CSDID with geographical fixed effect

[Yifei Liu]

Dear Statalist,

I have a question regarding the incorporation of fixed effects in conjunction with the CSDID command.

My objective is to assess the impact of participation in an auction system on a firm's investment choices. The actual auctions are conducted by distinct organization across various geographic regions (based on where the firms located, they attend regional auction). I am looking to validate my findings by incorporating an auction zone - year fixed effects.

I think I should run the following command:

 csdid2 investment i.year#i.zone, cluster(firm_id) time(year) gvar(first_treated_firm) method(dripw)

However, I have read another post regarding municipality fixed effects in CSDID. And the comment pointed out that adding such a fixed effect won't work (https://www.statalist.org/forums/for...fects-in-csdid).

I hope to seek further clarification on what is the best approach for my situation. In my case, only the treated group will ever fall into an auction zone. I am not sure what would be the right way to demean my dependent variable based on the auction zone.

Thank you very much.
Yifei

[George Ford]

I'd use the residuals like Fernando suggests. He coded the thing.

[FernandoRios]

And just to understand
year fixed effects makes no sense because in csdid all covariates are already interacted with year effects
In addition if you use variables that perfectly predicts treatment (or treatment outcome) you violate the overlapping assumption . This creates other problems for the estimation.
using residuals is just an auxiliary approach that mimics what did imputation does. And would help (as you say) work as robustness
hth