marketing analytics

The two-way fixed effects regression is the default for staggered geo rollouts. On simulated data where the true lift is known, it understates the effect by 41%. Here is why, and the estimator that fixes it, in R and Python.