#EconTwitter Recently there’s been an explosion of empirical applications of Difference in Difference estimation in economics. Unfortunately, DiD is often *less* credible than structural models. Why? Choice of covariates and functional form in DiD is more ad hoc, because... 1/3
1. We can’t know what conditioning variables (if any exist) are needed to satisfy parallel trends, 2. Some conditioning variables are likely to be colliders or otherwise endogenous, and 3. Additive linearity in covariates is almost certainly a misspecified functional form... 2/3
Comparable problems exist for structural models, but there economic theory can provide some guidance regarding covariates, endogeneity, and functional form. For DiD it’s generally little more than just hand-waving and robustness checks. 3/3.
