Targeting social assistance in fragile settings: An experiment on community-based targeting

Targeting is an important but challenging process in the design and delivery of social and humanitarian assistance programs. Community-based targeting (CBT) approaches are often preferred for their local information advantages, especially when data-driven methods are not feasible. However, how different variants of CBT approaches fare under various constraints and environments remains unclear. For example, it is not obvious whether agents involved in CBT maximize the number of beneficiaries or the intensity of transfers when given different levels of discretion or they face budget constraints. We implemented a clustered randomized control trial among community leaders in 180 villages in Ethiopia to evaluate how community leaders target and allocate resources when they face budget constraints and are in the presence (absence) of discretion. We find that under resource constraints, community leaders prefer to maximize the number of beneficiaries even at the expense of thinly spreading budgets (reducing average transfers to beneficiaries). Community leaders are keen to minimize exclusion errors even at the expense of increased inclusion errors, suggesting that community leaders may be sensitive to potential communal repercussions and hence prefer to accommodate beneficiaries who would otherwise be excluded based on survey-based measures and indicators of poverty. Consistent with this, we find that offering community leaders some level of discretion helps them reduce exclusion errors and include those most deprived or those affected by armed conflicts. Finally, we find that community leaders are more vulnerable to favoritism when real stakes (rather than hypothetical) are involved, budgets are relatively larger, and they lack discretion. We offer nuanced evidence about the implications of implementing CBT designs in the absence of incentives for community leaders to reveal how they use local information.