Check for updates
Article
PartyPolitics
Party Politics
_                                  .            .                                     . 2017, Vol. 23(5) 589-600
Party nomination strategies ©TheAuthor(s)2oi5
* Reprints and permission:
■ m   #1 f^y ■ 1^ I a   I ■          rwr^/\ |%^gt sagepub.co.ul</journalsPermissions.nav
III   IICAIUIC-II31  9piCIII». DOI: 10.1177/1354068815610974
tT                                     a                        a a         ^ journals.saeepub.com/home/ppq
Do preference votes matter: <Dsage
Audrey André
Department of Political Science, Vrije Universiteit Brüssel, Brussels, Belgium Sam Depauw
Department of Political Science, Vrije Universiteit Brüssel, Brussels, Belgium Matthew S Shugart
Department of Political Science, University of California, Davis, CA, USA Roman Chytilek
Department of Political Science, Masaryk University, Brno, Czech Republic
Abstract
A paradox in the comparative literature on electoral systems is that one of the most common systems in Europe - flexible-list proportional representation systems - may be the least understood. Any study of flexible-list systems must start by acknowledging a puzzle: why candidates spend time and effort striving to win preference votes when typically these votes make no difference between election and defeat. Offering the first comprehensive multi-country test of this key puzzle, we provide evidence from Belgium, the Czech Republic, and Slovakia that parties will promote to better list ranks in the next election those candidates who are successful at winning preference votes, thereby improving their prospects of election in the longer term and incentivizing them to cultivate personal reputations. Our findings have important implications for party scholars and practitioners when designing, or reforming, political institutions.
Keywords
candidate list position, flexible-list system, party strategy, preference votes
Introduction
A paradox in the comparative literature on electoral systems is that one of the most common systems in Europe may be the least understood. Used in Austria, Belgium, the Czech Republic, Denmark, Estonia, the Netherlands, Slovakia, and Sweden, flexible-list proportional representation systems1 are almost as common as closed-list and open-list PR systems. Given their widespread use, we should - at least try to - understand them better. Yet, oftentimes these systems are considered akin to closed-list systems; others lump them in with open lists; and still others ignore them altogether. Flexible-list ballots allow voters to mark a preference for a candidate and, because the candidate who wins preference votes at least equal to some legal threshold gains election irrespective of his or her position on the
ballot, candidates energetically cultivate personal reputations (André et al., 2015; Bráuninger et al., 2012). However, in practically all flexible-list systems there is strong evidence that very few candidates win substantial preference votes and thus many party seats are allocated in the order candidates appear on the ballot (see Andeweg, 2005; André et al., 2012; Beblavý and Veselkova, 2014; Bergman and Bolin, 2011; Lebeda, 2007; Můller, 2005).
Paper submitted 13 July 2015; accepted for publication 10 September 2015
Corresponding author:
Audrey André, Department of Political Science, Vrije Universiteit Brüssel, Pleinlaan 5 (office 2.63), Brüssel, 1050, Belgium. Email: aandre@vub.ac.be
590
Party Politics 23(5)
Any study of flexible-list PR systems must therefore start by acknowledging a puzzle: that is, why (re)election-seeking candidates spend time and effort striving to win preference votes when typically these votes make no difference between election and defeat. One reason is that parties benefit from preference-vote seeking. Thus they will promote to better ranks in the next election those candidates who are successful at earning preference votes, thereby improving their prospects of election in the longer term. This is the argument raised by Crisp et al. (2013) with regard to Slovakia (but see also Andeweg, 2005; De Winter, 2005; Muller, 2005). The Slovak legislator winning more preference votes, they find, is more likely to get a better list rank in the next election, all else equal. Parties reward preference-vote seeking, they find, even at the price of legislative voting unity because breaking ranks with the party may help earn votes.
This paper offers the first - as far as we know - multi-country empirical study of flexible-list systems - focusing on Belgium, the Czech Republic, and Slovakia - and presents the toughest test to date of this key puzzle. More specifically, our analysis goes beyond that of Crisp et al. (2013); first, by including districted PR systems where the number of ballot positions available to parties for freely moving candidates up and down a list is far more restricted compared to the nationwide Slovak party lists of 150 candidates. Second, our analysis includes all candidates, winning or losing, rather than only incumbent candidates. This is crucial to account for the intuition that the puzzle particularly pertains to the candidates who cannot realistically think to gain election from positions near the bottom of the ballot. Third, we test this puzzle using different modelling strategies and alternative operationalizations of the key variables.
This paper therefore contributes to the literature on political representation and electoral system effects by addressing the puzzle of personal vote incentives generated by flexible-list PR systems. In particular, we present new data on 14,255 candidates running in three consecutive elections in Belgium, the Czech Republic, and Slovakia. We demonstrate that in flexible-list PR systems candidates' strong performance in preference vote-earning is likely to gain them better list positions in the next election. In addition, we compare the direct and indirect impact of preference voting across the three countries, providing new evidence that flexible-list systems are not as similar to closed lists as previously thought. Our findings should be of interest to scholars of comparative politics, political parties, and candidate selection processes, as well as to political practitioners and electoral engineers.
Theory: Why parties reward personal vote-earning
List-PR systems differ in the manner in which they allocate seats not only to parties, but also to candidates. They have in common that, where voters have the option to mark a
preference for a candidate, these votes are pooled across all candidates running under the same party label to determine the number of seats the party gets. Yet, whereas closed-list systems allocate seats to candidates in the order they appear on the party ballot; candidates in open-list systems are elected in the order determined by the preference votes they poll. As such the literature has focused on the ensuing tension between party and candidate strategies in the latter (see Carey and Shugart, 1995). That is, open-list systems in particular have been found to incentivize candidates to court preference votes in an attempt to gain election: in these systems candidates will run more personal campaigns (Bowler and Farrell, 2011; Zittel, 2015), and - once elected - will do more constituency service (André and Depauw, 2013; Heitshusen et al., 2005), introduce particularized legislation (Crisp et al., 2004), and break from the party ranks more often (Carey, 2009; Sieberer, 2006). In closed-list systems, by contrast, they will concentrate exclusively on presenting to voters the coherent policy packages that their party pledges to pursue in office (see Kitschelt, 2000).
With regard to preference vote-seeking, however, flexible-list systems are something of a hybrid, expected to 'fall somewhere in between' closed-list and open-list systems (Shugart, 2005: 47). In these systems, the allocation of seats takes into account both a candidate's preference votes and his or her position on the party ballot (Karvonen, 2011; Katz, 1986; Marsh, 1985). It has been the assumption that candidates' vulnerability to defeat is the motivating force energizing them to seek preference votes. But why would those candidates insulated from defeat by their high position on the party ballot seek preference votes? Or indeed, why would candidates in the lower preelection ranks devote time and effort to campaigning for preference votes? Only a mere handful gain election from a position on the ballot that would not otherwise have seen them elected (see Andeweg, 2005; André et al., 2012; Beblavý and Veselkova, 2014; Bergman and Bolin, 2011; Lebeda, 2007; Muller, 2005). The answer, we argue, is in the party's nomination strategy.
Whereas candidates are assumed to want to maximize their probability of gaining election, parties will want to maximize their number of seats in the legislative arena (see also Bergman et al., 2013). Seats, after all, are the currency of power in the formation of cabinets and passage of legislation. Because preference votes are pooled across co-partisans, however, seat-maximizing parties can expect to benefit from fielding candidates capable of generating preference votes. A preference vote is invariably also a vote for the party. To the degree that a candidate's preference vote-seeking will bring in voters who would not otherwise have voted for the party, the party will increase its vote share and possibly gain an additional seat. There is ample evidence that voters' decision to vote (Adams and Merrill, 2003), as well as the party they vote for, can be affected by the candidates on offer (McDermott, 2009; Stone et al.,
André et al.
591
2010; Tavits, 2010). However, that a candidate will bring in additional votes is neither self-evident, nor very unlikely. A recent survey experiment for instance suggests that, if their preferred candidate had run under the label of a different party, as many as nine percent of Dutch voters would have followed their candidate and voted for the other party (van Holsteyn and Andeweg, 2010)!
As a result, where possible, nomination strategies should favor candidates capable of winning preference votes. The literature on candidate selection and recruitment has long recognized that risk-averse selectorates value incumbents and candidates with otherwise proven track records of winning votes (Gallagher and Marsh, 1988; Hazan and Rahat, 2010; Norris, 1997). Candidate selection is one of the defining functions of a political party and, more than its organization or even its manifesto, a party's candidates will define and determine the party's public face not only in the elections, but long after (Katz, 2001). Selectorates, studies indicate, look for quality candidates that will appeal to voters. Thus they favor candidates that are articulate, able to get on with people, and have gained public prominence in national or local politics (see Gallagher and Marsh, 1988; Norris, 1997; Put and Maddens, 2013). Of course, media coverage, campaign spending, and strong roots in the local community all increase a candidate's appeal to voters (Maddens and Put, 2013; Tavits, 2010). But what better predictor of a candidate's future ability in this regard than his or her preference vote tally in the past?
In particular, risk-averse selectorates do not want their parties to appear out of step with voters. The public impression that party cabals succeed in electing members to the Parliament, even if these have been disavowed by voters, can be damaging to a party (Marsh, 1985; see also De Winter, 1988).2 This concern arises especially in flexible-list systems because preference vote tallies are public proof of a candidate's support, or lack thereof, among voters, whereas the large majority of candidates gain election, or face defeat, dependent only on their position on the ballot. Clearly, it would be a risky strategy to disregard the voters' preferences altogether; nor is the observation new. To ward off partitocratic allegations of this kind, one early study noted, parties in the Netherlands 'take good care to present their candidates in a sequence that will not ensure the displeasure of their supporters' (Van Raalte, 1959: 86; see also Andeweg, 2005). Similar comments have been made regarding party nomination strategies in Austria (Miiller, 2005), Belgium (De Winter, 1988), and Sweden (Rustow, 1955).
Risk-averse selectorates should give candidates who beat expectations and are capable of winning more preference votes than anticipated a better list rank in the next election. This is exactly, Crisp et al. (2013) pointed out, what Slovak parties do: incumbents winning large preference vote tallies are moved up the list at the next election, giving them an incentive to seek preference votes even if
these cannot see them elected this time. If nomination strategies reward preference vote-earning, it strengthens the conclusion that parties are aware of candidate performance and respond to it. But it is important to acknowledge that selectorates face other demands as well - demands from different local branches, from party factions, interest groups, and social movements that their candidate be included. Parties in PR systems prefer to balance the ticket most notably in terms of sex, age, ethnic minorities, and place of residence (see Gallagher and Marsh, 1988). Ticket-balancing is an obviously rational strategy to widen a party's appeal to different voters. Many voters will only vote for a local candidate or a candidate who shares their gender or ethnic identity (for instance, André et al., 2012; Bengtsson et al., 2014; McDermott, 2009). An unbalanced ticket risks internal discontent and ultimately defeat at the polls (Gallagher and Marsh, 1988). But ticket-balancing also restrains selectorates when seeking to reward preference-vote earning. These restraints are likely to be stronger as the party ballot grows shorter and there simply aren't enough positions to accommodate all successful candidates. For this reason, we aim to contrast nomination strategies in the Slovak single nationwide constituency to two cases of districted PR: Belgium and the Czech Republic.
The cases of Belgium, the Czech Republic, and Slovakia
Belgium, the Czech Republic, and Slovakia are particularly interesting cases to study the importance of preference votes for party nomination strategies in flexible-list systems. All three elect members to their national parliament using single-tier PR systems that are thus 'simple' in Taa-gepera's (2007) sense.3 Whereas in Slovakia seats are allocated using the Hagenbach-Bischoff largest remainder method, Belgium and the Czech Republic employ the D'Hondt method of highest averages. Both methods are known to yield identical results, however (see Blais and Massicotte, 2002). We further contrast one nationwide to two districted PR systems, allowing for a wide variety in the length of party lists.4 Whereas all 150 seats in the Slovak parliament are allocated in a single nationwide constituency, there are 11 constituencies in Belgium and 14 in the Czech Republic. They range in magnitude from four to 24 and from five to 25 respectively.5 In all three countries a legal threshold of five per cent is in effect - at the level of the district in Belgium, but nationwide in the Czech Republic and Slovakia.6
In these three countries, parties submit lists on which candidates are ranked. Voters have the option of casting a single vote for the party list as a whole or of indicating preference votes for specific candidates on that list.7 Czech and Slovak voters can circle the names of up to four individual candidates on a list. Before the 2006 reform Czech voters could indicate up to two preferences. By contrast,
592
Party Politics 23(5)
Table I. Preference voting in Belgium, the Czech Republic, and Slovakia.
Elected candidates obtaining Candidates elected from
the preference vote threshold a lower list position
% Voters casting - -
preference votes threshold N % N %
Belgium	2003	66	Droop	35	23.3	18	12.0
	2007	61	Droop	33	22.0	16	10.7
	2010	58	Droop	35	23.3	1 1	7.3
Czech Republic	2002	29	7%	46	23.0	12	6.0
	2006	27	7%	42	21.0	7	3.5
	2010	44	5%	120	60.0	47	23.5
Slovakia	2002	68	10%	31	20.7	1	0.7
	2006	78	3%	63	42.0	7	4.7
	2010	74	3%	61	40.7	1 1	7.3
Note: The total number of seats in the Chamber of Representatives is 150 in Belgium and Slovakia and 200 in the Czech Republic. Source: Authors' own calculations based on the official election results obtained from http://polling20IO.belgium.be/ for Belgium, from http://www.volby.cz/ for the Czech Republic, and from http://slovak.statistics.sk/ for Slovakia.
voters in Belgium can tick the box of as many candidates as run on the same list.8 Surveys suggest that fewer than 10 per cent of Belgian voters indicate more than four, however (André et al., 2012). Table 1 reports the proportion of the valid ballots that indicate preferences for one or more individual candidates per country. The number of voters casting a preference vote is highest in Slovakia, where between 2002 and 2010 about 70 per cent marked a preference for a candidate. In Belgium about 60 per cent cast a preference vote between 2003 and 2010; whereas the number is much lower in the Czech Republic.9
Preference votes have the ability to alter the order in which candidates are elected: the candidates polling preference votes at least equal to some legal threshold are elected first, provided their parties have won sufficient seats to accommodate them. That threshold differs across countries and elections. In Slovakia, the threshold is three per cent of the party vote following the 2004 reform; before it was 10 per cent. In the Czech Republic, the threshold is five per cent of the party vote in the district; before the 2006 reform it was seven per cent. In both countries any remaining seats for the party are awarded to candidates who did not clear the threshold, following the order on the ballot. The lowering of the threshold in Slovakia, Table 1 indicates, doubled the number of elected candidates obtaining preference votes at least equal to the threshold. Most often they are prominent politicians whose high rank on the ballot would have ensured them election regardless. Post-reform, only five to seven per cent of Slovak legislators are elected from a position so low that they would not have been elected were it not for their preference votes (see also Beblavý and Veselkova, 2014). In the Czech Republic, the lowering of the threshold, combined with an increase from two to four preferences voters are allowed to indicate, has even trebled the number of elected candidates clearing the threshold. Post-reform, a record of one in four Czech legislators in 2010 gained election solely on the basis of their preference
votes, suggesting a dramatic change towards more open lists (see also Stegmaier and Vlachova, 2011; Voda and Pink, 2010).
By contrast, the threshold in Belgium is the Droop quota of the total party vote in the district (i.e. the party vote divided by the number of seats won, plus one) - resulting in a variable threshold ranging from eight to 50 per cent of the party vote. As a result, only about one in five Belgian legislators obtained the threshold, Table 1 indicates - about as many as in pre-reform Slovakia and the Czech Republic. However, contrary to the Czech Republic and Slovakia, seats are not allocated among the candidates not clearing the threshold in the order they appear on the ballot. Instead, (half of the) list votes are transferred in the order of the list.10 That is, list votes are added to the top candidate's preference votes until (s)he obtains the Droop quota and is elected. Then the remainder of the list votes is 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, third, these seats are allocated on the basis of candidates' preference votes alone. As such, fewer than 10 per cent of those elected to the Belgian Parliament are elected from positions lower on the list (who would not have been elected on the basis of their pre-election list position); that is about as few as in post-reform Slovakia. Due to the differences in the rules for intra-list allocation, not all of them polled more preference votes than the legal threshold, however.
Data
The data include 14,255 candidates in three consecutive elections, based on the official results obtained from the respective Ministries of the Interior. Included are all parties that have won at least one seat in any of the elections. The largest excluded party polled no more than 4 per cent in the Czech Republic, 3 per cent in Slovakia, and 1 per cent in
Andre et al.
593
Table 2. The number of returning candidates in three consecutive elections.
		Belgium		Czech Republic		Slovakia	
		N	%	N	%	N	%
eti	# cand. not running in et2	1051	70.3	1 189	70.2	1 131	76.5
	# cand. running again in et2	443	29.7	504	29.8	348	23.5
et2	# cand. not running in et3	910	60.8	1467	72.2	806	67.3
	# cand. running again in et3	586	39.2	566	27.8	392	32.7
Belgium. For reasons of comparability we further exclude 16 candidates who switched parties and 29 who switched districts. Despite the differences in electoral institutions, the number of candidates running is remarkably similar across the three countries. That is, the data include a total of 4625 candidates running in the 2003, 2007, and 2010 elections in Belgium; in the Czech Republic and Slovakia we have a total of 5757 and 3873 candidates, respectively, running in the 2002, 2006, and 2010 elections.
Testing the hypothesis that preference vote-earning is rewarded at the next election, by necessity, we look at returning candidates. About 70 per cent, Table 2 indicates, of the candidates running in any of the elections, did not run again in the election thereafter. By contrast, we concentrate on the 30 per cent who did run again and for whom we can compare their list position in election t and in election t+1. The proportion of returning candidates is roughly similar across countries and elections. Only the 2010 early elections in Belgium stand out in this respect: up to 39 per cent of the 2007 candidates ran again in 2010. Early elections present party selectorates with severe time constraints to have full slates of candidates ready in time which likely results in greater numbers of candidates being reselected.11 In all three countries incumbents are about three times more likely to run again in the next election than non-incumbents; even so in all three countries non-incumbents outnumber incumbents three to one among the returning candidates.
In order to ascertain the effect of preference vote-earning on a candidate's list position in the next election, it is clear the dependent variable is some measure of change in list position between elections. The top panel of Figure 1 depicts the distribution of changes in list position between election t and election t+1, suggesting both continuity and change. Continuity is illustrated by the number of candidates who run again in election t+1 from the exact same list position they ran from in election t. In Belgium, that number is 22 per cent; 14 per cent in the Czech Republic. But the main distinction is between both districted PR systems and Slovakia: in the latter, only 5 per cent of all returning candidates retain their list position at election t+1. At the same time, the distributions depict ample changes in list position across elections, although most move up, or down, only a few positions. But there are candidates who make large leaps, in Belgium but especially in Slovakia where lists of 150 candidates offer party selectorates the greatest
discretion for rewarding candidates' preference vote-earning or for sanctioning poor performance.
What is less clear, however, is how to measure a candidate's preference-vote earning, the independent variable of main interest. Previously, Crisp et al. (2013) used the raw numbers of preference votes polled. But raw numbers of preference votes are not comparable across districts or parties (being a function of population size and the number of voters using preference votes). More importantly, a party's decision whether to move up, or down, a candidate should not be affected by the presence of someone polling a larger number of preference votes in another district or for another party. For this reason, we reorder all candidates on a given list in function of the number of preference votes won. Preference vote-winning performance is then measured by the candidate's rank difference: that is, by subtracting the candidate's party list rank from his or her voter rank in terms of preference votes. A positive rank difference indicates that a candidate is ranked higher by the voters than by the party; a negative rank difference suggests the candidate has less support among voters than among the party selectorate. In other words, a candidate's rank difference is an indication of the difference in his or her prospects for election if an open-list PR system were used (based solely on candidates' rank in preference votes) and if a closed-list PR system were used (based solely on their list position).
The bottom panel of Figure 1 depicts the distributions in rank difference of the returning candidates in any of the three consecutive elections. In the Czech Republic and Slovakia, the distributions are roughly symmetrical. About as many performed above par, winning more preference votes than some co-partisans higher on the list (resulting in a positive rank difference), as there are who performed subpar, winning fewer preference votes than one or more co-partisans lower on the list (resulting in a negative rank difference). Only in Belgium are there a larger number of candidates with positive rank differences. In addition, in the districted PR systems, a greater number of candidates' voter rank does not match their party rank. About eight per cent of Slovak candidates have a rank difference of zero, indicating that they met expectations. Given position five on the list, for instance, they polled the fifth largest number of preference votes among those running on the same list. The number is twice as large in Belgium and the Czech Republic. Taken together, Slovakia amounts to
594
Party Politics 23(5)
BELGIUM
SLOVAKIA
-30 -20 -10   0    10   20   30 -30 -20 -10   0    10   20   30      -150 -100   -50 0
difference in list position between e, and et.i
SO    100 150
BELGIUM
CZECH REPUBLIC
SLOVAKIA
-30 -20 -10
10   20 30
30 -20 -10 0 10 20 30 preference rank difference at et
-150 -100 -50
50   100 150
Figure I. The distribution of changes in list position and rank difference.
something of a most likely case: the data suggest both a larger number of candidates performing out of sync and a larger number of list position changes. Belgium and the Czech Republic, by contrast, offer a much tougher test of the key hypothesis.
Method
In testing the hypothesis that preference vote-earning is rewarded at the next election, we capture changes in list position by estimating the candidate's list position at election t+1, using his or her list position at election t as a predictor. By including previous list position, we effectively control for all the individual properties, skills, and qualifications associated with candidate quality (see Jacobson and Kernell, 1981) that affect a candidate's list position in the first place. More precisely, OLS regression is used predicting the decimal logarithm of list position in order to correct for overdispersion: in the larger districts both great and small position rank numbers can be observed; whereas in smaller districts - where parties by law field fewer candidates - only small list position rank numbers are observed. Candidates' preference vote-winning performance is captured by their
rank difference. Because lower list position numbers indicate better positions on the ballot, we expect the effect of rank difference to be negative. To isolate the effect of rank difference from possible incumbency effects we reanalyze the models adding incumbency as an explanatory variable.
We present separate analyses for Belgium, the Czech Republic, and Slovakia, although we determined that pooling the data across the three countries would lead to identical conclusions. To correct for the interrelations between co-partisans' predicted list positions - a party has only one position of any given rank in a district - we cluster standard errors at the level of the party-list-in-the-district in each election. Because a logged dependent variable is not easy to interpret, moreover, the magnitude of the effects is assessed and compared by calculating the rank difference needed by a candidate to move up one position at the next election. That is, how much better must a candidate perform compared to his or her co-partisans, before the party selectorate will move him or her up the list one position?
In order to demonstrate that our findings are not driven by outliers whose voter rank is far removed from their party rank, we undertake an alternative specification in which the
Andre et al.
595
Table 3. OLS regression models of changes in list position.
Belgium Czech Republic Slovakia
Rank difference at et -0.010 (0.002)*** -0.010 (0.002)*** -0.002 (0.001)**
List position at et                 0.632 (0.035)*** 0.740 (0.021)*** 0.861 (0.031)***
Constant                           0.337 (0.033)*** 0.190 (0.020)*** 0.167 (0.055)**
N                                    1029 1070 740
LR(df) 392.14 (2)*** 728.27 (2)*** 778.86 (2)***
R2                                   0.32 0.49 0.65
Note: Entries are the parameter estimates and robust standard errors (in parentheses) of OLS regression models as well as the model's likelihood ratio test (LR). Standard errors are clustered by party list in each election. The dependent variable is the logarithm of a candidate's list position at et+!. The regression estimates are not affected by multicollinearity: the Variance Inflation Factor (VIF) never exceeds 1.19. *p < 0.05, **p < 0.01, ***p < 0.001, using two-tailed t values.
dependent and key independent variables are dichotomized. To further emphasize the robustness of our findings, we explore different modelling strategies and alternative operationalizations in the online supplementary material. First, negative binomial regression is used to take into account that list positions are positive integers. Second, Heckman selection models are added to stem concerns about a possible selection bias resulting from concentrating on returning candidates. Third, we offer an alternative operationalization of the key independent variable: while we argue that a candidate's rank difference best approaches what selectorates are looking for, we alternatively consider the number of preference votes a candidate polls as a percentage of the party total.
Results: The effect of personal vote-earning at et on list position at et+|
Party nomination strategies reward preference vote-earning, we find, estimating change in list position using a candidate's list position at election t as a predictor of their list position at election t+1. Across flexible-list PR systems, Table 3 demonstrates, the preference votes that candidates poll in an election are an important predictor of changes in list position in the subsequent election. Naturally, a candidate's list position at time t continues to be the strongest predictor of his or her list position at time t+1. If we had no information about a candidate's preference vote-earning, we would predict the candidate to be in roughly the same position as in the previous election. Solving the regression equation for different list positions predicts the candidate in position 10 for instance to run again from position 10 in Belgium, from position nine in the Czech Republic, and from position 11 in Slovakia. In all three countries, candidates further down the list are expected to move up a few positions, however. The lower the list position a candidate holds, the more he or she is likely to benefit from non-returning candidates vacating their higher-ranked slot.
More importantly, the candidate's rank difference at time t has a significant effect on the position he or she is
awarded at time t+1. Because position five for instance is lower down the list than position one, the effect shows up as a negative sign, as we see in Table 3. A positive rank difference, indicative of a candidate who wins more preference votes than anticipated on the basis of his or her list position, decreases the predicted position he or she will be in at time t+i; hence, the candidate will be moved up the list. Conversely, a negative rank difference, indicative of a candidate who performed subpar and polled fewer preference votes than co-partisans ranked lower on the list, increases the predicted position at time t+1. That is, the candidate will be demoted down the list in the next election.
Clearly, incumbents typically outperform non-incumbents and tend to be awarded better list positions. Incumbency and list position at time t are particularly strongly intercorrelated at 0.7; this is a direct result of how flexible-list systems operate. Yet, our finding that parties will reward preference vote-earning with better list positions in the next election should not be confused for incumbency effects. After including incumbency as an explanatory variable, Table 4 indicates, the regression coefficients are by and large unaffected: they are negative, statistically significant, and of roughly the same size. In spite of the strong intercorrelations, variance inflation factors are within acceptable margins. Considering that incumbents are most likely to benefit from increased media coverage during the term, in addition we can be confident that parties reward preference vote-earning at time t and our findings are not the result of mid-term changes in candidate profile between times , and t+1.
Interpretation of the magnitude of the effect is not straightforward, especially when moving from the single nationwide district in Slovakia to the districted PR systems of Belgium and the Czech Republic. To ensure correct interpretation, Figure 2 plots the rank difference required to be moved up one position on the list in the next election for different list positions. The Belgian candidate in position 20, for instance, requires a rank difference of four, all things equal. That is, he or she will be in position 19 in the next election, provided he or she obtains the 16th highest
596
Party Politics 23(5)
Table 4. OLS regression models of changes in list position, ruling out incumbency effects.
Belgium Czech Republic Slovakia
	b.	s.e.	b.	s.e.	b.	s.e.
Rank difference at et	-0.008	(0.002)***	-0.010	(0.002)***	-0.002	(0.001 )***
List position at et	0.555	(0.047)***	0.625	(0.031)***	0.768	(0.044)***
Incumbent	-0.109	(0.041)**	-0.166	(0.029)***	-0.146	(0.051)*
Constant	0.429	(0.050)***	0.331	(0.034)***	0.351	(0.075)***
N	1029		1070		740	
LR (df)	399.72	(3)***	763.22	(3)***	797.35	(3)***
R2	0.32		0.51		0.66	
Note: Entries are the parameter estimates and robust standard errors (in parentheses) of OLS regression models as well as the model's likelihood ratio test (LR). Standard errors are clustered by party list in each election. The dependent variable is the logarithm of a candidate's list position at et+|. The regression estimates are not affected by multicollinearity: the Variance Inflation Factor (VIF) never exceeds 2.29. *p < 0.05, **p < 0.01, ***p < 0.001, using two-tailed t values.
Figure 2. The estimated effect of rank difference.
number of preference votes of all candidates running on the same list. If he or she wins even more preference votes, the candidate may be moved further up the list.
In all three countries, the rank difference required to gain one position in the next election decreases as we move down the list. For positions near the top of the list, the results indicate, party selectorates are not as responsive to voters' preferences as other considerations enter their deliberation, most notably the composition of the parliamentary party after the election and any special knowledge each of its members may bring to it. Near the top of the list especially, large rank differences are required and rank differences are harder to achieve (up until the point when the rank difference required exceeds the number of co-partisans ranked higher than the candidate and the model predicts no change at all, depicted in Figure 2 by the dotted line). But as we move towards the bottom of the list, the rank difference required to gain one position decreases. In the bottom list positions, a rank difference of two to three is sufficient to be noticed, dependent on the country. The candidate who outperforms two to three candidates ranked higher on the list will be moved up the list, indicating
that in the districted PR systems in large districts party selectorates are more responsive to voters' preferences than in small districts.
How do the districted PR systems compare to the Slovak nationwide constituency? The comparison is made more difficult by the fact that a given list position may be a safe seat in one country, but a hopeless one in another. The fact that a greater rank difference is required for the Slovak candidate in position 13, for instance, to be moved up one position than the Belgian or Czech candidate in position 13 should not be taken to mean that Slovak party selectorates are less attentive to candidates' preference vote-earning. Rather, comparing the median candidate across the three countries - the candidate having as many candidates ranked above him or her as there are candidates ranked below him or her - we find Slovak parties to be at least as responsive on average as parties in the two districted PR systems. The median candidates in Belgium and the Czech Republic, at position 13, require a rank difference of five before the party will move the candidate up the list one position. In Slovakia, a rank difference of four suffices for the median candidate at position 75 to be promoted one position in the next election. However, outperforming four to five co-
Andre et al.
597
Table 5. Probit regression models using dummy coding of the dependent and independent variables.
		Belgium	Czech Republic			Slovakia
	b.	s.e.	b.	s.e.	b.	s.e.
Positive rank difference (dummy)	0.607	(0.078)***	0.637	(0.079)***	0.541	(0.063)***
Constant	-0.468	(0.051)***	-0.309	(0.053)***	-0.163	(0.086)#
N	1029		1070		740	
LR (df)	56.69		65.39		32.18	
Nagelkerke R2	0.07		0.08		0.06	
Pr(y= 1 | x=0)	0.32		0.38		0.44	
Pr(y = 1 |x = 1)	0.56		0.63		0.65	
Note: Entries are the parameter estimates and robust standard errors (in parentheses) of probit regression models as well as the model's likelihood ratio test (LR). Standard errors are clustered by party list in each election. The dependent variable is a dummy indicator for an increase in list position between et and et+|. The table further includes the predicted probabilities of an increase in list position. #p < 0.10,*p < 0.05, **p < 0.01, ***p < 0.001, using two-tailed t values.
partisans ranked higher is far easier to do when there are 74 of them than when there are only 12.
To stem concerns that our findings may be the result of a handful of extreme cases having very large observed rank differences, we recode the key dependent and independent variables into binary indicators. A change in position higher up the list is recoded to have a value of 1; 0 indicating a change lower down the list or no change. Similarly, a positive rank difference is recoded to a value of 1, indicating that a candidate ranks higher in terms of preference votes than his or her pre-election list position. Zeroes indicate that the candidate ranks lower in terms of preference votes than his or her list position, or polls as many preference votes as anticipated. Using probit regression, reported in Table 5, we find that any positive rank difference will be rewarded by a better list position in the next election. That is, we continue to find a significant effect of the candidate's preference vote-earning - even when disregarding any extra information we have about how well he or she performed. A candidate who is ranked higher by voters than by the party increases his or her probability of being promoted to a better list rank by 24 per cent in Belgium, by 25 per cent in the Czech Republic, and by 21 per cent in Slovakia. The observed effect, we conclude, is not confined to candidates whose voter support is particularly out of sync with their party list rank, but is a relatively common occurrence capable of incentivizing large numbers of candidates across flexible-list systems to seek preference votes.
Further robustness checks reported in the online supplementary material all provide additional support for our main conclusion: political parties in Belgium, the Czech Republic, and Slovakia are responsive to voters' candidate preferences. Negative binomial regression models, Heck-man selection models, and substituting rank difference for preference vote percentages all emphasize that party selec-torates reward preference vote-earning with promotion to a better list position at the next election.
Conclusion
Flexible-list systems are multifaceted hybrids, generating diverse and sometimes contradictory incentives on the part of candidates running for (re)election. Whereas at times up to 70 per cent of the voters mark a preference for an individual candidate, only for a mere handful of candidates do these preference votes make the difference between election and defeat. Presenting the first multi-country, comprehensive study of 14,255 candidates running in three consecutive elections in Belgium, the Czech Republic, and Slovakia, we demonstrate that, in flexible-list systems, preference votes have an important indirect effect on a candidate's prospects for a career in politics, capable of incentivizing them to spend considerable time and effort chasing preference votes. That is, using different modelling strategies and various operationalizations, we find that party nomination strategies reward preference vote-earning with better list positions in the next election, improving the candidate's probability of gaining election in the longer term. Despite important differences most notably with regard to list length, we find this indirect effect to be as strong in districted PR systems as in a system using a single nationwide constituency.
Flexible-list systems have frequently been thrown in with closed-list systems, accountability being collective and operating only through political parties. Yet, despite its limited direct effect, preference voting in flexible-list systems introduces an element of individual accountability operating at the level of the individual (would-be) representatives. That is, even though the flexible-list system precisely allows party leaderships to exert more effective control over access to the parliament than would an open list, party selectorates are still attentive to candidate performance and responsive to voter preferences. Because party nomination strategies will sanction preference vote-earning, voters' disapproval can effectively cut a candidate's career short, suggesting electoral competition in flexible-list systems to be more candidate-centered than previously anticipated. Parties' responsiveness in this
598
Party Politics 23(5)
respect is further supported by the growing public demand for flexible-list systems to be made more open and the trend across flexible-list systems to lower institutional thresholds through electoral reform (see Renwick and Pilet, 2015).
Research has only scratched the surface of how flexible-list systems operate, however. Future studies will have to explore the many tradeoffs parties face in these systems and the strategic choices they make balancing conflicting demands by turning to the key actors involved: that is, party selectors and candidates. Future interview data would be helpful to corroborate the behavioral pattern we observe. Anecdotal evidence strongly suggests that candidates in Belgium are well aware of the impact preference vote-earning has on the selection process: "Preference votes are super important for the next nomination process," one Member of Parliament told us, "at that time the preference vote tallies of the previous election are summoned and you are compared to the other candidates in the district." "As to your list position, preference votes are more important even than the work you do," another added. But, ultimately, we have few direct testimonies as to the motivations - short-term or long-term - of in particular lower-ranked candidates who cannot realistically hope to win a seat in Parliament. Future studies will also have to include more finegrained data on candidate quality and candidate media exposure that were beyond the scope of the present study. In addition, the simple fact that flexible-list systems prove to generate incentives more akin to those generated by open-list systems paves the way for more systematic data collection on preference voting - on voters' use of the preference vote and its implications - covering wider parts of the globe. As even basic data on the intra-party dimension continue to be lacking in most countries (Shugart, 2005), a whole field of study of which the relevance has been consistently underappreciated awaits further exploration.
Acknowledgements
Roman Chytilek gratefully acknowledges the research assistance of Filip Brstak and Miroslav Nemcok. The authors would like to thank Yusaku Horiuchi for helpful comments on an earlier draft of the article and Stefaan Walgrave for use of the interviews conducted as part of the INFOPOL project, supported by the European Research Council (advanced grant no. 295735) and the Research Fund of the University of Antwerp (grant no. 26827).
Declaration of conflicting interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: Matthew S Shugart acknowledges the support of the US National Science Foundation (NSF SES-0452573). Roman Chytilek acknowledges the support of the Czech Science Foundation (Project GA13-20548 S).
Notes
1. Flexible-list systems are also referred to as weak preferential systems (Karvonen, 2004) or semi-open systems (see for instance Maddens and Put, 2013).
2. Citizens have levelled similar allegations in mixed-member systems against members gaining election on the basis of their position on the party ballot in the proportional tier, even after being defeated in their district. They are commonly considered to be 'second-class representatives' or even 'zombies' (see Karp, 2002; Lundberg, 2006; Pekkanen et al, 2006).
3. Austria, Denmark, Estonia, and Sweden all have second-tier adjustment seats, providing candidates with complex electoral incentives on the basis of their probability to gain election in either of the two tiers.
4. Although the Netherlands elects members to the Parliament using a nationwide constituency, parties can and do nominate different candidates across the 19 sub-districts. For this reason the Netherlands is not 'simple' in its flexible-list aspect.
5. District magnitudes are based on the population census in Belgium but on turnout in the Czech Republic, mirroring as such differences in turnout between the districts (Voda and Pink, 2010).
6. That is, in the Czech Republic it occasionally happens that a list that would have sufficient votes in some district to win a seat is excluded for failing to clear the nationwide threshold.
7. In some other flexible-list systems, for instance in the Netherlands, voters may be required to give a preference vote - as is the case in some open-list systems like Finland and Poland.
8. Parties in Belgium field as many candidates as there are seats to be allocated in the district and half again as many substitute candidates (with a minimum of six). Members who resign mid-term (due to ill health, election to the regional parliaments, or promotion to the cabinet) are replaced by the first ranked substitute candidate, involving a similar distribution of the list votes across individual substitute candidates.
9. 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. Our estimations are based on the mean of the two, reflecting the increased total number of preference votes cast in the 2010 election -in part due to the increase in the number of preference votes each voter can indicate. The total number of preference votes marked, across all voters, has increased by about 170 per cent between the 2006 and 2010 elections (Stegmaier et al., 2014).
10. To be comprehensive, the list votes transferred to the highest-ranked effective candidates also comprise (half of) the ballots cast for substitute candidates only.
11. In 2010 early elections were called after the Liberal Party withdrew its support for the government. Elections are scheduled not more than 40 days after the dissolution of Parliament. Candidate lists in turn have to be submitted at least 28 days before the elections, leaving party selectorates little time to find new candidates willing to run.
André et al.
599
References
Adams J and Merrill SM (2003) Voter turnout and candidate strategies in American elections. Journal of Politics 65(1): 161-189.
Andeweg RB (2005) The Netherlands: The sanctity of proportionality. In: Gallagher M and Mitchell P (eds) The Politics of Electoral Systems. Oxford: Oxford University Press, pp. 491-510.
André A and Depauw S (2013) District magnitude and home styles of representation in European democracies. West European Politics 36(5): 986-1006.
André A, Depauw S and Martin S (2015) Electoral systems and legislators' constituency effort: The mediating effect of electoral vulnerability. Comparative Political Studies 48(4): 464-496.
André A, Wauters B and Pilet JB (2012) It's not only about lists: Explaining preference voting in Belgium. Journal of Elections, Public Opinion & Parties 22(3): 293-313.
Beblavý M and Veselkova M (2014) Preferential voting and the party-electorate relationship in Slovakia. Party Politics 20(4): 521-532.
Bengtsson A, Hansen KM, Hardarson ÓT, et al. (2014) The Nordic Voter: Myths of Exceptionalism. Colchester: ECPR Press.
Bergman T and Bolin N (2011) Swedish democracy: Crumbling political parties, a feeble riksdag, and technocratic power holders? In: Bergman T and Strom K (eds) The Madisonian Turn. Ann Arbor, MI: University of Michigan Press, pp. 251-296.
Bergman ME, Shugart MS and Watt KA (2013) Patterns of intra-party competition in open-list & SNTV systems. Electoral Studies 32(2): 321-333.
Blais A and Massicotte L (2002) Electoral systems. In: LeDuc L, Niemi RG and Norris P Comparing Democracies 2: New Challenges in the Study of Elections and Voting. London: Sage, pp. 40-69.
Bowler S and Farrell DM (2011) Electoral institutions and campaigning in comparative perspective: Electioneering in European parliament elections. European Journal of Political Research 50(5): 668-688.
Bráuninger T, Brunner M and Dáubler T (2012) Personal vote-seeking in flexible list systems: How electoral incentives shape Belgian MPs' bill initiation behaviour. European Journal of Political Research 51(5): 607-645.
Carey JM (2009) Legislative Voting and Accountability. Cambridge: Cambridge University Press.
Carey JM and Shugart MS (1995) Incentives to cultivate a personal vote: A rank ordering of electoral formulas. Electoral Studies 14(4): 417^39.
Crisp BF, Escobar-Lemmon MC, Jones BS, et al. (2004) Vote-seeking incentives and legislative representation in six presidential democracies. The Journal of Politics 66(3): 823-846.
Crisp BF, Olivella S, Malecki M, et al. (2013) Vote-earning strategies in flexible list systems: Seats at the price of unity. Electoral Studies 32(4): 658-669.
De Winter L (1988) Belgium: Democracy or oligarchy? In: Gallagher M and Marsh M (eds) Candidate Selection in Comparative Perspective: The Secret Garden of Politics. London: Sage Publications, pp. 20^6.
De Winter L (2005) Belgium: Empowering voters or party elites? In: Gallagher M and Mitchell P (eds) The Politics of Electoral Systems. Oxford: Oxford University Press, pp. 417-432.
Gallagher M and Marsh M (1988) Candidate Selection in Comparative Perspective: The Secret Garden of Politics. London: Sage Publications.
Hazan RY and Rahat G (2010) Democracy Within Parties: Candidate Selection Methods and Their Political Consequences. Oxford: Oxford University Press.
Heitshusen V, Young G and Wood DM (2005) Electoral context and MP constituency focus in Australia, Canada, Ireland, New Zealand, and the United Kingdom. American Journal of Political Science 49(1): 32-45.
Jacobson GC and Kernell S (1981) Strategy and Choice in Congressional Elections. New Haven, CT: Yale University Press.
Karp JA (2002) Members of Parliament and representation. In: Vowles J, Aimer P, Karp JA, et al. (eds) Proportional Representation on Trial. Auckland: Auckland University Press, pp. 130-146.
Karvonen L (2004) Preferential voting: Incidence and effects. International Political Science Review 25(2): 203-226.
Karvonen L (2011) Preferential vote in party list. In: Colomer JM (ed.) Personal Representation: The Neglected Dimension of Electoral Systems. Colchester: ECPR Press, pp. 119-134.
Katz RS (1986) Intraparty preference voting. In: Grofman B and Lijphart A (eds) Electoral Laws and their Political Consequences. New York: Agathon Press, pp. 85-103.
Katz RS (2001) The problem of candidate selection and models of party democracy. Party Politics 7(3): 277-296.
Kitschelt H (2000) Linkages between citizens and politicians in democratic polities. Comparative Political Studies 33(6-7): 845-879.
Lebeda T (2007) Electoral systems and voting behaviour: The case of the House of Deputies' elections in the Czech Republic. Contemporary European Studies 2(2): 25-48.
Lundberg TC (2006) Second-class representatives? Mixed-member proportional representation in Britain. Parliamentary Affairs 59(1): 60-77.
McDermott ML (2009) Voting for myself: Candidate and voter group associations over time. Electoral Studies 28(4): 606-614.
Maddens B and Put GJ (2013) Office effects and campaign spending in a semi-open list PR system: The Belgian/Flemish federal and regional elections 1999-2010. Electoral Studies 32(4): 852-863.
Marsh M (1985) The voters decide? Preferential voting in European list systems. European Journal of Political Research 13(4): 365-378.
Miiller WC (2005) Austria: A complex electoral system with subtle effects. In: Gallagher M and Mitchell P (eds) The Politics of Electoral Systems. Oxford: Oxford University Press, pp. 397^16.
600
Party Politics 23(5)
Norris P (1997) Passages to Power: Legislative Recruitment in Advanced Democracies. Cambridge: Cambridge University Press.
Pekkanen R, Nyblade B and Krauss ES (2006) Electoral Incentives in mixed-member systems: Party, posts, and zombie politicians in Japan. The American Political Science Review 100(2): 183-193.
Put GJ and Maddens B (2013) The selection of candidates for eligible positions on PR lists: The Belgian/Flemish federal elections 1999-2010. Journal of Elections, Public Opinion & Parties 23(1): 49-65.
Renwick A and Pilet JB (2015) Faces on the Ballot. The Personalization of Electoral Systems in Europe. Oxford: Oxford University Press.
Rustow D (1955) The Politics of Compromise. Princeton, NJ: Princeton University Press.
Shugart MS (2005) Comparative electoral systems research: The maturation of a field and the new challenges ahead. In: Gallagher M and Mitchell P (eds) The Politics of Electoral Systems. Oxford: Oxford University Press, pp. 25-55.
Sieberer U (2006) Party unity in parliamentary democracies: A comparative analysis. The Journal of Legislative Studies 12(2): 150-178.
Stegmaier M and Vlachova K (2011) The parliamentary election in the Czech Republic, May 2010. Electoral Studies 30(1): 238-241.
Stegmaier M, Tosun J and Vlachova K (2014) Women's parliamentary representation in the Czech Republic: Does preference voting matter? East European Politics & Societies 28(1): 187-204.
Stone WJ, Fulton SA, Maestas CD, et al. (2010) Incumbency reconsidered: Prospects, strategic retirement, and incumbent quality in U.S. House elections. The Journal of Politics 72(1): 178-190.
Taagepera R (2007) Predicting Party Sizes: The Logic of Simple Electoral Systems. Oxford: Oxford University Press.
Tavits M (2010) Effect of local ties on electoral success and parliamentary behavior: The case of Estonia. Party Politics 16(2): 215-235.
Van Holsteyn J and Andeweg RB (2010) Demoted leaders and exiled candidates: Disentangling party and person in the voter's mind. Electoral Studies 29(4): 628-635.
Van Raalte E (1959) The Parliament of the Kingdom of the Netherlands. London: Hansard Society.
Voda P and Pink M (2010) Candidates in parliamentary elections: Analysis of the preferential voting in the elections to the Chamber of Deputies of the parliament of the Czech Republic in 2006. World Political Science Review 6(1): 1-22.
Zittel T (2015) Constituency candidates in comparative perspective - How personalized are constituency campaigns, why, and does it matter? Electoral Studies 39: 286-294.
Author biographies
Audrey André is a Postdoctoral Researcher at the Vrije Univer-siteit Brüssel, Belgium. Her main research interests are in the fields of legislative and electoral studies, in particular the impact of electoral institutions on the demand- and supply-side of political representation.
Sam Depauw is an Assistant Professor and Postdoctoral researcher at the Vrije Universiteit Brüssel, Belgium. His research agenda focuses on political representation and party discipline.
Matthew S Shugart is a Professor of Political Science at the University of California, Davis, USA. He has published extensively on electoral systems and the constitutional design of executive-legislative relations.
Roman Chytilek is an Assistant Professor at the Department of Political Science and the International Institute of Political Science at the Masaryk University, Brno, Czech Republic. His research interests include elections, political parties, and experimental methods.