I know of at least five other seemingly different estimation methods that produce the same ATT, as well as the same moderating effects on D(i)*[X(i) - X1_bar]. Who can list some? I'm starting simple here and will eventually get to the general staggered case .... - ThreadSky

About ThreadSky

jmwooldridge.bsky.social • 103 days ago

I know of at least five other seemingly different estimation methods that produce the same ATT, as well as the same moderating effects on D(i)*[X(i) - X1_bar]. Who can list some?

I'm starting simple here and will eventually get to the general staggered case ....

Comments

paulgp.com•103 days ago

Synthetic control or did using X and lagged outcome as match variables?

pedrosantanna.bsky.social•103 days ago

I have done this same exercise a couple of times, especially when people say they have a new estimator 👀👀👀

jmwooldridge.bsky.social•103 days ago

I see I have others to check out! In addition to the original five others I was thinking about, there's a 6th that I've taught in the past that slipped my mind.

Answers to follow late afternoon/evening. Then on to more complicated settings.

Basic message: It's hard to beat a flexible regression.

jmwooldridge.bsky.social•103 days ago

The first one is cheating a little. I can replace the cross-sectional regression with imputation: estimate a linear regression function for Y(i,2) - Y(i,1) on 1, X(i) using D(i) = 0 (control units). Impute the residuals for the treated units, and average these over the treated. This gives ATT^.

maartenpvink.bsky.social•103 days ago

📌

zshahn.bsky.social•103 days ago

g-estimation

statwonk.bsky.social•103 days ago

y ~ D * X + (D | unit_id), random effects using MCMC and then compare posterior samples where D=0 and D=1.

statwonk.bsky.social•103 days ago

y ~ 1 + D * X, + cluster SE. Accurate but imprecise. A good baseline to prevent biased estimates. ie- basic Hausman test

Posting Rules

Be respectful to others
No spam or self-promotion
Stay on topic
Follow Bluesky's terms of service

Comments

Posting Rules

Reply