Abstract
In 2016, Hillary Clinton was the first woman to gain the presidential nomination from a major political party in the United States, yet she was unsuccessful. The current study explores barriers to being elected as president for women generally and Hillary Clinton specifically. Using the propositions and tools of affect control theory, we demonstrate how women’s political representation shapes cultural sentiments about women and the president. In a nationwide sample of Americans surveyed shortly before the election, we find women’s representation on the state level influences voter preferences through these cultural sentiments: More women in politics makes a woman president feel less deflecting, which is associated with a greater likelihood of voting for Clinton. We also demonstrate how sentiments about Clinton—as an individual, not merely a woman running for president—conflict with Democratic voters’ expectations for presidential qualities and behaviors, which may have further prevented victory in 2016.
Women have been running for and winning office in the United States for well over a century, with Jeannette Rankin being the first woman elected to Congress in 1916. However, today men still hold far more offices than women, and women’s representation in governing bodies across the United States varies greatly—representation of women in state legislatures varies from nearly 40 percent in Vermont to as low as 11 percent in Wyoming. Some states, like Tennessee, currently have no statewide elected women in their state’s executive branch, and both of their U.S. senators are male. Other states, such as Oregon, have multiple women in the executive branch, including their current governor.
Beyond concerns of pure equality—women make up about half of the U.S. population but about 20 percent of members of Congress—women’s representation matters because women legislators focus on policies that benefit women and are supported by women, which results in more policies being passed regarding women’s issues (Barnello and Bratton 2007; Boles 2001; Bratton and Ray 2002; Burrell 1994; Carroll 2001; MacDonald and O’Brien 2010; Norton 1999; Poggione 2004; Reingold and Smith 2012; Saint-Germain 1989; Thomas 1991; Thomas and Welch 1991, 2001). There is also evidence that the presence of women candidates and representatives increases female voters’ political interest and engagement in politics (Atkeson 2003; Burns, Schlozman, and Verba 2001; Hansen 1997). Due to the importance women have for legislative outcomes and their effects on women’s political participation more generally, it is essential to explore barriers to women’s ability to win elections.
Even though the number of women in different state houses and Congress varies, there is one elected position in which no woman has been elected to: the presidency. The 2016 election was the first in which a woman was the nominee for president from one of the two major parties. The 2016 election was close, and Democratic nominee Hillary Clinton won the popular vote, but she ultimately did not win the necessary votes to become the first woman president of the United States. It is impossible to predict what would have changed in our country with a woman president. However, it is reasonable to assume that a female president, like female legislators, would focus on somewhat different policies and regulations than a male president.
Since Jeannette Rankin, the number of women in Congress and running for Congress has greatly increased. Although the progression has not been a steady increase, the 2016 election saw a record number of major-party women running for Congress. It can thus be assumed that more women will run for president and we will soon see more women nominees and eventually a woman president. Political news sources such as The Washington Post have already mentioned several potential female candidates as frontrunners to run for president in 2020 (Cillizza 2017). The presidential nomination of Hillary Clinton and her loss in the 2016 election presents an opportunity to determine what factors may have hurt her chances of winning and what factors may affect future female candidates for president.
We argue that parity in political office actually influences voters’ decisions—that the political representation of women in the 50 states shapes cultural sentiments about women, presidents, and ultimately, voting intentions. We expect that in times and places where there are more women holding political office, it is more plausible to imagine a woman serving as the president of the United States. Conversely, in times and places where political positions—state seats, the governorship, and Congress—are more male dominated, it is harder to imagine a female as presidential. Thus, we posit that women’s representation in politics is not only important for the sake of equality or because women tend to focus on issues that affect women, children, and minorities but because their representation influences how Americans perceive women and politicians and thus their voting intentions.
We utilize affect control theory (e.g., Heise 2007; MacKinnon 1994; Robinson and Smith-Lovin 2006; Smith-Lovin and Heise 1988), a theory of culture, identity, and behavior, to outline and test these hypotheses in a nationally representative sample of Americans of voting age. Affect control theory (ACT) scholars argue that there are three core underlying dimensions of social interaction: evaluation, potency, and activity (EPA) (Osgood, Suci, and Tannebaum 1957). These EPA ratings are the mathematical representation of cultural sentiments (also called fundamental sentiments), our feelings about persons, objects, and actions given the culture we live in. Evaluation, potency, and activity are associated with esteem, power, and the readiness to act on social environments, respectively (Scholl 2013). An identity’s EPA profile shapes our expectations for that person, providing us with information about how to treat that person and how they might treat us. Thus, one could consider these sentiments the building blocks of social interaction.
Here, we calculate the distance between respondents’ ratings of woman and their ratings of president to measure how divergent the two concepts are for individuals—how confusing, or deflecting, a female president would be for this person. We use this variable, woman/president deflection, to link state-level representation of women in politics to voting intentions. Finally, we examine differences in sentiments between Democrats and Republicans and how their sentiments compare to those about former Secretary Hillary Clinton and President Trump. Although we establish potential cultural-affective barriers for women candidates generally and in this specific place and time, sentiment ratings for specific candidates can better provide information for predicting voting patterns.
In sum, we ask the following research questions.
Research Question 1: Is women’s political representation associated with individuals’ sentiments about woman and president identities?
Research Question 2: Do discrepancies between the sentiments for the identities woman and president explain voters’ intentions in the 2016 election?
Research Question 3: How do the average ratings of president among Democrats and Republicans compare to sentiments for Hillary Clinton and Donald Trump, respectively?
In the following, we provide an overview of research on women candidates and ACT’s core concepts. Next, we present our hypotheses, methods, and the results of our analyses. We conclude with a discussion of the implications of these findings for gender and elections broadly, in relation to Hillary Clinton specifically, and for the utility of ACT for predicting political behavior.
Women Candidates
Political scientists have long studied women candidates and their chances of winning elective office. Early findings in the 1970s and 1980s generally found that women candidates were discriminated against due to their gender, receiving fewer votes than male candidates (Diamond 1977; Githens and Prestage 1977; Rule 1981; Welch 1978). Research during this time suggested that this gender electoral performance gap was due to women not working in fields, such as law or business, that typically produced future politicians (Welch 1977). However, earlier studies did not account for the incumbency effect: In open seat races where no incumbent is running, women candidates do just as well as and sometimes even better than men candidates (Dolan 1998; Smith and Fox 2001). Recent researchers have found that when party and incumbency are controlled for, women candidates often perform just as well as men candidates do in elections (Burrell 1998; Carroll 1994; Darcy, Welch, and Clark 1994; Dolan 2004; Duerst-Lahti 1998; Seltzer, Newman, and Leighton 1997; Smith and Fox 2001; Thomas and Wilcox 1998). Studies show that one reason for this more recent success is due to the fact that the number of quality female candidates has increased—still, there is evidence that voters only support women candidates at similar rates to men when they are more qualified (for a discussion of theoretical and methodological debates, see Fulton 2012).
There is a plethora of research that outlines the barriers and forms of discrimination female candidates face. Women are less likely to be recruited than men for elective office, perceive themselves as being qualified for office, and want to run for elective office, including at high levels (Bledsoe and Herring 1990; Costantini 1990; Fox and Lawless 2004, 2010, 2011; Fulton et al. 2006). Once women become candidates, scholars have found that women are discriminated against by party members, and voters appear to prefer male-dominated legislatures (Niven 1998; Sanbonmatsu and Dolan 2009). Furthermore, Republican voters in particular seem to have even more of a preference for male candidates, and voters tend to perceive women candidates for office as being more left wing than male candidates (Alexander and Anderson 1993; King and Matland 2003; Koch 2000, 2002; McDermott 1997, 1998).
Voters have been found to have gender-specific stereotypes regarding male and female candidates that may result in women candidates being perceived as lacking important political skills (Fox and Smith 1998; Lawless 2004). These gender stereotypes may even cause some voters to have an underlying preference to be represented by a man or woman candidate (Sanbonmatsu 2002). Some specific political offices are more likely to be impacted by gender stereotypes as higher level political offices are stereotyped by voters as requiring masculine traits (Rosenwasser and Dean 1989). Cultural stereotypes and attitudes are important; in a cross-national survey, Paxton and Kunovich (2003) found a negative significant relationship between participants’ agreement with the statement “men are better in politics” and women’s parliamentary participation in their country.
Theoretical Perspective
Sociologists offer additional lenses through which we can understand women’s experiences in running for political office. In particular, the group processes tradition and sociological social psychology demonstrate a multitude of ways in which women experience barriers to selection and advancement in organizations and leadership contexts. Most relevant are a series of theories and empirical studies that suggest women as a lower status group are less easily granted legitimacy and are perceived as less competent than men when gender is a salient characteristic (e.g., Lucas 2003; Ridgeway, Johnson, and Diekema 1994). These status hierarchies influence interaction dynamics and conversation patterns in ways that inform decision making, self-assessments, and acceptance of influence (e.g., Balkwell and Berger 1996; Carli 1990; Johnson 1994; Thébaud 2010; Walker et al. 1996).
That women receive less consideration for hiring and promotion is also evidenced by a series of experimental and audit studies in which women receive fewer callbacks or offers and lower recommended starting salaries in a series of employment sectors, such as in service, the performing arts, marketing, and business (e.g., Correll, Benard, and Paik 2007; Goldin and Rouse 2000; Neumark, Bank, and Van Nort 1996). Furthermore, research shows that women and other low-status individuals experience stricter, higher standards for being perceived as competent and that women leaders receive backlash when violating gender expectations because agentic females disrupt status and dominance hierarchies (e.g., Correll 2004; Foschi 2000; Rudman et al. 2012). Although such research has not been conducted, theories of status, legitimacy, and double standards could be used to understand barriers to recruitment, advancement, and election among potential women candidates.
Affect Control Theory
Status and power considerations, while certainly helpful, cannot entirely explain gender differences in perceptions and behavior—gender socialization and cultural sentiments also shape reactions to, treatment of, and behavior from and by men and women (e.g., Balkwell and Berger 1996; Ridgeway and Diekema 1992). Affect control theory is one way in which we can understand how culture informs gender-differentiated interactional dynamics (e.g., Heise 2007; MacKinnon 1994; Robinson and Smith-Lovin 2006; Smith-Lovin and Heise 1988). According to Smith-Lovin (1992), “gender (when salient) influences our patterns of interactions by serving as an identity [italics added] with certain levels of status, power, and expressivity to be maintained” (p. 149). These three factors—status, power, and expressivity—parallel the three EPA dimensions of evaluation, potency, and activity (Heise 1987; Morgan and Heise 1988). The cultural sentiments (EPA profile) of an identity, such as president or woman, invoke others’ expectations for those actors, expectations that said actors should strive to confirm to in order to maintain smooth interaction. Thus, ACT provides a numerical scheme for understanding both the production of first-order, identity-based expectations and the striving for confirmation of these expectations through social action.
When collecting EPA ratings, respondents are asked to rate concepts from bad/awful to good/nice for evaluation, little/powerless to big/powerful for potency, and inactive/quiet/slow to active/noisy/fast on sliding scales ranging from −4.3 to 4.3 (Heise 2010). These three dimensions have been used in dozens of states, countries, and language cultures to pinpoint identities, behaviors, emotions, attributes, settings, and nonverbal forms of communication in three-dimensional semantic space. These EPA ratings are averaged and combined into dictionaries of cultural meaning such that we can assess how, for instance, Americans feel about presidents (1.27, 3.09, 1.06) and criminals (−2.58, −.25, .59), leading (2.05, 2.62, .57) and harassing (−3.04, .55, 1.56), a kind person (3.42, 2.95, −.04) and a cruel person (−3.50, .83, .59), and so forth (Robinson and Smith-Lovin 2016).
Predicting perceptions and behavior
Due to the consistency of measurement across components of interaction, we can predict what types of behaviors and emotions we can expect from a certain person, like a president. Using the Find Concepts option in Interact (Heise 2011), the theory’s publicly available, predictive software, we can see that some of the behaviors rated most closely to president are influence, discipline, and elect; we would expect a president to be persuasive and strong-willed, to feel proud and self-assured (Robinson and Smith-Lovin 2016). Conversely, we would expect a criminal to be immoral and selfish, feel irritated, and discourage, blame, and sneer because these concepts are closely rated to criminal in three-dimensional EPA space (Robinson and Smith-Lovin 2016).
Interact can also be used to simulate how interactants would behaviorally, emotionally, and cognitively respond to others’ actions. Each event produces transient impressions for the actor, behavior, and object, which are also measured on E, P, and A. When an actor behaves in ways that diverge from the cultural sentiments for the identity they are enacting, this causes confusion, termed deflection in the theory—it is the distance between the fundamental cultural sentiments and momentary transient impressions for the actor, behavior, and object in the event (MacKinnon 1994). To make sense of deflecting events, interactants will behave in ways that realign meanings for actors, modify interactants’ identities, or redefine the behavior witnessed. A president who yells at someone may be reasoned with, seen as narrow-minded, or redefined as a malcontent. Any of these actions, modifications, or redefinitions of identities would bring the sentiments produced by the event (transient impressions) closer in line to cultural sentiments about presidents.
Subcultures of meaning
EPA ratings, when collected in a particular culture, are assumed to be fairly stable across the population, which has been confirmed in empirical analyses (Heise 2010). However, exceptions exist when examining the ratings of concepts that are central to subcultures. Studies have shown, for example, that the EPA ratings of “kynd” behaviors at festivals differ among individuals who have participated longer and have more ties to jamband culture, where this term originated (Hunt 2010). Significant differences in EPA ratings exist among members of weight loss groups, Internet users, and members of LGBTQ-friendly churches (Graor 2008; King 2008; Smith-Lovin and Douglass 1992). These subcultural sentiments in turn influence members’ understandings of events and expectations for behavior. Given the evidence that Republicans tend to prefer male candidates (King and Matland 2003), we explore whether EPA ratings of woman and president differ significantly between political party members.
Analyses of a recently collected set of EPA profiles for over 2,000 identities, modifiers, and behaviors show that as a whole, there are very few gender differences in EPA ratings, which is consistent with Heise’s (2010) results (Boyle 2015). After adjusting for the large number of analyses conducted, less than 30 concepts were rated differently based on respondent sex. Among them are female, which women rated higher on evaluation, and male, which men rated higher on evaluation. Thus, given the rarity of gender differences in EPA ratings yet the significant difference for these gender terms, we also examine whether men and women rate these identities differently. Thus, we can offer one explanation for differing preferences for male candidates based on political party and respondent sex—we expect men and Republicans to find a female president more deflecting than women and Democrats.
Hypotheses
Although we do not in any way discount the tangible and overt barriers that women candidates face—either interpersonally or in the outcomes of recruitment or election processes—the current study utilizes a theory that links structure, culture, perceptions, and behavior. Affect control theory, as will be demonstrated here, shows how the structural composition of occupations—in this case, the gender composition of political positions—shapes perceptions of women, affects cultural sentiments about the presidency, and influences voting patterns. After assessing group differences in EPA ratings for woman and president, we test hypotheses linking women’s political representation, cultural sentiments, and voter intentions.
First, we expect that having more women in elected office makes a woman president less culturally discordant to individuals. Because individuals are more accustomed to seeing women in positions of power—political power in particular—seeing a woman run for president and imagining her in that role feels more authentic and less bewildering. Thus, while controlling for factors that typically drive voting patterns, such as age, race, sex, education, political party, and political ideology, we expect that women’s representation in political office decreases the deflection of a woman president and indirectly increases intentions to vote for Hillary Clinton in 2016 through this deflection variable.
Hypothesis 1: More women in elected office decreases woman/president deflection.
Hypothesis 2: Lower levels of woman/president deflection are positively associated with intentions to vote for Clinton.
Hypothesis 3: Women in elected office has a significant indirect effect on voting intensions through woman/president deflection.
In testing these hypotheses, we extend the utility of ACT to predicting political behavior, explain the effects of structural political representation on voting intentions, and shed light on an additional way through which we can understand Clinton’s loss in 2016. Beyond overt sexism about women’s place and more implicit biases and beliefs about women’s competence, we argue that general social psychological processes as outlined in ACT serve as an additional barrier to women in political office and in this case, to Clinton specifically.
Data and Methods
The current study utilizes data from several sources. The primary source is a survey of voters that was distributed on Amazon’s Mechanical Turk (also known as M-Turk) in October 2016, shortly before the presidential election. Survey responses allow us to test our hypothesis that sentiments about woman and president predict voting intentions. Second, we use the Institute for Women’s Policy Research (2015) report, which provides information about women’s political representation on the state level. We use these data to determine whether women’s political representation predicts participants’ sentiments about woman and president and preferences for Clinton.
Third, we use the results of a Quinnipiac University Poll, which reports the words most commonly used to describe Hillary Clinton and Donald Trump. The fourth and final source of data is the Georgia Dictionary, from which we gather the aggregated cultural sentiments (EPA ratings) for the words most often used to describe Hillary Clinton and Donald Trump. Together, these four sources of data allow us to better understand both barriers to women running for president generally and how sentiments about Hillary Clinton specifically may have partially prevented her election.
M-Turk Survey on the 2016 Election
The M-Turk website allows respondents to take a job for a small cash payment. In this instance, workers on M-Turk were paid $0.45 to take our short survey regarding candidate choice in the 2016 election. After providing informed consent, respondents were directed to the survey, which was programmed using Qualtrics’s data collection software. Respondents completed questions about their demographics; cultural sentiment measures (EPA ratings) for the identities man, woman, and president; and indicated their political preferences and voting intentions in the 2016 election.
Sample
Only respondents who were American citizens and age 18 or older were eligible to take the survey. Respondents who answered “not sure” to the question “Who do you plan on voting for in the November Presidential Election?” were excluded (N = 184) since we are interested in the perceptions of voters and their intentions for voting Clinton, Trump, or third party. Respondents in this M-Turk sample are slightly more likely to be white non-Hispanic (79 percent), have a college degree (62 percent), and indicate support for Clinton (55 percent) than in the general population. However, the sample is nationally representative, coming from 49 states (all but Wyoming), and it is more representative than samples of typical college students (the average age of our sample was 37).
Four respondents were removed because they reside in Washington, D.C., for which all state-level variables are not available. Fifty respondents were lost because they did not complete EPA ratings. Respondents lost due to missing data were slightly less likely to be Democrats and were significantly younger than those not missing these data. The final sample consists of 1,162 Americans, aged 18 or older, who provided voting intention and EPA data. Table 1 displays the characteristics of the survey respondents.
M-Turk Sample Demographics and Correlation with Study Variables.
Note: Values = Pearson correlation coefficients; Percentage = percentage of the sample; M = mean; SD = standard deviation.
p < .05.
Dependent variable: Voting intentions
Respondents were asked who they will vote for in the 2016 election. Their response options were Hillary Clinton, Donald Trump, a third-party candidate, and not sure. Fifty-five percent of respondents who selected a candidate said they would vote for Hillary Clinton, and 31 percent said they would vote for Donald Trump (14 percent said they would vote third party). Although this sample is more supportive of Clinton and third-party candidates than the general election results, patterns are reflective of polling done in early October when the survey was conducted. Many polls in early October had Hillary Clinton’s percentages in the high 40s with Trump in the mid-30s and third-party candidates hovering around 10 percent. The suppressed support for Donald Trump could also be due to our sample being more educated than the population as a whole as education level was highly predictive of voting behavior in 2016. Bivariate correlations (Table 1) indicate that predictors of our sample’s voting intentions reflect patterns in the nation as a whole. Non–college degree holders, whites, and males were all less likely to vote for Clinton.
Independent variable: Woman/president deflection
The first independent variable of interest is woman/president deflection, which is the distance between respondents’ EPA ratings of the identity woman and the identity president (Equation 1) (Heise 2007). Participants were asked to rate each of these identities on evaluation (bad/awful to good/nice), potency (little/powerless, to big/powerful), and activity (inactive/quiet/slow to active/noisy/fast) (Heise 2007). Each EPA dimension ranged from −4 to 4 (the dimensions typically range from −4.3 to 4.3). Woman/president deflection was then calculated using the following equation, where Pe is the evaluation rating of president, We is the evaluation rating of woman, and so forth. Thus, woman/president deflection represents the distance in three-dimensional semantic space between ratings of the identity woman and identity president, reflecting how much these two identities diverge in their cultural sentiments.
We expect this variable to directly predict voting intentions and mediate the effect of women’s political representation on voting intentions.
Independent variable: Women’s political representation
The Institute for Women’s Policy Research (IWPR) assesses gender equality and women’s issues across the United States. In 2015, they released the Status of Women in the States Report, which scores, ranks, and compares each of the 50 states in terms of political participation, employment and earnings, work and family, and four other areas. In the current study, we use one indicator, women in elected office, which accounts for female governors and state legislators, women in the U.S. Congress, and women in statewide executive office as of March 2015. States are ranked as “best” and “worst” states for women in elected office (New Hampshire held the best ranking, and Kentucky held the worst). We use this variable as our ranking of women in politics, which ranges from 1 to 50. To ease interpretation, we multiplied this number by −1, such that a higher women in politics ranking indicates that there are more women in elected office in that state.
Control variables
Respondents were also asked about their political affiliation and ideology. Political party identification is a 7-point scale ranging from a strong Democrat to a strong Republican, with Independent in the center. Conservative ideology is respondents’ ideological self-placement on a liberal (0) to conservative (100) scale. For descriptive purposes, we also separate Democrats, Republicans, and Independents in analyses. We controlled for respondent sex (male), race (white), and education level (college degree). Age is a continuous variable ranging from 18 to 81.
We also include two state-level variables in structural equation models. Specifically, we include a variable for the percent voted for Obama (2012), which is the percentage of the state that voted for Barack Obama in the 2012 presidential election, an indicator of how liberal a state is. We also include a dichotomous variable for northeast region, which is coded 1 for participants who reside in states that are located in the U.S. northeast census region since states in this region are significantly more likely to have women in elected office and to vote Democrat (analyses available on request). Thus, we partially ensure that it is not state liberalism or a general inclination toward Democratic candidates driving effects.
Words and Sentiments Associated with Hillary Clinton and Donald Trump
Quinnipiac University Poll
In the M-Turk survey of voters, we asked participants for EPA ratings of man, woman, and president, which does not provide information about Hillary Clinton and Donald Trump as specific individuals. However, data are available that inform this discussion. In August 2015, the Quinnipiac University Poll (2015) asked participants what word first came to mind when they think of Hillary Clinton. This survey was conducted between August 20 and August 25, 2015. The sample size consists of 1,563 registered voters—666 Republican/Republican-leaning voters and 647 Democrat/Democrat-leaning voters. Most participants were white (74 percent) and did not have a college degree (67 percent). Survey Sampling International used random-digit dialing to generate a call list of cell phones and landlines, which live interviewers called.
Participants were asked a number of questions about their demographics, party affiliation, and opinions of primary candidates before being asked, “What is the first word that comes to mind when you think of Donald Trump [Hillary Clinton]?” The report lists the number of times each word was used, including only words that were used at least five times. Fifty words were reported for Hillary Clinton, and 66 were reported for Donald Trump. These sets of words contain mostly identities (e.g., liar and woman for Clinton and idiot and businessman for Trump) and modifiers (e.g., strong and deceitful for Clinton and arrogant and outspoken for Trump). For Clinton, there were also a number of behaviors listed (e.g., murder, cheat), in addition to her husband’s name, Benghazi, and email. For Trump, change and hair were listed in addition to identities and modifiers.
Georgia Dictionary
We draw from the Georgia Dictionary (Robinson and Smith-Lovin 2016) to gather the EPA profiles for the words provided by Quinnipiac participants. The Georgia Dictionary is a set of EPA profiles for 488 identities, 495 behaviors, and 299 modifiers that were collected at the University of Georgia. Of the words used for Trump and Clinton in the Quinnipiac University Poll, 29 words used to describe Clinton and 32 words used to describe Trump were available. We used the 15 most commonly used terms for each candidate to estimate an EPA profile for Clinton (–0.20, 1.70, 0.52) and Trump (–0.19, 0.86, 1.38), which are listed in Table 5.
Analytic Strategy
In the first set of analyses, we report the means and standard deviations of demographic characteristics, political party identification, and political ideology. We report group differences in EPA ratings and deflection scores. Second, we test our hypotheses that women in politics will be negatively associated with woman/president deflection (Hypothesis 1) and that woman/president deflection will be negatively associated with voting for Hillary Clinton in the 2016 election (Hypothesis 2). We then determine whether woman/president deflection mediates the relationship between women in politics and voting intentions (Hypothesis 3). Because this is a mediating relationship and we aim to examine both direct and indirect relationships, we use structural equation modeling 1 in STATA 13. Voting for Clinton is a categorical variable (1 = will vote for Clinton, 0 = will vote for Trump or vote third party), so we use the generalized structural equation modeling (GSEM) function in STATA and use the probit option that marks Clinton as a dichotomous dependent variable. We use the standard SEM option, however, to report indices of fit, which are not available through GSEM.
In our final set of analyses, we compare the average EPA profile of president for Democrats and Republicans (Table 4) to the EPA ratings of terms used to describe Hillary Clinton and Donald Trump (Table 5). We plot these concepts to demonstrate the ways in which these candidates diverge from “typical” presidential expectations within their respective parties.
Results
Variation in EPA Ratings and Deflection Scores
In the first set of analyses, we examine group differences in EPA ratings for man, woman, and president and in deflection scores. Table 2 lists these EPA ratings by respondent sex, race, political party, and educational attainment. The results of a series of MANOVAs determine whether differences were statistically significant while controlling for other variables. In these analyses, the E, P, and A ratings for each identity were the dependent variables and sex, race, political party, education, age, and ideology were the independent variables.
Cultural Sentiments (EPA ratings) for Woman, Man, and President and Woman/President Deflection by Groups.
Note: MANOVA tests were conducted to establish significance, in which the E, P, and A ratings for each identity were the dependent variables and sex, race, political party, education, age, and ideology were the independent variables. E = evaluation; P = potency; A = activity; M = mean; F = F statistic; b = Unstandardized coefficients.
p < .05 (one-tailed tests).
On average, respondents rated woman (1.42, .33, .80) higher on evaluation, lower on potency, and lower on activity than man (.97, 1.16, 1.25). However, males rated the term woman significantly lower on evaluation than females did, and males rated the term man significantly lower on potency and activity than females did. Political party identification and conservative ideology also influenced EPA ratings of man and woman. Democrats rated woman higher and man lower on evaluation than Republicans; Democrats rated woman and man the highest on potency. More conservative (vs. liberal) ideology was positively associated with evaluation ratings of woman and positively associated with E, P, and A ratings of man.
Even when controlling for political party and ideology (males are less likely to be Democrats and are more conservative), there are significant gender differences in ratings for president—females rated president higher on evaluation and activity. Democrats rated president higher on E, P, and A (1.71, 1.85, 1.33) than Republicans (–.35, 1.36, .81) and Independents (.18, 1.31, .87), and more conservative respondents rated president lower on evaluation. Thus, women and Democrats perceive the role of the president in more positive terms, as a more benevolent, powerful, and active leader than men and Republicans, respectively. Democrats had the lowest levels of woman/president deflection (10.79), followed by Independents (14.75) and Republicans (21.41), a significant difference (F = 14.32, p < .05).
Voting Intentions
Next, we assess the relationships between having more women in elected office on the state level, woman/president deflection, and respondent voting intentions. To do so, we estimate structural equation models. Generalized SEM is the appropriate tool to use here given the categorical nature of our dependent variable (voting intentions). However, we first run this model using a standard structural equation model because GSEM does not provide goodness-of-fit indices or significance tests for indirect paths. As is customary, we first run a model that allows for a variety of directional paths, even those not hypothesized (available on request). Although this model has satisfactory fit (root mean square error of approximation [RMSEA] = .017, Comparative Fit Index [CFI] = .991, Tucker-Lewis Index [TLI] = .996), releasing nonsignificant paths can lead to an even better fit and greater clarity for our hypothesis tests.
Figure 1 displays this trimmed model, which has a better fit (RMSEA = .010, CFI = .999, TLI = .999). The STATA command called “estat teffects” tells us the coefficients and their significance for direct and indirect paths. Solid lines in Figure 1 indicate a significant direct effect, and a dashed line indicates a significant indirect effect. Although control variables for respondent race, sex, educational attainment, and age were included in models, they are not included in Figure 1 to improve clarity. After assessing goodness of fit and direct and indirect effects, this same model was run using GSEM; direct effects from this model are listed in Table 3.
Trimmed structural equation model predicting voting intentions in 2016 election.
Generalized Structural Equation Model Predicting Voting Intentions in the 2016 Election.
Note: b = unstandardized coefficients; SE = standard error.
p < .05 (one-tailed tests).
As hypothesized (Hypothesis 1), participants who reside in states where there are more women in elected office have lower woman/president deflection scores (b = –.057, p < .05). Second (Hypothesis 2), higher woman/president deflection scores are negatively associated with intentions to vote for Hillary Clinton (b = –.009, p < .05). Although women in politics does not directly predict voting for Clinton, woman/president deflection links women in politics to voting intentions (p < .05), lending support to Hypothesis 3. The Sobel-Goodman and binary mediation commands estimate the relationship between an independent, dependent, and mediating variable. These calculations indicate that woman/president deflection explains about 38 percent of the effect of women in politics on voting intentions, providing further support for Hypothesis 3.
“Presidential” Concepts and Estimated EPA Profiles for Clinton and Trump
The largest differences in EPA ratings for woman, man, and president are between Democrats, Republicans, and Independents, providing evidence for subcultural differences in sentiments. To demonstrate the substantive significance of these differences, we located concepts from the Georgia Dictionary that are rated closely to the average EPA profile for president among Democrats (1.71, 1.85, 1.33) and among Republicans (20.35, 1.36, .81) in our sample. The discrepancy in EPA ratings between members of the two major parties leads to quite different expectations for how presidents should look, talk, and behave. Among Democrats, “presidential” qualities include things like industrious, enterprising, and witty (Table 4); behaviors like direct, endorse, and challenge are close to president in EPA space (behaviors available on request). Republicans, on the other hand, expect a president to be strict, provocative, and dogmatic (Table 4)—to incite, rebuff, and punish.
Evaluation, Potency, and Activity (EPA) Ratings for Qualities Located Closest to the Concept President among Democrats and Republicans.
Note: Concepts and EPA ratings derived from the Georgia Dictionary (Robinson and Smith-Lovin 2016). Concepts are listed in descending order of closeness to president.
But how do Clinton and Trump measure up to the typical expectations from their respective parties? The words most commonly used when asked to provide the first word that comes to mind when they think of Hillary Clinton were both very negative and very positive (Table 5). The most common word was liar, but participants also listed intelligent, good, determined, and leader; woman was the 7th most commonly mentioned word. The name of her husband, former President Bill Clinton, was mentioned even more commonly than woman, demonstrating the fact that voters associate her with her husband. Among the top 20 words were crook and criminal, aligning her with overt law-breaking identities, and email and Benghazi, aligning her with corruption (corrupt was the 18th most common term used). The estimated EPA profile for Clinton using the 15 most commonly used words is –.20, 1.70, .52, which is far more negative on evaluation and somewhat more negative on activity than Democrats’ general expectations for a president—who they would expect to be cool, fashionable, and proud.
Evaluation, Potency, and Activity (EPA) Ratings for Words Used Most Frequently to Describe Hillary Clinton and Donald Trump in a Quinnipiac University (2015) Poll of Voters.
Note: Concepts derived from the Quinnipiac University (2015) Poll. EPA ratings derived from the Georgia Dictionary (Robinson and Smith-Lovin 2016). Concepts are listed in descending order of frequency of use. The word experience was listed, but we used the Georgia Dictionary term experienced. The word ass was used, but we used the Georgia Dictionary term asshole.
A number of deeply negative terms were also used to describe Donald Trump (Table 5). Participants questioned his intelligence, calling him an idiot, clown, or buffoon. There were also mentions of his interactional style (outspoken, aggressive), his arrogance (the most commonly used term), and questions of his sanity (crazy was the 10th most common). Many participants also held him in high regard. Participants mentioned his honesty along with his perceived accomplishments as rich, a businessman, and a leader. The average of the 15 most commonly used terms is –.19, .86, 1.38. Although these evaluations are not very positive, this profile is not terribly discrepant from Republicans’ EPA profile for president.
Figure 2 highlights the importance of these distinctions, where we plot the words rated most closely to Democrats’ EPA profile for president (dark blue circles) and Republicans’ EPA profile for president (dark red circles). The averages of these terms are labeled as Republican president and Democrat president. Next, using EPA profiles in the Georgia Dictionary (Robinson and Smith-Lovin 2016), we averaged the cultural sentiments for the 15 words used most frequently to describe each candidate in the Quinnipiac University Poll (2015) study. Figure 2 shows us that although the estimated EPA profile for Trump is rather negative, this is not in actuality very far from Republicans’ EPA profile for president (distance = .60). Conversely, Clinton’s estimated evaluation profile is also somewhat negative, yet she is located further from Democrats’ EPA profile for president (distance = 4.33). Thus, although both candidates were described with negative words, the profile for Donald Trump diverged less from his party’s ratings than that of Hillary Clinton.
Evaluation, potency, and activity (EPA) profiles for concepts rated closely to president for Democrats and Republicans and estimated EPA profiles for Hillary Clinton and Donald Trump.
Discussion
Women are unequally represented in American electoral politics at multiple levels, including at its highest levels in Washington, D.C. Political scientists offer various explanations for this underrepresentation, including the fact that women are less likely to be recruited to run for office (Fox and Lawless 2010). The 2016 presidential election is unique in that it is the first presidential election in the United States where a woman won the candidacy for a major political party. Hillary Clinton faced a man, Donald Trump, in the race to the White House. Although these individuals are demographically similar in a number of ways—race, class, sexual orientation—their difference in gender provides an opportunity to explore some of the mechanisms that prevent women from breaking the highest of glass ceilings.
Sociology offers a number of lenses through which we can view gender inequality in organizations and leadership. Affect control theory (e.g., Heise 2007; MacKinnon 1994; Robinson and Smith-Lovin 2006; Smith-Lovin and Heise 1988) in particular provides a multilevel approach to understanding one reason why Clinton—and other women seeking high-status, powerful positions in the United States—seemed to face unique, gender-based challenges that male counterparts were less vulnerable to. Primarily, does Hillary Clinton have the “stamina” for the job, or the “presidential look”? Trump’s quotes about Clinton are identity-based attacks on her ability to lead based on her strength and liveliness—she is also, according to Trump, a “bad, bad person.” Trump repeatedly called her “Crooked Hillary” and led chants of “lock her up,” invoking the same sentiments of criminality, dishonesty, and corruption present in (some) Americans’ perceptions (Berenson 2016; Martosko 2016; Tatum 2016).
For scholars familiar with ACT, these quotes from Donald Trump might evoke considerations of the evaluation, potency, and activity profiles for president and woman—in particular, how cultural sentiments for president and woman diverge and affect voting patterns. The current study employed ACT to highlight how structural gender inequality in important institutions (in this case, the political arena) translate into cultural sentiments. These sentiments in turn decrease the likelihood of voting for a woman, furthering still the problem of underrepresentation. To be more precise, having more women in elected office does not necessarily directly impact the raw EPA ratings of woman; rather, it affects the distance between woman and president. In actuality, women’s representation had more of an impact on evaluation ratings of president than on any single EPA dimension for woman. This poses a question that necessitates further research—does having more women in political office lead one to see women differently, leaders differently, or both? Furthermore, does this line of research translate into other arenas of occupational inequality?
If these processes can be generalized to other occupations, it is possible that the increased entry of women into male-dominated fields, including law, business, and politics, shifts sentiments about those positions. Rather than positively impact sentiments about women as more powerful, women’s entry into occupations previously held by males may instead change sentiments about that job. Since the movement of women in large numbers into jobs is associated with a pay decrease in that job (Levanon, England, and Allison 2009), sentiment shifts may be partially to blame for why these jobs come to be seen as less important, as one of the ways “gender bias sneaks into” salary decisions (quoting England in Miller 2016). These lines of inquiry could reveal ways that inequality operates and the gender pay gap persists, beyond more overt forms of discrimination and sexism.
Although effect sizes were small, in an election as tight as 2016, we suspect that these affective processes may have nudged some voters away from Clinton and toward Trump or a male third-party candidate, toward Bernie Sanders during the primaries, or even to abstain from voting altogether. Importantly, Clinton’s estimated EPA profile diverged much more from typical expectations for a president among Democrats than did Trump’s estimated EPA profile and his party’s expectations. Simply put, negative sentiments about Clinton likely hurt her more in the voting booth as participants predisposed to vote for her due to their party affiliation may have experienced more deflection than a Republican would be by a candidate located in a similar EPA space.
Among the most commonly used words for Clinton were liar, dishonest, untrustworthy, experience, and strong. The most commonly used words to describe Donald Trump were arrogant, blowhard, idiot, businessman, and clown. These words reflect the fact that first and foremost, Clinton and Trump are the least popular presidential candidates since Gallup began collecting such polling data in the 1950s (Saad 2016). Second, they remind us that voters were not simply evaluating a woman and a man for office but two specific, high-profile individuals who have saturated political, business, and mainstream culture for several decades. Although ACT has not been used in this way, it is possible that we hold cultural sentiments about specific celebrities, icons, and politicians with whom we are often confronted through media.
Affect control theory provides a new potential measure—EPA ratings—that can be used in unobtrusive ways to gauge voters’ expectations for and approval of men and women running for political office. For both candidates and elected politicians, affect control theory can tell us how to resolve voters’ deflection by demonstrating what kinds of characteristics and behaviors could bring sentiments about a person back in line with their expectations for that particular office and members of particular parties. Future research should aim to collect and analyze EPA ratings about candidates, especially as candidate pools become larger and more diverse, both during the campaign process and elected officials’ tenure.
Limitations and Future Research
There are some important limitations worth noting. The timing of this study could have been somewhat problematic as it was conducted in the immediate aftermath of the release of the Access Hollywood tapes in which Donald Trump used vulgar, sexually assaultive language about women (Fahrenthold 2016). Due to the ongoing campaign and later developments of the election, it might have been better to collect data immediately after the election. This would allow all campaign events to have taken place, and the respondents could have been asked which candidate they actually voted for.
Furthermore, although data were collected from participants in 49 U.S. states, the sample could better represent the demographics of the country. The sample was slightly more ideologically left wing, more college educated, and whiter than the country as a whole. This is likely due to the fact that the survey could only be completed by Amazon M-Turk users, which means the respondents likely had some level of technological competence and access. Future work should utilize samples that do not rely exclusively on M-Turk users, such as a survey conducted over the phone with random-digit dialing.
This study would also have benefited from the inclusion of additional survey measures, especially measures of socioeconomic status or social class, as the only measure included here is education. Other variables that might be of interest include the respondent’s family income, religiosity, and marital status. These additional measures might improve the quality of the analysis and give greater insight into who exactly felt the greatest woman/president deflection.
Future studies should employ a range of measures on gender attitudes and women’s status to build a more complex and complete understanding of how women’s representation in politics influences voting patterns. As mentioned earlier, there are a number of other traditions in sociology and political science that are relevant to our findings, such as status characteristics (e.g., Berger, Cohen, and Zelditch 1972; Wagner and Berger 1993) and expectation states (Berger 1988; Fisek, Berger, and Norman 1995; Ridgeway and Berger 1986). In the 2016 election, one could argue that gender was salient: It differentiated Clinton and Trump, and politics are considered a masculine, male-dominated domain in the United States. Although a debate or election would be considered outside the explicit scope of the theory, this differentiation could have potentially shaped performance expectations and evaluations of Clinton and Trump based on their gender (Correll and Ridgeway 2003). Another promising parallel literature that could be employed here are studies that examine the cognitive processing of gender schema and the effects of gender stereotypes and bias on candidate selection (e.g., Chang and Hitchon 2004; Riggle et al. 1992, 1997).
Like many researchers in these traditions, we suggest that experimental studies in which gender and other factors are manipulated could shed further light on responses to female candidates. Still, we argue that deflection processes are an appropriate approach to understanding perceptions of candidates and voting behavior given affect control theory’s mathematical explication of how the gender stratification of institutions shapes cultural sentiments—not just about women but about political offices in general—which in turn drive behavior. Whether these findings extend beyond the 2016 presidential election and to other institutions is an empirical question worth investigating.
Footnotes
Funding
The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: The authors would like to acknowledge Virginia Tech’s Open Access Subvention Fund for supporting the publication of this article.