Adaptive safety nets for rural Africa: Drought-sensitive targeting with sparse data

This paper combines remote-sensed data and individual child, mother, and household-level data from the Demographic and Health Surveys for 5 countries in Sub-Saharan Africa to design a prototype drought-contingent targeting framework for use in scarce-data contexts. To accomplish this, the paper: (i) develops simple and easyto-communicate measures of drought shocks; (ii) shows that droughts have a large impact on child stunting in these five countries—comparable, in size, to the effects of mother’s illiteracy or a fall to a lower wealth quintile; and (iii) shows that, in this context, decision trees and regressions predict stunting as accurately as complex machine learning methods that are not interpretable. Taken together, the analysis lends support to the idea that a data-driven approach may contribute to the design of a transparent and easy-to-use drought-contingent targeting framework.