PREDICTING PREFERENCE VOTE SHARES IN LOCAL ELECTIONS

Transcription

1 PREDICTING PREFERENCE VOTE SHARES IN LOCAL ELECTIONS Peter Thijssen & Kristof Jacobs Department of Political Sciences, University of Antwerp Prinsstraat 13, B-2000 Antwerpen, België Tel.:03/ , Fax: 03/ Paper to be presented in the workshop on Local participation in different contexts at the ECPR Joint Sessions of Workshops, April, 2005, University of Granada 1

2 1. Introduction Studies examining the determinants of the preference vote share of candidates often focus on one particular factor, in most cases incumbency (Cox and Katz, 1996; Ansolabehere and Gerber, 1997; Born, 2000) or to a lesser degree the ballot position of a candidate (Darcy and McAllister, 1990; Miller and Krosnick, 1998; Koppell and Steen, 2004). This focus seems legitimate since many studies find these two factors to be the most influential predictors of individual vote shares (Marsh, 1985; Desposato and Petrocik, 2003; Geys and Heyndels, 2003a). However, the importance of these factors may in a way be circumstantial. The existing work considers a wide variety of electoral contexts, but by far most of the studies focus on majoritarian electoral systems, more specifically single member plurality systems. Of the empirical studies cited above, this is the case for Ansolabehere and Gerber, 1997; Miller and Krosnick, 1998; Desposato and Petrocik, 2003; Born, 2000; Cox and Katz, 1996; Koppell and Steen, In this kind of electoral systems voters may tend to vote strategically for candidates they think have a fair chance of being elected. In this case a person prefers to vote for a so-so candidate with a good chance [of being elected], instead of voting for a great candidate with a poor chance (Popkin, 1991, 116). Therefore it is not inconceivable that in more proportional electoral systems other factors might be important. It is precisely in this area that we will make a contribution. 1 We will determine the most important factors predicting preference vote shares in proportional electoral systems. The Belgian electoral system provides such a case. It can be characterized as a proportional multiparty system with open lists. These lists can contain an ordinal 2

3 number of candidates up to the number of seats that are distributed. Voters may choose to cast a list vote or one or more preference vote(s). Especially the last characteristic is important. This multiple preference vote permits a voter also to choose candidates he or she prefers, not only those who have the best chance of being elected. In this context a voter will for instance not necessarily vote for the male candidate occupying the top position on the list, while he or she prefers a female candidate in a lower position. 2. The research case and the data More specifically this analysis is based on the Antwerp district council elections of 8 October We are concerned here with a special case, because it was the first time since the Antwerp municipal merger that elections were held at this level of local government (Thijssen and Van Assche, 2002). Although the district councils already existed since 1983, these councillors were appointed (not elected). Thus, strictly speaking there were no real incumbents. Given that incumbency is the single most important factor in predicting candidates vote share in city council elections (Krebs, 1998: 922), this research case meets a quasi experimental condition that is rarely satisfied in electoral research. If a factor isn t influential in this electoral context, chances are high it also will not have a significant impact in a more majoritarian context with incumbents. We focus specifically on the candidates of the six main political parties, as these submitted ballots for all nine districts. These six parties were AGALEV (green party), CVP (Christian democratic party), SP (social democratic party), VU-ID (Flemish 3

4 nationalist party), VLD (liberal party) and Vlaams Blok (extreme right-wing party). Each of the 1201 candidates to stand in the Antwerp district council election for one of these six parties was presented with a written questionnaire in the months of August and September Of these, 612 returned a completed questionnaire. Thus, the response rate was quite high (almost 51%) and, moreover, most subgroups appeared to be well represented. Nevertheless, the number of candidates of Vlaams Blok to return the questionnaire was notably smaller than in the case of the other parties (36.6% versus 55% for CVP, 43.5% for VLD, 62.7% for AGALEV, 55.5% for SP and 52.1% for VU-ID). Another relatively underrepresented group are the ethnic candidates. Only nine out of thirty ethnic candidates returned the questionnaire. A detailed analysis of the non-response can be found in appendix 1. As we intend to explain the electoral success of the individual candidates, it is selfevident that our dependent variable should be based on the number of preference votes. Since the nine districts vary in size, we divided this number by the total number of preference votes obtained by the six main parties in the district where the candidate in question stood. 4

5 Therefore the dependent variable can be written as: t jk x ijk 6 i= 1 j= 1 x ij where x ijk is the number of votes for candidate x ranked number i on the ballot of party j (AGALEV, SP, CVP, VU-ID, VLD, Vlaams Blok) in district k (Antwerpen, Berchem, Bezali, Borgerhout, Deurne, Ekeren, Hoboken, Merksem, Wilrijk). where t jk is the number of candidates on the ballot of party j in district k. 3. Multi-stage explanatory model and hypotheses As the factors that are central to the various theories are not always close to the dependent variable, both content-wise and temporally, it would be misleading to incorporate them simultaneously into the explanatory model. By analogy with the funnel of causality proposed by Campbell et al. (1966), we therefore start with independent variables that, in terms of content, are further removed from the dependent variable (cf. politically exogenous factors, such as demographic and sociocultural factors) and we gradually add variables that exhibit greater affinity with the political outcome to be explained (cf. politically endogenous factors) (figure 1). 2 [Figure 1 about here] One can argue that the order of step two and three can be switched. Indeed, social insertion and insertion in other public spheres stand both as close (or far) to the 5

6 electoral process. However, since we want to gain a maximum insight in the strength of the different factors, we decided to add the most powerful of the two (insertion in other public spheres) in the third step Demographic group membership Calhoun (1994) asserts that there is an increasing tendency among individuals who belong to minority groups to borrow a significant part of their identity from that group. Therefore, to the extent that certain socio-demographic groups are underrepresented in the political structures and/or on the ballots, it is not unthinkable that this will influence the voting behaviour of individual members of that group (Herring and Forbes, 1994). All the more so as they are encouraged in this sense by organised interest groups and, in some cases, even the government. Quite telling in this respect are the vote for a woman campaigns in Belgium (Carton, 1998), the American Women s Political Action Committees (Dolan, 2001) and the Latinooriented political campaigns in American presidential elections (Connaughton, S. and Jarvis, S., 2004). Members of underrepresented groups, such as women, ethnic minorities, youngsters, the elderly and the low-skilled, will according to our explanatory model give preference to members of their own group, believing that they will better protect their interests (Cutler, 2002). Moreover, it is not unthinkable that some members of politically overrepresented groups who take offence to the imbalanced representation will also show a tendency to vote for representatives of these underrepresented groups. Hypothesis 1: Candidates belonging to a socio-demographic group that is underrepresented in the political structures and/or on the ballots, such as women, 6

7 ethnic-minority candidates and, to a lesser extent, youngsters, elderly persons and the low-skilled, will receive significantly more preference votes. 3.2 Visibility through social activities (Social insertion) In Belgium many more preference votes tend to be cast in local elections than in other ballots. This finding seems to suggest that the social distance between voter and candidate is much smaller at the local level. The likelihood that the voters get to know certain candidates through interpersonal contact is, after all, much greater at local elections (Wauters, 2000: 15). This election thus offers us an opportunity to examine the impact of the social capital of a candidate on his or her vote share. It is, after all, not unthinkable that candidates with an extensive social network score better, perhaps even irrespective of which political party they represent. It is, however, almost impossible to gain direct insight into the social networks of the candidates. Therefore, we shall rely on an indirect measurement whereby we aim at the channels through which a candidate may construct a social network. A first measure that springs to mind is the intensity of membership of associations 3. To the extent that these associations are local 4, they constitute an ideal vehicle to construct a social network. A second channel that may contribute is the holding of a local profession 5, such as baker, teacher at the local school, or clerk with the local bank. The intensity of these customer relations is probably not as great as in the case of contacts within a local association. Nevertheless, these contacts do guarantee that a candidate is recognizable. 7

8 Hypothesis 2: Candidates who participate actively in local associations and/or who perform a profession through which they come into contact with local residents will receive significantly more preference votes Visibility through other activities In times when the social fabric of society appears to be eroding, media fame would seem to be an increasingly important substitute for interpersonal contacts (Stein and Fleischmann, 1987). With the arrival of regional television, local politicians now also have at their disposal an interesting additional medium (Lieske, 1989; Krebs, 1998). However, it should be pointed out that, again, the candidates for the district council elections received rather little attention. Nevertheless there are plenty of other opportunities to gain a high rate of visibility. This is for instance the case with wellknown entrepreneurs or people who are in showbizz. Given the fact that visibility is somewhat subjective persons for instance who are well-known by the lower educated, are not necessarily well-known by the higher educated, and vice versa-, external observers have a hard time determining whether a candidate is a VIP or not. This is why we will appeal to the leaders of the slating committees of each party in the nine districts to tell whether a candidate a VIP or not. 6 After all we can assume that the persons who recruited the candidates have a maximum insight in the candidates advantages. In comparison to the municipal elections, the proportion of VIP s standing in the Antwerp district council elections was rather small. Research by Fiske (1989) shows that, in such a situation, the well-known candidates have a large comparative 8

9 advantage over their not-so-well-known counterparts. In the land of the blind, the one-eyed man is king. Hypothesis 3: Candidates who acquired fame in non-political public spheres (VIP s), will receive significantly more preference votes. 3.4 Visibility through political activities (Political insertion) While we have thus far based ourselves mainly on politically exogenous factors, the following explanatory frames are founded firmly on politically endogenous factors. It is self-evident, for example, that a candidate may win votes by conducting an active campaign or thanks to activities which he/she undertook as part of a previous political mandate (Krebs, 1998). Although the analysis of the sample of election propaganda shows that the higher ranked candidates appear a lot more commonly in electoral propaganda, we were unable to assign individual scores for the intensity of campaigning. Therefore, we rely on subjective evaluations by the candidates, who assigned a score to themselves for the time invested in their campaigns. 7 Candidates who stand for a mandate which they have previously held generally score better than new candidates. However, as we pointed out in the introduction, this was, strictly speaking, an election without incumbents. After all, before 2000, the district councils consisted of individuals appointed by the parties and it possessed no autonomous competence. In this sense, it comes as no surprise that relatively little was said in the media about the activities of these former district councilors. Nevertheless, it is plausible that these councilors were slightly better known by the district residents thanks to political services rendered (Ackaert, 1994). The same 9

10 holds for municipal and provincial councilors, all of whom receive little media attention and were therefore generally not considered to be VIP s. But also candidates who by their name are clearly identifiable as relatives of a wellknown politician may profit from his or her fame (Mueller, 1970). Given the fact that the leaders of the slating committees are best placed to judge this kind of advantages, we will rely on their judgements to identify these relatives. Hypothesis 4: Candidates who campaigned intensively and/or previously held a political mandate, and are therefore possibly better known among district residents, and/or are clearly identifiable as relatives of a well-known politician, will receive significantly more preference votes. 3.5 Ballot position To the extent that voters are aware of the fact that the existing electoral system favours the highest-ranked candidates, they are perhaps inclined to vote for higher ranked candidates. After all, some voters may prefer to vote for a candidate who has a realistic chance of being elected. Research (Ackaert, 1994; Geys and Heyndels, 2003b) has shown that the rank of a candidate on the ballot has a significant impact on the number of preference votes obtained. Hypothesis 5: Candidates who occupy a higher position on the ballot obtain significantly more preference votes. 10

11 For this fifth hypothesis, the position on the ballot was considered to be a continuous variable. Perhaps this approach is not ideal, as it is reasonable to assume that the difference between positions 1 and 6 is more significant than that between positions 11 and 16. Moreover, the reason for voting for a higher-ranked candidate may also be that the voter wishes to spend as little time as possible in the voting booth. This may be the case, for example, if the voter knows none of the candidates. In such situations, the voter might not want to make the effort to go through the entire list of candidates. On this basis, we assume that the ranking effect manifests itself primarily in the case of the candidate heading the ballot. Hypothesis 5a: The candidate heading the ballot will receive significantly more votes. In accordance with that logic, another ballot position that may entail a higher vote share is the bottom position. This ballot position is also easy to find on a ballot list. Moreover, in Belgium it is a common practice to place formerly prominent politicians on these positions to support the ballot list. It is possible that this habit entailed an autonomous effect, so that even candidates who were never prominent would benefit from holding this ballot position. Hypothesis 5b: The candidate on the bottom of the ballot list will receive significantly more preference votes. As the Antwerp district council elections involved voting by computer and most ballots could not be represented in a single column on the computer screen, we must 11

12 also take account of the phenomenon of pseudo leading candidates, i.e. candidates who head the second or third columns. This may also give rise to a primacy effect. Hypothesis 5c: Candidates heading the second or third column of the ballot, will receive significantly more preference votes. It is also possible that some voters will simultaneously take into account the ballot position and one or more of the previously mentioned factors. As this gives rise to a myriad of combinations, we shall restrict our analysis to one of the most relevant combinations, i.e. the highest-ranked woman. After all, a voter who would like to see a representative of a political minority group elected can best vote for the highest ranked representative of that group. 8 Hypothesis 5d: The highest-ranked woman on a ballot will obtain significantly more preference votes. 3.6 Party affiliation The five previous explanatory logics took no account of the specificity of the party affiliation of the candidate. Still, it speaks for itself that many voters, including in local elections, first and foremost vote for a party, usually for ideological and/or strategic reasons. In such cases, only the candidates appearing on the list submitted by the favoured party are eligible for a preference vote. This means that candidates on the list of a popular party will achieve the best scores. 12

13 Hypothesis 6: Candidates appearing on the ballot of a popular political party obtain significantly more preference votes. 4. The results Table 1 provides an overview of the results of the six nested regression analyses. In our discussion of the effects of these various factors, we shall consider table 1 both horizontally and vertically. [Table 1. about here] 4.1. Demographic group membership If we consider the results of the first explanatory model in Table 1, it is noticeable that the demographic variables as such have hardly any impact. None of the five variables concerned has a statistically significant effect and, combined, they explain only 0,5% of the variance of the preference vote proportions per district. 9 Our first conclusion must therefore be that demographic factors have hardly any impact. Hypothesis 1 may therefore be rejected. Still, on closer inspection it appears that two demographic variables, namely ethnic minority background and gender, clearly gain in explanatory significance if other factors are added to the model. After all, starting from model 4, female candidates receive significantly more (α= 0,05) preference votes than their male counterparts (β= 0,08). The explanation for this suppressor effect is that female candidates, more so 13

14 than their male colleagues, are new to local politics and generally have no previous experience as a non-elected district council member. 10 This would appear to indicate that controlling for district council experience should enhance the explanatory significance of the variable female. After addition of the ballot position-related factor in model 5, this explanatory power would appear to wane somewhat (β= 0,07). This is, however, not really the case, as part of the effect is absorbed by the variable first woman. 11 After all, this variable pre-eminently incorporates the effect of the gender factor. The first woman is the leading candidate among the women on the ballot, as it were. It is therefore very interesting that the effect of the variable gender in the sixth and final model is significant for the first time at α= 0,01. Further, it also appears that ethnic-minority candidates obtain significantly more preference votes after the factors regarding ballot position are added to the explanatory model in model five. How can we explain this? Save for a few exceptions, ethnic minority candidates were assigned relatively low positions on the ballot. As a higher position usually yields a greater number of preference votes (ex post), this is clearly an electoral handicap. If we take into account the factors related to the ballot position of a candidate, we notice that ethnic-minority candidates receive significantly more votes (β= 0,10). 12 In sum, we may assert that certain salient characteristics of the candidates (Cutler, 2002: 478), i.e. gender and ethnic background, have a significant influence on the preference vote share obtained. 14

15 4.2. Social insertion A second explanatory logic encompasses the social insertion of a candidate, measured in terms of whether or not he or she has a locally representative profession and participates in associations. As with the demographic factors, the explanatory power of these aspects is limited: the variables from the second regression model combined explain just 1,6% of the variance. Whether or not one holds a locally representative profession actually never exhibits a statistically significant effect. Nevertheless, it pays electorally to be active in various kinds of associations. The standardized effect of this variable is statistically significant in the second and third models for α= 0,01 and amounts to respectively 0,11 and 0,12. However, the effect parameter of this variable declines rather rapidly as the politically endogenous variables in models four and five are added to the explanatory model. More specifically, we find that those who are well inserted in local associations are not only much more likely to have held a seat on the former, not directly elected, district councils, but that they are usually also offered a better position on the ballot for the new district council. 13 These factors in particular result in a substantial decline in the effect of participation in associations in models four (β= 0,08) and five (β= 0,06) respectively. However, despite their favourable positions on the ballot, these intensive participants perform only marginally better than candidates in comparable positions who participate very little if at all in local associations. In the analyses represented in Table 1, we only took account of the aggregated indicator in relation to the level of social insertion of the candidates. However, we also made an analysis on the basis of the nine different types of clubs and associations 15

16 that were considered. 14 This analysis reveals that membership of socio-cultural societies 15 in particular has a significance impact on the number of preference votes obtained by the candidate. This is maybe due to fact that these associations usually have a higher amount of members and/or are more active and hence visible. Yet, again, the effect of this variable is only statistically significant in the second and third regression models Insertion in other public spheres One of the most determining factors appears to be being a VIP. We notice this clearly from, among other things, the evolution of the general explanatory power of the models. The R² increases from 0,016 in model two (without VIP ) to 0,037 in model three (with VIP ). The beta coefficient of the variable remains relatively high after addition of the politically endogenous factors. In the sixth and final explanatory model, the standardized effect parameter is still 0,16. The variable even appears to have the greatest explanatory power after leading candidate, ballot position and being a candidate for the Vlaams Blok. From the introduction in the third model, the effect of this variable is significant at the 0,001 level. This remains the case until the final step. 16 As there tend to be a less VIP s in local elections than in federal elections, this finding does not really come as a great surprise. The hypothesis that candidates who have acquired fame in other public spheres receive more preference votes can therefore be retained. 16

17 4.4. Political insertion The fact that being known from other political activities plays an important role is very apparent from the evolution of the R². It rises in the transition from the third to the fourth regression model by 18,1%. Three variables out of four that are added further, even are initially statistically significant at α= 0,001. Only being a relative of a well-known politician is but significant at α=0,05. The fact that many politicians believe that campaigning will win them votes is quite apparent from the answers to the question regarding the intensity of their campaign: 38,4% of the candidates say they invested much to very much time in campaigning. Moreover, we learn from the fourth regression model in Table 1 that candidates who say they have spent very much time campaigning also obtained significantly more preference votes (α= 0,001). It should be noted, however, that the standardized regression coefficient for this variable halves from 0,18 to 0,09 after addition of the factors regarding ballot position in the fifth regression model. It seems probable that a degree of social desirability comes into play here. Candidates who are higher up the ballot will therefore report to devote more energy to their campaign because they are expected to. 17 On the other hand, they also receive more support from their parties. Be that as it may, the effect of this variable retains its statistical significance at α= 0,001 up to the final regression model. Likewise, candidates who were previously seated on the district council initially (fourth model) appear to obtain more preference votes (β= 0,21; α=0,001). However, here we see that the effect of the variable is no longer statistically significant from the fifth nested model (β= 0,04). This is primarily due to the fact that 15 of the 28 17

18 individuals in the sample who headed a ballot had previously held a seat on the district council, so that the latter variable incorporated part of the leading candidate effect. This also indicates that the parties tried to gain an advantage through their district councillors. 18 Whether or not the candidate had previously been elected to a position on the municipal council, the provincial council or parliament appears to have a significant positive influence on the number of preference votes obtained by that candidate. We may assume that these people can rely on a faithful following of voters who also voted for them on previous occasions. Another possible explanation is that these candidates had been in a position where they could provide services to the public, and thus acquired some fame and/or support. We should also add, however, that the beta value for this variable takes a plunge after the addition of factors regarding ballot position (from 0,16 to 0,11). Visibility through political activities even reflects on their relatives and yields an electoral bonus. Although this effect initially seems modest (β=0,10), it is statistically significant at α=0,05. When we control for ballot position effects the regression coefficient even becomes significant at α=0,001; although the beta-value of the effect drops slightly to 0,09. Even after the inclusion of party affiliation, the effect remains significant at 0,01. We may conclude that visibility through previous political activities influences the voter: candidates who campaign intensively and/or who previously held a(n elected) 18

19 mandate or are a relative of a politician receive more preference votes, irrespective of such factors as their position on the ballot and the party for which they stand Ballot position As was to be expected, ballot position-related factors, all of which concern the position of the candidate on the ballot, have an important impact on a candidate s vote share. The jump in R² is quite remarkable: the addition of ballot position-related factors in the fifth model results in a 45 percentage point rise in the explanatory power. A first variable that lies at the origin of this jump is ballot position. This is significant both in the fifth and in the sixth model at α= 0,001. The standardized regression coefficient of this variable is rather large. It amounts to 0,14 in model five and to no less than 0,17 in model six. However, as we have previously mentioned, the position on the ballot is considered here to be a continuous variable. It is quite possible that certain positions on the ballot have a specific attractiveness. One position that immediately springs to mind is that of the leading candidate, i.e. the candidate at the top of the ballot. The standardized regression coefficient of this variable (β= 0,63 in the sixth model) is many times greater than that of the second most important factor, namely ballot position (β= 0,17). The question is: how should we interpret this? It is certainly simpler to determine what the effect does not entail: leading candidates score significantly better than other candidates on a ballot, irrespective of, for example, their fame, previous political experience or gender. In concrete terms, it would appear that a candidate automatically obtains significantly 19

20 more votes if he/she occupies the top spot on a ballot, than in any other position. There are different possible explanations for this. First and foremost, it is quite conceivable that a voter will want to boost the position of his/her party s leading candidate vis-à-vis the leading candidates of other parties. A second possible explanation is that this is a way for the voter to express agreement with the order in which the candidates appear on the ballot. In this sense, a vote for a leading candidate may be typified rather paradoxically as a personalized list vote. Another hypothesis is that candidates who appear on the screen of the voting computer at the top of a column while they are not a party s leading candidate automatically obtain more votes. Yet the effect parameter of this factor never becomes statistically significant. However, it is not inconceivable that specific party electorates do take this ballot position into account. Some slating committee leaders of SP or the Vlaams Blok explicitly indicated that they considered this ballot position to be an important one. On a SP-ballot list in Wilrijk for instance a well-known former member of the non-elected district council, held the position of pseudo leading candidate and obtained quite a large vote share 19. The question remains nevertheless whether this had to do with the ballot position of this candidate or rather with his personal traits. For the time being, there is therefore insufficient evidence to conclude that these pseudo leading candidates will in any case obtain more votes. A third hypothesis in connection with ballot position-related factors which we tested is the combination of gender and position on the ballot. This variable seems to be very important. The combination of gender and position appears to have a greater effect than the variable gender as such. Many voters who want to vote specifically for 20

21 a woman apparently vote for the highest-ranked woman. This may happen for strategic reasons, or simply because it is convenient : the first woman on the ballot is after all also the most visible female candidate. Finally, the position at the bottom of the ballot is also influential on the number of preference votes obtained. The bottom position yields more preference votes, irrespective of the other variables that are introduced into the explanatory models. The effect is however much less powerful than that of, for example, the position at the top of the ballot. The beta coefficient for the variable bottom position on the ballot is 0,077 in the sixth and final model and is statistically significant at α= 0,01. The hypothesis that candidates who are given a higher position on the ballot, who are the highest-placed woman or the leading candidate will obtain more preference votes is thus confirmed Party affiliation It is possible that candidates obtain more/fewer preference votes simply because they appear on a more/less successful ballot. In order to test this hypothesis, we included five dummy variables relating to figuring on the ballot of either SP, CVP, VU-ID, VLD or Vlaams Blok. The fact that this factor plays (to a certain extent) is certainly apparent from the changes in the R² value. In comparison with model five, the sixth model offers an additional explanatory power of 4 percentage points. In practice, just two party ballots yield statistically significantly more (Vlaams Blok) or fewer (VU- ID) preference votes. 21

22 The fact that Vlaams Blok made substantial electoral gains for the fourth consecutive time in Antwerp also finds expression in that standing for Vlaams Blok as such guarantees a substantial number of preference votes. This variable appears to be almost equally as important as the candidate s position on the ballot (β= 0,17). The opposite was true for candidates of VU-ID. Individuals standing for this traditionally small party systematically scored worse (β= - 0,10). Candidates on ballots submitted by AGALEV 20, CVP, SP and VLD did not obtain significantly more/less preference votes. 5. Conclusion In this contribution, we have explored why certain candidates achieve significantly better results in terms of preference votes than others. We have focused on local elections under a proportional representation system, whereby the ballots contain as many candidates as there are positions for election and a voter may cast multiple votes. As a specific research case, we opted for the Antwerp district council elections of 8 October As we did not wish to restrict ourselves to variables that can be derived from available administrative data, we used survey data from a written questionnaire submitted to 1201 candidates. Consequently, we must settle for only a part of the population. However, given the response rate of almost 51% and a relatively balanced representation of the relevant subgroups, we believe that we can still paint quite a reliable picture. 22

23 Another interesting aspect of this research case is that, to an extent, we are concerned here with new elections. Since the Antwerp merger of 1983, elections had not been organized at this (sub-)local level. Consequently, there were strictly speaking no incumbents. Other research has after all shown that incumbents tend to dominate (local) elections (Lieske, 1989; Krebs, 1998). In this case, then, voters need to rely on a different heuristics when casting one or more preference votes. Thus one could assume in this situation that certain politically exogenous factors, such as socio-demographic group membership, intensity of social insertion and being a VIP would play a significant role. Let us summarize the most important findings: - Generally speaking, candidates from socio-demographic minority groups do not receive a larger preference vote share. This changes from the moment that we control for the (less favourable) positions which they occupy on the ballot. Unlike less visible socio-demographic minority groups (youngsters, the elderly, the lower skilled), women and ethnic-background citizens then obtain significantly more preference votes. It is striking that, additionally, we also found evidence that this was pre-eminently true of the highest-placed woman on the ballot. - Interestingly, we observe precisely the opposite elaboration pattern in the case of variables that gauge the candidates degree of social insertion. Generally speaking, candidates who participate intensively in local associations receive more preference votes than those who do not. However, this effect disappears almost completely if we control for the position they occupy on the ballot. The 23

24 political parties apparently anticipate this positive impact of active membership of associations by assigning these candidates to high positions on the ballot. - Further, it is undeniably so that being well-known in non-political or associational spheres is a much more important attribute for a candidate in local elections than interpersonal acquaintance resulting from participation in associations. While the standardized regression coefficient for membership of associations is 0,52 in the sixth and final step of our model, the corresponding value for being a VIP is 0,16 (α=0,001). It is also striking that this effect parameter is hardly affected by politically endogenous factors such as position on the ballot. - As might have been foreseen, the explanatory power of the regression models increases significantly if we add the politically endogenous variables to the model. Visibility through previous political activities influences the voter: candidates who campaign intensively and/or who previously held a(n elected) mandate and/or are a relative of a politician receive more preference votes. - By far the most important group of variables in terms of explanatory power were the factors relating to the position that the candidate occupied on the ballot. If we operationalize this characteristic in a continuous manner, the variable indeed appears to have a very strong impact. The higher an individual is placed on the ballot, the more votes he/she obtains. It is significant that this effect continues to play if we control for all other variables in our explanatory models. However, not all positions are equally relevant in this respect. Some positions apparently have a unique attractive power on top of the effect of the continuous rank. More in particular, this is the case for the first woman (β= 24

25 0,10) and the bottom position (β= 0,08). But by far the most important position in terms of electoral attractiveness is that at the top of the ballot (β= 0,62). Our results seem to indicate that even if the leading candidate were a fictitious figure, he or she would still obtain significantly more votes than the other candidates on the ballot. In other words, the indications are that a vote for the leading candidate on a particular ballot often fulfils the purpose of a personalized vote for the party (a personalized list vote ). - Finally, it appears that candidates standing for Vlaams Blok (β= 0,17) and VU-ID (β= -0,10) receive respectively significantly more and significantly fewer preference votes than candidates on the ballots of other parties. The fact that these are respectively the most and the least popular parties examined is an indication that there is a certain connection between list votes and preference votes. The latter finding, for that matter, brings us to a new research question. It is, after all, conceivable that the impact of the factors examined is connected with the political party under consideration. Unfortunately, in some cases we have too few research subjects at our disposal to arrive at reliable conclusions with regard to such interaction effects. Follow-up research can shed new light on this intriguing research question. Further, it is undeniably so that the electoral interest of the individual candidate does not always correspond with the electoral interest of the party. This is specifically the case in an electoral system where the voter is entitled to cast multiple preference votes. A candidate who has obtained only a small number of preference votes can be important to a party if these votes would otherwise have gone to a competing party. Future research should therefore also focus on the electoral interests 25

26 of the party. Finally it would be interesting to examine whether our findings also hold true in a more rural context 20 October 2004, 7846 words 6. Endnotes 1 The analyses reported here were funded through grants and contracts from the Fund for Scientific Research Flanders, the Ministry of the Flemish Community and the Antwerp City Council. 2 An analysis of the correlations between the variables included in our model(s) pointed out that none of the variables were highly correlated with each other. Multicollinearity therefore did not occur. 3 We based ourselves on the summed scores for the intensity of membership (no member = 0; passive member = 1; active member = 2; executive member = 3) of associations in nine societal areas (environment, youth cultural, socio-cultural, neighbourhood, religious, sports, counselling and trade unions). The image of the omnipresent local politician is certainly not confirmed. While the maximum score was 27, the average and median were respectively 4.57 and 4. 4 There are indications that this is indeed the case. We asked the candidates also to mention what they considered to be their most important memberships. The vast majority appeared to be locally relevant associations. 5 For this variable, we first and foremost compiled a list of professions in which one meets relatively many people. 110 candidates appeared to be in such professions. In the second instance, we ascertained for this subsample whether they exercised their profession within the perimeter of the electoral district. We based ourselves on the commuter time to and from work. Candidates in a locally representative profession should not spend more than 15 minutes commuting. In all, 38 candidates were retained on this basis. 6 We use oral interviews with 53 leaders of the slating committees (one per party in each district). These interviews were taken during the month of September in the year 2000, one month before the elections took place. Amongst others we asked them to judge the individual advantages of the candidates. These judgements allowed us to determine which candidates were VIP s. Candidates who 26

27 the leaders said to be well-known thanks to their involvement in voluntary associations or political activities were not included. This way this category remained clearly separated from the categories in the other steps in our model. 7 In practice, we used a 6-point scale ranging from a great deal of time (1) to no time at all (6). For that matter, the correlation between the scores on this scale and the position occupied on the list is 0.17 and is significant at the 0.01 level. 8 It would be interesting to ascertain whether such an effect also occurs in the combination of ethnic minority and ballot position. In view of the small number of relevant research subjects, we were unable to draw conclusions regarding this aspect. 9 In view of the great number of variables in the final model (21), we have opted to use the adjusted R² as a measure of explained variance % of the female candidates had previously sat on the district council, compared to 18.4% among the male candidates in the research population. 11 If we ignore the variable first-placed woman in the explanatory models, the beta of the variable woman would rise in the fifth and sixth step to It is possible that our limited number of ethnic background candidates is not proportionally representative for the population. An analysis at population level confirms our finding. The variable ethnic minority member is significant from the introduction of list-related factors (sig. = 0.012) and, in the final step, becomes significant at The correlation between social insertion and Previous mandate on a district council on the one hand and ballot position on the other are respectively 0,196 and 0,197. Both coefficients are significant on the 0,01 level. 14 The nine separate association variables substituted the aggregated variable, while all other variables were left unchanged. 15 As, in the case of political associations, it usually concerns party membership, this was not taken into account in the construction of the variable. 16 It can again be argued that we are concerned with a relatively small number of cases (there were only seven VIP candidates in the final step of our model), but an analysis at population level confirms this finding as well. 27

28 17 The correlation between the variable position on list and campaigning is statistically significant (r= **). The higher the candidate s position on the list, the more time he/she claims to have invested in his/her campaign). 18 This would appear to have been a deliberate strategy of SP and VLD in particular (4 out of 5 leading candidates each). CVP, on the other hand, appears primarily to have chosen candidates who had won their spurs at another policy level: three of the five leading CVP candidates in the sample had previously held a mandate at a higher level votes versus 534 votes for the candidate on top of the list and 360 for the candidate at the bottom of the ballot list. 20 In order to measure the effect of a candidature on an AGALEV list, we used the CVP as a residual category. 7. References Ackaert J. (1994) Het gebruik van de voorkeurstem bij de gemeenteraadsverkiezingen. Een terreinverkenning in de provincie Limburg, Res Publica 36: Ackaert, J. (1995) De gemeenteraadsverkiezingen van 9 oktober Analyse van de resultaten, Res Publica 37: Ansolabehere, S., Gerber, A. (1997) Incumbency Advantage and the Persistence of Legislative Majorities, Legislative Studies Quarterly 22: Born, R. (2000) Congressional Incumbency and the Rise of Split-Ticket Voting, Legislative Studies Quarterly 25: Calhoun, C. (1994) Social theory and the Politics of Identity, Oxford: Blackwell. Campbell, A., Converse, P.E., Miller, W.E., Stokes, D.E. (1966) The American Voter, New York: John Wiley & Sons. 28

29 Carton, A. (1998) Over de actie stem vrouw en de plaats van de vrouw in de politiek, in: Swyngedouw, M., Billiet, J., Carton, A., Beerten, R., De (on)redelijke kiezer, Leuven/Amersfoort: ACCO, pp Connaughton, S., Jarvis, S. (2004) Invitations for Partisan Identification: Attempts to Court Latino Voters Through Televised Latino-Oriented Political Advertisements, , Journal of Communication 54: Cox, G. W., Katz, J. N. (1996) Why Did the Incumbency Advantage in U.S. House Elections Grow?, American Journal of Political Science, 40: Cutler, F. (2002) The Simplest Shortcut of All: Socio-demographic Characteristics and Electoral Choice, The Journal of Politics 64: Dalton, R. J. (1999) Political Support in Advanced Industrial Societies, in: Norris, P., Critical Citizens. Global Support for Democratic Governance, New York: Oxford University Press, pp Darcy, R., McAllister, I. (1990) Ballot Position Effects, Electoral Studies 9: DeSantis, V., Renner, T. (1994), Term Limits and Turnover Among Local Officials, in: Municipal Year Book, Washington, DC: International City/County Management Association, pp Desposato, S., Petrocik, J. (2003) The Variable Incumbency Advantage: New Voters, Redistricting, and the Personal Vote, American Journal of Political Science 47: Dewachter, W., Janssens, P., Voorkeurstemmen bij gemeenteraadsverkiezingen. Een onderzoek aan de hand van de gemeenteraadsverkiezingen van 1982 en 1988 in Brabant, Tijdschrift van het Gemeentekrediet 48:

30 Deweerdt, M. (1994) Verkiezingen, in: Deweerdt, M., De Ridder, C. en Dillemans, R., Wegwijs Politiek, Leuven: Davidsfonds, pp Farrell, D. M. (2001) Electoral Systems. A Comparative Introduction, New York, Palgrave. Galligan, Y., M. Laver, G. Carney (1997) The Effect of Candidate Gender on Voting in Ireland, 1997, Irish Political Studies 14: Geys, B. Heyndels, B. (2003a) Ballot layout effects in the 1995 elections of the Brussels Government, Public Choice 116: Geys, B., Heyndels, B. (2003b) Influence of cognitive Sophistication on Ballot Layout Effects, Acta Politica 38: Herring, M., Forbes, J. (1994) The overrepresentation of the White Minority: Detroit s At-Large City Council, , Social Science Quarterly 75: Hessing, R. C. (1985) Bij voorkeur: een onderzoek naar het gebruik van voorkeurstemmen, Acta Politica 2: Huddy, L., Terkildson, N. (1993) Gender Stereotypes and the perception of Male and Female Candidates, American Journal of Political Science 37: Hooghe, M. (2003) Sociaal kapitaal in Vlaanderen, Amsterdam: Amsterdam University Press. Koppell, J., Steen, J. (2004) The Effects of Ballot Position on Election Outcomes, Journal of Politics 66: Krebs, T. (1998) The Determinants of Candidates Vote Share and the Advantages of Incumbency in City Council Elections, American Journal of Political Science 42:

31 Lieske, J. (1989) The Political Dynamics of Urban Voting Behavior, American Journal of Political Science 33: Mackay, F. (2003) The Scottish Parliament Elections, 2003 Women and the 2003 Elections: Keeping up the Momentum, Scottish Affairs 44: Marsh, M. (1985) The Voters Decide? Preferential Voting in European List Systems, European Journal of Political Research 13: McDermott, M. (1997) Voting Cues in Low-Information Elections: Candidate Gender as a Social Information Variable in Contemporary United States Elections, American Journal of Political Science 41: McNair, B. (1999) An introduction to political communication, London: Routledge. Miller, J., Krosnick, J. (1998) The Impact of Candidate Name Order on Election Outcomes, Public Opinion Quarterly 62: Mueller, J. E. (1970) Choosing among 133 Candidates, Public Opinion Quarterly, 34: Nay, O. (2001) Institutions and Representation: How Institutional Rules Shape Political Recruitment in French Regional Elections, West European Politics 24: Popkin, S. (1991) The Reasoning Voter, Chicago: University of Chicago Press. Prewitt, K. (1970) The recruitment of Political Leaders: a Study of Citizen- Politicians, New York: The Bobbs-Merill Company Inc. Putnam, R. (1993) Making Democracy work, Princeton: Princeton University Press. Smits, J. en Thomas, I. (1998) Het gebruik van de meervoudige voorkeurstem bij de parlementsverkiezingen van 21 mei 1995, Res Publica 40:

32 Stein, L., Fleischmann, A. (1987) Newspaper and Business Endorsements in Municipal Elections: A test of conventional Wisdom, Journal of Urban Affairs 9: Swanson, D., Mancini, P. (1996) Politics, Media, and Modern Democracy. An International Study of Innovations in Electoral Campaigning and Their Consequences, Westport, Connecticut, London: Praeger. Thijssen, P., Van Assche, D. (2002). In het oog, in het hart? De Antwerpse districtraadsverkiezingen en de kloof tussen burger en bestuur, Res Publica 44: Young, L. (1996) Women's Movements and Political Parties: A Canadian-American Comparison, Party Politics 2: Wauters, B. (2000) De kracht van de voorkeurstem in stad en dorp, Leuven: Afdeling Politologie. 32