Open azev77 opened 3 years ago
It may be worth comparing SCM w/ the new generation of Diff-in-Diffs estimators.
@vincentarelbundock compares TWFE/Callaway & Sant'Anna/IFE/Matrix Completion in a blog post.
The author has a disclaimer that he's not comparing the performance of alternative estimation strategies ...
That would be interesting. The disclaimer in myblog post is because I don't actually think that the factor structure in the DGP satisfies the Callaway & Sant'Anna assumptions. Baker et al (2021) has a different simulation where C&SA works better:
https://andrewcbaker.netlify.app/
Here's a stripped down version of that simulation that I coded up data.table
, but the full code is on Baker's website:
Reminder for self: here is a link to @andrewchbaker's DiD codes for "How Much Should We Trust Staggered Difference-in-Differences Estimates?"
Also @borusyak's https://github.com/borusyak/did_imputation compares 5 event study estimators in a simulated panel:
Also:
@kylebutts has did2s (R/STATA) w/:
I coded the classic Diff in Diffs (two way Fixed Effect) outcome model, which is a special case of Synthetic Controls.
I just discovered @junyuan-chen's new package DiffinDiffs.jl which has a few of the latest DiD techniques. Maybe, if there is interest, at some point there can be an opportunity to join forces? Maybe create an umbrella organization? (ProgramEvaluation.jl?)