Introduction

Graduation programmes aim to generate long-run socioeconomic impacts by helping low-income households engage in new or expanded income-generating activities. Under the premise that market frictions are not single-faceted, these programmes deliver a package of components, specifically a combination of consumption support, savings facilitation, skills training, coaching, and an asset transfer or lump-sum cash grant. This meta-analysis evaluates multifaceted programmes that include at least three of these five core components, covering both "core" Graduation programmes and broader economic inclusion initiatives that may be lighter in scope. In total, the analysis draws on 49 study records from 41 publications, reports, and working papers, covering 40 unique randomizations with 81 reported treatment variations and 109 canonical treatment-arm-wave observations.

Study sample and data

The final analysis sample includes programmes evaluated in 26 countries. While coverage is concentrated in Sub-Saharan Africa, it also spans South Asia, Southeast Asia, Latin America, and the Middle East. Approximately one-quarter of the studies occurred in fragile or post-conflict settings, including Afghanistan, Chad, and the Democratic Republic of Congo. The study tracks 12 outcomes grouped into consumption, income, food security, savings, assets, employment, psychological well-being, and female empowerment. Regarding costs, 48 unique treatment arms provided intervention cost estimates, which range from 102 USD to 3,090 USD per participating household. The mean cost is 819 USD, while the median is 445 USD.

Methodology

The research employs a Bayesian hierarchical framework to synthesise evidence. This approach separates true contextual heterogeneity from sampling variation, describes how impacts vary with programme composition, and derives persistence and cost-benefit estimates from uncertain impacts. For income and consumption, treatment effects are annualised. The study uses Hamiltonian Monte Carlo in Stan for estimation, with posterior distributions summarised using posterior means or medians and 95 per cent posterior intervals. Cost-benefit analysis translates posterior impact draws into ratios, discounting at an annual rate of 5 per cent. Three persistence scenarios are reported: "linear-decay" (benefits reaching zero after eight years), "estimated-persistence" (based on the sample’s follow-up slope over a 10-year horizon), and "permanent-persistence" (final effects persisting forever).

Results

The analysis finds large gains across economic and psychosocial outcomes. The posterior average treatment effect shows an increase in monthly household consumption of +38.1 USD (2020 PPP), representing an 11 per cent increase relative to the median control mean of 361 USD. Monthly household income rises by +28.0 USD (2020 PPP), a 15 per cent increase relative to the median control mean of 190 USD. Results on food security show statistically significant impacts of 0.20 standard deviation. The estimated effect on total assets is +282 USD (2020 PPP), or 40 per cent of the median control mean. Business ownership increased by 8 percentage points, a 27 per cent relative increase. Psychological well-being also improved, with mental health scores increasing by 0.10 standard deviations and life satisfaction by 0.20 standard deviations. Women’s agency improved by 0.09 standard deviations.

Persistence and cost-benefit analysis

Evidence suggests that consumption effects strengthen at later follow-ups, with an estimated annual slope of 47.8 USD PPP. The overall benefit-cost ratio is 4.34 under the preferred estimated-persistence scenario. Programmes implemented from 2016 onwards perform considerably better than earlier ones; benefit-cost ratios for recent programmes are 5.39, compared to 2.89 for those implemented through 2015. While programmes with large transfers (at least 400 USD PPP) have larger treatment effects, they are not more cost-effective. More expensive programmes generate larger treatment effects, but the improvements are not commensurate with the cost increases; thus, lower-cost programmes tend to be more cost-effective.

Conclusion

Graduation programmes generate large, durable improvements in the economic well-being of low-income households. The most striking finding is that programmes implemented after 2015 generate substantially larger treatment effects than earlier cohorts, representing an 87 per cent increase in benefit-cost ratios. This improvement cannot be explained by publication timing, shifts in component bundles, or mere cost reductions. Instead, it likely reflects a mechanism of continuous sector-community learning and improved implementation quality. The study suggests that development economics would benefit from greater attention to implementation science, identifying the granular choices that determine how a design is delivered.