A Tutorial on Applying the Difference-in-Differences Method to Health Data

Abstract

Abstract Purpose of Review Difference-in-differences analyses are a useful tool for estimating group-level decisions, such as policy changes, training programs, or other non-randomized interventions, on outcomes which occur within the intervention group. However, there is little practical advice on how to apply difference-in-differences to epidemiologic and health data. Here, we provide a tutorial on applying the difference-in-differences method to health services data, targeted at epidemiologists and other biomedical researchers. Recent Findings As epidemiologists increasingly engage in policy discussions, familiarity with difference-in-differences will be increasingly important. However, much of the literature on difference-in-differences is limited to econometrics examples where the types of data and questions encountered may differ from health research. There remain limited resources for epidemiologists and other medical researchers to learn how to implement difference-in-differences analyses without first having to familiarize themselves with econometric terminology and concepts. Summary This tutorial contains synthetic data, code, and worksheets for class instruction. We provide a step-by-step description of the difference-in-differences analysis including sensitivity checks, modeling decisions, and interpretation. In addition, we supply novel guidance on modeling difference-in-differences outcomes for count or score outcomes.