didregress vs xtreg fe for did analysis

[Pranshu Tripathi]

Panel data:1600 firms, 8 years, and 6 variables.
I have a time-based treatment dummy variable (pre and post-treatment), a group dummy variable (healthy and unhealthy), and a non-control continuous independent variable along with other independent variables acting as control variables. The non-control continuous independent variable has to interact with the other two dummy variables.
I want to ask whether I should go with the

didregress model
or

xtreg,fe model where I can create dummy interactions like

xtreg dep noncontrolindep controlindep i.treatmentdummy##i.groupdummy##c.non-controlcontiniousindependenttvariable, fe

Thanks

[Carlo Lazzaro]

Pranshu:
I would go -didregress- or -xtdidregress- as per your dataset.

[Pranshu Tripathi]

Okay, Carlos.

I have a few other questions, and please take into consideration that I have panel data.

If I do any one of these, then the model would be

didregress (outcome) (treatment_indicator), group(firm) time(timevar)

or

xtdidregress (outcome) (treatment_indicator), group(firm) time(timevar).




1) Are they both the same?

2) why"xtreg dep noncontrolindep controlindep i.treatmentdummy##i.groupdummy##c.non-controlcontiniousindependenttvariable, fe" is not right?

3) How to generate treatment_indicator. (I have created a dummy of the years based on a law passed in 2016. So, pre-2016 is pre-treatment (1), and post-2016 (2) is post-treatment. But these dummies have been generated in an excel sheet).

4. group(firm) means that group dummy where the firm is either healthy (1) or unhealthy(2). So do I have to go with group(groupdummy)?

5. Can we do a fixed effect regression with any one of these? (didregress/xtregress (outcome) (treatment_indicator), group(firm) time(timevar),fe?
5a. Can we cluster for errors into the model in the 5th point?

[Carlo Lazzaro]

Pranshu:
1) no, they are not. The two commands were developed for repeated cross-sectional studies and panel datasets DID, respectively. Therefore, as per your description, I would go -xtdidregress-;
2) I do not comment on the correctness of your -xtreg,fe- code, but I'm curious about an effective way of communicating a three-way interactions. Again, I'd stick with -xtdidregress-;
3) can't you simply import or, at worst, copy and paste it in a .dta file?;
4) you can add healthy 0/1 as a categorical predictor between brackets just after the -outcome-;
5) no;
5a) see -Standard error considerations- paragraph in -DID intro-, Stata .pdf manual-

[Pranshu Tripathi]

Thanks, Carlos. You are a savior.
Regards.