Abstract:
|
Difference-in-difference (DiD) designs are an important tool for observational causal inference and are a common strategy for evaluating the effects of changes in public policy. This paper focuses on the complications spatially organized data presents for doubly-robust DiD designs which combine techniques such as matching, IPW, and synthetic control with the conventional DiD approach. While issues such as the lack of comparability across groups and the importance of spatial spillover effects are both acknowledged within the DiD literature, there is limited work focusing on how space can complicate and bias doubly-robust DiD designs and vice versa. In this paper, we first address the ways space complicates matching, weighting, subclassification, and synthetic control methods when building comparable treatment and control groups. We then present the ways matching, weighting, subclassification, and synthetic control methods can complicate the study of spatial effects in DiD. We conclude by presenting Monte Carlo simulations of these problems and the bias they introduce, as well as recommendations for conducting applied DiD analysis of spatially organized data.
|