Party and Gender Stereotypes in Campaign Attacks
Research on negative campaigning has largely overlooked the role of stereotypes. In this study, we argue that the gender and partisan stereotypes associated with traits and policy issues interact with a candidate’s gender and partisanship to shape the effectiveness of campaign attacks. We draw on expectancy-violation theory to argue that candidates may be evaluated more harshly when attacks suggest the candidate has violated stereotypic assumptions about their group. Thus, attacks on a candidate’s “home turf,” or those traits or issues traditionally associated with their party or gender, may be more effective in reducing support for the attacked candidate. We use two survey experiments to examine the effects of stereotype-based attacks—a Trait Attack Study and an Issue Attack Study. The results suggest that female candidates are particularly vulnerable to trait based attacks that challenge stereotypically feminine strengths. Both male and female candidates proved vulnerable to attacks on policy issues stereotypically associated with their party and gender, but the negative effects of all forms of stereotype-based attacks were especially large for democratic women. Our results offer new insights into the use of stereotypes in negative campaigning and their consequences for the electoral fortunes of political candidates.
KeywordsNegative campaigning Stereotypes Traits Issue ownership Vote choice Gender Partisanship
In an interview in the spring of 2016, Donald Trump argued that U.S. foreign policy involves dealing with “very, very tough people.” He went on to contend that his opponents do not have what it takes to address security challenges facing the nation: “Somebody like Jeb…and by the way, Hillary is another one. I mean, Hillary is a person who doesn’t have the strength or the stamina… She’s not a strong enough person to be president” (ABC News 2015). This line of attack continued through the election, including in the first debate, where Trump asserted that Clinton lacked the “stamina” to be president. Are these types of attacks successful? Are they more successful against a female candidate like Clinton than against a male candidate like Bush? Against a Democrat or a Republican? In this paper, we evaluate how the content of campaign attacks influences political decision making. We focus on how attacks using traits and issues that are stereotypically associated with political parties and gender interact with candidate characteristics to influence voter behavior.
Candidate evaluations relate to beliefs about the traits and issue strengths of political candidates both as individuals and as members of a group (e.g., their party or gender). As such, campaigns focus on candidates’ individual strengths, playing up traits or issue stances they hope will resonate with voters. A candidate’s party and gender constrains these efforts, with Democrats and women viewed as more credible on policy issues involving social welfare or traits like compassion, whereas Republican and male candidates enjoy greater credibility on national defense or assertiveness, for example (Petrocik 1996; Huddy and Terkildsen 1993). Thus, voters may expect that a candidate has certain traits or issue competencies based on their gender or party affiliation.
Candidates do not control all elements of their campaigns, despite strategic choices to affirm stereotypic strengths or to stress competencies in non-stereotypic areas (Banda and Windett 2016). Indeed, campaign attacks can influence voters’ perceptions of candidate traits and policy strengths or weaknesses (Fridkin and Kenney 2011). As campaigns become increasingly negative, understanding the effects of specific attacks on candidate evaluations is essential (Mattes and Redlawsk 2015). But, we know very little about how the content of campaign attacks interacts with party and gender stereotypes to influence voter decision making.
In this paper, we pose two key questions about the substance of campaign attacks. First, do campaign attacks that utilize gendered or partisan traits and issues ‘work’? That is, do these attacks reduce perceptions of candidate strengths and support for that candidate? And, second, are attacks on the traits and issues typically associated with a candidate’s party or gender more successful than attacks on counter-stereotypical traits and issues? We draw on expectancy-violation theory to argue attacks on a candidate’s ‘home turf,’ or those traits or issues stereotypically associated with their party or gender, may be more effective in reducing support than criticisms that are not about stereotypic strengths of the group (Jussim et al. 1987).
We evaluate the role of stereotypes and expectancy violation in negative campaigning with two experiments: a Trait Attack Study (Study 1) and an Issue Attack Study (Study 2). Both were conducted on Amazon’s MTurk and manipulated the type of attack made on a candidate as well as the party and gender of the targeted candidate. We find that female candidates (particularly Democrats) consistently face harsher punishment from voters when attacks focus on feminine traits. At the same time, we find little evidence of partisan expectancy violation, with Democrats and Republicans equally likely to be harmed by stereotypic or non-stereotype attacks.
As one of the first examinations of the gendered and partisan dimensions of campaign attacks, our results offer new insight into the ways candidate characteristics and an environment saturated with campaign negativity shapes political cognition and voter decision-making. The results point to a need to integrate ideas about candidate strategy with theories of stereotyping and group-based heuristics to understand how campaign attacks shape views of political candidates. Our research also has implications for descriptive representation by identifying challenges facing female candidates of both parties.
Negative Campaigns and Candidate Evaluations
Political campaigns increasingly utilize attacks and negative materials (Geer 2012; Mattes and Redlawsk 2015); for example, over 90% of the political ads aired by Clinton and Trump in the last two weeks before the 2016 election were negative (Wallace 2016). Negativity is also pervasive in races for the Senate, House, and lower level offices (Banda and Windett 2016). Research on negative ads includes a robust debate about the (de)mobilizing aspects of negative advertising and campaign attacks (Geer 2012; Krupnikov 2011) and a focus on how candidate gender shapes the overall effectiveness of campaign attacks (Krupnikov and Bauer 2014; Fridkin et al. 2009). Our research builds on this extensive negative campaigning scholarship by delving into how the substance of negative ads shapes voter decision making and whether particular types of substantive attacks are more effective when waged against particular types of candidates.
The conventional definition of negative campaigning considers any ad that mentions one’s opponent to be negative (Mattes and Redlawsk 2015). Yet, voters differentiate among types of negative ads and reliably see distinctions between ads that are merely negative and those that are uncivil (Fridkin and Kenney 2008). Mattes and Redlawsk (2015) argue that voters effectively parse which negative ads are critical but also informational; this negative information can be useful, given the reluctance of candidates to discuss their own flaws. We argue that if the content of campaign attacks is important—because it conveys information to voters (as Mattes and Redlawsk (2015) argue)—then exposure to these gendered attacks should decrease evaluations of the gendered traits and issue competencies attacked and also shape vote choice.1 This leads us to our Campaign Attacks Hypothesis: attacks on specific traits or issues should decrease perceptions that the attacked candidate has those specific traits or competency on those issues. And, attacks on traits or issues should decrease willingness to vote for the attacked candidate.
Political Parties and Trait and Issue Strengths
Information contained in campaign attacks may not affect evaluations of all candidates equally. Instead, attacks may interact with group-based information that voters have about a candidate, including assumptions about the candidate based on stereotypes associated with their party or gender. A candidate’s political party affiliation and gender may create expectations about candidate traits and issue priorities that are contradicted by the campaign attack content. We expect that the extent to which stereotypic expectations are challenged by campaign attacks has implications for their effects on candidate evaluations and vote choice.
Americans draw issue ownership distinctions between the parties, associating some issues consistently with each party—such as education and healthcare for the Democrats or law and order and foreign policy for the Republicans (Pope and Woon 2009). These issue ownerships (Petrocik 1996; Sapiro 1981) provide consistent cues to voters about who will be good at managing various policy areas (Funk 1999). The public also holds beliefs about the traits of individual leaders within each party: voters see Republicans as holding traditionally masculine traits like being stronger leaders, while Democrats are viewed as more compassionate and empathetic, which are feminine traits (Winter 2010; Hayes 2005).
Little is known about how an opponent’s attacks on issues or traits owned by a party might influence candidate evaluations. We expect that partisan stereotypes create expectations about candidates’ strengths and weaknesses in the minds of voters. Attacks can challenge these expectations, shaping voter evaluations of candidates and vote choice. Expectancy violation theory argues that people respond more strongly to individual information that runs counter to group-based expectations, including those conveyed in social norms and stereotypes (Jussim et al. 1987; Prentice and Carranza 2002). When a campaign attack provides evidence that a candidate has failed where there are group-based expectations of success, such as Republicans handling security or Democrats addressing education policy, the voter may punish that candidate more harshly than when the candidate is attacked for failures on counter-stereotypical traits or issue areas, where competence was not expected. Based on expectancy violation theory, our Party-Expectancy Hypothesis argues that that attacks on Democratic candidates will be more effective when they focus on failures around the issues and traits owned by the Democratic Party, such as education or interpersonal skills, than attacks that focus on shortcomings on issues and traits owned by the Republican Party, such as national security and assertiveness or strong leadership. Conversely, we expect Republican candidates will prove more susceptible to attack on traits and issues owned by the Republican Party compared to attacks emphasizing the home turf of the Democratic Party.
Gender and Trait and Issue Strengths
Party is not the only simple cue that voters receive about candidates running for political office. Voters also use candidate gender to form expectations about candidate traits and issue competencies that work as heuristics in voting (McDermott 1998; Mo 2014). These gender stereotypes emerge from general beliefs about gender roles in society (Diekman et al. 2002). Stereotypes may shape the degree to which voters believe a female or male candidate will be good at handling certain issues or possess certain traits; voters may perceive women as having better interpersonal skills and men as stronger leaders (Huddy and Terkildsen 1993; Sanbonmatsu and Dolan 2009). Similarly, voters assume women will be better at handling some issues, like education and child care, while men are perceived as more capable of handling issues like foreign affairs and national security (Holman et al. 2016).
We argue that expectancy violation theory applies to gendered expectations as well. Political leadership tends to be associated with masculine tasks and characteristics (Eagly and Carli 2007; Huddy and Terkildsen 1993). As a result, female candidates can present a cognitive challenge to voters, as female leaders represent “largely divergent expectations about leaders and women,” while male leaders represent “redundant expectations” (Eagly and Karau 2002, p. 575). Research shows voters sometimes support female candidates more when they play to their strengths by highlighting feminine traits and policy priorities in their campaigns (Dittmar 2015) and may punish women when they present themselves as possessing masculine traits (but see Brooks 2013; Schneider 2014). For example, observers often commented that Hillary Clinton fell into this trap during her 2008 Presidential bid: “Clinton’s toughness had fueled a spiraling negative campaign narrative: that she wasn’t quite a woman and lacked a woman’s empathetic instinct” (Kornblut 2009, p. 32; Carroll 2009).
Our Gender-Expectancy Hypothesis argues that voters will punish candidates who appear to have violated gender-role expectations and are attacked on gender-stereotype traits and policy strengths (e.g. a female candidate lacks interpersonal skills or mishandles education policy) more than when compared to attacks on counter-stereotypic traits and policy strengths (e.g. female candidates lack assertiveness or the ability to handle security policy). A competing possibility, however, is that women are punished for not being seen as tough enough, given the divergent expectations of women and political candidates more generally (Eagly and Carli 2007). In this circumstance, voters may punish female candidates who cannot “prove” they are masculine enough for political office—that they are leaders, not ladies (Brooks and Geer 2007). Our Divergent Expectations Hypothesis thus suggests that female candidates may be particularly harmed by attacks on their masculine traits or policy strengths.
Gender, Party, and Trait and Issue Strengths
Expectations about party and gender interact in informing voters’ assessment of candidates (Sanbonmatsu and Dolan 2009; Bos and Schneider 2015). As we have discussed, Americans associate competence on feminine policy issues and traits with the Democratic Party and masculine policy issues and traits with the Republican Party (Petrocik 1996; Sapiro 1981; Hayes 2005). The association of gendered traits and issues with party labels then creates “feminized Democrats” and “masculinized Republicans” (Winter 2010). The overlapping nature of gender and party expectations may also mean that an attack on a partisan trait or issue competency may have gendered consequences and vice versa. Given the degree to which these expectations overlap, we examine them in conjunction with each other, referring to attacks on feminine traits and issues as “feminine attacks” and attacks on masculine traits and issues as “masculine attacks” throughout.
Beliefs about candidates based on party and gender may produce unique evaluations when gender and party intersect. For example, voters often stereotype Republican men as especially strong leaders given redundant expectations regarding their sex and party, while stereotypes of Democratic men and Republican women are more muddied, given divergent expectations associated with their sex and party (Holman et al. 2016). Our Gender × Party Attack Hypothesis argues that the effects of issue and trait attacks will likely vary based on a combination of a candidate’s party and gender, so that female Democrats will be most harmed by feminine attacks, while male Republicans will be least harmed. At the same time, male Republicans will be most harmed by masculine attacks, given the redundant expectations associated with male candidates and the Republican Party. The research provides significantly less guidance on expectations as to how these attacks might shape perceptions of male Democrats and female Republicans, given the divergent expectations associated with the gender and party. On one hand, research shows that party can be more impactful than gender in the application of gender stereotypes (Holman et al. 2016). This would suggest that masculine attacks would be more harmful to Republican women, whereas male Democrats could be more harmed by feminine attacks. On the other hand, if gender looms larger than party for candidates at these intersections, Republican women may be more disadvantaged by feminine trait and issue attacks while Democratic men suffer most from masculine trait and issue attacks.
Data and Methods
Experimental Design Overview
We conducted two experiments using Amazon’s Mechanical Turk to evaluate the effect of gendered campaign attacks on support for female and male candidates: A Trait Attack Study (Study 1) and an Issue Attack Study (Study 2). Modeled after a wide set of campaign and voting experiments (Krupnikov and Bauer 2014; Merolla and Zechmeister 2009), both experiments used a mock newspaper article to convey attacks on a hypothetical candidate. Given that campaign attacks receive a great deal of coverage from mainstream media, including newspapers (Mattes and Redlawsk 2015), a newspaper article about negativity in a campaign is an externally valid treatment.2 To further improve external validity, all the experimental treatments were developed from a set of real newspaper articles discussing attacks between candidates found via key word searches in newspaper archives.3 This approach provides the flexibility needed to manipulate the party and gender of the candidate and the substance of the attack while circumventing problems associated with using existing candidates or ads. The conditions were pre-tested for similarity and effectiveness.
In each of the two studies, we randomly assigned participants to read a mock newspaper article about a primary election contest for a state Senate seat. Research shows that female candidates tend to face a more competitive primary process, drawing larger fields from within their own party (Barnes et al. 2016; Lawless and Pearson 2008). Thus, evaluating these dynamics in primary races is important for understanding the obstacles women candidates face on their path to office. In addition, using a primary election contest means that both the attacked and attacking candidate come from the same party. Across our studies, we held the gender of the attacking candidate constant as male and only varied his partisanship, so that it always matched that of the attacked candidate, consistent with our primary contest setting. This increases the external validity of the treatment, given that well over 75% of all candidates in elections in the U.S. are male, and further reduced the need for additional subjects and experimental conditions.
In each of the articles provided to our survey respondents, we varied the gender and the party of the attacked candidate, allowing us to determine whether certain kinds of attacks are more effective against male or female candidates from each party. We also varied the type of attack waged against the candidate, including: no attack (the control condition), an attack on feminine or masculine traits (Study 1), and an attack on feminine or masculine issues (Study 2) to evaluate how qualitatively different attacks shape candidate evaluations and vote choice. The result is a Gender (2: Male, Female) × Party (2: Republican, Democrat) × Attack Type (3: None, Feminine, Masculine) experimental design for each study. The attacked candidate’s gender is conveyed via his or her first name (Patrick/Patricia)4 and gendered pronouns throughout the article; the partisanship of the candidates is repeated throughout the article.
Study 1: The Trait Attack Study
In the news articles used in Study 1, the attacking candidate accuses his opponent of lacking either feminine or masculine traits. The content of the feminine trait attack focused on how the attacked candidate (Patrick/Patricia Johnson) lacked traits that are traditionally associated with women’s strengths, such as being easy to work with and being helpful. The masculine trait attack treatment focused on masculine traits and claims regarding Johnson’s (the attacked candidate) inability to be assertive and a strong leader. There are a diverse set of potential traits that we could have utilized in our design that are gendered (Schneider 2014) and partisan (Hayes 2005). We selected gendered and partisan stereotypical traits that often appear in campaign attacks and news coverage of campaigns occurring frequently in the Wisconsin Ad Database from 2000 and 20045 and in newspaper articles about campaigns in this period.
Design and content of news stories
The attacked candidate has failed…
To demonstrate masculine traits like strong leadership and assertiveness
To demonstrate feminine traits like compassion and the ability to get along with others
To deliver on campaign promises about security policy
To deliver on campaign promises about education policy
Discussion of basic match-up between the candidates, with no attacks
Hepner campaign calls his new opponent ‘weak and ineffective’
Hepner campaign calls his new ‘opportunistic and divisive’
Hepner campaign says opponent ‘fails to deliver’ on security and safety
Hepner campaign says opponent ‘fails to make the grade’ on education
State senate race continues
“S/he just does not have the confidence to speak up and pursue his/her ideas in policy.”
“S/he just does not have interpersonal skills needed to work with other representatives toward our policy goals.”
“The bottom line is, s/he ran on a security and defense platform but did not make any policy changes in this area.”
“The bottom line is, s/he ran on an education platform but did not make any policy changes in this area.”
“Pundits already anticipate a tough fight leading up to the primary.”
“Johnson didn’t have “enough backbone or independence to get the job done.”
“Johnson “wasn’t a “team player” and refused to work with others to get the job done.”
“S/he might say s/he is a leader who cares about protecting our citizens, but s/he missed votes for major security funding bills several times in the last legislative session.”
“S/he might say s/he is a leader who cares about educating our children, but s/he missed votes for major education funding bills several times in the last legislative session.”
“Both candidates are equally matched in political experience.”
N per condition, by attacked candidate gender and party
Dem female: 77
Dem male: 67
Rep female: 73
Rep male: 76
Dem female: 74
Dem male: 75
Rep female: 80
Rep male: 66
Dem female: 68
Dem male: 63
Rep female: 67
Rep male: 69
Dem female: 70
Dem male: 71
Rep female: 68
Rep male: 67
Dem female: 70
Dem male: 70
Rep female: 76
Rep male: 72
Dem female: 70
Dem male: 73
Rep female: 68
Rep male: 67
Study 2: The Issue Attack Study
We modeled Study 2’s treatment after Study 1, replacing a focus on traits with policy competence. In the news articles used in Study 2, the attacking candidate accuses his opponent of lacking the ability to handle a stereotypically feminine (education) or masculine policy area (security). We again selected issues that have clear gender and party stereotype associations but are also reasonable in the context of a state legislative race. As was the case in the previous study, the attacked candidate’s gender is conveyed via his or her first name (Patrick/Patricia Johnson) and gendered pronouns throughout the article. We again held the gender of the attacking candidate, Michael Hepner, constant and only varied his partisanship, matching the candidates on party in a hypothetical primary race. The feminine attack focused on Johnson’s failures in education policies and broken campaign promises surrounding education reform, “The bottom line is, [he/she] ran on an education platform but did not make any policy changes in this area.” The masculine issue attack used similar claims about a failure to lead on security policy, accusing Johnson of working to “drain resources dedicated to public safety.” See Table 1 for treatment conditions and experimental design; the full text of the articles is available in the Online Appendix.
Balance and Manipulation Checks
We used participant demographics to observe balance in terms of respondent partisanship, ideology, gender, race, education, and income across all experimental conditions (see Tables A2 and A3 of the Online Appendix). As a manipulation and equivalency check for both studies, we asked the respondents “How critical would you say the tone of the article was?” Responses ranged from not critical at all (1) to extremely critical (5). On average, those in the treatment conditions reported significantly higher evaluations of the article’s critical tone relative to respondents in the control conditions (p < .001 for all comparisons in both studies). When comparing the attack conditions to each other, survey respondents reported statistically indistinguishable levels of criticism across the attack conditions which suggests equivalence in tone between the feminine and masculine types of attacks in each study. Additional data on this manipulation check is available in the Online Appendix, including that both Democratic and Republican respondents saw the treatments as equally critical and as more critical than the control.
We utilize trait evaluations, policy competence, and vote choice as our dependent variables, following other research in this area (Holman et al. 2016; Bauer 2017). Traits and issue strengths were measured using composite scales. In Study 1—the Trait Attack Study, we created two mean-centered variables from sets of candidate traits, corresponding to Feminine Traits (lenient vs. harsh, soft vs. hard, warm vs. cold, caring vs. distant, warm, feminine, sensitive, and cautious) and Masculine Traits (strong vs. weak, assertive, tough, masculine, active, and confident) (α > 0.80). These trait scales are common measures in extant scholarship (i.e., Rudman et al. 2001).
In Study 2—the Issue Attack Study, we asked respondents to rate how well the candidate would handle a variety of issues. We used exploratory factor analysis to identify a set of feminine issues (education, welfare, health, and child care) and a set of masculine issues (national security, crime, and tax reform) (Pope and Woon 2009). The resulting competency ratings were aggregated into mean-centered Feminine Policy Strengths and Masculine Policy Strengths scales (α > 0.77).6 Respondents in both studies also indicated how likely they were to vote for the attacked candidate on a scale from extremely unlikely (1) to extremely likely (7), which we use as a dependent measure of Vote.
We conducted Study 1 in March of 2015 and Study 2 in October of 2015 using Amazon.com’s Mechanical Turk (MTurk), which is an online marketplace where people can post tasks and pay individuals to complete those tasks. We restricted the sample to U.S. citizens who are 18 years of age or older. While MTurk’s respondents are not a representative sample of Americans, the population is more demographically and ideologically diverse than convenience samples like college students (Berinsky et al. 2012). Typically, MTurk samples over-represent men, liberals, and the highly educated compared to the general population (Paolacci et al. 2010); this is the case with our sample as well. Men were slightly over-represented in both samples, comprising 54% (Study 1) and 53% (Study 2). Both samples skewed in a liberal direction: of the Study 1 respondents, 50% are liberal, 21% moderate, and 28% conservative. For Study 2, 52% of respondents identified as liberal, 22% as moderate, 27% as conservative. Both samples were highly educated: 54% (Study 1) and 52% (Study 2) of participants reported holding at least a 4-year college degree. Table A1 in the Online Appendix has further information about the sample and benchmarking.
To evaluate the degree to which trait and issue based attacks are successful generally and also vary in their success based on the characteristics of the attacked candidate, we use regression models where Feminine and Masculine Traits, Feminine and Masculine Issue Competencies, and Vote Choice are the dependent variables, and the experimental conditions are independent variables, with independent analyses performed for each type of candidate (party, gender, party x gender).7 The full models are available in the Online Appendix. To ease in interpreting the results, we graphed the coefficients associated with the effect of the masculine and feminine attack in each model relative to the control condition. The graphs contain three key comparisons: (1) the effect of each attack as compared to the control group, which is measured by the distance from zero (indicated on each graph), (2) the effect of each attack by candidate type, which is measured by the difference between different types of candidates within the same treatment group (e.g. the effect of the feminine trait attack on the female Democrat v. female Republican), and (3) the effect of the two attack types, which is measured by the difference between the different treatment types across the same candidate type (e.g. effect of the feminine issue attack vs. masculine issue attack on the female Democrat). For the latter two comparisons, we conduct post hoc Wald tests and report significance in the text. It is worth noting that in some cases, the confidence intervals appear to overlap in the figures, yet the post hoc contrasts reported in the text are statistically significant at the p < .05 level. There is not a perfect correspondence between confidence interval overlap and the p-values for the estimated difference between two groups, with non-overlapping 95% confidence intervals reflecting a more conservative test than the standard p < .05 significance cut off, so the post hoc tests are preferred (Julious 2004). All post-estimation tests are two-tailed.
Does the Substance of Campaign Attacks Matter?
The right-side panel shows the results of Study 2 (Issue Attacks). Similarly, attacks on education reduced perceptions of the candidate’s ability to handle stereotypically feminine issues areas, including education, welfare, health, and childcare (p < .001). Attacks on the candidate’s security record decreased views of the candidate’s competency on stereotypically masculine issues including national security, crime, and tax reform (p < .001). Thus, respondents receive the specific messages conveyed by these attacks and judge candidates accordingly, consistent with Mattes and Redlawsk’s (2015) claims that campaign attacks provide important information in elections. The effects of the attacks are largely specific to the traits they emphasize. In Study 1 (Trait Attacks), the effect of the masculine trait attack on masculine traits is significantly larger than the effect of the feminine attack on masculine traits (p < .001, comparison of the hollow circles in the left figure), and the reverse is true for feminine traits, where the effect of the feminine attack is considerably larger than the effect of the masculine attack (p < .001, comparison of the solid circles in the left figure). The results are similar for the feminine attack in Study 2: the feminine issue attack had a much greater effect on evaluations of a candidate’s competence on stereotypically feminine policy issues, compared to the masculine issue attack (p < .001, comparison of the top and bottom solid circles in the right figure). However, both issue attacks depress views of masculine issue competencies and effect sizes are indistinguishable (comparison of the top and bottom hollow circles in the right figure).
We next look at the effect of the attacks on willingness to vote for the attacked candidate and find that all attacks reduce vote likelihood. In each case, the effect is significantly lower than the zero reference line (i.e., the control condition). In comparing the effectiveness of the two types of attacks (comparison of the top and bottom plus symbols), we find that the masculine attacks are generally more effective than feminine attacks (Study 1: p < .01; Study 2: p < .10).
Does the Substance of Campaign Attacks Interact with Candidate Characteristics?
Party and Attacks on Candidate Evaluations
Are attacks on traits and issues stereotypically associated with a candidate’s party more damaging to evaluations of candidates than attacks on strengths associated with the opposing party? To evaluate this question, we examine the effect of masculine attacks on Republican and Democrat candidates, as compared to the effects of feminine attacks. From our Party-Expectancy Hypothesis, we expected that feminine issue and trait attacks on Democratic candidates would be more effective than attacks that focus on the more stereotypically masculine issues and traits, and vice versa for Republicans. We find little evidence in favor of this hypothesis.
When directly comparing the relative effects of the trait attacks on the Republican versus Democratic candidate, we find that the effects of the feminine trait attacks on perceptions of the candidates’ feminine traits are more effective against the Democratic candidate (p < 0.07), while the masculine attack is equally effective in reducing perceptions of masculine traits for candidates from both parties. In terms of voting, the feminine trait attack harms the candidates of both parties equally; however, the Republican candidate faces more harm in terms of depressed vote likelihood from the feminine trait attack compared to the masculine trait attack (p < .01).
In Study 2 (Issue Attacks), we observe a similar pattern of results. Compared to the control condition (difference from zero in Fig. 2), both candidates experience lower feminine policy competence evaluations when facing the feminine issue attack (p < .001) and masculine policy competences when facing the masculine issue attack (p < .001). These attacks shape voting for the candidate—both the issue attacks significantly reduce the likelihood of voting for the Democrat and Republican candidate relative to the control (p < .001 for all candidate and attack types). Comparisons of issue attack effect sizes for Democrat and Republican candidates revealed few differences in terms of their effects on masculine and feminine policy competence ratings or vote likelihood. Overall, these results do not confirm our expectations.
Gender and Attacks on Candidate Evaluations
Are the feminine or masculine attacks more successful on female vs. male candidates? We find that the feminine trait attack harms the female candidate more than the male candidate, having a larger effect on feminine trait ratings (p < .01) and vote likelihood (p < .06). The masculine attack has a significant effect on the male and female candidate’s vote likelihood, but the difference between the effect on the male vs. the female candidate is not statistically significant. Thus, the Study 1 results suggest that female candidates are more susceptible than are male candidate to campaign attacks that focus on traits, but particularly those that focus on feminine traits.
In Study 2, the issue attacks produced fewer gender-specific effects overall. As in Study 1, each issue attack decreased perceptions of associated feminine and masculine policy strengths for all candidates relative to the control (p < .01 for all comparisons). When comparing the effects of each issue attack type on male versus female candidates, few differences emerge: willingness to vote for the female candidate is lower in the feminine issue attack than the masculine issue attack, suggesting female candidates are more susceptible to this kind of attack (Fig. 3, right panel) (p < 0.06), but both male and female candidates are harmed by both types of issue attacks.
We also are interested in evidence that may support our Divergent Expectations Hypothesis, where women may be particularly harmed by attacks on masculine traits and issues, given the incongruity between their gender and the general expectations of political office. We find little evidence of this—overall, the effect on masculine traits or issue competencies is similar for male and female candidates in the masculine attacks. Looking at the vote, we find that the feminine attack is more successful than the masculine attack in harming the vote for both the male and female candidate in both Study 1 (p < .05) and Study 2 (p < .06).
When comparing the candidate gender comparisons in Fig. 3 to the candidate party comparisons in Fig. 2, three interesting patterns emerge that speak to how attacks and candidate traits interact. First, trait attacks have more differential effects than issue attacks, so that party and gender shape the effectiveness of trait attacks far more than issue attacks. Second, evaluations of the candidate’s traits and issue competencies seem to operate in very similar ways across candidate types. Third, when we look at the effect of these trait attacks by candidate characteristics, female candidates are the most disadvantaged by both forms of the trait attacks. But how do gender and party interact to shape the effectiveness of these attacks?
Candidate Party and Gender and Stereotypic Attacks
Looking at vote likelihood (Fig. 4, left panel), we find that gender seems to matter more than party does in shaping the effects of the attack. Comparing the effects of the attacks on the vote relative to the control across candidate types, only the female Democrat is significantly harmed by the masculine trait attack; it is the only candidate type where the confidence interval does not span zero (p < .01). Further, the feminine trait attack harms both the female Democrat and (p < .01) the female Republican (p < .001) when compared to the control condition. The masculine trait attacks do not appreciably harm the male candidates relative to the control, while the feminine attack decreased vote likelihood among both male candidates (p < .10). Thus, we do not find support for our Party × Gender Expectancy Hypothesis, because gender is the primary driver of candidate evaluations and vote choice, not the interaction between gender and party. These findings contribute to our understanding of how evaluations of Republican women operate at the conjunction of party and gender.
The patterns differ when we look at the results of Study 2. Starting with the feminine issue attack, we find information consistent with our Party × Gender Expectancy Hypothesis. While the feminine issue attack harms feminine policy competency for all four candidate types relative to the control (p < .01), the female Democratic candidate is most harmed, in terms of substantive changes in evaluations of issue competency and willingness to vote for the candidate, by the feminine issue attack, while the male Republican is least harmed. The masculine issue attack depresses masculine policy competence for all candidate types (p < .01), except for female Democrats, who are not significantly harmed by the attack relative to the control. Both issue attacks harmed all candidate types in terms of vote likelihood relative to the control condition (p < .05), with similar effect sizes across the candidate types. When comparing the effects of the attacks to one another, the feminine issue attack is more effective than the masculine attack (p < 0.04), but only for the Democratic female candidate. Thus, the Party x Gender Expectancy Hypothesis holds, but only for the female Democrat.
As modern campaigns become increasingly saturated with negative advertisements and candidate attacks, we must understand not just the general effects of this environment, but how specific types of attacks influence perceptions of candidates and shape voter decision making. Our research disassembles the standard political attack and examines how attacks on traits and issue strengths stereotypically associated with a candidate’s party and gender influence voter preferences. The results of our two studies qualify the existing literature on gender stereotypes and political campaigns in an important way. While a significant body of research suggests that gender stereotypes have minimal effects in campaigns (e.g. Brooks 2013), we provide evidence that gender stereotypes can affect candidates, but do so in ways that are highly context dependent—varying as a function of an attacked candidates’ gender, party, and the substance of the attack itself. Such results are consistent with a growing body of literature that suggests gender stereotypes depend on context and other candidate characteristics, like party (e.g. Bauer 2017; Ditonto 2017; Holman et al. 2016).
In Study 1, we find that female candidates are particularly vulnerable to trait attacks. Trait evaluations and vote likelihood suffer when they face an attack on either a masculine or a feminine trait, though the effects are significantly larger for attacks on feminine traits. The picture is less clear for male candidates, who are less susceptible to both kinds of trait attacks. In Study 2, we find that female and male candidates are both vulnerable to gendered issue attacks, though again female candidates are more susceptible to attacks on a stereotypically female policy issue. These results lend support to our Gender-Expectancy Hypothesis, in that female candidates are more harmed by attacks that emphasize stereotypically feminine traits. However, they run counter to the expectation that women are especially vulnerable to perceived deficiencies in the traits and issues stereotypically associated with masculinity and leadership, captured in our Divergent Expectations Hypothesis. Violating expectations about femininity has a more pronounced effect for women.
We find more limited support for our Partisan-Expectancy Hypothesis. The female Democrat experiences a significantly reduced vote likelihood in each of the four attack scenarios, and the female Republican suffers the same fate in three of the four scenarios—she is not affected by the masculine trait attack. Men from both parties only experienced reduced vote likelihood when facing issue-based attacks. Thus, in terms of willingness to vote for a candidate in the face of gender-based attacks, it appears the Republican women’s partisanship protects them from masculine trait attacks, but not from feminine trait attacks, while the male candidates’ gender protects them from either set of trait attacks, regardless of party. This is consistent with our expectations about Republican women who, as we note above, face divergent expectations associated with their sex and party (e.g. Holman et al. 2016). Taken together, the party x gender results from Studies 1 and 2 suggest that reactions to these kinds of attacks may be more driven by gender than by party, but ultimately reflect integrated evaluations of a candidate’s gender and party. It may be that there is more flexibility in the expectations for behavior of partisans, as these are not expectations built on social roles.
These findings speak to a robust body of scholarship that suggests “Women… have to be ‘better’ than men in order to fare equally well” in political office (Lawless and Pearson 2008, 78). They also support the idea that female candidates face what Fulton (2012) calls a “valance” gap, where they must exhibit higher levels of non-policy related traits like integrity, competence, and leadership than their male counterparts to win the same vote share. Our results support this notion by demonstrating that women are particularly susceptible to trait-based attacks; thus, they may face a higher bar in their efforts to be viewed as capable (see also Ditonto 2017). These findings also have implications for campaign strategy. Female candidates should not simply “run as men” because they must also meet voters’ expectations about feminine strengths. Our findings problematize the common binary of “running as a man” or “running as a woman” because voters clearly value both masculine and feminine traits in female candidates. While female candidates face a greater penalty when perceived as lacking feminine strengths, neither men nor women can rely solely on the strengths stereotypically associated with their gender.
Investigating how male and female candidates can best respond to these types of attacks is an important avenue for future research. Our results indicate that gendered trait and issue attacks are effective, but we do not expect that candidates cannot counter them. Observational research on campaigns finds that candidates respond to their opponents’ actions in a strategic manner (Dittmar 2015; Windett 2014). Future research could evaluate the role of stereotypes in crafting effective strategic responses to gender-based campaign attacks, and whether responses are contingent on candidate gender and party. It may be that female candidates are often successful in counteracting stereotype-based attacks, thus neutralizing the cumulative effects of the stereotypes, which would fit some extant scholarship that finds few effects of stereotypes on voter behavior.
The increasingly negative environment in political campaigns in the United States requires further research into the substantive content of campaign attacks. Our research suggests that one of the major limitations of the literature on negative campaigning—and its effectiveness—may be the common practice of grouping dissimilar campaign attacks together and averaging across their effects. Instead, our findings suggest that scholars should evaluate the connection between the content of political attacks and the characteristics of the attacked candidates. Given that we consistently find that attacks on traits change the evaluations of the traits held by the attacked candidate, our results point to the continued importance of traits and trait-based campaigning.
Further research should delve deeper into qualitative differences in gender–stereotypic attacks and engage with a wider range of gendered traits. The traits we selected for this study were not an exhaustive list; we focused on traits that reflect widely held stereotypic strengths. Campaign attacks could focus on either a candidate’s stereotypic strengths or their stereotypic weaknesses. Future work could employ a wider range of trait and issue attacks in order to gain a richer understanding of the full spectrum of gendered attacks and their consequences. Work that expands the portfolio of gendered traits and issues employed in attacks may also offer additional insights into the unique vulnerabilities of female candidates based on their party, such as the differences observed for female Democrats and Republicans in Study 1.
Finally, our results raise additional questions about the extent to which party-based expectations moderate gender-based expectancies associated with trait and issue attacks. Our focus on primary races allowed us to explore how stereotypes influence candidate support outside a general election environment, holding the both candidates’ party constant. Research on primary races is rather limited in the current campaigns and elections and women and politics literatures, though preliminary research finds that women (particularly Republican women) face a tougher primary process than do men (Barnes et al. 2016; Crowder-Meyer and Lauderdale 2014). Our focus on a primary race is useful in demonstrating that these kinds of attacks are effective when made in intra-party contests, lending further support to the notion that primary races are gendered.
Future research should vary the party of the attacking and the attacked candidate, in order to further flesh out the role of party in gender-based attacks. Further, not all voters may be equally receptive to these attacks, particularly in general election contests. Because voters ultimately want their own party’s candidates to win and their party to exert more influence on policy, voters may care more about the failings of a candidate from their own party, particularly when those failings are associated with an issue or trait that the party stereotypically values. Future research might evaluate the extent to which voter partisanship shapes reception of these attacks.
Scholars have noted some differences between traits and issues in self-presentation and media and find that substantive attacks on traits are more effective than attacks without substantive content or issue based attacks (Dunaway et al. 2013; Fridkin and Kenney 2011; Brooks and Geer 2007), but little from this research offers a clear guidance as to whether attacks on traits or issue competence will interact more vigorously with candidate characteristics.
Reading a newspaper article likely offers a more conservative test of negative campaign effects relative to viewing a video of an ad, given that negative television ads can amplify emotional responses (Brader 2006). However, it does allow for a clean test of the causal relationship between those attacked and candidate perceptions.
We drew the control condition article from extant scholarship (Krupnikov and Bauer 2014).
Names used in the experimental treatment were drawn from previous experimental research (Holman et al. 2015).
For example, while uncompromising and weak both appear in the database, humility and protective do not, so we focused on the prior set of terms.
Women’s issues have been defined various ways in the literature, but commonly include issues pertaining to children, families, and social welfare generally (Holman 2014), which is consistent with the outcome of our factor analysis.
Modeling the effect of the treatment on these measures through an independent analysis of candidate type also ensures that we are comparing treatment conditions to the appropriate control condition. For example, estimating a regression model of the effect of the conditions on the Vote for the Republican candidates only compares votes in the masculine trait (or issue) attack conditions to the matched Republican control conditions, not all control conditions.
Data collection for the Trait Attack Study was funded by the 2015 Carrie Chapman Catt Prize. The authors would like to thank Angie Bos, Monica Schneider, Bas Van Doorn, J. Celeste Lay, Menaka Philips, the Gender and Political Psychology Writing Group, the Tulane Political Science Junior Scholar Research Group for their comments on drafts of this project, and our anonymous reviewers for their careful and constructive feedback. A previous draft of this paper was presented at West Virginia University and The College of Wooster. All data and code needed for replication is available on the Harvard Dataverse at https://dataverse.harvard.edu/dataset.xhtml?persistentId=doi:10.7910/DVN/OBGAHG.
- ABC News. 2015. ‘This week’ transcript: Donald trump. ABC News. November 22.Google Scholar
- Bos, A. L., & Schneider, M. C. (2015). Party and gender stereotypes. Journal of Women, Politics & Policy (Forthcoming).Google Scholar
- Brader, T. (2006). Campaigning for hearts and minds: How emotional appeals in political ads work. Chicago: University of Chicago Press.Google Scholar
- Brooks, D. J. (2013). He runs, she runs: Why gender stereotypes do not harm women candidates. Princeton: Princeton University Press.Google Scholar
- Dittmar, K. (2015). Navigating gendered terrain: Stereotypes and strategy in political campaigns. Philadelphia: Temple University Press.Google Scholar
- Eagly, A. H., & Carli, L. L. (2007). Through the labyrinth: The truth about how women become leaders. Cambridge, MA: Harvard Business Press.Google Scholar
- Holman, M. R. (2014). Women in politics in the American city. Philadelphia: Temple University Press.Google Scholar
- Kornblut, A. E. (2009). Notes from the cracked ceiling. New York: Crown/Archetype.Google Scholar
- Paolacci, G., Chandler, J., & Ipeirotis, P. G. (2010). Running experiments on Amazon Mechanical Turk. Judgment and Decision Making, 5(5), 411–419.Google Scholar
- Wallace, G. (2016). Negative Ads Dominate in Campaign’s Final Days. CNN, November 8. http://www.cnn.com/2016/11/08/politics/negative-ads-hillary-clinton-donald-trump/index.html.