Send message to Author

Paper short abstract

Food aid targeting involves two decisions: whom to reach and what to provide. Using a randomized trial in Egypt and machine learning, we show the deprivation-versus-impact trade-off is limited; the sharper choice is which nutritional objective to prioritize. Under a fixed budget, the costlier nutrit

Paper long abstract

Food aid confronts policymakers with a choice that poverty targeting does not: whom to reach and what to provide are separate decisions, and both govern the nutritional return on a fixed budget. It remains a principal channel through which governments provide services under binding budget constraints, and what that spending achieves depends on the composition of the transfer as much as on who receives it, since in-kind transfers can distort consumption. This paper examines targeting trade-offs across nutritional objectives in a nationwide food aid program in Egypt. Drawing on a randomized controlled trial, we use machine learning to characterize heterogeneity in treatment effects, and find substantial variation. Observable characteristics, of the kind used in proxy means tests, predict this variation differently across nutritional outcomes and transfer types. Building on recent work that weighs expected impact alongside deprivation, we show that the trade-off between reaching the most deprived and the most impacted households is limited. The sharper choice is which nutritional objective to prioritize. Accounting for program costs, we find that under a fixed budget, greater nutritional gains follow from allocating a larger share to the costlier nutrition-sensitive food box, with the clearest returns for food insecurity. Targeting on observable characteristics can thus extend the reach of a constrained budget, provided the criteria are matched to the objective and the transfer. By showing how impact heterogeneity can guide these decisions, the paper offers a route to rethinking food aid design and delivery when resources are fixed.