Governments and aid organizations spend billions each year on programs designed to lift families out of poverty: cash handouts, food vouchers, livestock grants, subsidized childcare, guaranteed public works jobs. Increasingly, policymakers want these programs to do something more specific — help women earn income, build savings, and gain a stronger voice in their households. But which forms of assistance actually deliver on that promise, and which fall short?
A new analysis published in Nature Human Behaviour pulls together evidence from nearly two decades of randomized experiments to answer that question. The pooled results suggest that most types of social safety nets do improve women’s economic standing and sense of agency, but the size of those effects varies substantially depending on the design of the program.
The question behind the review
Amber Peterman of the University of North Carolina at Chapel Hill and colleagues from institutions including the Technical University of Munich and the University of Zurich set out to synthesize what experimental evidence can tell us about a fairly specific question: When governments or NGOs deliver economic assistance in low- and middle-income countries, do women benefit in measurable ways?
Earlier reviews had leaned heavily on narrative summaries and focused mostly on cash transfers, leaving open questions about other forms of aid such as asset grants, subsidized care, or guaranteed employment schemes. Some observers have also raised concerns that certain programs, particularly cash transfers that require women to attend trainings or ensure their children’s school attendance, could backfire by piling additional unpaid work onto mothers.
The researchers wanted to test these ideas quantitatively across a broad body of evidence, and to look at seven distinct types of safety net programs side by side.
How the researchers built their sample
The team searched six academic databases in English, French, and Spanish, along with the publication libraries of organizations like the World Bank, J-PAL, and UNICEF. They screened nearly 4,000 abstracts and eventually included 115 papers reporting on 93 randomized controlled trials. Altogether, these studies covered 218,828 women across 45 countries, generating 1,307 individual effect estimates. About half of the studies had been published since 2019.
The researchers grouped the interventions into seven categories: unconditional cash transfers, conditional cash transfers (which require recipients to meet certain requirements, such as sending children to school), food or in-kind transfers, productive asset transfers (like livestock or business equipment), public works programs offering guaranteed short-term jobs, fee waivers or subsidies, and social care services such as subsidized childcare.
On the outcome side, they distinguished between “economic achievements” (things like labor force participation, earnings, savings, assets, and expenditures) and “agency” (measures of decision-making power, autonomy, self-efficacy, voice, and leadership). To make results comparable across such varied studies, they converted every effect into a standardized statistic called Hedges’ g, which expresses the size of an impact in standard deviation units.
What the aggregated evidence showed
Across all interventions and outcomes, the pooled effect was positive and statistically significant, with a Hedges’ g of 0.107. That figure lands in a similar range to what previous meta-analyses have found for vocational training programs aimed at young women, and it’s larger than the average effect of microcredit or savings interventions.
But the interesting patterns emerge when you break the results apart. Unconditional cash transfers, social care services, and asset transfers each produced effects of roughly 0.12. Public works programs showed a similar magnitude, though with weaker statistical significance because fewer studies exist on them. Conditional cash transfers, by contrast, produced a notably smaller effect of 0.059. And food or in-kind transfers showed no statistically significant impact overall.
The researchers offer a hypothesis for why conditional cash transfers underperform their unconditional counterparts. When women are required to attend mandatory trainings or shepherd children to health check-ups to keep receiving benefits, those requirements can eat into the time available for paid work and reinforce their role as primary caretakers. The authors argue that this pattern supports growing calls to drop conditionalities, though they note they can’t fully separate the effect of conditions from other design features.
Where the gains showed up
Looking across economic outcomes, the strongest impacts appeared in savings (Hedges’ g of 0.229), assets (0.235), and expenditures (0.177). Labor force participation and hours worked also increased, addressing a common concern that giving people money makes them work less. The evidence pointed in the opposite direction.
Care work was a rare exception. The researchers found no significant change in how much unpaid caregiving women did, though only 16 studies measured this in detail. Debt and loan outcomes also showed no clear effect.
Within the agency domain, the largest gains appeared for voice (Hedges’ g of 0.172), followed by autonomy and self-efficacy (0.105) and decision-making (0.087). Not enough studies measured aspirations, goals, or leadership to draw firm conclusions on those dimensions.
What didn’t predict success
One of the more surprising findings involves what the researchers didn’t find. The current thinking in the field is that programs targeting women specifically, offering larger benefits, or bundling cash with complementary services like training should produce bigger gains. When the team ran regression analyses looking for these patterns, most of them didn’t show up as significant predictors.
The authors offer two possible explanations. One is that the studies vary so much in how they implement these design features that any signal gets lost in the noise. Another is that many of the studies in their sample already incorporate some form of gender-focused design, leaving too little variation to detect what matters. They caution that this shouldn’t be read as evidence that thoughtful design doesn’t matter, only that the current evidence base can’t yet pin down which specific features drive results.
The money question
The researchers also looked at whether these programs pay for themselves. Only 25 of the papers reported any form of cost-benefit analysis. Among those that did, benefit-cost ratios were generally positive, ranging as high as 16.9 for an unconditional cash program in Tunisia. Internal rates of return in ten studies ranged from 6 percent to 73 percent.
But there’s a catch. Almost none of these calculations attempted to place a value on the gender-specific benefits — the increases in women’s earnings, savings, or decision-making power that this review documents. Most cost-benefit exercises count only household-level consumption and asset accumulation. The researchers argue this means existing cost-benefit ratios likely represent lower bounds, and that the field needs new methods for incorporating gender outcomes into economic evaluations.
Caveats worth noting
Several limitations shape how these findings should be read. The average intervention lasted 12 months and was followed up 14 months after ending — a fairly short window. Impacts on women could evolve differently over longer periods, either fading or compounding. The sample is also concentrated geographically, with 57 percent of effects coming from sub-Saharan Africa and 21 percent from South Asia.
The researchers also note that they can’t observe which studies chose to measure women’s outcomes in the first place. It’s possible that programs designed with women’s welfare in mind are also more likely to publish women-specific results, which could inflate the average effect sizes seen here.
For policymakers deciding where to put limited resources, the authors suggest their results point toward unconditional cash, asset-based programs, and care services as the modalities with the strongest evidence base for benefiting women. They also encourage practical steps like removing access barriers, extending coverage to excluded groups, and building linkages to complementary services — even where the meta-analysis couldn’t confirm the specific contribution of each design feature.
