Violence prevention interventions for women: a 2000–2024 evaluation review

The global evidence base for preventing violence against women has reached a turning point. A sweeping new review published in SSM – Mental Health analyzes a quarter century of intervention research and delivers a conclusion that is at once sobering and deeply practical: violence against women is preventable, but only about one in three evaluated programs actually works, and the single most powerful predictor of success is whether an intervention teaches people concrete skills for managing conflict in their relationships. The review, led by Rachel Jewkes of the South African Medical Research Council and the University of Exeter, together with Leane Ramsoomar, Pinky Mahlangu, Esnat Chirwa and Andrew Gibbs, examined every quantitative evaluation published between January 2000 and December 2024 that compared an intervention arm against a control group and measured physical or sexual intimate partner violence or non-partner rape.

The scale of the undertaking is considerable. The researchers identified 155 papers describing 185 distinct intervention arms, each representing an intervention tested against a counterfactual. To put that in perspective, the field has accelerated dramatically: 60 percent of all controlled evaluations ever conducted were published in just the seven years after 2018, compared with the 19 years from 2000 to 2018 that produced the remaining 40 percent. The median publication year was 2019. Yet the geographic distribution of this growing literature is strikingly lopsided. Sub-Saharan Africa accounts for 51.4 percent of all studies, North America for 23.2 percent, and Central and South-East Asia for 14.1 percent. The remaining regions — Latin America and the Caribbean, the Middle East and North Africa, East Asia and the Pacific, and Europe — together contribute barely a tenth of the global knowledge base, with Europe represented by only three studies. This means that policy makers in much of the world are being asked to act on evidence generated in very different social, economic and cultural contexts from their own.

The methodological core of the review is a bespoke quality-scoring instrument, developed by the authors because existing checklists such as CONSORT were judged either too vague or too cumbersome for this specific purpose. Each study was scored on six dichotomized criteria, weighted equally: sample size, with fewer than 300 participants per arm considered potentially underpowered; the use of an established multi-item measure of physical or sexual IPV or rape perpetration and victimization, such as the WHO instrument, the Demographic and Health Surveys module or the revised Conflict Tactics Scale; the presentation of baseline and endline prevalence figures by arm; the interval between baseline and endline assessment, with less than six months considered inadequate; the rate of loss to follow-up, with 40 percent or more treated as a serious bias risk; and whether the outcome measure cleanly captured acts of physical or sexual IPV rather than a composite score blending easier-to-change items. An intervention was classified as successful if it produced a statistically significant reduction, at p ≤ 0.05, in at least one measure of physical or sexual IPV or non-partner rape at the final follow-up point.

The headline result is that 71 of the 185 intervention arms — 32.4 percent — reduced violence against women on at least one measured indicator, with an additional 6.5 percent showing mixed results across subgroups. In 61.1 percent of evaluations, the intervention made no measurable difference. Success rates varied by region, from 39 percent in sub-Saharan Africa to just 18.6 percent in North America, though the difference was not statistically significant. Perhaps more revealing than the raw success rate is what predicted it. In multivariable logistic regression models adjusted for study quality, interventions that included a component addressing poor conflict and relationship skills — teaching communication techniques such as assertive expression, conflict de-escalation and the defusing of violence triggers — were more than three times as likely to reduce violence against women, with an adjusted odds ratio of 3.35 (95 percent confidence interval 1.58 to 7.11, p = 0.002). In sensitivity analyses using a stricter definition of success, that effect grew even stronger, reaching an adjusted odds ratio of 4.11.

This finding sharpens the field’s understanding of why certain celebrated programs work. Many of the most successful interventions to date — Stepping Stones in South Africa, Indashyikirwa in Rwanda, MAISHA in Tanzania — have been manualized, group-based programs using participatory learning methods, and nearly half of all evaluated interventions (48.1 percent) include such a component. But the new analysis suggests the active ingredient is not the participatory format itself so much as the skills-building content that targets the immediate relational context in which violence occurs, a point situated between the distant structural drivers of violence and the individual characteristics of women that so many programs have focused on. Notably, interventions that addressed only women’s disempowerment, without a critical gender analysis or skills component, showed an adjusted odds ratio below one, suggesting that empowerment rhetoric alone is likely an ineffective strategy.

The review also carries a warning about settings where prevention is most desperately needed. Interventions delivered in conflict-affected and humanitarian settings were dramatically less likely to succeed, with an adjusted odds ratio of 0.08 (p = 0.021) — meaning such programs were more than twelve times less likely to demonstrate impact than those in stable settings. Only 12 studies have ever been conducted in these environments, almost all in sub-Saharan Africa, despite the high prevalence of violence against women in crisis-affected regions. The authors conclude that current intervention models, largely designed for stable communities, do not adequately account for what is feasible in highly abnormal settings, and they call for a fundamentally new generation of programming for humanitarian contexts. Similarly, although more than a billion people worldwide live in urban informal settlements, only 11 studies have been conducted with these populations.

Economic interventions, long a favorite of the development sector, received a more nuanced verdict. Cash and asset transfers featured in 17.8 percent of interventions, savings or livelihood training in another 17.8 percent, and microfinance or loans in 5.4 percent, yet none of these approaches was individually associated with a significantly higher likelihood of success in the main analysis. However, a sensitivity analysis hinted at something important: when the analysis was adjusted for study context, interventions addressing poverty as a risk factor showed a significant association with success (adjusted odds ratio 2.70, p = 0.024). The authors interpret this cautiously, noting that poverty may matter most when programs also build relationship and conflict skills — a combination that reflects the structural, multi-layered model of violence risk that has animated much of the research in low- and middle-income countries.

The review also documents a troubling mismatch between who perpetrates violence and who is recruited into prevention programs. Although men and boys commit the vast majority of violence against women, only seven studies exclusively enrolled male participants, all published before 2019. Just over a third of interventions included men or boys at all, and half of those recruited men alongside their female partners as couples — a format used exclusively in sub-Saharan Africa and Central and South-East Asia that showed promising, if not statistically definitive, signs of effectiveness. Female sex workers, who face elevated violence from clients, police and partners alike, were the focus of only seven evaluations, and the effective ones mostly reduced violence from clients rather than intimate partners, revealing a substantial innovation gap. Only five studies combined the prevention of violence against women with the prevention of violence against children, despite the frequent co-occurrence of the two within households.

Quality of evaluation science itself emerged as a significant variable. Higher-quality studies were significantly more likely to report intervention success, with each point on the six-item quality score raising the odds of a positive finding by roughly 40 percent (adjusted odds ratio 1.43, p = 0.009). The authors read this partly as a reflection of genuine program strength — better-resourced, better-designed interventions tend to be better evaluated — but also as a caution about the literature’s weaker end. Nearly a quarter of studies had between 50 and 300 participants per arm, a range unlikely to detect effects without substantial risk of type I or type II error, and 11.9 percent were clear pilot or feasibility studies with 50 or fewer per arm. Some investigators, the authors note, recommended interventions for full-scale trials even when their small studies showed no hint of effect, a practice they argue must stop if the field is to advance efficiently. Reporting standards also varied widely across disciplines, with many studies omitting baseline and endline prevalence figures or estimated effect sizes, making cross-study comparison difficult.

The review’s conclusion offers practitioners and funders a clear, actionable hierarchy of priorities. First, build relationship and conflict management skills into every prevention program, because this is the factor most consistently and strongly associated with reducing violence. Second, treat poverty as a plausible target warranting further research, especially in combination with skills components. Third, invest in rigorous, adequately powered evaluations with follow-up periods of a year or more, since even the strongest programs in the field — such as Rwanda’s Bandebereho couples intervention, whose effects on violence persisted six years after baseline — were validated only through long-term, well-funded trials. Fourth, recognize that no approach is a magic bullet: community mobilization, participatory group programs, group counselling, microfinance-linked programming and professionally delivered counselling all have examples of success, while stand-alone individual psychotherapy, cash-only transfers, life skills training, safety planning and web-based tools currently enjoy weaker support. With 92 interventions across the global literature now rated in the top quality tiers and 71 demonstrably effective, the field has, for the first time, enough evidence to say not merely that violence against women can be prevented, but which ingredients make prevention succeed.