Two become one: improving the targeting of conditional cash transfers with a predictive model of school dropout

This paper analyses whether a common targeting mechanism of conditional cash transfers (CCTs), an income-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 and in targeting assessments. Using rich administrative datasets from Chile to simulate different targeting mechanisms, I compare the targeting effectiveness of a PMT with other mechanisms based on a predictive model of school dropout. I build this model using machine learning algorithms, one of their first applications for school dropout in a developing country. I show that using the outputs of the predictive model in conjunction with the PMT increases targeting effectiveness except when the social valuation of the poor and future school dropouts differs to a large extent. Public officials that value these two target groups equally may improve CCT targeting by modifying their allocation procedures.