Too much choice, too little impact: a multilevel analysis of the contextual determinants of preference voting

Abstract

Despite the rich and growing body of research addressing how turnout and party choice depend on the institutional context, far less is known about the impact of the political environment on voters’ propensity to vote for candidates – not parties. Recent single-country studies have focused almost exclusively on individual-level resource- and identity-based differences in preference voting. Combining data from the Comparative Study of Electoral Systems (CSES) and Participation and Representation in Modern Democracies (PARTIREP) election studies in six countries, this article provides the first comprehensive, cross-national test of the impact of macro-contextual factors on a voter’s decision to indicate a candidate preference, instead of simply casting a party list vote. It demonstrates that both the failure of preference votes to affect the allocation of seats and choice overload dissuade voters from marking a candidate name on the ballot. These contextual factors affect informed and uninformed voters differently, moreover. The findings have important implications for electoral scholars and political practitioners when designing electoral systems.

Keywords:

• Preference voting
• complexity
• choice set size
• effectiveness
• political sophistication
• flexible and open-list systems

Acknowledgement

The data used in this publication were collected by the Comparative Study of Electoral Systems (CSES) program and the PARTIREP project, funded by the Belgian Federal Science Policy (BELSPO). The authors would like to thank the organisers and participants of the Rome Workshop ‘Comparative Perspectives on Electoral Behaviour’ for their helpful comments on earlier drafts of this paper. Audrey André further acknowledges FRS-FNRS for financial support.

Notes

1. In Belgium, the allocation of seats proceeds slightly differently from the other flexible-list systems. High-ranked candidates are advantaged because (half of) the list votes are added to the top candidate’s preference votes until (s)he equals the legal threshold and is elected. Any remaining list votes are then added to the next candidate on the list and so on, until half of the list votes have been redistributed. Should there remain seats to allocate when the list votes are depleted, these seats are allocated on the basis of candidates’ preference votes alone.

2. The 2006 reform increased the number of candidates a voter can circle on the ballot from two to four. In addition, the legal threshold a candidate’s preference votes must exceed to gain election irrespective of his or her position on the list was lowered from 7% to 5% of the party vote in the district. Not only did the proportion of voters casting a preference vote witness a sharp increase (see Table 1), the proportion of elected candidates who obtained a number of preferences votes at least equal to the legal threshold also grew from 21% to 60%. In addition, the proportion of candidates lower on the list who would not have gained election if not for their preference votes increased from 3.5% to 23.5% (André et al. 2015).

3. Due to the smaller sample size combined with a larger number of districts (39), the rate of preference voting ranges more widely in Austria. The Austrian data include nine districts with fewer than ten respondents. If we exclude these districts from the aggregate-level analysis reported in Table 3, however, the findings remain unaltered.

4. The official results list only the total number of preference votes, not the number of ballots marking preferences for individual candidates, providing only a minimum and maximum number of voters casting preference votes. Estimations of the number of voters casting a preference vote are based on the mean of the two.

5. Some would argue that voters first decide on the party and only subsequently on the candidate (van Holsteyn and Andeweg 2010). Re-examining the evidence, however, we find the number of candidates a voter’s preferred party runs in the district to have a similar negative and significant effect on his or her probability of casting a preference vote. Because at this stage of the research we do not know very much about the sequence of choice – the evidence from STV systems points strongly to the heterogeneity of the electorate in this regard (see Marsh 2007) – we are hesitant to make strong a priori assumptions about voter decision processes and prefer to use the total number of candidates running in the district.

6. To justify the use of multilevel analysis, the two-level intercept-only model was tested against the single-level specification using likelihood ratio tests. Adding the random intercept reduces the deviance by 305, thereby strongly rejecting the null hypothesis that the variance at the district level does not differ from zero (p < 0.000).

7. Individual-level determinants of preference voting, such as age, sex, education, party identification or ideological position are not conditioned by either the choice set size, or the effectiveness of preference voting: no interaction term is significant at the conventional 5% level.

8. Because in random slope models the intercept-slope covariance is simultaneously estimated, the statistical significance of the slope variance is tested against the chi-bar distribution as recommended by Snijders and Bosker (2012). Even when using this more conservative test, the deviance difference of 6.4 is associated with a p-value smaller than 0.05.

9. The absence of an interactive effect between education and choice set size further corroborates our interpretation that cognitive limitations per se are not responsible for the effect, but rather voters’ propensity to ‘satisfice’.