predictnl after reghdfe

[Nicolas de Laubier]

Hello,

I am using Stata 17 with predictnl version 1.8.2 and reghdfe version 6.12.2.

I try to estimate the predicted probability after a regression of the log odds ratio on covariates and many fixed effects.

I am running the following commands:

reghdfe log_odds_ratio depvar  [pw=weights], absorb(year county_fe) cluster(state) resid
predictnl pred_prob=exp(predict(xbd))/(1+exp(predict(xbd))) , se(pred_prob_se)

and I get the following error message:

expression is a function of possibly stochastic quantities other than e(b)
r(498);

Note that if I don't use a non-linear function of the estiamtes and just use predict, things work.

reghdfe log_odds_ratio depvar  [pw=weights], absorb(year county_fe) cluster(state) resid
predict pred_log_odds_ratio, xbd

Thanks a lot!
Nicolas

[FernandoRios]

Hi Nicolas
The error that you get is more an informative message regarding theproblem.
So, what you re trying to predict is log_odds_ratio_hat as a function of your dep variables and fixed effects. However, standard errors for fixed effects (year and country) are not estimated.
Because of that predictnl is telling you there is a problem on doing that, because in addition to the point estimate, you are also requesting standard errors.

The only way to fix the error is, as far as i know, explicitly add the dummies in your model, rather than try to "absorb" them.
HTH