Most differences in differences papers these days have a treatment that turns on and/or off at different times. A specific group can be in the treated category or the control category depending on the year. What does this imply for our interpretation of a differences in differences estimate? Does this matter? Maybe not if a particular policy has one causal impact all the time, no matter what, but if the impact of a policy changes over time (and most things in life do change), then this matters. For a detailed and formal discussion of the issues, see Andrew Goodman-Bacon's NBER working paper, "Difference-in-Differences with Variation in Treatment Timing." For an easy (and fun!) to read explanation, see his twitter thread.
For all authors of econometrics/econometrics-y papers, I'd love to read more twitter threads like Andrew's describing the main ideas of your work. And if you can insert a gif in between, that just makes everything more enjoyable.
No comments:
Post a Comment