1. Introduction
With the Soviet Union's fall—where quotas had guaranteed women's political presence—women's legislative presence collapsed (Sundstrom, Reference Sundstrom2010; Thames, Reference Thames2018). Today, women's representation in much of post-Soviet Europe remains low. Women's political underrepresentation in post-Soviet democracies, which tend to be characterized by high levels of political corruption, is often assumed to be the result of patriarchal gender norms and voters’ preferences for placing men in leadership roles (Moser and Scheiner, Reference Moser and Scheiner2012). Our study offers the first experimental tests of whether or not Ukrainian voters prefer men candidates and also provides insight into how voters respond in democracies with high levels of corruption.
In democracies where corruption figures prominently in voters’ minds, it is also important to identify whether the success of anti-corruption platforms has a distinct gender dimension. Existing scholarship documents a link between women's electoral success and anti-corruption politics: in certain contexts, high levels of women's representation is associated with lower levels of corruption (Esarey and Schwindt-Bayer, Reference Esarey and Schwindt-Bayer2017). Furthermore, voters’ receptiveness to women candidates who run on an anti-corruption platform may be conditional on voter gender. Eggers et al. (Reference Eggers, Vivyan and Wagner2018) provide evidence that men and women voters react differently to the perception that candidates are corrupt, with women voters reacting especially positively to women candidates’ non-corrupt (good) behavior.
The existing literature on gender and politics leaves a number of important questions unanswered: Do voters in corrupt democracies overtly discriminate against women candidates? In corrupt contexts where women are political outsiders, do anti-corruption politics confer a gender advantage to women candidates? We answer these questions using two vignette experiments and a separate conjoint analysis.
The paper proceeds as follows. Section 2 reviews the research on the causes of women's political underrepresentation in democracies characterized by both high and low levels of political corruption, and outlines our expectations. Section 3 describes our first study where we embedded a vignette experiment in two separate waves of data collected in Ukraine in 2015 and 2016. The second wave replicates the first wave to increase the validity of our findings. In these vignette experiments, we asked respondents to indicate their willingness to vote for a set of hypothetical candidates who are said to be fighting corruption in Ukraine. We randomized respondents’ exposure to pictures of women or men candidates. To preview the findings, Study 1 offers no evidence that, ceteris paribus, voters prefer men candidates to women candidates. Study 1 also offers no evidence that men and women voters differ in their support for men or women candidates.
Although Study 1 represents an important first step in understanding whether voter discrimination is driving women's political underrepresentation in democracies with high levels of corruption, Study 1's drawback is that it offers a unidimensional test of only two hypotheses. Because candidates’ platforms and other attributes were held constant, Study 1 does not clarify how different combinations of attributes—such as how the intersection of candidate gender, platform promises, family status, experience, age, and ethnicity—impact vote choice. Therefore, Section 4 describes and presents the results of a separate conjoint analysis that directly addresses the limitations of Study 1. Conjoint analyses allow researchers to test a larger number of causal hypotheses and estimate the effects of multiple treatment components on a single behavioral outcome (Hainmueller et al., Reference Hainmueller, Hopkins and Yamamoto2014; Hainmueller and Hopkins, Reference Hainmueller and Hopkins2015). Our conjoint analysis also allows us to clarify whether women in Ukraine, an electoral democracy characterized by high levels of corruption, can capitalize on their status as political outsiders and gain an electoral advantage from running on anti-corruption platforms. Study 2 offers no evidence that women candidates in Ukraine can capitalize on their status as political outsiders in a corrupt context. This finding supports the theoretical expectation that voter behavior only disciplines men and women candidates asymmetrically in countries where electoral accountability is high (Esarey and Schwindt-Bayer, Reference Esarey and Schwindt-Bayer2017).
Replicating Study 1, Study 2 also offers no evidence that men or women voters prefer men candidates. However, we do find evidence that voters’ preferences can disadvantage women candidates, albeit in indirect ways. Specifically, we find that voters prefer candidates who are married, have children, and have more political experience. The challenges of caring for a family and working in politics may disproportionately disincentivize women with families from running for office in the first place. The preference for candidates who have more experience may also work against women candidates, as the legacy of women's political exclusion from Ukrainian electoral politics means politicians with the most experience tend to be men.
Section 5 then discusses the implications of our findings, which are consistent with existing cross-sectional research that suggests women candidates do not face overt voter discrimination at the polls but rather face barriers to running for office in the first place (Thames, Reference Thames2018). Our findings also support recent scholarship regarding how indirect barriers, rather than overt discrimination, can limit women representatives’ legislative influence (Senk, Reference Senk2020). We highlight how political parties and elites reinforce women's legislative underrepresentation in both highly corrupt democracies as well as in non-democratic contexts. Section 6 concludes with suggestions for how to increase women's representation in Ukraine and other democracies characterized by high levels of corruption.
2. Women's political representation
Just and legitimate political decision-making requires women's representation. Justice requires that citizens are included in political practices—such as voting, representing, or deliberating—to influence collective decisions that affect them (Warren, Reference Warren2017; Beauvais, Reference Beauvais2018). There is a well-theorized link between inclusion and democratic legitimacy (Fung, Reference Fung2013; Warren, Reference Warren2017). As Mansbridge (Reference Mansbridge1999, p. 651) explains, when women candidates descriptively and substantively represent women, they remind citizens that they are able to govern. The feeling of inclusion created through representation increases democratic legitimacy.
Substantial evidence shows that empowering women's political representation improves policy-making processes: women representatives tend to be more congenial, more cooperative, and less hierarchical in decision-making than men (Tolleson-Rinehart, Reference Tolleson-Rinehart1991). Women decision-makers are also more likely to engage in cross-party cooperation to promote women's interests (Swers, Reference Swers2002). Moreover, women's political representation has symbolic, role model effects that empower women citizens (Pitkin, Reference Pitkin1967; Burrell, Reference Burrell1996; Lawless, Reference Lawless2015). For instance, women representatives’ presence can boost women voters’ political efficacy (Alexander and Andersen, Reference Alexander and Andersen1993; Atkeson, Reference Atkeson2003; Reingold and Harrell, Reference Reingold and Harrell2010), political interest (Hansen, Reference Hansen1997), and knowledge of legislators’ substantive records (Jones, Reference Jones2014).
With respect to policy outcomes, women representatives tend to promote policies that disproportionately impact women (Thomas, Reference Thomas1994; Burrell, Reference Burrell1996; Bratton, Reference Bratton2005; Gerrity et al., Reference Gerrity, Osborn and Mendez2007). There is also an intriguing relationship between the presence of women representatives and corruption (Swamy et al., Reference Swamy, Knack, Lee and Azfar2001; Dollar et al., Reference Dollar, Fisman and Gatti2001). However, closer analysis reveals that there is only a link between women representatives and corruption in contexts with high electoral accountability (Esarey and Schwindt-Bayer, Reference Esarey and Schwindt-Bayer2017). In countries where, for instance, corruption is the norm and press freedom is not respected, no association exists between women representatives's presence and corruption. Corroborating research from the United Kingdom, a liberal democracy with high electoral accountability, suggests that the relationship between women representatives and corruption in contexts with high electoral accountability may be driven more by women candidates’ greater risk aversion than by voters holding women candidates to higher standards, since, on average, voters punish all candidates similarly for corruption (Eggers et al., Reference Eggers, Vivyan and Wagner2018).
While women's equal political representation is clearly a laudable goal, without quotas women tend to be underrepresented in electoral politics. Both external barriers—such as overt voter discrimination and unequal access to resources such campaign funds and social networks—and more subtle internal processes—such as cultural norms and stereotypical expectations—can function to exclude women from political practices (Young, Reference Young2000). There is evidence that both explicit and implicit attitudes (Mo, Reference Mo2015) and stereotype activation (Bauer, Reference Bauer2014) shape political behavior, raising concerns that women candidates may face overt voter discrimination. Considering the Soviet regime's legacy, many scholars point to patriarchal attitudes and voter discrimination to explain women's political underrepresentation in Ukraine (Moser and Scheiner, Reference Moser and Scheiner2012; Connolly and Ó Beacháin Stefańczak, Reference Connolly and Ó Beacháin Stefańczak2015).
2.1 Women candidates in electoral democracies with high levels of corruption
A great deal of research considers how voter biases impact women candidates’ electoral success in democracies with relatively lower levels of corruption such as the United States or United Kingdom. However, less work considers how voter biases impact women candidates’ electoral success in democracies where higher levels of corruption may create distinct gender-based opportunities and constraints. We are interested in the problem of representation in what can be thought of as “corrupt” or “electoral” democracies.Footnote 1
Ukraine offers an especially good case to examine whether voters discriminate against women candidates in democracies with high levels of corruption. In 2014, the year before Study 1 began, the Global Gender Gap Report ranked Ukraine 105 out of 142 countries in terms of women's political empowerment and 118 out of 142 countries in terms of the number of women in parliament.Footnote 2 Even compared to other democracies with high levels of corruption, Ukraine has a poor track record when it comes to women's political representation. If voter discrimination is driving women's underrepresentation in democracies with high levels of corruption, we should expect to find this mechanism at work in Ukraine.
Women's underrepresentation in Ukraine may be surprising to some, given Ukraine's history as a former Soviet state. After all, the Soviet Union espoused an official doctrine of gender equality, and in the early days of the foundling communist state, efforts were made to enhance women's social and political status (Racioppi and See, Reference Racioppi and See1995; Sundstrom, Reference Sundstrom2010). After the 1917 revolution, Bolshevik revolutionaries created the Zhenotdel (women's department) in the Communist Party. Inspired by Marx and Engels's writings, the Bolshevik revolutionaries legalized abortion and modernized family law to promote gender equality. However, these idealistic efforts were short lived. In 1930, Stalin declared the woman problem “solved” and closed the Zhenotdel. During this period, “feminist views were silenced” (Racioppi and See, Reference Racioppi and See1995, p. 821) and independent women's organizations were banned across Soviet states (Stockemer, Reference Stockemer2007).Footnote 3 Despite officially supporting gender equality, the Communist Party also promoted pro-natalist policies and actively reinforced the stereotypical view that women should be mothers and caregivers (Thames, Reference Thames2018).
Furthermore, even though quotas ensured women enjoyed relatively high parliamentary representation—reserving approximately one-third of national deputy's seats for women—women ultimately exerted little political influence in Soviet regimes. The quotas were mostly symbolic, as the legislatures had little power (Stockemer, Reference Stockemer2007, p. 479). In Ukraine specifically, quotas ensured that women held not only a third of the seats in Ukraine's Verkhovna Rada (parliament), quotas also guaranteed that women held approximately half of the seats in the oblast and local councils (Hrycak, Reference Hrycak2007). Despite this token representation, most women remained in minor offices and had little authority (Racioppi and See, Reference Racioppi and See1995; Sundstrom, Reference Sundstrom2010).
After the Soviet Union's collapse, post-communist states in Eurasia abolished quotas and women's representation in these countries dropped to just below 8 percent by 1995 (Matland and Montgomery, Reference Matland and Montgomery2003). As shown in Figure 1, women's representation in many democratic post-communist countries, such as Georgia and Ukraine, has lagged.Footnote 4
Data are scraped from The Inter-Parliamentary Union's (IPU) website and formatted into time-series cross-sectional data.
2.2 Explaining women's political underrepresentation in democracies with high levels of corruption
Most scholarship on women's underrepresentation in democracies with endemic corruption such as Ukraine, assumes there is “societal resistance to the idea of women as political leaders” (Moser and Scheiner, Reference Moser and Scheiner2012, p. 210). Some have suggested that the rejection of Western feminism has hampered efforts to achieve gender equality in countries such as Ukraine (Jaquette and Wolchik, Reference Jaquette and Wolchik1998; LaFont, Reference LaFont2001; Funk and Mueller, Reference Funk and Mueller2018), but most point to patriarchal gender norms and voter prejudice as the cause of women's electoral troubles (Moser and Scheiner, Reference Moser and Scheiner2012; Connolly and Ó Beacháin Stefańczak, Reference Connolly and Ó Beacháin Stefańczak2015). However, none of these studies directly test whether voter prejudice prevents women from successfully competing for seats.
The exception to this is Thames (Reference Thames2018), whose observational analysis of the impact of post-communist Ukraine's electoral systems on women's political underrepresentation represents the only effort to test whether explicit prejudice prevents women from winning seats in Ukraine. Thames (Reference Thames2018) finds that institutions and parties appear to be the main culprits behind women's legislative underrepresentation. In Ukraine, women were more likely to be nominated and win seats under proportional representation (PR) than in single-member district (SMD) elections. There is also clear evidence that political parties’ behavior is a key factor in women's legislative underrepresentation, exemplified by the way parties place women lower than men on party lists under PR and are especially unlikely to nominate women candidates under SMD elections. However, Thames's analysis of election results reveals no evidence that women candidates are discriminated against at the polls under any electoral system.
The null effect of voter discrimination mirrors findings from liberal democracies. In scholarship from the United States and other democracies with relatively lower levels of political corruption, “overt discrimination has fallen out of favor as an explanation for women's absence from electoral politics” (Lawless, Reference Lawless2015, p. 352). A growing body of evidence shows that when women run for US Congress they win at the same rates as men (e.g., Carroll, Reference Carroll1994; Cook, Reference Cook1998; Lawless and Pearson, Reference Lawless and Pearson2008; Anastasopoulos, Reference Anastasopoulos2016). Research from other democracies with low levels of political corruption also confirms that women do not suffer an electoral penalty (Sevi et al., Reference Sevi, Arel-Bundock and Blais2019). The tendency for parties to place women candidates in more competitive races (Thomas and Bodet, Reference Thomas and Bodet2013) and a gender gap in political efficacy and motivation (Lawless and Fox, Reference Lawless and Fox2005, Reference Lawless and Fox2010) can explain women's political underrepresentation in democracies with low levels of political corruption.
As we explained, Ukraine offers an especially good case to examine gendered electoral dynamics in democracies with high levels of political corruption, since Ukraine has both high levels of corruption and exceptionally low levels of women's representation. If voter discrimination is driving women's underrepresentation in highly corrupt contexts, or if women candidates are able to capitalize on their position as political outsiders to overcome voter discrimination, we should expect to find these mechanisms at work in Ukraine. This leads us to our first hypothesis:
H1(a) On average, subjects will react more favorably to men candidates than to women candidates.
In studies of voter preferences for social group members, it is also important to test whether support for a social group member is conditional on one's status as an in- or out-group member. With respect to gender, evidence suggests that women voters are more supportive of women candidates and may even be willing to cross party lines to support women candidates (Brians, Reference Brians2005). As such, we test the hypothesis that:
H1(b) The relationship between a candidate's gender and voter preference is conditional on subjects’ gender: men subjects will react more favorably toward men candidates than women subjects.
Studying the dynamics of voter behavior in Ukraine also lends insight into whether what counts as a gendered policy strength within corrupt contexts can indeed increase support for women candidates. Voters develop trait stereotypes of politicians’ gender (Huddy and Terkildsen, Reference Huddy and Terkildsen1993), and research on the United States reveals that women are associated with traits such as “compassion” (Shapiro and Mahajan, Reference Shapiro and Mahajan1986; Kahn, Reference Kahn1996; Herrnson et al., Reference Herrnson, Lay and Stokes2003). In the United States, women candidates who stress their strengths on issues that confirm gender trait stereotypes (often, compassion-related issues such as child or health care) garner more electoral support.
Less is known about whether women candidates in democracies with high levels of corruption can capitalize on gender dynamics by running on issues that are salient in corrupt contexts. Some research shows that women are viewed as less corrupt than men (Dolan, Reference Dolan2010, p. 72). Evidence from Latin America also suggests that women can capitalize on their positions as political outsiders to win votes (Morgan and Buice, Reference Morgan and Buice2013, p. 660). Funk et al. (Reference Funk, Hinojosa and Piscopo2021) find a bivariate association between the perception that women fight corruption and the percent of women nominated to party lists. Therefore, anti-corruption platforms may be a gendered policy strength in corrupt contexts, and running on an anti-corruption platform may disproportionately benefit women candidates.
However, a noteworthy negative correlation between the presence of women representatives and corruption exists—but only in contexts characterized by high electoral accountability (liberal democracies) (Esarey and Schwindt-Bayer, Reference Esarey and Schwindt-Bayer2017). In contrast, no such association is present in low electoral accountability contexts. Esarey and Schwindt-Bayer (Reference Esarey and Schwindt-Bayer2017) theorize that, because women are more risk averse than men and because voters hold women voters to higher standards in the polls, women will be less likely to engage in corrupt behavior in high-accountability systems. Eggers et al. (Reference Eggers, Vivyan and Wagner2018) use an experiment conducted in the UK to offer insight into the micro-level mechanisms of differential treatment by voters. The authors show that—although voters punish corrupt candidates similarly, regardless of candidate gender—women voters do react more positively to women candidates’ behavior, disproportionately rewarding good behavior by women candidates.
To date, the micro-level mechanisms of differential treatment by voters have only been studied in high-accountability, liberal democracies, and it is unclear how these dynamics function in highly corrupt, low-accountability democracies. Ukraine provides a useful case to identify whether anti-corruption platforms can increase support for women candidates in highly corrupt democracies. At least until Russia's 2022 full-scale invasion, Ukraine was a highly corrupt democracy, consistently ranking worse than 100th place, out of all country's ranked by Transparency International's Corruption Perception Index.Footnote 5 To identify whether contexts with high levels of political corruption shape voters’ willingness to vote for women candidates running on different platforms in similar ways to contexts with low levels of political corruption, we test the hypothesis that:
H2(a) On average, subjects will react more favorably to women candidates running on anti-corruption platforms.
Following the existing literature from contexts with low levels of political corruption (i.e., the UK (Eggers et al., Reference Eggers, Vivyan and Wagner2018)) we additionally test the hypothesis that:
H2(b) The relationship between a candidate's gender and policy issues is conditional on the subject's gender: women subjects will respond most favorably to women candidates running on anti-corruption platforms.
3. Study 1
3.1 Methods
In collaboration with the National Democratic Institute (NDI), in Study 1, we used vignette experiments embedded in subsamples of two nationally representative surveysFootnote 6 conducted in Ukraine to test the hypothesis that voters prefer men to women candidates, and to identify whether voter prejudice is conditional on voter gender.Footnote 7 For Study 1, data were collected from two independent (non-panel) survey waves. Wave 1 of Study 1 was carried out face-to-face using tablet devices at the end of 2015. Our experiment was embedded in a national survey. However, because our experiment required the use of tablet devices and sufficient tablets did not exist for the firm to survey the entire nationally representative sample on tablets (the remaining surveys were paper-based), Wave 1 of Study 1 relied on a representative sample drawn from the Ukrainian capital Kyiv (n = 466).
To further validate our findings, the experiment was replicated six months later, in the summer of 2016 (Wave 2 of Study 1). Again, because sufficient tablets did not exist, our experiment was primarily carried out on a representative sample from Kyiv (n = 337). However, there were sufficient tablets to deploy to some locations outside of Kyiv. To address the concern that the more cosmopolitan electorate of a capital city may be less biased toward women candidates than voters in rural areas and clarify whether our results would replicate in rural settings, we ran the experiment in 137 rural voting precincts where we sampled 594 respondents in Wave 2 of Study 1. The results from our rural sample are broadly similar to the Kyiv samples (see Appendix G.3).
In both waves of Study 1, subjects were randomly assigned to receive either a “man candidate treatment” or a “woman candidate treatment.” The experimental treatment was the exposure to a set of three pictures of fictional candidates in randomized order (either three fictional women or three fictional men candidates).
We asked each respondent to rate only one gender (to rate three pictures of men, or three pictures of women). We used three different images for each gender to prevent respondents from anchoring their ratings on the gender of the first image they saw. Because we are comparing respondents’ evaluations of sets of pictures of women and men candidates, it was important that the fictional candidates in each set were otherwise comparable. Each woman/man pair was approximately the same age, had similar skin tone denoting they were likely of Slavic origin (and White), and had the same hair color (with the exception of the oldest pair, where the man was bald and the woman candidate had dyed hair).Footnote 8 While research in the United States shows that some perceptions of women are conditional on party (e.g., Democratic women candidates are perceived as being more left-wing than Democratic men candidates) (Dolan, Reference Dolan2004), parties often change in Ukraine (D'Anieri, Reference D'Anieri2007, pp. 168-71), and many Ukrainians do not identify with any party. Moreover, parties are weakly ideological (Rovný, Reference Rovný2015; Way, Reference Way2015) and tend to be personalistic. Therefore, we do not include party in our analysis. We compared each matched pair of respondents’ similarities on a range of characteristics, including attractiveness, using the face++ machine learning API (see Appendix D.1). The fictional candidates were photographed against the same plain background and the images were cropped, so that the fictional candidates were the same size. The fictional candidates were also wearing the same type of clothes.Footnote 9 Figure 2 displays the fictional candidates’ pictures.
The three sets of pictures of women and man candidates.
The candidates in the vignettes were introduced as lawyers who decided to get involved in politics to fight corruption. Law is a typical professional background for many Ukrainian politicians. Telling respondents the profession of the candidate helped address a potential confound: that women and men come from different professional backgrounds and could have different relevant experience, which, in turn, could affect candidate evaluation. Fighting corruption is also a common platform across the political spectrum (Huss, Reference Huss2020).Footnote 10 For each of the three pictures of candidates, we told the respondent that:
This person has said “I am a lawyer, but I decided to get involved in politics to fight corruption.” On a scale of 0–10, zero being “I would never vote for this person” and ten being “I would absolutely vote for this person,” how likely would you be to vote for this person?
In our main analysis, we operationalize Y as the average of all three ratings. That is, the average treatment effect is measured by taking the difference between the average ratings for women candidates and men candidates. In the regression models used as robustness checks, we control for the order of picture exposure and a vector of demographic controls, which include age, age2, which captures the non-linear effect of age, the gender of the respondent respondent (woman) = 1, 0 otherwise), whether the respondent identifies as ethnically Ukrainian (ethnic ukrainian = 1, 0 otherwise), whether the respondent completed the survey in Ukrainian (survey in ukrainian = 1, 0 in Russian), and whether the respondent has greater than high school education (High+ = 1, 0 otherwise). Descriptive statistics and balance tables are presented in Appendix E.
3.2 Results
Study 1 allows us to test H1(a), whether subjects react more favorably to men candidates than women candidates on average, and H1(b), whether voter prejudice is conditional on voter gender, with men voters reacting more favorably to men candidates than women voters. As shown in Figure 3, our results offer no evidence for H1(a) in either wave of data collection. Operationalizing our outcome variable as respondents’ average willingness to vote for all three of the candidates who they rated, there is no statistical difference in either the distributions (using boot-strapped Kolmogorov–Smirnov tests (Sekhon, Reference Sekhon2011)) or the mean evaluations of men and women candidates (using t-tests) at the traditional (p < 0.05) level of significance.Footnote 11 Robustness checks using regression-based analysis also reach the same conclusions (see Appendices G.1 and G.2).Footnote 12 There is little evidence that voter prejudice is driving the immense and enduring gap in women's legislative representation in Ukraine.Footnote 13
The distribution of respondents’ self-reported willingness to vote for candidates Waves 1 and 2. The dark bars and dashed lines show the distribution of responses and mean scores for the women candidate treatments. The light bars and dotted lines show the distribution of responses and mean scores for the men candidate treatments. A * denotes a statistically significant difference in means between the treatment and control at 95 percent confidence levels.
With respect to H1(b), Study 1 also reveals no differential effect by subject gender. As shown in Figure 4, there is no evidence that either women or men voters rate anti-corruption women candidates more highly than otherwise similar anti-corruption men candidates.
Marginal effects of treatment by respondent gender.
While our simple design allows us to identify, ceteris paribus, whether women and men voters evaluate men candidates running on anti-corruption platforms more positively than women candidates running on the same platform, our vignette experiments only offer a unidimensional test of H1(a) and H1(b). The experiments in Study 1 do not vary the candidates’ platforms. Thus, while intriguing, these vignette experiments stop short of testing H2(a), that subjects will react more favorably to women candidates running on anti-corruption platforms, and H2(b), that the relationship between candidate gender and policy issues is conditional on subject gender. Furthermore, by holding possible confounding attributes—such as policy platform, candidate experience, and ethnicity—constant, we also reduce the complexity of the decision-making process that voters normally face, which has consequences for the external validity of our findings.
4. Study 2: conjoint analysis
To address Study 1's limitations and to test our remaining hypotheses, we ran a conjoint experiment on another wave of survey data collected by NDI in 2020. Conjoint analyses are excellent causal tools for testing hypotheses about decision-making because they are more realistic in information-rich environments such as elections (Hainmueller et al., Reference Hainmueller, Hopkins and Yamamoto2014; Hainmueller and Hopkins, Reference Hainmueller and Hopkins2015), and they can reduce social desirability bias (Horiuchi et al., Reference Horiuchi, Markovich and Yamamoto2021). Conjoint analyses also allow researchers to test multiple causal hypotheses and estimate the effects of multiple treatment components on a single behavioral outcome, thus going beyond unidimensional tests of one or two hypotheses (Hainmueller et al., Reference Hainmueller, Hopkins and Yamamoto2014).
4.1 Methods
Study 2 asked a sample of Ukrainian citizens to decide whether they would vote for sets of hypothetical candidates running in an election. Unlike Study 1, survey responses were not collected via a face-to-face interviewer using a tablet. Because data for Study 2 were collected during the COVID-19 pandemic, a sample (n = 2155) was recruited from the Info Sapiens opt-in panel to participate in an online survey via their personal devices between 20–24 July 2020.Footnote 14 One additional advantage of conducting the survey online is that it ruled out the possibility of interviewer effects.
Subjects were presented with a pair of candidates and a list of information for each candidate, and were told “Imagine that these candidates are from the party that you would consider supporting in local elections.” Each respondent saw four sets of paired profiles for a total of eight profiles (n = 17, 240). Respondents received information about each candidate, including the candidate's name (which cued gender and ethnicity)Footnote 15 As seen in Table 1, respondents saw five additional attributes that varied randomly across pairings. The attributes included one of five policy priority issues on which candidates commonly campaigned and were major priority areas for the Ukrainian electorate (education, health care, economy, security, or fighting corruption). In the Ukrainian context, individual candidates have significant leeway in setting their own campaign agenda, making this variation in priorities ecologically valid in the Ukrainian context.Footnote 16 Other attributes included the candidate's age (25, 35, or 55), marital status (married, divorced, or single), children (none or has two children), and political experience (none, 5 years, or 15 years experience).Footnote 17 With respect to the randomization of the attributes that varied across pairings, we imposed one restriction to rule out implausible profiles (Hainmueller et al., Reference Hainmueller, Hopkins and Yamamoto2014). Namely, we specified that 25- or 35-year-old candidates could not have 15 years of experience, as this combination is implausible. As such, we are careful to interpret the AMCE as being defined for only a subset of the linked values. Specifically, the causal effects of 15 years of experience are defined only for candidate profiles that include 55-year-old candidates.
Attributes for candidate profiles in conjoint experiment.
With respect to cuing gender, respondents were randomly assigned to receive either a woman's name or a man's name. Both the Ukrainian and Russian languages contain grammatical genders, and naming conventions are also gendered. It would thus be clear to readers whether a fictional candidate is a woman or a man based on their name. With respect to cuing ethnicity, following Frye (Reference Frye2015), subjects were randomly assigned to receive both first and last names that were clearly Ukrainian- or Russian-sounding. Our partners at NDI also suggested including names they believed were neutral-sounding.Footnote 18 Instead of limiting our analysis to only one or two features (such as gender and policy priority issues), our experimental design allows us to vary many candidate attributes simultaneously and to evaluate which attributes make candidates more appealing to voters.
After reviewing the candidate profiles, respondents were asked “Would you vote for these candidates?” (yes = 1). We note that this is not a forced choice conjoint in that respondents could say they would vote for one of the candidates, both of the candidates, or neither. This is more plausible in Ukraine's mixed member electoral system with non-mandatory voting, where electors have the choice of casting multiple votes (for candidates and parties) or none at all. Furthermore, Hainmueller et al. (Reference Hainmueller, Hangartner and Yamamoto2015) find this paired conjoint experiment design comes closest to achieving behavioral benchmarks in the context of immigration policy.Footnote 19
We estimated the corresponding AMCEs by regressing this outcome variable on dummy variables indicating each attribute value. For the two attributes that involved randomization restrictions (candidates ages 25 and 35), we use a conditionally independent randomization, and for the remaining attributes we have a fully independent randomization (see Hainmueller et al., Reference Hainmueller, Hopkins and Yamamoto2014, for a thorough explanation of the assumptions involved in conjoint analysis). In total, there are 15 AMCE estimates, with five attribute levels used as reference categories (t 0). The AMCEs represent the expected change in probability of considering voting for a candidate profile when a given attribute value is compared to the reference category.Footnote 20
4.2 Results
Figure 5 column 1 shows the AMCEs and 95 percent confidence intervals for each attribute value. As shown, voters do not choose candidates with men's names more often than those with women's names. In fact, the AMCE for women's names is statistically significantly higher, which would indicate a preference for women's names. Note, however, that this effect is substantively small, representing between a two and three percentage point increase in the probability of willingness to vote for a candidate when the candidate is a woman. Regardless, this finding runs contrary to the expectation that voters prefer men candidates. We interpret this as reinforcing the findings from Study 1, that voters do not overtly discriminate against women candidates.
AMCE estimates for vote (unconditional and conditional on respondent gender).
Ukrainian voters do have a clear preference for experience. We find that on average, voters express an 11 percentage point higher probability of stating they will vote for candidates with five years of experience, relative to a candidate with no experience. Recall that the attribute of having 15 years experience is conditional on a candidate's profile indicating that the candidate is 55 years of age. As such, the AMCE for 15 years experience should be interpreted to mean that, for candidates aged 55 years, voters express a 16 percentage point higher probability of stating they will vote for candidates with 15 years experience relative to a candidate with no experience who is 55.
The next important attribute explaining vote choice is prioritizing the issue of corruption. We find that, on average, voters express a seven percentage point higher probability of stating they will vote for candidates who say they prioritize fighting corruption relative to education. On average, having children, and prioritizing either healthcare or the economy (relative to education) also motivates vote choice in Ukrainian elections, but these effects are substantively smaller. It should be noted that, although the value and significance of an AMCE changes depending on the base category (a drawback of interpreting the results of any conjoint using AMCEs (Leeper et al., Reference Leeper, Hobolt and Tilley2019)), the AMCE for fighting corruption is always significant regardless of which issue is chosen as the baseline. In terms of marginal means, which Leeper et al. (Reference Leeper, Hobolt and Tilley2019) recommend reporting, the highest percentage of respondents say they will vote for candidates who prioritize fighting corruption (42 percent) compared to education (34 percent), security (33 percent), healthcare (38 percent), and economy (38 percent).Footnote 21
The AMCEs conditional on gender, shown in columns 2 and 3 of Figure 5, show that Ukrainian women voters do have slightly different preferences than men voters.Footnote 22 Indeed, Figure 5 shows that congruent with public opinion research from democracies with relatively lower levels of political corruption (Schaffner, Reference Schaffner2005; Shorrocks and Grasso, Reference Shorrocks and Grasso2020), candidates focusing on social issues—healthcare and educationFootnote 23—do significantly better among Ukrainian women voters as compared to men voters. Although all subjects are more likely to vote for candidates with children (relative to no children) and married candidates (relative to single candidates), the bump that married candidates or candidates with children get is larger (and statistically significant) among Ukrainian women. Similarly, although all subjects are more likely to vote for candidates 55 years of age with 15 years experience, older candidates with the most experience do even better among Ukrainian women than men voters at statistically significant levels.
The finding that voters prefer candidates who are married and have children is particularly noteworthy for a gender-based analysis. Although it is not one of our primary hypotheses, we conducted an exploratory analysis to identify whether voters particularly like women candidates who have children or are married relative to comparable men candidates. The results of this exploratory analysis reveal that, in fact, voters do reward women candidates who have children more than comparable men candidates (although they do not reward married women more than comparable married men candidates) (Appendix I.7).
To answer H2(a), that subjects will react more favorably to women candidates running on anti-corruption platforms, we examine the difference in marginal means between the mens’ and womens’ names on the policy issue area of corruption. We find that the difference between a woman candidate who says she is fighting corruption relative to a man candidate with the same policy stance, while slightly positive is not statistically different from zero (p = 0.07). Our study suggests that there is an electoral benefit for candidates who take an anti-corruption stance, but this benefit does not appear to be conditional on candidate gender.
To test H2(b), that the relationship between candidate gender and policy issues is conditional on subject gender—with women subjects responding most favorably to women candidates running on anti-corruption platforms—we subset the data and then examine the relationship between issue area (corruption) and candidate gender with only women respondents. Again, we find no statistically significant (p = .16) or substantively important change in the probability of women respondents’ reporting they would vote for a candidate when the candidate is a woman (relative to being a man), when both candidates say they would fight corruption. The patterns are similar in the full sample. Contrary to countries where there is high electoral accountability (Eggers et al., Reference Eggers, Vivyan and Wagner2018), in democracies with high levels of corruption—such as Ukraine, where there is low electoral accountability—there is no evidence that women voters (or any voters) react more positively to women candidates’ good behavior.
5. Discussion
Our findings speak to two important lines of research on gender and representation. First, the puzzle over why women are underrepresented in legislatures in countries with high levels of corruption. Do women voters face higher levels of voter prejudice, or are other mechanisms at play? Ours is the first study to test this question using an experimental design. Confirming Thames's observational (Reference Thames2018) findings, we found no evidence that Ukrainian voters prefer men candidates in our two vignette experiments (conducted in 2015 and 2016, respectively), nor our third conjoint analysis (conducted in 2020). Our ability to replicate the finding that, ceteris paribus, voters have no preference for men candidates in three experiments conducted over a five-year period offers robust support for the conclusion that Ukrainian women candidates do not face overt voter discrimination at the polls.
Although, on average, voters are not prejudiced against women candidates, our conjoint analysis shows that voters’ preferences still disproportionately punish women candidates. The preference for candidates who have more experience works in favor of men because the legacy of women's political exclusion from Ukrainian electoral politics means that men candidates tend to be more experienced. Furthermore, voters do prefer candidates with children. This can be punishing for Ukrainian women who are interested in politics. Balancing a political career and care-giving duties is harder for women in general, and in Ukraine, the decision to forego a family and focus on a political career incurs an electoral penalty. Furthermore, voters reward women candidates with children even more than comparable men candidates. Insofar as sexism is at work in Ukrainian elections, it manifests in how voters reward women candidates who reinforce norms related to child care-giving.
The second important line of research on gender and representation in democratizing contexts that we contribute to is the debate over women's ability to capitalize on their status as political outsiders and gain an electoral advantage from running on anti-corruption platforms (Morgan and Buice, Reference Morgan and Buice2013). The hypothesis that voters in democracies with high levels of corruption reward women for being political outsiders runs contrary to the theory that voters only punish (or reward) men and women candidates asymmetrically for bad (or good) behavior in countries where electoral accountability is already high (Esarey and Schwindt-Bayer, Reference Esarey and Schwindt-Bayer2017). Eggers et al. (Reference Eggers, Vivyan and Wagner2018) build on Esarey and Schwindt-Bayer's (Reference Esarey and Schwindt-Bayer2017) work, using experimental evidence to show why there's a negative association between women's legislative presence and corruption in contexts with high levels of electoral accountability. Eggers et al. (Reference Eggers, Vivyan and Wagner2018) show that in the UK, women voters disproportionately reward women candidates who are not corrupt.
The results of our experiment are theoretically congruent with Eggers et al.'s (Reference Eggers, Vivyan and Wagner2018) results, even though our findings are different. Eggers et al.'s (Reference Eggers, Vivyan and Wagner2018) study was conducted in the UK, and so cannot speak to the null association between women's legislative representation and corruption in contexts where electoral accountability is low; by contrast, our experiments speak directly to this question. As we show, in Ukraine, a country with very low electoral accountability, women and men voters reward anti-corruption candidates in similar ways and this finding is not conditional on candidate gender. Just as Eggers et al.'s (Reference Eggers, Vivyan and Wagner2018) experiment supports Esarey and Schwindt-Bayer's (Reference Esarey and Schwindt-Bayer2017) finding that there is a correlation between women's legislative representation and lower corruption in contexts with high electoral accountability (by showing that women voters are especially likely to support anti-corruption women candidates), our experiments support Esarey and Schwindt-Bayer's (Reference Esarey and Schwindt-Bayer2017) finding that there is no association between women's legislative representation and corruption in contexts with low electoral accountability.
6. Conclusion
Our experimental research offers little evidence that women candidates in democracies with high levels of corruption face overt voter discrimination because of their gender. Given women's dramatic underrepresentation in elected office in Ukraine and other democracies characterized by high levels of corruption, this may come as a surprise to some readers. However, these findings are replicated over three samples collected in 2015, 2016, and 2020. Five years is a long time, especially in newer democracies where the political context can change rapidly. The fact that our findings are replicated over multiple samples across a relatively wide time frame offers strong evidence that Ukrainian voters do not directly discriminate against women candidates.
However, our conjoint analysis does show that Ukrainian voters do prefer candidates who conform to traditional roles in the sense of being married and having children. Women feel the consequences of this most poignantly, as parental status matters more for women candidates’ electability than men candidates’. Not only is it likely that the challenges of caring for a family and working in politics disproportionately disincentivizes women with families from seeking a nomination or running, but women without children face a disproportionate electoral penalty. Furthermore, voters’ preference for candidates who have more political experience may also indirectly work against women candidates, as the legacy of women's political exclusion from Ukrainian electoral politics means the most experienced politicians tend to be men.
In addition to norms that disincentivize mothers from running for office, women face other institutional barriers to office—notably, the difficulty they face breaking into the old party networks. Future studies should pay greater attention to the role that gatekeepers in clientelistic networks play in maintaining gender inequality in democracies with high levels of corruption. Rather than focusing narrowly on voters’ prejudices and preferences, attention should also be paid to the role that institutions and parties play in empowering—or blocking—women from running for office. Thames (Reference Thames2018) shows that electoral institutions matter: more women were nominated and elected under PR than SMD elections. It seems likely that in corrupt contexts, clientelistic politics worsen the gender gap. This speaks to a burgeoning literature (e.g., Senk, Reference Senk2020) on how the indirect effects of institutional barriers prevent women from having legislative influence.
After our studies were carried out, Russia launched a full-scale invasion of Ukraine on 24 February 2022. It is difficult to predict the gendered consequences for Ukraine at this moment. Many women have volunteered and are serving in active duty in the war. These women could serve as a potential pool of candidates who can leverage their wartime experience to serve as post-war political candidates. On the other hand, many women and children fled Ukraine (men under 60 were barred exit).
With respect to institutional reforms to increase women's representation, if it continues its pre-war trajectory, Ukraine is returning to a PR system (albeit with a substantively different set of rules than its previous PR system). This PR system will hopefully improve women's representation. Another institutional solution for promoting women's representation in Ukraine might be to revisit the use of quotas. Additionally, Hrycak (Reference Hrycak2007) suggests that deepening ties with the European Union (EU), a process hastened by Russia's invasion, may be a powerful way to increase gender equality, because EU membership rules require member countries to enact equality of opportunity legislation. Along the same lines, Sundstrom et al. (Reference Sundstrom, Sperling and Sayoglu2019, p. 27) advocate for enshrining international norms that recognize women's rights as human rights in multilateral treaties, as women's rights are an essential component of human rights. These laws may also be more likely, given Russia's invasion.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/psrm.2022.46. To obtain replication material for this article, please visit https://doi.org/10.7910/DVN/P9E3WZ.
Acknowledgements
We thank Vincent Arel-Bundock, Sarah Bush, Volodomyr Kulyk, Maria Popova, Dietlind Stolle, and attendees at the 2017 annual meetings of the Midwest Political Science Association (MPSA) and the Caucasus Research Resource Center's (CRRC) 2016 methodological conference for feedback.
Open access
