Primary Submission Category: Difference in Differences
A Difference-in-Differences Framework to Estimate Causal Effects for Policy Interventions in the Presence of Heterogeneous Interference with an Application to the Philadelphia Beverage Tax
Authors: Gary Hettinger, Youjin Lee, Nandita Mitra,
Presenting Author: Gary Hettinger*
Public policy interventions are often evaluated using the difference-in-differences (DiD) approach, which does not directly account for a policy affecting nearby regions, particularly when these neighboring effects vary spatially. For example, an excise tax on sweetened beverages in Philadelphia (PHL) was associated with substantial decreases in volume sales of taxed beverages in PHL as well as increases in beverage sales of nontaxed bordering counties. The latter association may be explained by cross-border shopping behaviors of PHL residents, which may vary with border proximity, transportation access, and demographics. For example, past studies have found evidence of that the tax affects Philadelphia stores differentially depending on if they border other non-taxed Pennsylvania counties, New Jersey, or are entirely surrounded by Philadelphia zip-codes. Because such effects can offset the total effect of such interventions, particularly for specific sub-populations, understanding effect dynamics is essential to holistically evaluate public policies. Further, such insights may help predict policy effects under diverse implementation strategies. To address these concerns, we extend DiD methodology to robustly identify the causal effects of policy interventions under potentially heterogeneous interference exposure. Here, we present initial work demonstrating our framework with an evaluation of the PHL Beverage Tax policy.