Evan Munro, Neyman Visiting Assistant Professor in the Department of Statistics at UC Berkeley

Paper: Causal Inference under Interference through Designed Markets

Abstract: In many markets a centralized mechanism determines the allocation of goods. When an individual-level intervention affects submissions to the mechanism, program evaluation is challenging due to spillover effects that occur through the mechanism. For example, a bidder-level treatment that increases some bids in an auction will impact the market-clearing price, affecting all auction participants. We show that if the mechanism is truthful and has a “cutoff” structure, then interference is constrained, and it is possible to estimate the Global Treatment Effect (GTE) under a selection-on-observables assumption. Our proposed estimator is doubly-robust and has an asymptotic variance that meets the semi-parametric efficiency bound. We also characterize heterogeneous treatment effects under interference in this setting and propose estimators for the optimal targeting rule. Taking into account equilibrium effects notably diminishes the estimated impact of an information intervention on inequality in the Chilean school system.