The Prevalence of and Barriers to Bystander Intervention on Behalf of Sexual Assault and Intimate Partner Violence Victims

Abstract

Using newly available, U.S. nationally representative data from the No More study (N=1,307), this article investigates 1) knowledge of sexual assault and intimate partner violence (IPV) victims within one’s social network, 2) who intervenes, whom they intervene on behalf of, and how they intervene, and 3) the perceived barriers to intervening in IPV specifically. The findings reveal that knowledge of violence, the likelihood of intervening, and the intervention approaches taken all vary demographically and by violence type. Among respondents who have known a victim, one-third report having intervened for sexual assault, while one-half report having intervened for IPV. For both types of violence, respondents are more likely to have intervened on behalf of family or friends than on behalf of more distant network members. However, respondents are more likely to have solicited the help of authorities and less likely to have offered safe haven in instances of sexual assault than in instances of IPV. The most commonly cited barriers to IPV-intervention include fear of injury, fear of misinterpretation, and belief that IPV is a private matter, though these vary across demographic groups. These findings indicate that the decision to intervene is highly contextual—contingent on the individual characteristics of the intervener, situational characteristics of the violence, and the relationship between the intervener and the victim.

Nearly 1 in 3 women in the United States are estimated to experience intimate partner violence¹ in their lifetimes (Black et al. 2011); 1 in 5 are estimated to experience sexual assault² (Black et al. 2011). In recent years, these types of violence, particularly directed at women, have garnered much attention from the media, colored by rampant reports of sexual assaults in high schools, college campuses, and the military (Bidgood and Rich 2015; Coronel, Coll, and Kravtiz 2015; Kaplan 2015; Nieves 2014; North 2015; Oppel 2013), and public reports of intimate partner violence (IPV) perpetrated by professional athletes, police officers, and pop-stars (Bernard 2013; Cohen, Ruiz and Childress 2013; Sports Illustrated Wire 2015). These incidents and the rise in media attention focused on them have sparked recent governmental investigations into institutional negligence in responding to such violence (Cohen, Ruiz and Childress 2013; Department of Education 2015).

Both governmental investigations and the institutional responses to them espouse bystander intervention as a way to prevent violence against women. For instance, in 2014 the Obama Administration launched the “Not Alone” campaign, which provides free online resources for people who wish to intervene on behalf of someone they know who is experiencing violence. The U.S. Department of Education endorsed bystander interventions through the Dear Colleague Letter campaign, which threatens to withhold federal funding from public schools and universities if they do not report, investigate, and address violence on their premises (Department of Education 2015). Likewise, the Department of Defense launched a campaign in 2015 called “Eliminate Sexual Assault: Know Your Part. Do Your Part” (Department of Defense Sexual Assault Prevention and Response 2015).

Despite recent institutional pushes for bystander intervention, little research has examined the effectiveness of this strategy in the general U.S. population. Instead, because bystander approaches entail training members of a community to view themselves as important in preventing violence to other members of the community (Baynard et al. 2004), the vast majority of studies have relied on specific subsamples of the population with clear community boundaries. Primarily, these have included college students (Bennett, Banyard and Garnhart 2013; Foubert et al. 2010; Coker et al. 2011; Gidycz, Orchowski and Berkowitz 2011; Burn 2009), high school athletes (McCauley et al. 2013), and military personnel (Foubert and Masin 2012). Focusing on such subpopulations highlights unique, context-specific factors that inhibit (or promote) violence prevention. However, doing so limits our understanding of how bystander intervention differs across demographic groups, if at all. Identifying demographic differences is an important step toward refining theoretical models of how personal circumstances, such as gender norms or the resources people have to intervene with, may affect whether and how an individual intervenes.

Moreover, to our knowledge, only one existing study compares bystander interventions in sexual assault to bystander interventions in IPV (Palmer, Nickhasa, and McMahon, 2016) and this study is also limited to college students. While sexual assault and IPV can co-occur – sexual assault can be a form of IPV and many victims experience both (Black et al. 2011) – from the vantage point of interveners, the two types of violence may be perceived quite differently. For instance, the general public perceives the typical rape to be perpetrated by a stranger (Littleton & Axom 2003; Ryan 1998) but often fails to perceive forced sex as rape if it occurs in the context of a romantic relationship (Ferro et al. 2008; Monson et al. 1996). Even victims themselves often do not identify forced sex as rape if there is a romantic relationship between them and the perpetrator (Littleton, Breitkopf, & Berenson 2007). Given that definitionally sexual assault and IPV are overlapping but distinct, and perceptions of these forms of violence are quite different, bystanders’ willingness to intervene and their strategies of intervention may differ. If these strategies do indeed differ by demographic background or violence type, then policies aimed at increasing bystander intervention would need to anticipate and accommodate these differences in order to be effective.

Bystander Intervention in Sexual Assault and Intimate Partner Violence

The dominant bystander intervention model, developed by Latané and Darley (1970), identifies five steps to the intervention process – noticing the situation (step 1), identifying it as an emergency (step 2), taking responsibility to act (step 3), deciding what specifically to do (step 4), and choosing to do it (step 5). According to the only nationally representative study of bystander interventions in IPV, only half of individuals who have known a victim have ever intervened (Beeble et al., 2008). Thus, there are likely many barriers to intervening, at least in IPV, and these barriers may exist at every step.

In the U.S. population at large (Beeble et al., 2008) and among college students and military personnel specifically (Banyard, 2008; Burn, 2009), men are more reluctant than women to intervene in instances of sexual assault. Among college students, men’s (but not women’s) willingness to intervene depends on their perceptions of the victim’s situation (e.g., whether they believe a victim increased his or her own risk of victimization) (Burn, 2009). Research on college students also suggests that African Americans are more likely to report intervening in sexual assault than whites (Brown, Banyard & Moynihan 2014). Gender and racial differences among college students underscore the likelihood of demographic differences in intervention experiences in the population as a whole.

Bystanders are also more likely to intervene in sexual assault and IPV if they perceive that victims face significant danger than if they do not (Fischer et al., 2011). This may be because dangerous situations are easier to notice (step 1 of Latané and Darley’s (1970) model) and may be more clearly identifiable as emergencies (step 2). However, sexual assault and IPV may not always be recognized as dangerous situations if there is ambiguity about consent (Burn 2009); if victims are reluctant to disclose their private experiences to others (Dobash & Dobash 1979; Starzynski et al. 2005; Ahrens et al. 2010); or if victims minimize the extent or severity of violence (Dunham & Senn 2000) or do not identify their experiences as violence (Petersen et al. 2005). Perhaps because they more readily identify the signs of IPV and recognize them as dangerous, individuals who have previously been victims of IPV or who have been exposed to IPV as children are more likely to intervene than others (Beeble et al. 2008).

Among college students, existing research finds a greater willingness to intervene on behalf of personally known sexual assault victims than on behalf of strangers (Burn 2009; Bennett et al. 2013). This may be because knowing a victim encourages one to feel more responsible to act (step 3). College students also report a greater willingness to intervene on behalf of sexual assault victims when they feel supported by their peers, believe they know how to help, and are not in danger themselves (Bennett et al. 2013). Thus, norms about intervention and education helping individuals to identify instances of sexual assault and IPV may also be critical to the decision to intervene (step 4) (Brown, Banyard, & Moynihan, 2014; McMahon & Dick, 2011).

Once a person decides to intervene, a wide variety of strategies may be employed. These may include preventing violence by changing attitudes and beliefs that lead to assault or intervening in a potentially threatening situation before violence has begun; stopping violence mid-incident (which requires witnessing an event); or providing emotional or physical support after violence has occurred (e.g., Hoxmeier, Flay, & Acock, 2015; McMahon & Banyard, 2011). All three types of interventions (before, during, and after) are theorized to reduce an individual’s subsequent risk of violence (McMahon & Banyard, 2011).

Current Study

This study makes two contributions to the literature on bystander intervention. First, it uses a nationally representative survey of American adults to investigate demographic differences in bystander intervention in sexual assault and IPV. To date, only one other study has used nationally representative data to examine bystander intervention in IPV (Beeble et al., 2008). Second, this analysis examines how personal knowledge of victims, the likelihood of intervening, and intervention strategies differ between sexual assault and IPV.

Data and Methods

Sample

Data from the No More study (N=1,305), a U.S. nationally representative, cross-sectional survey of adults aged 15 and older, are made available by the Avon Foundation. This survey was administered in 2013 through GfK’s Knowledge Panel (commonly known as Knowledge Networks). The sample was drawn from a pre-recruited panel (Callegaro & DiSogra, 2008) of 50,000 respondents who were recruited through random-digit dialing (RDD) and address-based sampling (ABS) methods. By joining the panel, respondents agreed to participate periodically in online surveys and were provided Internet access and equipment if they did not already have it.

Respondents in the panel filled out an initial profile of basic demographic information when they enrolled. This was updated yearly from the time of enrollment. This study had a 65% profile completion rate for the demographic information (PROR) (Callegaro & DiSogra, 2008). The completion rate (COMPR) was 48%. Knowledge Networks’s samples closely match those of traditional RDD surveys and when weighted are representative of the United States as a whole (Chang & Krosnick, 2009; DiSogra, Dennis, & Fahimi, 2010).

The data were weighted to adjust for known sources of deviation from an equal probability of selection. To reduce the effects of non-coverage or non-response bias, a post-stratification adjustment was applied using demographic distributions from the most recent data from the U.S. Census Bureau and U.S. Department of Labor’s Current Population Survey for gender, age, race and ethnicity, education, census region, and whether the respondent lives in a city. The data were also weighted with regard to Internet access, information on which was collected at time of recruitment.

The No More data offer several advantages to this study. First, they are derived from a survey that included detailed questions about previous bystander experiences, including who respondents had intervened on behalf of and how they intervened. Second, whereas most research on bystander behavior has focused on specific subpopulations, the No More sample is nationally representative, allowing us to compare intervening behaviors across demographic groups and to make broader inferences about the U.S. population at large. Third, the No More survey included questions about past intervention experiences with sexual assault and IPV, whereas most studies focus on just one or the other. Fourth, individuals are more likely to report sensitive behaviors in computer-based surveys than in person (Schroder et al. 2003).

Measures

Previous intervention experiences.

We examine three factors related to previous bystander experiences: whether the respondent has ever known a victim of sexual assault or IPV (separately); if yes, what the relationship was between the respondent and the victim and whether the respondent intervened; and if yes to the latter question, how the respondent intervened.

Because sexual assault can occur both within and outside of relationships, some respondents may know individuals who have been victims of sexual assault via IPV. We are unable to identify such cases in our data because questions about sexual assault did not specify the assault context. Nevertheless, 45% of respondents who have known a sexual assault victim reported also knowing an IPV victim; 76% of respondents who have known an IPV victim also reported knowing a sexual assault victim. We assume at least some of these respondents know individuals who have simultaneously been victims of both. As such, we combine information on knowledge of violence types into one measure of knowing a victim: known none, known both, known sexual assault only, and known IPV only.

If a respondent reported knowing a sexual assault or IPV victim he/she was then asked, “What was the victim’s relationship to you?” Possible answers included: “family member,” “friend,” “acquaintance,” “friend of a family member other than my child,” and “friend of my child.” We code each relationship type separately, defined as (1) yes (shared this type of relationship with the victim) and (0) no.

If a respondent knew a victim and reported intervening he/she was then asked, “How did you step in to help?” Possible answers included “physically intervened,” “got an adult/ another person,” “reported to authorities,” “invited the victim to hang out/ provided safe haven,” “expressed sympathy,” and “told the abuser to stop.” Respondents were asked to check all that apply. We thus treat each intervention strategy separately and code all (1) for yes (intervened in this fashion) or (0) for no.

Barriers to intervention.

Irrespective of having previously intervened, all respondents were asked, “Which of the following, if any, could you imagine would prevent you from stepping in to help a victim of partner abuse/ violence whom you knew?” Respondents were allowed to check as many of the following answers as applied: “Afraid to get hurt,” “It’s private and I should stay out of it,” “Worried I would be called a liar,” “Worried I would be bullied at school,” “Afraid I would lose a friend,” and “Worried I was wrong and they were just joking around.” We treat these responses separately and code each one the same: (1) indicates the respondent perceived the item as a barrier to their willingness to intervene, and (0) indicates the respondent did not.

Demographic Characteristics.

Given the demographic differences in prior work on intervention, our multivariate models include various demographic covariates: respondents’ age (15–91 years); race (white, black, and Hispanic or other); gender; urban or non-urban residence; highest level of education (incomplete high school, complete high school, at least some college, B.A., or post-secondary); employment status (non-employed, full-time, part-time, self-employed, student, or retired); and geographic region (Northeast, South, Midwest, and West).

Analytic Strategy

Our analysis is organized into three parts. For both the first and the third part, the units of analysis are respondents; in the second part the units of analysis are instances of respondents knowing a victim. This is discussed more below.

The first analytic component offers a descriptive overview of who knows of victims within their social network—step one of Latané and Darley’s (1970) model. Here the units of analysis are individual respondents. This analysis graphs the prevalence of knowledge of victims by relationship to victim and violence type (Figure 1) and uses t-tests to formally compare the prevalence of relationship types to sexual assault and IPV victims. It also includes a multinomial logistic regression model that estimates demographic differences in the odds of knowing both types of victims, knowing victims of sexual assault only, and knowing victims of IPV only, relative to not knowing any victims (Table 2). To determine if only knowing victims of sexual assault differs from only knowing victims of IPV, etc., we rerun this model three times with a different reference category each time (Appendix A).

Figure 1: Respondents’ Relationships to Sexual Assault and IPV Victims Whom They Know Of.

Note: N=365 respondents who have known at lease one sexual assault victim. N=616 respondents who have known at least one IPV victim. Error bars denote 95% confidence intervals. Significant differences between sexual assault and IPV indicated by: *** p<0.001, ** p<0.01, * p<0.05.

Table 2.

Odds-Ratios Resulting from a Multinomial Logistic Regression Model Estimating Differences in Knowing Victim Types, Relative to Knowing None

	Both Types		Sexual Assault Only		Intimate Partner Violence Only
	β	SE	β	SE	β	SE
Age	1.01	(0.01)	1.01	(0.01)	1.00	(0.01)
Female	1.65*	(0.36)	0.93	(0.30)	1.53*	(0.32)
Race (ref: white)
Black	0.91	(0.42)	3.75*	(2.07)	1.08	(0.44)
Hispanic or other	1.19	(0.35)	0.77	(0.39)	0.64	(0.19)
Education (ref: <H.S.)
High school	0.65	(0.28)	1.13	(0.80)	0.74	(0.29)
Some college	0.88	(0.39)	1.01	(0.74)	0.44*	(0.18)
B.A.	0.69	(0.33)	1.37	(0.99)	0.46	(0.20)
Post-secondary	1.14	(0.53)	1.88	(1.38)	0.70	(0.30)
Employment (ref: non)
Full-time	1.11	(0.34)	0.85	(0.34)	0.86	(0.25)
Part-time	1.29	(0.55)	0.86	(0.51)	0.47	(0.21)
Self-employed	2.10	(1.00)	1.31	(0.85)	1.83	(0.89)
Student	0.37*	(0.18)	0.33	(0.28)	0.26***	(0.10)
Retired	0.37*	(0.15)	0.30*	(0.18)	0.38**	(0.14)
Urban	0.59	(0.18)	0.70	(0.37)	0.56*	(0.16)
Region (ref: Northeast)
South	1.60	(0.52)	1.28	(0.67)	1.01	(0.28)
Midwest	1.38	(0.46)	1.76	(0.99)	1.08	(0.31)
West	1.95	(0.69)	1.77	(0.90)	1.26	(0.39)
Constant	0.47	(0.35)	0.15*	(0.13)	1.85	(1.22)
Observations (respondents)	1,166		1,166		1,166

Note: Model estimated with sampling weights.

Robust standard errors in parentheses.

^***p<0.001,

^**p<0.01,

^*p<0.05

The second component investigates who intervenes, whom they intervene on behalf of, and how they intervene (steps 3, 4, and 5 of Latané and Darley’s model). Because respondents answered questions about specific victims separately, we reshape the data for this component such that each observation is an instance in which respondents have known a victim (1–4 observations per respondent who has known at least one victim). For example, if a respondent has known both a family member and a friend who have been victims of IPV, then this respondent appears twice in the data, with each relationship constituting one observation. Using instance-level data allows us to examine both within- and between-person differences in the odds of intervening, conditional on respondents’ relationship type to each known victim. With the data oriented per instance, we separately estimate the odds of intervening for sexual assault and IPV using logistic regression. To account for the non-independence of observations within respondents, we cluster standard errors by respondent. We then return to our original, individual-level data (where each respondent is represented once) to analyze how respondents who have known victims have intervened. We use graphed means to illustrate the prevalence of various intervention strategies (Figure 2) and employ t-tests to formally compare the prevalence of intervention strategies undertaken for sexual assault and IPV. As a supplement to this component, we use logistic regressions to estimate demographic differences in the odds of undertaking each strategy.

Figure 2: How Respondents Have Intervened on Behalf of Sexual Assault and IPV Victims.

Note: N=106 respondents who have intervened on behalf of a sexual assault victim. N=334 respondents who have intervened on behalf of an IPV victim. Error bars denote 95% confidence intervals. Significant differences between sexual assault and IPV indicated by: *** p<0.001, ** p<0.01, * p<0.05.

The third and final piece of our analysis analyzes the perceived barriers to intervening on behalf of IPV victims. These questions were asked of all respondents and referred to hypothetical barriers to intervention overall, not to specific past instances. As such, the units of analysis, like the first component, are respondents. We use graphed means and t-tests to determine the most common barriers (Figure 3) and logistic regressions to estimate the odds that each item is reported as a barrier (Table 4).

Figure 3: Percent of Respondents Reporting Each Item as a Barrier to Intervening on Behalf of A Hypothetical IPV Victim whom They Know.

Note: N=1,307 respondents. Error bars denote 95% confidence intervals.

Table 4.

Odds-Ratios Resulting from Logistic Regression Models Estimating Differences in the Barriers to Hypothetical IPV Intervention

	Injury	Private matter	Called a liar	Bullied	Losing friendship	Being wrong
Age	1.00	1.00	0.97	0.91***	0.97*	0.98
	(0.01)	(0.01)	(0.02)	(0.02)	(0.01)	(0.01)
Female	2.06***	0.61*	1.52	0.25*	0.98	1.01
	(0.35)	(0.14)	(0.77)	(0.14)	(0.34)	(0.23)
Race (ref: white)
Black	2.69**	0.72	1.63	2.43	0.94	0.83
	(0.89)	(0.29)	(1.67)	(2.37)	(0.51)	(0.38)
Hispanic or other	0.87	1.20	1.61	2.64	0.86	1.11
	(0.22)	(0.40)	(1.07)	(1.94)	(0.35)	(0.33)
Education (ref: <H.S.)
High school	1.13	1.10	0.49	2.12	1.35	1.36
	(0.38)	(0.47)	(0.37)	(1.69)	(0.88)	(0.67)
Some college	1.37	1.00	0.43	3.20	1.58	1.22
	(0.48)	(0.44)	(0.33)	(3.18)	(1.06)	(0.61)
B.A.	1.54	1.52	0.19	4.00	1.74	1.68
	(0.57)	(0.71)	(0.17)	(4.71)	(1.16)	(0.86)
Post-secondary	2.11	0.98	0.42	27.67**	1.90	1.29
	(0.80)	(0.47)	(0.38)	(32.15)	(1.34)	(0.67)
Employment (ref: non)
Full-time	1.00	0.52*	1.86	0.07	0.52	1.35
	(0.24)	(0.17)	(1.40)	(0.11)	(0.24)	(0.43)
Part-time	1.31	0.64	1.85	0.11	0.98	0.96
	(0.46)	(0.30)	(1.86)	(0.13)	(0.56)	(0.44)
Self-employed	0.81	1.23	2.33	1.21	1.22	1.34
	(0.31)	(0.54)	(2.46)	(1.86)	(0.77)	(0.60)
Student	1.29	0.71	2.47	0.46	0.95	3.03**
	(0.43)	(0.32)	(2.66)	(0.37)	(0.53)	(1.29)
Retired	1.15	0.95	20.77***	6.44	1.00	2.83*
	(0.36)	(0.40)	(17.33)	(12.85)	(0.75)	(1.24)
Urban	1.33	3.45***	0.74	0.59	1.88	1.08
	(0.32)	(1.29)	(0.40)	(0.36)	(0.89)	(0.34)
Region (ref: Northeast)
South	0.79	1.34	1.53	1.69	0.81	1.27
	(0.19)	(0.42)	(1.01)	(1.24)	(0.40)	(0.39)
Midwest	1.04	1.04	2.38	1.92	1.08	1.48
	(0.27)	(0.34)	(1.56)	(1.89)	(0.55)	(0.48)
West	1.00	1.01	1.29	3.19	1.57	0.93
	(0.27)	(0.38)	(0.95)	(2.68)	(0.78)	(0.34)
Constant	0.29*	0.08**	0.04*	0.12	0.07**	0.14**
	(0.15)	(0.06)	(0.06)	(0.15)	(0.07)	(0.10)
Observations (respondents)	1,175	1,175	1,175	1,175	1,175	1,175

Note: All models estimated with sampling weights.

Robust standard errors in parentheses.

^***p<0.001,

^**p<0.01,

^*p<0.05

All regression estimates are expressed as changes in odds-ratios in which values greater than 1 indicate a positive relationship with the outcome; values less than 1 indicate a negative relationship. All results are weighted to yield nationally representative estimates; all reported sample sizes are unweighted.

Results

Characteristics of bystanders and interventions

Table 1 presents descriptive statistics for our respondent-level data. Twenty-eight percent of respondents have known a victim of sexual assault, 53% have known a victim of IPV (Table 1). Thus, approximately 1.6 times as many respondents have known IPV victims than have known victims of sexual assault. However, 21% of respondents have known both.

Table 1.

Descriptive Statistics

	N	%
Total Sample	1307
Previous Bystander Experiences
Sexual Assault
Known a victim	365	28%
…And intervened	106	29%
Intimate Partner Violence
Known a victim	616	53%
…And intervened	334	54%
Demographic Characteristics
Sex: female	667	51%
Race
White	890	68%
Black	144	11%
Hispanic or other	261	20%
Education
Less than High School	103	10%
High school	403	36%
Some college	272	22%
B.A.	211	17%
Post-secondary	200	15%
Urban	1111	85%
Employment status
Non-employed or unemployed	317	24%
Full-time	458	35%
Part-time	131	10%
Self-employed	65	5%
Student	145	11%
Retired	183	14%
Age (15–91)		45.48 (mean yrs) 18.23 (st. dev.)

Note: Weighted respondent-level data used in this table.

Respondents’ relationships to victims are similar for sexual assault and IPV (Figure 1). For instance, 33% and 27% report knowing a family member who has experienced sexual assault and IPV, respectively. These similar percentages may reflect that sexual assault can occur within ongoing relationships.

For both types of violence, respondents are most likely to know of friends, family members, and acquaintances who have been victims (Figure 1), and significantly less likely to know of a friend of a family member’s or a friend of their child’s experience (t-tests confirm each of these six differences are significant at p<.001). Because knowing of others’ experiences is the first step in the bystander intervention model, these differences provide important information on whom individuals are theoretically most likely to intervene on behalf of.

The results of a multinomial logistic regression estimating differences in the odds of knowing different victim types, relative to not knowing any victims of either sexual assault or IPV, are presented in Table 2. Results of supplemental multinomial models in which knowledge of different types of violence victims serve as the reference category are available in Appendix A. We find that knowledge of victims differs systematically by demographic group. Compared to male respondents, female respondents have 65% higher relative odds of knowing both types of victims and 53% higher relative odds of knowing IPV victims only than of not knowing any victims (Table 2). Black respondents have 275% higher relative odds than white respondents of knowing sexual assault victims only than of knowing no victims (Table 2), and have higher odds of knowing sexual assault victims relative to knowing IPV only or knowing both (Appendix A.2). Compared to respondents who have not completed high school, respondents who have acquired some college education have 56% lower relative odds of knowing IPV victims only than of not knowing any victims. Students and retired respondents have lower relative odds of knowing both and of knowing IPV victims only than non-employed respondents (Table 2). Retired respondents also have lower relative odds of knowing sexual assault victims only than non-employed respondents (Table 2). Finally, compared to respondents residing in non-urban areas, respondents residing in urban areas have 44% lower relative odds of knowing IPV victims only than of not knowing any victims of either type. No variation in knowledge of victims across geographic regions is detected (Table 2 and Appendix A).

Who intervenes, whom they intervene on behalf of, and how they intervene

Table 1 reveals that 8% (n=106) and 26% (n=334) of all respondents (n=1,307) have intervened on behalf of a victim of sexual assault and IPV, respectively. In terms of respondents who have known victims, 29% have intervened for sexual assault victims; 55% for IPV victims.

Table 3 presents the results of logistic regression models estimating differences in the odds of intervening among respondents who have ever known victims, separately for sexual assault and IPV. The results on the left-hand side of the table suggest that the odds of intervening in instances of sexual assault do not vary with relationship to the victim (e.g. whether they are a friend, family member, etc.); however, they do differ with demographic background. Hispanic or other non-white, non-black respondents have 159% higher odds of intervening than white respondents. Self-employed respondents have 192% higher odds of intervening than non-employed respondents (Table 3), perhaps because a latent characteristic such as self-motivation simultaneously contributes to the likelihood of self-employment and intervention. No other demographic differences in sexual assault intervention are detected.

Table 3.

Odds-Ratios Resulting from Logistic Regression Models Estimating Differences in the Odds of a Past Intervention on Behalf of a Known Intimate Partner Violence Victim

	Sexual Assault		Intimate Partner Violence
	β	SE	β	SE
Relationship to victim (ref: family member)
Friend	0.76	(0.20)	0.93	(0.18)
Acquaintance	0.73	(0.24)	0.30***	(0.07)
Friend of family	0.38	(0.26)	0.54	(0.21)
Friend of child’s	--	--	--	--
Age	1.61	(0.43)	1.01	(0.01)
Female	2.25	(1.22)	1.28	(0.23)
Race (ref: white)
Black	2.25	(1.22)	0.99	(0.39)
Hispanic or other	2.59**	(0.87)	1.43	(0.33)
Education (ref: <H.S.)
High school	0.91	(0.44)	0.70	(0.23)
Some college	1.15	(0.59)	0.92	(0.31)
B.A.	0.95	(0.53)	0.57	(0.21)
Post-secondary	1.07	(0.57)	0.99	(0.36)
Employment (ref: non)
Full-time	1.59	(0.60)	0.90	(0.23)
Part-time	1.68	(0.84)	0.76	(0.27)
Self-employed	2.92*	(1.49)	1.06	(0.43)
Student	1.01	(0.61)	0.62	(0.22)
Retired	1.51	(0.79)	0.64	(0.23)
Urban	0.54	(0.20)	0.82	(0.19)
Region (ref: Northeast)
South	1.22	(0.54)	0.94	(0.24)
Midwest	0.98	(0.44)	0.90	(0.25)
West	1.24	(0.56)	0.81	(0.23)
Constant	0.13*	(0.11)	1.95	(1.12)
Observations (incidents of known violence)	364		635

Note: Model only include respondents who report knowing at least one sexual assault or IPV victim.

Robust standard errors, clustered by respondent, in parentheses.

^***p<0.001,

^**p<0.01,

^*p<0.05

The results on the right-hand side of Table 3 suggest that the relationship of the respondent to the victim matters for IPV: respondents have 70% lower odds of intervening on behalf of an acquaintance than on behalf of a family member (Table 3). However, no demographic differences in who intervenes on behalf of IPV victims is observed. A supplementary multinomial logistic regression, conducted at the person-level and limited to respondents who have known both types of victims, reveals no significant demographic differences in the odds of intervening for both types of violence, sexual assault only, or IPV only, relative to not intervening.

Figure 2 portrays the prevalence of different intervention strategies undertaken. For both types of violence, the two most common forms of intervention are offering the victim safe haven and offering the victim sympathy. However, respondents tend to intervene differently for sexual assault and IPV victims: only 28% of interveners have told the abuser to stop in instances of sexual assault, while 50% of interveners have done the same in instances of IPV (p<.001). Likewise, 47% of sexual assault-interveners have offered victims safe haven, compared to 60% of IPV-interveners (p<.05). Interveners are substantially more likely to involve authorities or adults when intervening on behalf of sexual assault victims—41% and 34% respectively have done so—than on behalf of IPV victims, in which case only 25% and 22%, respectively have done so (p<.001; p<.05). These differences highlight that steps 4 and 5 of the bystander intervention model (deciding what to do and choosing to do it) depend in part on the type of violence in question.

In analyses not shown, we conduct multivariate logistic regressions predicting each intervention strategy separately. Few demographic differences emerge. Nevertheless, we find that with regard to IPV, women are less likely to physically intervene than men (odds-ratio 0.28; p<.001), less likely to tell the abuser to stop (odds-ratio 0.46; p<.05), and more likely to tell an adult (odds-ratio 2.66; p<.05) (tables available upon request). We find no significant differences in the odds of sexual assault intervention strategies, likely due to the small sample size of respondents who have ever intervened on behalf of a sexual assault victim (n=106).

Barriers to Intervening on Behalf of IPV Victims

Finally, we investigate perceived barriers to intervening in hypothetical situations of IPV. Figure 3 presents bar charts of the frequencies with which respondents reported various barriers to their willingness to intervene. As this figure reveals, the most commonly cited obstacle to intervening is fear of physical injury (reported by 43% of respondents), which is cited nearly three times more often than not wanting to intervene in private matters and fear of misinterpreting the situation (t-tests confirm these differences, p<.001). The latter two barriers are both cited by 15% of respondents (Figure 3).

Table 4 presents the results of logistic regressions estimating differences in the odds of reporting barriers across demographic groups. We find, first, that women have 106% higher odds of reporting a reluctance to intervene for fear of physical injury than men, which coheres with women being less likely to physically intervene or to tell the abuser to stop in cases of IPV. Women also have 49% lower odds than men of articulating that their perception of IPV as a private matter would be a barrier to intervening (Table 4). Second, for each year older a respondent is, his or her odds of reporting a fear of being bullied or of losing a friendship as a barrier decrease by 9% and 3%, respectively (Table 4), reflecting either a cohort effect or that these reasons become less salient as respondents progress through the life course. Third, black respondents have 169% higher odds of reporting fear of injury as a barrier to intervening on behalf of an IPV victim than white respondents (Table 4). Fourth, with regard to socioeconomic differences, retired respondents have higher odds than non-employed respondents of reporting two related types of barriers: being called a liar and being wrong. Although the odds-ratio for retired respondents reporting “being called a liar” as a barrier, and the odds-ratio for respondents with post-secondary education reporting a fear of being “bullied” as a barrier are statistically significant, substantive conclusions from these two unusually large odds-ratios should be made with caution given our small sample size.

Discussion

This study investigated how bystander intervention experiences vary by demographic background and violence type—sexual assault or IPV. The analysis extended previous scholarship by considering bystander intervention within the U.S. population as a whole, rather than among select subpopulations, such as college students (e.g. Burn, 2009; Bennett et al., 2013). This is an important distinction because both the prevalence of violence and the resources people have to intervene with may vary with characteristics such as socioeconomic status, race, and gender. The findings further extended existing scholarship by demonstrating that knowledge of and responses to violence differ for sexual assault and IPV even though the two types of violence are clinically quite similar and often co-occur.

We employed Latané and Darley’s (1970) model of intervention. We were able to empirically examine each step. The first is noticing violence. Respondents were more likely to know victims of IPV than victims of sexual assault, which may be due to differences in prevalence or victims’ willingness to disclose. We also found substantial demographic differences in knowledge of sexual assault and IPV within one’s social network; respondents wee more likely to know victims who were friends or family than distant members of their social network.

With regard to the second step, identifying the situation as an emergency, the second and third most commonly cited hypothetical barriers to intervening in IPV were concerns that IPV is a private matter and fear of misinterpreting the situation. However, women were less likely than men to report a belief that IPV is a private matter as a perceived barrier. Our analysis of demographic differences in personal knowledge of victims suggests that this may be attributable, in part, to the fact that women have more experiences knowing IPV victims. These findings highlight the need for public education campaigns that decrease the tolerance of interpersonal violence, increase perceptions of the danger and harm inflicted by sexual assault and IPV, and promote the normativity of intervening (particularly among men).

In order to intervene, individuals must take on the responsibility to act (step 3). We found that individuals did not respond equally to all victims; in instances of IPV, respondents were significantly more likely to intervene on behalf of a family member than on behalf of an acquaintance. This latter finding is consistent with previous research among college students, which suggests that individuals are more likely to intervene for friends than for strangers (Burn, 2009; Bennett et al., 2013). We also found that Hispanic and other non-white, non-black respondents were more likely to intervene in sexual assault than white respondents. This is consistent with findings from other studies, which suggest that certain cultural norms promote group welfare among Hispanic individuals (Rizo & Macy, 2011).

Once individuals decide to intervene, they must decide what to do (step 4). Our study indicates differences in the approaches individuals take when responding to sexual assault and IPV. They are more likely to involve legal authorities when responding to sexual assault, but more likely to tell the abuser to stop when responding to IPV. This finding indicates that some people may believe that sexual assault is a crime typically perpetrated by strangers (Littleton & Axom, 2003; Ryan, 1998) that should be dealt with by legal authorities, whereas IPV is not a crime but a family issue best addressed by those close to or within the family. To dispel these myths, public awareness campaigns should highlight that sexual assault more frequently occurs among individuals who know each other than among strangers (see Black et al., 2011), and should further provide specific suggestions for intervention strategies that do not put bystanders in immediate danger.

Although we did not explore them, it is worth noting that there may be demographic differences in how bystanders respond to sexual assault and IPV. For instance, research suggests that minority groups, especially African Americans, are more distrustful of the police because of discriminatory police practices and fears of police brutality (Tyler 2005; Weitzer & Tuch 2004). This distrust may translate into varying degrees of reluctance to involve public authorities in instances of sexual assault and IPV across racial/ethnic groups.

Several limitations should be acknowledged. One is that we do not have information on the timing of previous interventions. This introduces the possibility of recall bias, especially among older respondents, and leaves open questions for future research about whether the types of interventions people pursue differ contingent on their position in the life course. Relatedly, we do not know when respondents became aware of victims’ situations. Respondents who were made aware long after the violence subsided will have been unable to intervene in most of the ways listed in the survey, which may downwardly bias our estimates of intervention prevalence. A second limitation is that no questions were asked about how respondents became aware of victims’ experiences (e.g., hearsay versus witnessing) or about the characteristics of violent incidents (e.g. age of victims, severity of violence). A third limitation is that we cannot compare the barriers of IPV intervention and sexual assault intervention because questions about hypothetical barriers to the latter were not asked. Lastly, questions about sexual assault did not differentiate between assault within and outside of intimate partnerships. Respondents who knew victims who had experienced sexual assault via IPV may downwardly bias our estimated differences in knowledge and intervention strategies. Nevertheless, our estimates reflect the reality that sexual assault is often perceived distinctly from IPV but may still occur within relationships.

Despite its limitations, this study provides important new information on bystander behavior in the U.S. population at large and highlights several impediments to the effectiveness of bystander interventions as a violence prevention strategy. The findings reveal important demographic differences in personal knowledge of victims, previous intervention experiences, and the perceived barriers to intervening. Further, they call attention to the fact that the likelihood of intervening is conditional both on the relationship to victims and the type of violence in question. These particularities may influence the effectiveness of bystander intervention as a violence prevention strategy.

Biographies:

Abigail Weitzman is an NIA-funded Postdoctoral Fellow at the Population Studies Center at the University of Michigan. She is also an affiliate of the Population Research Center and Assistant Professor in the Sociology Department at the University of Texas at Austin. She received her PhD in Sociology from New York University.

Kate Walsh in an Assistant Professor at the Ferkauf Graduate School of Psychology at Yeshiva University and a former Postdoctoral Fellow in the Department of Epidemiology at Columbia University. She received her PhD in Psychology from the University of Nebraska-Lincoln.

Sarah Cowan is an Assistant Professor in the Sociology Department at New York University and a former Robert Wood Johnson Health and Society Postdoctoral Scholar at Columbia University. She received her PhD in Sociology and Demography from the University of California Berkeley.