Causal Inference
Confounding in DiD: formal definition and variable selection strategies Daniela Rodrigues* Daniela Rodrigues Laura Hatfield
Confounding represents the key challenge to causal inference for non-experimental studies. Over the years, significant efforts have been made to define confounding and develop confounding variable selection strategies in cross-sectional studies. However, in the context of difference-in-differences (DiD), the definition of confounding has not been formalized. In addition, there is a lack of guidance on which confounding variables make the assumption of conditional parallel trends most plausible. In this work, we use causal diagrams to formalize the definition of confounding in DiD and propose strategies to aid in the selection of confounding variables in this context. We apply these developments to the study of the effect of Comprehensive Primary Care Plus on healthcare equity.