a1 American University
Research has shown that voters are willing to stereotype candidates on the basis of their gender, which can sometimes pose obstacles and sometimes prove advantageous for female politicians. But the literature is uncertain about how candidate gender interacts with candidate party affiliation to shape voters' perceptions. In this article, I draw on political psychology, the women and politics literature, and recent work on partisan “trait ownership” to suggest that the application of gender stereotypes will be limited by the salience of partisan stereotypes. I use nationally representative survey data and a content analysis of news coverage from the 2006 U.S. Senate elections to test the argument. Focusing on voter evaluations of candidate traits, I find that party stereotypes are more powerful than gender stereotypes, and that assessments of candidate attributes can be affected by news coverage when candidates are portrayed in ways that challenge traditional partisan images. The results suggest that gender stereotyping is limited by the relevance of party stereotypes, and that as the Republican and Democratic parties continue to polarize at the elite level, the importance of partisan stereotyping is likely to increase.
(Online publication June 06 2011)
Danny Hayes is Assistant Professor of Government in the School of Public Affairs at American University, Washington, DC 20016: firstname.lastname@example.org
Thanks for helpful comments and assistance of various sorts are due to Kristi Andersen, Deb Brooks, Elizabeth Cohen, Johanna Dunaway, Kim Fridkin, Matt Guardino, Jon Hanson, Jen Lawless, Jenn Merolla, Zoe Oxley, Sarah Pralle, Hans Peter Schmitz, Nick Winter, several anonymous reviewers, audiences at the Spring 2008 New York Area Political Psychology Workshop and the 2009 Midwest Political Science Association meeting, and seminar participants at Syracuse University. Thanks also to Daron Shaw and Brian Arbour for help acquiring the data and to Rebekah Liscio for research assistance.
List of Figures
Figure 1. Average trait ratings for male and female U.S. Senate candidates, by party, 2006 Cooperative Congressional Election Study. None of the differences between ratings of male and female candidates is statistically significant. The number of respondents for each trait comparison ranges from 463 to 466.
Figure 2. Average trait ratings for Republican and Democratic U.S. Senate candidates, by gender, 2006 Cooperative Congressional Election Study. Asterisks denote difference between candidate ratings as statistically significant, p < .05. The number of respondents for each trait comparison ranges from 463 to 466.
List of Tables
Table 1. Distribution of candidate gender and party combinations in U.S. Senate campaigns, 2006 Cooperative Congressional Election Study
Table 2. Portrayal of candidate traits in newspaper campaign coverage, 2006 U.S. Senate elections
Table 3. Model of Senate candidate trait attribution, 2006 Cooperative Congressional Election Study
Table 4. Model of trait attribution differential between Republican and Democratic Senate candidates, 2006 Cooperative Congressional Election Study
To be a woman running for office is to face the prospect of gender stereotyping. Decades of research has shown that voters tend to view female candidates as more liberal, less capable of handling national security and foreign policy, more adept at dealing with social welfare issues like education and health care, and less tough but more empathetic than male candidates. To be sure, the underrepresentation of women in office in the United States stems from factors well beyond public opinion (e.g., Lawless and Fox 2010). And whether these perceptions translate into obstacles or advantages for women depends on the political circumstances. But gender stereotypes can be “potent” (McGraw 2003, 402) and demand the attention of candidates for political office.
While political scientists have documented voters' willingness to apply stereotypes in evaluating hypothetical and fictional candidates in laboratory and survey experiments, the effect of gender stereotyping amid the cacophony of real-world campaigns is less certain. In particular, we know relatively little about how voters integrate multiple stereotypes of candidates into their judgments—in particular, the way that candidate gender interacts with candidate party affiliation to shape perceptions of politicians. Moreover, many of the studies most frequently cited as evidence for gender stereotyping were published a decade or more ago (e.g., Alexander and Andersen 1993; Huddy and Terkildsen 1993). With changes in attitudes about women's role in society and an increase (albeit small) in the number of female candidates for political office, an updated assessment of the prevalence of gender stereotyping is in order.
In this article, I draw on theories from political psychology, the women and politics literature, and recent research on partisan “trait ownership” in American elections to explore the conditions under which gender stereotyping is likely to occur. This synthesis suggests that the application of gender stereotypes will be limited by the greater salience of partisan stereotypes and that the media can play a role in shaping candidate evaluations.
In contrast to most previous studies, mine examines stereotyping outside the laboratory, using national survey data and a content analysis of news coverage from 30 U.S. Senate campaigns during the 2006 midterms. This allows for an examination of the extent to which social stereotypes about women influence evaluations of female politicians when gender is just one of the many pieces of information voters have at their disposal. My findings, which focus on the assessment of candidate traits, show that gender is much less powerful than candidate party affiliation in shaping voters' perceptions. In addition, voters view candidates more favorably when media coverage portrays them in ways that challenge traditional partisan stereotypes. The results suggest that the use of gender stereotypes in campaigns depends on their applicability, but that this is often limited by party stereotypes. And as the Republican and Democratic parties continue to polarize at the elite level, the importance of partisan stereotyping is likely to increase.
Stereotypes are relevant in politics because citizens are willing to devote only limited time to thinking about political matters. As a result, political judgments—whether about issues, events, or candidates—are often the result of a few salient cues. Stereotyping is the assignment of “identical characteristics to any person in a group regardless of the actual variation among members of that group” (Aronson 2004, 244). And among the most well-established stereotypes in person perception is an individual's sex (e.g., McKee and Sheriffs 1957), as women tend to be viewed as “warmer” and more empathetic than men, but less assertive and competent.
A long line of literature has found that the assignment of attributes based on gender extends to politics. Female candidates tend to be perceived as more liberal than their male counterparts (Alexander and Andersen 1993; Huddy and Terkildsen 1993; King and Matland 2003; McDermott 1997), and even more liberal than they actually are when compared to their congressional voting records (Koch 2000, 2002). As a result, women are advantaged among left-leaning voters, but at a disadvantage among conservatives, especially Republicans (King and Matland 2003; Sanbonmatsu and Dolan 2009). Similarly, citizens attribute policy expertise by sex, viewing women as more capable of dealing with social welfare issues—health care, education, and the like—while seeing men as better suited to handle foreign policy, defense, and crime (Alexander and Andersen 1993; Huddy and Terkildsen 1993; Lawless 2004). These stereotype-based assessments can pose obstacles for women (Fox and Oxley 2003; Kahn 1994, 1996), although female candidates can also benefit when the electoral environment makes women's issues salient (Paolino 1995) or when citizens are looking for an “outsider” candidate (Dolan 1998; Fridkin and Kenney 2009).
Gender, of course, is not the sum total of the information that voters possess about a candidate, and in recent work, attention has been turned to the circumstances under which gender stereotyping is most likely to occur. In particular, scholars have focused on whether party affiliation—itself a powerful stereotype that shapes perceptions of candidates (Rahn 1993)—conditions the effects of gender stereotypes (Dolan 2004; Huddy and Capelos 2002; Koch 2002; McDermott 1997). It is not clear from the existing literature, however, how gender and party stereotypes interact to shape candidate evaluation (McGraw 2003), or how other information from the political environment is integrated into judgments about candidates (see Winter 2010 for a compelling discussion of the relationship between gender and party in American politics).
Studies that have directly engaged this question have drawn different conclusions. Sanbonmatsu and Dolan (2009) conclude that even in the presence of party information, gender stereotypes remain powerful influences on voter attitudes. Using survey data in which respondents were asked questions about generic male and female candidates of each party, the authors find that voters use gender stereotypes to assess the issue competencies and positions of both Democratic and Republican candidates. For example, when asked whether a male or female Democrat would handle education better, voters gave substantial advantages to the woman. Male candidates were advantaged when the issue was crime. The pattern held when respondents were asked about a Republican man or woman. That, combined with evidence that voters believed female candidates of both parties would adopt more liberal positions on abortion, leads Sanbonmatsu and Dolan (2009, 490) to conclude that “[a]lthough it is often argued that any gender effect will disappear in the presence of the party cue, we find that gender stereotypes transcend party.”
Another study by Huddy and Capelos (2002) tested a “parallel processing” model (Kunda, Sinclair, and Griffin 1997; Kunda and Thagard 1996) that suggests individuals could integrate multiple pieces of stereotypical information about a person—for example, sex and party affiliation—in the process of forming an impression. For instance, a voter could evaluate a male Republican candidate as especially conservative, since both the gender and party cues suggest that the candidate will lean to the right. But when a candidate possesses cross-cutting cues—a female Republican, for example—the voter may assess the candidate as ideologically moderate (see Koch 2002). Huddy and Capelos's (2002) survey experiment, however, finds little support for the model. Instead, in almost every case, a candidate's party label dominated sex as a predictor of voter attitudes, suggesting very little integration of the two stereotypes. King and Matland's (2003) experimental study of Republican candidates found something similar: Voters did not see female GOP candidates as more empathetic or weaker leaders than male Republicans (King and Matland, Table 2). These disparate findings—on the one hand, that gender stereotypes transcend party, and on the other, that party stereotypes override gender images—suggest the need for additional theorizing in order to understand how party and gender may influence voters' perceptions when, as in general elections, both cues are available.
While this article does not explicitly examine the cognitive processes that lead to the use of stereotypes, the literature on attitude accessibility (e.g., Fazio 1995; Huckfeldt et al. 1999) provides a useful model for thinking about party and gender stereotyping. Research has shown that people often make political judgments based on the considerations that are most easily retrieved, those at the “top of the head” (Taylor and Fiske 1978; Zaller 1992; Zaller and Feldman 1992). Attitudes become more accessible the more frequently they are activated (Fazio, Powell and Herr 1983; Fazio and Williams 1986). For example, a person who regularly thinks about politics in ideological terms is more likely to draw on ideology as a cue in making political judgments (Huckfeldt et al. 2005). As a result, the more “chronically accessible” a consideration is, the more likely it is to become the foundation for political judgment (Huckfeldt et al. 1999).
Stereotypes are a subset of attitudes, and accessibility arguments in social psychology have been used to explain their application. The concept of “stereotype accessibility” suggests that when a social stereotype is easily retrieved, it is more likely to be used to make an evaluation (Stangor 1988). This finding has emerged in research on sex and race (Dijksterhuis, Macrae, and Haddock 1999; Payne, Lambert, and Jacoby 2002). Similarly, Huckfeldt and his colleagues (2005, 12) argue that “a heuristic is more useful if it is more readily available as an organizing principle and hence more frequently deployed in formulating political opinions.”
The question then becomes the circumstances under which gender and party stereotypes are likely to meet those terms. There is no doubt that the most frequently deployed heuristic in American politics is the political party (Lau and Redlawsk 2001; Rahn 1993). Voters routinely rely on party stereotypes to make “theory-driven” judgments about candidates' issue-handling abilities, issue positions, and other attributes (Conover and Feldman 1989; Hamill, Lodge, and Blake 1985; Petrocik 1996). More generally, citizens see the political world through a partisan lens, a view that is encouraged by political debate and news coverage that focuses intently on differences and conflict between Republicans and Democrats. “Candidates are partisan creatures, born of party primaries, vying for jobs in intensely partisan institutions,” write King and Matland (2003, 607). “Even more important, voters see candidates first and foremost as partisans.”
Moreover, the utility of a party heuristic is often validated and reinforced by policy differences between the parties. Consider a voter who casts a ballot for a Republican candidate with the expectation that the politician will promote conservative policies. When the voter's newly elected representative then votes in Congress for a tax cut, the use of the partisan heuristic is rewarded. As a result, party affiliation will likely be used to guide subsequent political judgments. This may be particularly true in an era of increasing elite polarization (Fiorina, Abrams, and Pope 2006; Hetherington 2001).
By contrast, candidate gender is less frequently promoted as a basis for political judgment. While the gender of a candidate is obvious to citizens, its ubiquity and availability do not imply that a sex stereotype will be employed. Its application depends not only on its mental activation but also on the appropriateness of the stereotype to the judgment task (Kunda and Spencer 2003). To be sure, the rise of “women's issues” can make a politician's gender especially salient in a campaign, but such circumstances represent the exception to the rule in American politics (Dolan 1998; Paolino 1995). Although sex pervades the structure and content of American politics (Carroll and Fox 2006; Winter 2010), most campaigns are not explicitly “gendered.”1
This perspective could help explain Sanbonmatsu and Dolan's finding that gender stereotypes matter more than party. In their design, respondents were asked whether, for example, “a Democrat who is a man” or “a Democrat who is a woman” is more capable of handling a particular issue. The structure of the question encouraged respondents to draw on a gender stereotype to make a judgment. Since the only piece of discriminating information between the two candidates was their gender, it was by default a salient and accessible cue. In other words, gender was primed as a criterion for judgment, whereas political party was made irrelevant.2 Thus, their study may tell us much about how voters assess male and female candidates in partisan primaries or nonpartisan elections. But it can tell us little about how voters evaluate candidates when multiple stereotypes about a candidate are potentially salient.
From this perspective, we would expect that when voters know both a candidate's gender and party affiliation, as they do in most elections, party affiliation is likely to exert a stronger effect on candidate evaluation. But since the influence of party and gender may also be conditioned by other information—especially media coverage—it is important to specify a role for information in the political environment. Work by Kahn (1996) and others suggests that when female candidates are portrayed in the news in gender-stereotypic ways, they are more likely to be viewed by voters in traditionally female terms. For example, if a candidate is portrayed in a typically “feminine” way—lacking toughness, for example—that may raise the accessibility of a gender stereotype and lead voters to judge the candidate on the basis of gender. If such news coverage activates latent gender stereotypes, the media may play a role in making those schematic images salient for voters.
Virtually no research, however, has combined data on citizens' political judgments with media coverage to examine how party and gender stereotypes interact. Given the central importance of information flow in conditioning the way citizens see the world, the absence of such an investigation leaves us without a full understanding of how and when stereotypes influence political judgments.
I test these arguments by examining the way that voters assess the personal attributes, or traits, of candidates for political office in the United States. Considerable work has examined the effects of gender on views of a candidate's ideology and issue competencies, but there is a dearth of research on how candidate gender influences perceptions of his or her personality traits. I focus on candidate traits for three reasons.
First, traits represent a domain in which sex and party stereotypes overlap considerably. Research in social psychology has shown that men and women tend to be perceived as representative of different personality attributes (Best and Williams 1990; Broverman et al. 1972; McKee and Sheriffs 1957). Women are more likely to be perceived as possessing traits associated with “warmth”—compassion and empathy, for example—whereas men are more likely to be seen as possessing traits associated with “competence”—leadership ability or assertiveness. Recent work on presidential elections, however, has suggested that similar cleavages exist in evaluations of Republicans and Democrats.
In a study of the seven presidential campaigns since 1980, I show that voters have perceived GOP candidates as stronger leaders than Democratic candidates, a stereotypically “masculine” characteristic (Hayes 2005). At the same time, Democratic candidates were always rated as more representative of “feminine” traits—compassion and empathy—than Republicans. The data suggest the existence of partisan “trait ownership” in American elections (Hayes 2005), a pattern derived from the issues each party is typically seen to handle most adeptly (Petrocik 1996). In the same way that voters apply party stereotypes to infer candidates' issue positions (Hamill, Lodge, and Blake 1985; Rahn 1993), they rely on party images in evaluating personal attributes. Thus, this overlap of sex and party trait images presents an opportunity to examine how voters respond to both cross-cutting and reinforcing stereotypes.
Second, traits are important because they are theorized to be the source of the differences in the perceptions of male and female politicians' ideological positions and issue competencies. The argument, articulated most clearly by Huddy and Terkildsen (1993), contends that because women politicians are likely to be seen as possessing feminine traits, but not masculine ones, they are stereotyped as adept at handling “compassion” issues, such as health care. Male politicians, likely to be seen as stronger leaders, are perceived as better able to handle foreign policy, defense, and crime. But though the argument seems plausible, Huddy and Terkildsen's study, like others (Kahn 1992; Rosenwasser and Seale 1988), does not directly test the trait-stereotyping hypothesis in the absence of an experimental manipulation. Rather, subjects were presented with information about hypothetical male and female candidates, and those experimental manipulations were used as predictors of subjects' assignment of different attributes, ideological positions, or issue competencies.3
Third, considerable work has shown that traits can have a direct and important influence on voting behavior itself. Trait perceptions affect candidate evaluation and vote choice at the presidential level (e.g., Bishin, Stevens, and Wilson 2006; Hayes 2009), and several recent articles have shown similar effects in subpresidential contests (Druckman 2004; Fridkin and Kenney 2011; Hayes 2010). In other words, trait evaluations do more than simply shape perceptions of candidates' ideological or issue profiles. For these reasons—and because much of the widely cited work on trait stereotyping is close to two decades old—it is important to clarify the role that gender and party play in influencing assessments of candidate traits.
In summary, the basic hypotheses to be tested are 1) that candidate trait evaluations reflect gender stereotypes, 2) that candidate trait evaluations reflect party stereotypes, and 3) that news coverage that portrays candidates in a gender-stereotypic fashion promotes gender stereotyping. The findings should yield a sense of how gender and party stereotypes interact to influence trait attribution, something not well understood in the existing literature (McGraw 2003).4
Much of the work on voters' perceptions of candidate traits relies on laboratory experiments, which give researchers firm control over the exposure of subjects to candidates' characteristics. But as noted, many of those studies are limited in the light they shed on individual information processing amid the cacophony of real-world campaigns. Because I am interested in the perceptions of candidates when voters have multiple cues from which to draw inferences—specifically, gender and party—I choose to use survey data from the 2006 elections.5
Survey data were collected through the Cooperative Congressional Election Study (CCES), a collaborative enterprise of more than three dozen academic institutions that organized a survey of Americans during the 2006 midterm election campaigns. The nationally representative online survey was conducted by Polimetrix, Inc. Each participating university was given the opportunity to design a questionnaire administered to a representative sample of 1,000 Americans, who also filled out a longer survey of “common content” questions. The key measures in this article were included in the University of Texas module in the CCES.6 The survey included a host of measures typical in congressional election studies: demographic questions, positions on issues, ideological views, candidate evaluation, and vote choice.
The module also included questions about the personality traits of the Republican and Democratic Senate candidates in each respondent's state, measures rarely included on nonpresidential survey instruments (Fridkin and Kenney 2011). Because of survey time constraints, roughly half of the 1,000-person sample was asked the trait battery. In all, the trait measures were asked of about 500 respondents from 30 states with Senate elections in 2006.7 Aggregating survey respondents from multiple states is a common technique in Senate election research (Kahn and Kenney 1999; Koch 1999; Lau and Pomper 2001).
In the course of the survey, respondents were asked to rate their Republican and Democratic Senate candidates on several personality traits. The questions included two “feminine” traits—“really cares about people like me,” a proxy for empathy, and “compassionate”—and one “masculine” trait—“provides strong leadership.”8 Following the standard wording of the National Election Studies presidential trait batteries, respondents were told that they were going to be asked about several characteristics of the candidates. They were then asked a question in this way: “Think about Amy Klobuchar, the Democratic candidate for U.S. senator. In your opinion, does the phrase ‘provides strong leadership’ describe Amy Klobuchar extremely well, quite well, not too well, or not well at all?” Responses were assigned a numerical value, with 1 representing “not well at all” and 4 representing “extremely well.” Respondents were asked to rate both Republican and Democratic candidates on each trait.9
The CCES sample does not constitute a random draw within each state, but the large number of contested Senate elections allow for an examination of trait assessment under various combinations of candidate party and gender. Table 1 displays the four different types of contests, the states that fall into each category, and the number of respondents in the data set. Not surprisingly, voters in a majority of states faced a choice between two male candidates. But 10 campaigns in 2006 featured at least one female candidate, and the CCES includes a large number of voters from these states. By aggregating across these races, it is possible to determine whether systematic patterns exist in the way voters attribute traits to candidates of different sex and party.10
Distribution of candidate gender and party combinations in U.S. Senate campaigns, 2006 Cooperative Congressional Election Study
Note: N reflects the number of respondents who rated both candidates on at least one trait. Respondents who did not rate either candidate on any trait are excluded.
The results will no doubt be influenced by the electoral environment of the 2006 campaign. Frustrated with the direction of the country, Americans handed the Republican Party a significant defeat, turning Congress over to the Democrats. As a result, assessments of Republican candidates were generally negative. Thus, it may be that the disadvantages faced by the GOP could limit the applicability of my findings to non-“wave” elections, a possibility I address later. Nonetheless, it is my judgment that the benefits of using the CCES data—the only data set that contains the appropriate measures for investigating how voters perceive the personal attributes of Senate candidates—outweigh the concerns associated with relying on a single election year.
If the deployment of stereotypes depends in part on the media environment, it is necessary to determine whether news coverage of these campaigns portrayed candidates in gender-stereotypic terms. A research assistant conducted a content analysis of news coverage from the largest newspaper in each of the 30 Senate states during the last month (Oct. 7–Nov. 7) of the campaign. Using the Nexis and Newsbank databases, we coded every article that mentioned at least one of the two Senate candidates. The articles were coded for a variety of attributes, including references to candidate traits.11 In all, the content analysis incorporates 1,345 stories, an average of 45 per state. The fewest stories (5) were published about the Wisconsin race, and the most (154) were published in the Virginia contest.
To determine whether news coverage systematically depicts candidates in gender-stereotypic ways, as previous research has found (e.g., Kahn 1996), the articles were coded to identify which traits a candidate was portrayed as possessing. I followed the procedures of Kahn and Kenney (1999), coding only for “manifest” trait mentions—that is, clear references to a candidate as possessing particular traits, rather than coding traits based on vague inferences that might be made by readers.12 The specific measure is the number of paragraphs in which trait mentions appeared. I then created categories for “feminine” and “masculine” traits, using the categories of the Personality Attributes Questionnaire (Helmreich, Spence, and Wilhelm 1981), a long-established scheme for identifying stereotypical “male” and “female” characteristics.13 I then summed the relevant traits together, which allowed me to determine whether male and female candidates are being systematically portrayed differently in the news.
The results are presented in Table 2. Looking first at the percentage of paragraphs referring to the candidates as possessing feminine traits, there are no differences in the portrayal of male and female candidates. Neither the comparison of Republican men and women nor Democratic men and women yields significant differences. In fact, male Republicans were portrayed in slightly (though insignificantly) more feminine terms than their women counterparts. It is difficult to draw strong inferences from the table, however, because of the small number of trait paragraphs about Republican women. Thus, the conclusions I draw in this study about the effect of news coverage on voter attitudes toward GOP women will remain necessarily tentative.
Portrayal of candidate traits in newspaper campaign coverage, 2006 U.S. Senate elections
Note: Cell entries represent the percentage of paragraphs referring to different traits out of all trait paragraphs. Raw Ns of paragraphs are in parentheses.
The same null findings are apparent for masculine traits. There was no significant difference in the portrayal of female and male candidates. In fact, contrary to previous findings about media stereotyping (Kahn 1996, e.g.), the four Republican women were actually described in more masculine terms than were their male counterparts. Thirty-nine percent of trait paragraphs about GOP women described them in masculine terms, compared to 27% of trait paragraphs about male Republicans. The same is true for Democrats, with 34% of trait paragraphs describing women in masculine terms and 27% for men. Neither of these differences, however, is statistically significant. In sum, by this measure, the media did not portray male and female candidates according to their gender stereotypic roles. Consistent with Fowler and Lawless's (2009) analysis of newspaper coverage of gubernatorial candidates, sex did not shape portrayals of candidate traits. This suggests that the media in 2006 may not have promoted gender stereotyping.
I turn now to an examination of the trait assessments of voters.
To determine whether candidate sex or party affiliation has an influence on candidate trait perceptions, I first present the average trait ratings for candidates of both parties. I do this in two ways. First, I examine comparisons between male and female candidates within each party. If gender stereotyping is evident, we should see differences between male and female Democrats, and male and female Republicans. I then compare perceptions of same-gender candidates of each party to determine whether party affiliation reveals differential trait ratings.
In the top panel of Figure 1, the average trait ratings for Republican female (lighter bar) and male (darker bar) candidates are presented. Following Bartels (2002a), I have transformed the ratings from the survey questions into a 0–100 scale to ease the interpretation of the data.14 If gender stereotyping is occurring, we would expect male Republicans to hold a significant advantage on “strong leader,” and female GOP candidates to possess advantages on “really cares” and “compassionate.”
Average trait ratings for male and female U.S. Senate candidates, by party, 2006 Cooperative Congressional Election Study. None of the differences between ratings of male and female candidates is statistically significant. The number of respondents for each trait comparison ranges from 463 to 466.
There is, however, no evidence of sex stereotyping. Female Republicans were rated as more empathetic than their male counterparts, but the difference is negligible and statistically insignificant (39.0 to 38.5). On the other “feminine” trait, compassionate, and on leadership, male GOP candidates were perceived more favorably, though insignificantly so (46.0 to 44.3 on compassion, 49.6 to 48.1 on leadership).
The bottom panel, which displays the comparisons for Democratic men and women, tells the same story. Once again, the data reveal no evidence of sex stereotyping. Female Democrats are perceived slightly more favorably on all three traits, but there are no statistically significant differences. In fact, the largest advantage for women emerges on leadership (53.6 to 49.5, p = .14), the ostensibly “masculine” trait. The differences on the feminine traits are small: 56.9 to 55.2 on compassionate, and 47.2 to 45.7 on really cares.
Figure 2 investigates whether trait perceptions are affected by Senate candidates' party affiliation. Evidence of partisan “trait ownership” at the presidential level (Hayes 2005) suggests that Democrats should be perceived as more compassionate and empathetic, whereas Republicans should hold an edge on leadership. The top panel of the figure compares perceptions of male candidates in each party. Just as Figure 1 controlled for party, this analysis controls for gender.
Average trait ratings for Republican and Democratic U.S. Senate candidates, by gender, 2006 Cooperative Congressional Election Study. Asterisks denote difference between candidate ratings as statistically significant, p < .05. The number of respondents for each trait comparison ranges from 463 to 466.
The results reveal that party affiliation is significantly related to trait attribution. Democratic men are perceived as more compassionate (55.2 to 46.0, p < .05) and empathetic (45.7 to 38.5, p < .05) than their GOP counterparts. Republicans hold a small and statistically insignificant advantage on leadership, 49.6 to 49.5.
The bottom panel presents the same comparisons for female candidates. Once again, there are substantial and significant differences in the attribution of feminine traits. Democratic women are rated as far more compassionate (56.9 to 44.3, p < .05) and empathetic (47.2 to 39.0, p < .05) than Republican women. Female Democrats also hold an advantage on leadership, though the difference (53.6 to 48.1) is not significant and, notably, the smallest among the three traits. There is clearly a baseline advantage in 2006 for Democratic Senate candidates, regardless of sex, but the only traits on which those advantages are significant are the feminine/Democratic measures.
Of course, it is likely that the anti-Republican sentiment in the country at the time of the 2006 midterms affected voters' ratings of the GOP candidates' traits. Displeasure with the Republican Congress, George W. Bush, and the war in Iraq probably permeated citizens' judgments about all aspects of the party's candidates (Jacobson 2007), including their personality traits. This may have led to unusually low ratings on the leadership measure in particular. Given the low baseline of Republican support, that the GOP did not “lose” these trait comparisons to the Democrats, even in the face of such public discontent, is telling. Whether this is evidence for “ownership” is not clear, but one interpretation is that the durability of party reputations influences citizens' responses to questions about candidate traits, even in a party's gloomiest days.
The descriptive data can only take us so far. It is possible that other factors could account for the patterns of trait evaluations. Incumbency and individual-level characteristics, such as party affiliation and ideology among others, seem plausible suspects. And it may also be that the portrayal of candidates in the news influences the extent to which voters perceive them in stereotypic ways.
To address these possibilities, Table 3 presents the results of ordered logit models predicting a respondent's trait score for each candidate on each trait. The dependent variable is a respondent's rating of a candidate on each trait, ranging from “not well at all” (low) to “extremely well” (high). I specify separate models for Republican and Democratic candidates for each of the three character traits. The models for Republicans appear in columns two through four, the Democratic models in columns five through seven. These models test whether candidate gender exerts an independent influence on trait evaluations and whether the influence of candidate gender is conditioned by the candidate's portrayal in the news.
Model of Senate candidate trait attribution, 2006 Cooperative Congressional Election Study
Notes: **p < .01; *p < .05
Dependent variable is the respondent's assessment of whether the trait described the candidate “not well at all,” “not well,” “quite well,” or “extremely well.” Cell entries are ordered logit coefficients. Robust standard errors are in parentheses.
The model includes a dummy variable coded 1 if the candidate is a woman. Its effect indicates whether men and women are evaluated in systematically different ways. A second dummy variable indicates whether the candidate is an incumbent, given the evidence that sitting senators are typically evaluated more favorably than challengers or open-seat candidates (Fridkin and Kenney 2011; Hayes 2010). I also include variables for the number of paragraphs in news coverage referring to a candidate as possessing a feminine or masculine trait. (To account for differences in the amount of coverage from race to race, the model includes a control for the total number of paragraphs published about the campaign.)
The model also includes interactions between candidate gender and trait coverage. If gender-stereotypic coverage activates gender stereotyping in trait assessment, then the coefficients for those variables should show significant effects. Specifically, we would expect gender-stereotypic coverage—coverage that portrays female candidates as possessing feminine traits—to suppress the evaluations of women on the trait of leadership. Given previous research (King and Matland 2003), it may be that female Republicans are substantially more disadvantaged by news coverage than female Democrats, relative to their male copartisans. At the same time, stereotypical portrayals of female candidates should increase their scores on compassionate and caring.
The model also includes a measure of the Republican candidate's advantage (or deficit) in campaign spending.15 Ceteris paribus, candidates with large spending advantages would be expected to be evaluated more favorably than their opponents. Finally, I include variables for a respondent's party identification and sex, as well as a 5-point ideology scale—with higher values indicating greater Republican identification and conservatism.16
If voters are assigning candidate traits on the basis of gender, we would expect female candidates to receive lower scores on leadership and higher scores on compassionate and caring. But that is not what the models reveal. Instead, in 2006, the only evidence of a gender effect is that Republican women were actually perceived as less compassionate than their male counterparts, in direct contradiction of the gender-stereotyping hypothesis. At the same time, there was no effect on the trait of leadership or caring. Among Democrats, men and women were not evaluated in systematically different ways. In short, there is no evidence of direct gender stereotyping.
There is minimal evidence that news coverage promoted gender stereotyping. The significant coefficient in the “Female X feminine trait coverage” variable for leadership for Republicans indicates that the more often female candidates were portrayed in the news as possessing feminine traits, the lower were their evaluations on the trait of leadership, relative to male Republicans. In other words, gender-stereotypic portrayals in the news prompted voters to evaluate female Republicans less favorably. I am hesitant to make much of this result, however, since there are only four races with GOP women in 2006 and little coverage that portrayed them in feminine terms. It is possible that these results reflect nothing more than idiosyncratic circumstances of those elections.
That is the only case in which the media play any role in promoting gender stereotyping. The remaining interactions with feminine coverage for Republicans and Democrats are insignificant. And the interactions with masculine coverage are significant only for Republicans on the traits of compassionate and caring. The positive signs mean that the more that Republican women were portrayed in masculine terms—strong, tough, aggressive—the more favorably they were evaluated on compassionate and cares relative to male Republicans. This suggests an overall positivity effect for candidates portrayed in masculine terms; that is, trait assessments are less the product of gender-stereotypic portrayals and more the product of being shown as possessing toughness and assertiveness, traits that voters value in politicians, regardless of their gender. This suggests that being portrayed as masculine serves to benefit female Republican candidates because voters give them higher evaluations, even on female trait dimensions. But taken as a whole, there appears to be little evidence that candidate gender affects trait attributions, and the effect of news coverage is inconsistent.17
These null results, however, do not demonstrate that voters' judgments are driven by party stereotypes. That conclusion is only suggested by the aggregate-level patterns in Figure 2. To determine whether candidate party is driving trait assessments at the individual level, we need to see that, controlling for relevant factors, the Democratic candidate advantage on Democratic traits remains higher than the Democratic advantage on the Republican trait, even when we account for candidate gender and other factors.
One way to test for partisan stereotyping is to examine the differences in trait assessments between two candidates running against each other. In the model presented in Table 4, the dependent variable is the difference in the assessments between two candidates' traits. To create that measure, I subtract each respondent's Democratic candidate's trait score from the Republican's trait score. The variable ranges from –3 to +3. Positive scores indicate a GOP advantage.
Model of trait attribution differential between Republican and Democratic Senate candidates, 2006 Cooperative Congressional Election Study
Notes: **p < .01; *p < .05
Dependent variable is the difference in a respondent's assessment of the Republican and Democratic candidates' traits. Higher scores indicate a Republican advantage. Cell entries are unstandardized OLS regression coefficients. Robust standard errors are in parentheses.
The key independent variables are dummies for the mixed-gender races—a Republican woman versus a Democratic man and a Republican man versus a Democratic woman.18 If the differences in assessments of two candidates of the same gender—a male Republican versus a male Democrat and a female Republican versus a female Democrat—are no different than in the mixed-gender races, then the coefficients on those dummy variables should be indistinguishable from zero. If so, that is evidence that variations in candidate gender do not explain variation in the trait scores. At the same time, we would expect the intercepts in the models for the traits of compassionate and cares to show larger advantages for Democratic candidates than in the leadership model. The combination of larger advantages for Democrats on Democratic traits and insignificant effects for the mixed-gender races would provide evidence that voters' assessments are the product of party stereotyping.
The results of these models are shown in columns two through four of Table 4.19 The intercepts for all the models are negative, confirming the results from Figure 2 that show a Democratic advantage in 2006 on all traits. Critically, the advantages for the Democratic traits—compassionate and cares—are larger than for the Republican trait of leadership (–2.81 and –2.94 compared to –2.29). In other words, Democrats held more substantial advantages on their “owned” traits, controlling for other factors.
The dummies for the mixed-gender races are all insignificant, which means that the differences in assessments of Democrats and Republicans are unaffected by the gender of the candidates. The differences in same-gender races are no different from the differences in the mixed-gender races. Thus, gender does not explain any variation. Instead, as evidenced by the intercepts, party does.
The last three columns in Table 4 run the same models, with the addition of the content of campaign news coverage. The news measures are, again, the number of paragraphs portraying the candidates as representative of male and female traits. I do not use differenced variables because doing so would obscure the source of any effects—that is, whether any effects are the product of stereotypic or counterstereotypic portrayals.
Media coverage matters. But perhaps surprisingly, news matters most when it challenges the public's default perceptions of the parties. Looking first at the leadership model, the more that Republicans are portrayed as feminine—in contrast to the traditional party image—the larger the GOP candidate's leadership advantage is. The same is true for the compassionate model and narrowly insignificant (p = 0.10) for caring. The more a GOP candidate is portrayed as feminine, the larger the advantage over the Democratic opponent. For instance, for every five additional paragraphs mentioning a Republican candidate's feminine qualities, there is a 0.6-point shift in the comparative compassionate trait assessment. This is not an enormous effect—the standard deviation for the variable, which runs from –3 to +3, is 1.5—but is large enough to suggest that media coverage has a substantively meaningful influence on public opinion. Obviously, larger changes in trait coverage would lead to larger shifts, but the number of trait mentions in the Senate coverage tends to be small.
The same phenomenon is evident for evaluations of Democrats. Democratic advantages increase when Democrats are covered in the news as strong, tough, assertive, and so forth—as playing against type. On each trait, such portrayals provide an advantage to Democratic candidates, as evidenced by the negative sign on the coefficients.20 In those models, the effect of an increase of five masculine paragraphs in the news leads to shifts in the comparative trait assessments of between 0.25 and 0.3, slightly smaller than the effects for Republicans.
At first glance, these patterns are counterintuitive: Why would counterstereotypic portrayals benefit candidates, especially when the news portrayals sometimes seem to be directly at odds with the traits? Why would portraying Democrats as tough cause voters to perceive them as more warm? Why would Republicans be seen as stronger leaders if reporters depict them as more caring?
The findings dovetail with my previous work about the impact on vote choice when candidates successfully “trespass” on their opponent's trait territory (Hayes 2005). The largest boosts in vote share from trait advantages come when candidates erode their opponent's advantages on partisan traits. Thus, Republicans do best when they overcome their expected disadvantage on compassion and caring, whereas Democrats benefit when they erode the GOP advantage on leadership. This result occurs because voters expect each party's nominees to possess specific, and different, personal qualities. A Democrat who establishes himself as an unusually strong leader, or a Republican who comes across as especially empathetic, for example, might win over voters looking for information to distinguish between the two candidates.
The effects of the news content variables find something similar in this new context: Voters are more likely to perceive the candidates favorably—regardless of the particular trait—when they are portrayed in ways contrary to the partisan stereotype. Voters have strong expectations about the traits that Republicans and Democrats should exhibit, and they reward candidates when they “overperform.” The benefits for Republican and Democratic candidates when the news portrays them in counterstereotypic ways are derived from the psychological power of differences between what voters expect and what they see (Aronson and Linder 1965; Bem and McConnell 1970). Thus, party reputations matter quite a bit. Gender, in this case, explains little.
Stereotypes are important in politics because they often influence, for good or for ill, voters' assessments of policies, parties, and candidates. For scholars interested in political behavior, understanding and analyzing the content and influence of stereotypes is critical, providing insight into the ways that citizens evaluate candidates, as well as the consequences for electoral outcomes and public opinion.
This study has suggested that voters are likely to use party stereotypes in making inferences about candidate traits, but that gender stereotypes are not as influential. Furthermore, the content of campaign news coverage can influence the favorability of trait assessments when it portrays candidates in ways that diverge from the default stereotypes of the Republican and Democratic Parties. Despite the literature showing that people perceive women as more compassionate and empathetic than men, and men as stronger leaders than women, I did not find this to be true in assessments of political candidates. At least in the 2006 U.S. Senate elections, party heuristics were more influential when voters judged politicians' personal attributes.
In interpreting the results, it is important to acknowledge that the findings might simply show that women have successfully neutralized sex stereotypes by conducting campaigns that lead voters to evaluate them in nongendered ways. Just as female candidates often focus their campaigns on the same issues as their male opponents (Dabelko and Herrnson 1997; Dolan 2005; Larson 2001; Williams 1998), women emphasize the personality traits they believe are most beneficial given the electoral context (Herrnson, Lay, and Stokes 2003; Huddy and Terkildsen 1993; Kahn 1996; Williams 1994), which may mean demonstrating their “masculine” side. In other words, because female candidates know that social stereotypes might put them at an electoral disadvantage, they undertake campaign strategies to present themselves to the public in counterstereotypical ways. If, in 2006, female candidates saw it as advantageous to downplay feminine traits in favor of masculine ones, then the data presented here may simply reveal that women successfully neutralized gender stereotypes, not necessarily that party stereotypes are more accessible as the basis for judgment, generally speaking.
One way this study addresses this alternative explanation is the inclusion of the media content variables. Media content is certainly affected by the nature of candidates' campaigns (Kahn 1996; Kahn and Kenney 1999). But given that the news media have traditionally been viewed as facilitating gender stereotyping, the fact that I have incorporated news coverage into the analysis and still fail to find evidence of gender stereotyping in voters' evaluations is important. News coverage matters, but it tended not to promote stereotyping; instead, it tended to benefit candidates when they were covered in ways that challenged partisan stereotypes.
The findings have both empirical and theoretical implications. Empirically, the results underscore the point made by other scholars that the process of candidate evaluation is complicated, and that candidate gender and party affiliation can raise a fairly complex set of considerations for voters (Dolan 2004; Winter 2010). If we want to understand how long-standing gender and party images influence political judgment, we need to study both simultaneously to determine which stereotypes are influential, under what circumstances. At the same time, the effect of the news coverage in my analysis suggests that moving traditional studies of stereotyping outside of the experimental contexts can be helpful, as doing so may lead to different conclusions than what laboratory studies have found. Future research on other election cycles and at other levels of office would also provide a better sense of whether these results are peculiar to 2006—a wave election that swept the Democratic Party back into power—or to the specific profiles of the 12 women who ran for Senate that year. This will require analysts to include questions about candidates' personal qualities on national surveys, something that has not been a regular part of congressional elections research, but should be.
Theoretically, thinking about variation in the accessibility of stereotypes may have implications beyond the study of women and politics. It may also be useful in identifying the circumstances under which candidate race or ethnicity is likely to serve as the basis for voter judgments. When campaigns are framed in terms of racial issues, race may be a salient and accessible stereotype by which voters come to judge candidates. If, on the other hand, the campaign environment makes the political party the salient cleavage, then a candidate's physical attributes may be de-emphasized in favor of party stereotypes.
Given the contemporary American political landscape, one broad implication is that the growing polarization of the Republican and Democratic parties will encourage voters to evaluate candidates as “partisan creatures,” as King and Matland (2003) put it. This would seem to be a favorable development; most observers would regard stereotyping based on party history and performance as preferable to stereotyping based on often inaccurate notions about men and women, whites and blacks. To the extent that polarization continues, I would expect party stereotypes to become increasingly powerful in shaping voter attitudes. That may be a good thing.
1. My argument acknowledges that in circumstances where the party cue is absent or irrelevant, gender stereotyping may indeed be quite powerful. For instance, in the 2008 Democratic presidential primary, gender surely played a role in shaping Democratic voters' attitudes toward Hillary Clinton and her male competitors. The focus here is on general elections that pit partisans against one another.
2. Similarly, Huddy and Capelos (2002) found that gender influenced voters' perceptions of a candidate's issue-handling ability only on “women's issues,” a domain in which gender is explicitly invoked.
3. To be sure, one of the difficulties of testing for gender stereotyping during actual campaigns is that female candidates often undertake strategies to preempt sex stereotyping. But accounting for female candidate portrayals in the news provides a way to capture the effect of counterstereotypic campaign activity. I address this further in the Conclusion.
4. My expectations are all of the ceteris paribus variety, in that candidate trait ratings will also be affected by an individual voter's predispositions, especially party identification (Bartels 2002b). As such, I control for partisanship and ideology in the analyses that follow. But my primary aim in this article is to determine whether there is a general tendency for candidates to be stereotyped on the basis of their sex, party, or both. Thus, my focus is primarily on the overall patterns, and only secondarily on the specific individual-level sources of variation.
5. It is of course possible to explore the interaction of party and gender stereotypes experimentally, as Huddy and Capelos (2002), for instance, have done. But there are numerous advantages to using an observational approach. First, the study removes the artificiality of a laboratory experiment in which sequestered subjects are forced to evaluate candidates in an unnatural setting. Second, using assessments of actual candidates rather than hypothetical ones improves the generalizability of the findings. Third, the observational approach allows me to measure news coverage as it occurs and test its effects in the “real world.” In an experiment, I would be able to test the effects of exposure to just a limited subset of a campaign's information environment.
6. More information about CCES and Polimetrix, Inc., the firm that conducted the Internet-based survey, is available at http://web.mit.edu/polisci/portl/cces/index.html. One postsurvey analysis shows that the CCES sample was slightly better informed, slightly younger, and slightly more educated than respondents in the 2004 National Election Study, but that these differences are not large (Hill et al. 2007).
7. This article does not analyze data from three states with unusual circumstances in their Senate races. In Indiana, Republican Richard Lugar faced no Democratic opposition. In Connecticut, Democrat Joe Lieberman won reelection as an independent, beating out a Republican and Democratic opponent. And in Vermont, Independent Bernie Sanders won election in a contest without an official Democratic candidate.
8. Respondents were also asked whether the candidate was “moral.” Those responses are not central to the research questions here, and so they are not included in the analysis. To be sure, a large battery of traits would have allowed for a more robust test of the gender stereotyping argument. Traits like “tough,” “assertive,” or “kind” might reveal other assessments indicative of sex stereotyping. But the limitation of the study to the three traits was necessary in order to pit the party and gender stereotypes against one another. My “trait ownership” argument ties the Republican and Democratic advantages to issues on which the parties are perceived to have an issue-handling edge (Hayes 2005). Other traits, such as “trustworthy,” do not map to issue advantages, and thus would not be expected to exhibit partisan patterns. And though it may be that gender stereotyping is more prevalent on traits not “owned” by one of the parties, the three traits used here are among the most central to candidate evaluation and vote choice. For example, recent studies of Senate candidate evaluation find empathy and leadership to be the traits most strongly related to vote choice (Fridkin and Kenney 2011; Hayes 2010).
9. The order in which the trait batteries were asked and the order of the candidates were randomized.
10. This is not to say that the distribution of male and female candidates is ideal. Clearly, the conclusions in the study would be more generalizable if data from 2006 could be combined with other election cycles. This combination would yield more female candidates and more variation in their profiles and the electoral environments in which they compete. It would also result in more respondents in states with two female candidates; the CCES data have only 64 respondents who faced such a choice in 2006. But absent appropriate data on trait perceptions from earlier elections—which do not exist—it seems reasonable to proceed and to encourage more research that can subject the present findings to further scrutiny.
11. Although I do not present these data, I also examined the percentage of paragraphs in each state's coverage that explicitly mentioned the candidates' party affiliation or gender. (Explicit mentions of party included references to the candidates as being members of a particular party, and explicit mentions of gender included references to candidates as a “female candidate” or other references that were gender specific.) By this measure, party clearly is more prominent than gender. That, of course, is hardly surprising. Most of the races involved two male candidates, rendering gender an irrelevant storyline for journalists. Predictably, races that involved female candidates were more likely to have gender mentioned than those involving males only. But even in those contests, gender was explicitly referenced only infrequently. For example, several of the contests with female candidates (Hawaii, Maine, Michigan, and Texas) contained no gender mentions. To be sure, gender works its way into campaign coverage more subtly than this measure can capture, and I am not suggesting that explicit mentions of gender are the only ways in which candidate gender is made accessible to voters. But the patterns are instructive in illustrating how news coverage encourages voters to think about politics in partisan terms.
12. In all, the coding included 78 different traits. A complete list is available from the author on request.
13. The “feminine” traits were compassionate, caring, patient, generous, respectful, weak, weak leader, passive, indecisive, incompetent, and unintelligent. The “masculine” traits were strong leader, tough, ambitious, decisive, independent, intelligent, competent, determined, confident, energetic, engaged, power hungry, not compassionate, uncaring, and stubborn.
14. Specifically, a score of 1 (“not well at all”) was converted to 0; 2 (“not well”) was converted to 33.3; 3 (“quite well”) was converted to 66.7; and 4 (“extremely well”) was converted to 100. The conversion of the scale has no effect on the statistical relationships among the ratings; it simply makes the scores easier to interpret.
15. Spending data were obtained from the New York Times 2006 Election Guide. The variable, which is scaled to $10,000s, takes negative values if the Democratic candidate outspent the Republican, positive values if the Republican held a spending advantage.
16. The models are estimated with Huber-White robust standard errors clustered by the respondent's state. This technique relaxes the assumption of the independence of observations, and accounts for the possibility of “intraclass” correlation in the trait assessments among individuals in the same Senate contest. In addition to the models shown in Table 3, I I have also run a series of other models in which I control for other individual-level characteristics, such as political sophistication and a battery of issue positions. Those alternative specifications yield the same conclusions as those presented. The results are available from the author on request.
17. I have also specified models with interactions between candidate gender and incumbency, candidate gender and respondent gender, and candidate gender and political sophistication. None of these models reveals statistically significant relationships.
18. In practice, that means the Republican woman vs. Democratic man dummy represents Maine and Hawaii. The Republican man vs. Democratic woman dummy represents the six states shown in Table 1.
19. The results in Table 4 are from an ordinary least squares regression. Because the dependent variable is an ordered scale, an ordered logit model is technically the more appropriate statistical technique. I have run the model as an ordered logit and found no effects for the mixed-gender races. However, there is no straightforward way to interpret the intercept shifts across the three traits. Thus, since the OLS model produces identical results while also allowing for a substantive interpretation of the constant, I present those models here. Results of the ordered logit models are available from the author on request.
20. I have also interacted the media variables with the dummies for the mixed-gender races. Two issues complicate the interpretation of those results, however. First, there is significant collinearity, as often happens when multiple interactions are included in regression models. Second, because there are just two states with a Republican woman running against a Democratic man, the estimates for those coefficients are highly unstable. With those limitations in mind, the results do not reveal any significant interactive effects. The counter-stereotypical portrayals appear to benefit men and women of both parties similarly.