The end of an armed conflict usually marks the beginning of political competition that is characterized by substantially higher levels of representation of women (Tripp Reference Tripp2015). This postconflict rise in women’s political representation can partly be explained by international pressure on postconflict regimes for democratization that is frequently paired with calls for the institution of gender quotas (Anderson and Swiss Reference Anderson and Swiss2014). Electoral institutions and democratization alone, however, do not fully explain the rise in political representation of women in postconflict societies (Hughes and Tripp Reference Hughes and Tripp2015). Armed conflicts arguably result in more fundamental changes in social gender relations. War violence often disrupts traditional gender roles and creates opportunities for women to redefine their interests and positions in social and political life (Tripp et al. Reference Tripp, Casimiro, Kwesiga and Mungwa2009).
Whatever the reason behind higher levels of women’s political representation in postconflict societies, it is clear that the benefits of their greater participation in political life and decision-making are real and substantively important (Wängnerud Reference Wängnerud2009). Higher proportions of women in the national legislatures of postconflict societies prolong peace through improved governance and prioritization of social welfare spending (Shair-Rosenfield and Wood Reference Shair-Rosenfield and Wood2017). States with greater participation of women in national leadership are less likely to resort to violence in foreign policy (Brysk and Mehta Reference Brysk and Mehta2014; Caprioli and Boyer Reference Caprioli and Boyer2001). Higher female representation also leads to better welfare and social outcomes for women, children, and families, regardless of how democratic postconflict societies are (Paxton and Hughes Reference Paxton and Hughes2014; Viterna and Fallon Reference Viterna and Fallon2008; Wang Reference Wang2013).
Although we have robust evidence of improvements in the level of women’s descriptive representation after armed conflicts and the substantive policy consequences of such improvements, we lack reliable tests of the relationship between gender and voter choice in postconflict societies. Do voters in postconflict societies actually prefer female electoral candidates, or do they exhibit the same kind of gender bias as voters in societies without recent histories of armed conflict? What effect does exposure to war violence have on voters’ preferences? Do voters in communities disproportionally exposed to violence have a higher or lower propensity to support female candidates?
Recent evidence from postwar Bosnia and Herzegovina suggests that communal exposure to violence may be associated with both a higher supply of women candidates in postwar elections and a lower likelihood of women actually winning those electoral competitions (Hadzic and Tavits Reference Hadzic and Tavits2021). Although the elegance of this account has theoretical appeal, its evidentiary basis is limited because of data and methodological issues. Moreover, we know little about the role played by political ideology in this whole process. The literature on gender and politics in postconflict societies rarely engages explicitly with the question of ideology, even though armed conflicts are often defined by ideological differences that can have a discernible gender dimension.
We answer these questions by analyzing a unique data set containing information on nearly 7,000 candidates running for parliamentary seats in three elections conducted under proportional representation (PR) rules with preferential voting in Croatia. We combine these candidate-level data with the ideological profiles of party lists, as well as with detailed electoral, socioeconomic, and demographic data on Croatia’s more than 500 relatively small municipalities that include statistics on the local population’s exposure to violence in the Croatian 1991–95 War of Independence. We supplement this aggregate-level analysis with an analysis of survey data collected from more than 2,000 Croatian respondents in the middle of the three electoral cycles under scrutiny.
In our view, Croatia is an excellent case for three reasons. First, its War of Independence belongs to an unfortunately sizeable class of post–Cold War conflicts, making our findings significant beyond the country’s borders. Second, its elections—particularly during the period of our interest—have been free and fair, allowing us to tap into voter choice without having to account for corrupt electoral practices, which are often present in postconflict polities. And third, the quality and granularity of the Croatian data, together with electoral rules featuring preferential voting, offer the perfect institutional setting in which to answer our research questions.
Our analysis convincingly demonstrates not only that Croatian voters exhibit a mild level of gender bias against female candidates, but that this bias is conditional on the local electorate’s ideology and exposure to war violence. Generally speaking, voters of right-wing parties are more biased against female candidates, as are voters in areas more affected by war violence. These findings are echoed in our analysis of survey data, where right-wing respondents and those who were more traumatized by the war are more likely to subscribe to traditional gender roles that would keep women at home and out of public life. Crucially, however, the effects of ideology and exposure to war violence on the electorate’s gender bias exhibit a strong interactive relationship, suggesting that bias against women in politics is strongest among right-wing voters in areas exposed to war violence and reversed among left-wing voters in areas exposed to war violence. We argue that this is likely the case because the classic dichotomy between security and social policy is amplified in areas more affected by war violence, where the needs for security and social services are greater than in areas that avoided destruction. Our findings highlight the long-lasting effects of armed conflict on politics and gender, as well as the need to better understand the relationship between gender, political ideology, and the history of violence in postconflict societies.
Conflict, Women, and Voter Bias
A higher level of representation of women in postconflict societies is an empirical fact (Tripp et al. Reference Tripp, Casimiro, Kwesiga and Mungwa2009). The question, however, is how to explain the causal mechanism behind it. The primary suspect is the introduction of gender quotas that often accompanies the signing of peace agreements, as well as the transition to more proportional electoral systems that is often part of institutional settlements for postconflict political competition (Anderson and Swiss Reference Anderson and Swiss2014; Tripp and Kang Reference Tripp and Kang2008). As much as electoral institutions may have an impact on women’s political representation—and in many contexts, they do—this only pushes the task of explaining the causal mechanism behind the observed trend of women’s higher representation down the road, because then we are left with explaining where electoral institutions such as gender quotas or PR rules come from.
The problem is compounded by the fact that neither electoral institutions nor the overall process of democratization alone can fully explain the rise in women’s legislative representation in societies coming out of violent conflict. Quantitative studies suggest that there is something inherent in the end of particularly violent conflicts—above what can be attributed to advantageous institutional solutions—that leads to higher participation and success of women in electoral competitions (Hughes and Tripp Reference Hughes and Tripp2015). What that something is, however, remains an open question. The answer may lie in the disrupted gender roles and the new opportunities for women in postconflict societies to take on leadership positions in various spheres of public life (Waylen Reference Waylen2007), or in the organizing efforts that many women undertake in societies recovering from violent conflict (Tripp Reference Tripp2016). Indeed, these organizing efforts—together with assistance from transnational and international actors—have been shown to be of critical importance in ensuring needed changes in institutions and the actual electoral success of female candidates in postconflict societies as diverse as Croatia (Irvine Reference Irvine2007), Rwanda, and Afghanistan (Tajali Reference Tajali2013).
We still do not know, however, what role voters play in this trend of increased representation of women in postconflict societies. Do voters in postconflict societies actually prefer female politicians to their male counterparts? The body of evidence for possible gender bias in voter choice in democratic societies not recovering from violent conflict is mixed and highly dependent on the electoral context and research methods employed. In consolidated Western democracies, it is difficult to say whether candidate gender has any impact on voter choice independent of other factors. For example, Danish and Finish voters exhibit, if anything, a slight pro-women bias (Blom-Hansen et al. Reference Blom-Hansen, Elklit, Serritzlew and Villadsen2016; Von Schoultz and Papageorgiou Reference Von Schoultz and Papageorgiou2021). Candidate gender seems not to matter at all for voters in Switzerland and Ireland (Lutz Reference Lutz2010; McElroy and Marsh Reference McElroy and Marsh2010). Voters are also not considered as directly culpable for the abysmally low levels of women’s political representation in the United States, partly on account of survey evidence suggesting candidate gender plays no role in voter decision-making (Dolan Reference Dolan2004, Reference Dolan2014).
However, experimental evidence from nonpartisan elections, as well as evidence from local elections and congressional elections when candidate quality is appropriately accounted for in the U.S. context, suggests that American voters do seem to be slightly biased against female candidates (Badas and Stauffer Reference Badas and Stauffer2019; Brockington Reference Brockington2003; Fulton Reference Fulton2014). The situation is similar in consolidating democracies like the Czech Republic, Estonia, Poland, and Indonesia, where voters have been found to harbor a general preference for male candidates (Allik Reference Allik2015; Dettman, Pepinsky, and Pierskalla Reference Dettman, Pepinsky and Pierskalla2017; Górecki and Kukołowicz Reference Górecki and Kukołowicz2014; Jurajda and Münich Reference Jurajda and Münich2014). Voters are not the only culprits in low levels of women’s representation, but often share the blame with the media and the leadership of political parties. The media commonly exhibits bias against female candidates even in Western democracies (e.g., Kahn Reference Kahn1994). And the leaders of political parties often do not give enough campaign resources to female candidates (Eder, Jenny, and Müller Reference Eder, Jenny and Müller2015) or do not place women high enough on electoral lists to be successful (Tavits Reference Tavits2010). In fact, once list placement, campaign resources, and media exposure are controlled for, voter bias against female candidates may be even nonexistent (Wauters, Weekers, and Maddens Reference Wauters, Weekers and Maddens2010).
While evidence of a general voter gender bias may be mixed, there is growing and convincing evidence that voters do use candidate gender as a heuristic device to make inferences about candidate policy positions and competences, personality traits, and ideological beliefs. When it comes to personality traits, women are stereotypically perceived as more sensitive, sympathetic, and caregiving than men (Bauer Reference Bauer2015; Johns and Shephard Reference Johns and Shephard2007). These traits can present serious problems for female candidates, particularly in low-information elections, because being sensitive, sympathetic, and caregiving is in conflict with the stereotypical voter preference for electoral candidates who are more aggressive and decisive (Huddy and Terkildsen Reference Huddy and Terkildsen1993). Emphasizing these traditional gender traits in campaigns is especially challenging for female candidates of conservative parties, whose voters are more likely to subscribe to both the stereotypical candidate preferences and views of female politicians (Bauer Reference Bauer2020).
Perhaps unsurprisingly, considering these gender trait stereotypes, female candidates are also perceived—often mistakenly—not only to be more left-wing than their male counterparts (Devroe and Wauters Reference Devroe and Wauters2018; McDermott Reference McDermott1997), but also to hold policy competences in areas traditionally considered to be both more left-wing and more “feminine,” such as social services, health, education, and culture. These stereotypes are so pervasive that even in countries like Sweden, which is commonly considered as about as free of gender bias in politics as possible, the advancement of female politicians is severely hampered if their policy competences are not congruent with what is commonly perceived as the political domain of women (Baumann, Bäck, and Davidsson Reference Baumann, Bäck and Davidsson2019). This type of gender bias is particularly pronounced in security policy, as well as during times of security crises. Extensive research, mostly in the U.S. context, shows that when security issues are salient (e.g., during a terrorist threat), voters expect leadership from their political representatives and believe that men are more competent to deal with the security challenges at hand (Falk and Kenski Reference Falk and Kenski2006; Lawless Reference Lawless2004; Merolla and Zechmeister Reference Merolla and Zechmeister2009). This is especially detrimental for the electoral chances of female Democrats, whose gender and party ID seem to reinforce voters’ stereotypes (Holman, Merolla, and Zechmeister Reference Holman, Merolla and Zechmeister2016). In fact, experimental evidence from the United Kingdom shows that at times of intense security threats, even female leaders with experience in security policy—such as the former prime minister Theresa May—suffer from voter backlash, especially among those who already subscribe to negative stereotypes about women (Holman, Merolla, and Zechmeister Reference Holman, Merolla and Zechmeister2022). Similarly, a country-level examination of women’s representation and international relations of democratic states from 1981 to 2007 suggests that the presence of an external security rival significantly decreases the level of female parliamentary representation, possibly because of the militarization of the society that finds itself under threat (Schroeder Reference Schroeder2017).
The evidence outside the context of Western democracies and particularly from postconflict societies is limited, but it suggests that a similar dynamic is likely at play. Experimental evidence from postwar Bosnia and Herzegovina implies that women may be less willing to participate in the political process if the past conflict is still electorally salient (Hadzic and Tavits Reference Hadzic and Tavits2019). What is more directly relevant for our argument, aggregate-level data from the same context suggest that communal exposure to violence lowers the level of representation of women—arguably becaue of voters’ continuing perceptions of threat (Hadzic and Tavits Reference Hadzic and Tavits2021). Although well argued, this finding is based on a methodologically problematic analysis of limited data—a difficulty often faced by scholars of postconflict societies, where data availability and quality can be lacking.
Although there are still questions about whether security-related gender stereotypes can have any decisive influence on voter choice or whether they are trumped by party preferences and stereotypes (Dolan Reference Dolan2014), the overall picture that emerges from the literatures on gender, voter choice, and postconflict politics gives us confidence to make the following set of empirical propositions. First, we propose that Croatian voters on average prefer male candidates to their equally qualified female counterparts ( H 1 ). We are cognizant of the mixed record of studies on voters’ gender bias, but we make our proposition on account of similar findings in consolidating democracies that are culturally and socioeconomically close to Croatia like the Czech Republic, Estonia, and Poland, where voters seem to have a general preference for male candidates (Allik Reference Allik2015; Górecki and Kukołowicz Reference Górecki and Kukołowicz2014; Jurajda and Münich Reference Jurajda and Münich2014). Second, we believe that this voter bias against female candidates should be conditional on voter ideology—that is, it should be stronger among right-wing voters ( H 2 ). We make this proposition for two reasons: (1) there is extensive correlational evidence from both nonconflict and postconflict societies that more women get elected on left-wing than on right-wing lists (e.g., Hughes and Tripp Reference Hughes and Tripp2015; Kenworthy and Malami Reference Kenworthy and Malami1999); and (2) the right wing of the political spectrum throughout Europe, and perhaps particularly in countries like Croatia (discussed further later in this article), has a clear gender dimension that sees politics as being predominantly for men.
Our third proposition builds on the literature on voters’ gender biases and stereotypes related to security, and it suggests that voters’ preferences for male candidates should be conditional on the level of communal exposure to war violence. We agree with Hadzic and Tavits (Reference Hadzic and Tavits2021) that voters in communities with greater experiences of war violence likely feel threatened long after the violence had ended. It is also likely that local politics in these communities is dominated by themes directly related to the war—such as reconstruction, veterans’ benefits, or refugees’ return—that make the recent war past politically salient. Coupled with voters’ general gender stereotypes when it comes to security, all of this should lead to female candidates being particularly disadvantaged in communities more exposed to war violence ( H 3 ). Here, however, we believe it is necessary to be cognizant of the politicized or even ideologized nature of politics related to the recent war past in war-affected areas. In the Croatian context, there has been a marked difference in the way the nationalist political right approaches politics related to the recent war past compared with the political left (Mochtak, Glaurdić, and Lesschaeve Reference Mochtak, Glaurdić and Lesschaeve2021). Right-wing politicians in Croatia and throughout the post-Yugoslav space not only keep the war past alive in political discourse, but also portray it in contentious terms that leave little room for reconciliation (Glaurdić, Lesschaeve, and Mochtak Reference Glaurdić, Lesschaeve and Mochtak2022). The left-wing parties, on the other hand, approach the politicization of war memories differently: in most cases, they either stay out of it or highlight the need for reconciliation. What is equally important, they approach the policy needs of voters in war-affected areas differently: rather than seeing them through the lens of the ended conflict, they present them through the lens of general social welfare.
In other words, the classic dichotomy between security and social policy (arguably as old as the modern nation-state) is amplified in war-affected areas, not only on account of the real needs of the population, but also on account of different strategies of the right-wing and the left-wing parties. The right “owns” the security issue, both physical security (national defense, fight against crime, etc.) and cultural security (norms, values, identity). The left, by contrast, “owns” social policies that focus on health care, education, fight against poverty, etc. By owning an issue, that party (or that side of the ideological spectrum) appears more competent, and voters who care about those issue will prefer the party that owns the issue. At the same time, gender stereotypes result in men appearing more competent on security issues, and women more competent on social issues. Thus, voters who care about security will be more likely to vote for right-wing parties, and, within the candidate lists of right-wing parties, to cast their ballots for men, because they appear to be capable of actually delivering on those security issues. Voters who care more about social welfare will prefer left-wing parties, and, within the candidate lists of left-wing parties, they will prefer women, again for reasons of perceived competence. We propose that the degree to which this holds will depend on whether those classic issues of security and social policy are electorally salient. We believe that one of the legacies of war lies in the fact that exposure to violence increases the salience of these two sets of issues even after the conflict has ended. In areas less affected by war, we can expect voters to care about other issues as well. Simply put, the patterns described in H 2 should be clearer in war-affected areas, and more affected by noise in areas less effect by the war. This is why we believe that the impact of ideology we proposed in H 2 should be further conditional on exposure to war violence. In other words, we believe that bias against female candidates should be particularly pronounced among right-wing voters in war-affected areas, and it should be mitigated among left-wing voters in war-affected areas ( H 4 ).
Women and Croatian Electoral Politics
Parliamentary representation of women in Croatia followed a similar trajectory to other postcommunist countries of Eastern Europe. Throughout the last years of communist rule, Croatian women enjoyed a somewhat higher level of representation than their counterparts in Western Europe, though substantially lower than the levels seen in the countries of the Soviet bloc (Montgomery Reference Montgomery, Matland and Montgomery2003). With the first democratic elections for the Croatian parliament (Sabor) in 1990, however, these figures plummeted and remained low during the last decade of the twentieth century, which was marked by the re-traditionalization of gender roles and popular backlash against “directive emancipation” in a martialized society defined by the 1991–95 War of Independence (Glaurdić Reference Glaurdić, Matland and Montgomery2003). Two additional factors did not help. First, throughout the 1990s, Croatia was ruled by the nationalist Croatian Democratic Union (HDZ, Hrvatska demokratska zajednica), which was actively committed to transforming the social role of women into that of little more than mothers and homemakers. Second, the majoritarian and mixed electoral rules in place arguably did not favor women. Figure 1 shows the evolution of women’s representation in the Croatian Sabor from 1978 to 2020.
Women candidates and MPs in Croatia, 1978–2020. For the nondemocratic “delegate elections” in 1978, 1982, and 1986, the candidate figures refer to the proportion of women in the pool of about 11,000 elected local-level delegates who elected national-level representatives from within their ranks. All figures always refer to the lowest (1978–90), lower (1992–2000), or single house of the Sabor (2003–20).
Prior to the advent of democracy, Croatian members of parliament (MPs), like those in the rest of Yugoslavia, were elected based on so-called delegate democracy rules. In practice, this meant that voters voted every four years for their local representatives/delegates (not all of whom had to be members of the ruling League of Communists), who then voted to determine which among them would be sent/delegated to the national/republican parliament (nearly all of whom ended up being members of the ruling League of Communists). For the 1990 elections, the Croatian League of Communists instituted two-round majoritarian elections in the mistaken belief that this would benefit it. It did not. Instead, the majoritarian electoral rules benefited the HDZ. The rules also polarized the electoral competition so that all political parties nominated far fewer women. This dynamic was repeated in 1992 and 1995, when the HDZ instituted a mixed electoral system in which voters cast two unrelated votes: one for a single candidate elected in a district under plurality rules and one for the national list. During this period, Croatia had one of the lowest levels of women’s parliamentary representation in postcommunist Europe.
This was perhaps not surprising considering the social changes brought about by the 1991–95 war and the re-traditionalization of gender roles promoted by the ruling HDZ and the Catholic Church. The war had devastating consequences for Croatian society, with more than 20,000 dead and 800,000 refugees in a country of just over 4 million inhabitants. Although most of the fighting was obviously done by men, the burden on civilians—most often women, children, and the elderly—was enormous, with the traumatic experiences of many of them dramatically affecting their lives years or even decades after the end of hostilities. Crucially, war also changed the prevailing perceptions of women in social and political life. The advancements toward equality that women previously achieved under socialism simply vanished during the first half of the 1990s (Stanić and Mravak Reference Stanić and Mravak2012).
Things changed dramatically with the 2000 election, for three reasons. First, in expectation of its electoral loss, the HDZ instituted PR electoral rules in which Croatia was divided into 10 districts each electing 14 MPs from closed lists. This allowed (or pushed) all parties to improve the gender balance of their tickets. Second, the election brought to power a center-left coalition led by the Social Democrats (SDP) that was far more progressive when it came to gender equality. Third, and most important, women of all stripes and colors organized in the immediate postwar period and campaigned for greater access to political power (Irvine Reference Irvine2007). Their activism and organizational drive closely mirrored similar efforts by women in other postwar societies (Tripp Reference Tripp2016). With the 2000 election, Croatia became the first postcommunist country to surpass the level of women’s representation in its last nondemocratic election. As the electoral system solidified and remained unchanged in the decade that followed, women made slow but steady progress. The newly moderated HDZ even passed a new Law on Gender Equality in July 2008, mandating a 40% gender quota for all party lists that was to be enforced “within three regular elections” (Hrvatski sabor Reference sabor2008).
Things changed for the worse, however, with the 2015 election. The government led by the SDP changed the electoral rules once again, this time opening the lists and allowing voters to cast either a list vote or a preferential vote for one candidate who could rise on the list ranking if they received more than 10% of the party’s vote in the electoral district (Hrvatski sabor Reference sabor2015). The new law also mandated the 40% gender quota, which would disqualify all lists that did not satisfy this provision. In response to some parties’ complaints, the Constitutional Court struck down the gender quota mandate provision of the law just prior to the election, leaving in place minor financial penalties for lists that were not gender balanced (Gergorić Reference Gergorić2015; Ustavni sud Republike Hrvatske 2015). One-fifth of party lists did not satisfy the gender quota, including the HDZ in all 10 electoral districts and the SDP in 5.Footnote 1 More women were nonetheless nominated, but the party leadership placed them much lower than before. Interestingly, this trend closely parallels the trend observed across Europe in European Parliament elections, in which parties placed women lower under open-list than under closed-list PR rules (Lühiste Reference Lühiste2015).
Figure 2 shows the extent of this tendency graphically using the examples of the HDZ- and SDP-led coalition lists since the institution of PR rules in 2000. It is important to note that between 75% and 90% of MPs during this period were elected from these two lists—in other words, these figures are the ones that truly matter. Three things are immediately apparent from Figure 2. First, the Social Democrats have consistently nominated more women. Second, the upward trend in nominations of women is clear for both parties. And third, the drop in average women’s list position between the 2007 peak, when there was basically no difference between the average list positions of men and women (i.e., they were around 7.5 for both the HDZ and the SDP), and the 2015 nadir is stark. Together with the fact that the SDP-led coalition fared worse in 2015 and 2016 than in 2011, this was the crucial reason why the proportion of women elected from lists fell from 22.1% in 2011 to 15.7% in 2015 and 12.9% in 2016, before bouncing back to 22.1% in 2020.Footnote 2
Nominating women under PR rules by the HDZ and the SDP, 2000–20.
Data and Method
Our analysis is based on the results of three rounds of parliamentary elections (2015, 2016, and 2020) held under open-list PR rules. The fact that these elections took place more than two decades after the end of Croatia’s War of Independence should make the testing of our hypotheses particularly stringent. As noted earlier, the Croatian electoral system divides the country into 10 relatively large multimember electoral districts. However, thanks to detailed reporting of precinct-level vote tallies for all individual candidates, we were able to pair electoral results with economic, sociodemographic, and war-related data on Croatia’s 556 municipalities. Our sample consists of 6,986 candidates running on 97 electoral lists of all ideological profiles. Since we aim to explain voters’ decision to support a particular candidate on their party list, we model the dependent variable as the number of votes cast for a candidate, expressed as a proportion of all votes cast for the candidate’s party slate in a given municipality. This choice follows the standard established in the literature for preferential voting (e.g., Allik Reference Allik2015; Dettman, Pepinsky, and Pierskalla Reference Dettman, Pepinsky and Pierskalla2017; Van Erkel and Thijssen Reference Van Erkel and Thijssen2016; for a useful survey of preferential voting systems, see Passarelli Reference Passarelli2020). We weight the observations by the absolute number of votes behind each proportion, because we want to make sure that the results are not biased in favor of dynamics valid for small parties and municipalities.
Since our unit of analysis is actually candidate * municipality and the disaggregation of electoral results expands the number of observations, we need to account for this in order to avoid underestimating the regression coefficients’ standard errors and overestimating statistical significance (Rabe-Hesketh and Skrondal Reference Rabe-Hesketh and Skrondal2012). Additionally, our data are by nature multilevel, with candidates embedded in lists and lists embedded in municipalities, and our dependent variable is a proportion. This is why we opt for binomial multilevel regression with random intercepts for the candidate, list, party/coalition, and municipal levels. A linear model with log-transformed vote proportions as the dependent variable instead of a binomial model was also a possibility (e.g., Allik Reference Allik2015; Van Erkel and Thijssen Reference Van Erkel and Thijssen2016). However, since our robustness checks showed the results of both approaches as similar, we opted for the more straightforward binomial model.
Our principal variable of interest on the candidate level is gender, but our models also include a set of candidate characteristics that the literature (e.g., Aguilar, Cunow, and Desposato Reference Aguilar, Cunow and Desposato2015; Blom-Hansen et al. Reference Blom-Hansen, Elklit, Serritzlew and Villadsen2016; Dettman, Pepinsky, and Pierskalla Reference Dettman, Pepinsky and Pierskalla2017; Devroe and Wauters Reference Devroe and Wauters2018; Jankowski Reference Jankowski2016; Tavits Reference Tavits2010; Van Erkel and Thijssen Reference Van Erkel and Thijssen2016) commonly considers as having an impact on voter choice: candidate age, incumbency, geographic closeness to the electorate (modeled as the log-transformed distance between their residence and the municipal center to capture the decreasing marginal effect of distance on voting), and their position on the list (captured here with list position, first list place, and last list place). All these variables—except incumbency, which is more or less common knowledge among the electorate—are on the ballot materials in one form or another and thus easily observable to voters. On the level of party lists, we control for the number of women on the list to account for the size of the pool of female candidates that voters of a particular party in a given district can choose from; list vote in a given municipality; as well as party list ideology, which is a variable of particular interest for our second hypothesis. We follow the literature’s standard (e.g., Armingeon et al. Reference Armingeon, Knöpfel, Weisstanner and Engler2014) and code ideology as left (1), center-left (2), center (3), center-right (4), or right (5) after closely examining party programs.
We also control for contextual factors commonly considered as affecting voters’ propensity to cast preferential votes: education, unemployment, urbanization, and the electorate belonging to the ethnic majority (in this case, Croats) (André, Wauters, and Pilet Reference André, Wauters and Pilet2012; Wauters et al. Reference Wauters, Thijssen, van Aelst and Pilet2016). Moreover, considering our interest in examining whether greater exposure to war violence of the electorate affects its propensity to vote for female candidates, we also include a variable measuring the number of war disabled—that is, persons whose disability was directly caused by war operations, per 1,000 municipal inhabitants. We opt for this approach to modeling the exposure of the population to war violence for several reasons. First, there simply are no reliable war death figures for both civilians and military personnel on the local level in Croatia. Second, the disability figures were collected in Croatia’s 2011 census and therefore are more reliable than similar figures from various social service institutions, where individuals may have incentives to alter their disability or war records to procure benefits. And third, the disability figures capture the war exposure of the population that inhabited the municipalities at a time as close as possible to the elections. This reduces potential issues caused by migration flows during and after the war (which would have been a problem, for example, for the war-related deaths). This variable has been used in several other studies (e.g., Glaurdić and Vuković Reference Glaurdić and Vuković2015, Reference Glaurdić and Vuković2016), where it was shown to correspond rather well to the movement of the front lines between the combatants. To assuage any concerns about possible bias of this measure, however, we also conduct robustness tests using the figures for Croatian military deaths acquired from the Croatian Memorial and Documentation Center of Homeland War (Glaurdić, Lesschaeve, and Mochtak Reference Glaurdić, Lesschaeve and Mochtak2018). In our robustness tests, we use the cube-transformed figures for military deaths and war disability to examine whether the skew in the distribution had any effect on the results. We report the results of our robustness tests in the Supporting Information online. Finally, we include dummy variables for the 2015, 2016, and 2020 elections in all our models to account for any electoral cycle differences. Table 1 gives an overview of all variables used in this analysis.
Descriptives of variables, electoral results
We supplement our analysis of the actual results in three rounds of elections with the analysis of survey data collected in the fall of 2018—that is, in the midst of the period under scrutiny—using a sample of more than 2,000 Croatian adults (Lesschaeve, Glaurdić, and Mochtak Reference Lesschaeve, Glaurdić and Mochtak2022). To achieve representativeness, survey respondents were recruited using quota sampling based on about 300 social strata identified in Croatia according to gender, age, level of education, and region. Discrepancies between the sample and the population were further reduced using survey weights throughout the analyses (Kalton and Flores-Cervantes Reference Kalton and Flores-Cervantes2003). In addition to standard demographics and information on political affiliation and interest, the survey collected data on respondents’ experiences of war trauma (operationalized as the sum score of answers to six yes/no questions measuring symptoms of war-related trauma, following Ringdal and Simkus Reference Ringdal and Simkus2012) and their views on traditional gender roles that would keep women committed to family life and child-rearing. Table 2 shows the descriptives of the variables used in the analysis of survey data.
Descriptives of variables, survey data
Results
We show the results of our analyses of electoral results in Table 3, where the four models correspond to our four hypotheses. The results of our robustness tests with alternative variables capturing the exposure to war violence are presented in Tables A1–A3 in the Supplementary Materials. Notably, they show nearly identical findings to those shown in Table 2. Here we should note that the data and our models allow for testing of our hypotheses only on the intraparty level (i.e., controlling for voters’ party choice). In other words, our data do not allow us to test whether voters, for example, opted to vote for a left-wing party because it had more women on its list or for a right-wing party because it had fewer women on its list. The data do allow us to test, however, whether voters had any gender bias once they made their choice of which party to support. Our first hypothesis suggested that female candidates were electorally penalized by voters. Model 1 provides strong support for this hypothesis as the effects of gender are highly statistically significant and in the expected direction. Female candidates received 0.26 percentage points fewer preferential votes than comparable male candidates. Considering that candidates on average received 2.2% of total votes given to their party list and its candidates, this is a substantively small, though not trivial, effect. Just like voters in Estonia (Allik Reference Allik2015), Poland (Górecki and Kukołowicz Reference Górecki and Kukołowicz2014), and Belgium (Marien, Schouteden, and Wauters Reference Marien, Schouteden and Wauters2017), Croatian voters do indeed penalize female candidates when allocating their preferential votes.
Determinants of preferential vote
∆ AIC indicates the change in AIC when comparing the model to an intercept only model.
* p < .05;
** p < .01;
*** p < .001.
When it comes to control variables on the individual level, we find no support for candidates’ age having an effect. We do find, however, that candidates’ incumbency brings them substantial electoral benefits (1.4 percentage points), as does their geographic closeness. In line with a string of studies in other contexts (e.g., Arzheimer and Evans Reference Arzheimer and Evans2012; Jankowski Reference Jankowski2016; Van Erkel and Thijssen Reference Van Erkel and Thijssen2016), we find that candidates living, for example, 100 kilometers away from the municipal center can expect 3.6 percentage points fewer preferential votes than comparable candidates living in the municipal center. What matters most, however, is the candidates’ placement on the electoral list—another finding that is in line with the literature (e.g., Blom-Hansen et al. Reference Blom-Hansen, Elklit, Serritzlew and Villadsen2016; Dettman, Pepinsky, and Pierskalla Reference Dettman, Pepinsky and Pierskalla2017; Koppell and Steen Reference Koppell and Steen2004). Obviously, list position is not random, but a proxy for the candidates’ preelection standing in the party hierarchy and, at least to some extent, among voters. List position, first list place, and last list place should, thus, be seen as the most important set of control variables in our models. The results of our analysis confirm that higher-placed candidates receive more preferential votes, and this is particularly the case for candidates who are placed first or last on the list (last place is usually reserved for party mavericks).
On the list level, we find that preferential voting was more prevalent among voters of smaller and right-wing parties (i.e., those with fewer list votes and with higher values of ideology). On the municipal level, we find the incidence of preferential voting to be higher in less urbanized municipalities with higher unemployment and lower levels of education—a sign perhaps that voters in “forgotten” areas are more eager to change party hierarchies (cf. André, Wauters, and Pilet Reference André, Wauters and Pilet2012). Preferential voting, on the other hand, was less prevalent in areas more affected by war violence, signaling that voters in those areas (most of them voters of right-wing parties) were more accepting of the candidate rankings provided by the party leadership.
Our second hypothesis proposed that gender bias was at least partly conditional on voters’ ideology (i.e., that it was more pronounced among voters of right-wing parties). We test this hypothesis in Model 2, in which we interact gender with ideology and find the results statistically significant at the 0.05 level and in the expected direction. Just as it has been shown in other contexts (e.g., McDermott Reference McDermott1998), Croatian voters of right-wing parties are indeed more biased against female candidates than voters of left-wing parties. To be exact, voters of fully left-wing parties (i.e., those with ideology = 1) allocate 0.1 percentage points more to male than to otherwise comparable female candidates, whereas for voters of fully right-wing parties (ideology = 5), this difference is 0.4 percentage points.
In the third hypothesis, we proposed that gender bias was also at least partly conditional on the exposure of the local electorate to war violence. We argued that voters in areas more exposed to war violence would be, for a variety of reasons, more biased against female candidates. The results of our Model 3, in which we interact gender with our proxy for exposure to war violence, war disabled, clearly support H 3 , as the results are highly statistically significant and in the expected direction. Voters living in areas that were completely unaffected by war violence (i.e., war disabled = 0) give 0.3 percentage points in preferential votes more to male candidates than to comparable female candidates. This difference increases to 0.4 percentage points in areas with high levels of war violence (mean war disabled + 1 standard deviation = 28). We show this effect—substantively perhaps limited, but not trivial—graphically in Figure 3. Here we should note that we treat the predictions as group means and calculate whether they differ significantly at different values of war disabled. We put a star next to each value of war disabled on the x-axis if the p-value of the difference between the two lines is smaller than 0.05. Our proposition that voters in areas affected by violence would be more biased against female candidates in elections was grounded in several strands of literature that suggest such a dynamic might be true due to legacies of violence affecting voters’ hierarchies of policy preferences or even gender norms. More research is needed to uncover the actual reasons and mechanisms behind this finding—in addition to confirming it in other postconflict contexts—but we believe it highlights the long-lasting challenges of the exposure to conflict violence to postconflict social progress and recovery.
Predicted preferential votes for male and female candidates, conditional on the level of local exposure to war violence.
Here we have to note that it is possible that the effect we have identified is caused by omitted variable bias. The geographic pattern of violence in the Croatian War of Independence reflected the local ethnic balance, strategic value of territory, and the distribution of the Yugoslav Army forces that sided with the Serb rebels. In other words, it was independent of the local population’s possible gender bias. However, it is also the case that the violence was higher in areas that were comparatively less socioeconomically developed, with some areas having a history of armed conflict and militarized public administration going back to Habsburg times (Tkalec Reference Tkalec2020)—lower levels of socioeconomic development and history of militarized social organization being factors that may have led to more traditional gender norms when it comes to involvement in politics. Some of these problems should be alleviated by our inclusion of municipality-level variables for education, unemployment, and urbanization, whose geographic pattern today is little different than it was decades ago. Nevertheless, we have to allow for the possibility that people in war-affected areas exhibited higher levels of gender bias in electoral decision-making before the 1991–95 violence, even when controlling for these variables capturing their communities’ socioeconomic development. However, lack of any useful municipality-level data on political gender bias from that period, coupled with the fact that Croatia had no democratic elections prior to 1990, precludes us from answering this question definitively.
Finally, our fourth hypothesis suggested that this interactive relationship between exposure to war violence and gender was conditional on voters’ ideology. We noted the two completely opposite ways in which the political left and the political right politicized the recent war past in Croatia and approached policy making directed at vulnerable populations and areas affected by war violence (Glaurdić, Lesschaeve, and Mochtak Reference Glaurdić, Lesschaeve and Mochtak2022; Mochtak, Glaurdić, and Lesschaeve Reference Mochtak, Glaurdić and Lesschaeve2021). This is part of the reason—in addition to the real needs of the local population, which are heightened—why we suggested that, in the Croatian context, voters of right-wing parties in war-affected areas should be even more prone to prioritize security over other policy areas and thus even more inclined to favor male candidates. Conversely, we suggested that voters of left-wing parties in war-affected areas should be more prone to prioritize policy areas such as social welfare and thus more inclined to favor female candidates. In other words, we suggested that the dynamics underlying the classical dichotomy between security and social policy should be amplified in war-affected areas. The results of Model 4 support our propositions, as the coefficients are all statistically significant and in the expected direction. Since they do not lend themselves to easy interpretation from the table, we present them graphically in Figure 4, with the top graph showing predicted preferential votes for male and female candidates, conditional on the level of local exposure to war violence for left-wing parties and the bottom graph showing the same for right-wing parties. Here we note again that we put a star next to each value of war disabled on the x-axis if the p-value of the difference between the two lines at that value is smaller than 0.05.
Predicted preferential votes for male and female candidates, conditional on party ideology and the level of local exposure to war violence.
In areas completely unaffected by war violence (war disabled = 0), voters of both left-wing and right-wing parties give female candidates 0.2 to 0.3 percentage points fewer preferential votes than to comparable male candidates. In areas heavily affected by war violence (mean war disabled + 1 standard deviation = 28), however, male advantage jumps to 0.5 percentage points for right-wing parties. For left-wing parties, in contrast, it moves in the opposite direction and disappears completely. in other words, it is clear that the electoral prospects of male and female candidates in left- and right-wing parties evolve in opposite directions as the impact of the past conflict in a community increases. We suggest that this was likely the case because of the greater needs of the local population for both security and social policy, as well as because left- and right-wing parties stress different issues when seeking to connect with voters in war-affected areas, with the right emphasizing community security and the left focusing on people’s well-being. We believed this would reinforce the stereotypes of the issues male and female candidates are supposedly more capable of handling—that is, the focus by right-wing parties on security would benefit male candidates on their lists, while the focus on more traditionally feminine issues by the left in war-affected communities would increase the electoral prospects of women on their lists. More research, however, is needed to confirm this interpretation, as well as to study this dynamic in other contexts. We need to better explain the politicized or even ideologized nature of gender in postconflict societies that is likely highly contingent on local and temporal context. Without understanding the interaction between gender, ideology, and violence in a given postconflict society, we cannot fully understand the nature of social change brought about by that violence nor the magnitude of the challenge presented by that social change for postconflict progress and recovery.
We supplement our analysis of aggregate-level electoral results with an examination of similar questions using individual-level survey data. As noted earlier, the survey we use was conducted in the midst of the period of our interest. Its primary aims were somewhat different from the aims of the first part of our analysis, but it still offers exceptionally useful data that can help us answer whether the dynamics we observe on the aggregate level also translate to the level of individual voters. The dependent variable of interest is the respondents’ agreement with a single statement measured on a Likert-type scale: “The government should encourage women to stay at home to take care of the children.” We believe this statement, while not corresponding perfectly to the aggregate-level part of our analysis, does strike at the core beliefs underlying voters’ evaluation of electoral candidates based on their gender. Here we are particularly interested in establishing whether people’s experiences of war-related trauma have an impact on their propensity to subscribe to traditional gender roles.
As Table 4 shows, that is indeed the case. Respondents’ higher levels of war trauma are strongly positively correlated with their belief in traditional gender roles—which offers a good parallel with the individual level of our H3 that was posed on the aggregate level. This effect survives the inclusion of a battery of control variables, including respondents’ ideological orientation on the left-right spectrum. Unsurprisingly, people subscribing to traditional views on gender are disproportionally men, those of lower education, and those of right-wing ideological orientation—the latter finding corresponding perfectly with our H2. Our theoretical discussion suggested that voters in communities with greater experiences of war violence likely feel threatened long after the violence had ended, which—coupled with voters’ general gender stereotypes when it comes to security—leads to female candidates being particularly disadvantaged in communities more exposed to war violence. This dynamic is likely valid on the individual level as well—that is, voters more traumatized by war likely place a premium on security that, together with their possible belief in gender stereotypes on security policy, leads them to believe that women’s place is in the family and not in politics. Obviously, other mechanisms could also be at play, and more research is needed to properly expose what drives this finding. What matters for us, however, is that our hypotheses H2 and H3 have useful parallels in survey data. Here we should note that interacting respondents’ left-right self-placement with war trauma did not yield any statistically significant finding. This should not, however, be seen as going against our H4 as that hypothesis was made on the aggregate level and was rooted in our understanding of the different policy and campaign strategies left- and right-wing parties use when seeking to connect with voters in war-affected areas as opposed to those in areas that were not directly affected by war violence. The survey unfortunately did not capture fine-grained information on respondents’ communities and their exposure to war violence, so this proposition from the aggregate-level analysis could not be appropriately replicated with individual-level data.
Effects of war trauma on belief in traditional gender roles
Notes: Ordinary least squares regression. Dependent variable = “To what extent do you agree with the statement ‘The government should encourage women to stay at home to take care of the children’”?
* p < .05;
** p < .01;
*** p < .001.
Conclusions
In 1997, just two years after Croatia’s War of Independence, a representative of the HDZ told his female colleague from the center-left Social Liberals on the floor or the Sabor that she should “talk less and give more births!” Four years later, a far-right MP told the leader of the People’s Party, Vesna Pusić, during a parliamentary debate that “God made her for a mattress, not to lecture others” (Jutarnji list Reference list2010). Such public outbursts of misogyny were commonplace in Croatia’s political discourse in the immediate aftermath of the War of Independence. After years of re-traditionalization of social gender roles during the war, women started to regain some social and political influence in the second half of the 1990s (Irvine Reference Irvine2007). This provoked a harsh pushback from the right wing of the political spectrum that still believed in the role of women as—at best—mothers and caregivers in a martialized society. Although such episodes are no longer customary in Croatia’s political mainstream, women are still strongly underrepresented in national politics. Our analysis convincingly shows that at least part of the problem may lie in the bias that Croatian voters exhibit against female electoral candidates. More importantly, our analysis also shows that this gender bias—even more than two decades after the War of Independence ended—is conditional on voters’ ideology and their communities’ exposure to war violence. We believe this is yet another testament to the long-lasting and often not immediately apparent effects armed conflicts have on societies.
Our analysis, of course, poses as many questions as it answers. First, and most obviously, there is the need to account for the portability of our argument. Our propositions need to be tested in other postconflict contexts using not only aggregate-level but also individual-level data. We need to somehow square the observation that postconflict societies often have higher levels of women’s representation with our findings that exposure to war violence makes electorates actually less likely to support female electoral candidates. Second, we need to understand more about the interaction of ideology, conflict, and gender in postconflict societies. Croatia’s War of Independence was not a conflict between two sides that had different views of the role of women in society. It was a conflict in which the dominant ideological force with a firm grip on power on both sides was the nationalists—and they had very clear ideas regarding the “proper place” of women in public life. Our analysis demonstrates that the interaction between ideology, conflict, and gender matters in postconflict politics, but this interaction needs to be studied and better theorized further, using data and examples from other postconflict contexts, especially those where the fighting forces may have had different ideological orientations and views of women in society.
Finally, there remains the need to better explain the actual causes and mechanisms behind the impact of communal exposure to war violence and gender bias. We suggested that the essence of the story may be in the different hierarchies of policy preferences of voters in war-affected areas, particularly when it comes to security. This proposition, however, needs to be empirically confirmed because it is possible that exposure to war violence leads to changes in more fundamental views of voters when it comes to the social role of women or to gender and leadership. In our view, these questions offer a solid path forward for the study of gender bias and voter choice in postconflict societies. Together with the principal findings of our analyses highlighting the long-term social effects of exposure to war violence, they are an important contribution to the broader study of gender and politics in postconflict societies.
Supplementary Materials
To view supplementary material for this article, please visit http://doi.org/10.1017/S1743923X22000654.
Open access
