Doubly robust estimation of policy-relevant causal effects under interference.

This article focuses on developing and applying a methodological framework to estimate the causal effects of public policies in the presence of interference between geographically adjacent regions, specifically addressing bypass effects where individuals circumvent policy restrictions by crossing borders. The authors adapt doubly robust (DR) difference-in-differences (DiD) estimators combined with exposure mappings to separately estimate the Average Treatment Effect on the Treated (ATT) in Philadelphia, Pennsylvania, and the Average Treatment Effect on the Neighbouring Control (ATN) in bordering counties, using retail sales data of sweetened beverages. Their analysis reveals that the Philadelphia beverage tax (PBT) significantly reduced taxed beverage sales within Philadelphia stores while increasing sales in neighboring counties, with notable seasonal and geographical heterogeneity in effects. Simulation studies demonstrate that the proposed DR estimators provide robust and less biased causal effect estimates compared to traditional DiD methods, especially under confounding and treatment effect heterogeneity, highlighting the importance of accounting for spatial interference in policy evaluations.