{Adjusting treatment effect estimates by post-stratification in randomized experiments}