/blog/2021/08/25/twfe-diagnostics/

[andrewheiss]

Exploring Pamela Jakiela's simple TWFE diagnostics with R | Andrew Heiss
https://www.andrewheiss.com/blog/2021/08/25/twfe-diagnostics/

[andrewheiss]

Originally posted by ercbk on 2021-09-17 01:22:51

Thanks for writing this. I don't understand the homogeneity check in
3.2 though. She says that if the linearity relationship varies by
treatment then this breaks assumptions for TWFE models. I've only
looked at her paper and the Chaisemartin paper, and the only assumptions
I saw for TWFE models in general were the "common trends" and the
"strong exogeneity" assumptions. I think this is more likely to be
about the "common trends" assumption, and my understanding of that one
is that it pertains to the effect across time for a particular group.
I'm sure there's a fundamental connection between those two concepts,
but I can't see it. Or maybe it's a different assumption all together.

[andrewheiss]

Originally posted by Apo on 2022-09-12 10:22:26

Hi! Thank's a lot for sharing how you coded your way through this
paper. What is your final conclusion, after performing the last checks
(excluding later years and excluding countries one by one? Do you
finally trust your model or would you go for a most fancy/robust
approach?