Indonesia let villages spend aid freely, then added a bonus that made it faster

Indonesia's community-driven development machine — the Kecamatan Development Program from 1998, consolidated into PNPM Mandiri in 2007 — grew into a World Bank-backed operation covering more than 6,000 subdistricts and 70,000 communities, with money flowing straight to village plans instead of ministry projects. Aid circles had long argued between two orthodoxies: community discretion versus performance-based funding. Discretion might drift; performance pay might distort effort and, worse, let better-off communities chase bonuses away from the poorest.

In 2007 the government's Generasi program put both ideas in one design and randomized the question across 264 subdistricts of roughly twelve villages each. Villages elected eleven-member management teams that ran social mapping and spent grants on anything from hiring midwives to opening schools, measured against twelve indicators for child and maternal health and education. Eighty-three subdistricts got no grants, eighty-eight got unconditional grants, and ninety-three got grants with a bonus pool — 20% of funds, split by indicator performance among villages within the same subdistrict.

After 2.5 years, funded villages scored 0.04 standard deviations better across the twelve indicators: more monthly child weight checks, 7–12-year-old enrollment up, and malnutrition among under-threes down 2.2 percentage points from a 22.8 percent baseline — three times that effect in the worst-starting villages. Incentivized villages moved faster on health in the first 18 months, and the feared side effects never appeared: no manipulation of immunization or attendance records, and bonus money did not flow to wealthier areas.

Give the money to the village; bonus the metrics

Discretion found the local fixes a central menu could never list — the village knows whether it needs a midwife or a classroom.

The bonus pool was set inside subdistricts, so villages competed only against near-identical neighbors; design, not trust, prevented rich-village capture.

Keeping 80% of each grant unconditional meant incentives could not starve the work the indicators didn't measure.

A randomized design made the answer legible to a finance ministry — the program scaled on evidence, not anecdote.

Discretion and incentives aren't rivals: let communities choose the means, pay a modest bonus for measured ends, and bucket the competition so neighbors — not the whole country — are your benchmark.

On the strength of the evaluation, the Government of Indonesia moved exclusively to incentivized grants and raised annual funding, reaching about 3.7 million women and children in roughly 5,400 villages a year over four years, aimed where stunting and service gaps ran deepest. The parent PNPM program, by then one of the most studied development operations in the world with more than seventy evaluations, kept refining the model the experiment had validated.