search

grf-labs / grf

Generalized Random Forests
https://grf-labs.github.io/grf/
GNU General Public License v3.0
957 stars 250 forks source link

Question: staggered adoption design with event study #1057

Open kimpeijn opened 2 years ago

kimpeijn commented 2 years ago

Hi, thank you very much for helping out.

I am working on a paper where everyone is treated at some point in time, and the year of treatment differs between people. I am estimating then the treatment effects separately at 0, +1, up until +10 years after treatment (later I obtain the ATE using the appropriate weights). The panel is unbalanced. Some people enter the data at -10 and leave at +10, while others enter at -6 and leave at +1.

Is it possible to estimate heterogeneous treatment effects in this setting? How should I approach this?

Thank you. Kim

erikcs commented 2 years ago

Hi @kimpeijn, I think @muhlbach applies causal forest to a similar panel data setup in his thesis, if you look at section 2.3.3

kimpeijn commented 2 years ago

Thank you @erikcs . It is similar, except that he does not estimate treatment effects separately for each period following treatment. The approach makes sense in his thesis, but I am not sure how to extend it to my setup.

finnkrueger commented 5 months ago

@kimpeijn did you find an implementable solution for your problem ?