Introduction
National elections in Australia, Britain, Canada, and the USA in 2019–2020 saw a record-breaking number of women elected to public office (Allen Reference Allen2020; CAWP 2020; Hough Reference Hough2022; Raman-Wilms Reference Raman-Wilms2019). These successes appear to bolster an emerging scholarly consensus that voters will support women candidates: despite well-documented biases, victory rates for female and male candidates in Australia, Canada, Britain, and the USA are near-identical (Burrell Reference Burrell1992; Black and Erickson Reference Black and Erickson2003; Lawless and Pearson Reference Lawless and Pearson2008; Sevi et al. Reference Sevi, Arel-Bundock and Blais2019; Thomsen Reference Thomsen2020). Additionally, a slew of experimental research shows that voters support men and women candidates equally (Dahl and Nyrup Reference Dahl and Nyrup2021; Kage et al. Reference Kage, Rosenbluth and Tanaka2019; Saha and Weeks Reference Saha and Weeks2022), or even prefer female candidates (Clayton et al. Reference Clayton, Robinson, Johnson and Muriaas2020; Teele et al. Reference Teele, Kalla and Rosenbluth2018).
However, other evidence suggests that sexism can shape electoral outcomes. This scholarship focuses on voter antipathy towards women, per Glick and Fiske’s (Reference Glick and Fiske1996) concept of hostile sexism, and shows a clear association between sexist attitudes and electoral choices, such as support for Donald Trump (Cassese and Barnes Reference Cassese and Barnes2019), the British Conservatives (De Geus et al. Reference De Geus, Ralph-Morrow and Shorrocks2022), and the Australian National Party (Beauregard Reference Beauregard2021). Sexist attitudes are also associated with policy preferences (Beauregard et al. Reference Beauregard, Holman and Sheppard2022; Beauregard and Sheppard Reference Beauregard and Sheppard2021). Since women remain under-represented in legislatures globally, this evidence of sexism’s relationship with party preferences suggests some voters might be reluctant to support women candidates, despite empirical evidence on average failing to reveal such bias. However, because research on sexism typically focuses on correlations between voters’ gender attitudes and party rather than candidate choice, it is unclear whether sexist attitudes impose electoral penalties on women seeking office.
Conflicting results thus characterize these related literatures: much experimental research suggests voters are not biased toward women candidates, yet sexism studies – typically using observational data – indicate that sexist attitudes are prevalent and consequential for party choice. Here, we bridge this divide and offer novel evidence of how sexism affects candidate choice. We match information on voter choice and attitudes from post-election surveys to candidates’ characteristics, using both official and our original candidate data. We focus on four democracies with district-based systems where voters cast a ballot for a specific candidate (Britain, Canada, Australia, and the USA), thus enabling us to analyse the relationship between sexist attitudes and votes cast for female and male candidates. The datasets yield information on approximately 26,000 voters and 5,346 candidates. We analyse candidate choice and abstention to understand how votes are conditioned by voter sexism and candidate gender. In each country we consider, women are under-represented amongst political candidates and constitute just over 1/3 of elected representatives. Moreover, sexist attitudes are held by a significant minority of voters (see, for example, Cassese and Barnes Reference Cassese and Barnes2019; de Geus et al. Reference De Geus, Ralph-Morrow and Shorrocks2022; Beauregard Reference Beauregard2021). And yet, experimental studies have found that voters either prefer female candidates in Australia (for example, Kang et al. Reference Kang, Sheppard, Snagovsky and Biddle2021), Canada (Chen et al. Reference Chen, Thomas, Gosselin and Harell2024), the UK (Carnes and Lupu Reference Carnes and Lupu2016) and the USA (for example, Saha and Weeks Reference Saha and Weeks2022), or evaluate male and female candidates similarly (for example, Campbell and Cowley Reference Campbell and Cowley2014; De Geus et al. Reference De Geus, McAndrews, Loewen and Martin2021; Teele et al. Reference Teele, Kalla and Rosenbluth2018). Our case study countries therefore provide an ideal context in which to ascertain whether sexist attitudes coexist with support for real female candidates.
After accounting for variation between male and female candidates – especially the performance of their party at the previous election at the district level – we find voters equally likely to choose a woman or a man in Britain, Canada, and Australia. In line with recent experimental work, US voters are more likely to vote for a woman. We find no effect of voter sexism on the likelihood of voting for a woman in any country, once we account for the relationship between sexism and party choice, which is statistically significant in every country, in line with existing findings. We find that in the USA only, sexist partisans faced with a woman co-partisan candidate are less likely to vote, suggesting they prefer to abstain rather than vote for an alternative. Our results, then, suggest that partisan preferences matter more than sexist attitudes when it comes to vote choice, in line with some previous findings (for example, Dolan Reference Dolan2014), which may explain why experimental findings do not find an anti-woman bias amongst voters despite the prevalence of sexist attitudes in the electorate. However, there are two caveats to this: first, in light of much experimental evidence which finds a slight preference amongst voters for women candidates (Schwarz and Coppock Reference Schwarz and Coppock2022), it is notable that we do not observe this in our observational study and, second, we also find that abstention may be an important electoral choice for sexist partisans that is often missed by existing observational and especially experimental work.
We offer five key contributions. First, through descriptive analysis of candidate data, we argue that experimental studies do not capture the patterns of gendered candidacies in elections across parties and districts. Second, unlike experimental studies, we have data on real-world votes and find no difference in preferences for male or female candidates in three of four countries. We reflect later on the implications of this in light of recent experimental work. Third, unlike many studies on voter sexism, we measure candidate rather than party choice. We find little evidence that sexists are less likely to vote for women, but confirm that sexism remains important to voter decision-making. Fourth, we offer a test of abstention alongside candidate choice as an important behavioural option for sexist voters, one often ignored in both experimental studies – despite their claims of replicating real-world decision making – and observational studies. We show that, especially in the US, abstention rather than candidate choice might be a better way to understand sexist voters’ behaviour when facing a woman candidate. Finally, we offer a comparative study of hostile sexism across four countries. Because much literature on both women candidates and the relationship between sexism and electoral candidates focuses on the USA, it is unclear to what extent these findings apply elsewhere. We compare levels of sexism in the four countries, analyse the relationship between sexism and candidate choice, and find evidence that Americans, uniquely, are slightly more likely to vote for women.
The paper proceeds as follows. We provide an overview of gender and politics research which focuses on voter bias against and towards women, and explain why we think observational studies that combine voter with candidate data can provide valuable insights. Next, we discuss theories and evidence about the role of sexism and gender attitudes in electoral outcomes. We draw on this scholarship to establish hypotheses, then describe our approach to data collection and analysis. We discuss our results and finally identify further areas of research.
Sexism and Voting for Women Candidates: How Should We Study It?
Women remain significantly under-represented in politics. Around a third of the Australian and Canadian Houses of Representatives and British House of Commons are women; in the US House of Representatives, less than a quarter. Many studies aim to identify the hindrances to women entering public office. One category focuses on structural, supply-side obstacles, such as gender differences in political socialisation, caregiving responsibilities, and under-involvement in pipeline professions that precede political careers (Fox and Lawless Reference Fox and Lawless2014; Fulton et al. Reference Fulton, Maestas, Maisel and Stone2006; Thomsen and King Reference Thomsen and King2020). The other category of scholarship – which our paper contributes to – focuses on demand-side barriers by seeking to identify the factors which cause parties and voters to support women candidates (or not), including party recruitment and selection processes for candidates (Bjarnegard and Kenny Reference Bjarnegard and Kenny2016; Caul Reference Caul1999; Teele et al. Reference Teele, Kalla and Rosenbluth2018). Gender stereotypes often provide a theoretical underpinning here, with the literature suggesting that voters do indeed hold stereotypic assumptions about gendered personality traits which inform baseline preferences for male or female candidates, and beliefs about leadership and policy competencies (Sanbonmatsu Reference Sanbonmatsu2002; Holman et al. Reference Holman, Merolla and Zechmeister2011; Huddy and Terkildsen Reference Huddy and Terkildsen1993; Lawless Reference Lawless2004). However, it is far from clear that gender stereotypes in fact influence political behaviour, with multiple studies suggesting that partisanship is a more important determinant of vote choice (Dolan Reference Dolan2014; Hayes Reference Hayes2011; Hayes and Lawless Reference Hayes and Lawless2016).
There is little empirical evidence that contemporary voters discriminate against women candidates. The literature points to initial voter bias against women that dissipates over time. In Australia, women candidates in the 1970s faced a voter penalty of up to five percentage points, which shrank to 0.3 per cent by 2004 (King and Leigh Reference King and Leigh2010). Canadian federal elections in the 1920s produced a gender gap of around 2.5 percentage points, which by 2015 was virtually zero (Sevi et al. Reference Sevi, Arel-Bundock and Blais2019). A gap of up to 3 per cent in 1970s Britain was statistically insignificant by 2010 (Kelley and McAllister Reference Kelley and McAllister1984; Campbell and Heath Reference Campbell and Heath2017). US scholarship since the 1970s has contended that candidate sex does not influence election outcomes (Darcy and Schramm Reference Darcy and Schramm1977); recent studies support this (Thomsen Reference Thomsen2020).
However, gender-neutral electoral returns do not mean the absence of bias, and some observational studies suggest lingering voter antipathy towards women. Pearson and McGhee (Reference Pearson and McGhee2013) found that female Democratic and Republican congressional candidates were more experienced than male candidates, and that Democratic (but not Republican) women ran in more politically favourable districts. But women did not have a higher victory rate, so this electoral parity may imply voter discrimination. Fulton (Reference Fulton2012) similarly concluded that female congressional candidates of higher quality than their male counterparts were no more likely to win their races, and Fulton and Dhima (Reference Fulton and Dhima2021) show that, when qualifications are held constant, female Democratic candidates are less likely than males to win. Sevi’s (Reference Sevi2023) analysis of the electoral fortunes of incumbent candidates found that while male incumbents enjoyed a small increase in their vote share, female incumbents experienced a modest reduction. Although these vote share results are not statistically significant, Sevi (Reference Sevi2023) notes that they are consistent with gender bias and may point to women legislators being held to higher standards or experiencing a greater degree of scrutiny than their male counterparts.
A criticism of such studies is that, given their observational nature, a causal relationship between candidate gender and vote choice cannot be identified. In a bid to explicitly identify how candidate gender influences vote share, political scientists have therefore increasingly turned to candidate choice experiments, typically using conjoint designs, which ask participants to rank or choose between candidates who vary across randomly assigned dimensions (gender, party, ideology, marital status) (Ono and Yamada Reference Ono and Yamada2020; Winter, Reference Winter2023; Teele et al. Reference Teele, Kalla and Rosenbluth2018). Although these myriad studies have not always produced consistent results, Schwarz and Coppock’s (Reference Schwarz and Coppock2022) meta-analysis found that far from unearthing voter bias against women, these studies revealed a pattern in favour of them across many countries. The authors (2022) suggest that voter behaviour towards women candidates has changed: whilst older studies often showed negative bias, post-2014 studies show voters are more likely, on average, to vote for women. Although Schwarz and Coppock (Reference Schwarz and Coppock2022) note that this positive bias towards women candidates is somewhat unexpected theoretically, they nevertheless observe that it is remarkably consistent across studies published after 2014.
However, an absence of hostility towards women candidates in experimental settings does not prove the absence of discrimination. First, survey experiments may not reflect real-world decision-making (McDonald Reference McDonald2020); participants may respond favourably to a hypothetical woman yet reject one at the ballot box (Schwarz and Coppock Reference Schwarz and Coppock2022). Second, conjoint experiments can mislead when seeking to identify proportions of voters who endorse particular candidate characteristics (Abramson et al. Reference Abramson, Kocak and Magazinnik2022); instead of accurately reflecting preference distributions, they can assign too much weight to a minority who intensely prefer a particular option. Third, as observed by Arel-Bundock et al. (Reference Arel-Bundock, Briggs, Doucouliagos, Mendoza Aviña and Stanley2022), many political science tests are underpowered; it is therefore possible that experimental findings which purport to reveal a preference for women voters are in fact reporting an inflated effect.
Perhaps most importantly, experimental studies often omit a crucial behavioural outcome: abstention. Candidate choice experiments are typically conjoint; while this design is hailed for its ability to mimic real-life scenarios by asking respondents to simultaneously consider multiple pieces of information (Hainmueller et al. Reference Hainmueller, Hopkins and Yamamoto2014), they are also usually ‘forced choice’ and compel participants to select from a series of options. While this approach has been justified on the grounds that it may increase respondent engagement with the task and lead them to more carefully consider candidate profiles (Hainmueller et al. Reference Hainmueller, Hangartner and Yamamoto2015), it also deprives respondents of an option open to many real-life voters, namely, to abstain from voting. Indeed, so rarely is an abstention option given in candidate choice experiments that out of the sixty-seven studies considered by Schwarz and Coppock’s (Reference Schwarz and Coppock2022) meta-analysis, we could only identify six which provided respondents with the option to abstain or not select any candidate. This is an important omission: Miller and Ziegler’s (Reference Miller and Ziegler2024) replication of two forced-choice conjoint experiments found that providing respondents with an abstention option can produce different conclusions. In the context of a candidate choice experiment, it is possible that including an abstention option may mean that a sexist respondent who is offered a choice between a female co-partisan and a male opponent may select neither.
As candidate choice experiments have proliferated, studies relying on surveys and observational data to understand women’s electoral fortunes have decreased. This, in our view, is regrettable, because each method has its own strength. Experiments excel at identifying causal mechanisms, while observational research enables political scientists to study voters who have been ‘exposed to real-world treatments’ by encountering political information in their natural environment (Banducci et al. Reference Banducci, Schoonvelde, Stevens, Barabas, Jerit and Pollock2017, 228). We therefore join Dolan and Lawless (Reference Dolan and Lawless2023) in contending that an analysis of real-life vote choice remains an important piece of the gender and elections puzzle.
Theory and Expectations
Although experimental scholarship gives us little reason to suspect that voters discriminate against women candidates, recent studies focusing on the association between political behaviour and sexism – as captured by the Ambivalent Sexism Inventory (ASI) tool – do indicate that gender attitudes have electoral consequences. Developed by psychologists Glick and Fiske (Reference Glick and Fiske1996), the ASI contains twenty-two items that measure two types of sexism: hostile and benevolent. Hostile sexism is ‘antipathy toward women who are viewed as usurping men’s power’ (Glick and Fiske Reference Glick and Fiske2001, 109); benevolent sexism refers to subjectively positive but ultimately disempowering beliefs about women and is characterized by reverence for women in the roles of wives, mothers, and romantic love objects (Glick and Fiske Reference Glick and Fiske1996).
Hostile (but not benevolent) sexism has been shown to influence political attitudes and behaviour. Most of this scholarship is from the USA, showing associations between hostile sexism and support for Donald Trump in 2016 (Cassese and Barnes Reference Cassese and Barnes2019; Schaffner et al. Reference Schaffner, Macwilliams and Nteta2018); antipathy towards female congressional candidates in 2016 (Winter, Reference Winter2023); endorsement of Republican House candidates in the midterm 2018 (Schaffner Reference Schaffner2022); and favourable assessment of male gubernatorial candidates (Dolan and Lawless Reference Dolan and Lawless2023). Research elsewhere has concentrated on the relationship with party choice. In Britain, De Geus et al. (Reference De Geus, Ralph-Morrow and Shorrocks2022) found an association between hostile sexism and the Conservative Party vote in 2019. In Australia, hostile sexism mattered for vote choice in 2019, but only for the National and Green parties, not the major (Labour and Liberal) parties (Beauregard Reference Beauregard2021). Some British and Canadian studies have found that voters (especially women) motivated by a desire to see more women in parliament are more likely to support female candidates in legislative elections (Campbell and Heath Reference Campbell and Heath2017; Goodyear-Grant and Croskill Reference Goodyear-Grant and Croskill2011), indicating that gender attitudes may matter in non-US contexts, but the role of negative attitudes to women – sexism – rather than positive attitudes has largely not been tested. An exception, Gareau-Paquette et al. (Reference Gareau-Paquette, Léal, Leblanc, Taylor, Vandewalle and Dassonneville2024), finds inconsistent evidence that traditional gender attitudes lowers voter support for women candidates in Canada.Footnote 1
The USA, then, provides the only context where political scientists have found an association between hostile sexism and candidate choice (Cassese and Barnes Reference Cassese and Barnes2019; Schaffner et al. Reference Schaffner, Macwilliams and Nteta2018; Winter, Reference Winter2023). While these studies indicate that hostile sexists will be less likely than others to vote for women, there is also reason to anticipate that US findings may not be replicated in other countries. First, because gender was unusually salient and politicized during the 2016 US presidential election, it may have exerted an outsize influence (Cassese and Barnes Reference Cassese and Barnes2019; see also Campbell and Heath Reference Campbell and Heath2017). Second, many studies which found an association between hostile sexism and unwillingness to support women focused on presidential and gubernatorial candidates (for example, Cassese and Barnes Reference Cassese and Barnes2019; Dolan and Lawless Reference Dolan and Lawless2023). Scholarship suggests that voters prefer candidates with masculine personality traits for executive positions (Huddy and Terkildsen Reference Huddy and Terkildsen1993), so findings from these contests may not apply to down-ballot decision-making.
Given that more recent experimental studies have tended to find a pro-woman bias amongst voters (Schwarz and Coppock Reference Schwarz and Coppock2022), and our focus is on recent elections, we should expect that overall voters are more likely to support female than male candidates (‘Woman Candidate hypothesis’, below). However, the importance of sexist attitudes to some candidate choice decisions in the USA and party choice decisions elsewhere suggests that sexist attitudes might be politically relevant and that some sections of the electorate might be less likely to support women. Particularly, hostile sexists should be more likely to vote for a man, given the choice (‘Sexism hypothesis’).
Scholarship has also often found that partisan attachment to a party is more important than gender stereotypes or candidate gender to voter decisions (Dolan Reference Dolan2014). But the hostile sexism literature indicates that associations between inegalitarian gender attitudes and vote choice persist after controlling for partisanship (Cassese and Barnes Reference Cassese and Barnes2019; Winter, Reference Winter2023). Sexist attitudes, then, may be relevant even for strong partisans, and their sexism and partisanship may work against each other when faced with a co-partisan woman candidate. We suggest that sexist partisans might be unlikely to switch to another party’s male candidate in this scenario, because of their partisan identity, but might be more likely to abstain to avoid voting for a woman (‘Abstention hypothesis’).
We express our three hypotheses fully below. Based on our arguments above, we test these hypotheses observationally in the next section:
Woman Candidate hypothesis: On average, voters are more likely to vote for a woman candidate than a man when faced with a choice in their district.
Sexism hypothesis: The likelihood of voting for a woman decreases for those with higher levels of hostile sexism.
Abstention hypothesis: Sexist partisans who have a co-partisan woman candidate in their district will be more likely to abstain from voting than partisans with lower levels of sexism.
We remain agnostic about whether we should see differences between Australia, Britain, Canada, and the USA. Although previous studies have established similar levels of sexism and an electoral role for sexist attitudes in Australia, Britain, and the USA (De Geus et al. Reference De Geus, Ralph-Morrow and Shorrocks2022; Beauregard Reference Beauregard2021), there are differences, as sexist attitudes in Australia are associated only with minor-party voting (Beauregard Reference Beauregard2021).
There are important between-country differences regarding gender equality and partisanship that might lead to differences in the role of sexism for voter choice. First, gender equality: the World Economic Forum’s 2018 Global Gender Gap Index (GGGI) ranks the UK and Canada at 15 and 16, Australia at 39, and the USA at 51. It is unclear what role gender equality may play as a moderating factor in the link between sexism and vote choice. On the one hand, citizens of gender egalitarian countries are more likely to value gender equality in the public sphere (Yu & Lee, Reference Yu and Lee2013); voters in these countries may be less likely to object to women candidates. On the other hand, sexist men in gender-egalitarian countries are less likely than in gender-inegalitarian countries to endorse participating in collective actions to benefit women, such as supporting gender-egalitarian politicians (Kosakowska-Berezecka Reference Kosakowska-Berezecka, Besta, Bosson, Jurek, Vandello, Best, Wlodarczyk, Safdar, Zawisza, Żadkowska, Sobiecki, Agyemang, Akbaş, Ammirati, Anderson, Anjum, Aruta, Ashraf, Bakaitytė, Bi, Becker, Bender, Bërxulli, Bosak, Daalmans, Dandy, de Lemus, Dvorianchikov, Etchezahar, Froehlich, Gavreliuc, Gavreliuc, Gomez, Greijdanus, Grigoryan, Hale, Hämer, Hoorens, Hutchings, Jensen, Kelmendi, Khachatryan, Kinahan, Kozlowski, Lauri, Li, Maitner, Makashvili, Mancini, Martiny, Milošević Đorđević, Moreno-Bella, Moscatelli, Moynihan, Muller, Ochoa, Adebayo, Pacilli, Palacio, Patnaik, Pavlopoulos, Piterová, Puzio, Pyrkosz-Pacyna, Rentería-Pérez, Rousseaux, Sainz, Salvati, Samekin, García-Sánchez, Schindler, Sherbaji, Sobhie, Sulejmanović, Sullivan, Torre, Torres, Ungaretti, Valshtein, Van Laar, van der Noll, Vasiutynskyi, Vohra, Zapata-Calvente and Žukauskienė2020). If a gender-egalitarian context serves as a reminder of women’s encroachment into previously male-dominated spaces, then sexists in relatively equal countries may be particularly confronted by the prospect of women in political office (Kosakowska-Berezecka et al., Reference Kosakowska-Berezecka, Besta, Bosson, Jurek, Vandello, Best, Wlodarczyk, Safdar, Zawisza, Żadkowska, Sobiecki, Agyemang, Akbaş, Ammirati, Anderson, Anjum, Aruta, Ashraf, Bakaitytė, Bi, Becker, Bender, Bërxulli, Bosak, Daalmans, Dandy, de Lemus, Dvorianchikov, Etchezahar, Froehlich, Gavreliuc, Gavreliuc, Gomez, Greijdanus, Grigoryan, Hale, Hämer, Hoorens, Hutchings, Jensen, Kelmendi, Khachatryan, Kinahan, Kozlowski, Lauri, Li, Maitner, Makashvili, Mancini, Martiny, Milošević Đorđević, Moreno-Bella, Moscatelli, Moynihan, Muller, Ochoa, Adebayo, Pacilli, Palacio, Patnaik, Pavlopoulos, Piterová, Puzio, Pyrkosz-Pacyna, Rentería-Pérez, Rousseaux, Sainz, Salvati, Samekin, García-Sánchez, Schindler, Sherbaji, Sobhie, Sulejmanović, Sullivan, Torre, Torres, Ungaretti, Valshtein, Van Laar, van der Noll, Vasiutynskyi, Vohra, Zapata-Calvente and Žukauskienė2020).
Second, between-country differences exist regarding political polarization. Since the 1980s, the extent to which citizens feel negatively towards parties other than their own – ‘affective polarisation’ – has increased more rapidly in the USA than any other OECD country (Boxell et al., Reference Boxell, Gentzkow and Shapiro2024). Polarization increased relatively little in Canada during this period and declined in Australia and Britain (Boxell et al., Reference Boxell, Gentzkow and Shapiro2024). Because out-party aversion is a stronger predictor of voting behaviour than in-party fondness (Finkel et al., Reference Finkel, Bail, Cikara, Ditto, Iyengar, Klar, Mason, McGrath, Nyhan, Rand, Skita, Tucker, Van Bavel, Wang and Durckman2020), it is therefore possible that sexist partisans in the highly polarized USA will be more likely to support a woman candidate than an equally sexist voter in a less-polarized country.
Third, although voting is voluntary in Canada, the UK, and the USA, it is compulsory in Australia. The impact that compulsory voting may have on women candidates is unclear. One of the few comparative studies to directly address this question is Studlar and McAllister (Reference Studlar and McAllister2002), who found that OECD countries with compulsory voting between 1950 and 2000 had a 3.5 per cent reduction in women’s representation, although the authors do not suggest an explanation for this result. On the one hand, compelled voters spend less time seeking out and engaging with political information (Singh and Roy Reference Singh and Roy2018); this could potentially undermine women candidates who fare worse than their male counterparts in certain low-information contexts (McDermott Reference McDermott1997). On the other hand, compulsory voting means Australians will not have the option to abstain from voting, which could potentially benefit women candidates seeking the support of sexist co-partisan voters. There is also a possibility that compulsory voting could benefit women candidates by increasing the number of female voters (Reeves and Smith Reference Reeves and Smith2024). This explanation, though, is premised on ‘affinity’ effects, whereby voters support candidates who are descriptively similar to themselves; while some studies (for example, Dolan Reference Dolan1998) suggest that women voters support women candidates, this has not been borne out by more recent research (for example, Ono and Burden Reference Ono and Burden2019).
Finally, there are between-country and between-party differences in candidate selection procedures with gendered effects. In particular, the Australian Labour Party, the New Democratic Party (Canada), and the Labour Party (Britain) use party gender quotas. Research suggests that party gender quotas can increase the quality of female candidates by fostering internal party infrastructure that aims to increase women’s political skills (Beauregard and Taflaga Reference Beauregard and Taflaga2023), and by incentivising parties to closely scrutinize the characteristics of all candidates rather than falling back on traditional male networks (Aldrich and Daniels Reference Aldrich and Daniel2024). Although it is possible that gender quota parties may produce particularly high-calibre female candidates, there is no direct evidence to suggest that this will alter the vote choices of sexists. We reflect in the results section on the possible implications of party quotas.
Data
Case Selection
We focus on four democracies: Australia, Britain, Canada, and the USA. Most studies that explore attitudes toward women candidates focus on the US, but those findings do not necessarily translate elsewhere, as we have shown: cross-national research is important (De Geus et al. Reference De Geus, Ralph-Morrow and Shorrocks2022; Beauregard Reference Beauregard2021). The four countries have different histories of women’s representation, especially in the executive. Australia, Britain, and Canada have had a female head of government, though the duration differs starkly.Footnote 2 The countries score differently on measures like the GGGI. But the countries have several helpful similarities. All four legislatures operate under district-based majoritarian electoral systems. All four electoral systems are based on candidate rather than party choices. And, as Figure 1 shows, the patterns of women’s representation in the legislatures are similar. There are small differences in levels – the USA lags behind – but each country shows a slow increase in the female percentage. Importantly, it appears that growth in all four has stagnated at around 35 per cent. This reinforces the need to understand women’s persistent under-representation in politics in these democracies.
Women’s Representation in Legislature, 2002–2022.
Data from World Bank Development Indicators, proportion of seats held by women in national legislatures (per cent).
Data Collection
Candidate Data
We collected data on candidates in the 2019 Australian federal election, the 2019 British general election, the 2019 Canadian federal election and the 2020 US House of Representatives election. For Australia, we used data on candidates, district, party affiliation, and votes received from the Australian Electoral Commission, which we coded to include candidate gender and incumbency status. For Britain, we used House of Commons Library data (2020), which lists name, party affiliation, gender, constituency, incumbency status, and vote share of candidates. For Canada, we drew on Johnson et al.’s (Reference Johnson, Tolley, Thomas and Bodet2021) dataset, which lists vote shares, candidate party affiliation, gender and incumbency status. Our US analysis uses a Centre for American Women and Politics (CAWP) (2020) dataset, which focuses on Congressional women candidates and includes district, party affiliation, and race; we also coded incumbency status and vote share and added equivalent information for male candidates. For all countries, we added party performance at the previous election at the district level, measured by the distance a candidate’s party was from the previous winning candidate (Australia 2016; Britain 2017; Canada 2015; US 2016). Focusing on the main parties in contention (see Table 2), our four independent candidate datasets comprise 5,346 candidates: Australia, 462; Britain, 2,407; Canada, 1,421; USA, 1,056.
Individual-level Data
We matched these candidate datasets to existing data on voters, linking survey respondents to candidates who stood in their district. In all countries, we chose datasets that included questions from Glick and Fiske’s (Reference Glick and Fiske1996) ASI. The datasets also needed to contain information on the constituency/district where respondents voted, which we used to map candidates to voters. For Australia, we used the cross-sectional Australian Election Study (McAllister et al. Reference McAllister, Bean, Gibson, Makkai, Sheppard and Cameron2019). For Britain, we used the British Election Study Internet Panel (BESIP) Wave 19 (Fieldhouse et al. Reference Fieldhouse, Green, Evans, Mellon and Prosser2020) from December 2019. For Canada, we used the 2019 Canadian Election Study (CES) Online Survey (Stephenson et al. Reference Stephenson, Harell, Rubenson and Loewen2020), including pre-election and post-election waves. For the USA, we used the 2020 Cooperative Election Study (CES) (Schaffner et al. Reference Schaffner, Ansolabehere and Luks2021), which includes pre-election and post-election waves. After matching voter data to candidate data, we had complete observations on approximately 26,000 voters deciding on 5,346 candidates across four countries.
Methods
Vote Choice Models
To test the Woman Candidate and Sexism hypotheses, we use conditional logit models. Standard logit regression models are inappropriate here because voters are choosing between several candidates, and often multiple candidates are women. We therefore transform the data into ‘long’ format, with as many observations per respondent as candidates in their district. The dependent variable in the first set of models is vote choice, which is 1 if the respondent votes for the candidate, 0 if not. The conditional logit model estimates the within-individual effect of candidate-level characteristics on voting for that candidate, analogous to an individual fixed-effects model. Because these are within-individual effects, we do not control for variables that do not vary within an individual (socio-demographics, attitudes) because these would drop out of the analysis. We control for candidate party, which should capture party-based reasons for candidate selection, incumbency status, and party performance at the previous election, which capture potential differences in candidacy types for male and female candidates.
Independent Variables
Our key independent variables in the vote choice models are candidate gender and voter sexism. Candidate gender is a binary variable: female or male. We collected data on non-binary candidates, but there were too few for meaningful analysis, so they are excluded. Voter sexism is measured using hostile-sexism items from the ASI, listed below (Table 1). Respondents answered on a Likert scale from (1) Strongly Disagree to (5) Strongly Agree; we then reverse-coded where appropriate so that higher numbers consistently indicated higher levels of sexism. Although this is a slimmed-down list, these statements have been associated with party choice in Britain (De Geus et al. Reference De Geus, Ralph-Morrow and Shorrocks2022), Australia (Beauregard Reference Beauregard2021), and the USA (Cassese and Barnes Reference Cassese and Barnes2019).Footnote 3 All options have been found to load strongly onto the same factor, and, except for the ‘interpreting innocent remarks as sexist’ item, are recommended for a reduced version of the hostile-sexism scale (Schaffner Reference Schaffner2022). We created a scale for hostile sexism and coded it to run from 0–2, where <1 indicates non-sexist responses, >1 sexist, and 1 overall neutral. For our analytical sample, the scales have the following Cronbach’s alpha in each country: Britain: 0.75; Australia: 0.83; Canada: 0.68; USA: 0.80.
Hostile Sexism Survey Items
Control Variables
Our control variables, measured at the candidate level, are intended to include factors which might impact a respondent’s vote choice other than candidate gender or sexist attitudes. They include candidate party, incumbency status, and party performance at the previous election in the district. Table 2 has the full coding. Due to the nature of the conditional logit models, we exclude individual-level control variables (for example, respondent sex, education level), and our estimates should be treated as within-individual effects.
Coding of the Control Variables for Vote Choice Models
We confine our conditional logit analysis to voters and to a sample with complete answers on all variables listed below. We also restrict our analysis to districts where voters could choose between a man or a woman. Table 3 shows the number of districts in each election and the sample with a mixed-sex race, which we use for our analysis. Because of the USA’s high number of men-only races and the exclusion of districts where a candidate ran unopposed (which only applies in the US), we have substantially fewer US districts compared to the total.
Total and sample districts in each election
After excluding all-male or all-female races, we are left with the following sample sizes of unique voters and total observations (voters by candidates in district) in each country: Australia: 1,467 voters, 4,558 observations; Canada: 3,488 voters, 14,477 observations; Britain: 5,495 voters, 21,387 observations; USA: 15,407 voters, 30,822 observations.Footnote 4
Abstention Models
To test the Abstention hypothesis, we use standard logit models with a dependent variable: 1 if a respondent abstained, 0 if they voted. For Britain, respondents in the post-election survey were asked if they voted in the 2019 general election: 9.8 per cent said no, a strong under-reporting of abstention rates given the recorded turnout of 67.3 per cent.Footnote 5 Similarly, in Canada, we use a post-election survey self-reported turnout question with a reported abstention rate of 7 per cent, again a strong under-reporting given the recorded turnout of 67 per cent.Footnote 6 In Australia, we use a combined measure of abstention or an informal (spoiled) ballot, given Australia’s compulsory voting. The 5.1 per cent abstention rate is reasonably close to the 91.89 per cent registered turnout.Footnote 7 For the USA, we can use a validated turnout measure available in the CES, which records whether there is a valid voting record for a respondent for 2020. Here, there is a chance of slight over-estimation of abstention, given that ‘matches are only made with records when there is a high level of confidence that the respondent is being assigned to the correct record’.Footnote 8 The measure also records whether there is a voting record; it’s possible someone voted only for the presidential ticket and abstained from the House ticket, but this cannot be captured by the variable. Nevertheless, the resulting abstention rate for this variable in our models is 35.5 per cent, mapping closely to the reported turnout of 66.8 per cent.Footnote 9 This is consistent with previous research, which finds that using vote validation can address the problem of social desirability bias when measuring turnout via surveys, and that levels of self-reported turnout are reduced when using validated vote measures (DeBell et al. Reference DeBell, Hillygus, Shaw and Valentino2024). It should be noted that since those who vote are more likely to respond to surveys (DeBell et al. Reference DeBell, Krosnick, Gera, Yeager and McDonald2020), there is likely systematic bias in our measure of abstention, especially in the three countries for which we do not have validated vote measures. We return to this issue in the conclusion.
We restrict the abstention models to voters with a co-partisan candidate in their district, and thus also restricting the models to partisan voters. This reduces our sample sizes to the following compared to the vote choice analysis described above: Britain: 4,098/5,495; Australia: 1,003/1,467; Canada: 3,209/3,488; USA: 10,595/15,407. This allows us to identify if voters who score high on sexism are more likely to abstain when the candidate from their preferred party is female. To identify co-partisanship between voters and candidates, we use whether respondents identified with or felt closest to the candidate’s party. The measure of party identification varied across datasets, and the full question wording can be found in the supplementary information. The main independent variable here is sexism, measured the same way as for the conditional logit models, which we interact with candidate gender. We add several control variables. A recent meta-study of voter turnout across forty-four countries suggests several key factors: ‘competitive elections, concurrent elections, economic globalization, inflation, previous turnout, proportional representation, spending decentralization, and some geographical dummies’ (Frank & Martínez i Coma Reference Frank and Martínez i Coma2023, 630; see also Blais Reference Blais2000). As we run the models separately for each country, we do not control for factors such as the electoral system, which do not vary within countries, but we do include regional dummies.Footnote 10 We also use individual-level variables that often impact turnout. Recent studies predominantly point to education and measures of economic hardship or economic circumstances as crucial predictors (Wilford Reference Wilford2020; Kostelka and Blais Reference Kostelka and Blais2021; Blais Reference Blais2000). We therefore control for age, education, employment status, union membership, and social class. In Britain, we use social grade; in Canada, the USA, and Australia, we use household income as a proxy for social class. Finally, we control for candidate incumbency status and party performance at the previous election as other potential factors in partisans’ abstention decisions.
Results
Before running our regression models, we look at the descriptive distribution of our key independent variables – candidate gender and hostile sexism – across four countries. Table 4 compares levels of sexism in Australia, Britain, Canada, and the USA for respondents in our analytical sample.Footnote 11 Nearly one-third fall on the hostile-sexist end of the scale (above 1), with Canada lowest (22 per cent) and Britain highest (33 per cent). Hostile sexists are a significant minority in every country. Canada shows the highest percentage on the not-hostile-sexist end of the scale (63 per cent), compared to Australia (56 per cent), the USA (59 per cent), and Britain (48 per cent). Whilst the differences are not large, Canada shows the lowest levels of hostile sexism and Britain shows the highest.
Levels of Hostile Sexism in Australia, Britain, Canada, and the USA (per cent)
We also compare women candidates’ profiles across the countries. Table 5 shows the percentage of women by party, incumbency, and election outcome – note these are for the main parties in each country and do not include minor or independent candidates. A higher proportion were women in Canada (41 per cent) than elsewhere (Britain 38 per cent; Australia 37 per cent, the USA 35 per cent). However, the figures suggest that female candidates are standing in different types of seats. In all countries, women make up a smaller share of incumbents than they do candidates overall, with the smallest disparity in Britain and the largest in Canada. Similarly, women are under-represented amongst elected candidates compared to their overall numbers, with the disparity again lowest in Britain and highest in Canada. This may be partly because women are placed in seats where their party struggled more, especially in the USA and Canada. In the USA, the average party distance from the winner was 15 points for women and 12 for men. In Canada, the difference is more striking, with women running in seats where their party had an average distance from the winner of 42 points, compared to 34 for men. This replicates previous findings from Canada that female candidates are placed in less ‘winnable’ seats (Thomas and Bodet Reference Thomas and Bodet2013).
Profile of Women Candidates in USA (2020), Britain (2019), Canada (2019), and Australia (2019)
The figures in parentheses give the number of women candidates in each category.
In Britain and Australia, women were not placed in less ‘winnable’ seats in terms of the distance from the winner at the previous election – although there is evidence from previous research that this may vary by party (Curtice et al. Reference Curtice, Fisher, English, Ford, Jennings, Bale and Surridge2021). However, women were still elected in lower numbers than would be expected given their proportion amongst all candidates. This could be because in these countries, the centre-right party (Conservatives in Britain; Liberal/National coalition in Canada) won the election. These parties, as also shown by Table 4, and as previously discussed, do not have gender quotas and field fewer women candidates than other parties, resulting in fewer women elected in these elections. In all countries, left-wing (Labour/NDP/Democrats) and Green parties fielded more women than did right-wing or Conservative parties. This is consistent with previous research (for example, O’Brien Reference O’Brien2015).
Support for Women Candidates
We now move to the analysis of vote choice. First, we test whether female candidates hold an electoral advantage amongst voters (‘Woman Candidate hypothesis’). This is a salient finding from recent experimental literature (Schwarz and Coppock Reference Schwarz and Coppock2022). Table 6 shows the model results, and Figure 2 plots the average marginal effect of voting for a woman over a man in the models with and without the control variables. There is no advantage or disadvantage for women candidates in Britain and Australia, in models with and without the control variables. In Canada, there is a significant disadvantage in the first model without controls (negative and statistically significant marginal effect), but this disappears once we add controls for party, incumbency and previous party performance. In the USA, there is a disadvantage for women candidates in the model without controls, but a small positive advantage (positive and statistically significant marginal effect) when the controls are added. It is therefore only in the USA where voters are more likely to support women, once we take into account systematic differences between male and female candidates. In the other countries, there is no effect of candidate gender on vote choice. This suggests that experimental-study results which find pro-woman bias amongst voters may not travel to all country contexts or to many real-world scenarios.
Regression Models for Voting for Women Candidates in Australia, Britain, Canada, and the USA
* p<0.05; ** p<0.01 – Models presented are conditional logit models, coefficients are odds ratios, full models available in supplementary information (pp.2-3).
Regression Models for Voting for Women Candidates by Voter Sexism, Candidate Gender and Candidate Party in Australia, Britain, Canada and the USA
* p<0.05; ** p<0.01 – Models are conditional logit models, coefficients are odds ratios, full models available in supplementary information (pp.4-7). Base category political parties are Australia = Liberal-National Coalition; Canada = Conservative; UK = Conservative; USA = Republicans. In our final set of models (Table 8), we measure sexism’s effect on abstention when partisan voters face a female co-partisan candidate. The Abstention hypothesis proposed that partisan sexist voters will abstain rather than vote for a woman from their party or a man from another. Because here we are interested in the decision to vote or not, rather than party choice, we exclude the party candidate and the sexism interaction. We instead focus on the interaction between sexism and candidate gender to test our hypothesis. Importantly, abstention rates are under-reported in the British and Canadian surveys, especially; the variable in the USA is likely most accurate in capturing non-voting. Again, we restrict our sample to races where voters face a male and female candidate. Table 8 shows a significant main effect of sexism on abstention in Australia, Britain, and Canada. This suggests that here, (partisan) sexists are more likely to abstain in general than (partisan) non-sexists. We also see descriptively that abstainers (both partisan and non-partisan) have higher levels of sexism than voters in all countries. In the USA, where we have the most reliable measure of abstention, 32 per cent of abstainers are hostile sexists versus 26 per cent of voters. The equivalent figures for the other countries are Britain: 42 per cent/33 per cent; Australia: 40 per cent/29 per cent; Canada 30 per cent/22 per cent. It is notable that we see the same pattern in Australia, albeit with much lower levels of reported abstention, even with compulsory voting.
Average marginal effect of voting for a female candidate versus a male candidate, calculated from models in Table 6.
The variable which makes the difference between uncontrolled and controlled models in the USA and Canada is party performance at the previous election at the district level. Given that women in the USA and Canada are more likely to run in seats where their parties are less competitive, as discussed above, this means contextual factors need to be taken seriously as at least a partial explanation for women’s lack of electoral success. This points to explanations which emphasize party-selection procedures (Bjarnegard and Kenny Reference Bjarnegard and Kenny2016) rather than voter bias as the key reason.Footnote 12
Second, we test the Sexism hypothesis, analysing the effect of voter sexism and its interaction with candidate gender and party in Table 7. In the first column for each country, we see a statistically significant interaction between candidate gender and voter sexism: sexists are less likely to vote for women in Britain and the USA, but not in Canada or Australia. At face value, this seems to confirm that voter sexism negatively affects women’s electoral chances, at least in Britain and the USA. In the second set of models, however, we add an interaction between candidate party and voter sexism. Previous studies show that sexism is associated with party choice in these countries (Cassese and Barnes Reference Cassese and Barnes2019; De Geus et al. Reference De Geus, Ralph-Morrow and Shorrocks2022; Beauregard Reference Beauregard2021), and different parties also field women candidates at very different rates. When we include this interaction, the significant interaction between candidate gender and voter sexism in Britain and the USA disappears, suggesting this initial result was because of the association between voter sexism and party choice. The UK Conservatives and US Republicans are much less likely to field women candidates, so the association between sexism and support for these parties confounded the relationship between sexism and candidate choice by gender. Since the baseline party in Table 7 is the Conservative or Republican Party option, we observe that highly sexist voters are significantly less likely to vote for a candidate who represents more left-of-centre, regional or Green options. We also see this null result when we run the models for men and women separately.
Regression Model for Abstention in Australia, Canada, Britain and the USA
* p<0.05; ** p<0.01 – Models are logistic regression models, full models available in supplementary information (pp.12-16). Coefficients are odds ratios, Abstention is coded as 1 = Abstention, 0 = Voted.
There are reasons to believe partisanship may be an important variable in the relationship between vote choice and candidate gender: bias against women is pronounced amongst voters without a strong partisan attachment (Fulton Reference Fulton2014; Ono and Burden Reference Ono and Burden2019; Winter, Reference Winter2023), likely because they have no partisan identity to rely on and may use other candidate characteristics. In the supplementary information (pp. 8–11), we run these models for those with and without partisan identification, enabling us to test whether sexist non-partisans might be less likely to vote for a woman because they do not have a partisan tie to a party and can thus more easily switch their vote. This does not change the results for any country.
We also run the models for those who identify with the main right-wing and left-wing parties (supplementary information, pp. 8–11), to identify if there are differences between party supporters in how sexism affects their likelihood to vote for women. We see no difference between these groups in any country except Canada. In Canada, once we disaggregate our sample based on party, we see that Conservative party identifiers are less likely to vote for women if they score higher on hostile sexism, with a statistically significant interaction at p<0.05. This is the only subsample across the countries that supports our hypothesis that sexists will be less likely to vote for women. We therefore interpret this with caution, but suggest that further research is warranted on how sexism affects support for women candidates in Canada (such as recent work by Chen et al. Reference Chen, Thomas, Harell and Gosselin2023; Mansell et al. Reference Mansell, Harell, Thomas and Gosselin2022).
In our final set of models (Table 8), we measure sexism’s effect on abstention when partisan voters face a female co-partisan candidate. The Abstention hypothesis proposed that partisan sexist voters will abstain rather than vote for a woman from their party or a man from another. Because here we are interested in the decision to vote or not, rather than party choice, we exclude the party candidate and the sexism interaction. We instead focus on the interaction between sexism and candidate gender to test our hypothesis. Importantly, abstention rates are under-reported in the British and Canadian surveys, especially; the variable in the USA is likely most accurate in capturing non-voting. Again, we restrict our sample to races where voters face a male and female candidate. Table 8 shows a significant main effect of sexism on abstention in Australia,Britain, and Canada. This suggests that here, (partisan) sexists are more likely to abstain in general than (partisan) non-sexists. We also see descriptively that abstainers (both partisan and non-partisan) have higher levels of sexism than voters in all countries. In the USA, where we have the most reliable measure of abstention, 32 per cent of abstainers are hostile sexists versus 26 per cent of voters. The equivalent figures for the other countries are Britain: 42 per cent/33 per cent; Australia: 40per cent/29 per cent; Canada 30 per cent/22 per cent. It is notable that we see the same pattern in Australia, albeit with much lower levels of reported abstention, even with compulsory voting.
However, only in the USA do we see a statistically significant interaction between sexism and having a woman co-partisan candidate, where sexist partisans are more likely to abstain. Either the hypothesized relationship exists only in the USA, or this finding is because we have a much more accurate measure of abstention there. We also plot the predicted probability of abstaining in each country for those faced with a co-partisan male or female candidate (Figure 3). In Australia, Britain, and Canada, we see that those at the sexist end of the scale (>1) are about 10 percentage points more likely to abstain than those at the lower end (<1), but this is not statistically significantly different between male and female candidates. In the USA, the probability of voting for a man remains the same across different levels of sexism, but sexists are about 20 percentage points more likely to abstain when faced with a co-partisan woman than non-sexists are.
Predicted probability of abstaining for those with co-partisan female versus co-partisan male candidates.
Conclusion
Experimental studies on gender bias increasingly suggest that women candidates do not face electoral penalties and may even enjoy an advantage amongst voters. However, sexist attitudes have been important when understanding party choice in recent elections. Missing from the literature is a focus on candidate choice, using real-world data on women candidates and voter sexism. Relying on four nationally representative voter surveys and four candidate datasets, we provide evidence on whether voters support men and women candidates at different rates, and the extent to which sexist attitudes are associated with willingness to vote for women.
First, we find that political context structures electoral choice in important ways often not captured by experiments. Political parties differ substantively in the number of women they run; US voters are especially likely to face only male candidates, while women often run in harder-to-win seats in Canada and the USA. Thus, conjoint or vignette experiments offering equal choices between male and female candidates fail to capture important real-world features of elections where women are under-represented amongst candidates, run predominantly for left-wing parties, and contest more difficult seats. We should therefore observationally analyse voter willingness to support women candidates, to further illuminate how candidate gender conditions vote choice in real-world scenarios.
Second, after analysis of real-world candidate choice, we find no electoral penalty toward women – but also no electoral bonus, as some meta-analyses suggest (Schwarz and Coppock Reference Schwarz and Coppock2022), except in the USA, where women enjoy a slight favour. This is perhaps consistent with evidence that US women candidates appear to be more qualified than men (for example, Fulton Reference Fulton2012). Evidence on gender differences in candidate qualifications or experience in our other countries is limited, although Allen et al. (Reference Allen, Cutts and Campbell2016), using All-Women Shortlists in Britain, find no evidence of difference in candidate quality between those elected and other MPs (male and female), perhaps suggesting limited differences between men and women generally in this context. Future research could usefully understand how much men and women differ in qualifications, experience and expertise outside the USA, to contextualize the finding that voters are equally likely to support women at the ballot box on average.
Our findings could be taken positively: voters are not biased against women candidates, so parties should have no concerns about putting forward women in elections where voters vote for a specific candidate. But there is evidence, at least from Britain and Australia, that survey respondents sometimes report a higher likelihood to vote for women in experimental studies (Schwarz and Coppock Reference Schwarz and Coppock2022; Vivyan and Wagner Reference Vivyan and Wagner2015). This suggests voters may be biased somehow against women – their hypothetical preference does not extend to ‘real-world’ women. This shows the need to use observational studies along with experimental ones to appreciate how much external validity experimental studies have. Our results suggest the external validity of candidate choice experiments, which investigate gender, might be limited here, or at least reveal little about voter propensity to support women in actual elections.
Third, our results confirm the importance of sexism for understanding elections in the countries we study, although differently from expected. We examine, novelly, the impact of sexist attitudes on candidate rather than party support. However, we found sexism much more relevant for party support than for candidate choice, and we found little evidence that sexists are less likely than non-sexists to vote for women. This is interesting given the above discussion; if our findings indicate voter bias against women in real-world elections, this suggests this bias is exhibited by sexist and non-sexist voters – perhaps stemming from more engrained gender stereotyping rather than voter sexism.
We find evidence across all countries that sexists are more likely to vote right-wing. This is consistent with research using models which control for other values and attitudes (Beauregard Reference Beauregard2021; Cassese and Barnes Reference Cassese and Barnes2019; De Geus et al. Reference De Geus, Ralph-Morrow and Shorrocks2022). Right-wing voters are thus more likely to hold sexist attitudes. However, right-wing parties would be wrong to infer that these voters do not support women candidates. Our evidence suggests they are just as willing to do so as their non-sexist counterparts (except perhaps in Canada), likely because party considerations are more important, and they assume that a woman in their preferred party shares their values. This raises interesting questions about what happens to women candidates on the right who make feminist claims or attempt to represent women substantively. Future research could explore whether right-wing or sexist voters are less likely to vote for such candidates.
Fourth, we show that abstaining is a potential option for voters who might be ‘cross-pressured’ by their values and partisanship (De Geus Reference De Geus2019). Partisan sexists in the USA faced with a woman co-partisan candidate are more likely to abstain than partisan non-sexists, even when controlling for other determinants. So, although women in the USA enjoy a slight electoral advantage, their candidacy might still depress turnout amongst some voters, even if those voters do not go as far as switching parties. We find no evidence of this elsewhere, but note that abstention is measured more robustly in US data, and compulsory voting in Australia may make this question less relevant. Though observational studies should be used more in this area, perhaps experimental studies should include abstention to understand its extent in other contexts, since abstention rates are difficult to measure. Future studies may also want to further probe this US paradox – where women candidates seemingly face both an electoral advantage and turnout disadvantage – via an intersectional analysis which highlights how candidate race interacts with candidate gender (see Van Oosten et al. Reference Van Oosten, Mügge and van der Pas2024). Race is highly salient to US politics (Stephens-Dougan Reference Stephens-Dougan2021), and a record-breaking number of women of colour stood in the 2020 congressional elections (Dittmar Reference Dittmar2021). Existing experimental studies which focus on evaluations of minoritized women candidates vis-à-vis their white counterparts have produced mixed results (Gershon and Monforti Reference Gershon and Monforti2019; Schwarz and Coppock Reference Schwarz and Coppock2022), thereby underscoring the need for further research in this area.
We also add a comparative element to work examining sexism and voting behaviour, which has so far focused on single-country case studies. We show that whilst levels of sexism vary across the four countries, they are roughly comparable, with around one-third of voters on the hostile-sexist end in each. Our results do not clearly show any differences according to the level of gender equality in each country, nor do we see partisanship exerting more influence on candidate choice in the USA, as we might have expected. Instead, we see US exceptionalism: the only country with an advantage for women candidates overall. This may be because much fewer races there feature women candidates as a proportion of the whole compared to the other countries, where most districts have at least one. Previous research has identified that women in the USA run in more ‘women-friendly’ districts in terms of their demographics (for example, Ondercin and Welch Reference Ondercin and Welch2009), and so voter characteristics in places where women stand could be driving the positive result we observe. We note that we only focus on a single election for our four cases, and further research on the extent to which these patterns persist in subsequent elections would be useful.
We have shown that the extent to which voters support women candidates is more complicated than experimental studies show. Experimental studies do not capture the real-world variation in women candidates across parties and districts. Nor is the positive advantage women candidates enjoy in many experimental studies replicated in three of our four countries, perhaps indicating voters’ greater willingness to support women hypothetically but not in reality. We find little evidence that sexism conditions how much voters support male over female candidates, indicating that any bias is rooted in other attitudes or psychological processes. Future research should work towards integrating findings from experimental and observational studies to fully understand the extent to which women candidates face bias at the ballot box.
Supplementary Material
Supplementary material for this article can be found at https://doi.org/10.1017/S0007123425000390.
Data Availability Statement
This study’s datasets are available from the following repositories:
Britain
Fieldhouse E, Green J, Evans G, Mellon J and Prosser C (2020) British Election Study Internet Panel Wave 19. V1.0-2 https://doi.org/10.5255/UKDA-SN-8810-1. https://www.britishelectionstudy.com/data-objects/panel-study-data/
House of Commons Library (2020). Constituency Data: Election Results. https://commonslibrary.parliament.uk/research-briefings/cbp-8749/
House of Commons Library (2017). Data File: Detailed Results by Constituency. https://commonslibrary.parliament.uk/research-briefings/cbp-7979/
Canada
Stephenson L, Harell A, Rubenson D and Loewen PJ (2020) 2019 Canadian Election Study - Online Survey. V1 https://doi.org/10.7910/DVN/DUS88V. http://www.ces-eec.ca/2019-canadian-election-study/
Johnson A, Tolley E, Thomas M and Bodet MA (2021) A new dataset on the demographics of Canadian federal election candidates. Canadian Journal of Political Science 54(3), 717–725. V1. https://doi.org/10.1017/S0008423921000391. https://dataverse.harvard.edu/dataset.xhtml?persistentId=doi:10.7910/DVN/MI5XQ6
Results for 2015 from Elections Canada: https://www.elections.ca/content.aspx?section=res&dir=rep/off/42gedata&document=summary&lang=e
Australia
McAllister I, Bean C, Gibson R, Makkai T, Sheppard J and Cameron S (2019). Australian Election Study, 2019. https://doi.org/10.26193/KMAMMW, ADA Dataverse, V2. https://dataverse.ada.edu.au/dataset.xhtml?persistentId=doi:10.26193/KMAMMW
For full candidates’ data, see Shorrocks R, Ralph-Morrow E, de Geus R (2025) Replication Data for: Voter Sexism and Electoral Penalties for Women Candidates: Evidence from Four Democracies, https://doi.org/10.7910/DVN/AKYBJJ, Harvard Dataverse.
USA
Schaffner BF, Ansolabehere S and Luks S (2021) Cooperative Election Study Common Content, 2020. Harvard Dataverse, Release 2. https://doi.org/10.7910/DVN/E9N6PH.
MIT Election Data and Science Lab, 2017, ‘U.S. House 1976–2022’. https://doi.org/10.7910/DVN/IG0UN2, Harvard Dataverse, V13.
Center for American Women and Politics (CAWP) (2020). Congressional Women Candidates Database. https://live-ru-cawp-next.pantheonsite.io/election-watch/past-candidate-and-election-information
For full candidates’ data, see Shorrocks R, Ralph-Morrow E and de Geus R (2025) ‘Replication Data for: Voter Sexism and Electoral Penalties for Women Candidates: Evidence from Four Democracies’. https://doi.org/10.7910/DVN/AKYBJJ, Harvard Dataverse.
Acknowledgements
The authors would like to thank Stephen Fisher, the Politics Departmental Research Seminar at the University of Oxford, and the three anonymous reviewers for their helpful comments. Roosmarijn de Geus wishes to acknowledge that the research and the majority of writing for this publication were conducted during her time as a lecturer at the University of Reading.
Financial support
Support for this research was provided by the Leverhulme Trust ECF-2017-081 and University College London’s Department of Political Science’s Departmental Research Funds.
Competing interests
None [to disclose].
Open access
