To identify studies meeting selection criteria for this review, we first conducted searches of the following online databases between May and August of 2009 and repeated these searches in March and April of 2010 and May of 2012: PsycNet, PsycExtra, PubMed, ERIC, Sociological Abstracts, MEDLINE, Web of Knowledge, Dissertation Abstracts International, and GoogleScholar. Search terms included combinations of the following: (intervention, prevent*, program, effectiveness, efficacy or evaluation) and (perpetration, rape, rapist, sex*, coercion, violence, aggression, assault, offender, or abuse). Second, manual reviews of issues from relevant journals (i.e., Aggression and Violent Behavior, Journal of Adolescent Health, Journal of Interpersonal Violence, Journal of Women’s Health, Prevention Science, Psychology of Violence, Sexual Abuse: Journal of Research and Treatment, Trauma, Violence, & Abuse, Violence Against Women, Violence & Victims) published between January 2008 and May 2012 were also conducted to identify recent work in this area that may not have been cataloged yet in searchable databases. Third, to identify unpublished evaluation reports, solicitations were sent to relevant email lists and e-newsletters, including Prevent Connect, VAWnet, and the Sexual Violence Research Initiative. Fourth, for each article or report identified, we scanned the reference list to identify and retrieve additional reports that might meet inclusion criteria. During each of these iterative search steps, we were over-inclusive to ensure that all abstracts with the potential for inclusion were identified. The initial searches identified more than 10,600 reports, from which 330 were retained for full-text retrieval because they appeared to describe an outcome evaluation of a sexual violence prevention strategy.

Studies were eligible for inclusion if they examined the effectiveness of primary prevention strategies for sexual violence perpetration and were published in print or online between January 1985⁴ and May 2012. Journal articles, book chapters, and reports from government agencies or other institutions were included. Efforts were made to gather unpublished manuscripts, conference presentations, theses, and dissertations (see above). Because the focus on this review is to summarize the evidence base for the primary prevention of sexual violence perpetration, this review did not include studies that exclusively examined secondary and tertiary prevention approaches (e.g., treatment or recidivism prevention), strategies targeting victimization prevention (i.e., risk reduction), or etiological research. In order to avoid double-counting studies, existing reviews and meta-analyses of interventions for sexual violence prevention were excluded.

Only studies that compared one intervention condition to a no-treatment or waitlist control group (i.e., experimental and quasi-experimental designs) or that utilized a single-group pre–post design were included in this review, as the goal was to ascertain changes or differences in the outcomes following exposure to a specific treatment program. Thus, we excluded studies in which data from two different intervention groups were combined and compared to a control group as it was not possible to determine which intervention was responsible for any observed changes on the outcome measures. In addition, we excluded studies in which the intervention and the comparison conditions received different sexual violence prevention programs, because these studies examine the relative benefits of one program compared to another program as opposed to an individual program’s overall effectiveness relative to no intervention. Similarly, studies in which the comparison condition included a combined sample of control participants and participants who received a different sexual violence preventative intervention were also excluded. Because our focus was to examine the effectiveness of strategies to prevent sexual violence, studies that did not measure outcomes relevant to sexual violence perpetration were excluded (see below for a description of the outcomes included).

Of the 330 full-text reports retrieved, 226 reports were excluded. Reports were excluded because they did not describe an outcome evaluation study (45%; n = 101; e.g., review or meta-analysis, program description, theoretical paper, etiological research), did not measure sexual violence-related outcomes (11%; n = 25), evaluated a victimization prevention strategy only (10%; n = 23), did not evaluate a primary prevention strategy (8%; n = 18; e.g., sex offender treatment or recidivism prevention), did not utilize a research design with a comparison group or pre–post measurement (7.5%; n = 17), or met other exclusion criteria (8.1%; n = 27; e.g., non-English language). In addition, we identified several reports that described outcomes from the same study (e.g., a dissertation and a peer-reviewed journal article). In these cases, the peer-reviewed journal article was coded as the primary source and other reports were excluded as a duplicate report (3%; n = 7). In some cases, the excluded reports (e.g., dissertations) were used to provide supplemental information about the sexual violence prevention program or the evaluation design during the coding process. Numerous attempts were made to retrieve all reports identified in the initial searches, including contacting the first author directly and utilizing inter-library loan resources to obtain print copies. However, another eight reports (3.5%) identified through database searches could not be retrieved and were excluded as unavailable. These missing reports were nearly all dissertations and most were published more than 15 years ago; thus, this review may underrepresent these older dissertations.

Studies were classified as having either a rigorous or non-rigorous evaluation design. Rigorous evaluation designs included experimental studies with random assignment to an intervention or control condition (e.g., randomized controlled trial [RCT], cluster RCT) or rigorous quasi-experimental designs, such as interrupted time series or regression-discontinuity, for strategies where random assignment is not possible due to implementation restrictions (e.g., evaluation of policy). Other quasi-experimental designs (e.g., comparison groups without randomization to condition, including matched groups) and pre–post designs were considered non-rigorous evaluation designs, for the purposes of examining effectiveness in this review, consistent with standards of prevention science and evaluation research (e.g., Eccles, Grimshaw, Campbell, & Ramsay, 2003; Flay et al., 2005; Shadish, Cook, & Campbell, 2002).

In addition to design considerations, studies meeting criteria for a rigorous evaluation design were required to have at least one follow-up assessment beyond an immediate post-test assessment. Prior research has established the presence of a rebound effect on attitudinal and knowledge outcomes for sexual violence prevention programs wherein effects are seen immediately after the program but are not evident at longer-term follow-up (Anderson & Whiston, 2005; Brecklin & Forde, 2001; Carmody & Carrington, 2000). In addition, studies without a follow-up assessment often conducted the pre-test and the post-test measurement and the intervention all within the same session, increasing the potential influence of demand characteristics and test–retest effects. Thus, studies that did not include at least one follow-up measurement beyond immediate post-test, regardless of the research design, were also considered to be non-rigorous.

To identify prevention strategies with rigorous evidence of effectiveness, we developed criteria to classify specific interventions based on the strength of evidence of effectiveness for preventing sexually violent behavior. These criteria, illustrated in Fig. 1, emphasize sexual violence behavioral outcomes and rigorous experimental research designs that permit inferences about causality. Based on these criteria, interventions were placed into one of five categories: Effective for Sexual Violence Behavioral Outcomes includes those interventions with evidence of any positive impact on sexual violence victimization or perpetration in at least one rigorous evaluation. Interventions categorized as Not Effective for Sexual Violence Behavioral Outcomes were evaluated on sexual violence outcomes using a rigorous evaluation design and had consistently null effects on those measures. Interventions categorized as Potentially Harmful for Sexual Violence Behavioral Outcomes include those with at least one negative effect on sexually violent behavior in a rigorous evaluation. Interventions categorized as More Research Needed included those with evidence of positive effects on sexual violence behavior in a non-rigorous evaluation or positive effects on sexual violence risk factors or related outcomes in a rigorous evaluation. Interventions were considered to have Insufficient Evidence if they were not published in a peer-reviewed journal or formal government report, if they measured outcomes at immediate post-test only without a longer follow-up period, if they found null effects on sexual violence behavioral outcomes using a non-rigorous design; and/or if they only examined risk factors or other related outcomes using a non-rigorous design (regardless of the type of effect).

We attempted to identify and combine findings from multiple studies or reports examining the same intervention based on the program name or description and used outcomes from the most rigorous evaluation(s) available to categorize the program’s effects. In some cases, researchers may have evaluated modified versions of the same program over time; findings from these evaluations were considered together if the program name did not change and there were no indications that modifications to the structure or content of the program model over time substantially altered the core content or strategy.

Evaluation of sexual violence perpetration prevention programs peaked in the late 1990s and again in 2010 and 2011 (see Fig. 2). Table 1 describes characteristics of the 140 studies and interventions, including the research design, study population, intervention length, setting, participant and presenter sex, and mode of delivery. Notably, almost two-thirds (n = 84; 60%) of the included studies examined one-session interventions with college populations; these programs had an average length of 68 min. The majority of studies utilizing pre–post designs measured outcomes at immediate post-test only (n = 13, 56.5%). Studies with quasi-experimental designs measured outcomes most often at post-test (n = 12, 34.3%) or with a follow-up period of one month or less (n = 10, 28.6%). In contrast, evaluations using experimental designs had the lowest proportion of studies with post-test only outcomes (n = 19, 23.2%) and the highest proportion with follow-ups at 5 months or longer (n = 17, 20.7%).

To examine changes in evaluation methodology over time, we compared studies published in 1999 or earlier (n = 73; 52.1%) to those published in 2000 or later (n = 67; 47.9%). Before 2000, 63% (n = 46) of published studies were RCTs, 30.1% (n = 22) used quasi-experimental designs, and 6.8% (n = 5) used pre–post designs; 28.8% (n = 21) assessed outcomes at immediate post-test only and only 6.8% (n = 5) followed participants for 5 months or longer. Since 2000, 53.7% (n = 36) of published studies were RCTs, 19.4% (n = 13) were quasi-experimental, and 26.9% (n = 18) were pre–post designs; 34.3% (n = 23) of these studies measured outcomes at immediate post-test only, but another 26.9% (n = 18) of studies assessed outcomes after at least 5 months.

Table 2 summarizes patterns of intervention effects by study characteristic and outcome types. Studies with mixed effects across outcome types and follow-up periods were most common (41.4%; n = 58). More than one-quarter of studies (27.9; n = 39) reported only positive effects and another 21.4% (n = 30) reported only null findings. Nine studies (6.4%) had at least one negative finding suggesting that the intervention was associated with increased reporting of sexually violent behavior (Potter & Moynihan, 2011; Stephens & George, 2009), rape proclivity (Duggan, 1998; Hillenbrand-Gunn, Heppner, Mauch, & Park, 2010), or attitudes toward sexual violence (Echols, 1998; McLeod, 1997; Murphy, 1997). Peer-reviewed studies and government reports tended to have positive or mixed findings more often than dissertations and unpublished manuscripts. Examination of outcomes by study design suggested that evaluations employing more rigorous methodologies (i.e., experimental or quasi-experimental designs with comparison groups) were less likely to identify consistently positive effects than studies using a pre–post design. Similarly, studies that examined outcomes at immediate post-test only were more likely to identify positive effects than studies with a longer follow-up period.

Looking at the pattern of intervention effects by outcome type, results suggest that null effects were more common and positive effects less common on sexually violent behavior and rape proclivity outcomes than on other outcome types. Specifically, about half of all studies measuring sexually violent behavior or rape proclivity found only null effects (47.6%; n = 10); very few studies (4.8%; n = 4) reported only significant, positive effects on these main outcomes of interest. In contrast, the majority of studies measuring knowledge, bystanding behavior or intentions or skills found consistently significant positive effects on these outcomes. No clear pattern was evident for studies assessing attitudinal or affective/arousal outcomes.

To examine the potential impact of intervention length, we estimated the average intervention exposure (i.e., sessions × length) for studies with positive, mixed, negative, and null effects. Findings indicate that interventions with consistently positive effects were about 2 to 3 times longer, with an average length of 6 h (SD = 11.4), than interventions with mixed (M = 3.2 h; SD = 6.6), negative (M = 2.2 h; SD = .9), or null (M = 2.8 h; SD = 4.3) effects.

As shown in Table 3, only three interventions (based on 3 studies; 2.1%) were categorized as effective for sexual violence behavioral outcomes: Safe Dates (e.g., Foshee et al., 2004, 2005), Shifting Boundaries building-level intervention (Taylor, Stein, Mumford, & Woods, 2013; Taylor et al., 2011), and funding associated with the 1994 U.S. Violence Against Women Act (Boba & Lilley, 2009). Five interventions (based on 11 studies; 6.4%) were found to be not effective for sexual violence behavioral outcomes and three interventions (based on 2 studies; 2.1%) reported evidence suggesting that they were potentially harmful. Another ten interventions (based on 17 studies; 12.1%) were categorized as needing more research in order to understand their effects. Findings within each of these categories are discussed below. The majority of studies reviewed (n = 108; 77.1%) provided insufficient evidence to adequately evaluate the effectiveness of the intervention for preventing sexual violence; these studies were unpublished manuscripts or dissertations which had not been subjected to independent peer review (n = 53; 38%), measured outcomes at immediate post-test only (n = 57; 41%), and/or examined only risk factors or related outcomes for sexual violence using a non-rigorous design (n = 71; 51%). Interventions with insufficient evidence are not included in Table 3 due to the large number of studies in this category and the lack of practical value for this information when the findings are inconclusive.

In the last three decades, a sizable literature has emerged examining the effectiveness of strategies to prevent sexual violence perpetration with more than 100 evaluation reports identified since 1985. The number of studies published in the last two years of this review increased notably, suggesting a possible resurgence of research interest in this area. However, our results suggest that the sexual violence prevention evaluation literature has not seen a steady increase in publications over time to mirror the large increases in other types of sexual violence research. A bibliometric analysis of sexual violence research found that publications with the keywords “rape,” “sexual assault,” or “sexual violence” increased over 250% between 1990 and 2010, from approximately 5990 citations in 1990 to about 15,400 citations in 2010 (Centers for Disease Control & Prevention, 2012). Despite this marked increase in general research attention to sexual violence, the current review suggests that the prevention evaluation literature has remained relatively stagnant both in terms of quantity and quality. In part, this trend may reflect the relatively limited resources available during this period for development and rigorous evaluation of sexual violence primary prevention approaches (Jordan, 2009; Koss, 2005). Fortunately, funding for sexual violence evaluation research has increased over the last decade. For example, CDC funded 27 research projects with a focus on sexual violence between 2000 and 2010, resulting in the increased availability of more than $19 million in federal funding for the field; more than half of these projects involved prevention evaluation research (Centers for Disease Control & Prevention, 2012; DeGue, Simon, et al., 2012). Although this funding represents a large proportional increase in federal dollars available for sexual violence research, the total research funding available remains low compared to other forms of violence and other areas of public health (Backes, 2013; DeGue, Massetti, et al., 2012).

In addition to limiting the quantity of evaluation research studies, fiscal constraints may have also resulted in less rigorous research designs, as large randomized controlled trials of prevention strategies are generally considered costly to implement. Indeed, this review found two-thirds of the evaluation studies conducted over nearly 30 years examined brief, one-session interventions with college populations, approaches that are relatively inexpensive to implement and evaluate. In terms of measurement, few of these studies (n = 11) measured sexually violent behavior, and none found consistently positive effects on these key behavioral outcomes. Of course, the predominance of brief awareness and education strategies in the literature not only reflects resource limitations for research but also implementation challenges in the field. Many colleges may limit access to students to only one class period or have policies requiring only 1 h of relevant training—spurring the development of programs to fit this need. Nevertheless, future research is needed that rigorously evaluates a more diverse and comprehensive set of prevention approaches with various populations.

Although the vast majority of preventative interventions evaluated to date have failed to demonstrate sufficient evidence of impact on sexual violence perpetration behaviors, progress is being made. Findings from several large, federally-funded⁷ effectiveness trials of comprehensive, multi-component primary prevention strategies have been published more recently, with interventions targeting a broader, and younger, segment of the population (e.g., Foshee et al., 2004, 2012; Miller et al., 2012b; Taylor et al., 2013) with additional evaluations underway (e.g., Cook-Craig et al., in press; Espelage, Low, Polanin, & Brown, 2013; Tharp, Burton, et al., 2011). This new research is providing the primary prevention practice field with additional evidence on which to base decisions about resource allocation and implementation in order to prevent sexual violence. However, as we discuss below, more rigorous evaluation research on various prevention approaches is needed before we can expect to see measurable reductions in sexual violence at the population level.

Emphasizing rigorous evaluation and behavioral outcomes, we developed and applied a set of criteria to identify specific interventions with more or less evidence of effectiveness for the primary prevention of sexual violence perpetration in order to serve as a guide for decision-making. Communities and organizations are increasingly interested in and required to implement evidence-based interventions with an expectation of achieving reductions in sexual violence. Table 3 is intended to serve as a resource and tool for this purpose. Although we believe that this approach has many practical advantages, it has notable limitations as well. Most importantly, it is limited by the ever-growing and evolving nature of the evaluation research literature. Over time, additional effective interventions will be identified, some will be found to be ineffective, and others will find that their effects can be replicated—or not—in different populations. The current review provides only a snapshot of knowledge regarding “what works” currently to prevent sexual violence. Practitioners are encouraged to consider this information in the context of the needs, goals, and resources of their organization and to supplement this summary with additional information about the strategy and new research findings as they become available. This summary may also be useful in identifying promising strategies in need of further research or when developing new comprehensive strategies that combine the strengths of multiple evidence-based approaches. Future research investments should reflect the best available science and theory, and move beyond approaches that have proven ineffective or insufficient.

There have been substantial gains in the field of sexual violence prevention over the last 30 years with regard to public education and awareness, legal protections for victims, federal funding and infrastructure for prevention work, community mobilization, and research on the prevalence, etiology, and prevention of sexual violence. But important gaps remain, hindering progress toward the ultimate goal of reducing sexual violence at the population level. Rates of sexual violence remain alarmingly high, and we still know very little about how to prevent it.

The field’s ability to identify effective strategies for sexual violence perpetration is severely constrained by the quality of the available research. Without more rigorous research designs that examine the primary behavioral outcomes of interest, it is not possible to determine with sufficient confidence whether a strategy is likely to have the desired outcomes or be cost-effective. Evaluation research need not always involve an RCT; other rigorous quasi-experimental designs, such as regression discontinuity or interrupted time series, may be the most practical options for evaluating policy or environmental strategies, for example. In contrast, less rigorous designs, such as pre–post studies, may be the best approach for formative research to develop and refine strategies. Likewise, measurement of risk characteristics provides important information about potential mediators and moderators of effectiveness, but without identification of true causal risk factors, these outcomes cannot tell us whether or not a program “works.” Thus, measurement of key behavioral outcomes, including perpetration behavior, is a critical component of rigorous effectiveness research. Of course, the use of RCTs and behavioral measures represent only the minimum criteria necessary to allow for causal inferences from the data; these design features alone do not ensure that a study is well-conducted or reliable. Additional factors, such as sample size and retention, measurement validity, group equivalence, and appropriate data analysis, are also important in determining whether study findings represent valid evidence of effectiveness (Coalition for Evidence-Based Policy, 2010). Thus, a critical priority for the field to ensure the growing availability of effective, evidence-based prevention strategies for sexual violence involves improving study rigor. The limited available resources for prevention should be directed toward methodologies most likely to advance practical knowledge of what works.

There is also a need in the field to consider not only statistical significance, but also the magnitude or clinical significance of any effects identified. If a strategy is widely implemented, even a small effect on perpetration behavior may have a large impact. However, a small effect on an attitudinal or knowledge outcome, for example, may not have any practical value. One limitation of this review is that we examined this field as it is — categorizing outcomes by the direction of effect rather than by the size of the effect. The broad scope of the current review and the wide variability in the quality and design of included studies made meta-analytic techniques impractical. However, prior meta-analyses conducted with smaller subsets of the literature have found relatively small effects, especially on the most direct or proximal outcomes (Anderson & Whiston, 2005; Brecklin & Forde, 2001; Flores & Hartlaub, 1998). As noted by these prior reviews (Breitenbecher, 2000; Schewe & O’Donohue, 1993), more attention to issues of clinical significance is needed within the sexual violence evaluation literature to better understand the value and potential impact of specific strategies when applied at the population-level.

The lack of effective prevention strategies for sexual violence is due not only to a lack of rigorous evaluation to identify those effects but also to the nature and quality of the approaches being developed and evaluated. A key conclusion from this review is that a large portion of research (and, presumably, programmatic) resources, to date, have been invested in brief psycho-educational strategies that are not consistent with the principles of prevention and have not demonstrated effectiveness despite numerous evaluations. Prevention strategies based in a coherent theory of change with a plausible likelihood for impact on sexual violence perpetration and addressing a broader range of risk and protective factors for sexual violence may be more likely to be effective. With most of the attention in existing programs focused on knowledge and attitudes, many sexual violence risk factors—well-grounded in theory—have been ignored. For example, childhood exposure to violence, general delinquency and aggression, and early sexual behavior have consistent empirical support across numerous studies and are included in the Confluence Model, a well-supported theoretical model of sexual violence perpetration, but they are rarely addressed in sexual violence prevention efforts (Tharp et al., 2013). Explicit attention to an expanded range of risk factors in intervention development and a broader set of behavior change theories, such as those identified by Paul and Gray (2011), may result in more integrative and effective models of prevention.

The lack of community- and societal-level prevention approaches for sexual violence perpetration also remains a critical gap in this field. The social–ecological model (Bronfenbrenner, 1977) conceptualizes violence as a product of multiple, interacting levels of influence at the individual, relationship, community, and societal levels of the social ecology. Most prevention strategies evaluated thus far have focused solely or primarily on creating change at the individual level, with a few also addressing peer influences or small-scale social norms change through bystander intervention or school poster campaigns. Individual-and relationship-based approaches are likely key pieces of the prevention puzzle given the plethora of risk correlates identified at these levels. But, achieving long-term behavior change with such programs is unlikely when they are delivered in a social, cultural, or physical environment that counteracts those messages and discourages safe, healthy behaviors or rewards violent behavior (DeGue, Holt, et al., 2012). This would necessitate drawing on theory and lessons learned in other areas of public health to identify innovative policy, environmental, and structural approaches that support and encourage healthy behavior, positive social norms, and non-violence. Such strategies would target modifiable risk and protective factors that are characteristic of communities and that are empirically or theoretically associated with sexual violence (e.g., neighborhood disorganization, availability of alcohol; Casey & Lindhorst, 2009; DeGue, Massetti, et al., 2012; Tharp et al., 2013). In late 2013, CDC released a funding opportunity announcement to encourage innovation in this area by providing potential funding for the rigorous evaluation of such an approach (See RFA-CE-14-005 on www.grants.gov). More work is needed to develop and identify community-level measures, indicators, or proxies of sexual violence beyond law enforcement record data for use in evaluating these outer-level approaches (DeGue, Holt, et al., 2012).

Continued progress is needed toward the development and rigorous evaluation of effective, comprehensive, theory-based primary prevention strategies for sexual violence perpetration that address risk and protective factors at multiple levels of the social ecology. This progress is dependent on the innovation and methodological expertise of sexual violence researchers and program developers, well-directed research funding, and support from prevention professionals implementing these strategies in the field. Sexual violence is a complex problem with social, structural, cultural, and individual roots. By designing prevention efforts that are equally complex, multifaceted, and embedded within our lives and environments we can enhance their effectiveness. With increasing pressure to demonstrate effectiveness and economic efficiency, stronger evidence of impact will be required to justify investments in evidence-based primary prevention strategies. Thus, further investment in rigorous evaluation research is critical to ensuring sustained movement toward the identification of evidence-based strategies for the prevention of sexually violent behavior. Such research should focus on comprehensive, theory-based strategies across levels of the social ecology and build on the best available research evidence to identify a complement of effective approaches for implementation and move us closer to ending sexual violence in communities.