Pinned post
How much data do you need to conduct an informative staggered diff-in-diff? In our new working paper @fhollenbach and I simulate the power of #DiD estimators, and find that you might need *a lot*, even to detect large effects. We also provide suggestions for improving power 1/