Search

Centralised targeting registries are increasingly used to allocate social assistance benefits in developing countries. This paper provides the first attempt to identify the relative importance of two key design issues for targeting accuracy: (1) which households to survey for inclusion in the targeting registry and (2) how to rank surveyed households. We evaluate the performance of Indonesia’s Unified Database for Social Protection Programmes (UDB), the largest targeting registry in the world, which is used to provide social assistance to more than 25 million households. Linking administrative data with an independent household survey, we find that the UDB system is more progressive than previous targeting approaches used in Indonesia, leading to a decrease in benefit leakage to non-poor households. However, if poor households are not surveyed in the first place, even a perfect ranking method cannot prevent their exclusion. Under a simulation that considers enumerating and estimating proxy-means testing (PMT) scores for all households (as in a census), we estimate a one-third decrease in undercoverage compared to focusing on households that have been registered in the UDB. Investigating household- and community-level correlates of misenumeration and misclassification, we find evidence that local communities use different definitions of poverty and have better information on the welfare status of their members.

In developing countries, identifying the poor for redistribution or social insurance is challenging because the government lacks information about people's incomes. This paper reports the results of a field experiment conducted in 640 Indonesian villages that investigated two main approaches to solving this problem: proxy-means tests, where a census of hard-to-hide assets is used to predict consumption, and community-based targeting, where villagers rank everyone on a scale from richest to poorest. When poverty is defined using per-capita expenditure and the common PPP$2 per day threshold, we find that community-based targeting performs worse in identifying the poor than proxy-means tests, particularly near the threshold. This worse performance does not appear to be due to elite capture. Instead, communities appear to be using a different concept of poverty: the results of community-based methods are more correlated with how individual community members rank each other and with villagers' self-assessments of their own status than per-capita expenditure. Consistent with this, the community-based methods result in higher satisfaction with beneficiary lists and the targeting process.