Introduction

Sexual assault at higher education institutions in the USA has recently garnered national attention (Brown et al. 2015; Taylor 2015; White House Task Force to Protect Students From Sexual Assault 2014). The detrimental physical, emotional, and psychological effects include sexually transmitted infections, unwanted pregnancies, substance use, and post-traumatic stress disorder (Chen et al. 2010; Martin et al. 2011b). Higher education institutions are being urged to implement and improve prevention and treatment for sexual violence (White House Task Force to Protect Students From Sexual Assault 2014).

A critical first step in improving sexual assault prevention and treatment is identifying the prevalence of sexual assault and populations disproportionately affected by sexual assault (White House Task Force to Protect Students From Sexual Assault 2014). Studies have shown that gender, sexual orientation, and race/ethnicity separately influence risk for sexual assault among undergraduate students. Cisgender (i.e., not transgender) women are at heightened risk for sexual assault compared to cisgender men (American College Health Association 2012, 2013, 2014; Cantor et al. 2015; Krebs et al. 2016; Sinozich and Langton 2014). Transgender people are at higher risk for sexual assault than their cisgender counterparts (Cantor et al. 2015; Coulter et al. 2015; Krebs et al. 2016), and gay/lesbian and bisexual people are at greater risk of sexual assault than heterosexuals (Blosnich and Bossarte 2012; Blosnich and Horn 2011; Krebs et al. 2016; Martin et al. 2011a). Meanwhile, racial/ethnic differences in sexual assault among college students are inconsistent. White women are sometimes more likely than Black (Gross et al. 2006; Koss et al. 1987; Mohler-Kuo et al. 2004), Hispanic (Koss et al. 1987), and Asian (Cantor et al. 2015; Koss et al. 1987) women to experience sexual assault—but not always (Cantor et al. 2015; Krebs et al. 2016; Mohler-Kuo et al. 2004). Studies often only sample cisgender women or students from a few post-secondary institutions, thereby creating small and limited samples that preclude researchers from examining how gender identity, sexual identity, and race/ethnicity intersect and modify the risk for sexual assault victimization.

To advance sexual assault prevention and treatment efforts, intersectional approaches can elucidate whether and how specific subpopulations are more vulnerable to sexual assault. Intersectionality posits that myriad systems of stigma and discrimination (e.g., racism, heterosexism, and sexism) functioning at multiple levels (e.g., structurally and interpersonally) combine and interact to confer disproportionate risks among populations with marginalized social identities (Crenshaw 1989, 1991). Commonly, intersectional analyses examine how social identity markers (e.g., race/ethnicity, sexual identity, and gender identity) interact to modify risks of certain health outcomes (Bowleg 2008; Institute of Medicine 2011). In previous studies (Corliss et al. 2014; Talley et al. 2014), interactions between race/ethnicity, gender identity, and sexual identity show heightened health risks (e.g., alcohol and cigarette use) among certain subgroups. Yet no studies to date have examined how sexual identity, gender identity, and race/ethnicity taken together may contribute to increased vulnerability for sexual assault among undergraduate students. This is partially because quantitative studies of intersectionality necessitate investigating interactions, which require large sample sizes (Bowleg 2008). Intersectional approaches can allow researchers to identify heterogeneity in sexual assault prevalence and illuminate predictors of sexual assault for specific subgroups.

We analyzed survey data from undergraduate students collected at 120 post-secondary institutions to explore the intersections of gender identity, sexual identity, and race/ethnicity. First, we investigated and explored the independent effects of sexual identity, gender identity, and race/ethnicity on past-year sexual assault. We hypothesized that LGBT people would be at greater risk for sexual assault than their cisgender, heterosexual counterparts (Blosnich and Bossarte 2012; Blosnich and Horn 2011; Cantor et al. 2015; Coulter et al. 2015; Krebs et al. 2016; Martin et al. 2011a). However, we did not have specific a priori hypotheses about race/ethnicity because of inconsistent previous findings (Cantor et al. 2015; Gross et al. 2006; Koss et al. 1987; Krebs et al. 2016; Mohler-Kuo et al. 2004). Second, we explored the interaction effects of sexual identity, gender, and race/ethnicity on sexual assault. We did not have a priori hypotheses for interaction effects because these analyses were exploratory in nature.

Methods

Study Design

We used cross-sectional survey data from the National College Health Assessment (NCHA) survey administered through the American College Health Association. Schools were included in this dataset if they self-selected into the NCHA data administration (American College Health Association 2012, 2013, 2014). To ensure the representativeness of samples, only schools that sampled randomly selected students, sampled students in randomly selected classrooms, or sampled all students at their school were included in the final datasets made available to researchers. The current study used data collected in Fall 2011 (44 schools), Fall 2012 (51 schools), and Fall 2013 (57 schools) from 120 unique higher education institutions in the USA. Schools were located in the northeast (37.5%), south (29.2%), midwest (17.5%), and west (15.8%). Overall, 30.8% of the schools were doctorate-granting universities, 30.8% were master’s colleges and universities, 26.7% were baccalaureate colleges, 10.0% were associate’s colleges, and 1.7% were special focus institutions (Center for Postsecondary Research 2011). Half were public schools.

During a specific time period selected by each school’s administration, sampled students completed paper or web surveys depending on the format offered at their institution. Response rates were high for paper surveys (mean response proportions range, 71 to 100% from 2011 to 2013), but lower for web surveys (mean response proportions range, 16 to 21% from 2011 to 2013). In total, 88,975 students completed the surveys, including 73,791 self-identified undergraduate students included in the current study. Each school’s Institutional Review Board (IRB) approved the original study procedures, and the University of Pittsburgh’s IRB deemed the current secondary analyses as exempt.

Measures

Past-Year Sexual Assault

Sexual assault was assessed with three yes/no items after the initial stem “Within the past 12 months”: (1) “were you sexually touched without your consent”; (2) “was sexual penetration attempted (vaginal, anal, oral) without your consent”; and (3) “were you sexually penetrated (vaginal, anal, oral) without your consent.” Consistent with the previous work (Blosnich and Horn 2011; Coulter et al. 2015), we combined these three items into a single variable because they were low prevalent events (range, 1.6–5.9%) and internally consistent (Kuder-Richardson coefficient = 0.72; Kuder and Richardson 1937). If participants responded “yes” to any of these items, we classified them as having been sexually assaulted. If they responded “no” to all items, we classified them as having not been sexually assaulted.

Key Exposure Variables

Sexual identity was assessed with the following question: “What is your sexual orientation?” Participants selected one of the following options: heterosexual, gay/lesbian, bisexual, or unsure. Gender identity was assessed with the following question: “What is your gender?” Participants selected one of the following options: female, male, or transgender. As done previously (Diemer et al. 2015), we refer to participants as cisgender women, cisgender men, and transgender people, respectively. Race/ethnicity was assessed with the following question: “How do you usually describe yourself?” Participants selected one or more of the following options: White; Black; Hispanic or Latino/a; Asian or Pacific Islander; American Indian, Alaskan native, or native Hawaiian; biracial or multiracial; or other. Due to small cell sizes for some groups, we created five categories: White only, Black only, Hispanic or Latino/a only (referred to as Latino hereafter), Asian or Pacific Islander (API) only, and other.

Covariates

Age in years was assessed continuously, but we created the following categories based on sample size and interpretability—18, 19, 20, 21, 22, 23–24, 25–29, 30–39, and 40 years or more. We also measured year in school (1st, 2nd, 3rd, 4th, or ≥5th year undergraduate students) and survey year (2011, 2012, or 2013).

Analyses

We conducted analyses in Stata version 14 (College Station, TX), and set statistical significance at 0.05. Among the 73,791 undergraduates who completed the surveys, we removed 0.8% of the participants who were missing data on any sexual assault items. Of the remaining participants, 0.6% had missing data for gender identity, 1.0% for sexual identity, 0.6% for race/ethnicity, and 0.7% for age. Because total missingness was low (3.2%), we used listwise deletion, creating an analytic sample of 71,421 participants.

We examined the bivariate relationships between independent variables and sexual assault. To examine multivariable associations of sexual assault, we used multilevel logistic regression models with random intercepts, allowing sexual assault to vary by school. In the unconditional model (i.e., no predictor variables), sexual assault varied significantly between schools (variance component [VC] = 0.12; standard error [SE] = 0.02), justifying the use of multilevel modeling. Controlling for covariates, we fit a model containing the main effects of key exposure variables, followed by a model containing all three-way and two-way interactions between key exposure variables. When using the total sample, however, models examining interactions of gender identity by race/ethnicity did not converge. This occurred because of small cell sizes and multicollinearity (several variance inflation factors were greater than 8.0). To remedy this issue, we fit models separately for cisgender and transgender people.

Among cisgender people, the unconditional model showed that sexual assault varied between schools (VC = 0.12; SE = 0.02). We tested three-way interactions between sexual identity, gender, and race/ethnicity, which were not significant (p = 0.49). We removed them, and our second model included all two-way interactions between key exposure variables. Because the majority of significant interactions were by gender identity, we also fit stratified models containing main effects only for cisgender men and women separately. We estimated the past-year predicted probabilities of sexual assault for key exposure variable subgroups averaged across covariates.

Among transgender people, the unconditional model showed sexual assault did not vary between schools (VC = 0.12, SE = 0.42), therefore we used standard logistic regression. Since the number of transgender people was small (n = 177), we only examined the main effects. We estimated the predicted probabilities of sexual assault for each race/ethnicity and sexual identity subgroup averaged across the covariates.

Results

Sexual Assault Among Total Sample

Table 1 shows the demographic characteristics of the total sample and by gender identity. Table 2 shows the unadjusted prevalence of past-year sexual assault of the total sample and by gender identity. For gender identity, sexual assault was highest among transgender people (20.9%), followed by cisgender women (8.6%), and lowest among cisgender men (3.6%). For sexual identity, sexual assault was highest among bisexuals and people unsure of their sexual identity (15.7 and 12.6%, respectively), followed by gays/lesbians (9.8%), and lowest among heterosexuals (6.4%). For race/ethnicity, past-year sexual assault was greatest among Blacks and people with other race/ethnicity (8.7 and 8.6%, respectively), followed by Whites (7.0%), and lowest among Latinos and APIs (5.4 and 5.3%, respectively).

Table 1 Characteristics for the total sample and stratified by gender identity: 2011–2013
Full size table
Table 2 Prevalence of past-year sexual assault: 2011–2013
Full size table

As shown in model 1 in Table 3, cisgender women had higher odds of sexual assault than cisgender men (adjusted odds ratio [AOR] = 2.46; 95% confidence interval [CI] 2.27, 2.66), and transgender people had significantly higher odds of sexual assault than both cisgender men (AOR = 3.93; 95% CI 2.68, 5.76) and women (shown by non-overlapping CIs). Gays/lesbians (AOR = 1.92; 95% CI 1.63, 2.26), bisexuals (AOR = 2.37; 95% CI 2.12, 2.65), and people unsure of their sexual identity (AOR = 1.95; 95% CI 1.66, 2.28) had significantly higher odds of sexual assault than heterosexuals. Compared to Whites, APIs (AOR = 0.75; 95% CI 0.66, 0.85) and Latinos (AOR = 0.81; 95% CI 0.72, 0.92) had significant-ly lower odds of sexual assault, while Blacks (AOR = 1.28; 95% CI 1.13, 1.44) and people with other race/ethnicity (AOR = 1.23; 95% CI 1.13, 1.34) had significantly higher odds of sexual assault. Compared to 18-year-olds, people aged 22 and above had significantly lower odds of past-year sexual assault (AORs range, 0.29–0.65).

Table 3 Multivariable logistic regression results of past-year sexual assault: 2011–2013
Full size table

Sexual Assault Among Cisgender Men and Women

Model 2 in Table 3 displays results from models with two-way interactions between gender, sexual identity, and race/ethnicity; models 3 and 4 display results of stratified models for cisgender men and women, respectively. Gender identity modified some of the effects of sexual identity on past-year sexual assault. There was a significant difference between gays/lesbians and heterosexuals for cisgender men (AOR = 3.50; 95% CI 2.81, 4.35; Model 3), but not for cisgender women (AOR = 1.13; 95% CI 0.87, 1.46; model 4). The difference in past-year sexual assault between people unsure of their sexual identity and heterosexuals was greater for cisgender men (AOR = 2.92; 95% CI 2.10, 4.08) than for cisgender women (AOR = 1.68; 95% CI 1.40, 2.02).

Gender identity also modified effects of race/ethnicity on sexual assault, with smaller differences for Whites compared to Latinos, Blacks, and people of other race/ethnicity for cisgender women versus men. The Black–White AOR was 1.97 (95% CI 1.49, 2.61) for cisgender men and 1.16 (95% CI 1.01, 1.34) for cisgender women; we found a similar pattern for people of other race/ethnicity compared to Whites. For cisgender women, Latinos had lower odds of sexual assault than Whites (AOR = 0.73; 95% CI 0.63, 0.84); but for cisgender men, there were no significant differences between Latinos and Whites (AOR = 1.21; 95% CI 0.93, 1.57).

Sexual Assault Among Transgender People

Model 5 in Table 3 displays multivariable associations of sexual assault among transgender people. Compared to White transgender people, Black transgender people had significantly higher odds of sexual assault (AOR = 8.26; 95% CI 1.09, 62.82).

Predicted Probabilities of Sexual Assault

Table 4 shows the predicted probabilities of past-year sexual assault averaged across covariates and stratified by gender identity. Heterosexual, White, and API cisgender men had the lowest predicted probabilities of sexual assault (2.6–2.9%). Predicted probabilities for gay, bisexual, and unsure cisgender men (7.7–9.4%) were similar to predicted probabilities for heterosexual and gay/lesbian cisgender women (7.7–11.8%), and lower than bisexual cisgender women (15.9%). Predicted probabilities were generally higher among transgender versus cisgender people, with the highest being for Black transgender people at 57.7%.

Table 4 Predicted probabilities of past-year sexual assault for sexual identity and race/ethnicity by gender identity: 2011–2013
Full size table

Discussion

Our exploratory results indicated that sexual assault disproportionately affects several sexual, gender, and racial/ethnic minority subgroups. Transgender people had higher odds of sexual assault than cisgender men and women, and Black transgender people had significantly higher odds of sexual assault than White transgender people. Like in previous studies (American College Health Association 2012, 2013, 2014; Cantor et al. 2015; Krebs et al. 2016; Sinozich and Langton 2014), cisgender women had higher odds of sexual assault than cisgender men, but gender modified the effects of sexual identity and race/ethnicity on sexual assault. For example, being bisexual, unsure of one’s sexual identity, Black, or of other race/ethnicity increased the odds of sexual assault more for cisgender men than for cisgender women. Though these relative differences were greater among cisgender men than women, the absolute differences for these groups were relatively similar.

Our paper utilized an intersectionality framework (Crenshaw 1989, 1991) to examine how multiple marginalized identities intersect to form and amplify varying risks for sexual assault. An intersectional frame can also be used to understand the causes of sexual assault in these subgroups. The etiology of sexual assault is complex, stemming from factors at multiple levels of the social ecological model (Heise 1998). Each factor is tied to one’s social identities in specific ways, thereby influencing sexual assault risks. For example, discrimination may contribute to dehumanizing specific groups of people and putting them at risk for violence. Transgender people face discrimination interpersonally (e.g., being misgendered) and structurally (e.g., by policies and laws dictating bathroom usage), and Black people also face interpersonal discrimination (e.g., biased language). The intersections between these oppressions may cultivate unsafe environments for Black transgender people, placing them at substantially greater risk of sexual assault. Additionally, examining perpetrators’ characteristics (e.g., relationship with survivor) and motivations can highlight specific intervention targets (e.g., malleable risk factors among specific populations). For example, common motivations for sexual assault include power and anger (Nicholas 1977), and perhaps perpetrators targeting LGBT people are more likely to engage in “corrective” rape (i.e., trying to “cure” LGBT people through sexual assault; Martin et al. 2009). Our findings suggest that intersectional approaches in epidemiologic research may elucidate the individual and contextual factors contributing to the elevated sexual assault risks among specific subgroups.

Prevention and Treatment Implications

Few interventions have been shown to be effective in preventing sexual assault (DeGue et al. 2014). Even fewer sexual assault interventions are tailored for racial/ethnic minority populations (DeGue et al. 2014), and none have been evaluated with LGBT populations (DeGue et al. 2014). Despite cisgender heterosexual women being at high risk for sexual assault, we found that transgender people and cisgender bisexual women have even higher odds of sexual assault. If sexual assault prevention solely focuses on cisgender heterosexual violence (e.g., cisgender man-on-woman violence), it may invalidate LGBT people’s assault experiences and be ineffective for LGBT populations. To our knowledge, studies have not examined whether universal sexual assault interventions equally reduce sexual assault across all sexual orientation, gender, and racial/ethnic subgroups. If universal interventions are less effective among minority subgroups, researchers and practitioners can augment existing universal interventions by explicitly addressing homophobia, biphobia, transphobia, and racism, and they can develop new or adapt existing sexual assault interventions for LGBT populations and racial/ethnic minorities. Prevention programs may be more effective when tailored to the cultural beliefs and norms of the target population (Nation et al. 2003; Small et al. 2009). Furthermore, compared to people aged 22 or less, people who are aged 25 or older have lower odds of sexual assault, perhaps because they are atypical undergraduate students and less engaged with campus life, making campus-focused prevention less appropriate for this population.

Because providing trauma-informed care to sexual assault survivors requires understanding survivors’ histories and the contexts of their experiences (The National Sexual Assault Coalition Resource Sharing Project and National Sexual Violence Resource Center 2013), higher education institutions should include staff training on sexuality, gender, and race/ethnicity. Reporting sexual assault to college administrators can be difficult and stigmatizing (Rennison 2002). Meanwhile, many sexual, gender, and racial/ethnic minorities may feel mistrustful, unwelcomed, invisible, or discriminated against, which makes reporting their experience of sexual assault even more difficult (Boulware et al. 2003; Frankowski 2004; Grant et al. 2011). Because many of these minority subgroups are at heightened risk for sexual assault, developing programs that are welcoming and affirming of survivors with marginalized identities is crucial. Staff who treat sexual assault survivors should be aware that sexual assault during college may not be their first or most profound assault event (Carey et al. 2015; Martin et al. 2011a). While our study could not assess whether the assaults occurred on campus, sexual assault confers numerous mental, physical, and reproductive health impacts (Chen et al. 2010; Martin et al. 2011b). Therefore, regardless of sexual assault location, it is critical for administrators to understand the vulnerability their students experience as they adopt prevention programming on their campuses. A trauma-informed approach to working with LGBT survivors, in particular, is necessary as assault is astoundingly prevalent among LGBT youth (Friedman et al. 2011; Rothman et al. 2011). Consequently, treatment of sexual assault among LGBT undergraduates may uncover a history of childhood and adolescent trauma.

Limitations and Strengths

We must consider our results within their limitations. Sexual assault was self-reported and is often underreported (Rennison 2002). Because surveys assessed past-year sexual assault, some reports of the sexual assault may have occurred while students were not at college, especially for first-year students. Nevertheless, higher education institutions should still be prepared to provide trauma-informed care because the effects of sexual assault can be long lasting and impact students’ health during college (Chen et al. 2010; Martin et al. 2011b). We also combined sexual touching, attempted penetration, and completed penetration, were unable to examine the tactics used during sexual assault (e.g., physical force, incapacitation, and no affirmative consent), and had no information about perpetrators. Sexual identity was used to measure sexual orientation, and we may have found different results if we used measures of sexual behavior or attractions (Matthews et al. 2013). Additionally, the gender identity question may under-identify transgender people. For example, some people whose current gender identity differs from their sex assigned at birth may identify solely as female or male, but not transgender. This measure did not allow transgender individuals to identify as transgender men or transgender women. Because of small cell sizes and multicollinearity, we could not include transgender people in analyses with interactions and collapsed some race/ethnicity subgroups. Interactions are an imperfect way to test intersectionality, but are among the best quantitative techniques to date (Bowleg 2008).

In spite of analyzing data from 120 institutions, generalizability may be limited. Selection bias may have occurred because participation rates were low in some schools, and we were unable to adjust for the response rate within each school. We could not adjust for participants who completed surveys at multiple waves, which may result in non-independence among participants; however, we mitigated this bias by controlling for survey year in our models. Additionally, participation rates were less than optimal for web surveys, which is common across web surveys (Cook et al. 2000). Moreover, post hoc analyses (data not shown) revealed that survey administration (web versus paper survey) was not associated with sexual assault in any of our models, and did not substantially affect any of our results, thereby introducing minimal bias.

Despite these limitations, our study has some key strengths. We utilized data from 120 higher education institutions and included cisgender men and transgender people as well as cisgender women. We used multilevel models to control for the clustering of students within institutions, which is uncommon in the literature to date. As one of the largest studies to examine sexual assault at higher education institutions, we had enough statistical power to examine intersections by gender, race/ethnicity, and sexual orientation among cisgender people—thereby addressing one of the five major recommendations put forth in the 2011 Institute of Medicine report about LGBT health (Institute of Medicine 2011).

Conclusions

Sexual assault is a problem disproportionately affecting certain sexual identity, gender identity, and race/ethnicity subgroups. Epidemiologic research and interventions should consider intersections of gender identity, sexual identity, and race/ethnicity to better tailor sexual assault prevention and treatment for college students.