To document the peer-reviewed evidence comparing unconditional direct cash transfers with traditional program-based development aid on welfare outcomes, administrative efficiency, and long-run beneficiary impact — and to assess the institutional and political economy reasons why the better-performing intervention remains a marginal share of global development spending.

Systematic review of randomized controlled trial evidence on cash transfer programs including GiveDirectly Kenya studies (Haushofer & Shapiro 2016, 2022; Egger et al. 2019), Mexico's PROGRESA/Oportunidades 25-year evaluation data, J-PAL meta-analyses of cash versus in-kind transfer experiments, and World Bank and USAID administrative cost data. Supplemented by political economy literature on aid agency organizational incentives.

I. THE EVIDENCE SUMMARY

Direct cash transfers to the extreme poor work. This is not a provisional finding pending replication. It is one of the most robustly replicated findings in development economics, accumulated over 25 years of randomized controlled trials across multiple continents, with follow-up periods ranging from 2 years to a decade. The evidence is strong enough that I will state what it shows before describing why institutions continue to behave as if it does not exist.

GiveDirectly, the NGO that gives unconditional lump-sum cash to households living on less than $2/day (primarily in Kenya and Uganda), has been studied more rigorously than virtually any other development intervention.

Haushofer and Shapiro (2016, Quarterly Journal of Economics) randomized 1,008 Kenyan households to receive transfers of approximately $1,000 and followed them for 9 months. Outcomes: 37% increase in asset values, 52% increase in food security, significant reductions in psychological stress and domestic violence, with no increase in alcohol or tobacco consumption.

The last finding is worth emphasizing because it was the prediction of every skeptic, and it has failed to materialize in every major study.

The 2022 follow-up by the same authors, nine years after randomization, found that consumption gains persisted at approximately 40% above control levels. Nine years later. The skeptics' preferred outcome — that recipients would spend the money unwisely and revert to baseline — did not occur. What occurred was sustained investment in productive assets, children's education, and housing.

Egger et al. (2019, Quarterly Journal of Economics) studied a larger GiveDirectly program in Kenya at sufficient scale to detect general equilibrium effects — the concern that cash transfers cause inflation in local markets by increasing demand without increasing supply. They found no net inflationary effect. Local markets responded with supply increases. This was the last major structural objection to cash transfers at scale, and it failed to materialize.

Mexico's PROGRESA program — later renamed Oportunidades, then Prospera — has been evaluated continuously since 1997 and represents the world's most studied conditional cash transfer. The conditions (school attendance, health clinic visits) add administrative complexity but also political cover.

The 25-year evaluation literature shows sustained effects on educational attainment, nutrition, and inter-generational poverty reduction across two generations of recipients. The program covered 6 million families at peak enrollment.

J-PAL's 2022 meta-analysis of 37 cash transfer programs across 20 countries found consistent positive effects on consumption, food security, psychological wellbeing, and child outcomes, with effect sizes larger than most comparable in-kind programs at equivalent cost.

II. THE COST COMPARISON

The administrative efficiency data is where the institutional story becomes uncomfortable.

GiveDirectly's operating costs run at approximately 8-15% of total expenditure depending on program type — meaning 85-92 cents of every dollar reaches a beneficiary.

USAID's program delivery overhead varies by program type but averages 30-40% across its portfolio, with some food aid programs running 50-60 cents of program delivery cost per dollar received by beneficiaries (GAO 2011, 2018 reports on food aid delivery inefficiency).

The World Food Programme's in-kind food delivery programs run overhead at 40-50% in logistics-intensive settings.

This is not a criticism of the competence of these agencies. It is a description of what program delivery costs when it involves procurement, logistics, staff, monitoring, evaluation, and reporting infrastructure. Cash transfers are administratively cheaper because money travels through mobile payment systems and does not need to be procured, stored, shipped, or inventoried.

III. WHY INSTITUTIONS PREFER THE WORSE OPTION

The political economy of development aid is not mysterious once you look at it directly.

Development agencies employ tens of thousands of skilled people whose jobs consist of designing, implementing, and evaluating programs. Those programs require sector experts, logistics staff, procurement offices, monitoring and evaluation units, field coordinators, and reporting specialists. A shift to cash transfers at scale would eliminate most of those functions.

This is not a novel insight — it was stated explicitly in the 2005 Paris Declaration on Aid Effectiveness and has been noted in every major aid effectiveness review since. It is stated, discussed, and not acted upon.

NGO funding dynamics compound this. Most development NGOs are funded by grants tied to specific program designs. A funder who wants to address malnutrition writes a grant for a nutrition program, not for cash that recipients might spend on nutrition or might spend on something they consider more urgent. The funder retains control over what gets addressed.

Cash transfers return that control to recipients. For funders who believe their program design knowledge is the value-add, this is threatening.

I want to be careful to distinguish between two claims. The first claim — that cash transfers work better than programs on average — is strongly supported by the evidence. The second claim — that all programs should be replaced with cash — is not what the evidence shows.

There are interventions where in-kind provision is clearly superior: vaccination programs, infrastructure, public health systems. The evidence supports a substantial reallocation toward cash, not a complete replacement of programmatic aid.

The current allocation — approximately 3% of bilateral aid in cash or cash-equivalent form — is not consistent with the evidence. It is consistent with institutional incentives.

A development economist I know from Monterrey — Pilar, who spent six years evaluating PROGRESA/Oportunidades for Mexico's CONEVAL and now studies aid effectiveness at ITESM — told me that the best thing PROGRESA did, scientifically, was randomize its rollout by village lottery in 1997. That accident of implementation produced 25 years of high-quality evidence.

She said most programs never produce that evidence because they are never randomized. I asked why they are never randomized. She said: 'Because that would require admitting you do not know if it works before you start.' She said this while reviewing a funding proposal from an NGO that had been running the same program design since 2004 without a control group.

She did not seem surprised. She seemed like someone who had explained this before.