Gender and LGBT Affinity: The Case of Ontario Premier Kathleen Wynne

Abstract

When a party selects an out lesbian as its leader, do women and LGBT people evaluate that leader more positively? And do they become more likely to vote for that party? We answer these questions using the case of Kathleen Wynne, premier of Ontario, Canada, from 2013 to 2018. We draw on four large-sample surveys conducted by Ipsos before and after the 2011 and 2014 Ontario elections. We compare shifts in best premier choice and vote choice among non-LGBT men, non-LGBT women, LGBT men, and LGBT women from 2011 to 2014. We find gender and LGBT affinity in leader evaluations. However, we find that only non-LGBT women and LGBT men were more likely to vote Liberal after Wynne became leader. This article contributes to research on affinity by examining LGBT affinity in a real-world election and the intersection of gender and LGBT affinity.

A commonplace assumption is that members of marginalized groups are more favorable to candidates and leaders who share their identity or identities. Although scholars have given considerable attention to gender and racial/ethnic affinity, little work has considered LGBT affinity.¹ Everitt and Horvath (2021) provide experimental evidence that lesbians and gay men are more likely to support lesbian and gay candidates. However, we know from work on other groups that it is possible for experiments to show evidence of affinity even when real-world elections do not. In addition, although work has increasingly examined the intersection of gender and racial/ethnic affinity, we know little about the intersection of gender and LGBT affinity. When a party selects an out lesbian as its leader, do women and LGBT people evaluate that leader more positively? And do they become more likely to vote for that party?

These questions are difficult to address for three reasons. First, many surveys do not ask questions about LGBT identities. Second, many surveys that ask about LGBT identities do not have enough LGBT respondents to examine LGBT affinity. Third, many parties have never had an out LGBT leader. We overcome these challenges using the case of Kathleen Wynne, former premier of the Canadian province of Ontario (2013–18). She was both the first woman premier of Ontario and the first LGBT first minister anywhere in Canada. Because we have large-sample surveys with an LGBT identity question that were run before and after she became leader, we have a rare opportunity to examine gender and LGBT affinity by examining voters’ evaluations of her and their willingness to vote for her party.

We pool four large-sample surveys conducted by Ipsos before and after the 2011 and 2014 Ontario elections (N = 16,818 in 2011 and N = 13,100 in 2014). This period covers the last Ontario election before Wynne became the Liberal leader and the first election in which she was the leader. These surveys are unusual for the time period in asking respondents whether they are LGBT.² They also have substantial numbers of LGBT respondents given their large sample sizes (N = 515 in 2011 and N = 536 in 2014). We estimate shifts from 2011 to 2014 among non-LGBT men, non-LGBT women, LGBT men, and LGBT women in leader evaluations (measured using a question about which leader would be the best premier) and vote choice. The 2011 election is a particularly useful baseline for the 2014 election because the party vote and seat shares did not change very much between the two elections, and the leaders of the other major parties—the center-right Progressive-Conservative (PC) Party and the social democratic New Democratic Party (NDP)—did not change over this period.

We find the expected gender and LGBT affinity patterns for best premier evaluations among non-LGBT women, LGBT men, and LGBT women. However, we only find (relatively) clear evidence for affinity voting among non-LGBT women and LGBT men. Unfortunately, even these large-sample surveys do not have enough LGBT women to say for certain whether LGBT women were more likely to vote Liberal after Wynne became leader or how LGBT women compare to non-LGBT men, non-LGBT women, or LGBT men in their swings in Liberal vote choice. Despite this limitation, our results support the idea that both women and LGBT people can have affinity for a lesbian leader in a real-world electoral context.

We contribute to the literature on affinity in three ways. First, we present evidence that non-LGBT women can feel affinity for a lesbian party leader even though they do not share a sexual identity. In this case, their shared gender identity outweighs any possible prejudice based on sexuality.³ Second, we demonstrate that LGBT men and women have LGBT affinity for a lesbian leader. This suggests that LGBT affinity may exist in other real-world elections. We encourage researchers to examine LGBT affinity in other contexts to determine the conditions under which it manifests. Third, we expand intersectional analyses of gender affinity by looking at the intersection of gender and sexuality. Our findings suggest that it is possible that gender and LGBT affinity in leader evaluations may not translate into vote choice equally for all subgroups. We need both intersectional and disaggregated analyses of gender and sexual identity because not all subgroups of LGBT people behave the same way.

Gender, LGBT, and Intersecting Affinities

There is an established literature examining whether members of marginalized groups are more favorable toward candidates from their own group(s). There are several reasons to expect affinities in candidate evaluations and/or voting behavior. Besco (2019) provides a useful review of possible mechanisms for affinity, which he groups into interest- and identity-based explanations. He identifies three interest-based explanations. To start, voters from a marginalized group might assume that a politician from their group will adopt certain policies, especially those that are favorable to the group. If those voters also support these policies, there will be affinity through policy stereotyping. Second, voters may assume that a politician from their group will have a particular ideology. If the stereotype of the politician matches the opinions of voters from the same group, there will be demographic affinity through ideological stereotyping. Third, if voters think a politician from their own group will act in their best interests in some general and nonspecific way, we may see affinity through a mechanism of general group interest.

Besco (2019) also provides three identity-based explanations of affinity. First, we may see affinity in evaluations of politicians and in voting because of in- group bias. According to social-identity theory, individuals have a need for positive self-esteem. When they evaluate members of the same identity group positively, they fulfill this need. Similarly, individuals may vote for a candidate from the same identity group because they see “victories as shared victories and losses as shared losses” (Besco 2019, 39). Second, if voters are more persuaded by messages from politicians of their own group, we may see affinity through persuasion effects. After all, work shows that the source of a message affects its persuasiveness and that demographic matching can make the source more convincing (Besco 2019). Finally, when voters see vote choice as a way to express their own identity, we may see affinity voting through expressive voting.

Much of the work on affinity focuses on gender. This includes work about voters’ evaluations of and propensity to vote for women candidates, especially in the United States (see, e.g., Badas and Stauffer 2019; Brians 2005; Cook 1994; Dolan 1998, 2008; Paolino 1995; Plutzer and Zipp 1996; Sanbonmatsu 2002; Zipp and Plutzer 1985). Gender affinity work has also been done in other countries, however, including Canada (Cutler and Matthews 2005; Goodyear-Grant 2010; Goodyear-Grant and Croskill 2011). The results for gender affinity are mixed: in some elections, there is gender affinity, while in others there is not, and in yet other cases, men vote for women candidates in greater proportions than women voters (see the discussion in Goodyear-Grant and Croskill 2011, 225). Much of this research focuses on affinity between voters and local candidates rather than between voters and party leaders. However, there is more evidence of gender affinity for party leaders (see, e.g., Banducci and Karp 2000; Cutler 2002; O’Neill 1998). Many of the mechanisms that Besco (2019) identifies could explain gender affinities.

There is relatively little work on LGBT affinity. Although we recognize that different subgroups of LGBT people have different political attitudes and behaviors (Guntermann and Beauvais 2022; Jones 2021; Strolovitch, Wong, and Proctor 2017), LGBT (or GLBT) is a “political coalition and umbrella identity category” that has gained traction since the late 1990s (Murib 2017, 14). We take LGBT as a higher-order coalitional identity that has lower-order subgroup identities (e.g., lesbian, gay, bisexual, or transgender) nested within it. Moreau, Nuño-Pérez, and Sanchez (2019) demonstrate that a substantial number of LGBTQ Latinx respondents express linked fate with the coalitional identity. This makes it plausible to examine affinity with the higher-order coalitional identity (LGBT) rather than only with lower-order subgroup identities.

The main study of LGBT affinity is a survey experiment conducted on Canadian voters (Everitt and Horvath 2021). The authors find that lesbian and gay voters were more likely to support lesbian and gay candidates. Their results are nonsignificant for bisexual, “other” sexual identity, and “other” gender identity respondents, but the point estimates suggest an affinity for lesbian and gay candidates among these other subgroups. Although experiments are useful for isolating the impact of a candidate’s identity or identities on vote choice, they may produce findings that do not appear in observational studies. Experimental work tends to find affinity even where observational work does not (for a discussion of this, see Bird et al. 2016). In the context of real-world elections, candidates are often presented with more information than in an experimental condition. Although we do not know whether LGBT affinity exists in real-world elections, we expect that—as in the case of women—it will be more likely in cases of party leaders than local candidates.

We see several reasons to expect mechanisms that produce affinity to exist in the LGBT case. Past work provides evidence to suggest that interest-based mechanisms for affinity are plausible. LGBT politicians are often stereotyped as more liberal or left-leaning (Jones and Brewer 2019; Loepp and Redman 2022). In contexts in which LGBT voters are also more liberal or left-leaning, including Canada and the United States (Egan 2012; Guntermann and Beauvais 2022; Hertzog 1996; Perrella, Brown, and Kay 2012, 2019), there could be affinity through policy or ideological stereotyping. In addition, Everitt and Raney (2019) show that respondents overall (rather than LGBT respondents specifically) assumed that Wynne would represent the interests of lesbians and gay men better than her competitors. If this was also true among LGBT people, we might expect affinity through the mechanism of general group interest. Although variables related to identity-based explanations have received less attention, we see good reason to think these mechanisms may also produce LGBT affinity. It is possible, for example, that LGBT people’s need for positive self-esteem will lead them to evaluate more positively and/or vote for LGBT politicians. This seems especially plausible given that in-group bias tends to be stronger among smaller groups (Besco 2019).

The Intersectional Turn in Affinity Studies

As with other work on identity, scholars of affinity have increasingly conducted intersectional analyses that account for more than one axis of identity at a time. Much of this work focuses on racialized women voters and candidates. Some scholars hypothesize that racialized women will face a double disadvantage because they face both racism and sexism (Moncrief, Thompson, and Schuhmann 1991). Others, however, suggest that racialized women candidates can benefit from “multiple community identifications,” gaining support as both racialized candidates and as women candidates (Smooth 2006, 411).

Results from both experimental and observational studies of intersectional affinity paint an even more complicated picture. In a fictionalized candidate choice experiment, Philpot and Warton (2007) find some support for the idea that Black women candidates benefit from support as both Black candidates and women candidates, but the affinity disappears when controlling for partisanship. In an experiment in Canada, Goodyear-Grant and Tolley (2019) find that Chinese women voters were not the most favorable group to a (fictional) Chinese woman candidate running against a white man candidate. Instead, the Chinese woman candidate’s strongest supporters were white women, followed by Chinese women, white men, and, finally, Chinese men. In other words, there was a more straightforward gender affinity effect rather than an ethnic affinity effect or combined gender and ethnic affinities. In a study of the 2014 Toronto mayoral election, Bird et al. (2016) examine gender and ethnic affinity voting in a race in which Olivia Chow was the only woman and the only racialized candidate among the three major candidates. Although the authors find strong evidence of ethnic affinity voting, gender was only related to vote choice when it interacted with ethnicity. In fact, white women were less likely than white men to support Chow, even after controlling for egalitarian attitudes, ideology, and partisanship. Overall, work on intersectional affinities shows that sometimes multiple affinities are present, sometimes one affinity dominates, and yet other times there are complicated interactions between affinities.

We contribute to this literature by examining gender and LGBT affinity. Although work has examined affinity at the intersection of gender and race, we do not know much about how gender intersects with sexuality. Our study begins to explore this intersection with the case of Kathleen Wynne, the first woman and the first lesbian premier of Ontario. We consider two possible outcome variables. First, we look at how voters rated Wynne (best premier rating). Second, we look at vote choice (affinity voting).

Hypotheses

If we do see gender and/or LGBT affinity, they may follow different patterns. We consider four main possibilities for gender and LGBT affinity: (1) only gender affinity, (2) only LGBT affinity, (3) both gender and LGBT affinity (that is, a double advantage), and (4) affinity among only LGBT women (that is, a narrow advantage or double disadvantage). We also consider whether affinity in evaluations translates into vote choice.

Non-LGBT women and LGBT men are key to distinguishing among the four main possibilities, because they generate different expectations for these two groups. We do not have clear expectations for these groups from past work, which tends to focus on only one axis of identity or gender combined with axes other than sexuality. Findings from past work on gender, for example, almost certainly include LGBT respondents, but researchers do not often analyze them separately. When researchers pool non-LGBT and LGBT respondents, the results are likely driven by the non-LGBT respondents. In addition, the candidates or leaders included in past studies of gender affinity are overwhelmingly cisgender and straight.

Given past work that finds women voters are more likely to evaluate women leaders more positively and/or vote for parties with women leaders, our first set of hypotheses focuses on the possibility of only gender affinity. If we find that both non-LGBT and LGBT women (but not LGBT men) evaluate an out lesbian leader more positively ( H_1a ), we can say that gender affinity explains the results. After all, there would be gender affinity even among non-LGBT women who do not share an LGBT identity with the leader. Similarly, if we find that non-LGBT and LGBT women (but not LGBT men) become more likely to vote for a party when it selects an out lesbian leader, that would also be consistent with gender affinity voting ( H_1b ). We might see only gender affinity (and not also LGBT affinity) if gender is salient while LGBT identity is not, or if LGBT men do not feel affinity for a lesbian leader because they do not share a gender identity and/or lower-order sexual identity with her.

H_1a (Gender Affinity): Both non-LGBT and LGBT women will be more likely to evaluate an out lesbian positively when she is selected as party leader than non-LGBT and LGBT men because of a shared gender identity.

H_1b (Gender Affinity Voting): Both non-LGBT and LGBT women will be more likely to vote for a party when it selects an out lesbian party leader than non-LGBT and LGBT men because of a shared gender identity.

Another possibility is that there is only LGBT affinity. Given past work that has found that lesbian and gay voters are more likely to vote for lesbian and gay candidates, our second set of hypotheses considers the possibility of only LGBT affinity effects. If we find that both LGBT men and women (but not non-LGBT women) evaluate an out lesbian leader more positively ( H_2a ), we can say that LGBT affinity explains the results. This pattern would suggest there is a strong LGBT coalitional identity that leads LGBT men to feel affinity with an LGBT party leader of a different gender. We would attribute this pattern to LGBT affinity but not to gender affinity based on non-LGBT women’s lack of affinity. Again, if this pattern is also present in vote choice, it would fit with LGBT affinity voting ( H_2b ). We might see only LGBT affinity but not gender affinity if LGBT identity is salient while gender is not, or if non-LGBT men do not feel affinity for a lesbian leader because they do not share a sexual identity with her. This latter possibility might be more plausible in cases in which there are also non-LGBT women party leaders.

H_2a (LGBT Affinity): Both LGBT men and LGBT women will be more likely to evaluate an out lesbian positively when she is selected as party leader than non-LGBT men and women because of a shared LGBT identity.

H_2b (LGBT Affinity Voting): Both LGBT men and LGBT women will be more likely to vote for a party when it selects an out lesbian party leader than non-LGBT men and women because of a shared LGBT identity.

A third possibility is a double advantage based on both gender and LGBT affinity. In this scenario, those who share a gender (non-LGBT and LGBT women) and those who share an LGBT identity (LGBT men and women) would all evaluate an out lesbian party leader more positively. This set of hypotheses differs from the first two in that more than one affinity is at work ( H_3a and H_3b ). If this is the case, it suggests that it might be desirable (at least under certain circumstances) to select out LGBT women as party leaders. This is the opposite of the double disadvantage expected under intersectional theory (see, e.g., Crenshaw 1989, 1991).

H_3a (Double Advantage—Gender and LGBT Affinity): Non-LGBT women, LGBT men, and LGBT women will be more likely to evaluate an out lesbian positively when she is selected as party leader than non-LGBT men because of shared gender and/or LGBT identity.

H_3b (Double Advantage—Gender and LGBT Affinity Voting): Non-LGBT women, LGBT men, and LGBT women will be more likely to vote for a party when it selects an out lesbian party leader than non-LGBT men because of shared gender and/or LGBT identity.

A fourth possibility can be understood as a narrow advantage and/or double disadvantage based on integrated gender and LGBT identities. If we find that only LGBT women (and neither non-LGBT women nor LGBT men) evaluate an out lesbian leader more positively ( H_4a ) or are more likely to vote for that party ( H_4b ), we can say that gender and LGBT identity only work together as an integrated identity. This can also be understood as a double disadvantage, whereby an out lesbian leader benefits from neither gender affinity among non-LGBT women nor LGBT affinity among LGBT men. We might expect to see this pattern if “lesbian” is taken as a distinct identity (at a specific intersection of gender and sexuality) rather than as implying or combining both a woman gender identity and a sexual minority identity. This would mean that non-LGBT women would see lesbians as an out-group (based on sexuality) and not an in-group (based on gender). Similarly, this would mean that LGBT men would see lesbians as an out-group (based on gender) and not an in-group (based on LGBT coalitional identity). This has important implications for the boundaries of gender and sexual identity.

H_4a (Narrow Advantage—LGBT Women Affinity): LGBT women will be more likely to evaluate an out lesbian positively when she is selected as party leader than non-LGBT men, non-LGBT women, and LGBT men because of shared gender and LGBT identities.

H_4b (Narrow Advantage—LGBT Women Affinity Voting): LGBT women will be more likely to vote for a party when it selects an out lesbian party leader than non-LGBT men, non-LGBT women, and LGBT men because of shared gender and LGBT identities.

Past work has shown that even if there is affinity in candidate or leader evaluations, it does not necessarily translate into vote choice. One of the main reasons it might not is partisanship (Dolan 2008). In Canada, women and LGBT people have partisan ties that might overwhelm affinity. Women in Canada are less likely than men to support parties of the right and more likely to support the center-left Liberals and the social democratic NDP (Erickson and O’Neill 2002; Gidengil et al. 2005; Gidengil et al. 2013). LGBT people in Canada are far less likely than non-LGBT people to support the Conservatives (Perrella, Brown, and Kay 2012, 2019). They tend to split their support between the Liberals and the NDP. LGBT men are more likely than LGBT women to support the Liberals and less likely to support the NDP (Guntermann and Beauvais 2022; Perrella, Brown, and Kay 2012, 2019).

Studies of voter-candidate affinity generally address concerns that affinity is an artifact of partisanship by controlling for party identification. In this study, we examine changes in leader evaluations and vote choice before the party selects an out lesbian leader (i.e., when the party was led by a non-LGBT man) and after. The other major parties did not change their leaders between these elections. The changes between elections adjust for the baseline support for each party and its leader.

H₅ (Partisanship Overriding Affinity in Vote Choice): There will be affinity in leader evaluations (best premier choice) but not in vote choice because partisanship outweighs shared identities in vote choice. This may happen for one or more gender and sexual identity groups.

The Ontario Case

We take the case of Kathleen Wynne in Ontario as a rare opportunity to examine gender and LGBT affinity in a real-world election. Comparatively, there are few out LGBT first ministers or party leaders. Wynne was Ontario’s first woman premier and Canada’s first out LGBT first minister. In addition, this case provides a wealth of survey data that includes an LGBT identity question, which is relatively rare for this period. Finally, we have comparable survey data not only from the 2014 election (i.e., the first election after Wynne became leader) but also from the previous election (2011). This allows us to examine gender and LGBT affinity by estimating the swings among non-LGBT men, non-LGBT women, LGBT men, and LGBT women from 2011 to 2014.

During the period of study, Ontario had a three-party system. The three major parties were the PCs, the Liberals, and the NDP. All three of these parties have formed governments in Ontario. From 2003 to 2018, the Liberal Party formed the government. From 2003 to 2013, the Liberal leader was Dalton McGuinty (a non-LGBT man). From 2003 to 2011, the Liberals formed a majority government; after the 2011 election, they were reduced to minority government status. In 2012, McGuinty announced his resignation as Liberal leader and the party held a leadership race. When Wynne won the Liberal leadership in 2013, she succeeded McGuinty as premier of Ontario. She became leader when the Ontario Liberal Party was in decline and her predecessor was unpopular (Thomas 2018). She then led her party into an early election in 2014.

We do not have much guidance from past research on how non-LGBT men, non-LGBT women, LGBT men, and LGBT women would vote differently in Ontario politics during this period. However, past work on Canadian federal politics generally finds that women are more likely to vote NDP than men are (Gidengil et al. 2013). In addition, past work covering this period suggests that LGBT people tend to be more likely to vote Liberal or NDP and less likely to vote Conservative than non-LGBT people (Perrella, Brown, and Kay 2012, 2019). Among LGBT people, LGBT men are more likely to vote Liberal and less likely to vote NDP than LGBT women (Perrella, Brown, and Kay 2012, 2019). Although this research only examines federal politics, we expect (and find) that these types of patterns also hold in Ontario politics.

We have good reasons to believe that gender and LGBT affinity may be relatively salient in this case. Wynne’s identities as a woman and as a lesbian were more notable because she was the first woman premier and the first out LGBT leader. For these reasons, her identities were of historic importance. This is particularly true for LGBT people. This was not the first time that women in Ontario had the opportunity to vote for a major party led by a woman, but it was the first time LGBT people in Ontario could vote for a major party led by an out leader. We have evidence in particular that both news media and voters stereotyped Wynne as a lesbian. In a review of media coverage of three out lesbian or gay leaders of political parties in Canada (Kathleen Wynne, Prince Edward Island Liberal premier Wade McLaughlin, and Parti québécois leader André Boisclair), Lalancette and Tremblay (2019) find that Wynne received press coverage as a “good, respectable lesbian/gay” (similar to McLaughlin but not Boisclair). Everitt and Raney (2019) find evidence that voters stereotyped Wynne as a lesbian in the 2014 Ontario election. In particular, they assumed that she would represent the interests of lesbians and gay men better than her competitors.⁴ With that said, both Wynne and the news media “may have been particularly careful in how they discussed her uniqueness” (Thomas et al. 2021, 398). Neither Wynne nor the news media portrayed her as “the LGBT” leader (Thomas et al. 2021). Wynne downplayed her own past LGBT activism (Everitt and Raney 2019), but her government approved pro-LGBT reforms to the sex education curriculum and voted to ban conversion therapy for minors.

Notably, the 2011 and 2014 Ontario elections were very similar apart from the change in Liberal leader. The swings in vote share from 2011 were generally small (+1 percentage point for the Liberals, –4 percentage points for the PCs, +1 percentage point for the NDP, and +2 percentage points for the Greens and other minor parties). The swings in seats were also generally small (+5 Liberal, –9 PC, +4 NDP). Many local incumbents elected in 2011 sought reelection in 2014. Both elections showed a tendency for the Liberals to perform better in cities than in rural areas (Cross et al. 2015). Wynne and McGuinty had similar ideological positions. Both McGuinty and Wynne favored spending on public services and restricting the size of deficits. Both also had unambiguously pro-LGBT rights positions. McGuinty’s government introduced discussions of sexual orientation and gender identity into the Grade 3 curriculum before the 2011 election and mandated that both public schools and publicly funded Catholic schools allow students to form Gay-Straight Alliances after the 2011 election. Most importantly, the leaders of the other two major parties, Tim Hudak (PC) and Andrea Horwath (NDP), were the same in 2011 and 2014. These similarities strengthen the research design because there are no changes in leadership of the other major parties that could explain changes in the two main outcome variables (best premier evaluations and vote choice).

Data and Methodology

We draw on four large-sample nonprobability online surveys fielded by Ipsos before and after the 2011 and 2014 Ontario elections. Before each election, Ipsos conducted a preelection “invitation” survey to recruit respondents for its postelection exit survey. Some of the invitation survey respondents did not participate in the exit survey, but many of them did. In addition, Ipsos recruited additional respondents for the exit survey who did not participate in the invitation survey. All four surveys included an LGBT identity question (2011: “Are you a member of the gay, lesbian, bisexual or transgender community?”; 2014: “Do you consider yourself to be gay, lesbian, bisexual or transgender?”). We use these questions to code our main explanatory variable, a combined gender and sexual identity variable (1 = non-LGBT men, 2 = non-LGBT women, 3 = LGBT men, 4 = LGBT women).

We pool the invitation and exit surveys to increase LGBT men and LGBT women subsamples in each year, which helps address the small size of these populations. All four surveys have questions on respondents’ choice of best premier (among the three major party leaders) and vote intentions or vote choice. We code several variables measured using the same questions across the invitation and exit surveys as though all the respondents came from the same survey. These variables include age, gender, self-identification as a “visible minority” (a term from Canadian employment equity law that indicates someone who is neither white nor Indigenous), country of birth, region, education, LGBT identity, and best premier choice.⁵ For respondents who participated in both the invitation and exit surveys, we use data from the exit survey to code their values on these variables wherever possible (for example, vote choice in the exit survey over vote intention from the invitation survey). We provide details on the pooling of the surveys, along with the question wordings and variable coding for all variables used in this study, in the Supplementary Material.

Our research design focuses on estimating the shifts in best premier choice and vote choice among non-LGBT men, non-LGBT women, LGBT men, and LGBT women from 2011 to 2014. We use these over-time shifts among each group as evidence of affinity in best premier choice and vote choice. Our approach follows the logic of a difference-in-differences design. In a difference-in-differences design, researchers use data on observations before and after some of them receive a “treatment” to estimate the treatment effect. The observations before the “treatment” serve as the baseline for each observation. If researchers take the difference between the before and after time periods, they adjust for the baseline level of the outcome of interest for each observation. Then, they can compare the differences in these differences across the observations that do and do not receive the treatment. In an ideal world, we would use this design with panel data in which the same respondents answered surveys in 2011 and 2014, which would provide more confidence that the swings from 2011 to 2014 are causal. Although we do not have panel data, we can still use the 2011 best premier choice and vote choice estimates for each group as an aggregate baseline tendency for each group to (1) pick the Liberal leader as best premier and (2) vote Liberal. This approach provides more confidence that Wynne is driving the results than simply relying on a cross-sectional survey conducted in 2014 because it adjusts for each group’s baseline on best premier choice and vote choice.

These estimated swings may be a “ceiling” on how much affinity for Wynne may have shifted political attitudes and behavior in the 2014 Ontario election. Elections are complicated events, and many factors can affect leader evaluations and vote choice beyond the ones we can examine in this study. Although we know that these results are not attributable to changes in which individuals are leading the other major parties or substantial changes in parties’ policy positions on issues that are more salient to women or LGBT people, we cannot rule out the possibility that there may be some other factors that also contribute to these swings. At the same time, we believe it is possible that LGBT affinities for leaders can be even larger than the swings we identify here. Some LGBT respondents may not have been aware that Wynne was a lesbian. If everyone had been aware, it is possible that these swings may have been even larger.

These data have three important limitations. First, none of these surveys has a measure of party identification. Commercial polling firms in Canada do not reliably ask party identification questions in their surveys, often preferring alternatives such as vote choice in the last election. In a simple cross-sectional analysis of the 2014 election, this would be a major problem because these groups almost certainly vary in their levels of Liberal party identification. Although we would adjust for individual-level party identification if we could, our research design helps mitigate this problem by examining the swings among each group from 2011 to 2014. This approach adjusts for each group’s baseline level of support for the Liberals. Second, because of the question wording, we cannot disaggregate subgroups of LGBT people, such as lesbians versus bisexual women. There is likely variation in affinity among subgroups of LGBT respondents (Guntermann and Beauvais 2022), but the data only let us examine LGBT as an umbrella category. Third, we do not have questions that allow us to examine mechanisms of gender or LGBT affinity. We do not have questions that ask about, for example, group consciousness or linked fate, which would help us test mechanisms more directly. Despite these limitations, we nonetheless believe that these data provide a rare opportunity to look for evidence of gender and LGBT affinity for a lesbian leader in a real-world election.

The Ipsos surveys are unrepresentative of the voting age citizen population on several demographic variables, including age, gender, education, racialized identity, country of birth, and region. We construct poststratification weights on these variables separately for each year using raking (or iterative proportional fitting).⁶ We provide details on the weighting in the Supplementary Material. We present the weighted results in the main analyses and the unweighted results in the Supplementary Material.

After we construct the weights, we use multiple imputation with chained equations (MICE) to handle missing data within each year’s pooled data set. This approach also helps bridge across the entrance and exit surveys in cases in which there are minor differences between the surveys, such as using vote intentions in the entrance survey and vote choice in the exit survey or including or not including an explicit “don’t know” in the best premier question. MICE helps bridge across the models because we have many respondents who answered both questionnaires. This allows us to simulate how the respondents who only participated in one survey would have responded if they had participated in both surveys. Scholars have employed this approach, using data from one survey to impute missing data on other surveys, to examine how respondents would have responded to alternative question wordings (Eckman 2022). We include additional details on MICE in the Supplementary Material.

In the Supplementary Material, we run four replications of our results to examine the extent to which our methodological decisions affect the results. First, we rerun our analysis without the weights but with multiple imputation. We find that the weights tend to make the sample more pro-NDP and less pro-Liberal, tend to shrink our estimates of affinities among LGBT men and women somewhat, and tend to increase the confidence intervals around our estimates. Second, we rerun our analyses (as much as possible given missing data limitations) without the weights and without multiple imputation. These results are likely to be misleading because of distortions among the sample, but they suggest that non-LGBT men are more hostile to Wynne than in our other results and that swings among LGBT men and LGBT women in best premier choice are even larger than we report in the main text. Third, we rerun our analyses using only “likely voters”—that is, dropping invitation survey respondents who did not participate in the exit survey and indicated they were least likely to vote. The results are generally very similar to our main analyses, except the confidence intervals are larger because we drop some respondents. Fourth, we rerun our analyses using an alternative approach to handling missing data on LGBT identity in which we exclude respondents who have missing data on this variable before we construct the weights or use multiple imputation, rather than imputing LGBT identity. We find that our results do not change substantially depending on whether we impute missing data on this variable or exclude respondents with missing data. If anything, our analyses in the main text may generate smaller estimated swings than the ones in which we exclude respondents who have missing data on LGBT identity. Overall, none of these decisions substantially changes our conclusions. We note minor deviations from our main results where appropriate in the analysis.

Analysis

We begin with a bivariate analysis of best premier choice and vote choice among non-LGBT men, non-LGBT women, LGBT men, and LGBT women in each election. Figure 1 displays the percentage picking each major party leader as the best premier with 95% confidence intervals separately for non-LGBT men, non-LGBT women, LGBT men, and LGBT women in each year. In 2011, non-LGBT men were most likely to pick Liberal leader McGuinty as best premier, then PC leader Hudak, then NDP leader Horwath. The results are nearly identical for non-LGBT men in 2014, even with Wynne as the Liberal leader. In 2011, non-LGBT women were most likely to pick Horwath, then McGuinty, then Hudak. This pattern may be evidence of gender affinity for Horwath, who was the only woman leader at the time. However, in 2014, non-LGBT women became much more likely to pick Wynne as best premier, then Horwath, then Hudak. There is no evidence that non-LGBT women had a bias against Wynne, since they were much more likely to pick Wynne as best premier in 2014 than they were to pick McGuinty in 2011 (by 9.2 percentage points). Instead, these results are consistent with gender affinity among non-LGBT women in 2014.

Figure 1. Estimated percentage selecting each party leader as best premier, by gender and sexual identity and year.

The results shown in Figure 1 are even more striking for LGBT men and women. In 2011, LGBT men were most likely to pick McGuinty as best premier, then Horwath, then Hudak. However, this pattern became even more pronounced in 2014: LGBT men were 24.4 percentage points more likely to pick Wynne as best premier in 2014 than they were to pick McGuinty in 2011. There was a similar swing among LGBT women. In 2011, LGBT women mostly picked Horwath as best premier over McGuinty and Hudak. LGBT women were 27.4 percentage points more likely to pick Wynne as best premier in 2014 than they were to pick McGuinty in 2011. Overall, these results for leader evaluations fit with H_3a (Double Advantage—Gender and LGBT Affinity).

Figure 2 displays a similar plot for vote choice. Figure 2 includes the three major parties—the Liberals, the PCs, and the NDP—along with an additional category for respondents who voted for another party (usually the Greens). In 2011, non-LGBT men were most likely to vote PC, then Liberal, then NDP. In 2014, the order of the parties was the same for non-LGBT men, but the Liberal vote share increased to 34.1%. By contrast, in 2011, non-LGBT women were about equally as likely to vote Liberal as PC, and the NDP was a close third. In 2014, non-LGBT women were most likely to vote Liberal over the PCs and the NDP. Non-LGBT women were 4.9 percentage points more likely to vote Liberal in 2014 than they were in 2011. This result fits the expectations of gender affinity voting.

Figure 2. Estimated percentage voting for each party, by gender and sexual identity and year.

We do not find as strong evidence of LGBT affinity in vote choice as we do in best premier choice, especially among LGBT women. We focus our analysis on LGBT men and women on voting for the Liberals and the NDP because we find few LGBT men and women voting PC in either election, as expected from work on federal politics. According to Figure 2, LGBT men were about equally likely to vote Liberal and NDP in 2011.⁷ In 2014, LGBT men became more likely to vote Liberal than NDP. These results suggest a swing of 12.8 percentage points to the Liberals among LGBT men, largely at the expense of the NDP. By contrast, in 2011, LGBT women were more likely to vote NDP than Liberal. In 2014, LGBT women were still more likely to vote NDP (43.6%) than Liberal (36.2%), though the point estimates suggest that the Liberals may have done better with LGBT women in 2014. The finding that LGBT men were more likely to vote Liberal than LGBT women generally fits with work on Canadian federal politics during this period (Perrella, Brown, and Kay 2012, 2019). These results for vote choice most closely fit H_3b (Double Advantage—Gender and LGBT Affinity Voting), except the effect is small and nonsignificant among LGBT women. H₅ (Partisanship Overriding Affinity in Vote Choice) only appears to hold for LGBT women.

We estimate the swings in best premier choice and vote choice from 2011 to 2014 among non-LGBT men, non-LGBT women, LGBT men, and LGBT women using weighted logistic regression models. We take the differences in these swings across each group as plausible estimates of affinity in best premier choice and vote choice in 2014. We run three sets of models for each outcome variable.

In Model 1, we include the gender and sexual identity variable, the year, and an interaction term between gender and sexual identity and year. This model allows us to estimate the predicted percentage picking the Liberal leader as best premier or voting Liberal across gender and sexual identity categories, along with the change in each category from 2011 to 2014. Given the many similarities between the 2011 and 2014 elections other than the change in Liberal leader, these over-time comparisons are plausibly attributable to the change in Liberal leader from McGuinty to Wynne.

In Model 2, we add in age, country of birth, visible minority identity, and Indigenous identity, along with interactions between these variables and the year. Model 2 accounts for variables that are plausibly causally prior to gender and sexual identity that may shape both gender and sexual identity and political attitudes and behaviors. It does not include variables that may be consequences of gender and sexual identity, such as community size or religious identity. (LGBT people tend to move to cities and identify as nonreligious in large part because they are LGBT.)

Finally, in Model 3, we add every possible control variable that is in common across the four surveys, including education, region, religious identity, importance of religion in daily life, income, community size, household gun ownership, and self-reported vote choice in the previous election.⁸ We take Model 2 as our best attempt to estimate the “total effect” of gender and sexual identity. In doing so, we generate an estimate of the swing among each group that does not take place through any mediating variables, such as community size or religion. By contrast, Model 3 is our best attempt to estimate the “controlled direct effect” of gender and sexual identity. Model 3 presents conservative estimates of the relationships between gender and sexual identity and our outcomes of interest because it only estimates the relationships between gender and sexual identity and our outcomes of interest that do not take place through mediating variables. Model 3 may actually attribute some of the swings among LGBT people from 2011 to 2014 to variables on which LGBT people tend to differ from non-LGBT people, such as community size or religion, when gender and sexual identity are really “doing the work.” All three models show similar patterns to the bivariate analyses, but the estimated swings for LGBT men and LGBT women are generally much smaller in Model 3.

Figure 3 shows the estimated percentage point change in picking the Liberal leader as best premier from 2011 (under McGuinty) to 2014 (under Wynne) from our three models. The results are generally similar across all three models. For non-LGBT men, these swings have very small point estimates, and they are not significantly different from zero. Non-LGBT women were about 9 percentage points more likely to pick Wynne as best premier in 2014 than McGuinty in 2011. This swing fits with past work that finds gender affinity in leader evaluations (see Banducci and Karp 2000; Cutler 2002; O’Neill 1998). Although the small LGBT men and women subsamples generate large confidence intervals, they provide clear evidence that LGBT men were much more likely to pick Wynne as best premier in 2014 than McGuinty in 2011.The same is true for LGBT women, who had the largest estimated swing from 2011 to 2014 in picking the Liberal leader as best premier.

Figure 3. Estimated percentage point change from 2011 to 2014 in choosing Liberal leader as best premier with 95% confidence intervals, by gender and sexual identity, weighted logit.

Although Figure 3 helps illustrate the general patterns among each group, the most appropriate test of our affinity hypotheses is to examine the differences in these swings across each pair of groups (the differences-in-differences). We plot the differences-in-differences in Figure 4. Across all three models, Figure 4 shows that non-LGBT women, LGBT men, and LGBT women had larger swings in best premier choice to the Liberals between 2011 and 2014 than non-LGBT men did. These swings suggest that there is some affinity in leader evaluations among each group. LGBT men had larger swings than non-LGBT women when we include no controls or only clearly prior controls, but the difference between these two groups becomes much smaller (about 5 percentage points) and nonsignificant when we include all the controls. This nonsignificance does not provide evidence against affinity, but it does suggest that LGBT affinity may not be much larger than gender affinity once accounting for other variables that may distinguish LGBT men from non-LGBT women. Figure 4 presents evidence that LGBT women had larger swings in best premier choice to the Liberals than non-LGBT women did. Even after adjusting for every possible available control variable, we still estimate that the swing among LGBT women was about 8 percentage points larger than the swing among non-LGBT women, though this difference is nonsignificant. (This difference is significant when we do not include weights.) Finally, we find small and nonsignificant differences in the swings between LGBT men and LGBT women. However, we are cautious in interpreting the differences between LGBT men and LGBT women because of the size of these subsamples.

Figure 4. Pairwise differences in estimated percentage point swings from 2011 to 2014 in choosing the Liberal leader as best premier with 95% confidence intervals, weighted logit.

Overall, our models suggest that both gender and LGBT affinity exist in leadership evaluations. Again, these results correspond to H_3a (Double Advantage—Gender and LGBT Affinity). Our results also suggest that LGBT affinity may be stronger than gender affinity, but we would need even more data to know for sure. Across all three models, LGBT men and LGBT women have larger estimated swings than non-LGBT women do, though the differences in the swings become nonsignificant among these three groups in Model 3, which includes the full set of control variables. We cannot clearly tell from our analyses whether gender and LGBT affinity are interactive or additive. The uncertainty around our estimates for LGBT men and LGBT women are too large distinguish between these two possibilities.

These shifts in evaluations of party leaders did not translate into vote choice for all these groups, however. Figure 5 displays the estimated percentage point changes in Liberal vote choice from 2011 to 2014 among non-LGBT men, non-LGBT women, LGBT men, and LGBT women from our three models. Again, the three models generate similar results. We do not find a significant shift in voting Liberal among non-LGBT men. Non-LGBT women were 4–6 percentage points more likely to vote Liberal in 2014 than in 2011. This result is consistent with gender affinity voting among non-LGBT women. Across all three models, LGBT men became substantially more likely to vote Liberal in 2014 than they were in 2011 (by 10–14 percentage points). This swing in Liberal voting is significant despite the small size of the LGBT men subsample. This result fits the expectations of LGBT affinity voting among LGBT men. However, we do not find clear evidence that LGBT women swung to the Liberals after the party chose a lesbian as leader. Although the point estimates are about 3–5 percentage points, this swing is too small to be significant given the size of the LGBT women subsample.⁹ Even though LGBT women had the largest swing in choosing Wynne in 2014 over McGuinty in 2011 as best premier, they did not vote Liberal in substantially larger numbers the way LGBT men did.

Figure 5. Estimated percentage point change from 2011 to 2014 in Liberal vote choice with 95% confidence intervals, by gender and sexual identity, weighted logit.

Although Figure 5 presents a clear pattern of swings to the Liberals among non-LGBT women and LGBT men, the results are less clear in Figure 6 when we examine the differences in the swings across pairs of groups. Overall, our point estimates suggest that non-LGBT women, LGBT men, and LGBT women were all more likely to vote Liberal in 2014 than in 2011 than non-LGBT men were. If our sample estimates reflect the real world, then these are substantively large swings to the Liberals among non-LGBT women, LGBT men, and LGBT women. However, almost none of these comparisons are statistically significant, largely because of the uncertainty around our estimates that compare the small LGBT men and LGBT women subsamples. The exceptions are the difference in the swings among non-LGBT men and non-LGBT women in Model 3 and the difference in the swings among non-LGBT men and LGBT men in Model 2. Our results present the strongest support for a difference in the swings between non-LGBT men and non-LGBT women. Not only do we find a significant difference in the swings in Model 3, but we also find consistently significant differences between non-LGBT men and non-LGBT women in our unweighted results (see the Supplementary Material). Given the uncertainty around our estimates of the differences-in-differences, we are cautious in interpreting our results. Our estimates are consistent with the possibility that non-LGBT women, LGBT men, and LGBT women all moved to the Liberals the same amount in 2014. They are also consistent with the possibility that LGBT women actually swung away from the Liberals after they selected Wynne as the Liberal leader or that they were no more likely to swing to the Liberals than non-LGBT men.

Figure 6. Pairwise differences in estimated percentage point swings from 2011 to 2014 in Liberal vote choice with 95% confidence intervals, weighted logit.

Overall, our analyses of vote choice are suggestive of both gender and LGBT affinity. Although the difference-in-difference estimates are not statistically significant given our small LGBT subsamples, they are all consistent with H_3b (Double Advantage—Gender and LGBT Affinity). They provide the strongest support for gender affinity among non-LGBT women and LGBT affinity among LGBT men, both of whom clearly swung Liberal in 2014. However, in this sample, LGBT women do not generally show strong evidence for affinity in vote choice. Even though many LGBT women were more likely to think Wynne would be the best premier than McGuinty, and they were the group that had the largest swing in best premier choice, it is entirely possible that these positive evaluations of Wynne did not translate into voting Liberal for LGBT women in 2014. We simply cannot tell with much confidence how much LGBT women swung to the Liberals in 2014, let alone how different their swing was from non-LGBT women or LGBT men. As a result, we cannot rule out H₅ (Partisanship Overriding Affinity in Vote Choice) for LGBT women.¹⁰ This result is surprising because the logic of affinity would generally suggest that LGBT women would be the most likely group to have affinity for a lesbian party leader because they have the potential for both gender and LGBT affinity.

Conclusion

Our results are most consistent with the hypothesis that Wynne benefited from a double advantage—that is, she benefited both from gender and LGBT affinity. This double advantage is clearest in best premier choice. Non-LGBT women, LGBT men, and LGBT women all show evidence for affinity in best premier choice. Notably, our results suggest that LGBT affinity in leader evaluations may be as strong, if not stronger, than gender affinity among women. Our point estimates consistently suggest that the swings in best premier choice among LGBT men and, especially, LGBT women were at least as larger as the ones among non-LGBT women. Even though our LGBT women subsample is small, we can still say with confidence that LGBT women showed even more affinity for Wynne than non-LGBT women did.

Our results are not as clear for affinity in vote choice. Non-LGBT women and LGBT men clearly were more likely to vote Liberal in 2014 than they were in 2011. However, our data do not allow us to precisely estimate how large the swing is among LGBT women. Our best guess is that LGBT women were more likely to vote Liberal in 2014 than in 2011, but we cannot tell with any confidence whether they had a larger or smaller swing to the Liberals than non-LGBT women or LGBT men. These results suggest that the large gender and LGBT affinities in leader evaluations we observe with Kathleen Wynne may not produce large swings in vote choice.

The unexpectedly weak results for LGBT women suggest that leader evaluations might matter differently to vote choice for different gender and sexual identity groups. We suspect that LGBT women are less likely to be crossover voters who will switch parties based on an affinity with the leader. It could be that many LGBT women care more about party identification and/or policy than about the identity of a party leader. This would be consistent with evidence that LGBT women in Canada are more tied to the NDP, which has traditionally been at the forefront of pro-LGBT policies.

These results suggest that LGBT affinity may exist in other real-world contexts. Comparative work could be especially useful for determining the conditions under which LGBT affinity in evaluations translates into LGBT affinity voting. We recommend that future research disaggregate gender and sexual identity further. When we disaggregate LGBT men and women, we find differences in their political behavior. However, these data do not permit further disaggregation of LGBT identities. We expect to find variation within LGBT people based on sexual identity (lesbian, gay, bisexual, queer, etc.) and transgender and nonbinary identities. We need disaggregated analyses so that larger subgroups do not drive the results.

Our findings provide cautious optimism for those seeking to increase the political representation of those who are intersectionally marginalized. Not all intersectionally marginalized party leaders will be doubly (or multiply) marginalized. In fact, parties that select such leaders might face a double (or more) advantage. Of course, Wynne was a conventional leader in many respects. She is white and took an “insider” path to political leadership. We may not see such encouraging findings if we look at politicians at other intersections of marginalization. Future research could examine the conditions under which intersectionally marginalized party leaders face advantages or disadvantages among different groups of voters.

Our results are also encouraging because they suggest that many voters use more expansive boundaries of group membership than we might expect. It was not clear that non-LGBT women would feel affinity with an out lesbian party leader. After all, although non-LGBT women share a gender identity with Wynne, they do not share a sexual identity. We might expect out-group prejudice on sexuality to overwhelm any in-group favoritism on gender. Likewise, although LGBT men share an LGBT identity with Wynne, they do not share a gender identity. It would be possible for LGBT men to draw on more narrow identities within the LGBT coalition and not see Wynne as an in-group member. Instead, LGBT men appear to have drawn on an LGBT coalitional identity that included Wynne. Future work could pursue additional intersectional analyses of gender and LGBT affinity that would help further map group boundaries, showing us who voters “count” as in-group members.