puller received fewer votes if respondents could cast a list vote, compared to the situation in which respondents could not. The option to cast a list vote indeed has a negative effect on voting for the list puller.

Table 2. Voting for the list puller.

	Belgium		The Netherlands
	Odds ratio	95% CI	Odds ratio	95% CI
List vote	0.306	0.218–0.430	0.630*	0.434–0.914
Multiple pref. votes	1.443*	1.034–2.014	1.122	0.773–1.629
Political interest^a
Somewhat interested	1.276	0.876–1.860	0.955	0.600–1.519
Highly interested	1.361	0.802–2.308	1.809	0.871–3.760
Party member	1.332	0.810–2.190	0.792	0.386–1.624
Evaluation difference^b
List puller < party	0.193**	0.121–0.308	0.358**	0.225–0.569
List puller > party	2.204**	1.505–3.229	1.542	0.980–2.428
Education^c
Middle	1.008	0.660–1.541	0.746	0.460–1.211
High	0.798	0.506–1.258	0.322**	0.198–0.525
(constant)	1.606		3.386**
N	713		530
χ^2	154.001	(DF = 9. sig. < .01)	64.175	(DF = 9. sig. < .01)
Nagelkerke R^2	0.259		0.154

*p < .05, **p < .01.

^aReference category: not interested.

^bReference category: evaluation list puller = evaluation party.

^cReference category: low.

We further expect that the option to cast a list vote has no influence on voting for other candidates. If correct, we would expect a non-significant odds ratio for the list vote. However, in Table 3, we see that respondents who have the option to cast a list vote are less likely to vote for other candidates in both Belgium and the Netherlands. The odds ratios for list vote in this model are actually very close to the list vote odds ratios in the model for list pullers. In other words, the option to cast a list vote equally affects votes for the list puller as well as other candidates. These results refute the idea that votes for other candidates are more sophisticated and more “preferential” in nature than votes for the list puller in situations where a list vote is not possible. In our experimental study, we see that the list puller and other candidates are equally affected by the introduction (or abolition) of a list vote.

Table 3. Voting for other candidates.

	Belgium		The Netherlands
	Odds ratio	95% CI	Odds ratio	95% CI
List vote	0.325**	0.232–0.455	0.582**	0.401–0.845
Multiple pref. votes	4.001**	2.852–5.614	1.400	0.962–2.037
Political interest^a
Somewhat interested	1.436	0.981–2.103	1.355	0.846–2.169
Highly interested	1.659	0.978–2.814	1.035	0.497–2.158
Party member	1.397	0.857–2.277	1.399	0.691–2.833
Evaluation difference^b
List puller < party	2.009**	1.312–3.076	2.434**	1.542–3.843
List puller > party	0.659*	0.448–0.968	0.745	0.473–1.174
Education^c
Middle	0.958	0.624–1.470	1.266	0.782–2.050
High	1.089	0.695–1.707	1.932**	1.191–3.132
(constant)	0.462**		0.307**
N	713		530
χ^2	143.220	(DF = 9. sig. < .01)	42.648	(DF = 9. sig. < .01)
Nagelkerke R^2	0.244		0.106

*p < .05, **p < .01.

^aReference category: not interested.

^bReference category: evaluation list puller = evaluation party.

^cReference category: low.

The second set of expectations state that the option to cast multiple preference votes will mostly benefit candidates lower on the list and have no effect on voting for list pullers. For Belgium, in the model for voting for other candidates (Table 3), the odds ratio is highly significant. Respondents who have the option to cast multiple preference votes are four times more likely to (also) vote for another candidate than voters who do not have that option. However, opposed to what we expected, the option to have multiple preference votes also positively affects voting for the list puller (see Table 2). However, the effect for the list puller is much smaller than the effect for other candidates, so, in general, our expectation that lower placed candidates benefit more from multiple preference votes is supported. In the Netherlands, given the lower use of the option to cast multiple preference votes, it is more difficult to grasp the effect of this option. Based on the results of the experiment, we should conclude that the option to cast multiple preference votes has no significant effect on voting for the list puller (Table 2). However, unlike Belgium, we do not find a significant effect on voting for other candidates in the Dutch case. This difference between Belgium and the Netherlands remains robust when we combine the cases in one model with an interaction term between the Netherlands and multiple list votes (online Appendix F). Thus, the findings with regard to the effect of multiple preference votes are mixed. In the next section, we will discuss the nature of these differences.

Discussion

This paper set out with the aim of studying the extent to which electoral rules influence electoral behaviour with regard to preference voting. In line with a recent trend, we conducted an experiment in order to answer this question. We focused both on the act of casting a preference vote and on the distributions of these votes.

The main finding of our experiment is that electoral rules are important and strongly shape voting behaviour. Belgian voters were already used to the options of a list vote and multiple preference votes. Consequently, in the experimental groups where these options were available, they also used them. For Dutch voters, these options were new. Nevertheless, about 20% of the Dutch respondents did make use of this option when it became available to them, and the same can be said for multiple preference votes. A second important finding of this study is that our results show that – in the first place for the Netherlands – the idea behind the distinction between a vote for the list puller as a party vote and a vote for another candidate as a preference vote may be too strict. We expected that in systems where a list vote is not possible, many voters cast a vote for the first candidate because this is the most “simple” option when one lacks information about the candidates. Thus, once an even simpler option is available in the form of a list vote, we expected voters to go for the list vote instead. For voters who cast a vote for another candidate, we expected no change as this is considered to be a more sophisticated type of voting behaviour (Marsh 1985), reflecting a “real” preference. Yet, this expectation did not hold. Both the list pullers and other candidates equally lose votes once the option to cast a list vote becomes available. Why voters who cast a more sophisticated type of vote in one situation switch to a “simpler” type of vote in another situation is puzzling and raises the question whether a “preference vote” for a list puller and a preference vote for another candidate are as different from each other as is often assumed, especially in the Netherlands. Finally, we found that the option to cast multiple preference votes mostly benefits lower ranked candidates, but this only holds in Belgium.

Although we argued that it is difficult to compare the results of the experiment in Belgium and the Netherlands, it is still interesting to look at the differences. It is not possible to draw strong conclusion from these differences, but it could help to gain insight in where there is room for further research. We find that our expectations that other candidates benefit from the option of multiple preference votes and candidates on top of the list do not, only receive support in Belgium. Additionally, while a substantial number of Dutch respondents use the option to cast a list vote or multiple preference votes, this percentage is by no means as high as the number of Belgian respondents using these options. We can only speculate about the nature of these differences. Perhaps a reason is that the Netherlands has on average stronger leaders, due to the single national district, giving less incentive to voters to cast a list vote or to vote for candidates with a lower ballot list position. Yet, this is unlikely as in the experiment we made the Belgian lists as strong as the Dutch lists by putting on party leaders. Also the fact that the number of list votes is still higher for voters of the N-VA, which has a strong leader in Bart De Wever, goes against this explanation. A more likely explanation is that differences can be accounted for by voting habits. Belgian respondents were already familiar with the options to cast a list vote or multiple preference votes. For Dutch voters, on the other hand, these rules were new, and consequently, respondents may have stuck to their usual behaviour. This could indicate that changes in the electoral system do not immediately result in changes of voting behaviour, but that voters need some time to “learn” using the new rules. This is in line with actual election results in Belgium. The option to cast multiple preference votes was introduced in 1995, but the option was hardy used in the elections of 1995 and 1999. In 2003 and 2007, however, the average number of votes per “preference ballot”⁷ increased (Wauters and Weekers 2008). In 2010, this number again increased although it dropped slightly in 2014. In general, an increasing use of multiple preference votes was visible. Further research, for example by studying first time voters, should shed more light on whether voting habits indeed can account for the differences. However, if this is the case, it means that the experimental method that is currently used to estimate effects of electoral rules may actually underestimate these effects.

Our findings have two important implications. First, our results suggest that preference votes might be less preferential (with regard to a specific candidate) than has often been assumed. For systems without list votes, it is often argued that whereas votes for list pullers might not really be preferential, votes for all other candidates are based on conscious decisions and therefore truly preferential. However, our findings cast doubt on this common assumption, since both voters who cast a vote for the list puller and voters who cast a vote for other candidates go for a list vote when presented with this possibility. Of course, we cannot fully exclude the possibility that, for some voters, a list vote still contains a preference for a candidate, as it could be that they prefer candidates at the top of the list and think that a list vote will help those candidates to get elected. However, considering the effect sizes we find, it is unlikely that all voters who casted a list vote instead of a preferential vote think so strategically. Many voters may not be aware that by casting a list vote, one technically supports the candidates at top positions. Especially in the Dutch case, where voters are not used to these new rules and where in the description we did not explain all the implications of these rules, many voters might not have been aware of the practicalities of the list vote. In this case, our findings of voters who switch their vote might be conservative, indicating that at least for a group of voters who switch their preference vote might be less preferential than has often been assumed. What this discussion above all shows is that further research is necessary to get a better understanding of the meaning of a preference vote and a better insight into the degree of preference in a so-called preference vote.

Second, while we argued that we should be reserved in comparing the results of the experiment in Belgium and the Netherlands, the fact remains that we found different results in both countries. We speculated that these differences might be the result of voting habits. If this is true, it means that we underestimate the effects. In real life, and at least over time, the results might even be stronger. This has also implications for other experiments. Even if an experiment is well designed, the extent of the effect might still be influenced by contextual factors outside the influence of the researcher. Therefore, we stress the importance of conducting experiments in different contexts to strengthen the robustness of the findings.

Acknowledgements

The authors thank Joop van Holsteyn, Rudy Andeweg, Peter van Aelst, Stefaan Walgraeve and Michiel Johnson as well as the three anonymous reviewers for their useful comments on earlier versions of the paper.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes on contributors

Marijn A. M. Nagtzaam is a PhD candidate at the Institute of Political Science of Leiden University. His PhD research is focused on ‘Second-Order Electoral personalization’. His study will look in more detail at intra-party preference voting, its causes and its consequences.

Patrick F. A. van Erkel is a PhD candidate at the Department of Political Science at the University of Antwerp and member of the research group M2P (media, movements and politics).

Additional information

Funding

This work was supported by Research Foundation Flanders: [Grant Number G026513N].

Notes

1 A distinction can also be made on the basis of the relative weight of preference votes. However, this is related to what happens with the votes and much less with the type of vote(s) voters can cast. Only in some urban regions does this have a small effect (Wauters, Verlet, and Ackaert 2012). Therefore, we will not consider this distinction.

2 In the context of our experiment, we consider all votes for individual candidates as preference votes, to distinguish them from list votes.

3 For more information, we refer to the SSI website https://www.surveysampling.com.

4 The time respondents spent on this page was registered. These data led to two conclusions: (1) voters who used the option to cast a list vote or to cast multiple preference votes on average spent more time on the page with the explanations of voting rules than voters who did not use any of these options (2) based on a comparison between Belgium and the Netherlands, there is no reason to expect that this influenced the experiment. At most, the estimation of the effects we report is conservative.

5 Candidate lists are available on request.

6 The average number of preference votes for the 2007 federal elections in Belgium was (in Flanders) 2.63 (Wauters and Weekers 2008, 64), 2.75 in 2010 and 2.55 in 2014 (own calculations), so comparable with the results of our experiment.

7 A term used to distinguish ballots with preference votes from ballots with a list vote.