A Generalized Difference-in-Differences Estimator for Causal Inference with Staggered Treatment Adoption
Lee Kennedy-Shaffer, PhD
Abstract
Staggered treatment adoption arises in the evaluation of policy impact and implementation in a variety of settings. This occurs in both randomized stepped-wedge trials and non-randomized quasi-experimental panel data settings using causal inference methods based on difference-in-differences analysis. These have been used to evaluate health and economic policies; group-based interventions in education, health care, and other settings; and the impact of sudden events. In both types of setting, it is crucial to carefully consider the target estimand and possible treatment effect heterogeneities to estimate the effect without bias and in an interpretable fashion. I propose a non-parametric approach to this estimation that unites the two settings. By constructing an estimator using two-by-two difference-indifference comparisons as building blocks with arbitrary weights, the investigator can select weights to target the desired estimand in an unbiased manner under assumed treatment effect homogeneity and minimize the variance under an assumed working covariance structure. This provides desirable bias properties while using the comparisons efficiently to mitigate the loss of precision. I will first describe these two settings and their role in causal inference and policy evaluation. I then present simple settings to illustrate the underlying ideas of this new method. Finally, I will show an example using the method to estimate the effects of COVID-19 vaccine financial incentive lotteries in U.S. states; these are compared to analyses using previous methods.
