lukestein’s Twitter Archive—№ 3,769

So if causal inference is so easy that we need no tools other than ”A/B tests,” explain to me why panel variation with *highly* non-random selection is informative about counterfactuals @roybahat/1038986277981310977