Difference-in-differences with firm and time fixed effects; -reg- vs. -xtreg- command

[Max Zorn]

Hello Statalist,

I want to conduct a difference-in-difference analysis with time and firm fixed effects using panel data. I want to run both -reg- and -xtreg- commands to calculate the DID estimator. I am not sure whether in my -reg- command I account for firm-fixed effects if I write it as follows:

reg y i.treat##i.post i.time (other covariates), vce(cluster firmID)

For the sake of completeness, here is my -xtreg- command:

xtreg y i.treat##i.post i.time (other covariates), fe

Could someone kindly provide me with some feedback on whether these commands are correct, and particularly whether #1 (the -reg- command) accounts for firm-fixed effects?

Kind regards,
Max

[Dimitriy V. Masterov]

The first specification does not include firm FEs. You can accomplish that with -areg-, though the asymptotics are arguably a bit different.

[Max Zorn]

Thanks for your feedback Dimitriy!

Follow-up question: could I include firm-fixed effects in my -reg- specification as follows:

reg y i.treat##i.post i.time i.firmID (other covariates), vce(cluster firmID)

or is this generally not possible with a -reg- command? I've already tried to do it like this, however, there are too many firmIDs and Stata cannot compute it...

[Jesse Wursten]

I'd recommend using reghdfe instead

 
 reghdfe y i.treat##i.post (other covariates), absorb(time firmID) vce(cluster firmID)

[Dimitriy V. Masterov]

Both -areg- and -reghdfe- would work.