How partisanship and sexism influence voters’ reactions to political #MeToo scandals

Abstract

Influential theories of motivated reasoning, as well as real-world anecdotal examples, would suggest that voters do not always penalize legislators from their own party for alleged immoral behavior, such as sexual harassment. But very little empirical evidence exists on how voters react to sexual misconduct allegations, especially since the start of the #MeToo movement. We examine how partisanship and sexist attitudes shape individuals’ reactions to sexual harassment allegations about a politician. Using a pretest–posttest online experiment, we randomize both the party affiliation of the accused legislator as well as the severity of the allegations. Overall, we find some evidence of partisan bias, but that there may be a limit. Subjects were more forgiving of an accused co-partisan legislator than a legislator of the opposing party in their overall evaluation and their perceptions of punitive repercussions, but their levels of electoral support decreased just as much for co-partisans as they did for opposing partisans. Importantly, these reactions are strongly conditioned by sexism; as subjects’ levels of sexism increase, the otherwise large and negative effect of allegations on evaluations of favorability and electoral support disappears.

Introduction

In late 2017, millions of people began sharing their experiences of sexual harassment on social media with the phrase and hashtag “Me Too.” What became known as the “#MeToo movement” originated in Hollywood, but has since grown to raise awareness about many forms of sexual misconduct by all sorts of perpetrators, including politicians. Although accusations have impacted elites from both parties, the public’s response has been polarized along partisan lines. ¹ Influential theories of motivated reasoning depict how partisanship affects voters’ perceptions of many phenomena, including politicians who are mired in sex scandals (Barnes et al., 2018; Doherty et al., 2014; Fischle, 2000; Funk, 1996). Yet most research in this area has focused on scandals involving marital infidelity, and even rarer are studies that examine reactions to scandals in the post-MeToo era. In cases of sexual misconduct, does partisan motivated reasoning still have the same influence? Other questions remain, such as how voters react to different kinds of sexual misconduct allegations and whether such allegations simply reduce voters’ expressive evaluations of a politician or actually influence their vote choice.

In this article, we examine how partisanship and sexist attitudes shape individuals’ reactions to #MeToo allegations about a politician. Using a pretest–posttest online experiment, we randomize both the party affiliation of the accused legislator as well as the severity of the allegations (sexual assault compared to sexist jokes). Overall, we find some evidence of partisan bias, but that there may be a limit. Subjects were more forgiving of an accused co-partisan legislator than a legislator of the opposing party in their overall evaluation and perceptions of punitive repercussions, but their levels of electoral support decreased just as much for co-partisans as they did for opposing partisans. Importantly, reactions are strongly conditioned by sexism; as subjects’ levels of sexism increase, the otherwise large and negative effect of allegations on evaluations of favorability and electoral support disappears.

Partisan motivated reasoning and reactions to scandal

Motivated reasoning has been one of the most widely used theoretical frameworks to understand how people process information (e.g. Druckman and Bolsen, 2011; Redlawsk, 2002; Taber and Lodge, 2006). People have two general goals when seeking and processing new information: to reach an accurate conclusion, or alternately, a particular directional conclusion (Kunda, 1990). When people feel emotionally invested in issues, they are more likely to be motivated by directional goals and process information in a way that aligns with their pre-existing attitudes, rather than prioritize the accuracy of that information.

Considering the intense emotional attachments that drive affective polarization (Iyengar et al., 2012; Mason, 2018), partisanship is a strong source of directional motivated reasoning (Druckman et al., 2013; Green et al., 2002; Sigel, 1964). Even if individuals do not truly alter their attitudes based on partisan biases, they may expressively engage in partisan cheerleading just to satisfy their directional motivations (Schaffner and Luks, 2018). But are there limits to this type of behavior? If politicians engage in immoral behavior, at what point will like-minded partisans penalize them? Several members of Congress in both parties have resigned in the wake of various misconduct allegations in recent years, suggesting that, at least anecdotally, legislators mired in sexual misconduct scandals are punished (or fear being sanctioned) by constituents. This aligns with research showing that moral judgments play a critical role in voters’ assessments of candidates. All else equal, voters favor politicians who are free of scandals such as tax evasion, marital infidelity, or corruption (Basinger, 2013; Carlson et al., 2000; Doherty et al., 2011; Funk 1996; Sigal et al., 1988). On the other hand, voters’ evaluations of politicians accused of unethical behavior are shaped by their preconceived perceptions (Fischle, 2000) and partisanship can be a stronger driver of voters’ responses to immoral behavior than their moral foundations (Walter and Redlawsk, 2019).

The extent to which voters penalize a co-partisan legislator embroiled in a #MeToo scandal may depend on the severity of the allegations. With less “serious” accusations (such as accusations of making sexist jokes, compared to sexual assault), are voters willing to afford co-partisans more leniency? Shared partisanship has been found to decrease suspicions of politicians accused of election fraud (Barnes and Beaulieu, 2014), but we argue that directional motivations are less likely to drive behavior when they impose higher cognitive discomfort, such as in the case of sexual misconduct. Given the increased sensitive nature, if accusations reach a certain level of severity, voters’ moral foundations may overcome partisan motivated reasoning.

We also argue that it is important to consider voters’ attitudes towards women. Recent studies have demonstrated how ambivalent sexism influences evaluations of political elites (Cassese and Barnes, 2018; Cassese and Holman, 2019) and even how sexism shapes reactions to sex scandals compared to corruption scandals (Barnes et al., 2018). For example, Barnes et al. (2018) find that hostile sexists are less likely to punish male candidates than women candidates for sex scandals, suggesting that sexist stereotypes could affect voters’ reactions to sexual misconduct specifically. Other research shows that people who endorse sexist stereotypes of women are less likely to think that sexual harassment warrants legal action (Wiener et al., 1997). But as we suspect there is with partisan motivated reasoning, is there a limit? As allegations increase in severity, do politicians face a penalty even from voters who hold sexist attitudes?

Research design

We fielded a pre-registered experiment on 2,806 subjects using Mechanical Turk (MTurk) in October, 2019. ² The experiment follows a 2x3 pretest–posttest design in which subjects evaluated a fictional male legislator, whose partisanship was randomly assigned, before and after exposure to one of three randomly chosen news stories about that legislator (see the Online Appendix for the full text of the vignettes). In the pretest, subjects indicated their agreement on a five-point scale with four statements selected for the purposes of this study from Glick and Fiske’s (1996) hostile sexism battery, such as “most women interpret innocent remarks or acts as being sexist” and “women are too easily offended.” The statements have high reliability (α = 0.95). We take the mean score across all four items to create one variable for subjects’ pretest levels of sexism. The mean of this variable is 2.4 (SD=1.2).

Subjects were then shown a description of a senator and asked to evaluate how favorably they felt towards him and how likely they would be to vote for his reelection (both measured on a 0–10 point sliding scale) in order to get their baseline ratings. ³ The only randomized component at this point was his party affiliation (Democrat or Republican). In the pretest, subjects gave a mean favorability rating of 6.9 (SD=2.1) and a mean 6.0 (SD=2.9) likelihood to support reelection, indicating that subjects at first felt relatively warmly towards the senator.

After answering several demographic questions, subjects were randomized into one of three conditions which exposed them to a news story about the same legislator. The story either reported allegations of sexual assault, reported allegations that he repeatedly made sexist jokes, or for the “control” condition, described the legislator’s recent trip to a county fair. In both the sexual assault and jokes conditions, two women separately come forward about the senator’s purported behavior and he is claimed to deny the accusations. In addition to the legislator’s claim of innocence, these treatments included another strong elite cue: “Many prominent [Democrats/Republicans] in Congress have also risen to Frederickson’s defense, citing his ‘integrity’ and claiming that the allegations lack clear evidence” where the partisanship of the apologists was randomized in line with the legislator’s partisanship.

Several considerations guided our design choices for the experimental vignette. First, in a relatively artificial survey environment, we considered the possibility of both social desirability bias and experimenter demand effects. The language that the legislator used to defend himself and with which he was defended by his colleagues serves to offer incentives against these effects. When politicians deny allegations instead of apologizing, they are perceived in a more positive light (Sigal et al., 1988). Individuals are also guided by elite cues when forming their opinions on candidates (Bisgaard and Slothuus, 2018; Gilens and Murakawa, 2002). These components also increase the ecological validity of the vignette; news coverage of similar allegations often includes quotes from the accused politician and reactions from other elites.

After reading the news story, subjects again evaluated how favorably they felt towards the senator and how likely they would be to vote for his reelection on 0–10 point sliding scales. Subjects that were randomized into the joke and assault conditions also answered whether they agreed with four statements about the appropriate course of action regarding the allegations, such as that the senator should apologize, resign, or that nothing needs to be done (α = 0.96). We again take the mean score across all four items in order to create a single variable for subjects’ levels of punitiveness against the accused legislator (mean=2.9, SD=1.0).

Finally, while we are most interested in how sexism conditions treatment effects, and thus will primarily use pretest responses to the sexism items, we recorded responses to these items posttest as well in order to measure changes in sexism post-treatment (mean=2.4, SD=1.2, α= 0.96).

Results

Effects on favorability, electoral support, and sexism

We begin by briefly reporting the difference in means on the post-treatment measures of favorability and electoral support. On the 0–10 scale, the mean favorability rating for subjects in the control condition was 7.1, while the mean rating for those in the assault condition was 4.4 (−2.7, p<0.001) and for those in the sexist jokes condition 5.3 (−1.8, p<0.001). In terms of how likely they would be to vote for the legislator, the mean in the control group was 6.2, in the assault condition was 4.0 (−2.2, p<0.001), and in the jokes condition was 4.8 (−1.4, p<0.001). Clearly, being exposed to either allegation decreased both kinds of support in relation to the control. While the differences between the assault and jokes conditions are less than the differences between these and the control, they are still statistically significant at the 0.001 level, indicating that our manipulation of the severity of the allegations was effective.

Turning to how these ratings changed from pretest to posttest, we use the percent change in favorability and electoral support in order to guard against floor effects. That is, since we are primarily interested in partisan motivated reasoning and since opposite party subjects begin lower on the scales in the pretest, their movement in the posttest is obscured if we use the raw change in magnitude from pre to post; estimates are thus falsely inflated for same party subjects. Nonetheless, in the Online Appendix, we present all analyses showing the raw change from pretest to posttest instead of percent change.

Overall, the assault and jokes conditions resulted in a decrease in favorability and electoral support post-treatment. The mean percent change for both favorability and electoral support was slightly over 10 percentage points larger for subjects in the assault condition than the jokes condition, again confirming that subjects reacted more strongly to the allegations of sexual assault compared to sexist jokes.

There are mixed results with regard to shared partisanship. Figure 1 shows the effect of each treatment compared to the control group on percent change in favorability. While both treatments result in a substantively large and statistically significant percent decrease in favorability, subjects are more forgiving when they share the legislator’s partisanship. After being exposed to the story about sexual assault allegations, opposite party subjects decreased their favorability by 38.6% compared to the control condition, while same party subjects decreased their favorability by 31.9%. In the sexist jokes condition, there was a 28.5% decrease in favorability for opposite party subjects and a 20.4% decrease for those who share the legislator’s partisanship. The differences between same party and opposite party subjects are statistically significant, suggesting that partisan motivated reasoning conditions how strongly individuals react to allegations.

Figure 1.

Treatment effects on percent change in favorability from pretest to posttest.

A different pattern emerges for electoral support. While subjects still had more negative reactions to the assault allegations than the jokes condition, there were no meaningful differences within conditions based on shared partisanship. As shown in Figure 2, the decrease in likelihood of supporting the reelection of a senator accused of assault was 30.3% for same party subjects and 30.6% for opposite party subjects (+0.3, p=0.94); in the sexist jokes condition, the decrease was 18% and 20.5% respectively (+2.5, p=0.5).

Figure 2.

Treatment effects on percent change in electoral support from pretest to posttest.

We next consider whether either treatment increased subjects’ levels of sexism, measured using the mean score on the four-item sexism battery. Note that this analysis was not registered as a formal hypothesis and thus should be considered exploratory. Nonetheless, since we included measures of sexism both pre- and post-treatment, there is an opportunity to explore whether subjects’ reported gender attitudes shift in response to allegations about sexual misconduct, especially those that are refuted by in-party elites. Figure 3 suggests that this may be the case. Compared to subjects in the control condition, there is a small but statistically significant percent increase in sexism for subjects who share the party with the accused legislator, but not for those who are of the opposite party.

Figure 3.

Treatment effects on percent change in sexism from pretest to posttest.

The Online Appendix presents additional results for pre-registered exploratory research questions on the treatment effects conditional on gender, party affiliation, age, and news interest. We find that Democrats, women, young millennials (subjects aged 18–24), and subjects that follow the news regularly penalize accused legislators more than Republicans, men, older subjects, and those who do not follow the news often, respectively. While we did not have strong expectations regarding these exploratory analyses, it is clear some heterogeneity exists in a way that generally aligns with existing observations about the #MeToo movement. For example, since the movement has been pioneered primarily by Democrats (Graf, 2018), it is possible that Democrats feel a greater ownership over the issue of sexual assault. Future research is needed to further understand how different groups in the mass public react to sexual misconduct allegations of varying severity.

Punitiveness and perceptions of consequence

We are also interested in investigating how partisan considerations shape determinations of appropriate consequences for allegations. We use the mean agreement on the repercussion questions (with 5 representing the most punitive and 1 representing the least punitive) to analyze how shared partisanship and the severity of the allegations influences subjects’ punitiveness. Recall that subjects in the control condition did not answer these questions. We therefore simply show the mean scores in both treatment conditions by shared partisanship, rather than coefficient estimates compared to a control baseline category (see Figure 4). We follow the advice of Julious (2004) and use 84% confidence intervals since we compare the two sexual harassment conditions to each other rather than evaluating whether the effects of the conditions differ from zero (though using 95% confidence intervals does not change the results)

Figure 4.

Mean agreement with punitive repercussions by shared partisanship.

Subjects of the opposing party in the assault condition had a mean punitive score of 3.2, while the mean score for subjects of the same party was 3.0 (−0.2, p<0.001). In the jokes condition, both opposite and same party subjects were less punitive compared to the assault condition (p<0.001). However, opposite party subjects were still significantly more punitive in this condition than same party subjects. Remarkably, opposite party subjects were only slightly less punitive for a legislator accused of making sexist jokes than same party subjects were for a legislator accused of sexual assault.

The results when we look at each item separately follow the same general pattern (with one exception). That is, subjects that share the legislator’s party were more likely to agree that no repercussions are necessary because “it was a long time ago” and “until there is more evidence.” They were also significantly less likely to agree that he should resign from office than opposite party subjects. The single exception to this pattern is whether subjects agree that the legislator should apologize for his alleged actions. The mean agreement on this statement is about 2.9 in both the assault and joke conditions, regardless whether subjects share the legislator’s party.

Effects conditional on sexism

So far, we have shown that individuals do react negatively to sexual misconduct scandals, especially when they involve politicians of the opposite party. Considering the large body of literature on how gender attitudes and stereotypes shape evaluations of legislator behavior, we additionally registered a set of hypotheses regarding the conditional effects of sexism. Are subjects that endorse sexist stereotypes about women less reactive to sexual harassment allegations?

Figures 5 and 6 show the marginal effect of both the assault and jokes conditions compared to the control across subjects’ pretest levels of sexism. ⁴ Both treatments produce a smaller decrease in favorability and likelihood of supporting reelection as sexism increases. In fact, the treatment effect approaches 0 as sexism increases. In other words, the most sexist subjects react about the same way to sexual assault and sexist jokes allegations as they do to the control news story about the legislator attending a county fair. With that being said, the effects are larger in magnitude in the sexual assault condition. While the treatment effect all but disappears for the subjects with the highest levels of sexism in both the assault and jokes conditions, subjects who still place high on the sexism scale (4 out of 5) decrease their favorability ratings by about 25% for the legislator accused of sexual assault but only 10% for the legislator accused of sexist jokes.

Figure 5.

Treatment effects on percent change in favorability conditional on sexism.

Figure 6.

Treatment effects on percent change in electoral support conditional on sexism.

Finally, we also hypothesized that subjects were less likely to agree with more punitive repercussions as their pretest levels of sexism increase. We find support for this expectation as well, as seen in Figure 7. As subjects’ sexism increases, the less likely they were to agree with more punitive repercussions for the legislator accused of either sexual assault or sexist jokes.

Figure 7.

Effect of sexism on mean agreement with punitive repercussions for legislator.

Discussion

In this article, we report a set of findings that contribute to understandings about how individuals react to allegations of sexual harassment against a politician. In particular, the evidence presented here makes two main contributions. First, there has been very limited research on how motivated reasoning operates in the context of #MeToo. Given the recent rise in awareness of sexual harassment and the varied public responses based on the type of allegation, how voters respond to such cases deserves attention in its own right.

Second, one reason for the relatively minimal understanding about voters’ reactions to sex scandals may be scholars’ focus on scandals that differ in kind (e.g. extramarital affairs vs financial fraud) rather than degree. By examining reactions to allegations of sexual assault compared to making sexist jokes, and by measuring both favorability and electoral support, we investigate whether there is a limit to the conditional effects of shared partisanship or sexist attitudes that many influential theories would predict.

The results show that there may be a limit with partisanship, but not for sexism. Even in the sexual assault condition, subjects who scored highest in sexism had a treatment effect that was statistically indistinguishable from 0. These findings suggest that in trying to understand how the #MeToo movement influences elections more broadly, we must consider the ways that sexism forms an ideology that serves to uphold and legitimize structural injustices (e.g. Jost et al., 2004). While both partisan motivated reasoning and sexism significantly shape how people respond to sexual harassment allegations, the prevalence of sexism may hinder acceptance of the #MeToo movement in a way that partisanship does not.

Supplemental Material

Online_Appendix – Supplemental material for How partisanship and sexism influence voters’ reactions to political #MeToo scandals

Supplemental material, Online_Appendix for How partisanship and sexism influence voters’ reactions to political #MeToo scandals by Mia Costa, Trevor Briggs, Ajaipal Chahal, Jonathan Fried, Rijul Garg, Sophia Kriz, Leo Lei, Anthony Milne and Jennah Slayton in Research & Politics