ABSTRACT
Chapter 7 explains how to use the difference-in-differences (DiD) method to adjust for confounding when one of either additive equi-confounding, additive equi-confounding on the log scale, or additive equi-confounding on the logit scale holds. It pairs these three assumptions with a linear, loglinear, or logistic DiD model. It shows that the resulting estimator targets the average effect of treatment on the treated. The chapter motivates the DiD methods by a pre- versus post-pandemic comparison of employment rates and a pre- versus post-negative interest rate adoption comparison of bank net interest margins. DiD methods are compared to standardization and it is shown that for a simple binary dataset, that the linear DiD estimator will differ from the standardized estimator of the average effect of treatment on the treated except in trivial cases. R code is also provided.
