The case that cash curbs partner violence rests on fragile surveys

Every two months, Peru's Juntos program pays about $60 to households below a poverty threshold. A new NBER working paper by Jorge Agüero, Ursula Aldana, Erica Field, Veronica Frisancho, Ignacio Rodriguez Hurtado and Javier Romero uses that rhythm to ask an uncomfortable question about how we know what we know. When women in rural and peri-urban Peru answered questions about sexual violence by their partners in the standard way, aloud to an interviewer, reported rates hovered around 10% whatever the time since the last payment. When women marked their answers privately on a ballot, reported rates rose in the weeks after a payment, to around 20%, roughly double, and then fell back. The transfers were the same; only the way of asking changed.

The problem

Few findings in development economics have been as welcome as the evidence that cash transfers to women reduce intimate partner violence. It has been used to argue that social protection does more than relieve poverty: it shifts power inside the home. Almost all of that evidence, however, rests on women telling an enumerator, face to face, whether a partner has hit, threatened or forced them. Our question is narrow: how much of the conclusion that cash transfers reduce partner violence survives once we take seriously how violence is measured? It matters because underreporting unrelated to a program only adds noise, while underreporting that moves with the program, or with the cash it delivers, can bias estimates in either direction.

What the evidence shows

Start with the averages. A mixed-method review by Buller, Peterman, Ranganathan, Bleile, Hidrobo and Heise identified 14 quantitative and eight qualitative studies from low- and middle-income countries; 11 and five of them, respectively, found that transfers decrease violence, and only two showed mixed or adverse effects. A meta-analysis by Baranov, Cameron, Contreras Suarez and Thibout reached a similar conclusion: transfers are associated with less violence or no effect on average, with significant reductions in physical and emotional violence and controlling behaviors, though some studies find increases for subgroups, such as women with little schooling whose husbands have even less. Theory allows either sign. Money can ease stress and conflict over scarce resources and strengthen a woman's position if she could leave. It can also provoke a partner who feels his status threatened, or who uses violence to extract the money. Buller and co-authors group the channels into economic security and emotional well-being, intra-household conflict, and women's empowerment.

The individual studies are better identified than most in the field. In northern Ecuador, a randomized trial by Melissa Hidrobo, Amber Peterman and Lori Heise found that cash, vouchers and food transfers to women reduced controlling behaviors and physical or sexual violence by 6 to 7 percentage points, with no difference between modalities. In Kenya, Johannes Haushofer and co-authors found that unconditional transfers to women averaging $709 reduced physical and sexual violence by 0.26 and 0.22 standard deviations, and that physical violence also fell against women in treated villages who received nothing. In Bangladesh, Roy, Hidrobo, Hoddinott and Ahmed found that six to ten months after a program ended, women who had received transfers alone reported no less violence than a control group, while those who had also received nutrition behavior-change communication reported 26% less physical violence. Effects depend on design, and may not outlast the payments.

Latin America adds warning signs. Using a specialized survey of Mexico's Oportunidades program, Bobonis, González-Brenes and Castro found beneficiary women 40% less likely to suffer physical abuse but more likely to receive violent threats without abuse. With the same experimental data, Angelucci found that small transfers reduced violence by 37%, while large ones increased aggressive behavior among husbands with traditional views of gender roles. In Peru, Juan-José Díaz and Victor Saldarriaga exploited the staggered rollout of Juntos across municipalities, together with its eligibility rule, in a difference-in-differences design, which compares changes over time in places the program reached with changes in places it had not yet reached. They found that Juntos reduced physical partner violence by 25–30%, apparently through less frequent drinking and alcohol-related aggression by male partners. Their data come from the Demographic and Health Surveys, which ask face to face.

That is where measurement bites. Survey researchers have tried two fixes. List experiments offer full anonymity: a woman reports only how many statements on a list apply to her, and prevalence is inferred by comparing groups whose lists do and do not include the sensitive item. In Nigeria, Cullen found partner violence up to 35% higher with lists than with direct questions, and misreporting was associated with education and vulnerability, characteristics that development programs often target. In poor urban Lima, Agüero and Frisancho found no significant difference between methods, even across 16 subgroups. The new Peruvian paper suggests one reason for the divergence: lists are hard to explain and yield only group averages, and in rural Peru they produced higher rates than direct questions on only three of seven items, none significantly. The ballot, which offers privacy without anonymity, raised reports of any violence by 17% across 5,562 randomly assigned women, with the largest gaps among women with little schooling. Underreporting also distorted analysis. With standard data, a schooling reform appears to reduce violence; with ballot data, the effect disappears.

Put together, these findings qualify the average result rather than overturn it. The payment-cycle pattern appears for sexual violence; for physical violence the paper finds no difference between methods in how reports move over the cycle, which is reassuring for estimates such as Díaz and Saldarriaga's. The two exercises also answer different questions. One compares households with and without a program; the other follows recipients through the weeks between payments. A transfer can lower average violence and still bring short spikes after payday that face-to-face surveys miss. The cycle analysis relies on the timing of payments, not on random assignment, and on how recent events weigh in women's recall; the authors note they cannot entirely rule out that financial stress changes how women respond to privacy.

The ODA view

We read the evidence as follows. The finding that cash transfers to women reduce physical partner violence on average is among the better-supported in social protection research: it rests on several randomized trials on three continents, a quasi-experimental study of Juntos and a meta-analysis. It is, however, a claim about averages, about physical violence and about violence that women are willing to describe to an interviewer. Claims about sexual violence, about short-run dynamics and about which women benefit are weaker, because those are exactly the margins where reporting appears to fail. Two practical implications follow. Evaluations of transfer programs, including those of Juntos and its counterparts elsewhere in the region, should measure violence with private self-completion methods such as the ballot, which is cheap and familiar to anyone who has voted. And programs should treat the weeks after payment as a period of risk, linking transfers to the services that already exist for women, such as Peru's Women's Emergency Centers, and to the kind of complementary activities that in Bangladesh made reductions in violence last.

What we still don't know

The ballot evidence comes from one sample of women in rural Peru, interviewed in 2018 near emergency centers for women; it needs replication elsewhere, including where enumerators and respondents share less trust. We do not know whether violence around payday depends on how often and how payments are made, or whether digital payments into women's own accounts change the pattern. Administrative records, such as the injuries, police reports and emergency calls used in a new study of dry spells in Mexico, offer a check on surveys, but they capture only violence that reaches the state. Finally, there is a trade-off for monitoring: if household surveys switch to private methods, measured violence will rise, and statistical offices will need overlapping samples so that a better measure is not mistaken for a worse reality.