13 per cent output never changed in a binary choiced panel data

[Rain Au]

Dear Statalist-Members,

I am new in Stata and data analysis.

- I have a panel data with binary dependent variable
-13% of the dependent variable never change during 7 waves(which is T).

When I tried to do xtlogit fixed effects, 13% observations have been omitted because of all positive or all negative outcomes.
I dropped these unused observations by using drop if !e(sample).
Then, I did pooled regression, xtlogit random effects, xtprobit random effects. Hausam test showed that I have to use xtlogit fixed effects.

THE QUESTION IS: Is there any better method to deal with a data like this(binary dependent variable in panel data and a big part of them never change)?
Does 13% observations be omitted influence the accuracy of result?

Any help is highly appreciated!
Thank you,

Rain Au

[Carlo Lazzaro]

Rain:
welcome to this forum.
If there's no within-panel variation in the regressand and/or the regressors predict the outcome perfectly, the ML esrimator suffers and, as such, Stata omits the culprits.
As data are what they are, I would stick with -xtlogit,fe- without dropping anything and explain in my researh report/article why 13% of the observations were omitted.

[Rain Au]

Rain:
welcome to this forum.
If there's no within-panel variation in the regressand and/or the regressors predict the outcome perfectly, the ML esrimator suffers and, as such, Stata omits the culprits.
As data are what they are, I would stick with -xtlogit,fe- without dropping anything and explain in my researh report/article why 13% of the observations were omitted.

Thanks a lot! Another dummy variable in the data represents the observations' countries. This variable is omitted, too. Because of never change. If I still want to get the relationship between the countries' variable and the binary dependent variable.
Would you mind giving me some suggestions?

[Carlo Lazzaro]

Rain:
the -fe- estimatopr (even when conditional, like in -xtlogit-) wipes out all time-invariant variable.
Therefore, you simply cannot estimate the coefficient for -i.country- under this specification: you should switch to -xtlogit,re- provided that it's the right estimator given your dataset.

[Rain Au]

Rain:
the -fe- estimatopr (even when conditional, like in -xtlogit-) wipes out all time-invariant variable.
Therefore, you simply cannot estimate the coefficient for -i.country- under this specification: you should switch to -xtlogit,re- provided that it's the right estimator given your dataset.

Mr. Lazzaro:
Thank you so much!

[Carlo Lazzaro]

Rain:
Carlo is enough! Thanks.