While the literature on vote choice in Canada is large and well-developed, there is still little understanding of whether the sex and ethnicity of local candidates impact individual voting decisions. At the level of federal elections, research strongly suggests that candidates are less important to the overall voting calculus of the Canadian electorate than are parties and leaders (Blais et al., Reference Blais, Gidengil, Dobrzynska, Nevitte and Nadeau2003; Cunningham, Reference Cunningham1971). While factors such as incumbency and strength of the local candidate have some influence, particularly for non-partisans (Roy and Alcantara, Reference Roy and Alcantara2015), it is unclear whether candidate sex and ethnicity matter in this way. One expectation is that there may be gender or co-ethnic “affinity” voting, in which women give greater support than men to female candidates and ethnic minorities give more support than whites to minority candidates. There is evidence that voters' sociodemographic proximity to federal party leaders plays a role in individual vote choice (Cutler, Reference Cutler2002), but the findings for the characteristics of local candidates are mixed. Research into real elections at the federal level shows little effect of voters' affinities on candidate choice (Cutler and Matthews, Reference Cutler and Matthews2005; Goodyear-Grant, Reference Goodyear-Grant, Anderson and Stephenson2010; Goodyear-Grant and Croskill, Reference Goodyear-Grant and Croskill2011; Landa et al., Reference Landa, Copeland and Grofman1995; Murakami, Reference Murakami2014) though experiments involving manipulations of the characteristics of fictional candidates suggest that ethnic and, to a lesser extent, gender affinities do matter (Besco, Reference Besco2015a, Reference Besco2015b; Bird, Reference Bird and Bilodeau2016; Tolley and Goodyear-Grant, Reference Tolley and Goodyear-Grant2014). One possibility is that party ties in real elections overwhelm the potential impact of affinity effects; if so, we would be more likely to find gender or ethnic affinity voting in Canada in local elections, where partisanship is less salient.Footnote 1
Our study tests for the presence of gender and ethnic affinity voting in the non-partisan 2014 Toronto mayoral election, where Olivia Chow was the only woman and only visible minority among the three major candidates.Footnote 2 This case provides an important analytical advantage in the search for affinity effects. First, the non-partisan nature of the contest allows us to parse out the confounding factors of party leaders and partisan cues that tend to supersede candidate characteristics in elections at other levels. Experimental studies can isolate the impact of candidate sex and race on vote choice, but their external validity is questionable; by presenting bare bones scenarios that leave respondents with little other information on which to form an opinion, we cannot assess whether candidate sociodemographics matter in the context of other information that voters might possess. They may also be subject to social desirability effects. Second, the enormous social diversity of Toronto makes it an especially useful site to examine the propensity of particular ethnic minorities to vote for other minorities. Third, understanding affinity voting at the municipal level is important in its own right. Women and visible minorities are surprisingly few in number in elected municipal office in Canada's largest cities, and it is vital to learn whether candidate sociodemographics and voter bias play some role in this underrepresentation.Footnote 3 Equitable representation at this level also matters, as we know that municipal careers are often used as a springboard to higher office (Deckman, Reference Deckman2007; Sanbonmatsu et al., Reference Sanbonmatsu, Carrol and Walsh2009)
Our analysis contributes to a nascent body of research looking at the interactive effects of voter and candidate sex and ethnicity on voter choice in Canada and is the first to examine this in the context of a non-partisan municipal election in a non-experimental manner. Our findings suggest that in a meaningful and high-profile election, where information is abundant but party cues are less salient,Footnote 4 ethnic affinity plays an important role in voter choice. We also show that gender was related to vote choice but only when it interacted with race. Specifically, we find that minority and non-minority women have very different preferences when presented with a female minority candidate.
Gender and Ethnic Affinities
Do women vote for women and men for men? Do visible minorities vote for minority candidates and white voters for white candidates? The “affinity voting” thesis postulates that voters invoke baseline preferences for candidates on the grounds of shared gender, racial or other highly visible sociodemographic characteristics (Brians, Reference Brians2005; Matson and Fine, Reference Matson and Fine2006; McDermott, Reference McDermott1997, Reference McDermott1998; Plutzer and Zipp, Reference Plutzer and Zipp1996; Sanbonmatsu, Reference Sanbonmatsu2002; Tolley and Goodyear-Grant, Reference Tolley and Goodyear-Grant2014). Affinity voting is thought to be rooted in social psychological processes that may be either cognitive, identity-based, or some combination of the two.Footnote 5
Cognitive explanations emphasize the role of social stereotypes and schemata that allow us to impute quickly and with little cognitive effort a broader set of traits and characteristics to a given stimulus (Dolan, Reference Dolan2008, Reference Dolan2014; Fiske and Kinder, Reference Fiske, Kinder, Cantor and Kihlstrom1981; Markus and Zajonc, Reference Markus and Zajonc1985; Huddy and Terkildson, Reference Huddy and Terkildsen1993; Koch, Reference Koch2000, Reference Koch2002; McDermott, Reference McDermott1998; Schneider and Bos, Reference Schneider and Bos2014). Sociodemographic cues can be especially useful in low-information elections, where detailed candidate information may be difficult for most voters to obtain (Conover and Feldman, Reference Conover and Feldman1989). Cognitive-based affinity voting is thus a positive bias based on a voter's assumptions concerning some shared sociodemographic trait.
Social identity explanations, on the other hand, emphasize the role of social membership and feelings of group identification and solidarity (McClain et al., Reference McClain, Carew, Walton and Watts2009; Plutzer and Zipp, Reference Plutzer and Zipp1996; Tajfel, Reference Tajfel1981). Shared group identification may become politicized into group consciousness, at which point it assumes heightened salience as a basis for candidate choice. Under this account, affinity voting occurs when a candidate is readily and widely identified by her membership in the salient group, is presented or perceived as “acting for” that group's interests (however defined), and where group members are motivated by a set of beliefs about the group's social standing and determination to improve its status and realize its collective interests.
We are agnostic as to whether affinity is a heuristic shortcut or social identity process, and set aside this research question for another study. Furthermore, we recognize that if female voters have policy preferences that align more closely with positions taken by female candidates—that is, where female voters and candidates are more likely than male voters/candidates to prioritize affordable child care or favour gender parity in municipal governance—it will be difficult to distinguish whether same-sex vote choice is based on gender affinity itself (whether a heuristic shortcut or a social identity process) or on more dispassionately “reasoned” ideological/policy congruence. Below, we review the research findings on the subject of voter affinities, first for women, then for visible minorities, before turning to discuss the intersection of sex and race and the question of inter-ethnic affinity. Following that, we introduce the context of the Toronto mayoral election and then present our research methodology, including our empirical strategy for isolating affinity effects from ideological/policy congruence. Finally, we present and discuss our findings.
Gender affinity
While affinity is an individual-level phenomenon, aggregate evidence can be suggestive of general trends. Such aggregate-level evidence provides little indication that female candidates in Canada face generalized discrimination from voters.Footnote 6 Black and Erickson's study (2003) of the 1993 federal election found that female candidates actually receive more votes than similarly situated men when constituency characteristics are controlled. Young, analyzing the 1997 and 2000 federal elections and controlling for incumbency and party competitiveness, found modest evidence of reduced vote share in ridings with female incumbents but concludes that “if there is [voter] discrimination, its magnitude is so limited that it is unlikely to decide the outcome of local contests” (Reference Young, Sawer, Tremblay and Trimble2006: 51). While few have looked closely at whether female candidates face bias in municipal elections in Canada (Seigel et al., 2001; Tardy et al., Reference Tardy, Tremblay and Legault1997), studies in the US have found substantive differences in voter evaluations of male and female council candidates at that level (Brown et al., Reference Brown, Heighberger and Shocket1993; Crowder-Meyer et al., Reference Crowder-Meyer, Gadarian and Trounstine2015; Huddy and Terkildson, Reference Huddy and Terkildsen1993; Saltzstein, Reference Saltzstein1986). Crowder-Meyer and colleagues (Reference Crowder-Meyer, Gadarian and Trounstine2015) find that women fare better in wards and city clerkships compared to at-large and mayoral elections but that female mayoral candidates must be of higher quality than male candidates to achieve equal levels of electoral success. They argue that institutional rules of local politics interact with stereotypes about women's competency at specific tasks, and that this affects the type of office women choose to contest as well as their success rate across different types of elections.
Individual-level evidence of gender affinity voting is similarly mixed and appears to be weaker in Canada than in the United States. Cutler (Reference Cutler2002) found that voters were more likely to support a federal party headed by a leader of the same sex. In contrast, a study of voter choice between a male and female candidate for Vancouver mayor found no gender affinity effect, either before or after controlling for ideological- and party-gender overlap (Cutler and Matthews, Reference Cutler and Matthews2005).Footnote 7 In a study drawing on data from the Canadian Election Study, Goodyear-Grant and Croskill looked closely at the behaviour of “flexible” voters (those without a strong party attachment). They conclude that “women voters in Canada do not seem to cast a greater proportion of their ballots for women candidates than men voters do, even when conditions seem fertile for such an occurrence” (2011: 245). Finally, using a vote choice experiment with Canadian subjects, Tolley and Goodyear-Grant (Reference Tolley and Goodyear-Grant2014) report an 8-point difference in support for a female candidate among women compared to men, though this effect was not sustained after controlling for respondents' education, political knowledge, right-left ideology, and egalitarian gender attitudes. There is no consensus, therefore, on the presence or strength of gender affinity voting in Canada.
In contrast, in the American context gender affinity effects have been observed using experimental methods (McDermott, Reference McDermott1998; Sanbonmatsu, Reference Sanbonmatsu2002) and in actual elections among non-partisans and weak identifiers (Plutzer and Zipp, Reference Plutzer and Zipp1996). In partisan contests, several studies have found that male and female voters interact differently with Democratic and Republican women candidates (Brians, Reference Brians2005; Dolan, Reference Dolan2008; King and Matland, Reference King and Matland2003; Koch, Reference Koch2002; Sanbonmatsu and Dolan, Reference Sanbonmatsu and Dolan2009; Winter, Reference Winter2010). While affinity voting occurs, female Democratic candidates appear to garner more support than female Republican candidates from female crossover voters, possibly due to the more consistent direction that stereotypes of women and Democrats take.
Ethnic affinity
In the US, there is abundant evidence that white voters prefer white candidates, while minorities prefer minority candidates (Barreto, Reference Barreto2007; Bobo and Gilliam, Reference Bobo and Gilliam1990; Collet, Reference Collet2005; Philpot and Walton, Reference Philpot and Walton2007). There is also modest evidence of co-ethnic voting in Britain (Fisher et al., Reference Fisher, Heath, Sanders and Sobolewska2014) and France (Brouard and Tiberj, Reference Brouard, Tiberj, Bird, Saalfeld and M2011). Such patterns are especially strong in vote choice experiments presenting fictionalized white or minority candidates (McConnaughy et al., Reference McConnaughy, White, Leal and Casellas2010; McDermott, Reference McDermott1998; Moskowitz and Stroh, Reference Moskowitz and Stroh1994; Sigelman et al., Reference Sigelman, Sigelman, Walkosz and Nitz1995; Terkildson, Reference Terkildsen1993). They are less consistent in actual elections, presumably because the salience of issues, candidate characteristics, campaign styles and other local contextual factors can trump baseline ethnic and racial preferences (Citrin et al., Reference Citrin, Green and Sears1990; Highton, Reference Highton2004; Kaufmann, Reference Kaufmann2003a, Reference Kaufmann2003b, Reference Kaufmann2004; Pettigrew, Reference Pettigrew1976; Stein et al., Reference Stein, Ulbig and Post2005). At the municipal level, Kaufmann (Reference Kaufmann2004) shows that white and black voters' perceptions of interracial conflict in the community are a key factor shaping their evaluation and support of black mayoral candidates.
The context of racial politics in the US is very different than that in Canada and so we expected to find different results with respect to co-ethnic voting across the two countries. What is perhaps more remarkable is the paucity of research on the effects of candidate ethnicity on voter choice in Canada. We have identified just eight studies that address this issue.Footnote 8 Four of these draw on aggregate voter returns (Berdahl et al., Reference Berdahl, Adams and Poelzer2011; Black and Erickson, Reference Black and Erickson2006; Landa et al., Reference Landa, Copeland and Grofman1995; Tossutti and Najem, Reference Tossutti and Najem2002); three are based on experimental designs (Besco Reference Besco2015a, Reference Besco2015b; Bird, Reference Bird and Bilodeau2016; Tolley and Goodyear-Grant, Reference Tolley and Goodyear-Grant2014); and one uses both experimental and voter survey data (Murakami, Reference Murakami2014). Only one of these (Bird, Reference Bird and Bilodeau2016) focuses on the municipal context, though in a fictionalized scenario.
The results of the studies that rely on ecological data are mixed. While Toussutti and Najem (2002) and Black and Erickson (Reference Black and Erickson2006) found no relationship between visible minority candidacy and electoral success, Landa and colleagues (Reference Landa, Copeland and Grofman1995) and Berdahl and colleagues (Reference Berdahl, Adams and Poelzer2011) found that voter support for the party of an ethnic minority or Aboriginal candidate increased with the density of the ethnic or Aboriginal population in the riding. However, the small sample size and lack of controls (especially for incumbency) necessitate caution in interpreting these positive findings. Experimental designs have produced more consistent findings of ethnic affinity voting in Canada (Besco, Reference Besco2015a, Reference Besco2015b; Bird, Reference Bird and Bilodeau2016; Tolley and Goodyear-Grant, Reference Tolley and Goodyear-Grant2014).Footnote 9 As in the US, these analyses all find fairly strong evidence that visible minority voters prefer minority candidates, while white voters prefer white candidates. However, it is difficult to know whether the results would persist in a more robust political context. Recent work (Murakami, Reference Murakami2014) finds negative effects of candidates' ethnic minority background in a vote choice experiment but also finds that partisanship washes away the effect when real election data (from the 2008 Canadian Election Study) are used. The only exception is a very small set of voters who demonstrated an a priori strong negative affect towards ethnic minorities and opposed policies benefiting ethnic minorities. Murakami concludes that the influence of candidates' ethnicity on vote choice in Canadian federal elections is “negligibly small or extremely limited to only a small set of voters” (2014: 127).
Inter-minority ethnic affinity
An important question concerning ethnic affinity voting is whether we should expect visible minorities of all ethnic backgrounds to support Olivia Chow, who is from Hong Kong. Research in the US has found that racialized minorities are not necessarily willing to support minority candidates from other ethnic groups over white candidates (Casellas, Reference Casellas2009; Kaufman, Reference Kaufmann2003b; Ramakrishnan et al., 2008; but see Collet, Reference Collet2005). Drawing on an experiment with Canadian subjects, Besco (Reference Besco2015b) assesses the support for a Chinese (versus white) candidate among South Asian respondents, and for a South Asian (versus white) candidate among Chinese respondents. He finds some evidence of “rainbow coalition” affinity effects, and no evidence of inter-minority discrimination.
Assessing inter-ethnic affinity voting in Canada is crucial. There may be less inter-minority conflict in Canada's relatively peaceful multicultural mosaic than in other settings, but there are almost no ridings where a single ethnic group constitutes the majority of the electorate. Furthermore, Siemiatycki and colleagues (Reference Siemiatycki, Rees, Ng and Rahi2001) suggest that differences in class, language, national background, migration experiences, religion, income and residential location, even among seemingly cohesive and geographically clustered groups such as the Chinese, present enormous challenges of mobilization in Toronto. Thus, electoral success for minority candidates in Canada depends critically on support across a multitude of different ethnic communities. Toronto provides an excellent case in which to determine whether support for a visible minority candidate differs depending on whether the candidate is a member of one's own or a different ethnic group.
The intersection of gender and race
Finally, what happens when a visible minority woman appears on the ballot? There are two potential outcomes of note. The first is that such candidates may be subject to both gender and racial stereotypes, which could produce a double disadvantage in their quest for elected office (Moncrief et al., Reference Moncrief, Thompson and Schuhmann1991). On the other hand, growing evidence suggests that the gender gap in office holding is smaller among visible minorities than among non-minorities (Bejarano, Reference Bejarano2013; Bird, Reference Bird, Bird, Saalfeld and M2011; Celis et al., Reference Celis, Erzeel, Mügge and Damstra2014; Scola, Reference Scola2007, Reference Scola2013). Thus, the other potential outcome is that, as Wendy Smooth has argued, minority women candidates can benefit from the crossover appeal gained from their “multiple community identifications” (Reference Smooth2006: 411). Philpot and Walton (Reference Philpot and Walton2007) find some support for this hypothesis. In a fictionalized candidate choice experiment, they show that black women candidates were preferred over white male candidates by black female, black male and white female voters and were preferred over black male candidates by black and white female voters. However, in their analysis of national exit poll data from US House of Representatives elections, they found that controlling for partisanship washed away the affinity effects; the overwhelmingly Democratic affiliation of black female candidates explained their support among those voter groups.
In Canada, only experimental research has examined the combined effects of voter sex and race on support for a female minority. Tolley and Goodyear-Grant (Reference Tolley and Goodyear-Grant2014) replicated Philpot and Walton's experiment with a (fictional) Chinese candidate for the visible minority condition. Closest to the conditions of the present study is their “affinity congruence” treatment, in which they compared preferences of Chinese women voters (relative to other subjects) when presented with a Chinese female and a white male candidate. Surprisingly, they found that Chinese women were not more likely than other participants to favour a Chinese woman candidate in this condition. Rather, white women (at 70%) were her strongest supporters, followed by Chinese women (63%), white men (59%) and finally Chinese men (55%). The results in this particular fictional contest suggest a straightforward gender affinity effect, rather than any interaction of gender and ethnic affinities; however their findings across multiple paired contests suggest that “among Chinese Canadians, race is a more salient vehicle for political identification than gender” (Tolley and Goodyear-Grant, Reference Tolley and Goodyear-Grant2014: 21). Thus, while we acknowledge the potential for diminished or accentuated affinity effects when ethnicity and gender intersect, the sole Canadian finding leads us to question whether any interaction will be found at all. We see our study as taking a step toward adjudicating between the conflicting evidence.
Taken as a whole, the literature on gender and ethnic affinity voting leads us to identify three key research questions for our study:
-
1. Did female voters disproportionately support Olivia Chow?
-
2. Did Chinese voters support Chow more than other ethnic groups?
-
3. Did the intersection of gender and ethnicity accentuate, diminish or have no effect on affinity biases?
The accumulation of research on affinity voting suggests that race is more salient than gender as a basis for candidate choice. Further, evidence on the interaction of ethnic and affinity voting is inconclusive, and we have had no previous opportunity in Canada to examine individual-level voters' responses (net of party influences) when a visible minority woman seeks elected office. Turning to the Toronto case, and a more in-depth and contextually specified analysis of voting patterns for a high profile Chinese woman candidate, can thus be a useful step in the development of better theoretical models for explaining the intersection of gender and ethnic affinity voting.
The 2014 Toronto Mayoral Election
Our review of the research in Canada suggests little evidence of gender affinity effects. There is somewhat stronger evidence of ethnic affinity effects, though the findings are contingent upon the methodology used. Observational studies of aggregate and individual voter choice in partisan elections show that the ethnic background of candidates matters little, once partisanship and incumbency are taken into account. Experimental studies consistently find affinity effects; however, these findings leave us with questions about external validity.Footnote 10
To address the persistent uncertainties concerning the role and impact of affinity voting in Canada, we draw upon individual-level data from the 2014 Toronto mayoral contest. This election was exceptionally high-profile in nature and saw a record high turnout of 54.7 per cent. John Tory was elected mayor with 40.3 per cent of the vote, while Doug Ford and Olivia Chow came in second and third, with vote shares of 33.7 per cent and 23.2 per cent respectively. Since the city was created in its present form by merging several previously existing municipalities in 1997, Toronto has never had a female or minority mayor. On two occasions women have come in second place (Barbara Hall in 1997 and Jane Pitfield in 2006), but Olivia Chow's third-place showing in 2014 is the best result ever achieved by a visible minority candidate.
The matter we consider here is whether support for Chow differs among male and female, and minority and non-minority, voters. There are a number of reasons to suspect that visible minorities, as well as women, might demonstrate stronger support for Chow when compared to other voters. First, Chow was the only woman and only minority among the three main contenders for the mayoralty, so voting for one of her main competitors meant voting for a white man. Second, while no candidate in this election ran a campaign that specifically targeted voters on the basis of their sex or ethnicity (see Collet, Reference Collet2008), there were nonetheless several gendered and racial cues that made Chow's characteristics on these dimensions especially salient. Chow received a public endorsement from local Chinese leaders who urged members of the community to support her, citing her record on issues including immigration and multilingual emergency services (Alcoba, Reference Alcoba2014). She received a similar endorsement from a group of notable Toronto women who praised her campaign's focus on affordable childcare and equality for women (Houpt and Church, Reference Houpt and Church2014). Chow's campaign also faced several incidents of overt racism that were well publicized. For example, at one debate in September a Ford supporter heckled her from the audience, saying that she should “go back to China.” And a day before the election, one of the city's major dailies, the Toronto Sun, published an editorial cartoon that depicted Chow in slanted glasses and a Mao Zedong-style tunic, standing on the coattails of Jack Layton, her late husband. In an interview with CP24, Chow called the cartoon “racist and sexist.”Footnote 11 These cues may have served to make both gendered and racialized group identities more salient to voters, and thus to catalyze group membership as a basis for vote choice (Plutzer and Zipp, Reference Plutzer and Zipp1996: 34).
On the other hand, it is not self-evident that affinity voting should have played a strong role. First, while race has been shown to be a salient basis for candidate choice in some US cities, it is less clear whether we should expect similar results in a city like Toronto, which is widely seen as a model of multicultural urban governance, immigrant integration and racial harmony (Bloemraad, Reference Bloemraad2006; Frisken and Wallace, Reference Frances and Wallace2000; Good, Reference Good2009; Graham and Phillips, Reference Graham, Phillips, Banting, Courchene and Seidle2007; Lo, Reference Lo, Price and Benton-Short2008; Siemiatycki and Isin, Reference Siemiatycki and Isin2007). Second, the abundant information about the candidates and issues, as well as the high voter interest in the Toronto mayoral election, may reduce voters' need to rely on “easy” cues such as candidate sex or race. Each of the three mayoral candidates had an exceptionally high public profile and well-known issue positions, partisan backgrounds and ideologies. Chow, a former Toronto city councillor, had a national profile as an NDP member of Parliament alongside her husband and leader of the official opposition, the late Jack Layton. John Tory was former leader of the Progressive Conservative Party of Ontario. Doug Ford, the third main contender, entered the race after his brother Rob Ford, the incumbent, dropped out after being diagnosed with cancer. Rob Ford, of course, was well known in Toronto and became something of an international celebrity after he admitted to substance abuse, including smoking crack cocaine while in office, an issue that was widely reported by media in Canada and abroad. Both Ford brothers were known to be well connected to key members of the federal Conservative party, and Doug had publicly considered running for the leadership of the provincial Progressive Conservatives.
Over the summer and fall leading up to the October election, the candidates faced off in dozens of debates,Footnote 12 including two that were televised. An indication that voters were attentive and interested in the election is the turnout rate, which, at nearly 55 per cent, exceeds the average across Ontario (43%) and rivals the 2014 provincial and 2011 federal elections (51% and 61% respectively). Thus, while we would be unsurprised to see affinity effects operating in very “low information” elections, where there is little else to go on when making a vote choice (Matson and Fine, Reference Matson and Fine2006), to see such effects in the present study would suggest that they are powerful and robust amidst a wealth of readily available information.
Data and Methodology
We evaluate the influence of sex and visible minority status upon vote choice in the 2014 Toronto mayoral race using data from the Toronto Election Study. The TES is a two-wave internet survey of Torontonians that was conducted around the time of the Oct. 27 election. Respondents were interviewed in the weeks before election day (after the candidate nomination deadline of September 12) and then again in the week after the election.Footnote 13 The TES includes a variety of questions about sociodemographic characteristics, attitudes and behaviour, similar to those contained in many national and provincial election studies, and it is the first dataset of its kind that allows for a thorough consideration of voting behaviour in a non-partisan Canadian municipality.Footnote 14
Our analysis consists of two stages.Footnote 15 We begin by describing the voting patterns of TES respondents according to sex and ethnicity and establish that female and visible minority voters were indeed more likely to support Olivia Chow than male or white voters. We then conduct a multivariate analysis to explore the nature of gendered and ethnic differences in vote choice. Our key variables of interest are sex (male/female) and ethnicity (white/Chinese/other visible minorityFootnote 16 ).Footnote 17
Assessing the impact of affinities on vote choice is complicated by the fact that sex and ethnicity correlates can overlap with ideological, policy and interest-based sources of candidate support.Footnote 18 The empirical strategy that we adopt here is to see whether female (ethnic minority) voters are more willing than men (non-minorities) to vote for same-group candidates after controlling for ideological and policy preferences. We do this by including in the model measures of egalitarian attitudes towards women (“Do you think that there should be more women on council?” and a 100-point feeling thermometer on attitudes towards feminists) and visible minorities (“Do you think that there should be more visible minorities on council?” and an identical feeling thermometer on feelings towards visible minorities). We also control for more general ideological and partisanship factors. Chow was widely perceived to be to the ideological left of the other two major mayoral candidates and was linked in the minds of most voters to the NDP.Footnote 19 It is therefore important to control for any overlap between voter sex/ethnicity and these ideological and partisan correlates of her support.Footnote 20 Indeed, the inclusion of such variables is necessary to isolate the relationships between gender, ethnicity and vote choice. If the controls co-vary with gender or ethnicity, failing to account for them may inhibit our ability to identity the effects of gender and ethnicity. If the sex and ethnicity variables have statistically significant relationships to vote choice after the addition of controls, we can be confident that any observed differences in the voting behaviour of our sociodemographic groups of interest (men versus women, visible minorities versus white) are driven by these characteristics, rather than by ideological or policy congruence. If, however, the addition of controls causes sex or ethnicity to no longer display a statistically significant relationship to vote choice, it would suggest that ideological and policy preferences, rather than sociodemographic status itself, are behind the observed gender and ethnic voting patterns.Footnote 21
Results
Describing vote choice
Prior to conducting a multivariate analysis to disentangle the sources of gender and ethnic affinity voting, it is useful to present simple descriptive data. Such an analysis reveals the extent of the difference in voting patterns of men and women, and our three ethnic groups. Figure 1 displays the share of voters in each category that supported each of the three major mayoral candidates.Footnote 22
Figure 1 Vote Choice by Gender and Ethnicity
Figure 1 suggests that Chow performed best among women and visible minorities. Women were 6.4 percentage points more likely to support Chow than were men (p < 0.01). In terms of ethnicity, 19.1 per cent of white voters supported Chow, which is less than either of the minority groups (p < 0.05), though there is no statistically significant difference between Chinese (29.7%) and non-Chinese visible minorities (26.7%). Such a finding suggests some support for the “rainbow coalition” hypothesis.
Explaining vote choice
What accounts then for the patterns observed in Figure 1? Did affinity voting determine ballot choice, or did ideological/policy congruence and attitudinal dispositions associated with group membership drive voting decisions? In order to address this question, a multivariate analysis is necessary. Accordingly, Table 1 contains the results of several logistic regression models, where the dependent variable is vote choice (Chow = 1, other = 0).Footnote 23 We consider a series of different model specifications to disentangle the effects of gender, ethnicity (including the interaction of the two) and attitudinal and ideological factors on vote choice. From models I to III we introduce two sets of controls (egalitarian and ideological). This approach is replicated in models IV to VI, which also consider the interaction of the gender and ethnicity variables.
Table 1 Gender, Ethnicity and Vote Choice (Multivariate Analysis)
Entries report log-odds and standard errors (in parentheses).
***: p < 0.01, **: p < 0.05, *: p < 0.10
Table 1 reveals several findings of note. First, Model I indicates that, without accounting for controls or interactions, both gender and ethnicity are positively associated with a Chow vote. Such a finding is not surprising given the results in Figure 1. A difference between the effects of gender and ethnicity becomes apparent in Model II, however, when attitudinal variables are added and the effect of gender disappears. In this model, attitudes towards women on council and feminists are both significant predictors of a Chow vote, and accounting for these variables nullifies the effect of the gender variable. This pattern holds when ideological measures are added (Model III). For female voters, therefore, it appears to be feminist and egalitarian attitudes and left-wing ideological dispositions, rather than shared membership in the female group, that drives the relationship between gender and Chow support.
In contrast, the effect of ethnicity is robust to the addition of controls in both models II and III; both ethnicity variables (“Chinese” and “non-Chinese visible minority”) retain their statistical significance. In fact, opinions of whether there should be more minorities on council and attitudes towards minorities are insignificant in Model III, suggesting that the relationship between ethnicity and vote choice is based on shared group membership, rather than egalitarian attitudes concerning visible minorities. The relationship between ethnicity and vote choice is therefore very different than the one between gender and vote choice.
Table 1 also suggests that ethnic affinity effects are stronger among Chinese voters than other ethnic minorities. Indeed, the magnitude of the “Chinese” variable actually increases after controls are added (from models I to III), while the effect of the non-Chinese visible minority variable changes only slightly. These results suggest that, after accounting for important attitudinal variables, Chinese voters were more likely to support Chow than were other ethnic minorities, who in turn, supported Chow at greater rates than did white voters.
Finally, the results of models IV to VI suggest that gender and ethnicity have an interactive effect upon vote choice, and this relationship becomes most evident once attitudinal controls are added. Though Model I shows a relationship between vote choice and both gender and ethnicity, the interactions of gender with both ethnicity variables in Model IV (which lacks controls) provides only weak evidence of an interactive effect—the coefficient for the non-Chinese interaction term is significant at p < 0.10 and the Chinese interaction term is insignificant (though positive). However, in Model VI, both interactions are positive and significant, indicating that the interactive effects of gender and ethnicity are found most strongly after the addition of egalitarian and ideological controls. Such a finding suggests the effects of gender upon vote choice are different for minorities than they are for whites, and that these effects are driven by sociodemographic characteristics—thus affinity effects—rather than attitudes.
Interestingly, the female constituent term is negative and significant in models V and VI, which suggests that, after controlling for attitudinal considerations, white female voters were less likely to support Chow than their male counterparts. Females were thus more likely to support Chow than men but only if they were visible minorities. Such a finding reveals that gender does indeed matter for vote choice; its effects are simply conditional upon ethnicity.
To present the results of Table 1 in a more intuitive manner, we present Figure 2, which shows the predicted probability of voting for Chow for each gender and ethnicity group. Values were estimated using postestimation following Model VI, where the values for gender and ethnicity were manipulated and all other values remained unchanged. The figure provides information on the magnitude of the differences between the various sociodemographic categories.
Figure 2 Predicted Probability of Supporting Chow, by Gender and Ethnicity
Figure 2 illustrates in detail the interactive relationship between gender, ethnicity and vote choice. These findings line up quite closely with Philpot and Walton's expectations (2007) regarding the intersectional structure of voter preferences in the US except that, in our case, white women demonstrate the least affinity toward Chow after accounting for ideological and attitudinal dispositions. Among both visible minority groups, however, women were more supportive than men. So while gender has an impact within ethnic groups, the effects are not in a consistent direction.
Overall, the combined effect of gender and ethnicity are significant. White men have a predicted probability of supporting Chow of only 21.3 per cent, while the comparable value for Chinese women is 38.1 per cent. There is little doubt, then, that gender and ethnicity played important roles in the outcome of the 2014 Toronto mayoral election.
Discussion
The 2014 Toronto mayoral election offers a unique setting in which to study affinity voting patterns. The presence of a candidate who is both female and a visible minority, and whose opposition consisted of two white men, provides an ideal opportunity to assess the impact of gender and race, both individually and combined, upon voter behaviour. The availability of survey data from the Toronto Election Study affords us the chance to consider these matters in the context of a genuine election, using individual-level data.
Our analysis reveals several findings of note. First, we find strong evidence of ethnic affinity voting: both Chinese and non-Chinese visible minority voters were more likely than white voters to support Chow, even after controlling for a range of attitudinal, partisan and ideological considerations. This result helps to resolve previous conflicting findings between experimental studies (showing significant effects) and observational studies of federal elections (showing no effects). It suggests that candidate sociodemographics can be relevant to voters beyond strictly low-information contexts. It may thus be the particular importance of parties and the lesser importance of the “personal vote” that trumps affinity voting for local candidates in federal elections, rather than the wealth of information per se.
We also find that Chinese voters are different than non-Chinese visible minority voters in their patterns of support. While there is evidence of inter-minority ethnic affinity, support for Chow among voters from her own ethnic group is almost twice as strong as that from minorities of other ethnic backgrounds. We interpret this as evidence of an inter-ethnic affinity effect among visible minorities.
Our results also point to a fascinating result with respect to the relationship between gender and vote choice. By itself, Model III in Table 1 would seem to suggest that, after controlling for egalitarian attitudes and ideological preferences, gender loses any relationship with vote choice. Such an interpretation is congruent with many other observational studies, from Canada and elsewhere, that argue that women are no more or less likely than men to support female candidates, and that it is rather partisanship and ideology that account for the correlation of voter sex with preferred candidate sex (Dolan, Reference Dolan2008; Ekstrand and Eckert, Reference Ekstrand and Eckert1981; Goodyear-Grant and Croskill, Reference Goodyear-Grant and Croskill2011; Paolino, Reference Paolino1995). By interacting the race and gender variables, however, we reveal that gender is indeed related to support for Chow. After differentiating between voters on the basis of race, TES data suggest the relationship between gender and vote choice depends upon the race of voters. We observe null results for the gender variable when it is not interacted with race because the impact of voter gender on candidate choice runs in opposite directions for white and visible minority voters. Controlling for egalitarian attitudes, ideology and partisanship, white women were less likely to support Chow than were white men, while visible minority women (Chinese or otherwise) offer more support to Chow than did their male counterparts. The reason that white women were less supportive of Chow than were white men is not immediately clear, and future research should consider the mechanisms underlying this result.Footnote 24
In conclusion, we wish to briefly speculate as to what our findings say about the prospects for other visible minority female candidates. The 2014 Toronto mayoral contest, like all municipal elections, was unique. Factors related to the urban political context vary from city to city and are consequential for understanding the presence of women and minorities as mayors (Kaufman, Reference Kaufmann2003b, Reference Kaufmann2004; Smith et al., Reference Smith, Reingold and Owens2012). If our results with respect to gender and ethnicity are applicable elsewhere, however, then the success of visible minority female candidates appears to depend heavily upon the turnout patterns across gender and ethnicity categories. TES data reveal that, though Chow had an advantage among visible minorities, this group voted at a significantly lower rate than white voters (this gap is estimated at seven percentage points). Turnout among visible minority women was also significantly lower (16.5 points) than among visible minority men (both differences are significant at p < 0.01). Thus, those individuals who were most likely to support Chow (minority females) were the least likely to vote of any gender/ethnicity combination. Even in a city such as Toronto, which has a political context that should stimulate ethnic minority representation (that is, a liberal electorate, a sizable visible minority population and a deep candidate pool where minorities enjoy considerable personal and professional resources and access to effective advocacy organizations), such a pattern presents a significant obstacle for a visible minority candidate to overcome. Should our findings with respect to gender and ethnic affinity hold in cities with fewer visible minorities, it is clear that minority female candidates face an uphill battle when attempting to win office at the local level.
Appendix I: Toronto Election Study Questions
Note that all variables have a range from 0 to 1
Sociodemographic characteristics (dummy variables): Gender (male/female), ethnicity (coded as white/Chinese/other visible minority)
Vote choice (nominal variable): Which mayoral candidate did you vote for?
Attitudes towards women (interval-level variable) : Do you agree/disagree that there should be more women on city council? Attitudes towards feminists (0–100)
Attitudes towards visible minorities (interval-level variable) : Do you agree/disagree that there should be more visible minorities on city council? Attitudes towards visible minorities (0–100)
Ideological placement of candidates (interval-level variables) : On a scale from 0 to 10, where 0 means left and 10 means right, where would you place (Doug Ford, Olivia Chow, John Tory)?
Partisanship (dummy variables) : In Federal politics, do you usually think of yourself as a: Liberal, Conservative, NDP, Green, Other, None of the above, Don't know. How strongly do you associate with [answer to question above] party? Very strongly, Fairly strongly, Not very strongly, Don't know. Following Blais et al. (Reference Blais, Nadeau, Gidengil and Nevitte2002), only those respondents who report a “very” or “fairly” strong attachment to a party are coded as partisans.
Quality control question (dummy variable) : To ensure that your browser is downloading the content of this survey correctly, please select option “four” below.
Appendix II: Descriptive Statistics
