Lincom after xtpois

[davidbrown]

Dear Statalist members,

I have read that it doesn't make much sense using margins after non-linear fixed effects models (one of the links:https://hbs-rcs.github.io/2017/02/14...fixed-effects/).
As I understand it, the reason why you can't predict the dependent variable is that the models do not calculate the fixed effects.

Then I wondered about lincom. I think it is OK to use lincom unlike margins because althoug it is a similar post estimation with margins, but it does not predict the dependent variable but just linear combinations of coefficients. I wonder what other members think about using lincom after fixed effects non-linear models.

Thanks,
David

[Clyde Schechter]

Yes, you can use -lincom- after -xtpois, fe- and similar fixed effects non-linear models. The reasoning against using -margins- doesn't apply to -lincom- because the things that -lincom- can be used to estimate don't depend on the fixed effects.

That said, you can use -margins, dydx()- after -xtpois, fe- and similar fixed effects non-linear model. You will probably get results saying "not estimable," but if you apply the -noestimcheck- option, you will get results. And those results will be usable because -margins, dydx()-, like -lincom- doesn't need to know the fixed effects. (By contrast, if you use -noestimcheck- after -margins- (without dydx()) you will get results, but those results will be wrong.)

[davidbrown]

That clears up a lot of concerns I had.
Thanks, Clyde!