n.V1-4.04 | WHICH PART OF IT ACTUALLY WORKS

01| "Bottom-up" bundles at least four separable things: who holds decision rights over money, who employs the frontline workers, who the implementer answers to, and whether anything survives donor withdrawal. Two programmes described identically can differ on all four. A community that votes on which project gets built but cannot move a budget line, staffed by outsiders, reporting to a donor, funded for three years, scores local on one dimension and external on three.

02| Because the components vary independently in real programmes, and because several have been experimentally manipulated on their own, the bundle can be taken apart. When it is, the components do not perform equally — and the one the sector emphasises most is not the one with the strongest results.

Who chooses

03| Two experiments manipulate decision rights while holding delivery constant.

04| Olken randomly assigned 49 Indonesian villages to select development projects either through representative meetings or through direct plebiscites. Plebiscites produced dramatically higher satisfaction, greater knowledge of the project, higher perceived benefit and greater willingness to contribute. The effect on which projects were actually chosen was much smaller. His conclusion is explicit: direct participation can substantially increase satisfaction and legitimacy even where it barely changes the decisions.

05| Alatas, Banerjee, Hanna, Olken and Tobias ran an experiment across 640 Indonesian villages comparing three ways of identifying poor households — a proxy means test using assets, community targeting in which villagers rank everyone from richest to poorest, and a hybrid. Against a consumption threshold, community targeting was somewhat less accurate than the proxy means test, though not by enough to change poverty outcomes much for a typical programme. It produced higher satisfaction. And elite capture does not explain the accuracy gap: communities appear to be applying a different concept of poverty than consumption.

06| Together these establish something important and uncomfortable. Transferring decision rights reliably produces legitimacy, satisfaction, knowledge and willingness to contribute. It does not reliably produce better allocation against externally defined metrics. Where the two diverge, the divergence may be a real disagreement about what poverty is rather than a failure of local judgement — which leaves a question the evidence sharpens and cannot answer: whose concept should the programme optimise against?

Who answers to whom

07| This component has the strongest hard-outcome evidence of the four, and it is the one nobody markets.

08| Reinikka and Svensson studied a Ugandan newspaper campaign that published the size of central-government capitation grants so parents and head teachers could check what arrived. In the mid-1990s roughly 20 to 24 cents on the dollar reached schools. By 2001 the average school received more than 80 percent of its entitlement. Areas with higher newspaper penetration showed higher enrolment and higher test scores.

09| Björkman and Svensson randomised community-based monitoring of primary health providers in Uganda — village meetings plus report cards on staff performance — and found large increases in utilisation, reduced child mortality and increased child weight. A follow-up found the improvements persisted with minimal continued support.

10| Neither intervention transferred a decision right, employed anyone locally, or changed who funded what. Both changed the direction in which providers were answerable. Both moved outcomes the decision-rights experiments did not move: leakage, enrolment, test scores, mortality.

11| The condition is worth naming. Accountability mechanisms need information a community can act on and a provider who can be sanctioned. Both Ugandan cases had an identifiable provider and a measurable entitlement. Where neither exists, the mechanism has nothing to grip.

Who does the work

12| Björkman Nyqvist, Guariso, Svensson and Yanagizawa-Drott randomised a community health promoter programme across 214 Ugandan villages, with promoters recruited locally and paid through margins on the health products they sold. After three years: under-five mortality down roughly 27 percent, infant mortality 33 percent, neonatal 28 percent. No clinics, drugs or staff were added to the wider system.

13| Bold, Kimenyi, Mwabu, Ng'ang'a and Sandefur varied the employer while holding the contract constant in Kenyan schools. Teachers on identical fixed-term contracts raised test scores when hired by an international NGO and produced zero impact when hired by the government, with bureaucratic and political opposition producing delays and divergent readings of the same terms.

14| Both are cited as evidence for local delivery and both support something narrower. The first varies locality and incentive structure together. The second varies the employing institution, not the locality of staff. Neither isolates "is the worker from here" as a variable. No trial we located delivers the same protocol through local versus non-local workers holding employer, incentives and supervision constant. Given how much of the bottom-up argument rests on that claim, the absence is striking.

What survives withdrawal

15| The weakest-evidenced component, and the one asserted most often.

16| What exists is indirect. Blattman, Fiala and Martinez found Ugandan grant recipients' gains had dissipated by year nine as the control group caught up. Casey, Glennerster and Miguel found short-run public goods gains from a community-driven programme in Sierra Leone and no sustained impact on collective action, decision-making or the inclusion of marginalised groups. Both describe what happens after external input stops. Neither manipulates funding independence as a design variable, because no funder has run that experiment.

17| Organisations that generate their own revenue — BRAC's social enterprises are the standard example — are cases, not evidence. One organisation succeeding while self-funding does not establish that self-funding caused it, and BRAC differs from its comparators on many dimensions at once.

The bundle, whole

18| Beath, Christia and Enikolopov evaluated Afghanistan's National Solidarity Programme across 500 villages — elected village councils, block grants, mandated female participation. It improved access to drinking water and electricity, acceptance of democratic processes, perceptions of economic wellbeing, and both the participation of women and attitudes toward them.

19| That is the bundle delivering. It does not tell you which part delivered, which is exactly the problem: the largest and best-evaluated participatory programmes are evaluated as packages, so the field has accumulated evidence that the bundle sometimes works without learning what in it works.

What the decomposition shows

20| Ranked by strength of evidence for effects on hard outcomes: accountability direction first, and by a distance. Frontline delivery second, with confounded attribution. Decision rights third, with clean evidence for a narrower claim. Funding independence untested.

21| This inverts the sector's emphasis. Decision rights are treated as the core of local control — Arnstein's ladder is built around them, and every localisation target is a proxy for them. The component with the best evidence for moving outcomes people care about is accountability direction, which is separable from decision rights and considerably cheaper. Publishing entitlements in a newspaper is not a transfer of power in anyone's typology. It moved sixty cents on the dollar.

22| What this does not support is the claim that decision rights do not matter. Legitimacy and satisfaction are real outcomes; they plausibly affect participation and maintenance over horizons these trials do not cover; Olken's own framing is that participation matters even where it does not change decisions. The narrower finding is that decision rights and outcome quality are separable, and a programme optimising for one should not assume it gets the other for free.

23| The practical consequence: if the components separate this cleanly in the evidence, they can be specified separately in design. Any claim to be bottom-up should say which of the four it holds and on what basis. "All of them" is a claim requiring evidence, not a description.