Since women’s suffrage, women constitute roughly 50 per cent of voters in all democracies. In parliaments, however, their representation is rarely proportional. In most democracies, becoming a candidate for elected office involves being selected by a political party (for example, Bochel and Denver Reference Bochel and Denver1983; Berz and Jankowski Reference Berz and Jankowski2022); meanwhile, becoming a viable candidate requires having access to the necessary resources to compete in an election (Hazan and Rahat Reference Hazan and Rahat2010). Because party elites select candidates for office and distribute resources that can make them competitive (for example, high placement on their candidate lists or sufficient funds to carry out campaigns), they are instrumental in determining the pool of (viable) women candidates from which voters can choose (for example, Norris and Lovenduski Reference Norris and Lovenduski1995; Kenny and Verge Reference Kenny and Verge2016; Luhiste Reference Luhiste2015; Murray Reference Murray2010). While candidate selection may provide party leaders with an opportunity to address women’s underrepresentation, they may not be willing to do so: strategic considerations about how to more effectively advance their parties’ goals of winning and performing in office may conflict with the goal of promoting greater gender equality in politics. We ask: do party elites consider women’s underrepresentation when deciding who to nominate?
When faced with the task of forming candidate lists for parliamentary races, party leaders’ preferences are often strategically shaped by two main goals: 1) winning seats for their parties and 2) building a list of people who, if elected, will represent their parties’ interests (Galasso and Nannicini Reference Galasso and Nannicini2015). Concerns about electoral competition and party loyalty may motivate party leaders to nominate incumbents – that is, people with experience in delivering party interests during and after elections (Gallagher and Marsh Reference Gallagher and Marsh1988; Put, Muyters and Maddens Reference Put, Muyters and Maddens2021; Piscopo et al. Reference Piscopo and Hinojosa2022; Carson, Engstrom and Roberts Reference Carson2007). Since, in most democracies, the overwhelming majority of parliamentarians are men, party selectors’Footnote 1 decision to pursue a strategy focused on rewarding incumbents may contribute to the maintenance of women’s underrepresentation. Women’s underrepresentation in party lists, however, may in itself pose a problem to parties’ competitive standing: voters may interpret men-dominated candidate lists as a signal of parties’ disregard for promoting gender equality and representing the interests of women voters (Rahat, Hazan and Katz Reference Rahat and Hazan2008; Rehmert Reference Rehmert2022; Weeks et al. Reference Weeks2023) – so party leaders may also have the incentive to address gender imbalances in their parties’ lists.
Since candidate selection often takes place behind closed doors (Gallagher and Marsh Reference Gallagher and Marsh1988; Kenny and Verge Reference Kenny and Verge2016) and (not) promoting gender equality in representation is a sensitive issue, disentangling how competing pressures affect party leaders’ predisposition to nominate women is a challenging task. Given these obstacles, to explore how candidate selection shapes women’s underrepresentation, many studies have turned to examining what is observable: not party leader’s preferences, per se, but the gender breakdown of candidate nominations. These studies have provided valuable insight into how party ideology (for example, Caul Reference Caul1999; Weeks and Allen Reference Weeks and Allen2023), the characteristics of selectors (for example, Cheng and Tavits Reference Cheng and Tavits2011; Niven Reference Niven1998; Wylie Reference Wylie2018), as well as formal (for example, Hinojosa Reference Hinojosa2012; Aldrich Reference Aldrich2020; Luhiste Reference Luhiste2015) and informal institutions (for example, Bjarnegård and Zetterberg Reference Bjarnegård and Zetterberg2019; Piscopo Reference Piscopo2016) contribute to women’s (under)representation in candidate lists. Yet, we still know little about party leaders’ preferences and, more specifically, their willingness to consider women’s underrepresentation during candidate selection processes.
We contribute to this debate by leveraging a conjoint survey experiment with party elites. Conjoint experiments are a fitting approach for exploring party elites’ preferences in candidate selection, as the method allows researchers to isolate the effect of variations in a single factor (for example, candidates’ gender), while also varying several other candidate characteristics (for example, age, education) – as would be the case in non-hypothetical, on-the-ground candidate selection processes. Containing responses from 1,389 party elites from nineteen nationally relevant parties in Austria, Germany, and Switzerland, this is, to the best of our knowledge, the most extensive study of party elites that employs a conjoint experiment and the first cross-country study of this type.
Because party elites can use aspirants’ characteristics to assess how likely they are to attract voters and represent their parties’ interests if elected, party leaders tend to favour certain aspirant traits (for example, Rehmert Reference Rehmert2022; Doherty, Dowling and Miller Reference Doherty and Dowling2019; Berz and Jankowski Reference Berz and Jankowski2022). Since the characteristics of their parties’ full lists may also affect their parties’ standing against rivals, party leaders are also likely to have preferences regarding their parties’ list composition as a whole. Specifically, because voters could interpret obvious gender imbalances in parties’ lists as a lack of effort to engage with women voters and prioritize gender equality, we anticipate that women’s underrepresentation in earlier candidate selections will likely influence party selectors’ choices between subsequent aspirants of different genders.
As such, and in contrast to other research that uses conjoint experiments to study candidate selection, we treat party elites’ choices between aspirants not as independent but as decisions taken in the context of broader candidate lists. Explicitly, when exposing respondents to pairs of aspirants, we vary not only information about individual-level aspirant characteristics but also about the share of women already incorporated into the party’s list. Embedding this randomized prime into the conjoint experiment thus allows us to examine how the degree of women’s underrepresentation in parties’ lists affects party elites’ likelihood of selecting women – and, consequently, of addressing women’s underrepresentation.
We find that women’s underrepresentation in candidate lists is a crucial factor shaping party elites’ decisions to select women: the worse women are signalled to be represented on the party list, the better the chances of women aspirants being selected. As our sub-group analyses indicate, awareness of women’s underrepresentation significantly impacts selectors for whom promoting gender equality may be a less crystallized priority (Caul Reference Caul1999; O’Brien Reference O’Brien2018): centrists and men. Being cued to consider women’s underrepresentation also reinforces preferences for women aspirants among those for whom gender equality may be a core value (left-wing selectors and men) but does not affect those for whom opposing gender equality may bring electoral advantages (right-wing elites). Reinforcing prior literature, we also find that addressing women’s underrepresentation does not compromise candidate ‘quality’ (Besley et al. Reference Besley, Folke, Persson and Rickne2017; Weeks and Baldez Reference Weeks and Baldez2015).
Our findings shed light on the potential role that signalling underrepresentation may have on party elites’ likelihood of selecting women aspirants. While civil society campaigns to promote greater diversity in politics have generally focused on motivating women to run for office or on encouraging voters to cast ballots for women candidates (Piscopo Reference Piscopo2018; Gatto, Russo, and Thome Reference Gatto, Russo and Thome2021), few have targeted party elites. Our study suggests that efforts focused on signalling to party elites the reputational costs of gender imbalances in their candidate lists may have positive consequences for promoting women’s political representation – particularly beyond the ‘usual suspects’ (Celis and Erzeel Reference Celis and Erzeel2015).
(Gendered) Candidate Selection
Voters can only cast ballots for candidates whose names appear on them. It is thus no surprise that, in seeking to understand the sources of women’s underrepresentation in elected office, gender scholars have often turned to the processes through which candidates are selected (for example, Norris and Lovenduski Reference Norris and Lovenduski1995; Kenny and Verge Reference Kenny and Verge2016; Murray Reference Murray2010). As prior literature shows, formal and informal institutions are important in shaping the outcomes of candidate selection (Luhiste Reference Luhiste2015; Bjarnegård and Zetterberg Reference Bjarnegård and Zetterberg2019; Gatto and Wylie 2022; Aldrich Reference Aldrich2020), but so are the people leading these processes: in many democracies, aspirants must first be selected by party leaders before they can attain a nomination (Bochel and Denver Reference Bochel and Denver1983). Because of their roles in selecting candidates and distributing resources, party leaders determine the pool of candidates from which voters can choose – and, consequently, the potential make-up of elected bodies (for example, Rehmert Reference Rehmert2022; Doherty, Dowling and Miller Reference Doherty and Dowling2019; Berz and Jankowski Reference Berz and Jankowski2022). Understanding party leaders’ preferences when selecting candidates is, thus, critical to examining the potential barriers to women’s political representation.
In broad terms, when forming candidate lists, party leaders are concerned with advancing the interests of their parties. Most often, this requires satisfying at least two different goals: 1) party competitiveness in elections and 2) party cohesiveness and loyalty in the post-electoral period (Galasso and Nannicini Reference Galasso and Nannicini2015; Rehmert Reference Rehmert2022; Berz and Jankowski Reference Berz and Jankowski2022). To ensure electoral competitiveness, party selectors must consider whether aspirants can attract voter support (Bochel and Denver Reference Bochel and Denver1983; Lopes da Fonseca Reference Lopes da Fonseca2017). Incumbents tend to have an advantage in this area since performance in previous races gives party selectors more direct insight into aspirants’ political capital (Carson, Engstrom and Roberts Reference Carson2007; Fouirnaies and Hall Reference Fouirnaies and Hall2014; Put, Muyters and Maddens Reference Put, Muyters and Maddens2021; Piscopo et al. Reference Piscopo and Hinojosa2022).
Party leaders must also consider whether their selected candidates will likely maintain party cohesiveness in the post-election period (that is, when they take office) (Berz and Jankowski Reference Berz and Jankowski2022; Greene and Haber Reference Greene and Haber2015). As with electoral competitiveness, incumbents also tend to hold an advantage in this area, as party leaders have information about their legislative behaviour and party service.
Incumbents’ advantages in candidate selection are not restricted to the greater amount of information available to candidate selectors; they are also a product of dynamics that ensure intra-party stability. First, as experienced party and legislative insiders, incumbents are generally better placed than newcomers to deliver on their parties’ legislative goals. Second, candidate (re)selection serves as a reward for those with track records in advancing partisan interests (Schmuck and Hohendorf Reference Schmuck and Hohendorf2023), so party leaders who fail to use candidate (re)selection to reward incumbents who have served their parties risk losing credibility and reducing incentives for accountability (Hazan and Rahat Reference Hazan and Rahat2010, 124–125). Because incumbents tend to overwhelmingly be ethnic-majority men (Murray Reference Murray2014), considerations about aspirants’ levels of political capital, party loyalty, and service reward may discourage the nomination of diverse party lists.
Nonetheless, when deciding on how to cast ballots, voters may evaluate not only the individual qualities of a specific candidate but also diversity in parties’ broader candidate lists (Rahat, Hazan and Katz Reference Rahat and Hazan2008; Rehmert Reference Rehmert2022). Although voters may have expectations about the diversity of party lists across several candidate characteristics – such as age, socio-economic class, and racial or ethnic background – the format of candidate lists may render voters particularly aware of gender-based imbalances. In many countries, the design of ballots means that voters would have to seek out information about candidates’ age, socio-economic class, and racial or ethnic backgrounds but may be able to infer candidates’ genders based on their first names more accurately. In addition, international efforts to promote women’s political representation (Krook Reference Krook2006; Bush and Zetterberg Reference Bush and Zetterberg2021) are likely to have made gender inequality in politics a more salient issue than the underrepresentation of other groups. In efforts to comply with voter expectations and avoid backlash, party leaders are thus likely to consider women’s underrepresentation when forming candidate lists (Clayton, O’Brien and Piscopo Reference Clayton2019; Weeks et al. Reference Weeks2023).
However, by design, women’s underrepresentation is inherently a product of the overrepresentation of men as incumbents (Murray Reference Murray2014), so parties’ efforts to address underrepresentation may incur the decision of deselecting incumbents who may possess ‘tested’ qualities of electability and party loyalty. Under certain circumstances, the deselection of men candidates may not affect men incumbents; this is the case of parties’ decisions to nominate women to unelectable positions at the bottom of their lists – a symbolic measure unlikely to promote women’s election (Schwindt-Bayer Reference Schwindt-Bayer2009; Hughes et al. Reference Hughes and Paxton2019).
In other words, addressing women’s underrepresentation requires giving women candidates the possibility of winning. In systems with pre-ordered candidate lists, this essentially means nominating women to viable list positions (Luhiste Reference Luhiste2015), which are highly contested and often reserved for party insiders (Galasso and Nannicini Reference Galasso and Nannicini2015).
Given our interest in understanding party elites’ willingness to address women’s underrepresentation, we focus on their likelihood of nominating women to prominent positions on their parties’ lists – that is, ‘safe’ positions with a high chance of election. We expect that while attempting to signal a commitment to gender equality in representation, party elites will also seek to minimize the costs of demoting men incumbents. As such, we anticipate that selectors become more likely to nominate women aspirants when gender imbalances in the candidate lists are more pronounced. Following this rationale, we hypothesizeFootnote 2 that:
H1 The lower the share of women among preselected candidates, the higher the likelihood that party elites will nominate women aspirants to viable placements.
While we expect the preferences of all party elites in candidate selection processes to be shaped by their efforts to maximize their parties’ goals in pre- and post-election periods, we grant that addressing women’s underrepresentation does not equally help all party elites achieve these goals.
Through internal party quotas or their manifestos, left-wing parties are more likely than other parties to make explicit aims to promote gender equality (Krook and Childs Reference Krook and Childs2010). Due to the need to comply with these internal quotas, selectors from left-wing parties have greater incentives to nominate women. In addition, left-wing voters are also more likely to have a higher commitment to gender equality than their right-wing counterparts (Norris Reference Norris1997; Caul Reference Caul1999; O’Brien Reference O’Brien2018); as such, they are likely to be more demanding of (and attuned to) women’s descriptive representation in the candidate lists of the parties they support. There may also be heterogeneity in the preferences of elites from the same party. For example, selectors who perceive themselves as more to the left of their parties may use candidate selection to signal and advance their normative commitments to greater diversity. Together, preferences driven by partisan interests, voter demands, and personal commitments provide greater motivation for left-wing party selectors to nominate gender-balanced candidate lists – and potentially deviate from their preferences towards other aspirant attributes to accomplish this. Thus, we expect that:
H2 The lower the share of women among preselected candidates, the higher the likelihood that left-wing party elites will nominate women aspirants to viable placements.
It is also plausible that party leaders’ willingness to address underrepresentation is shaped by their gender. Be it due to gender affinity, a normative commitment to gender equality, or being embedded into networks of women in politics, women tend to be more supportive of women’s political representation than men; when they gain power within their parties, women advance their preferences by facilitating women’s representation (Kittilson Reference Kittilson2006; Kenny Reference Kenny2013; Norris and Lovenduski Reference Norris and Lovenduski1995; Thomsen and Swers Reference Thomsen and Swers2017). As prior work has shown, women party leaders are more likely than men to recruit women with leadership potential and use their positions of power to promote gender balance in party lists (for example, Cheng and Tavits Reference Cheng and Tavits2011; Piscopo Reference Piscopo2016); as such, there tends to be a strong correlation between the share of women among party leaders and the share of women in candidate lists and elected positions (for example, Luhiste Reference Luhiste2015; Ashe and Stewart Reference Ashe and Stewart2012; Niven Reference Niven1998; Wylie Reference Wylie2018). Thus, we also anticipate that:
H3 The lower the share of women among preselected candidates, the higher the likelihood that women party elites will nominate women aspirants to viable placements.
We test our hypotheses with data from a conjoint survey experiment conducted in November and December 2019 with 1,389 party elites from nineteen parties in Austria (N=219), Germany (N=385), and Switzerland (N=785) – making this, to our knowledge, the largest and first cross-national conjoint experiment with party elites to date.
Data and Case Selection
We employ data from Austria, Germany, and Switzerland. Besides cross- and within-country variation in levels of women’s descriptive representation,Footnote 3 our three cases also offer comparable formal institutional settings of candidate selection, even though the electoral systems of each of these countries feature distinct idiosyncrasies: In all three cases, 1) elections employ a PR system and 2) state party elites play a prominent role in the candidate selection process. In Germany and Switzerland, party lists are determined by party sub-organisations at the state levels (Länder and Cantons); in Austria, party lists are composed at the regional, state, and federal levels.
While all three cases have common formal institutional features that make them comparable, they also provide relevant variation regarding the openness of their candidate lists. This allows us to observe whether party leaders’ willingness to address women’s underrepresentation varies across systems. In Germany’s closed-list PR system, voters cannot alter party lists or split votes. This means that concerns over individual candidates’ electability should be less salient, and that party selectors may be guided by other types of considerations – such as descriptive representation – when forming candidate lists (Rehmert Reference Rehmert2022). Because voters cannot alter party lists and select their preferred candidates, party elites’ failure to consider gender balance (Detterbeck Reference Detterbeck2016) may be punished at the ballot.
Contrasting Germany’s closed-list system, in Switzerland’s open-list PR, voters may choose to vote for a party list but can also split their votes between different party lists and candidates. Swiss voters can also allocate up to two votes per candidate. In this system, candidates (not only parties) compete with one another; this gives party elites greater incentives to nominate highly electorally competitive aspirants. In this context, concerns about the gender balance of party lists may take a back seat to the priority of nominating a list of candidates with strong individual electoral support.
The Austrian case sits in the middle of the other two. While Austrian voters cannot split their votes between parties, they can influence individual candidates’ party list positions via preferential votes. However, as voters are not allowed to split votes, Austrian parties should also have an incentive to produce gender-balanced party lists.
In sum, our three cases allow us to examine party elites’ efforts to address women’s underrepresentation in systems where they have substantial power over women’s electoral fate and explore whether different institutional incentives impact party elites’ likelihood of considering women’s underrepresentation when composing party lists.
Austria, Germany, and Switzerland also offer useful within-case variation. While none of these countries have national legislated quotas, some of their parties have instituted voluntary party quotas. In fact, the Social Democratic Party of Austria and the German Greens were early adopters of internal voluntary quotas in 1985 and 1986, respectively (IDEA 2024). In addition to within-country variation in the use of formal institutions, parties’ reliance on informal institutions also varies within countries (Bjarnegård and Zetterberg Reference Bjarnegård and Zetterberg2019; Bjarnegård and Kenny Reference Bjarnegård and Kenny2015). To account for potentially different within-country party-level incentives and practices of candidate selection, we aimed to fully sample elites from all nineteen nationally relevant parties in our three cases.Footnote 4
We define party elites as all members of the executive boards of state-level party sub-organisations in Germany and Switzerland and state and federal party organisations in Austria. To identify these individuals, we searched the websites of the state-level party organizations of all nineteen major parties from our cases. These efforts resulted in a list of 4,641 individuals (694 in Austria, 1,815 in Germany, and 2,131 in Switzerland), which corresponds to a complete list of party elites in 97 per cent of all state-level parties across the three countries. While we could not obtain all the names of the party elites for every party in every state, according to our estimates, the population should be at most 5,000 individuals. Of the 4,641 party elites identified, we found corresponding email addresses for 4,206 people (90.6 per cent).
From November to December 2019, we sent emails and reminders inviting these party elites to participate in the survey.Footnote 5 Out of the 4,206 individuals contacted, 1,389 completed the survey. This corresponds to a response rate of 33 per cent.Footnote 6 Although there is no systematic review of surveys conducted with party elites, experience from recent studies suggests that our response rate is high for an elite survey (Bailer Reference Bailer, Martin, Saalfeld and Strøm2014). Notably, our response rate is higher than that of recent conjoint experiments conducted with party elites: 20 per cent for Rehmert (Reference Rehmert2022), 20.7 per cent for Berz and Jankowski (Reference Berz and Jankowski2022), and 16.7 per cent for Doherty, Dowling and Miller (Reference Doherty and Dowling2019).Footnote 7
It is worth noting that party convention delegates have the final decision-making power over the party list for some of the parties that comprise our sample. We opt for surveying party elites over convention delegates for several reasons. First, while, formally, party convention delegates have the final say in the candidate selection process, their actual power varies considerably across parties. For example, German party elites usually draft a party list, and while party convention delegates are allowed to alter this list, whether they do varies substantially across parties, with the proposed and final lists being identical or very similar in some parties and notably different in others (Höhne and Schüttemeyer Reference Höhne and Schüttemeyer2019). In general, in parties with strong regional branches, such as in the cases of Germany and Switzerland, regional party elites are the most powerful group in the candidate selection process (Norris and Lovenduski Reference Norris and Lovenduski1995; Luhiste Reference Luhiste2015).
The second reason for opting for party elites instead of delegates is the feasibility of studying all relevant parties in three countries simultaneously. All party leadership members’ names and contact information are usually publicly available, whereas this is not the case for party convention delegates.
Thirdly, whereas all parties have some executive board elected by the party membership or delegates, the way parties delegate members to party conventions varies from country to country, subnational unit to subnational unit, and party to party.
Survey Design
When choosing between aspirants, party selectors face a multidimensional decision-making process that involves considering how the characteristics of individual aspirants improve the electoral chances of their overall party lists. Conjoint experiments provide a way of ‘mimicking’ this process, as the method allows us to analyse how different aspirant attributes influence party selectors’ decision to choose specific candidates over others (Hainmueller, Hangartner and Yamamoto Reference Hainmueller, Hangartner and Yamamoto2015).
In conjoint experiments, respondents are typically confronted with a profile or set of profiles of one or several objects with randomly varied relevant attributes. Respondents are then asked to choose between these alternatives or to rate them. Since the method allows researchers to mimic scenarios of real-world complex decision making in which individuals are expected to consider various factors, conjoint experiments often have high external validity (Hainmueller, Hangartner and Yamamoto Reference Hainmueller, Hangartner and Yamamoto2015).
Unsurprisingly, conjoint experiments are widely used in studies analysing candidate evaluation by voters,Footnote 8 but the technique has also been recently used to study candidate selection by political elites (for example, Rehmert Reference Rehmert2022; Berz and Jankowski Reference Berz and Jankowski2022; Doherty, Dowling and Miller Reference Doherty and Dowling2019). In addition, conjoint experiments have been shown to mitigate social desirability bias (Horiuchi, Markovich and Yamamoto Reference Horiuchi, Markovich and Yamamoto2022), a salient concern when studying preferences towards sensitive issues, such as women’s underrepresentation.
While conjoint experiments can take many forms, in this paper, we employ a paired conjoint forced-choice design. Party selectorates always have to choose candidates to compose their party lists and are usually contemplated with scenarios in which they have to choose between two (or more) aspirants; as such, this design more accurately reflects the candidate selection process than single vignettes and single-profile conjoint designs. Specifically, in this type of setup, respondents are presented with two profiles and asked to choose one. Because the attributes that make up each profile are randomly assigned, the method allows us to estimate how particular characteristics shape respondents’ likelihood of selecting profiles – and, as such, to understand respondents’ preferences for specific profile traits.
Table 1 shows the attributes of aspirants and candidate lists randomly assigned to form aspirant profiles for the conjoint. In the absence of information about aspirants’ electoral performance – as is the case for newcomers – party selectors can use other individual-level characteristics such as educational attainment and non-political/electoral professional experience as shortcuts to evaluate aspirants’ electoral viability and likelihood of following the party line. The individual-level attributes chosen for the conjoint thus reflect information that is usually available to party selectorates and that may be used as proxies for aspirants’ political capital and party loyalty.
Example of randomly assigned aspirant and candidate list attributes
Note: Attributes were completely randomized for each aspirant profile. The only exception is the aspirants’ gender. Respondents were shown a man-man and woman-woman aspirant pair with a probability of 25 per cent each. A man-woman aspirant pair was shown with a probability of 50 per cent.
Specifically, and similar to other studies (for example, Rehmert Reference Rehmert2022; Berz and Jankowski Reference Berz and Jankowski2022), profiles presented to survey participants included key socio-demographic information about aspirants (their age and gender), as well as indicators of electability (educationFootnote 9 and parliamentary experience), and partisan ideological congruence (ideology and priority policy issue). We chose to provide information on aspirants’ ideological position relative to that of respondents. We deemed that employing aspirants’ relative ideological position was more appropriate as both the aspirants and the party elites forming candidate lists are members of the same party. This means their general ideological positions should be similar, and ideological differences should be minor and only be understood in relative terms. For example, a conservative party would not nominate a left-wing candidate but could potentially appoint a candidate who is left-wing relative to the median conservative selector.
Finally, as shown at the bottom of Table 1, respondents also received information about the gender ratio of candidates already selected for their party lists. This allows us to estimate party elites’ preferences for aspirants’ individual-level attributes conditional on the signalling of different levels of women’s representation among previously selected candidates. In other words, this strategy enables us to examine whether the make-up of overall party lists influences party selectors’ choices between candidate profiles.Footnote 10
As shown in Table 1, the lowest level of women’s representation signalled to respondents was 10 per cent, while the highest signalled gender parity. Since women are rarely overrepresented in party lists (Höhne Reference Höhne2020; Gilardi Reference Gilardi2015), information about the gender breakdown of already selected candidates never tipped to favour women. Information about the gender ratio of already selected candidates was randomized with each new pair of aspirants presented to respondents. Primes were fully randomly assigned and, thus, were independent of respondents’ previous choices (that is, the gender ratio did not become more men-dominated if a respondent chose only men in earlier rounds of the experiment).
After answering a short battery of questions to confirm their status as party selectors and provide information about their socio-demographic characteristics and ideological self-placement, respondents were given a brief explanation of their task in the conjoint experiment. As prior research has shown, nominating women to unelectable positions at the bottom of candidate lists allows parties to signal compliance with gender egalitarian social norms and expectations without costs of displacement to party insiders (for example, Luhiste Reference Luhiste2015). Given our interest in understanding the selectors’ decision-making process under a scenario that could incur costs to (men) incumbents, respondents were tasked to choose aspirants for an electorally viable position on their parties’ list. This condition was kept constant (that is, was non-randomized) and was reinforced three times in the explanation of the task. The text used was as follows:
In the next part of the survey, you are presented with eight pairs of fictitious aspirants seeking a candidacy in your party.
You will receive some information about each candidate, such as their age and policy priority area. These aspirants are applying for a prominent placement on your party’s list for the upcoming national parliamentary elections.
You will also receive information about the composition of the list of candidates who have already been selected.
Based on the information provided, please select the aspirant that you would prefer to place on a prominent placement in your party’s candidate list. Please assume your chosen candidate will represent your party and your voters in the next parliament.
Upon confirming their understanding of the task, each respondent selected candidates from eight paired forced-choice experiments. This provides us with over 26,000 evaluated profiles. An example of the paired aspirant profiles presented to respondents is in Table 2.
Randomly assigned aspirant and candidate list attributes
Following recent methodological developments in the analysis of conjoint experiments, we estimate marginal means (MM) effects to test our hypotheses (Leeper, Hobolt and Tilley Reference Leeper, Hobolt and Tilley2020), which, unlike average marginal component effects (AMCES) (see, for example: Hainmueller, Hopkins and Yamamoto Reference Hainmueller, Hopkins and Yamamoto2014), do not require the display of a reference category, thus facilitating the visualization of results (Rehmert Reference Rehmert2022). Since respondents had to complete eight choice tasks, the observed outcomes of these choice tasks are not independent. As is standard practice in these cases, we cluster standard errors at the respondent level.
Results
H1 poses that party elites are more likely to select women aspirants when women are underrepresented among (the primed gender ratio of) previously selected candidates. Figure 1 shows respondents’ preferences for aspirants of different genders by different primed levels of women’s underrepresentation in their parties’ preselected lists.Footnote 11 The figure displays MM and their corresponding 95 per cent confidence intervals for each primed level of women’s representation. The symbols (that is, circles) indicate respondents’ preference for a given characteristic when all other attribute levels are held constant. Symbols left of the dashed centre-line (below 0.5) indicate negative favourability, while symbols right of the dashed centre-line indicate a positive favourability (that is, a higher chance of being selected). Because we are interested in how the primed gender ratios in candidate lists affect party elites’ likelihood of selecting women, interpreting the effect of the prime on respondents’ preferences requires evaluating whether or not confidence intervals overlap vertically (that is, across levels of the prime).
The impact of underrepresentation on the selection of women aspirants (MM).
Note: Compiled by authors. The unit of analysis is the individual respondent. Estimates correspond to MM effects with the full sample of party elites (N = 1,389). The figure displays respondents’ preferences for men and women aspirants at different levels of primed information about women’s representation in candidate lists. Interpreting the effect of the prime on respondents’ preferences requires evaluating whether or not confidence intervals overlap vertically (that is, across levels of the prime). For example, the probability of selecting women is higher at 30 per cent underrepresentation than when primed at 40 and 50 per cent, but indistinguishable when primed at 20 and 10 per cent. For full estimates, see Appendix L.
As shown in Figure 1, even when women are not underrepresented (that is, when primed as 50 per cent among previously selected candidates, our baseline), respondents have a pro-woman bias.Footnote 12 More specifically, in scenarios in which the primed information is of gender parity, 55.8 per cent of selected aspirant profiles are of women. This is consistent with previous studies, which also find that, in conjoint experiments, party elites are more likely to choose women over men (Rehmert Reference Rehmert2022; Doherty, Dowling and Miller Reference Doherty and Dowling2019; Berz and Jankowski Reference Berz and Jankowski2022).
Our primary interest is in assessing whether women’s underrepresentation – that is, moving away from the baseline of gender parity – further increases respondents’ preferences for women aspirants. Consistent with our expectations, we find that women aspirants’ advantage over men grows the more underrepresented women are in the party list, indicating party elites’ effort to address women’s underrepresentation. At a 40 per cent share of women candidates on the list, 59.9 per cent of the chosen profiles are of women. Women’s advantage over men at this level of the prime is 4.1 percentage points (pp) higher than at gender parity. When respondents are informed that women compose 30, 20, or 10 per cent of candidates previously elected to the list, the share of women aspirant profiles selected reaches 63.3 per cent – 7.5 pp points higher than at baseline.
Notably, while the effect is substantively the highest when women are described to be 10 per cent of preselected candidates on the list, this effect is statistically indistinguishable from the effect sizes of primes informing respondents that women are 20 or 30 per cent of the lists. These findings support H1, indicating that women aspirants enjoy a greater advantage the more the party list is skewed against them – but also that women’s advantage has a limit.
Cross-country variation in formal and informal norms of candidate recruitment, as well as cross-party variation in adherence to party-level gender quotas (Kenny and Verge Reference Kenny and Verge2016; Davidson-Schmich Reference Davidson-Schmich2006), mean that party leaders may operate under different incentives for candidate selection, which, in turn, could impact their preferences towards addressing women’s underrepresentation. We find, however, that party elites’ preferences for women candidates under different conditions of underrepresentation are not sensitive to institutional variations in candidate selection. As shown in Appendices E and F, our findings hold in sub-country analyses and are robust to analyses considering whether respondents belong to parties with or without gender quotas. For example, effects are equally as prominent in Germany’s party-centric closed-list PR system, where voters cannot alter candidate-list order, as in Switzerland’s more personalized open-list PR system, where voters can split their votes between party lists and candidates. Selectors operating in parties with internal targets for women’s candidacies are more responsive to our primes, but signalling women’s underrepresentation is also effective in moving the preferences of elites from parties that do not have quotas.
To win seats and elect representatives loyal to their parties, selectors must engage in complex decision-making processes that involve balancing the characteristics of individual aspirants and their overall parties’ candidate lists. This means that addressing women’s underrepresentation may require (at least some) deviation from party selectors’ average preferences on individual-level characteristics associated with electoral competitiveness and party loyalty.
As introduced earlier, in addition to information about aspirants’ gender and the gender composition of preselected candidates, survey respondents were also given information about aspirant characteristics potentially associated with electability (education and parliamentary experience) and party loyalty (ideology and main policy focus area). This allows us to explore whether party elites’ preferences towards aspirants’ features besides gender also change when they are primed to consider the level of gender imbalance in their parties’ lists. As shown in Appendix G, our analyses indicate that party elites do not deviate from their preferences towards indicators of electability or partisan ideological congruence to address women’s underrepresentation.
Although tackling gender imbalances in candidate lists may help party elites avoid potential reputational costs and remain competitive in elections, selectors do not all have the same pressure to nominate gender-balanced slates and, as such, may have different preferences when it comes to addressing women’s representation. Even in contexts where they face similar pressures, party elites’ personal commitments to addressing women’s underrepresentation are also likely to vary.
Party elites’ ideology
To explore potential variation in preferences, we first evaluate the effect of our primes on the candidate selection strategies of left-wing, centrist, and right-wing party elites. Since individual respondents’ self-perceptions may somewhat deviate from externally coded measures of their respective parties’ ideologies, we create subsamples by using respondents’ answers to the pre-treatment question ‘how would you describe your ideological placement?’, which had as possible answers: left, centre, and right. This also allows us to pick up potential variations in selectors’ ideology within parties. Out of the 1,389 respondents, 563 (40.5 per cent) placed themselves on the left, 569 (40.9 per cent) on the centre, and 257 (18.5 per cent) on the right.
As shown in Figure 2, and in line with our main results, at gender parity, women have a higher chance than men of being selected by party elites from all ideology-based subsamples. However, the extent of women’s advantage over men varies substantially according to respondents’ ideology. Consistent with previous work on party ideology and commitments to gender equality (for example, Caul Reference Caul1999; O’Brien Reference O’Brien2018), we find that left-wing selectors have a stronger preference for women aspirants than those who self-place as centrists or right-wing. At baseline, 58.7 per cent of aspirant profiles selected by left-wing party elites are of women. At this same level of the prime, the preferences of centrist and right-wing respondents are indistinguishable from one another at 53.8 and 53.6 per cent, respectively.
The impact of underrepresentation on the selection of women aspirants, by respondents’ ideology (MM).
Note: Compiled by authors. The unit of analysis is the individual respondent. Estimates correspond to MM effects with the subsamples (Left = 563; Centre = 569; Right = 257). The figure displays respondents’ preferences for men and women aspirants at different levels of primed information about women’s representation in candidate lists, by respondents’ ideology. Interpreting the effect of the prime on respondents’ preferences requires evaluating whether or not confidence intervals overlap vertically (that is, across levels of the prime). For full estimates, see Appendix L.
As anticipated by H2, we expect cues of women’s underrepresentation to affect party elites from different ideological groups heterogeneously. As Figure 2 illustrates, left-wing respondents are the most opposed to gender imbalances in party lists, and being primed to consider women’s underrepresentation at any level (that is, 10, 20, 30, and 40 per cent) improves women’s chances of being selected to a similar extent (as indicated by overlapping confidence intervals). More specifically, among left-wing respondents, women aspirants’ advantage over men increases from 18.4 pp at gender parity to 35.3 pp when the primed share of preselected women is 10 per cent (a difference of 16.9 pp).
The effect of cueing gender imbalances in party lists is even more pronounced among centrist respondents. Whereas women aspirants enjoy an advantage over men of 8 pp at gender parity, when women are primed to represent 10 per cent of preselected candidates, their advantage over men is 27.6 pp (a difference of 19.6 pp). Critically, priming centrists to take women’s underrepresentation into account during candidate selection moves their preferences towards women aspirants from being indistinguishable from those of right-wing selectors (when women are primed to be 50 and 40 per cent of preselected candidates) to being indistinguishable from those of left-wing elites (when women are primed to be 30 and 10 per cent).
The most striking result, perhaps, is that of the effect of our primes on right-wing party elites. Although women aspirants have an advantage over men also among right-wing respondents (of 7.1 pp), as indicated by the overlapping confidence intervals in Figure 2, right-wing respondents’ selection of women candidates is unaffected by the primed level of women’s underrepresentation. Effectively, the results show that while women also enjoy an advantage among right-wing party elites, among this group, a woman aspirant has about the same chance of being selected at gender parity as in cases where only 10 per cent of previously selected candidates are women. This suggests that, on average, addressing underrepresentation is not an issue right-wing party elites consider when forming candidate lists. As shown in Appendices H and I, respectively, these findings also largely hold in sub-country analyses and analyses with subsamples of respondents in parties with and without gender quotas.
Party elites’ gender
Besides ideology, it is also possible that party elites’ gender shapes their commitments to addressing underrepresentation in candidate selection. Of the nineteen parties included in this study, women are only (slightly) overrepresented among state party elites from the Austrian and German Greens. Women are somewhat or substantially underrepresented in the other seventeen parties from which our sample was drawn (see Appendix B). Across the population of party elites from the nineteen parties included in the study, 33.7 per cent of party elites are women. Within our sample, 524 respondents (36.1 per cent) are women.
In Figure 3, we break down our analyses by the self-reported gender of party elites. As shown, at baseline, women respondents are more likely than men to favour women aspirants. This is consistent with previous studies, which find that women party leaders are more active in recruiting and nominating women (for example, Ashe and Stewart Reference Ashe and Stewart2012; Cheng and Tavits Reference Cheng and Tavits2011).
The impact of underrepresentation on the selection of women aspirants, by respondents’ gender (MM).
Note: Compiled by authors. The unit of analysis is the individual respondent. Estimates correspond to MM effects with the subsamples (Men = 865; Women = 524). The figure displays respondents’ preferences for men and women aspirants at different levels of primed information about women’s representation in candidate lists, by respondents’ gender. Interpreting the effect of the prime on respondents’ preferences requires evaluating whether or not confidence intervals overlap vertically (that is, across levels of the prime). For full estimates, see Appendix L.
Our primes cueing respondents of the level of women’s underrepresentation among preselected candidates affect both women’s and men’s likelihood of choosing women aspirants. Contrary to H3, however, we find that the effect of our primes is strongest among male selectors. Specifically, at gender parity, women aspirants already hold an advantage of 21 pp over men aspirants among women party elites. At the lowest level of primed women’s representation (10 per cent), women aspirants’ advantage over men increases to 29.7 pp (a difference of 8.7 pp).
Meanwhile, among men selectors, women aspirants’ advantage over men for the same primed levels increases much more drastically, from 6.4 pp to 26 pp, a difference of 19.6 pp. Overall, these results suggest that while women selectors support women aspirants at all levels, men selectors’ preferences towards women aspirants are more sensitive to change when given cues about women’s underrepresentation. Similar to our results about the impact of our primes on centrist party leaders, this result points to a potential strategy for promoting women’s greater representation beyond the ‘usual suspects’ (Celis and Erzeel Reference Celis and Erzeel2015). The findings of our analyses by party elites’ gender also hold to subgroup analyses by country and by parties with and without gender quotas (Appendices J and K).
Discussion
Together, our findings indicate that becoming aware of gender imbalances in their parties’ lists motivates some selectors to prioritize women. As we show, respondents of different ideological backgrounds and genders display different baseline preferences for women aspirants. Consistent with earlier work that attests to the importance of ideology in improving women’s opportunities in candidate selection (Caul Reference Caul1999; O’Brien Reference O’Brien2018), we find that left-wing elites have a higher preference for women aspirants even at baseline – a testament to their normative preference for promoting women’s political representation. Nonetheless, when primed to consider women’s underrepresentation, not only left-wing selectors but also centrists become more supportive of nominating women. In fact, the effects of our primes are the strongest among the latter. Notably, respondents are most responsive when women are said to be less than 30 per cent of candidates in party lists, suggesting that selectors are willing to accept some gender imbalance, as long as it is not as evident to voters.
Conversely, our primes do not affect the preferences of right-wing respondents. Among party elites for whom gender imbalances could disengage voters, priming women’s underrepresentation improves the chances of women aspirants; meanwhile, the potentially lower demand for gender-balanced slates among their core constituents seems to render right-wing selectors, on average, less prone to addressing women’s underrepresentation. This is consistent with prior research showing that conservative voters do not tend to demand policies to address gender-based inequalities, which reduces conservative politicians’ incentives to prioritize gender equality in decision making (Araújo and Gatto Reference Araújo and Gatto2022).Footnote 13
In other words, while normative commitments to gender equality seem to be critical to forming party leaders’ attitudes towards the general recruitment of women aspirants, priming women’s underrepresentation moves the preferences of some groups of party elites further in favour of women aspirants. Taken together, our results suggest that party elites choose to address women’s underrepresentation by selecting more women aspirants when doing so might be electorally advantageous to their parties (or, at least, when failing to do so could harm their party’s reputation and electoral chances). Consistent with our theoretical expectations, these findings indicate that party elites’ preferences in candidate selection processes reflect strategic decisions about their parties’ electoral standing.
While in alignment with conjoint experiments that reveal a pro-woman bias among party elites (Rehmert Reference Rehmert2022; Doherty, Dowling and Miller Reference Doherty and Dowling2019; Berz and Jankowski Reference Berz and Jankowski2022), our results contrast with observational studies that show that candidate selection processes are gendered in ways that disadvantage women (for example, Norris and Lovenduski Reference Norris and Lovenduski1995; Bjarnegård and Kenny Reference Bjarnegård and Kenny2015; Gatto and Wylie 2022; Murray Reference Murray2012).
There are several potential reasons for this mismatch. First, gender biases operate in nuanced ways that may not be fully captured by our (or any) experimental setting. For example, because women tend to undermine their political qualifications (Lawless and Fox Reference Lawless and Fox2005; Carnes and Lupu Reference Carnes and Lupu2016), those who reach aspirant status tend to be, on average, more qualified than men (Lawless and Fox Reference Lawless and Fox2010; Dahl and Nyrup Reference Dahl and Nyrup2021). So, although our conjoint randomly assigns aspirant characteristics, party elites’ selection of profiles in this setting may already embed their priors about women’s higher competence, leading them to interpret whichever attributes are assigned to women’s profiles as of higher quality. This may further boost selectors’ preferences for women candidates in the conjoint setting.
Nonetheless, even if selectors tend to perceive women aspirants as more qualified than men, in actual contexts of candidate selection, women’s historical exclusion from the political domain may put them at a disadvantage, as they are often less likely than men to have access to influential networks and conform to expectations about party service and incumbency – factors that could lead them not to be prioritized in candidate selection and the distribution of resources, despite their qualifications (Teele, Kalla and Rosenbluth Reference Teele, Kalla and Rosenbluth2018; Luhiste Reference Luhiste2015; Piscopo et al. Reference Piscopo and Hinojosa2022). Although we have sought to address this in our design by randomly assigning information about aspirants’ political experience, party service and incumbency often offer subtle and largely unobservable sources of advantage (for example, personal bonds with party leaders) – which are, in themselves, gendered (Kenny and Verge Reference Kenny and Verge2016).
Second, as gender scholars have long established, women’s political underrepresentation is a product of supply and demand factors (Norris and Lovenduski Reference Norris and Lovenduski1995; Schwindt-Bayer and Mishler Reference Schwindt-Bayer and Mishler2005). In the context of the conjoint experiment, the supply of women aspirants is unlimited, while in natural environments, women often represent a lower share of contenders (Krook Reference Krook2010). This means that even if party elites wanted to address underrepresentation in candidate selection processes, it is possible that there would not be a sufficiently large pool of women aspirants from which they could choose. Critically, even if a pool of women candidates was numerically available, their profiles might not match what party selectors were looking for. Additionally, since women aspirants can self-select into their party of choice, it is plausible to expect that, in contexts where women are increasingly more left-wing than men (for example, Giger Reference Giger2009), right-wing parties could face a bigger challenge than other parties to find suitable women aspirants.
As prior studies have shown, however, although party leaders often justify women’s underrepresentation as a supply-sided issue (Krook and Childs Reference Krook and Childs2010; Krook Reference Krook2010), in cases where gender quotas have been nationally legislated, parties across the ideological spectrum have managed to transform the gender composition of their candidates drastically (Franceschet, Krook and Piscopo Reference Franceschet2012; Paxton and Hughes Reference Paxton and Hughes2015). In other words, the supply of candidates is not entirely exogenous to the actions of party elites, so when the incentives of candidate nominations change, party leaders find a way to respond to new demands. In addition, party leaders’ active recruitment may in itself increase women’s intention to run for office, thus boosting the supply of women aspirants (Fox and Lawless Reference Fox and Lawless2010). All in all, the scholarship shows that while gendered opportunity structures may lead women to ‘opt-out’ of actively declaring their political aspirations, formal and informal incentives that increase party leaders’ efforts to incorporate more women into their structures can foster greater levels of women’s candidate emergence (Piscopo and Kenny Reference Piscopo and Kenny2020).
Conclusion
Although advances in this area are undeniable, women’s political underrepresentation continues to persist throughout the world. As gatekeepers to candidate selection and the distribution of resources (Luhiste Reference Luhiste2015; Piscopo et al. Reference Piscopo and Hinojosa2022), party leaders are in a pivotal position to tackle underrepresentation (Norris and Lovenduski Reference Norris and Lovenduski1995; Kenny and Verge Reference Kenny and Verge2016). But is addressing women’s underrepresentation a concern party elites have when forming candidate lists?
Prior studies have shown that party leaders’ preferences in candidate selection processes are often shaped by their efforts to maximize their parties’ electoral viability and minimize the risks that those elected will deviate from partisan interests (Galasso and Nannicini Reference Galasso and Nannicini2015). Since most aspirants possessing these ‘tested’ qualities may be ethnic-majority men (Murray Reference Murray2014), candidate selection strategies that focus on individual-level characteristics may result in heavily homogeneous lists; nonetheless, the underrepresentation of politically marginalized groups – particularly of women, whose first names may make them more identifiable as underrepresented in candidate lists – may in itself compromise parties’ competitive standing with the electorate (Rahat, Hazan and Katz Reference Rahat and Hazan2008; Rehmert Reference Rehmert2022; Weeks et al. Reference Weeks2023).
To explore whether addressing women’s representation is a factor shaping candidate selection processes, we employ a conjoint experiment with 1,389 party elites in Austria, Germany, and Switzerland – making this, to our knowledge, the most extensive and first cross-country conjoint experiment with party elites.
As part of the conjoint experiment, party leaders were asked to select between pairs of aspirants to occupy a prominent position on their parties’ candidate lists. Besides information about the profile of each aspirant, party leaders were also primed about the gender ratio of candidates preselected to their party lists (ranging from 10 per cent of women to gender parity). This design allows us to isolate the effect of considerations about women’s underrepresentation in candidate selection.
Aligned with our expectations, we find that party elites’ likelihood of selecting women aspirants increases the lower women’s representation among previously selected candidates. Reinforcing earlier findings that gender quotas do not hinder the ‘quality’ of politicians (Besley et al. Reference Besley, Folke, Persson and Rickne2017; Weeks and Baldez Reference Weeks and Baldez2015), our results indicate that efforts to address underrepresentation through the selection of women do not lower selectors’ standards of candidate characteristics associated with electability and partisan alignment.
Notably, being prompted to consider women’s underrepresentation has an even more sizable impact on centrist selectors; this effect distances centrists from the preferences of right-wing respondents and approximates them to left-wing elites’ high inclination for women aspirants. By contrast, right-wing selectors are unaffected by cues of underrepresentation. While, similar to others, right-wing party elites also display a slight pro-woman bias, this is no different when women’s underrepresentation is primed to be 50 or 10 per cent among previously selected candidates.
Also aligned with previous scholarship attesting to the relevant role of women party leaders in promoting women’s political representation (for example, Karpowitz, Monson and Preece Reference Karpowitz, Quin Monson and Robinson Preece2017; Cheng and Tavits Reference Cheng and Tavits2011), we find that women party elites display higher support for women aspirants than men selectors; women’s higher baseline preferences for women aspirants, however, means that they are not as substantively affected by our primes as men respondents.
Together, the results of our sub-group analyses indicate that cueing party elites about women’s underrepresentation improves women aspirants’ chances of being selected among the ‘usual suspects’ (left-wing and women selectors), as well as among those for whom gender balance in representation may not be a crystallized criterion of candidate selection but could still be an electoral advantage (centrist and men party elites) – but not among those for whom signalling opposition to gender equality may be electorally strategic (right-wing selectors). All in all, our findings indicate that women’s underrepresentation may become an essential nomination criterion when made salient during candidate selection processes.
Although it is possible that only individuals already committed to gender equality would spontaneously and recurrently seek out this information outside the experimental setting, our study suggests that awareness campaigns that remind party leaders of the potential reputational costs associated with nominating a gender-imbalanced slate may prompt other groups of party leaders to consider the importance of addressing the underrepresentation of women in candidate selection. Campaigns and interventions that seek to increase the salience of women’s underrepresentation during periods of party conventions may thus be an additional strategy for promoting gender equality in political office – particularly among groups for whom this may not be a salient issue. Future studies could more directly test this by explicitly assessing whether pressures from civil society to address gender imbalances in candidate selection produce similar effects to the ones we uncover.
Future scholarship may also explore whether our findings travel to other contexts. While some respondents belonged to parties with voluntary party quotas, none of our cases have legislated national-level gender quotas. Party selectors operating under systems of legislated quotas may already incorporate gender criteria into their candidate selection strategies; the strength of quota designs may also matter for how elites not only select candidates but also distribute valuable resources, such as campaign funds or high placement on their parties’ lists (Schwindt-Bayer Reference Schwindt-Bayer2009; Hughes et al. Reference Hughes and Paxton2019).
Finally, party leaders may also make calculations about the descriptive representation of other groups (for example, Jensenius Reference Jensenius2016; Weeks and Allen Reference Weeks and Allen2023; Celis and Erzeel Reference Celis and Erzeel2017). Future work may examine whether selectors’ willingness to address the underrepresentation of women also applies to other politically marginalized groups. Unlike women, who are the majority of the voter population across many democracies and whose political representation is the target of international campaigns, signalling the underrepresentation of other groups – such as racial and ethnic minorities, LGBTQI+ people, or people with disabilities – may not trigger the same effect. Relatedly, future studies may also investigate gatekeepers’ preferences on list composition more holistically. For example, how do considerations about women’s underrepresentation interact with party leaders’ preferences on regional representation and the incorporation of marginalized groups?
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S0007123424000723.
Data availability statement
Replication data for this article can be found in the Harvard Dataverse at https://doi.org/10.7910/DVN/XRNZC7.
Acknowledgements
We are thankful to Victor Araújo, Fabrizio Gilardi, Thomas Kurer, Katharina Lawall, Débora Thomé, and participants of the URPP Conference on Gender and Inequality for valuable feedback and suggestions. All remaining errors are our entire responsibility.
Financial support
None.
Competing interests
None.
Ethical Standards
This study was reviewed by the University of Zurich’s Ethics Committee and approved on 10 April 2019 (approval number 19.4.11). More information about research ethics is available in Appendix A.
Open access
