Community Targeting at Scale

Community-based targeting, in which communities allocate social assistance using local information about who is poor, in experimental settings leads to nuanced allocations that reflect local concepts of poverty. What happens when it is scaled up, by either by making the stakes high, or by replicating the process nationwide? We study this by examining community targeting in both a high-stakes experiment, in which villages determined who would receive the Indonesian conditional cash transfer program - worth almost USD 1,000 over 6 years - and in a nationwide scaleup, whereby Indonesia used community-based meetings to allocate COVID-transfers to over 8 million households. We find that both the experimental scale-up and the massive national scale-up had broadly similar performance to the original experimental study. We find strongly progressive targeting as measured by baseline household consumption, though - as in the pilot - not quite as strong as if they had used a fully up-to-date proxy means test. In both scale-ups, we also find that the villages gave additional weight to locally-valued characteristics beyond pure consumption, such as widowhood, recent illness, and food expenditure shares, again echoing the findings from pilots. The results suggest that community targeting can perform well at scale, as predicted by the experimental study