Nonlinear Difference-in-Differences with Repeated Cross Sections

I show how to extend to repeated cross sections the nonlinear difference-in-differences estimators with panel data for staggered interventions in Wooldridge (2023). Implementation of pooled quasi-maximum likelihood is straightforward for both lags only and leads and lags (event study) estimators. Leading cases are logit, fractional logit, and Poisson regression. The average treatment effects on the treated can be aggregated by exposure time, with simple computational tricks making robust or clustered standard errors easy to obtain. One can easily include cohort specific trends when the key parallel trends assumption appears to be violated.