A0805
Title: Resource-efficient policy targeting under heterogeneous partial interference
Authors: Laura Forastiere - Yale University (United States) [presenting]
Elena Dal Torrione - Yale University (United States)
Chan Park - University of Illinois Urbana Champaign (United States)
Abstract: In many empirical studies, units are interconnected, and a unit's outcome may depend on the treatment of others, leading to interference. When interference is heterogeneous, treating individuals with specific characteristics can influence the population average outcome differently, either through their direct response or their impact on others. For instance, policymakers may minimize resource use by vaccinating individuals identified as superspreaders to achieve a target reduction in disease incidence. Under heterogeneous clustered interference, a method to estimate optimal stochastic treatment allocations is proposed, in which an individual's treatment probability is determined by both individual- and cluster-level covariates. The approach minimizes the expected marginal treatment probability within a cluster while ensuring a specified outcome level is met. Although the resulting optimization problem is non-convex, it is efficiently solved using difference of convex functions algorithms. To evaluate the methodology, theoretical guarantees are provided, analyzing how the excess risk bound depends on the function class complexity and cluster size. Additionally, a simulation study is conducted, and the method is applied to a water, sanitation, and hygiene (WASH) intervention in Senegal. The estimated policy is compared to alternative approaches, with the method achieving greater resource efficiency compared to policies with homogeneous treatment probabilities within clusters.
