Still puzzling over IPW with staggered treatment. A common/convenient thing is to assume timing is random. Can we do better estimating survival probabilities and weight by those instead of static treatment propensities? Anyone seen this done? Someone smarter than me should write this paper. #EconSky - ThreadSky

About ThreadSky

bengilbertecon.bsky.social • 456 days ago

Still puzzling over IPW with staggered treatment. A common/convenient thing is to assume timing is random. Can we do better estimating survival probabilities and weight by those instead of static treatment propensities? Anyone seen this done? Someone smarter than me should write this paper. #EconSky

Comments

bengilbertecon.bsky.social•456 days ago

Thinking in terms of DR-DID type models here.

bengilbertecon.bsky.social•456 days ago

So if me and you have the same X’s and the same propensity score but you get treated a year before me, that should tell us something about different preferences that might be correlated with our responses to (or compliance with) treatment.

bengilbertecon.bsky.social•456 days ago

In a stacking estimator you could estimate a different propensity score for each stack/cohort. I imagine you could do similarly with the sample restrictions in LP-DID.

bengilbertecon.bsky.social•456 days ago

But maybe estimating survival probabilities is easier/more widely applicable across DID estimators? Could also account for censoring if some units were more likely to receive treatment after the sample window.

apoorvalal.com•455 days ago

this paper does something along the lines you're suggesting - it estimates a survival model for treatment timing

https://arxiv.org/abs/2107.13737

bengilbertecon.bsky.social•455 days ago

This is the second time you've helped me on this question. Thank you! At first glance this paper is exactly what I had in mind, and from some superb authors.

Posting Rules

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

Comments

Posting Rules

Reply