Two Become One: Improving the Targeting of Conditional Cash Transfers with a Predictive Model of School Dropout

This paper offers a methodology to improve targeting design and assessment when two or more groups need to be considered, and trade-offs exist between using different targeting mechanisms. The paper builds from the multidimensional targeting challenge facing conditional cash transfers (CCTs). I analyze whether a common CCT targeting mechanism, namely, a proxy means test (PMT), can identify the poor and future school dropouts effectively. Despite both being key target groups for CCTs, students at risk of dropping out are rarely considered for CCT allocation or in targeting assessments. Using rich administrative data sets from Chile to simulate different targeting mechanisms, I compare the targeting effectiveness of a PMT and other mechanisms based on a predictive model of school dropout. I build this model using machine learning algorithms. Using two novel metrics, I show that combining the outputs of the predictive model with the PMT increases targeting effectiveness except when the social valuation of the poor and future school dropouts differs to a large extent. More generally, public officials who value their key target groups equally may improve policy targeting by modifying their allocation procedures.