137 / 2024-04-21 22:31:32

Eligibility threshold selection in social assistance programs: Targeting the poor from self-targeted population

social assistance,self-targeting,self-selection

Social assistance programs (SAPs) aim to utilize limited budget to serve as many households in need as possible. Targeting is the process of identifying and reaching out to households truly in need. However, targeting is challenging due to 1) limited prediction power on households’ welfare levels, and 2) the applicants of SAPs are biased samples of the whole population due to self-targeting effect. These factors can result in high exclusion errors and leaving budget underutilized in SAPs. To improve targeting performance, we focus on the eligibility threshold decision, which determines applicants’ eligibility based on their predicted welfare levels. We first characterize the self-targeted applicant population in distribution shift and population scale, constructing corresponding estimators with performance guarantees; Then we provide a new policy embedding the estimators into eligibility threshold selections. The numerical results show that the proposed policy significantly outperforms the benchmark policy from practice and covers more people truly in need, especially when welfare prediction power is limited. Beyond SAPs, self-selection and threshold-based control decisions are widespread in many other areas, where we believe the proposed method can also provide potential benefits.