In difference-in-differences settings, should you allow for unit-level fixed effects or just treatment-cohort (group) fixed effects? With imputation, Jeff Wooldridge's paper shows it doesn't matter and group fixed effects can be *way* faster.
Notes on how I think about non-traditional difference-in-differences settings. I first describe the intuition behind the basic two-stage diff-in-diff method and then use that insight to describe how to approach non-traditional settings (e.g. continuous treatment, treatment turning on and off, and multiple doses during period)
Introductory notes with detailed derivations of influence functions. I took these notes while trying to learn the material myself.