Disparate Impact in Difference-In-Differences

• In Event: Causal Inference and Algorithmic Discovery for Criminology and Public Policy

Abstract

Difference-in-differences is a common method used to evaluate the impact of a law or policy. Often, researchers using the specification will also be interested in evaluating whether the policy had disparate impact on different groups, e.g. by race. Typical approaches include comparing subgroup-specific treatment effects, either in absolute terms or relative to the pretreatment mean. We show that these methods may not correspond to the research question of interest and propose an alternative method of testing for disparate impact that uses complier analysis tools from the instrumental variables literature. Using data from Philadelphia to evaluate the disparate impact of Philadelphia's "No-Cash-Bail" policy on bail decisions across different racial groups, we demonstrate how these different methods can yield strikingly different results. The complier analysis method, which we argue is the best metric for measuring disparate impact, shows that White defendants disproportionately benefited from the reform.

Authors

• Aurélie Ouss, University of Pennsylvania

• Megan T. Stevenson, University of Virginia