boottest with xtlogit giving an error message

[Andia Kozmai]

Dear Statalist Community,

I am analysing panel data with Stata 17. I am running two xtlogit regressions and trying to use the boottest command and I get two different error messages for both regressions. (Without clustered SEs, I still get the error messages.) Note that with the same dataset and xtgls regressions, the boottest command with clustered SEs works.

Case 1:

xtset group phase

xtlogit num price max trend fin i.phase

boottest price, cluster(group)

The xtlogit command shows an output, but the boottest command shows:

Could not generate scores from xtlogit with predict.

r(198);

Case 2:

xtset sub phase

xtlogit rob profit price max trend fin i.phase

boottest, cluster(group)

Again, the xtlogit command shows an output, but the boottest command shows:

Overriding estimator's cluster/robust settings with cluster(group)

(h0(1) assumed)

Constraint 1 not found and will be skipped.

r(111);

I am a bit confused about these problems with xtlogit and boottest, and what I have read until now has not helped me. I would really appreciate if anybody understands a bit more about the error messages and can explain to me what is going on, and/or can recommend a way to solve it.

Thanks a lot and best regards!
Andia

[David Roodman]

After a nonlinear model like xtlogit, boottest can only perform the score bootstrap of Kline and Santos. To do this, it needs to be able to get the scores from your model fit, using the predict command. At “help xtlogit postestimation##predict”, you will see that the score option is not available for the default random effects xtlogit model. That is why the first example fails.

The second example fails because it doesn’t specify a null hypothesis.