CANDIDATE SEX AND VOTING BEHAVIOR

Over the last few years a growing body of research has examined the impact of candidate sex on voters and has addressed the question of whether women are more likely (and men less likely) to support women candidates—and if so, why. The most straightforward possibility relates to identity politics, where voters have a baseline preference for candidates of their own sex (Sanbonmatsu Reference Sanbonmatsu2002). As Johnston et al. (Reference Johnston, Blais, Brady and Crête1992) argue, “the more an agent resembles oneself the more he or she might be expected reflexively to understand and act on one's own interests.” This can provide a powerful motivation to vote for candidates or leaders who embody our own demographic characteristics. Evidence to support this contention has been found in other contexts, specifically with respect to ethnicity—in the United States (Sigelman et al. Reference Sigelman, Sigelman, Walkosz and Nitz1995; Tate Reference Tate1993; Terkildsen Reference Terkildsen1992), Europe (Bergh and Bjørklund Reference Bergh and Bjørklund2011; Teney et al. Reference Teney, Jacobs, Rea and Delwit2010), and India (Heath, Verniers, and Kumar Reference Heath, Verniers and Kumar2015)—but evidence with respect to gender has been far more mixed, and candidate sex has been found to influence voting behavior in some elections but not others (Dolan Reference Dolan1998, Reference Dolan2001, Reference Dolan2004; Dolan and Lynch Reference Dolan and Lynch2014).

In addition to the possibility of a straightforward identity politics effect, where voters prefer candidates of their own sex, the literature suggests preference for, or hostility to, women candidates might be influenced by the perceived feminism of the candidate. In their analysis of exit polls from the 1992 elections, Eric Plutzer and John Zipp found evidence to suggest that a simple pattern of identity politics, whereby women vote for women is not a sufficient explanation for who votes for women candidates (Plutzer and Zipp Reference Plutzer and Zipp1996). According to their evidence, it would seem that in a number of cases women are indeed more likely to vote for women, but this effect is most pronounced when the woman candidate is perceived as being profeminist. When candidates are perceived as “acting for,” rather than merely “standing for,” women, they stimulate voting based on gender identity (34).

This line of work has been developed in a number of different directions to examine the character traits and perceptions of male and female candidates. Recent research suggests that voters tend to assess men and women candidates differently by using stereotypes about their perceived strengths and weaknesses (Cutler Reference Cutler2002; Huddy and Terkildsen Reference Huddy and Terkildsen1993a; McDermott Reference McDermott1997, Reference McDermott1998; Schwindt-Bayer, Crisp, and Malecki Reference Schwindt-Bayer, Crisp and Malecki2008). Gender stereotypes portray women as approachable and empathetic and men as assertive leaders (Alexander and Andersen Reference Alexander and Andersen1993; Burrell Reference Burrell2008; Huddy and Terkildsen Reference Huddy and Terkildsen1993b; Kahn Reference Kahn1996; King and Matland Reference King and Matland2003; Lawless Reference Lawless2004; Leeper Reference Leeper1991; Paul and Smith Reference Paul and Smith2008; Sapiro Reference Sapiro1981/1982). Voters tend to use these gender stereotypes to assess men and women candidates’ policy competencies differently, with women seen as more competent on matters of health and education and men seen as more competent on matters of the economy and terrorism (Dolan Reference Dolan2010). Perhaps as a result of gender stereotypes equating women with higher levels of compassion and empathy, studies have found that women candidates are perceived as more liberal than men running for election (Koch Reference Koch2000, Reference Koch2002; McDermott Reference McDermott1997), and voters who are worried about levels of honesty and integrity in government have been more likely to vote for women candidates (Dolan Reference Dolan2004; McDermott Reference McDermott1998). Although women candidates often believe they have to be better than men to succeed (Lawless and Fox Reference Lawless and Fox2010), this is a misperception, and as candidates become known to voters, gender stereotypes play far less of a role. Recent evidence suggests that gender stereotypes in candidate evaluations have no effect on vote choice (Brooks Reference Brooks2013; Dolan Reference Dolan2014a, Reference Dolan2014b; Dolan and Lynch Reference Dolan and Lynch2014).

Studies in the United States have also tended to find stronger patterns of candidate gender voting among independents, who do not have such strong partisan ties and so may be more influenced by candidate traits (e.g., Fulton Reference Fulton2014). Other studies have found that candidate gender voting is shaped by partisan affiliation and that people are less likely to vote for Republican women candidates than Democrats (King and Matland Reference King and Matland2003; Lawless and Person Reference Lawless and Person2008). Dolan (Reference Dolan2004) suggests that this is because issue and trait stereotypes of women and of Democrats are generally consistent with each other, while stereotypes of women and of Republicans are more at odds with each other.

One area that has received less attention, though, is public opinion toward the question of women's representation in elected office (Dolan Reference Dolan2010, 70). Previous work in this area has examined public attitudes toward descriptive representation (Campbell, Childs and Lovenduski Reference Campbell, Childs and Lovenduski2010) and how these attitudes shape evaluations of women candidates (Dolan Reference Dolan2010; Rosenthal Reference Rosenthal1995), but as yet there has been little scholarly work on how these attitudes directly influence actual behavior. In this article we contribute to this emerging literature by examining whether public attitudes toward the descriptive representation of women condition the extent to which candidate sex influences voting behavior. We expect voters who support increasing the descriptive representation of women to be more likely to vote for women candidates than those who do not, but we expect the effect to be more pronounced among women than men—as the issue is likely to be more salient to women.

GENDER AND THE 2010 BRITISH GENERAL ELECTION

In order to test our research question, we focus on the 2010 British election. As set out above, previous research has found that the extent to which gender is politically salient depends upon the political context. For example, Dolan (2014, 24) argues that the role candidate gender plays in voter calculations varies with the electoral context and is more important in some contests than others. She found that women were more likely to vote for women candidates in 1992 than in other election years because of the high salience of gender in the 1992 election, which was commonly referred to as the “year of the woman” (Dolan Reference Dolan2001, Reference Dolan2004). Plutzer and Zipp also single out 1992 because of the salience of gender in the election and thus thought that should women ever vote for women, they would certainly do so in this year.

Although it is hard to think of a similar example where gender has played such a significant role in election reporting in the British case, 2010 provides us with the best opportunity in recent times to do so. Since 1997 the Labour Party easily has been the most feminist-oriented political party in Britain, but after his election as Conservative leader in 2005, David Cameron sought to challenge the Labour Party on this terrain. Cameron made improving the representation of women in parliament a significant part of his bid to “modernize” or decontaminate the Conservative brand. He needed to shake off the “nasty party” image, and embracing gender equality, along with environmentalism and other liberal values, was part of the strategy (Childs and Webb Reference Childs and Webb2012). In his first speech as party leader Cameron stressed the importance of improving the representation of women in politics and talked about work-life balance as one of the most pressing issues facing contemporary British families. His controversial A-list policy—which required that women make up 50% of candidates on the party's preferred candidate list—was substantially watered down in response to a backlash from party members. However these internal divisions kept the issue on the media's agenda.

In addition to controversies over the selection of women candidates, the issue of women voters, or, more specifically, mothers’ voting preferences, became a highly salient feature of the media coverage of the election. In the long campaign all three party leaders participated in web chats on the mumsnetFootnote ¹ website at their own request, and the news frame “the mumsnet election” was directly referred to in 43 national newspaper articles.Footnote ² In fact Labour's election campaign coordinator, Douglas Alexander, was on record saying, “Labour needs to win back middle-income female voters with children in marginal seats.”Footnote ³ Each of the three main parties devoted considerable space in their manifestos to flexible working hours, parental leave, and preschool child care provision. The Conservatives pledged to make Britain the most family-friendly country in Europe, and the Labour and Liberal Democrat manifestos also made specific reference to sex discrimination in the work place and gender disparities in pay. The salience of “family issues” and other traditional values is measured by the Manifestos Research Group over time (Volkens et al Reference Volkens, Lehman, Merz, Regal and Werner2014).Footnote ⁴ As Figure 1 shows, these issues were far more salient in 2010 than they had been previously.

Figure 1.

Salience of “family issues” and other traditional values in party manifestos.

Source: Manifesto Research Group.

The 2010 election saw more women candidates and more women winners than ever before, creating a step-change in the partisan balance of women's representation in parliament. As a result of selecting more women candidates in winnable seats the Conservatives more than doubled the number of Conservative women MPs, from 17 in 2005 to 48 women in 2010 (Campbell and Childs Reference Campbell, Childs, Geddes and Tonge2010). The effect of these changes in the supply of women candidates has important methodological implications. When only the Labour party put forward women candidates with any serious chance of winning, it was only possible to examine whether voters would offer support (or not) to the Labour party on the basis of the candidate's sex. In 2010, for the first time parties other than Labour put forward women candidates in winnable seats, making it possible to try and untangle the relationship between candidate sex, political party and vote choice to a much greater extent than previously. Regardless of the salience of gender as an electoral issue, it would simply not be possible to sensibly carry out this type of analysis on earlier British elections since too many of the women candidates stood for only one of the major parties.

Despite the controversy surrounding the selection of so many women as Conservative candidates, by and large those women who did stand for election did just as well as their male counterparts (see Table 1). Although male Conservative candidates tended to do slightly better than female Conservative candidates, both in those seats where the Tories were the incumbent (51.4% vs. 49.7%) and in those seats where they were the challenger (28.7% vs. 26.3%), the differences are relatively small and may just reflect differences in the winnability of the seats for which men and women were selected. If we compare the average swing for men and women Conservative candidates (which controls for the winnability of the seats), the differences are even smaller and do not reach conventional levels of statistical significance. Where women stood as the Conservative candidate, the party did 3.4 percentage points better than they did in 2005, and where men stood, the improvement was 3.9 percentage points. Thus there does not appear to be any evidence to suggest that putting forward so many women candidates cost the Conservatives votes. Similarly, there is little to suggest that women candidates for Labour did any worse than men. Women candidates for the Liberal Democrats did slightly worse than men in the seats which the party held in 2005: where women stood there was a 3.4 percentage-point swing against the Lib Dems, and where men stood, there was a 0.4 percentage point swing toward the Lib Dems, but, again, this difference was not statistically significant, and we should probably treat these results with caution since there were only 12 women candidates standing for the Lib Dems in seats they won in 2005.

Table 1.

Performance of male and female candidates at the 2010 election

Note: P-values from a between groups comparison of means, ANOVA.

These findings show that, on average, men and women candidates performed more or less equally well for each of the three main parties, which suggests that the overrepresentation of men in Parliament owes more to inequalities in the selection process than it does to discrimination by voters. However, as Dolan (Reference Dolan2010) cautions, just because real-world election results suggest that women do not suffer disproportionally at the ballot box because of their sex, we should not assume that this means that voter attitudes about gender are irrelevant to politics, or even irrelevant to vote choice. We can only infer so much from aggregate level data about individual behavior, and an individual's attitudes toward descriptive representation and gender stereotypes may shape an individual's vote-choice decision when faced with a woman candidate.

MATERIALS AND METHODSFootnote ⁵

Previous studies on candidate sex effects have been hampered by the small number of women candidates who run for office in winnable seats. This creates problems at the analysis stage since inferences about whether or not women are more likely to support women candidates are often based on very small samples of women candidates. This increases the likelihood of outliers—or unique personalities—driving the results, and just as we would be unwise to infer too much about gender voting from whether someone supported Mrs Thatcher or not, we should also treat with caution the results of analysis based on relatively few women candidates, since prominent female politicians may in a sense transcend gender. The increased supply of women candidates in winnable seats at the 2010 election helps us to overcome this problem and provides a valuable opportunity to study the effect of gender away from the glare of the media spotlight on one or two key personalities and examine how candidate gender influences voting behavior across a range of different political contexts.

In order to test our hypothesis about the effect of candidate sex on voting behavior, we link data on candidate sexFootnote ⁶ and constituency characteristics to cross-sectional survey data. The individual-level data on voters come from the 2010 British Election Study.Footnote ⁷ The survey was administered to a random sample of 3,075 respondents in 205 constituencies. After linking the survey data to the candidate data, our sample includes 154 women candidates (54 Labour; 58 Conservative; and 42 Liberal Democrat). This provides us with a robust sample to test the impact of candidate sex on voting behavior.Footnote ⁸

Attitudes toward Descriptive Representation

Our key theoretical variable of interest is attitudes toward the descriptive representation of women. To measure these attitudes, we draw on a special module in the British Election Study, which includes the following four items on gender equality:

1. 1. Women are better representatives of women's interests than men.

2. 2. Most men are better suited to politics than are most women.

3. 3. Women need to get more involved in politics to solve problems that concern them.

4. 4. A husband's job is to earn the money; a wife's job is to look after the home and family.

For each item, respondents could agree strongly, agree, neither agree nor disagree, disagree, disagree strongly, or refuse to give an answer. Descriptive statistics are shown in Table 2, which reports the mean for each item on a scale of 1 to 5, where 1 is strongly agree and 5 is strongly disagree. With respect to attitudes about descriptive representation, both men and women were more likely to agree than disagree with the statement that women are better representatives of women's interests than men, and women need to get more involved in politics to solve problems that concern them (as illustrated by mean values greater than 3, where 3 indicates neither agree nor disagree). Women were significantly more likely to agree with these statements than men, particularly on the issue of women getting more involved in politics; where 67% of women agreed with the statement compared to 53% of men.

Table 2.

Attitudes toward gender equality

Note: Items are measured on a 1−5 scale where 1 = strongly agree and 5 = strongly disagree. *p < 0.05; **p < 0.01; ***p < 0.005.

The other two items are less directly related to issues of descriptive representation and tap more into attitudes about traditional gender roles. Both men and women were more likely to disagree than agree with the statements that men are better suited to politics than women and that a husband's job is to earn the money (as illustrated by mean values less than 3). Once again women were significantly more likely than men to disagree with these statements. These results reflect the findings of Dolan's 2013 U.S. study: “women are more likely than men to have a baseline preference for women candidates, to want to see more women in office, to support gender parity in elected office and to support same-sex representation” (Dolan 2014b, 69).

In order to check whether these items do indeed tap into attitudes about descriptive representation and traditional gender roles specifically rather than attitudes toward gender equality in general, we undertake a principal components analysis of the responses to all four items (Table 3). These items load on to two clear dimensions (Campbell, Childs, and Lovenduski Reference Campbell, Childs and Lovenduski2010). As such, rather than simply combining items relating to gender equality in the BES data we create two additive scales. The first combines items one and three to create a scale measuring attitudes toward the descriptive representation of women. The second scale, measuring attitudes toward traditional gender roles, combines items two and four. Each scale includes a range of valid responses from zero to eight, where eight represents a more feminist position: either high levels of support for the descriptive representation of women or opposition to traditional gender roles.

Table 3.

Principal components analysis of attitudes toward gender equality

The mean score for women on the descriptive representation scale is 4.79 compared to a mean score of 4.49 for men. (The sex difference is significant at the 0.01 level.) The mean score for traditional gender roles is 5.34 for women and 5.03 for men. (The sex difference is significant at the 0.01 level.) Our key hypothesis concerns attitudes toward descriptive representation: we expect voters who are in favor of the descriptive representation of women to be more likely to vote for women candidates, but we expect the effect to have a more pronounced effect among women than men, as the issue is likely to be more salient to women. By way of comparison we also examine public attitudes toward traditional gender roles, which according to prior theory should be empirically unrelated to vote for women candidates (Brooks 2013; Dolan Reference Dolan2014a, Reference Dolan2014b; Dolan and Lynch Reference Dolan and Lynch2014).

Modeling the Impact of Candidate Sex on Vote Choice

Our main hypothesis is to investigate the link between attitudes toward descriptive representation and support for women candidates. Do women who hold strong feminist views about the importance of getting more women into politics put these views into practice when they have the chance by voting for women candidates? Or do partisan ties override feminist principles?

There are a number of challenges involved in estimating the effect of candidate traits on voting behavior in multicandidate elections, which make common econometric choice models, such as multinomial logit, inadvisable (see Alvarez and Nagler Reference Alvarez and Nagler1998). One solution is the conditional logit model, which is “conditional on the characteristics of the choices” (Alvarez and Nagler Reference Alvarez and Nagler1998, 56) and so can take into account candidate characteristics in a parsimonious manner. In order to estimate the effects of candidate sex on voting behavior, we therefore specify a conditional logit model (Long and Freese Reference Long and Freese2005). The model uses stacked data, with three rows for each respondent, one for each party. The conditional logit model allows us to fit how the choice between party alternatives is affected by the characteristics of the candidates. In formal terms Long and Freese (2005: 307) specify the model as:

$$\Pr (y_i = p/x_i, z_i ) = \displaystyle{{exp\left( {z_{ip} \gamma + x_i \beta _p} \right)} \over {\mathop \sum \nolimits_{j = 1}^J exp\left( {z_{ij} \gamma + x_i \beta _j} \right)}},$$

where z_im contains values of the alternative specific variables for alternative p and case i. In our model there are three alternatives for party p: Labour, Conservative, and Liberal Democrat. Then ϒ is a parameter indicating the effect of the alternative specific variables on the probability of voting for one party over another. In general, for each variable z_k, there are J values of the variable for each case, but only the single parameter ϒ_k. This enables a single coefficient for the effects of candidate sex to be estimated regardless of the candidate's party. We also tested for evidence of variation in these coefficients across parties. As in a multinomial logit model, x _i contains case-specific independent variables for case i, and β_p contains coefficients for the effects on alternative p relative to the base alternative (Labour).

The control variables present in the models for both men and women are Party Like Score (an eleven-point scale from 0 = strongly dislike to 10 = strongly like the party), which captures both positive and negative disposition to each of the parties; Contacted by party (1 if the respondent was contacted by the party during the campaign, or 0 otherwise), which captures mobilization and campaign effects; Incumbency, which captures incumbency advantage, and Distance from Contention (for third- and lower-placed parties this is the difference between the 2005 share of the vote for the party and that for the second-placed party, and 0 if the party came first or second in the constituency), which captures supply side factors to do with the allocation of tickets to women candidates. In particular women selected for Labour from all-women-shortlists (AWS) tend to be selected in competitive seats more often than women candidates from other parties. Finally we control for Candidate sex (1 if the candidate is a woman, or 0 if a man), which is our key theoretical variable of interest. Descriptive statistics are shown in the appendix. All of these controls are alternative-specific, so their values can vary across the party options for a given respondent. This enables a single coefficient for each variable to be estimated regardless of party. In practice this is equivalent to assuming that the magnitude of the effect for each of these variables is constant across party options. We tested the validity of this assumption by specifying party-interactions (not shown), none of which were significant.

The model also allows case-specific variables as predictors. The attitudinal variables that we examine as a potential moderator of candidate sex effects are attitudes toward descriptive representation and attitudes toward traditional gender roles discussed above. Both of these variables are measured on a 0–8 scale, where higher values indicate a more feminist position. Where a control variable is case-specific (say X), two coefficients are estimated (Con × X and LD × X) for the effects on voting Conservative and Liberal Democrat, respectively, each relative to the baseline Labour (just as a regular multinomial logit/probit model would have two such coefficients).Footnote ⁹ In order to test our main hypotheses we also specify an interaction effect between candidate sex and attitudes toward gender.