-margins- after -eprobit-

[Parul Gupta]

I am using -eprobit- for estimation. I am demonstrating here using the State example.

webuse heartsm.dta, clear

eprobit attack age bmi, entreat(exercise = bmi i.gym) vce(robust)

To get the ATET, I am calculating margins.

margins r(0 1).exercise if exercise, contrast(effects nowald) post

My sample size is large and I have a lot of fixed effects in the model, and the margins computation is taking very long. I have to submit soon and I cannot wait for it to run for days.

I am using the -nose- option after margins to suppress standard errors, which speeds up the computation.

However, I would like to report the statistical significance of the estimates, which I cannot get with the nose option. Are there standard errors/CI/p-values from the main model that I can report along with the margins to indicate the statistical significance?

[Felix Bittmann]

Deleting not required observations can speed up the computation:

eprobit attack age bmi, entreat(exercise = bmi i.gym) vce(robust)
keep if exercise == 1
margins r(0 1).exercise if exercise, contrast(effects nowald) post

[Parul Gupta]

Thanks, it didn't help unfortunately.

[Richard Williams]

I would consider just reporting the p-values etc from the main model and the marginal effects without them. If a reviewer insists on them you can let your computer spend a week or so cranking them out.

[Parul Gupta]

Which coefficient should I be looking at in the main model?