What factors are associated with recent intimate partner violence? findings from the WHO multi-country study on women's health and domestic violence

Picture yourself dropping a pin on a world map and asking a simple, brutal question: in this place, how often does violence happen inside a relationship? The answer swings wildly. In the World Health Organization's multi-country study, lifetime physical or sexual violence against ever-partnered women ranged from about one in seven in an Ethiopian province to nearly three in four in Japan. For the past year, the spread was four percent at the low end and more than half at the high end. Those are not small fluctuations. They are entire worlds apart. And underneath those numbers sit the consequences we know too well: injuries and chronic pain, depression and post-traumatic stress, disrupted pregnancies and infections, a shadow that stretches across health and life. Now, if we want to understand why risk is higher in one place and lower in another, the first rule is simple: make the measurements the same. Past research struggled here. Different surveys asked different questions in different ways, and interviewer effects—what's asked, how it's asked, who asks it—can change what gets disclosed. Even the big Demographic and Health Surveys, when they added a domestic violence module, allowed a lot of country-level tailoring. Abramsky and colleagues designed this WHO study to push in the opposite direction. Same questionnaire everywhere. Same training for interviewers. Same field procedures. Then watch what patterns line up across countries, and which ones don't. They built the sample to be population-based and comparable: ten countries and fifteen sites. Capitals, big cities, rural provinces—some countries had two locations, others one because of cost and logistics. Female interviewers visited sampled households and spoke with one randomly chosen woman between 15 and 49. In total, twenty-four thousand ninety-seven interviews were completed in 14 languages. Ethics were front and center: privacy safeguards, careful protocols for safety and referrals, and approvals from the WHO's committee and national boards. This is not a trivial point. In this kind of work, protecting respondents is part of getting the science right. The study focused on ever-partnered women and on recent events. Women reported whether specific acts of physical or sexual violence had occurred, and if they had happened in the last 12 months. For analysis, the team compared two groups: women who reported past-year intimate partner violence—what they called current violence—and women who had never reported partner violence. They excluded women who had experienced violence in their lifetime but not in the past year, to avoid washing out differences by mixing in past events with current risk. Of the nineteen thousand five hundred seventeen women who had ever had a partner, that left fifteen thousand two hundred seven for the "prior to relationship" models and fifteen thousand fifty-eight for the "current situation" models. Here's the clever bit in the analysis. They split predictors into two time bins: factors that occurred before the relationship began—things like childhood abuse or education—and factors inside the current relationship—like union type or alcohol use. That separation helps keep early-life exposures from being masked by later-life circumstances that might lie on the causal path. At each site, they ran logistic regression models, first bivariate and then multivariate, to estimate odds ratios. If you need a quick translation: an odds ratio above one signals higher odds of violence; below one signals protection. They treated changes smaller than about five percent—odds ratios between 0.95 and 1.05—as no meaningful association. Clustering within sites turned out to be small, with intra-class correlations under 0.06, so they presented results without extra clustering adjustments. What emerged, across a very diverse set of places, is a tight core of factors that keep showing up. Education is the headline. When both partners had completed secondary school, the odds of past-year violence dropped in twelve of the fourteen sites that had data, and in ten of those, the drop was statistically solid. That "both educated" group also tended to have the lowest odds among all education combinations in ten of fourteen sites. When only one partner had reached that level, the signal was still generally protective, but not as consistent. In parallel, higher household socioeconomic status—think better material conditions at the household level—tracked with lower risk in fourteen sites, with around eight showing clear significance when comparing the highest to the lowest group. Together, those two paint a picture: schooling and household resources seem to change something fundamental in how relationships navigate conflict. If that's the protective side, the risk side has a standout too: alcohol. Harmful drinking by a partner was linked with higher odds of violence in every site that measured it—fourteen out of fourteen—with significance in twelve. When the woman herself had alcohol problems, risk rose in ten of eleven sites, with about half showing statistical significance. And when both partners had alcohol problems, the risk was higher everywhere—eleven out of eleven—with seven significant. A detail worth holding onto: frequent drunkenness among men typically produced larger jumps in risk than problematic drinking by the woman, and the combination—both partners drinking heavily—was worse than either alone in the vast majority of sites. It's a classic dose-response pattern, and it lines up with what many clinicians and advocates see on the ground. Beliefs matter too. In relationships where attitudes that justify a husband beating his wife were present, violence was more likely. That pattern appeared in thirteen of fifteen sites, and it cleared the significance bar in eight. That's not a culture-blame story; it's a norms-shape-behavior story. When violence is seen as acceptable under certain conditions, it happens more. Then there's the past. Childhood adversity—being abused as a child, witnessing your mother being abused—left a long shadow. In bivariate analyses, childhood sexual abuse was associated with current partner violence in fourteen of fifteen sites. It remained important in multivariate models that accounted for other factors. And again, concordance amplified risk: when both the woman and her partner had been abused in childhood, the odds of current violence were particularly high in roughly eleven of fourteen sites. You can hear the echo—violence taught early is replayed later—unless something interrupts the pattern. Age played a role, though more quietly. Younger women tended to report higher past-year violence in twelve sites. The partner's age showed similar hints in simple analyses but overlapped too heavily with the woman's age to be modeled together. Other life-course markers added texture: women exposed to non-partner violence as adults, or with partners who'd been in fights with other men, faced higher risk in many sites. And when partners had outside relationships, risk jumped. In several places, the adjusted odds ratios for a partner's concurrent relationships hovered around two and a half, with confidence intervals that kept them clearly above one. How couples are structured and how they live together also seemed to matter. Cohabiting without formal marriage tended to be riskier than being formally married in most sites that could look at it—ten of twelve showed higher risk for cohabiting, with three statistically clear. Interestingly, not living with a partner at all was often linked with lower risk—eight of nine sites went in that direction, though only a couple were significant. In settings where polygamy is practiced, having a husband with more than one wife raised risk in every site that asked the question, with three of six showing significance. You can read these results as signals about commitment, stability, and trust breaches inside unions—none of them perfectly captured by a single variable, but together pointing toward dynamics that elevate or dampen conflict turning into assault. Urban versus rural? There wasn't a single story there. Some places showed differences by setting, others didn't, and the directions varied. It's a reminder that the urban-rural label bundles a lot of different exposures—income opportunities, density, anonymity, social control—and those bundles differ by country. One theme kept surfacing across these domains: concordance. When both members of a couple shared a protective factor—say, both had finished secondary school—the odds of violence were lower than when only one had. Flip it, and when both carried a risk factor—both had alcohol problems, both had been abused in childhood—the odds were higher than either alone. It sounds intuitive, but the consistency across countries gives it real weight. In prevention, that points you toward couple-level strategies that try to shift both sides of the dyad at once. Let's pause on what this study could and couldn't do. It was cross-sectional—one slice in time. That means you can't nail down what came first for every association, and you have to be careful about causal language. Some groups were small, especially at extreme exposures, which makes estimates wobbly. The team relied on women's reports about their partners' characteristics, which can introduce misclassification, though if anything that often dilutes associations rather than making false ones appear. And the analysis made a choice to present results unadjusted for clustering because clustering looked small. All that said, the comparability across fifteen sites is a genuine strength, as is the careful split between early-life and current relationship factors. If you pull back and look for the common threads that travel well across borders, a short list stands out. Keep girls and boys in school through secondary education. Across settings, that was the most consistent protective signal, and it was strongest when both partners had it. Lift household socioeconomic conditions, because they track with lower risk, even if part of that association likely runs through the same factors we just named. Change norms that make violence seem permissible; in this data, those attitudes marked higher risk almost everywhere. Treat harmful drinking like the preventable and reversible risk factor that it is; when one or both partners drink heavily, the odds of harm go up. And cut the intergenerational cord by preventing and responding to childhood abuse; the signal from early trauma to adult partner violence is strong and stubborn. There's also a humility embedded in these findings. Some associations vary by place. Formal marriage tended to be protective relative to cohabitation in many sites, but not all. Urban-rural differences weren't uniform. And while the patterns with alcohol, attitudes, education, and childhood trauma were broad, they weren't absolute. That argues against one-size-fits-all programs. It points instead to packages that are anchored in consistent levers—schooling, norms, alcohol, early trauma—but tuned to the local context. So where does this leave us? With a map that's both sobering and actionable. Sobering because the burden of intimate partner violence remains high in too many places, and the spread between sites is enormous. Actionable because several levers show up again and again, and they are things societies can actually move. As Heise and Jewkes framed years ago, risk lives not in one person alone but in the space between two people, shaped by what each brings from before the relationship, their standing within it, and the norms around them. Abramsky and colleagues put numbers on that idea across fifteen very different places, and the numbers line up. Change what couples bring into that space—education, sobriety, beliefs about violence—and you change what can happen inside it.