Pedro’s diff in diff checklist: step 1) plotting the rollout
When the “diff in diff credibility revolution" started to grow, and we had a half dozen or more different methods for estimating the same parameter (the ATT) under very similar if not the same assumptions (SUTVA, no anticipation, conditional parallel trends, and overlap), I remember reading online people hoping for a checklist. The checklist they seemed to want would allow them to pick between:
And that just gets us through six of them, but you can add even more. Without clear guidance as to which one to use, many researchers just implement all of them, presented in a single graph, which I have called the “all the diff-in-diffs” event study graph.
And I am sympathetic to that personally though I know some consider it an unprincipled approach. But my philosophy has been that if there’s two ways to do something, and you ex ante really are uns…
Keep reading with a 7-day free trial
Subscribe to Scott's Substack to keep reading this post and get 7 days of free access to the full post archives.