Introduction

How do parties respond to local electoral volatility in parliamentary elections? While election campaigns and results are becoming more and more nationalized (Caramani, 2004), political parties still need to deal with subnational variations in electoral support. A possible response to these variations is to change the within-district distribution of their candidates over subsequent elections. Parties that suffer electoral losses in certain local areas might be inclined to raise the number of locally rooted candidates from those areas on the candidate list for the next elections. The increased local presence of candidates could then lead to electoral recovery.

This argument is based on the strong theoretical claim that candidates with local roots (for example, place of residence, birthplace, holding local office) are electorally more successful in their own local areas compared with non-locally rooted competitors (Parker, 1983; Marsh, 1987; Blais et al, 2003; Shugart et al, 2005). If this is true, then political parties should be able to strategically manipulate their candidate lists in response to local electoral setbacks. But do parties indeed change the geographical distribution of candidates on their lists in order to increase electoral support in specific areas?

This article analyzes the effect of local electoral volatility on the geographical composition of party lists. We expect that changes in local party results will affect the share of local candidates in subsequent elections. In other words, parties adapt the within-district distribution of candidates on the basis of previous election results.

The within-district distribution of candidates is the outcome of intraparty candidate selection processes. Previous research has shown that these processes can vary both over time and parties, in terms of voting procedures, candidacy, centralization and inclusiveness of the selectorate (Bille, 2001; Lundell, 2004; Hazan and Rahat, 2010). Arguably, the parties’ ability to manipulate or change their candidate lists from one election to another strongly depends on the nature of their candidate selection processes. Intuitively, the level of territorial decentralization in candidate selection is expected to exhibit the strongest effect, but recently scholars also debated on the effect of selectorate inclusiveness on the composition of candidate lists. Therefore we analyze which level of decentralization and inclusiveness will provide parties with the greatest flexibility to manipulate the geographical composition of their candidate lists. Data on the composition of candidate lists and election results of eight Belgian political parties for seven consecutive Lower House elections (1987–2010) are used to answer the research question addressed in this article.

Party Candidates and the Electoral Value of their Local Roots

Both the field of political science and electoral geography have extensively dealt with the effects of local candidate roots in elections. A strong consensus exists that localism plays an important role in electoral politics, and that the local roots of party candidates should be considered important determinants of voting behaviour and candidate selection outcomes. However, the two fields have developed different approaches and operationalizations of the local roots concept, and identified different mechanisms to explain its importance in political recruitment and elections.

In political science literature, it has been firmly established that local candidates are valuable electoral assets in various types of electoral systems. Shugart et al (2005) argue that in electoral systems with personal vote-seeking incentives, local roots function as informational shortcuts for voters to simplify their voting decisions, resulting in a higher number of preference votes for these candidates (Tavits, 2009, 2010). Other authors claim that the sensitivity of party selectorates to local candidates is not restricted to this particular group of candidate-centred electoral systems. In their seminal work on candidate selection, Gallagher and Marsh (1988) stated that local roots are invariably sought by party selectorates. Indeed, even in party-centred systems, party selectorates make an effort to recruit local candidates to campaign in different regions and legislators are oriented toward their local interests (Hazan, 1997, 1999; Latner and McGann, 2005). In these systems, parties consider local candidates and their local networks as valuable local campaign resources that they aim to distribute in the way that maximizes the total party vote (Latner and McGann, 2005, p. 713). So although the nature of the campaign might be different in candidate-centred vis-à-vis party-centred electoral systems, the literature states that local candidates are always considered to improve the party result in local areas.

In the political geography literature, there is a much stronger emphasis on local roots operationalized as the place of residence of candidates. Many scholars have investigated the impact of local contexts on electoral behaviour, among which Key (1949) and Cox (1969), who described the existence of the friends and neighbors effect. This effect explains why voters tend to support candidates living in, or close to their respective areas and has been replicated in various types of electoral systems (see, for example, Arzheimer and Evans, 2012; Górecki and Marsh, 2012). Putnam (1966) argues that locally dominant parties or candidates have strong organizations and will be able to outdo the competition in terms of campaign resources.

In addition, the ‘neighborhood effect’ stresses the importance of social distance between voters and candidates in electoral behaviour. Voters tend to have more knowledge of their local candidates because their communication networks are highly localized and they are more likely to be exposed to information about a nearby candidate. This is often summarized by the observation that ‘people who talked together voted together’ (Miller, 1978; Pattie and Johnston, 1999). The implication is that place of residence is a strong predictor of both voting behaviour by the electorate and electoral success for the party candidate, through the mechanisms of social interaction and personal contact (Górecki and Marsh, 2012, pp. 565–566).

These theories have far-reaching consequences for political parties in both majoritarian and proportional electoral systems: Parties should consider candidates with local roots as extremely valuable electoral assets (Bochel and Denver, 1983; Gallagher and Marsh, 1988; Childs and Cowley, 2011). This implies that parties will prefer to recruit candidates with some level of experience in local politics, for instance through holding a local political office. The more local officeholders on the party list, the better the list will perform.

Concerning the place of residence of party candidates, the story gets more complicated for the parties. To enjoy the benefits of locally rooted candidates all over the electoral district, parties need to geographically balance their lists with candidates from different areas of the district. As a result, a geographically diverse set of candidates is needed, but the nature of this exercise strongly depends on the electoral system. In majoritarian systems, where district magnitude usually equals one, parties will nominate the candidate with the broadest electoral appeal (Tremblay, 2012). In other words, parties nominate one candidate from one particular area in the district. In proportional electoral systems with higher district magnitude, parties nominate groups of candidates on candidate lists. This allows them to balance candidate characteristics in terms of gender, ethnicity, age, profession, but also their place of residence (Valdini, 2012).

The existence of friends and neighbors and neighborhood effects stimulates parties in proportional systems to nominate candidates from various areas within the electoral district. If parties only select candidates from the largest city of the district, they risk losing potential votes in other areas of the district because of a lack of candidates that are locally rooted in those areas. From this point of view, candidate positions on the party list could be considered as scarce resources, which should be carefully distributed within districts (Latner and McGann, 2005). In the next section, we elaborate on what strategies parties might develop in terms of the within-district distribution of their candidates.

Within-District Distribution of Party Candidates and Local Electoral Volatility

By distributing their candidates over the electoral district, parties enjoy the electoral benefits of local candidates all over the district. But, as already mentioned in the introduction, local variations in voter support might urge parties to reconsider the within-district distribution of candidates before subsequent elections. But how exactly will parties change the geographical distribution of candidates in response to within-district electoral volatility? We define two types of party behavior in this respect. The first type is called the expansion strategy, which defines parties that increase the presence of local candidates in areas where they lost votes in previous elections.

In this first strategy, parties respond to a local electoral setback in election t−1 by increasing the number of candidates with place of residence in that area for election t. The higher the electoral setback in the previous election, the larger the increase in candidate share on the list in the subsequent election. Parties following the expansion strategy will increase candidate shares in losing areas at the expense of winning areas. In sum, this type of party aims to increase its vote share by investing more local candidates in areas where they previously lost votes. Hence, for parties following this strategy, the changes in local party shares from election t−2 to election t−1 are negatively correlated with the changes in local candidate shares from election t−1 to election t, in the sense that local areas where the party vote has decreased, the number of candidates will consequently increase.

Arguably, this strategy requires a certain amount of coordination by the party elite or selectorate drafting the candidate lists. Candidate selection processes are highly complex puzzles where parties balance candidate characteristics, factional interests and strategic considerations. In order to increase candidate shares in response to losing votes in specific areas, the party elite needs to exert some level of control over candidate selection outcomes.

A second type of strategic behaviour is the consolidation strategy. Party selectorates following this strategy aim to defend areas where they increased their vote share over the last two elections. These parties increase the presence of candidates in areas with a positive election result: An increase in party vote share over the last two elections leads to an increase in candidate shares in those areas. In other words, the consolidation strategy implies that parties will increase candidate shares in winning areas at the expense of losing areas. These parties will essentially give up on areas where their electoral losses have been too large, and focus on the positive electoral swings during the previous elections. Hence, for parties following the consolidation strategy, the changes in local party shares from election t−2 to election t−1 are positively correlated with the changes in local candidate shares from election t−1 to election t.

Arguably, the consolidation strategy is a very natural strategy for political parties. It implies that parties mainly recruit candidates from areas where they have been electorally successful. This is a logical course of action: parties will always tend to select more candidates in electoral strongholds, since local party branches and political networks of parties have been more developed in those areas than elsewhere. In addition, local electoral successes might make politically motivated individuals more inclined to run as candidate in parliamentary elections. In local areas where the party is weak, it may have more difficulty finding local candidates (Carty et al, 2003). In other words, the causality between local candidate presence and party vote shares might also work the other way around: not only could candidates with local roots increase vote shares in their area, but high vote shares will also increase the recruitment of local candidates in those areas. In this sense, the consolidation strategy marks a self-reinforcing effect which parties will experience if they do not actively decide to shift to the expansion strategy.

These nomination strategies can be followed by parties in districted as well as non-districted electoral systems. In systems with a nationwide constituency (for example, Israel, The Netherlands, Slovakia), parties face these strategic considerations on a larger scale with longer candidate lists, but essentially the nature of this puzzle is similar to the process in every separate district of districted electoral systems. Here too, parties treat list positions as campaign resources that have to be wisely distributed over the electoral district, which coincides with the nation’s border (Latner and McGann, 2005). In districted electoral systems, the same exercise takes place in each of the separate electoral districts. The next question is which type of selectorate would be best fit to complete this puzzle successfully.

The Effect of Candidate Selection Processes

In sum, parties can develop two types of party behavior with regard to the within-district distribution of candidates. The parties’ ability to shift between these behaviors, however, depends on the nature of their candidate selection process. If the (national) party elite strongly controls the selection of candidates, parties will be more inclined to apply the expansion strategy. If, on the other hand, the selection process involves several other intraparty stakeholders, it will be more difficult to shift toward the expansion strategy, and the party will more likely apply the consolidation strategy.

The selection of candidates for parliamentary elections is one of the core functions of political parties (Sartori, 1976). Therefore, it is not surprising that this topic has received considerable research attention. The analytical framework of Hazan and Rahat (2010) provides a useful set of variables for the comparative analysis of candidate selection processes. Their model disentangles four dimensions, among which inclusiveness of the selectorate and decentralization are the most important. The selectorate, on the one hand, is the body that selects the candidates, and can be composed of only one person, or several people, up to the entire electorate of the nation. This dimension can be measured on a continuum from exclusive selectorates, where a very limited group of selectors take control, to inclusive selectorates, such as the party members or the electorate. Decentralization, on the other hand, measures the extent to which local party branches nominate party candidates. In highly centralized methods, the national party level has complete control over the nomination process.

Scholars have argued that high inclusiveness in candidate selection has detrimental consequences for party management: candidates and MPs depend on party members and not party elites for re-nomination on candidate lists. First, this might lead to decreased party cohesion, which threatens government stability and functioning of democracy in general (Hazan and Rahat, 2006). Second, since inclusive candidate selection form a complex coordination game, the resulting candidate lists are often not very representatively composed. For instance, Rahat et al (2008) find that selection methods with membership ballots are less likely to produce candidate lists with women on safe positions.

Highly exclusive methods, however, do lead to more representative sets of candidates. Selectors in exclusive party committees are more inclined to base their decisions on evaluations of the collective party good. In other words, these party committees are asked to construct a list of candidates that they think has the best chance of maximizing the electoral result, and keeping the intraparty turmoil to a minimum.

The argument is largely similar from the perspective of within-district distribution of candidates: in the case of exclusive candidate selection methods, the party elite has more flexibility to control candidate selection outcomes, and thus the geographical composition of candidate lists. This is not the case with inclusive candidate selection methods, where the number of stakeholders in the candidate selection process is too large to develop a clear response to local electoral volatility. In line with this argument, we formulate the following hypothesis:

Hypothesis 1

  • Parties with exclusive candidate selection methods are more likely to follow the expansion strategy than parties with inclusive candidate selection methods.

Figure 1 summarizes the expected relationship between the inclusiveness of the selectorate and the within-district distribution strategy of political parties. While exclusively organized candidate selections will respond to negative electoral swings with increases in candidate shares, the more inclusively organized parties will rather decrease their candidate shares under these circumstances.

Figure 1
figure 1

Hypothesized relationship between inclusiveness of the selectorate and within-district distribution strategy.

Full size image

The second candidate selection dimension, territorial decentralization, is intuitively the most relevant aspect of selection processes when analyzing the geographical composition and distribution of candidate lists. Still, the effect of the centralization dimension of candidate selection is not that straightforward. Arguably, to develop a clear strategy with regard to within-district candidate distribution, the party should at least have a bird’s eye perspective over the electoral district: any involvement of the sub-district-level party branches would complicate this exercise. More specifically, if local (municipal) party branches are involved in candidate selection, these branches will not be inclined to lower their share of candidates to the advantage of other local branches, regardless of the electoral result in previous elections.

So any level of decentralization involving sub-district party branches would be problematic for party elites inclined to follow the expansion strategy. If candidate selections are taking place at the district level, the party at least has an overview of the entire electoral district and is able to respond properly to local electoral volatility. But arguably, centralized candidate selection methods are even more preferable: centralized parties can develop a clear strategy both within and between districts in response to local electoral volatility. National party elites would be able to impose the expansion strategy within all electoral districts. If, on the contrary, every district party decides autonomously, strategies will strongly diverge. For instance, some district parties might be more inhibited to take into account the member rates of the municipal party organizations while distributing candidates, whereas a national party selectorate selects candidates from a more remote distance to the local party branches and subdistrict party organizations. This distance allows them to coordinate and commit to the expansion strategy. Hence, the second hypothesis of this article:

Hypothesis 2

  • Parties with centralized candidate selection methods are more likely to follow the expansion strategy than parties with decentralized candidate selection methods.

Figure 2 visualizes the theoretical logic of this second hypothesis. Centralized candidate selection methods will respond to local electoral setbacks with increases in local candidate shares, at the expense of areas where their vote shares increased.

Figure 2
figure 2

Hypothesized relationship between centralization and within-district distribution strategy.

Full size image

Case Selection, Data and Methods

Being a multilevel democracy with separate regional party systems, Belgium is a country case with complex territorial dynamics. Moreover, its proportional representation electoral system for parliamentary elections works with multimember districts. Contrary to single-member districts, electoral competition in multimember districts is fought between multiple parties with often long lists of candidates. Multimember districts thus allow parties to distribute their candidates over the various parts of the electoral district.

The Belgian electoral system is a flexible-list system: Voters can support the party list as a whole or cast preference vote(s) for candidates on party lists (Shugart, 2005). While the allocation of seats is determined by both list order and preference votes, the threshold of preference votes to overcome list order is infrequently reached. These flexible-list systems therefore more strongly resemble closed- than open-list systems (De Winter, 2005). The importance of rank order implies that not every candidate has the same perspective on winning a seat. The highly ranked candidates have considerably more chances of getting elected to parliament than candidates placed in the lower positions on the party list. This is not only caused by the mechanical effects of the electoral system, but also because of the higher campaign visibility of highly ranked candidates and the voters’ propensity to vote for them (Miller and Krosnick, 1998).

Previous research on preference voting in Belgium has demonstrated that the local roots of candidates are important determinants of individual-level and list-level election results (Maddens and Put, 2011; Put and Maddens, 2015). Moreover, André et al (2012) find that the Belgian voters’ propensity to cast a preference vote is partly related to their familiarity with candidates, for instance through contact over local casework. Recently, André et al (2015) argued that candidates who earn a great deal of preference votes in their district are often rewarded with re-nomination and even better list positions in the next elections. In sum, Belgian preference voting literature suggests that the mechanisms underpinning the theoretical arguments in this article hold true in the Belgian case: local candidates are electorally valuable, and party selectorates take into account previous election results in their nomination strategies for subsequent elections.

Belgium is an interesting case for the analysis of parties’ nomination strategies for a number of reasons. First, it has an electoral system that strongly resembles the flexible-list and latent-list systems in other European countries (for example, Austria, Czech Republic, Estonia, The Netherlands), making the conclusions based on its case potentially generalizable to a broader set of country cases. Second, while the average district magnitude for the Belgian Lower House elections has gradually increased over the last decades, these elections were generally organized in medium-sized districts and not in extremely small or large electoral districts, which further strengthens the generalizability of the Belgian case. Third, as we will discuss later on, the strong variation in district magnitude and intraparty candidate selection processes makes Belgium a suitable case to test the abovementioned hypotheses.

While examining the within-district distribution of candidates on Belgian party lists, one should take into account these differences between types of candidates. In this article, candidates assigned to realistic list positions are double-weighted and candidates in unrealistic positions only counted one time.Footnote 1 The used dichotomy realistic–unrealistic list positions is similar to previous work (Put and Maddens, 2013). In practice, Belgian party selectorates often distinguish between realistic and non-realistic list positions with previous election results as benchmarks. For example, if a party won two seats in a district during the previous election, the upper three positions could be considered realistic: the first two positions led to a seat in the previous election, and the candidate on the third position stands a realistic chance of getting elected depending on her number of preference votes and/or a positive result by the party list.Footnote 2 We register the number of seats parties won in the district in the previous election, and consider that number of positions plus one at the top of the list as realistic.Footnote 3

The next crucial decision is the selection of a geographical level as unit of analysis. Up until this point, we have talked about within-district candidate distribution and its relationship with parties’ vote shares in local areas. But what is an appropriate operationalization of these ‘local areas’ in the Belgian case? As already mentioned, Belgium has a rather complex territorialization, and this is also noticeable at the subdistrict level, where there are municipalities, cantons and arrondissements.

First, the lowest geographical level are of course the municipalities, in which parties also have their lowest organizational units (that is, local party branches). While municipalities are a meaningful geographical level to analyze electoral trends, the fundamental problem is that there is no data available at the municipality level for Belgian parliamentary elections.Footnote 4

Second, electoral cantons are the lowest level for which there are data available on Belgian parliamentary elections. However, cantons are merely administrative units in which groups of municipalities coordinate and organize parliamentary elections. This level has no political or electoral meaning, and cantons are very unequal in size (in terms of population). Moreover, data aggregation at the canton level has shown that there is an insufficient number of candidates at this level for meaningful analysis.

Finally, the third sub-district geographical unit are the arrondissements. These have long formed the contours of the electoral districts for parliamentary elections in Belgium. Before the electoral district reform of 2002, all districts were arrondissements or mergers of arrondissements. Since 2002, Lower House elections are organized in electoral districts that coincide with the Belgian provinces. However, the arrondissement level still has its political relevance since various Belgian parties still organize themselves to some extent at the arrondissement level (Wauters et al, 2003). As a result, these arrondissements take on the role of ‘local areas’ in this article, and form the unit of analysis. It is important to stress that we do not take into account local areas that form an electoral district on their own. If an arrondissement forms a separate electoral district, parties are not able to distribute candidates over various local areas within the district, since there is only one local area.Footnote 5

We gathered data on party candidates from eight Belgian political parties, four in each language group. These parties were permanently represented in parliament during the seven consecutive elections we have data for (1987–2010).This gives us a total of 520 observations or local areas (that is, arrondissements) that will be analyzed in the empirical part of this article.Footnote 6 The two most important variables are the place of residence of Belgian election candidates and the party vote shares in the local area. We calculated the number of candidates per municipality for every election and party and aggregated these data to the local area level (that is, arrondissement). For every party list, we calculated the percentage of candidates from a local area in the district. This gives us the candidate share for every local area. To measure the change in candidate share (ΔCS), the dependent variable in this article, we calculate the following:

This is the relative change in candidate share between election t−1 and election t for every local area. Furthermore, the change in party share (ΔPS) is calculated according to the following formula:

The two hypotheses should be tested by means of a multivariate analysis where intra-party candidate selection methods and other control variables are taken into account. To test for the effect of candidate selection methods, we enter the level of decentralization and inclusiveness of the selectorate as independent variables to the analysis. Data on Belgian parties’ selection methods were collected on the basis of party regulations, news articles and existing literature on these parties.Footnote 7

The two candidate selection variables, selectorate and decentralization, are measured on ordinal 3-point scales.Footnote 8 First, the selectorate dimension has three categories (see Figure 3): (i) exclusive selection, where non-selected nomination committees dominate the selection process, without any form of member influence; (ii) The middle category defines selection procedures where party delegates form the selectorate; (iii) the most inclusive category are the selection procedures with member influence, for instance primaries and member polls.

Figure 3
figure 3

Scale of inclusiveness of the selectorate.

Full size image

The second dimension is decentralization, which is also divided in three categories (see Figure 4): (i) the highest level of centralization is reached if the national party level has a strong grip on the selection process, for example, by practicing veto power or through modification of candidate lists in the final phase of the process; (ii) the second category are candidate lists that are the outcomes of interaction between the national and district level; and (iii) the most decentralized methods are those where the district party level (or lower) autonomously drafts the lists without interventions by the national party level.

Figure 4
figure 4

Centralization scale.

Full size image

The multivariate analysis in the next section also takes into account the district magnitude, measured as the number of seats to be distributed in the electoral district. District magnitude is strongly related to the number of candidates on the lists and thus the level of flexibility that party selectorates have in changing candidate shares within districts. During the period of investigation, Belgium has experienced strong variations in district magnitude as a result of two electoral district reforms. Figure 5 shows the three phases of electoral districts between 1987 and 2010. In the 1987 and 1991 Lower House elections, the electoral districts were fairly small and averaged 7.1. Afterwards, in the 1995 and 1999 elections, some of the smaller districts were merged that led to a modest increase in average district magnitude (7.5). Finally, since the second electoral district reform of 2002, the Lower House elections take place in fairly large multimember districts (average DM=13.6).Footnote 9 In each of the three maps depicted in Figure 5, the bold lines represent the electoral district borders and the thinner lines show the borders of arrondissements within districts. While in the first phase the majority of electoral districts coincided with one single arrondissement, the two subsequent electoral district reforms increased district sizes and with it the number of districts consisting of multiple arrondissements. In the third and final electoral district phase, only two electoral districts (that is, Leuven and Walloon Brabant) coincide with one arrondissement.

Figure 5
figure 5

The distribution of arrondissements over electoral districts (upper left: 1987–1991; upper right: 1995–1999; lower: 2003–2010).

Full size image

As mentioned earlier, the between- and within-election variation in district magnitude makes Belgium all the more interesting as an empirical case to analyze determinants of within-district candidate distribution. We also take into account the size of the arrondissements by including their relative population size in the district as control variable. Besides adding DM as independent variable, we add dummies for election years to control for changes in electoral laws that might have an effect on candidate selection processes and parties’ geographical strategies (for example, district reforms as depicted in Figure 5, gender quota laws, changes to the weight of preference vote). For instance, the first election after a district reform might show considerably more fluctuations in candidate shares, as parties still need to adapt their strategies to the newly reshaped districts.

Analysis

How do intraparty candidate selection processes affect the within-district distribution of candidates? In the theoretical section, the argument was made that exclusive and centralized candidate selection processes allow more flexibility to apply an expansion strategy. Inclusive and decentralized processes, on the other hand, will push parties toward the consolidation strategy, where local candidate shares are positively related to local vote shares. But what is the variation in candidate selection processes in the data sample, which deals with the lower house elections between 1987 and 2010?

Figure 6 shows the distribution of 520 observations over the different categories of the centralization and selectorate dimension. With regard to centralization, only in a minority of cases does the national level completely control the selection process. In 70 per cent of the observations in our data set, the party district level at least has some influence on the composition of the candidate list. Put differently, the majority of local areas in our data set were part of an electoral district where candidate selection was highly decentralized. However, differences between the categories are negligibly small as in 34 per cent of the cases, candidate lists were drafted through cooperation between the district and national party level. Finally, 30 per cent of local areas were part of an electoral district, where the national party level completely controlled the outcome of the process.

Figure 6
figure 6

Relative frequency of every category of selectorate (left) and decentralization (right).

Full size image

The second candidate selection dimension, inclusiveness of the selectorate, tells a different story. In this dimension, observations are not as equally distributed over the three categories: no less than 61 per cent of the observations involved members in the candidate selection process. Party scholars indeed found that the practice of primaries and member polls became popular again during the 1990s in Belgian politics (Deschouwer, 1994; Fiers and Pilet, 2006; Verleden, 2013). During the 1970s and 1980s, the majority of Belgian parties applied more exclusive procedures (De Winter, 1988). The system with delegates (28 per cent) and especially the nomination committees (11 per cent) have been more and more out of use: while selection by delegates have been applied by both the Flemish and Francophone socialist parties during the period under investigation, nomination committees were almost exclusively used by the Francophone liberal party.

How do these candidate selection dimensions affect the within-district distribution of candidates, measured by ΔCS? We model this by means of an OLS regression analysis. The key independent variables in the model, selectorate inclusiveness, decentralization and electoral swing, are all included as categorical variables. The main theoretical argument presents an interactive logic between political parties’ vote swings, selection processes and changes in local candidates. We argue that testing this interactive logic should be done by estimating separate coefficients for each of the possible ‘scenarios’ depicted in Figure 1 and Figure 2. As a result, the continuous variable ΔPS is added as a categorical variable, disentangling observations where parties lost and won votes.Footnote 10

Table 1 presents the results of the OLS regression model with changes in candidate shares in the arrondissement (ΔCS) as dependent variable. Positive regression coefficients will point to increases in candidate shares. The first column presents the estimated coefficients, the second column shows the standard errors. Apart from the dummy variables controlling for the election years, only three coefficients are significant at the 0.10 level. First, district magnitude is negatively associated with changes in local candidate shares. More precisely, the results suggest that an increase in district magnitude significantly decreases ΔCS. Put differently, arrondissements that are part of larger districts will in general be less likely to increase candidate shares, but instead maintain status quo or even decrease the number of local candidates.

Table 1 OLS regression model with changes in candidate share as dependent variable
Full size table

Second, the coefficient for ‘negative swing’ is also significant, and has a theoretically more meaningful interpretation than the effect of district magnitude. It seems that parties who suffered an electoral loss in an arrondissement are significantly more likely to increase the number of local candidates in that area. The significance of this main effect supports the expansion strategy: parties respond to local electoral losses by increasing their candidate shares in those areas.

Third, moving on to the interaction effects, only one is statistically significant at the 0.10 level:Footnote 11 the interactive term between a negative electoral swing and selection processes where the national level has full control. In other words, parties with centralized candidate selection processes are more likely to respond to local electoral losses with decreases in candidate shares. This is at odds with the second hypothesis, which stated that parties with centralized candidate selection would rather pursue the expansion strategy. On the contrary, parties even lower the number of local candidates on the lists in areas that have underperformed over the most recent elections, which points to the consolidation strategy. So instead of supporting local areas where the local party has suffered an electoral defeat, centralized parties seem to be inclined to ‘punish’ local party branches and invest more list slots and local candidates in other areas of the electoral district.

With regard to the two candidate selection variables, selectorate and decentralization, we find that the main effects of both dimensions do not significantly affect changes in candidate shares. In sum, the results show only weak support to the idea that the applied strategies in response to local electoral volatility is dependent on the nature of intra-party candidate selection methods. Only centralization has a significant interaction term, but the sign of the coefficient is at odds with the earlier mentioned hypotheses. Concerning the inclusiveness of the selectorate, however, there are no conclusive findings to report: the interaction terms of changes in party share and selectorate do not contribute to the explanatory power of the model (Adjusted R 2=0.12).

Conclusion

This article has examined the effects of changes in local vote shares on the within-district distribution of party candidates. By studying the geographical composition of party lists and election results for eight Belgian parties in seven Lower House elections, we analyzed whether there is a clear party strategy in responding to local electoral volatility using the geographical composition of party lists. In addition, we examined whether these responses are related to particular types of candidate selection procedures.

The multivariate analysis has produced only little substantial results. The results have shown that, in general, parties that experienced a local electoral setback respond with increases in their local candidate shares, which supports the logic of the expansion strategy. Moreover, only one of the many interaction terms between electoral swing and candidate selection procedures reached significance: if parties decrease their vote shares in local areas and have centralized candidate selection methods, they are significantly more inclined to decrease candidate shares in those areas. This provides partial support for the consolidation strategy, where parties decrease candidate shares in electorally successful areas to increase candidates in losing areas.

It appeared very difficult to pinpoint a more general relationship between local electoral volatility, changes in list composition and candidate selection procedures. There are several explanations for these difficulties. First, for some parties it might be the case that there simply is no relationship here: some parties do not perceive list positions as scarce campaign resources, but rather focus on campaign spending efforts to respond to local vote fluctuations. The geography criterion might not be equally important for all parties, resulting into geographically unbalanced candidate lists. Second, this contribution only measured the quantity of candidates in local areas. Arguably, by including the quality of local candidates as well, we might get a different picture of parties’ selection strategies. Local candidate quality could, for instance, be operationalized as holding local office, high local campaign spending efforts or being active in local civil society organizations. For reasons of data availability, the quality of local candidates operationalized in any of these forms could not be included in our analysis.

Third, while some parties might take previous election results into account during candidate selection processes, their opposing strategies blur the image of a general relationship. In other words, it might be interesting to work with hierarchical models, where the effect of changes in vote shares varies over the different parties. That way we would also be able to treat the candidate selection variables as Level-2 predictors that vary over time and parties, but not within parties during an election. In this study, the number of observations and parties were too limited for estimating such hierarchical models. More data on the composition of candidate lists for a larger set of political parties is needed to expand the analysis and apply these more advanced estimation techniques.

Fourth, even if parties think about applying these strategies, they might not succeed because of the complex nature of intra-party candidate selection methods. More precisely, candidate lists could be considered the outcome of intraparty struggles between different factions, constrained by the rules of the candidate selection process (that is, centralization, selectorate, candidacy and voting procedures). If candidate selection is indeed primarily a coordination game where party elites balance the interests of their party factions, there is of course little room for strategic maneuvers such as the ones analyzed in this contribution.