FEMALE REPRESENTATION IN OLPR

There is strong evidence that proportional representation systems are more beneficial for the election of women than plurality/majority systems (Kittelson and Schwindt-Bayer Reference Kittelson and Schwindt-Bayer2012; Matland Reference Matland1998), particularly in the long term (Thames Reference Thames2017). However, it is still being discussed whether OLPR systems prevail over closed-list systems in this respect (Schmidt Reference Schmidt2008). Thames and Williams (Reference Thames and Williams2010) argue that the ability of voters to change ranks might undermine women's representation, since female candidates usually do not perform well in systems with personal vote incentives.

Candidate selection under OLPR systems is more complex than under closed-list proportional representation or first-past-the-post systems. According to the well-known Carey and Shugart (Reference Carey and Shugart1995) classification, OLPR systems generate more incentives for candidates to cultivate a personal vote. Under OLPR, these incentives are stronger with increasing district magnitude, as the number of copartisans against whom each candidate is competing increases (Shugart, Valdini, and Suominen Reference Shugart, Valdini and Suominen2005). These personal features of candidates are thus taken into account by party leaders nominating candidates in order to maximize the party electoral result. In many OLPR settings, including Poland, party leaders also determine the ballot order of candidates, which affects candidates’ electoral chances because of the existence of strong ballot position effects (e.g., Blom-Hansen et al. Reference Blom-Hansen, Elklit, Serritzlew and Villadsen2016; Marcinkiewicz Reference Marcinkiewicz2014; Miller and Krosnick Reference Miller and Krosnick1998).

Nonetheless, the institution of an “open list” allows voters to modify the ranking proposed by parties with the use of preference votes. Their distribution is used to determine which candidates receive the seat won by a given list. The research on intraparty competition demonstrates that preference votes matter, particularly if they are mandatory and party magnitudes are high (Gendźwiłł and Raciborski Reference Gendźwiłł and Raciborski2014; Gendźwiłł and Marcinkiewicz Reference Gendźwiłł and Marcinkiewicz2017). Therefore, studies of elections held under OLPR rules usually deal with two stages of electoral choice: party choice and voter choice. They are represented by ballot ranking and preference votes distribution.

Ranking the candidates can be considered a final step in the selection process, which is dependent on the supply of candidates willing to stand for office and the demand of party gatekeepers (Lovenduski Reference Lovenduski2016; Norris and Lovenduski Reference Norris and Lovenduski1995). Kenny and Verge (Reference Kenny and Verge2016) argue that both candidate supply and demand are structured by the wider social and political framework and produce gendered outcomes: the underrepresentation of women as aspirants, candidates, and elected politicians. The existing research demonstrates that party elites, who are predominantly male, tend to exclude women from the distribution of rewards because of stereotypes and as a result of the out-group effect (Niven Reference Niven1998). Informal selection criteria usually favor men over women (Verge Reference Verge2015); even highly qualified women are less likely to be recruited to run for office (Fox and Lawless Reference Fox and Lawless2010). Even if women are selected, parties tend to place them in unwinnable positions on party lists (Luhiste Reference Luhiste2015).

The findings concerning the effect of gender on preference voting patterns are mixed. While there is evidence that gender-based voting negatively affects women (Giger et al. Reference Giger, Holli, Lefkofridi and Wass2014; Holli and Wass Reference Holli and Wass2010; Kukołowicz Reference Kukołowicz2013), once the analyses take into account candidate political resources and ballot placement, the dominant conclusion is that voters tend to marginally favor female candidates (Black and Erickson Reference Black and Erickson2003; Kunovich Reference Kunovich2012; Siemieńska Reference Siemieńska2005; Stegmaier, Tosun, and Vlachová Reference Stegmaier, Tosun and Vlachová2014). More precisely, the models controlling for ballot placement do not verify whether voters generally prefer women over men but whether the potential imbalance in their preferences systematically corrects the parties’ bias, already inscribed in the ballot placement pattern. Matland (Reference Matland, Ballington and Karam2005, 105) even argues that “women do better with voters than they do with the party committees putting together party lists, that is, the preferential vote leads to a greater representation of women.” Several analyses demonstrate that there is, on average, no significant gender bias in voters’ choices (Jankowski and Marcinkiewicz Reference Jankowski and Marcinkiewicz2017; McElroy and Marsh Reference McElroy and Marsh2010). It is clear, however, that voters’ choices under OLPR most frequently follow the pattern of ballot ranking provided by the parties; voters to a large extent prefer the same candidates as party elites.

THE EFFECTS OF LEGISLATIVE GENDER QUOTAS IN OLPR SYSTEMS

There is evidence that gender quotas are more efficient in closed-list systems than in open-list systems (Jones and Navia Reference Jones and Navia1999), particularly if the “zipper rule” (i.e., alternating men and women candidates at the top of the list) is applied (Millard Reference Millard2014). The existing literature demonstrates that the outcomes of gender quotas depend on the quota size, additional placement rules concerning top ballot positions, and sanctions for noncompliance (Paxton and Hughes Reference Paxton and Hughes2015; Schwindt-Bayer Reference Schwindt-Bayer2009). There is also evidence that left-wing parties are more likely to implement quotas (Caul Reference Caul1999; Millard Reference Millard2004).

As the quota threshold is usually set at a level higher than the average share of female candidates in previous elections, the most immediate, direct effect of quotas is an increase in the proportion of female candidates on the lists. However, the share of women among elected representatives usually increases much slower, as it is an outcome of a nonrandom selection.

It is obvious that the significant change in proportion between men and women directly affects women's chances of being elected: the increasing share of female candidates in comparison with pre-quota elections should proportionally increase their electoral chances if no other factors influence the selection process. Nonetheless, as this inflow of female candidates is a result of an intervention, not a gradually progressing political empowerment of female candidates, it may be expected that the average “electoral capital” of female candidates will decrease after the introduction of quotas, while simultaneously the average electoral capital of male candidates will increase. Researchers frequently approximate this electoral capital, a favorable combination of personal vote-earning attributes, by incumbency and previous electoral experience in national or subnational elections, the factors that are taken into account by party leaders (Htun and Jones Reference Htun, Jones, Craske and Molyneux2002; Shugart, Valdini, and Suominen Reference Shugart, Valdini and Suominen2005).

Numerous studies demonstrate that the general pattern of incumbency advantage may in fact be a disadvantage to the election of women (Schwindt-Bayer Reference Schwindt-Bayer2005). In OLPR systems, incumbents, who are predominantly male, are more likely to receive more promising ballot positions and, consequently, more preference votes (Matland Reference Matland, Ballington and Karam2005). Party leaders usually argue that it is rational to put experienced and well-recognized candidates on the top of the ballot. Sineau (Reference Sineau, Ballington and Karam2005) discusses the case of 2002 French legislative elections, in which large parties preferred to pay the fines prescribed by the quota law rather than have more women candidates and “sacrifice” incumbents.

Several authors mention that quota regulations may create a stereotype that “quota women” are underqualified (Franceschet and Piscopo Reference Franceschet and Piscopo2008; Nanivadekar Reference Nanivadekar2006). Nonetheless, the decreasing average electoral capital of female candidates immediately after the introduction of quotas does not imply that there are fewer experienced female candidates in absolute terms. It usually means that this group simply becomes a smaller part of the significantly enlarged group of all female candidates. Analogously, a significant number of male candidates with lower electoral capital simply are not placed on the ballot.

It seems obvious that ballot ranking may be used to support or counteract quota regulations. The pattern in which fewer women are placed in favorable rather than distant positions can be explained—at least to a certain extent—by differences in the distribution of electoral capital between men and women. But the question remains open whether candidate ranking patterns are additionally biased against women. Party elites, dominated by men, may be interested in preserving the pre-quota status quo, or at least mitigating the immediate impact of the quota regulations.

To sum up, in our model assessing the consequences of the introduction of quotas, several interrelated factors should be taken into account (Figure 1).

Figure 1. Factors influencing female representation in an OLPR system and assumed changes after the introduction of legislative gender quotas.

We assume that the individual electoral chances of female candidates under an OLPR system—that is, the chances of both obtaining a viable ballot position and ultimately being elected—are conditioned mechanically by the share of female candidates on lists. However, along with the change in the share of female candidates, we can observe simultaneous changes (1) in the distribution of electoral capital among candidates of each gender; (2) in the parties’ gender bias, that is, systematic differences in ballot placement of women and men that are not explained by differences in their electoral capital or cannot be attributed to the parties’ residual preference for incumbents; and (3) in voters’ additional gender bias, that is, systematic preferences for candidates of a particular gender expressed in preference voting yet explained neither by their ballot placement nor by their level of electoral capital.

MAIN HYPOTHESES

In our analyses, we examine how parties (through a change in ballot ranking patterns) and voters (through a change in preference voting patterns) reacted to the introduction of quotas.Footnote ² We define the gender bias in ballot ranking as a significant effect of a candidate's gender on his or her individual chance of being placed in favorable ballot position. Analogously, the bias in preference voting relates to the impact of a candidate's gender on his or her individual chance of winning a seat. We ask whether a negative bias exists and, more importantly, whether it changed in the post-quota period.

We do not observe equal chances of women being nominated for viable positions, even if the main personal vote-earning attributes describing the electoral capital of candidates are controlled; we attribute this effect to gender bias in ballot ranking patterns (Dahlerup Reference Dahlerup2007, 84–85). Analysis of the literature leads us to the first hypothesis concerning the pre-quota and post-quota periods:

In fact, if such a formulation implies a control for incumbency, it becomes a very conservative estimation of gender bias in ballot ranking. We assume that gender bias, which is attributed to party leadership, is to a certain extent captured by incumbency status and previous electoral experience, which are not “pure” control variables but outcomes of previous electoral selection processes. In other words, if previous nominations have been already biased in favor of male candidates, they will probably affect subsequent nominations through incumbency status.

As we already discussed, the increased share of female candidates on the list should proportionally increase the share of women occupying viable list positions, once candidates of both genders have equal levels of electoral capital. However, after the introduction of quotas, politically valuable resources are still prevailingly at male candidates’ disposal. Once women become more numerous and men become less numerous among candidates, the introduction of quotas should decrease the individual chances of female candidates being placed high on party lists.

Nonetheless, we assume that ballot ranking patterns could be additionally influenced by bias against female candidates, if political parties counteract the quota regulations. If this is true, women's electoral chances will decrease after the introduction of quotas, not only because of the decrease in their average electoral capital but also because of the additional bias expressed in the ballot ranking.

We acknowledge that preference voting to a large extent resembles the pattern of ballot ranking, determined by party elites. For that reason, in the models explaining individual probabilities of receiving a seat, we control for ballot position effects. We also take into account electoral capital, as it can be important not only for the party establishment but also for the electorate. Once these factors are controlled, the remaining effect of a candidate's gender on his or her chance of receiving a seat can be attributed to the additional gender bias in preference voting patterns. Taking into account the dominant (yet not univocal) tone of existing studies, we assume that voters are positively discriminating in favor of women against men, hypothetically acting against the bias already expressed in the ballot rankings.

Once we attempt to assess the change after the introduction of legislative quotas, we seek to confirm the results presented by Górecki and Kukołowicz (Reference Górecki and Kukołowicz2014), which provide evidence in favor of a “paradoxical effect” of quotas on preference voting. This effect means, in brief, that once quota regulations introduce more women on lists, the preference votes of voters who clearly prefer female candidates are less numerous and more dispersed, which decreases women's chances of ultimately being elected:

We test these hypotheses empirically in the subsequent sections of the article. Additionally, we reflect on how the results estimated in the metrics of individual electoral chances could be expressed in the metrics of electoral outcomes.

GENDER QUOTAS IN POLISH SEJM ELECTIONS

In this study, we use the results of Sejm elections held in Poland under the OLPR system between 2005 and 2015, including two elections before and two elections after the introduction of gender quota regulations.

The presence of women in the Polish parliament slowly but systematically increased after the transformation. While in the first democratically elected Sejm (1991–93), there were only 9.6% women representatives, in the eighth term (2015–19), this share increased to 27%. Poland adopted a 35% legislative gender quota in January 2011, before the parliamentary elections held in autumn, yet voluntary party quotas were used before: in the 2001 elections, three parties (the Freedom Union, the Labour Union, and the Democratic Left Alliance) used 30% quotas for the first time (Dubrow Reference Dubrow2011; Fuszara Reference Fuszara2013; Gwiazda Reference Gwiazda2017). The first outcomes of quota regulation seemed disappointing: the difference between the shares of female legislators elected in 2007 and 2011 was rather small (an increase from 20% to 24%), yet one can notice a visible change in the share of female candidates (Figure 2).

Figure 2. Shares of female candidates and legislators in the Sejm, 2005–15. Source: National Electoral Commission, own calculations.

The first look at the candidate data reveals that the electoral capital at the disposal of female candidates was, on average, smaller than in case of their male counterparts (Table 1), which is in accordance with our initial expectations. Górecki and Kukołowicz (Reference Górecki and Kukołowicz2014) also argued that female candidates after the introduction of quotas in Poland were not only more numerous but also—on average—“lower-quality” candidates in comparison with their male counterparts. One can find substantial differences between parties in the candidates’ electoral capital, resulting mainly from the volatility of the party system and the emergence of new parties. However, more important for our reasoning are the intraparty differences between male and female candidates in this respect.

Table 1. Average electoral capital of male and female candidates in Polish Sejm elections, 2005–15

Source: National Electoral Commission, own calculations.

It is worth mentioning that the candidates’ electoral capital is not necessarily related to their professional competences—for example, Kunovich (Reference Kunovich2003), who examined party lists in Poland between 1989 and 1997, found that equally qualified women were placed lower than their male colleagues on most of the party lists.

The pooled data from Polish Sejm elections also demonstrate that the visible increase in women's share on the party lists after the introduction of quotas was not evenly distributed among the ballot positions—fewer women were placed in favorable than distant positions (Figure 3). (Fuszara Reference Fuszara2013; Jankowski and Marcinkiewicz Reference Jankowski and Marcinkiewicz2017; Kunovich Reference Kunovich2012). More precise analyses (e.g., Gwiazda Reference Gwiazda2017; Millard Reference Millard2014) reveal differences between parties in the placement of male and female candidates; the authors attribute these differences mainly to internal regulations. However, it can be assumed that other factors (such as party size, ideology, its newness and the dynamics of electoral support) may serve as alternative explanations. In fact, it is difficult to draw systematic conclusions about differences between parties based on single-country studies with low numbers of electoral cycles.

Figure 3. Shares of female candidates placed at different ballot positions, 2005–15. Source: National Electoral Commission, own calculations.

DATA AND METHOD

In our empirical analyses, we use the official electoral data obtained from the National Electoral Commission. The original data set contains information on 14,887 candidates, nested in lists, electoral districts, and years; in this study, we include only the lists that received at least one seat in a district (Table 2). The number of electoral districts was constant (41) across the period analyzed, as was the total number of seats (460). The average number of candidates per seat varied between 13.45 in 2007 and 23.17 in 2005. The district magnitudes in Sejm elections ranged from 7 to 20, with average value of 11.2.Footnote ³

Table 2. Sejm elections included in the empirical analysis

Source: National Electoral Commission, own calculations.

In order to analyze the candidates’ individual electoral chances, we employ binary logistic regression using two different dependent variables. In order to capture the gender bias in ballot ranking, we estimate the probabilities of being nominated for a viable position at the top of the list. In order to capture the gender bias in preference voting patterns, we estimate the probabilities of being elected.

The operationalization of a “viable position” in the Polish version of the OLPR system is not obvious. For example, Kunovich (Reference Kunovich2012) considered the top three positions as a simple definition of a prominent ballot placement, relating her approach to the 35% threshold of the candidate gender quota. On the other hand, Górecki and Kukołowicz (Reference Górecki and Kukołowicz2014, 71), followed by Jankowski and Marcinkiewicz (Reference Jankowski and Marcinkiewicz2017), related the definition of a viable position to party magnitude. They defined “viable” or “promising” positions as top-N positions on each list, where for parties winning seats in a given district, N equals twice the party magnitude. For parties not winning seats in a district, they proposed to consider the top two positions as viable.

We modify the latter approach and use a more restrictive definition of “viable position.” For the lists winning seats in a given district, the viable ballot positions are {1, … m}, where m is the party magnitude.Footnote ⁴ More importantly, too broad definition of a viable position can actually obscure the real placement strategies, which are unfavorable for female candidates, and thus could lead us to underestimate negative ballot ranking bias and, possibly, overestimate the positive influence of quotas on female candidates’ chances.

The main explanatory variable in our analyses is, obviously, the candidate's gender. In order to test the hypotheses concerning change after the introduction of quotas, we include three dummy variables coding the year when the election was held. It allows us to verify the hypotheses regarding the general effect of introducing quotas (i.e., conducting tests on linear combinations of respective model parameters). The interaction between post-quota year (2011, 2015) and female variables should demonstrate the indirect effects of gender quotas—the changes in either party nomination strategies or voters’ preferences.

In order to account for differences in the levels of electoral capital, we control for the features of candidates that, in light of previous studies, affect their electoral chances: age (and age squared, in order to allow for a nonlinear relationship), incumbency status, and previous electoral experience (status of a candidate, regardless of the ultimate result). We differentiate incumbency and candidate statuses in national (parliamentary) or subnational (municipal, county, regional) elections. We assume that these variables capture well the political experience and popular recognition of the candidate (Crisp et al. Reference Crisp, Olivella, Malecki and Sher2013); we may assume that they also approximate the candidate's status within the local party organization.Footnote ⁵ In light of the “career politician” model (Kunovich Reference Kunovich2003; Shabad and Slomczynski Reference Shabad and Slomczynski2002), these traits are important to understand “candidate value,” which affects parties’ ballot ranking decisions, as well as candidate recognition, which can additionally boost his or her popularity among voters.

Additionally, in the models estimating the chances of receiving a seat, we control for the well-known ballot position effects. Based on the results presented by Marcinkiewicz (Reference Marcinkiewicz2014), we distinguish the first- and last-position ballot order effects. In order to control for ballot order effect, we use standardized ballot position—by transforming the original ballot positions to the <0,1 > range, we account for differences in list lengths. In order to control for first- and last-position effects, we include relevant dummy variables.

All models include a control for party magnitude (the number of seats received by a party list in a district). Based on the previous studies, we assume that the size of a local party delegation, expressed by party magnitude in a given district, can serve as an important concurrent explanation of gender bias in ballot ranking strategies and voters’ choices (Matland Reference Matland1993).

Continuous variables (age, standardized ballot position, and party magnitude) are centered around their grand means. Because of the nested structure of the data (candidates nested in lists), robust standard errors and significance levels are reported with correction for clustering at the list level.

ELECTORAL CHANCES VERSUS ELECTORAL OUTCOMES: COMBINING TWO ANALYTICAL PERSPECTIVES

The first step of our empirical analysis clearly refers to the metrics of individual chances of receiving a viable ballot position or a seat. The models estimating average individual chances of female candidates being elected or nominated for a viable position allow us to control for personal vote-earning attributes. Yet such an approach poses analytical challenges once applied to study the effects of gender quotas, as candidate chances are heavily dependent on the supply of candidates (Jankowski and Marcinkiewicz Reference Jankowski and Marcinkiewicz2017). The latter is obviously affected by the quota regulation.

Many studies of legislative gender quotas analyzing candidate-level data use the metrics of preference vote counts or shares. The conventional approach in the literature is to use the candidate's list vote share (e.g., Jankowski and Marcinkiewicz Reference Jankowski and Marcinkiewicz2017) or the absolute number of preferential votes cast for a candidate (a count-dependent variable often requiring negative binomial regression; e.g., Górecki and Kukołowicz Reference Górecki and Kukołowicz2014). However, both variables are mechanically dependent on the list length.

One could argue that the ultimate goal of any quota legislation is expressed not in terms of individual electoral chances but in terms of electoral outcomes measured at the aggregate level. Quotas are introduced to increase the share of females in legislative assemblies, not the average electoral chances among particular groups of candidates.

In order to combine the analytical perspective focused on individual chances with the one focused on the share of candidates, we propose to recalculate counterfactually the estimates of individual chances and aggregate them in order to present the predictions concerning the share of female candidates on the lists and in assemblies. Our reasoning can be summarized in the following steps, which follow the estimation of individual chances of being placed in a viable list position and being elected:

1. 1. First, we use the parameters of a model describing the election preceding the introduction of quotas (2007) to compute the individual chances of candidates being nominated for a viable position and their chances of being elected.

2. 2. Given these predicted probabilities, we can compute the expected number of male and female candidates placed at viable positions (or winning a seat) by summing the predicted probabilities for each gender. Computing the predicted proportion of females is straightforward: the predicted number of females is divided by the sum of the predicted number of females and the predicted number of males. Such a computation applied to the same election described by the initial model used to estimate probabilities matches exactly the observed numbers of candidates nominated for viable positions (or winning a seat), irrespective of the set of predictors used in a model.

3. 3. We use the model referring to the pre-quota elections to compute the predicted probabilities of being nominated for a viable list position (or winning a seat) by candidates running in the post-quota election. Subsequently, we can determine the expected number of male and female candidates nominated for a viable position (or elected) if the mechanisms of selection (described by the model) have changed since the reference pre-quota election. Knowing these parameters, it is easy to compute respective female shares. However, the set of candidates for whom probabilities are recalculated differs from pre-quota election both in terms of candidates’ numbers and their electoral capital. For that reason, including different sets of predictors in the models makes a difference in terms of the counterfactual interpretation of the results (Table 3).

4. 4. Consequently, the differences between the effects computed using models with different sets of predictors can be used to estimate the importance in shaping the overall (net) effect of quotas of different factors: (1) change in number of female and male candidates, (2) change in average candidate electoral capital, (3) change in party bias, and (4) change in voters’ additional bias.

Table 3. Counterfactual interpretations of the recalculation from “individual chances” to “electoral outcomes” metrics