Primary Submission Category: Causal Inference in Networks
Regression Discontinuity Designs Under Interference
Authors: Elena Dal Torrione, Tiziano Arduini, Laura Forastiere,
Presenting Author: Elena Dal Torrione*
Interference takes place whenever a “treatment” on one unit affects the outcome of another unit, and such a phenomenon can occur in regression discontinuity designs (RDD). For instance, in conditional cash transfer programs for education, eligible children’s schooling choices may affect the schooling choices of their ineligible peers. We propose an extension of the continuity-based framework to RDD to identify and estimate a set of causal estimands in the presence of interference. In this setting, assignment to effective treatment is determined by a unit’s score and the scores of other units—for example, her neighbors. Unlike the standard RDD, embedding the exposure mapping function as a summary of other units’ treatment may give rise to complex, multidimensional frontiers. We provide a method to characterize such frontiers for a broad class of exposure mapping functions and derive generalized continuity assumptions to identify the proposed estimands. Next, we develop three estimation methods that can handle high-dimensional—and potentially heterogeneous—score spaces, and evaluate their empirical performance in a simulation study. Finally, we apply the presented methodology to the PROGRESA/Oportunidades data to estimate the spillover effects of financial aid to families on children’s school attendance.