(Sex) Crime and Punishment in the #MeToo Era: How the Public Views Rape

Abstract

One of the core tasks of a well-functioning state is providing fair and adequate criminal justice. Recent events have raised concerns that the US exhibits a “culture of rape,” wherein victims are often disbelieved and blamed. Scholars have not yet examined how the public understands rape and how it should be punished, despite the important role that public pressure has played in the #MeToo era. We present an empirical conceptualization of rape culture to generate predictions for how various attributes of rape incidents affect the likelihood that they are perceived as punishable crimes. In a series of conjoint experiments, we demonstrate that details relating to the victim’s consent and credibility significantly decrease participants’ propensities to support reporting to police or to recommend a severe punishment for the perpetrator. The results show that emphasizing certain legally irrelevant features of rape strongly affect whether the public views an incident as severe or worthy of punishment.

Understanding criminal justice is key to comprehending contemporary politics. Limiting violence and dispensing justice among citizens, in turn, are among the most important roles of the state (Weber 1919). Consequently, the recent wave of high-profile rape and sexual abuse allegations in the context of the #MeToo movement has raised questions about whether and how well the state is able to address the crime of sexual assault, and the broader consequences for US society. Research has found a negative impact of crime victimization on trust in the criminal justice system, political institutions and satisfaction in democracy (Blanco and Ruiz 2013). Prominent sexual assault allegations—such as those against director Harvey Weinstein, actor Kevin Spacey, journalist Charlie Rose and Supreme Court Justice Brett Kavanaugh—have provoked significant debate about whether or not the U.S. is characterized by a “culture of rape,” wherein victims are disbelieved and blamed for their own violations.¹ Activists and advocates worry that such a social environment may not only enable rapists and intimidate victims but ultimately undermine the criminal justice system’s handling of sexual assault cases.

We argue that the public’s beliefs about rape determine their reactions to and interpretations of sexual crime. These, in turn, shape the current, critical era of social change around gender issues, which the New York Times has termed “the #MeToo moment,” and which the Economist has called “the defining movement of the Trump era.”² Public opinion and debate on rape culture have undeniably increased the stakes in contemporary U.S. politics. Between 2017 and 2018, nine members of Congress resigned or declined to run for re-election, and three congressional candidates lost or ended their campaigns, all after being accused of sexual misconduct. Yet the debates that surrounded the confirmation of Justice Kavanaugh in late 2018 indicated that the public remains deeply divided in its views on sexual misconduct and its severity.

Our analysis joins a burgeoning literature in political science that has explored how the public’s perception of complex issues can distort political and societal responses to those issues, often in manners inconsistent with legal frameworks. For instance, Huff and Kertzer (2018: 55) explore the public’s views on terrorism, arguing that “what ordinary citizens think terrorism is is a crucial prerequisite to understanding how they react to it” (emphasis in original). We focus on public perceptions of rape not because public opinion can offer insight into how rape ought to be viewed and sanctioned, but rather because it plays an outsized—but understudied—role in shaping how society perceives rape, and treats perpetrators and victims. Previous research shows that public opinion shapes criminal justice policy. It does so, according to Pickett (2019: 418), by creating “a range of acceptable policies.” Because policymakers follow trends in public opinion, when attitudes shift in a particular direction, policymakers become more sensitive to those opinions. Because the current trend in public opinion attitudes is toward heightened concern about sex crime, the attitudes examined in our experimental studies indicate the potential for a substantial influence on policy.

Beyond policymakers, ordinary citizens are frequently called upon to make decisions relevant to rape and its adjudication. These include rendering informal judgments, such as whether or not to report a rape incident to authorities, offering support to rape survivors, or disclosing the details of an attack to the news media. Members of the public can also take on more formal institutional roles, such as serving on juries tasked with deciding the fate of an accused perpetrator, or removing elected officials who made controversial decisions in rape cases. Through activism, protest and voting, the public can alter the extent to which political elites tolerate or condone rape and sexual misconduct.³

In short, public perceptions about rape shape how the US political system deals with sexual violence, including whether or not such cases enter the criminal justice system, and how they are treated once they do so. As a result, the public’s biases critically influence how well the criminal justice system serves rape survivors. In this way, our work extends foundational studies of how well public institutions represent constituents—and whether they act in their best interests—in contexts ranging from welfare offices (Soss et al. 2008) to state legislators (Butler and Broockman 2011) and election officials (White et al. 2015), to decision-making by judges (Kastellec 2013).

To analyze the effects of the public’s perceptions of rape, we conducted a series of conjoint experiments involving approximately 5400 US adults.⁴ We supplemented our main study—a nationally representative survey from NORC's AmeriSpeak panel—with additional studies on convenience samples recruited on Amazon Mechanical Turk (MTurk), yielding both experimental and qualitative data. We show that ordinary people understand rape not only based on objective facts, but also on numerous subjective factors relating to the perpetrator, the victim and the context of the assault. We find that these subjective, legally superfluous case details substantially influence decisions related to the reporting and adjudication of rape cases. When provided with information related to underlying notions of consent and the victim’s credibility, such as a victim’s prior relationship to the perpetrator, seemingly promiscuous sexual history, or prior felony record, survey participants were significantly less likely to recommend a case for reporting to the police, all else equal. The sex of the victim also mattered: respondents were far less likely to choose for reporting to the police cases involving male, rather than female, victims. Additionally, participants were significantly less likely to recommend a case for a more severe punishment when provided details related to two key features of rape culture, victim-blaming and credibility, such as information about the incident location (a party or a public park) and a victim’s sexual history.

In the analysis that follows, we first offer an original empirical conceptualization of rape culture. Based on extensive interviews with experts, we identify four key, measurable features of rape culture: victim blaming, empathizing with perpetrators, assuming the victim’s consent, and questioning the victim’s credibility. Based on this framework, we design an experiment to test the influence of people’s perceptions of rape cases on criminal justice. Specifically, we examine how exposure to details that are legally irrelevant to determining guilt, but that observers and advocacy groups have identified as invoking rape culture beliefs, influence how people perceive such cases.⁵ We focus on two key stages of the criminal justice process when the public’s influence is potentially greatest: crime reporting⁶ and sanctioning.⁷ We then offer additional robustness checks to determine whether any of the observed biases are unique to rape cases or extend to other violent crimes such as armed robbery cases. We also test whether treatment effects vary across different demographic groups, and present qualitative data from open-ended questions to corroborate our main findings. We conclude by discussing the implications of our findings for the study of rape culture in American politics.

A Unifying Empirical Framework of Rape Culture

Scholars and activists have defined rape culture broadly as “a set of values and beliefs that provide an environment conducive to rape” (Boswell and Spade 1996), where “rape is often not acknowledged as a crime and its victims are frequently blamed for their own violation” (Vogelman 1990). Two mechanisms are central to this conception. The first is a distortion of perceptions of consent such that rape is neither considered a forcible act nor acknowledged as a violent crime. The second is a skewing of attributions of blame, by shifting personal responsibility away from the perpetrator and toward the victim (Bradbury and Fincham 1990).

Both scholarly and public debates converge around four main dimensions of rape culture, beliefs and values that contribute to an “environment conducive to rape”: (1) Victim blaming, defined as attributing to the victim personal responsibility for the assault (e.g., Suarez and Gadalla 2007); (2) empathizing with perpetrators, wherein blame is deflected away from the offender (e.g., Smith and Frieze 2003); (3) assuming the victim’s consent when none was given (e.g., Swauger et al. 2013); and questioning the victim’s credibility, often by focusing on unrelated life decisions or personality traits of the victim, as a means of portraying the victim as not trustworthy and retroactively implying the victim’s consent (e.g., Ellison and Munro 2009).

Many of these patterns feature prominently in the public discourse surrounding recent controversial rape cases, but they tend to be more prevalent in some cases than others. To derive a set of case features most likely to invoke these four main dimensions of rape culture, we solicited detailed feedback from two dozen experts, including academic researchers, journalists, activists, and policy practitioners. We employed several of the most commonly cited case features to design our experimental treatments (see the Online Appendix, Section 1) and drew on Baum et al. (2018) for the components identified in our investigations and those selected for inclusion in this study.

The Effect of Case-Level Characteristics on Perceptions and Handling of Rape Cases

A variety of legal and extra-legal factors may distort the investigation and adjudication of crimes in general and rape cases in particular. For this study, we focus on the effects of legally irrelevant case parameters on decision-making during the reporting and adjudication stages of the criminal justice process.⁸ In particular, we explore how cultural beliefs about rape affect reporting of the crime of rape, and whether false assumptions about rape and its victims alter perceptions of what constitutes an adequate punishment.

Racial Prejudice and Rape

Much of the previous work suggests that the race and ethnicity of both the rape victim and the perpetrator may affect outcomes of criminal justice processes, including rape trials (Wuensch et al. 2002; Feild 1979). Considerations of race and ethnicity often evoke complex emotions and represent power structures deeply entrenched in American society, potentially leading to discrimination. With historical roots in slavery and beliefs about racial supremacy, the rape of a white woman by a black defendant has traditionally been interpreted as a “theft of white men’s ‘property’” (Feild 1979: 263; Brownmiller 1975). As one respondent wrote in our qualitative study, “since the perp was of black origin I would say he’s probably more guilty.” This literature suggests that cases involving black perpetrators will be more likely to be recommended for reporting or punishment than cases involving a white perpetrator, all else equal. On the other hand, recent research (Chudy 2020: 4) has found that racial sympathy, a racial attitude that captures “white distress over black suffering,” can activate opposition to policies thought to harm blacks. For example, one participant in our qualitative study explained that he had selected a case with a white perpetrator rather than a black perpetrator “because black men are treated unfairly and unjustly systematically.” This research suggests an alternative prediction: that cases involving black perpetrators will be less likely to be recommended for reporting or punishment than cases involving a white perpetrator, all else equal.

Similarly, researchers have suggested that rape cases involving black victims tend to be punished less harshly than those involving white victims because rape reports by black women are often considered suspect (Vrij and Fischer 1997), reflecting a sense that “they may have received just what they wanted” (Feild 1979: 264). This prejudice was apparent in the trial of former Oklahoma City police officer Daniel Holtzclaw, who was convicted of raping or sexually assaulting thirteen African–American women, all with criminal histories. Holtzclaw reportedly targeted this population because he believed them to be especially unlikely to report the crimes. One victim stated at trial, “I didn't think anyone would believe me. I'm a black female.”⁹ From this, we hypothesize that participants will be less likely to recommend for reporting or punishment cases involving black victims than cases involving white victims, all else equal.

Conversely, drawing again on the research on racial sympathy, research has found evidence of support for policies perceived to benefit blacks. Again, respondents in our qualitative study expressed these views explicitly. For example, a participant selected a case with a black victim rather than a white victim because “the black woman needs representation more so than a white man would since racial minorities usually do not have the same recourse to law that white people enjoy.” Based on this, we offer a competing hypothesis, that participants will be more likely to recommend for reporting or punishment cases involving black victims than cases involving white victims, all else equal.

Class, Socioeconomic Status and Rape

Socioeconomic status (SES) can also affect how people perceive sex crimes. However, the psychological mechanisms proposed by scholars differ from those associated with race. A high SES is “interpreted as an achieved rather than an ascribed status and thus, unlike race, as a reflection of the character of the defendant” (Gleason and Harris 1975). In other words, male suspects with high SES are often assumed to be “desirable” to women, such that rape seems out of character—or “unnecessary”—for them. As a result, cases involving a defendant with a high SES in their communities tend to be judged less harshly (Black and Gold 2008; Gleason and Harris 1975). Relatedly, women are sometimes accused of fabricating accusations of rape by high SES perpetrators in order to gain money, publicity or status. For instance, the media debates surrounding the rape allegations and financial settlement by a hotel maid against former managing director of the IMF Dominique Strauss-Kahn are consistent with this latter conjecture (Saletan 2011). In addition, President Trump spoke about the 18 women who have accused him of sexual misconduct, highlighting ulterior motivations: “I’ve had a lot of false charges made against me. I’m a very famous person, unfortunately…People want fame, they want money, they want whatever.”¹⁰ As a respondent wrote in our qualitative study, “…owning a business puts a person in a higher level of authority. They might feel as though they can do whatever they want to.” We hypothesize that participants are less likely to recommend for reporting or punishment, and to perceive as less severe, rape cases involving high SES perpetrators relative to cases involving low SES perpetrators, all else equal.

Gender Roles, Socio-Sexual Behavior and Rape

Gender role expectations also shape perceptions of a victim’s credibility and notions of consent. One of the most commonly identified factors that can undermine a rape victim’s credibility in the court of public opinion—and in the courtroom—is past sexual history. Individuals tend to judge women who have had multiple sexual partners as promiscuous, improper and unchaste (Feild 1979; Viki and Abrams 2002). Women with an active sexual history are often less likely to be believed when they report a rape and are more likely to be blamed for what happened to them (Schult and Schneider 1991; Viki and Abrams 2002).¹¹ Data from our qualitative study revealed that some respondents shared these biases. As one respondent wrote, “The victim has had multiple partners so that implies she is more than average open to physicality.” Another wrote, “I chose case number 1 over case number 2 because the person in case number 2 had multiple partners and probably wouldn’t feel as awful as the person in case number 1.” Finally, another respondent wrote, “Victim 1 has a track record of sleeping around so she may have hooked up with the assailant before, or he didn’t do what she asked and she’s lying to get him in trouble.”

Similarly, details about the victim’s past criminal record may lower her credibility and as a result, the likelihood that a victim reports a given case and that the perpetrator is punished.¹² One respondent noted, the “victim could be lying. She has a record and so you know she’s not perfect.” Another stated, “she has a criminal record so she might not be the most moral in telling the truth.”

Further, the relationship between the victim and perpetrator prior to the incident appears crucial to perceptions of consent. Although research suggests that 78% of all sexual violence occurs between acquaintances (Department of Justice 2013: 1), many people still conceptualize rape as stranger-on-stranger violations (Abrams et al. 2003; Swauger et al. 2013; Mancini and Pickett 2016) and perceive such cases as more severe. As one participant commented, “a crime against a stranger is a crime against all of us.”¹³ By contrast, rape committed by a friend or a partner evokes notions of implicit consent, and observers often do not consider these acts serious crimes since the victim had previously agreed to a relationship, and potentially also sex (Monson et al. 2000). Along these lines, a participant stated, “[victim and perpetrator] were acquaintances so one might have a question as to whether they were romantically involved and the charges were bogus.” Thus, we hypothesize that participants will be less likely to recommend for reporting or punishment cases involving acquaintances of the victim.

Socio-sexual behavioral norms may also play an important role in attributions of blame in the context of rape. Contemporary notions of masculinity suggest that the male sex drive is natural and uncontrollable, and that “boys will be boys” (Schrock and Schwalbe 2009). Sexually suggestive behavior by a woman, such as flirting or dressing provocatively, may be interpreted as an open invitation to sex. When a rape victim’s behavior or appearance are deemed sexually provocative, observers may assume that she “asked for it,” shifting responsibility away from the perpetrator and onto the victim, while also stripping rape of its violence. Indeed, evidence suggests that a victim’s perceived physical attractiveness or dress at the time of the incident may lead to distortions in individuals’ judgments about blameworthiness and incident severity (Whatley 2005; Ellison and Munro 2009). Responses to our qualitative study included numerous examples of such perceptions. As one participant observed, “The victim (…) was wearing a club outfit which can give the wrong impression and can leave a person open to being attractive [sic].” Another wrote, by way of explaining why they had selected a case in which the victim was wearing a work outfit over another in which the victim wore a clubbing outfit: “She was not in a place that would encourage rape, and she was not wearing any clothes that might get a man aroused.”

Building on classical feminist literature that distinguishes between public and private spaces, and women’s proper place in the latter (Brownmiller 1975; Benedict 1992), the “good girls don’t get raped” narrative may also apply when an incident takes place in a location that suggests risk-taking or “contributory negligence” on the part of the victim (Bryden and Lengnick 1997: 1333). In other words, victims of rapes that occurred in locations where “well-behaved” women would not choose to spend time (at a nightclub or in a park late at night) tend to be held more individually responsible (Feild 1979; Bryden and Lengnick 1997). We hypothesize that participants will be less likely to support reporting or recommending for a more severe punishment incidents that took place in locations that might suggest negligence and risk-taking on part of the victim, all else equal.

Finally, we consider the sex of the victim. Few existing studies systematically explore how the criminal justice system perceives and treats victims of male-on-male rape.¹⁴ Male-on-male rape remains stigmatized and is complicated by persistent norms of masculinity: in the eyes of many, “real men” do not get raped, and the men who are raped are at fault for not adequately protecting themselves or fighting back. Public opinion research has found that most people believe that typical sex crime victims are vulnerable women and children (Mancini and Pickett 2016). As one of our study participants wrote, “A female victim gets a little more sympathy and favor. And while I know it isn’t right, the male victim may/should be better able to defend himself in the attack.” Another respondent noted how implausible male-on-male rape seemed to him: “None of these cases are realistic as far as I’m concerned. What kind of guy is going to let himself be raped by another guy?” We therefore hypothesize that our participants will be less likely to recommend for reporting or punishment cases involving male victims of rape, and will perceive such cases as less severe, all else equal.

Table 1 summarizes our hypotheses, organized by their applicability to the four rape culture dimensions. While these dimensions help to organize the main components of rape culture, they are not mutually exclusive. For instance, a victim who was dressed in a clubbing outfit may be blamed for the attack. As one participant wrote in our follow up study, “Although it doesn’t excuse the crime, wearing a night club outfit in a park is just asking for trouble.” Alternatively, a clubbing outfit may be perceived as giving consent. Another participant wrote, “the defense will say that [the clubbing outfit] suggested she was interested in sexual activity. This doesn’t make it right by any means, but is unfortunately the world we live in.”

Table 1 Main case-level hypotheses, by rape culture dimension

Research Design & Estimation Strategy

Independent Variables and Conjoint Design

We employ a choice-based conjoint experimental design to test the effects of legally irrelevant case details on two outcomes related to decision-making at different stages in the criminal justice system: reporting and punishment of a crime. Absent biased beliefs about rape, its victims, and its context, such details should be irrelevant for decisions about reporting or the level of punishment.

In our main study, we randomly assigned participants to review five pairs of either rape or armed robbery cases. The case profiles presented to participants are composed of nine legally irrelevant case attributes that we hypothesized would potentially invoke biased perceptions of rape cases. These vignettes provide information about the victim and perpetrator, as well as the circumstances of the incident. For victim characteristics, we include details on sex, race, past relationship history, and criminal record. In order to create a case narrative similar to those with which participants would be confronted in news reports or in court proceedings, we aggregate case attributes into three broader categories: details on the victim, the perpetrator and the circumstances of the incident. For the perpetrator, we include details on race, socioeconomic status (as signaled by profession) as well as his prior relationship to the victim. Finally, we present a set of circumstantial details, such as the incident location and the victim’s attire at the time of the incident.¹⁵ See Table 2 for a list of attributes and their values.

Table 2 Attributes in case profiles

We randomly manipulate attribute values for each profile, allowing us to simultaneously test the effects of a number of independent variables on our outcome measures. In order to address any attribute ordering effects, we also randomize the sequence in which the attributes appear in the profiles. We assign all attribute values with equal probability, with the exception of sex. All perpetrators were male, reflecting the reality that 97% of those arrested for rape in the US are male,¹⁶ Conversely, 70% of victims were female and 30% were male, mirroring the fact that women comprise a majority of rape victims. In combination, the nine case components and corresponding attribute values allow for 1728 different profiles, representing a large bandwidth of rape and armed robbery cases as they occur in real-world settings (e.g. acquaintance and stranger rape, armed robberies at home and outside the home).¹⁷

Dependent Variables: Crime Reporting and Punishment

In two different conditions of the experiment, we measure choice-based outcomes pertaining to different decision-making scenarios that are particularly relevant in the context of rape incidents. In both conditions, we randomly assign participants to assess and compare either five pairs of rape or five pairs of armed robbery cases. In the first study, participants review pairs of criminal cases and then choose the case in which they would be more likely to advise the victim to report the incident to the police. In the second study, participants receive a prompt asking them to imagine that they have been summoned and selected for jury duty and that the perpetrator has been found guilty. We then ask them which of the two cases they would recommend for more severe punishment.

Qualitative Evidence

To verify that our results are not artifacts of the conjoint methodology but rather reflect genuine bias in individuals’ reasoning, we conducted an additional follow-up study on MTurk, fielded in May 2018. Similar to our AmeriSpeak study, we assigned MTurk participants to review five pairs of randomly generated rape cases before asking them which case they would recommend for reporting for a more severe punishment. We then asked participants to explain their choice in two or three sentences via an open-ended question. This generated a wealth of qualitative data.¹⁸

We coded the open-ended responses along the four major themes of rape culture: victim-blaming, empathy (or lack thereof) for the perpetrator, doubts about the victim’s trustworthiness, and false notions of consent. In addition, we used a five-cluster implementation of the Latent Dirichlet Allocation (LDA) as an unsupervised method to model topics and themes that emerge from the qualitative data (Blei et al. 2003).

Control Measures

Following the experiment, all participants answered a standard repertoire of demographic questions as well as a number of attitudinal questions. In particular, we used a self-reported measure of participants’ political ideology to estimate heterogeneous treatment effects.¹⁹

Estimation Strategy

To estimate the marginal effects of each of the nine case attributes, we follow the statistical approach for conjoint experiments proposed by Hainmueller et al. (2014). Conjoint experimental designs offer an alternative to widely used vignette techniques, allowing us to consider a large number of potential profiles to which participants could potentially be exposed. Because we randomly assign case attributes for each profile, we can estimate the average marginal component effect (AMCE) of each attribute over the average distribution of all other attributes, regardless of whether participants review every potential profile.

The AMCE represents the average difference in the likelihood that participants choose a given rape or armed robbery case for reporting or punishment when comparing two attribute levels (black vs. white victim). We estimate the AMCEs using a benchmark OLS regression,²⁰ including an outcome indicator for whether a participant selected a given profile for reporting or punishment and factor variables for each of the attributes (excluding reference categories). The beta-coefficients on these factor levels represent the component-specific effects (AMCEs) of each attribute value on the outcome of interest. For example, the AMCE for the attribute indicating a victim’s race represents the average difference in the probability of a participant selecting a case for reporting or a more severe punishment if the victim is black compared to white, where the average is computed for all other possible combinations of the other attributes included in the model. Because each respondent reviewed multiple case pairs, we follow standard practice and cluster standard errors at the participant level to account for individual-level biases.²¹

In addition to AMCEs of case-level attributes, we are also interested in differences across participant-level social values, here measured by self-declared political ideology. We add participant-level interactions between case characteristics to our benchmark regression models. Lastly, to determine whether any observed biases are unique to the context of rape, we add crime-level interactions to test whether there are differential effects of the case components on our outcomes of interest, depending on whether participants review rape or armed robbery cases.²²

Benefits of Conjoint Experimental Design

Using a conjoint design to study implicit biases in decision-making in the context of rape may raise a number of concerns. First are issues of external validity and experimental realism: do participants’ decisions in an online survey experiment accurately reflect how they would respond in real life? While most survey experiments are by design simplifications of real-world behavior, previous research has shown that conjoint experiments in particular can produce externally valid results (Hainmueller et al. 2015). In our study, we offer participants decision-making scenarios that are related to real-world situations they might encounter, such as advising friends to report a crime or making difficult choices when serving on juries. In addition, our research design allows us to observe hypothetical decisions rather than merely collecting self-reported attitudes, which are subject to social desirability biases—especially on sensitive topics like rape.

Second, our design choice requires participants to select one of two cases rather than allowing them to pick both (or neither). In the real world, participants would almost never have to weigh crime cases against each other. Even if forced to do so, they might choose both cases for reporting or severe punishment, or neither of them. However, the benefit of conjoint experiments is that they allow researchers to tease out underlying, or latent, preferences that influence individuals’ decision-making. Forcing participants to decide between two cases reveals the subtle distinctions participants make, which would otherwise go undetected if they were to evaluate one case at a time, or if they could select both cases, or neither case.²³ To be clear, in a state of the world devoid of rape culture, we should not observe a systematic preference for certain types of rape scenarios over others. Rather, we would expect individuals to pick between cases at random, yielding no statistically significant differences between scenarios. For example, a participant may consciously understand that the crime of rape is similarly severe regardless of whether the victim was wearing business attire or a clubbing outfit. So, given the option of indicating that there is no difference between two scenarios in which the only variation concerned the victim’s attire, that participant would likely do so. However, if forced to choose one or the other case, bias is the only logical reason for participants in the aggregate to systematically rate a case involving a victim in business attire as more worthy of reporting or a severe punishment than a case involving a victim wearing a clubbing outfit. If we find a statistically significant difference, this suggests the presence of systematic latent bias.

We further substantiate our design choice with a second follow-up study fielded on MTurk in April 2019 (N = 2010) where respondents were assigned to one of four conditions, in which they: (1) evaluated a single case at a time, deciding whether or not to recommend the given case for reporting or severe punishment (i.e., single case—binary choice); (2) reviewed a single case at a time and for each case, rated how likely they would be to recommend the case for reporting or severe punishment (i.e., single case—rating); (3) assessed two cases at a time, choosing which case to report or severely punish (i.e., two cases—binary choice, as in the main study); or (4) reviewed two cases at a time and for each case, rated how likely they would be to recommend this case for reporting or for severe punishment (i.e., two cases—rating).

We find that no matter how the outcome variable is measured, the patterns we observe in our main study replicate in many of these conditions. However, the effect sizes are largest and most significant when we forced respondents to choose between two cases or when we gave them a binary choice to report an incident, or not, or severely punish a perpetrator, or not. Again, this is what we would expect since the binary choice—employed in our main study—is intentionally designed to tease out underlying biases that might go unnoticed were respondents offered rating-based answer options. Our qualitative response data collected in a separate MTurk study further highlight this important function of our binary-choice design. Several participants acknowledged that they found the cases very similar, and therefore had difficulty deciding between them, only to subsequently provide a justification for their selections that align well with our predictions. As one participant wrote, “I picked the first one because this person wasn’t trying to be sexy or draw attention to herself [in contrast to the clothing of the victim in the other case scenario]. Both deserve justice but if i [sic] had to pick one it would be the first.”²⁴ These replications, combined with the qualitative data, offer strong additional empirical support for our binary-choice design strategy and its appropriateness for the context of adjudicating rape cases.²⁵ Absent this design, we would almost certainly underestimate the extent of rape culture bias among participants. Conversely, employing this design should not result in an overestimate of rape culture, since, as noted, absent bias participants’ choices should be randomly distributed.

Analysis and Results

Data and Sample Size

We fielded a nationally representative survey of 1012 US adults as part of NORC’s AmeriSpeak panel in December 2017.²⁶ We randomly assigned 60% of participants to review rape cases while the remaining 40% reviewed armed robbery incidents. Each participant reviewed ten cases, bringing the total number of observations (cases reviewed) to 10,120. Just over half of survey participants were female, and about two-thirds were white. Over half self-identified as Democrats or leaning towards the Democratic party. Age groups were roughly equally represented, as were regional groups, with slightly more participants residing in the South. Nearly all participants had at least a high school degree, and roughly one-third had earned a BA or higher degree. Throughout our analysis, we use the survey weights provided by NORC to make the sample representative of the U.S. population.²⁷

Main Findings

We summarize our main results in Fig. 1.²⁸ The coefficient plots display the effects of individual attribute values on the probability that survey participants selected a given rape case for reporting (reporting condition) or for a more severe punishment (punishment condition). All estimates are based on our benchmark OLS model with standard errors clustered on the individual participants. In this and all subsequent figures, the lines surrounding each point estimate indicate 95% confidence intervals. The reference category to which we compare a given attribute level is in parentheses. We present in solid lines confidence intervals for only those estimates that are statistically significant at the 0.05-level; otherwise, we use a dashed line. As predicted, we find that legally irrelevant details about the circumstances of the incident matter a great deal for individuals’ decisions to recommend a given case for reporting or a more severe punishment.

Fig. 1

Effect of case attributes on decisions related to rape reporting and punishment. Note: Coefficient plots display the estimated effects of randomly assigned case details on the probability that a case is recommended for reporting (left panel) or selected for more severe punishment (right panel)

Beginning with victim blaming, we find strong evidence for H2, H3 and H4. Incident location (H2) yielded large effects: incidents at a party were about 6 percentage points less likely to be recommended for reporting (SE = 0.031, p < 0.10). In the punishment condition, the effect was even stronger: cases that took place at a party were 17.6 percentage points (SE = 0.030) less likely to be chosen for a more severe punishment than those occurring in the victim’s home. In addition, participants were 10.4 percentage points less likely to select cases for a more severe punishment when the incident took place in a park (SE = 0.035).

Participants chose cases for reporting 4.8 percentage points less often when the victim was described as wearing a night club outfit rather than work attire (H3) (SE = 0.027, p < 0.10). Lastly, the sex of the victim (H4) proved to be one of the most salient factors influencing individuals’ decision-making. Participants were 16.4 and 5.6 percentage points more likely to recommend cases with female victims for reporting or punishment, respectively, than cases involving male victims (SE = 0.032 and SE = 0.030, respectively). This suggests that many respondents perceive male-on-male rape as a less severe violation.

Turning to empathy for the perpetrator (H5 and H6), here we find no support. Perpetrator SES (H6) did not meaningfully influence participants’ decision-making. With respect to implications of consent (H7), we find that information related to the perpetrator’s previous relationship to the victim decreases participants’ likelihood of recommending a given case for reporting. Knowing that the victim and perpetrator were acquaintances reduces participants’ likelihood of choosing a case for reporting by 11.8 percentage points (SE = 0.024). In addition, participants were around 9 points less likely to recommend for both reporting and a more severe punishment cases involving victims with multiple sexual partners (SE = 0.027 and SE = 0.030, respectively), and 5.4 percentage points less likely to reporting cases involving single victims (SE = 0.028, p < 0.10), relative to married victims (H8).

We turn finally to case details related to the victim’s credibility (H9). Participants were 6.4 points less likely to select for reporting victims with a felony record (SE = 0.027), compared to victims with no record (H9).

Surprisingly, we found little evidence of racial bias (H1 and H5), either in terms of victim blaming or empathy for the perpetrator. Participants did not systematically discriminate against black victims (H1a) or perpetrators (H5a). Rather, we found evidence of racial sympathy: they tended to be more favorable towards black victims (H1b) and perpetrators (H5b). In the punishment condition, for example, cases involving black victims were 4.7 percentage points more likely to be recommended for more severe punishment, compared to cases with white victims (SE = 0.024, p < 0.10). By contrast, participants were 5.0 and 6.2 percentage points less likely to select cases with black rather than white perpetrators for reporting or punishment, respectively (SE = 0.025 and SE = 0.024, respectively).

These estimates suggest a pattern of racial sympathy that we also explicitly observed in the previously described MTurk iteration of this experiment (see Online Appendix, section 4). Such responses may reflect a growing public awareness of racial biases in the criminal justice system, brought to the fore through social movements like Black Lives Matter, and echo recent scholarship that finds increasing evidence of racial sympathy (Chudy 2020). The results also align with Lancaster’s research on sex panics, which provides a nuanced interpretation of the role of racial bias in evaluations of rape perpetrators. Lancaster (2011: 92) argues “in modern sex panics, the black man is not necessarily the imagined perpetrator, the statistically preponderant object of fear and loathing.”

In addition, it is noteworthy that treatment effects of certain case attributes seem to differ across the reporting and punishment conditions. In particular, participants are significantly more likely to select incidents that occurred at a party or in a park when they are assigned to the reporting rather than the punishment condition. By contrast, individuals are significantly less likely to select acquaintance rape cases in the reporting than in the punishment condition (although the effect of seeing an acquaintance relative to a stranger rape case is negative in both conditions). While we did not have theoretical priors concerning differences in response patterns across choice tasks, these differences might arise due to their slightly different contexts. In particular, we explicitly instructed participants in the punishment condition, but not in the reporting condition, that they should assume the perpetrator is found guilty before they make their choice. Overall, however, the directions of the effects are largely comparable across these two conditions.

To better understand the substantive meaning of our findings, we predicted the probabilities of a case profile being recommended for reporting or more severe punishment based on our OLS regression models with clustered standard errors. Figure 2 displays the likelihood of reporting or punishment for selected case profiles in the 1st, 25th, 50th, 75th, and 99th percentiles of the probability distribution.

Fig. 2

Predicted probabilities of being chosen for reporting or severe punishment for selected case profiles

In the reporting condition—represented in the first percentile, with only 18.4% (SE = 0.048) likelihood of being reported—are cases that took place at a party, involving a black male victim who has multiple partners and has a minor felony record, as well as a black perpetrator who owns a business and was acquainted with the victim. At the other extreme, occupying the 99th percentile, are incidents that took place at a park and involved a black female victim, who is married, has no criminal record, and did not know the perpetrator. Such a case is likely to be reported with a 71.6% probability (SE = 0.041). Similarly, in the punishment condition, cases that took place at a party between acquaintances and involved a victim with a felony record and multiple sexual partners are over 40 percentage points less likely to be recommended for a severe punishment than incidents involving married victims who were assaulted by a stranger in their own home (predicted probabilities of 24.1 (SE = 0.038) and 68.6% (SE = 0.048), respectively).

Profiles

Overall, our results demonstrate that some of the case-, perpetrator- and victim-specific factors that affect people’s desire to report and to punish rape induce key aspects of rape culture. These findings provide evidence that exposing individuals to legally irrelevant case details can bias decisions regarding the reporting and punishment of sex crimes in important ways. However, some case details—invoking different dimensions of rape culture—matter more than others. In particular, information related to notions of consent and credibility (H7 and H8), such as the victim’s sexual history and the perpetrator’s relationship to the victim, influence participants’ likelihood of recommending a given case for reporting and punishment. Case details related to victim blaming, such as the sex of the victim and the location he or she was in at the time of the incident (H2 and H4), also affect criminal justice-related decisions.

Robustness Checks

Contrasting Rape to Robbery

While our findings confirm many of our hypotheses, it remains unclear whether the observed biases are unique to the context of sex crimes. To address this question, we randomly assigned 40% of our participants to review cases of armed robbery. We generated case profiles from the identical array of case attributes, and we asked participants to choose the case they would recommend for reporting or more severe punishment.

Our results, shown in Fig. 3 (below), indicate that those legally irrelevant case details that influence reporting and punishment decisions among our participants for rape cases matter less for armed robbery. In the latter case, only incidents that took place at a party are significantly (by 14.7 percentage points) less likely to be recommended for reporting (SE = 0.037). In addition, participants were 6.5 points (SE = 0.026) less likely to select cases for reporting when the perpetrator was identified as black, again providing evidence of racial sympathy.

Fig. 3

Effects of case attributes on reporting and adjudication decisions, conditional on violent crime type

Similarly, participants were far less likely to recommend a case for punishment when the robbery took place at a party or in a park (by 25 percentage points (SE = 0.048) and 21 percentage points (SE = 0.046), respectively). Again, respondents were 7.2 points less likely to select cases involving black perpetrators (SE = 0.032). Cases involving female victims, in turn, were 14.7 percentage points more likely to be chosen for more severe punishment (SE = 0.029). By contrast, case details related to the victim’s trustworthiness or to notions of consent, which were influential in the context of rape cases, do not appear to matter for reporting or sanctioning of robbery cases.

To determine whether rape and robbery cases are treated differently overall, we conducted additional tests on nested models for the reporting and punishment conditions. Specifically, we estimated OLS regressions without and with interaction terms for rape (see also Online Appendix, sections 2.4.3 and 2.4.4). We then conducted an F-test to compare whether these nested models were significantly different from each other. For both the reporting and punishment conditions, participants indeed viewed rape incidents as statistically significantly distinct from robbery cases overall.²⁹

Causal Effects of Case Attributes by Partisan Ideology

We have argued that biases and distorted beliefs about rape and its victims are rooted, in part, in traditional gender role expectations (Viki and Abrams 2002; Whatley 2005). If such biases indeed structure decision-making related to the adjudication of rape cases, we would expect these effects to be strongest among subgroups that hold more traditional values. One implication is that self-described conservative participants should be less likely than their self-described liberal counterparts to select for reporting or a more severe punishment gender-role defying victims and unconventional scenarios (Berkowitz and Lutterman 1968; Tetlock et al. 1989; Wetherall et al. 2013).

Our study used respondents’ political ideology as a proxy for the extent to which participants likely adhere to traditional social values.³⁰ In Fig. 4, we consider the causal effects of case attributes on preferences separately for subgroups of liberal, conservative, and moderate respondents, while using joint models with treatment-covariate interactions to determine differential effects across groups.³¹ In fact, we find that conservative participants respond differentially to case characteristics that invoke notions of consent and trustworthiness of the victim. In the reporting and punishment conditions, conservatives were 9.7 and 10.9 percentage points less likely than liberals, respectively, to choose a case when the victim was portrayed as having multiple partners, relative to cases with married victims. Similarly, in the punishment condition, conservatives were 12.9 points less likely than liberals to select a case when the victim and perpetrator knew each other. An F-test on the baseline model and the interaction model found that in both the reporting and the punishment conditions, liberals and conservatives, overall, handled rape cases differently depending on their personal self-reported ideology.³² Together, these findings help corroborate our claim that the biases we observe are indeed rooted in cultural beliefs and values.³³

Fig. 4

Effects of case attributes on reporting and adjudication decisions, by participant ideology

Measuring Bias Qualitatively

To further assess the extent to which the reasoning behind participants’ decision-making was consistent with our theoretical expectations, we conducted a follow-up survey experiment using a sample of 507 MTurk participants. Similar to our main study, we assigned participants to review pairs of randomly generated rape cases before asking them to decide which case they would recommend for reporting or a more severe punishment. We then asked participants to explain their choice in two or three sentences via an open-ended question. We collected a total of 1521 open-ended responses which we coded along the four major dimensions of rape culture: victim-blaming, empathy (or lack thereof) for the perpetrator, doubts about the victim’s trustworthiness, and false notions of consent.³⁴

With respect to the conjoint analysis, the results from the follow-up study align closely with our main findings (see Fig. 5 and Table 3). In the reporting condition, participants again preferred female to male victims to a large degree. They were also less likely to recommend a case for reporting when the victim and the perpetrator knew each other or when the victim had a felony record. In the punishment condition, participants showed the same preference for cases involving female victims. In addition, they were less likely to recommend cases for a more severe punishment when the incident took place at a party or in a park, if the perpetrator was black, and when the victim was wearing a night club outfit at the time of the incident.

Fig. 5

Main effects for reporting punishment conditions (MTurk Follow-Up Study) Note: Coefficient plots display the estimated effects of randomly assigned case details on the probability that a case is chosen in the reporting (left panel) and punishment condition (right panel)

Table 3 Rape culture elements in the open-ended survey responses

In both the reporting and punishment conditions, the majority of comments related to victim blaming (60.1% and 55.3%, respectively) referenced the location of the incident, often arguing that attacks in the victim’s home are more severe, especially when assaulted by a stranger. Also striking is the extent to which male-on-male rape is downplayed. In both conditions, about one-fifth of all comments that were coded as victim blaming language assert that cases with female victims are more severe and that male victims ought to be able to defend themselves. In addition, more than half of the comments that were coded as questioning the victim’s credibility mentioned her or his criminal record, indicating that victims with a minor felony record are less believable.

Anecdotal evidence supporting the face validity of our treatments—that is, the absence of legally relevant distinctions between the case profiles—also emerged in the qualitative responses. Some participants recognized that the cases they reviewed were very similar and acknowledged that the decision was difficult for them. In about 6.0% and 10.6% of responses in the reporting and punishment conditions, respectively, participants refused to distinguish between the cases. As one participant put it, “ ‘Gender’ is irrelevant. ‘Race’ is irrelevant. Being a ‘Stranger’ is irrelevant. What they were wearing is irrelevant. A person’s record is irrelevant. None of it has to do with whether someone was or wasn’t assaulted and whether it should be reported. If you were assaulted, then you report it, its [sic] very simple.”

In addition to the coding, we follow Hainmueller and Hopkins (2015) and employ a five-cluster implementation of Latent Dirichlet Allocation (LDA) using the R package topicmodels (Gruen and Hornik 2011). The findings from this unsupervised analysis of open-ended survey responses, displayed below in Fig. 6, support the results from the conjoint analysis. For example, in the reporting condition, clusters 1 and 5 relate to circumstances of the incidents, with respondents discussing both the location of the assault and the outfit of the victim. By contrast, cluster 2 represents comments that are related to the sex of the victim, with words like “less,” “likely” and “believe” indicating many participants’ beliefs that male-on-male rape incidents are less severe and less believable. Cluster 4 relates to questions about the victim’s credibility, with references to the criminal record and felony charges of the victim. In the punishment condition, we observe similar patterns, with the location of the incident and the outfit of the victim featuring in clusters 1 and 3. Cluster 4 once again hints at minimizing male-on-male rape, as we observed both in the conjoint data and from qualitative responses. Finally, cluster 5 references terms related to the victim’s credibility, such as “felony” or “criminal record.”

Fig. 6

Topic clusters based on latent Dirichlet allocation (MTurk Follow-Up Study)

Conclusion

In this study, we sought to delineate the features of rape culture in contemporary American society. We proposed and tested an empirical conceptualization of the components of rape culture using a conjoint experiment designed to explore whether legally irrelevant case details associated with rape culture affect criminal justice-related decision-making. We found substantial effects on decisions related to the reporting and adjudication of rape cases. Yet some factors matter more than others. Most notably, those that may distort perceptions of consent, such as a victim’s prior sexual history or relationship to the perpetrator, influence both reporting and adjudication. Case parameters that evoke victim blaming, such as the incident location, also have substantial effects. Lastly, sex matters: participants were far more likely to support female than male victims.

By contrast, the socioeconomic status of the perpetrator had little impact on decisions, suggesting that individuals are less prone to empathize with certain types of perpetrators than is often assumed in the popular discourse. There is no evidence of systematic racial bias against black victims and perpetrators; if anything, across our studies, participants tended to support them at higher rates.

To determine whether these biases uniquely apply to the context of rape cases, we compared response patterns for rape cases to robbery incidents. Some of the case details influenced reporting and adjudication decisions in the robbery conditions as well, most notably the location of the incident and the victim’s sex. However, we find strong evidence that decision patterns differ systematically for robbery and rape cases, which confirms the notion that American society exhibits cultural biases unique to rape, viewing it differently from other violent crimes. We also found some evidence of subgroup-specific effects. Most notably, ideology—which we employed as a proxy for adherence to traditional social values—seems to play a role in the evaluation of rape cases, with conservative participants being less likely to recommend for reporting or a more severe punishment cases involving victims who are male, have more promiscuous sexual histories, are single, or have a criminal record.

The results contribute to the developing body of political science research that seeks to evaluate the quality of constituent representation provided by public institutions across all three branches of government. Our findings echo some of these previous studies, suggesting some of the ways in which bias may disrupt the criminal justice processes around rape and ultimately reduce institutional responsiveness to certain groups. The policy implications of this research are potentially profound. The details included in our case profiles are those often featured prominently in news stories surrounding rape allegations (Baum et al. 2018). When journalists emphasize the salience of these features, they may unwittingly wield a powerful effect on how real-world rape cases are assessed and handled. Understanding what type of information induces bias, and through what psychological mechanisms, can potentially help explain the large reporting gap for rape. Rape victims may explicitly or implicitly be discouraged by their social environments from reporting their violations, perceiving (correctly, our results show) that they will be disbelieved or discounted.

Moreover, legally irrelevant details often also become known to jury members during trials, whether unintentionally through witness testimonies or deliberately as part of the defense strategy. Our findings may provide insights into, and guidance for, prosecutorial decisions and strategies. Research has shown that certain case characteristics make it more or less likely for prosecutors to take a case (Beichner and Spohn 2005). Systematic evidence of the kind of details that lead to distortions may facilitate the redesign of investigative, prosecutorial, and trial incentives and procedures surrounding rape cases in order to reduce the prevalence and influence of such biases.

By invoking rape culture beliefs, emphasis on legally irrelevant case details may distort public perceptions of prominent rape cases and alter people’s views concerning appropriate responses to such incidents. As more media reports on rape incidents surface that involve politicians and public figures, our findings may provide some insights as to when we ought to expect a backlash against sexual misconduct in the form of heightened activism, public pressures to prosecute or punish prominent perpetrators, or even electoral defeats. The public’s impressions of whether a victim is credible or a particular crime is severe can determine the career trajectory of the accused; this is exemplified by the debates over whether the comedian Louis C.K. waited “long enough” to return to stand-up after a 9-month hiatus.³⁵

Finally, our study holds implications for both future research and policy. In the former case, the next logical extension of this work would be to employ similar methodologies to explore the effects of implicit biases associated with rape culture on key actors in the criminal justice system, such as police, attorneys, and judges. Researchers should also investigate whether rendering these implicit biases explicit helps mitigate their effects. If so, the implication for policy would be to develop policies aimed at doing just that. Taken together, our findings show that how the public views rape matters—for people, politics, institutions and democracy—even more so in the era of #MeToo.