A meta-analysis of 49 studies finds “Graduation” anti-poverty packages boost income, consumption and well-being, but cost-effectiveness varies with design and delivery.
Graduation-style anti-poverty programs—bundles of support meant to help very low-income households build sustainable livelihoods—are moving quickly from pilot projects to large-scale government and nonprofit rollouts. A new study takes stock of what two decades of evidence say about how well these programs work, and under what conditions they deliver the most value for money.
The authors, Samantha Carter, Edward Jee, Dean Karlan, and Patrick Premand, analyze results from 49 studies covering 81 distinct treatment arms using a Bayesian hierarchical meta-analysis. They define “Graduation programs” as interventions that include at least three of five common elements: short-term consumption support, assistance or incentives to save, skills training, ongoing coaching, and a one-time lump-sum transfer.
Across the full evidence base, the average impacts are sizable and extend beyond incomes alone. The analysis finds that monthly consumption increases by about 28 (PPP), or 15%. Households also show gains in savings and asset holdings, improved food security, better mental health, and advances in women’s empowerment, suggesting these packages can shift both economic stability and psychosocial well-being.
But the paper emphasizes that average effects hide important patterns in how program design and delivery relate to results.
One clear finding concerns the size of the initial transfer. Programs that provide “large” transfers—defined here as at least $400 (PPP)—tend to achieve stronger impacts. However, those larger benefits do not translate into superior efficiency: big-transfer versions are not more cost-effective than programs with smaller transfers once costs are accounted for.
Costs more broadly show a similar trade-off. Higher-priced Graduation programs do produce larger improvements, on average. Yet the gains do not rise proportionally with spending. Put differently, the meta-analysis indicates diminishing returns: as program budgets expand, outcomes improve, but not enough to keep pace with the extra cost. The implication is that lower-cost models often deliver better cost-effectiveness, even if their raw impacts are somewhat smaller.
The most striking pattern is a time trend. Graduation programs launched from 2016 onward perform substantially better than those implemented earlier. The paper reports benefit-cost ratios averaging 5.39 for post-2016 implementations, compared with 2.89 for programs through 2015. Notably, the authors do not find that this improvement is explained by shifts in overall program cost or by changes in high-level program architecture (the mix of components). Something else appears to be improving over time.
To probe whether this is unique to Graduation programs, the study looks to other extensively evaluated interventions. It finds a comparable upward trend in impacts for financial education when examining behavior change outcomes, but not for unconditional cash transfers. Based on these patterns, the authors propose a plausible mechanism: ongoing learning across the sector—researchers, implementers, and funders—leading to better “how-to” knowledge and stronger implementation quality even when the headline program design looks similar.
The paper also documents the extensive collaborative effort behind assembling and checking the dataset, including contributions from research assistants, partners involved in data infrastructure, and technical feedback from workshops and reviewers. The authors note they used AI tools for limited tasks such as coding assistance and producing some figures and tables, while retaining responsibility for the final content and any errors. They also stress that the views expressed do not necessarily reflect those of the World Bank, its affiliates, or the National Bureau of Economic Research.
Why it matters
Graduation programs are increasingly being chosen as flagship tools for reducing extreme poverty, so understanding both their typical impacts and their cost-effectiveness is central for public budgets and donor strategy. The evidence suggests these packages can generate broad improvements—but that spending more does not automatically buy proportionally better results. The post-2016 performance jump also raises a crucial question for policymakers: what implementation practices, management approaches, or delivery innovations are driving better outcomes, and how can they be replicated as programs scale?