Panel Multinomial Logit Model

[Phindile Nkosi]

Good day,

Kindly assist running a panel multinomial logit model. My dependent variable is categorical, with 5 categories (cooking). I ran a pooled multinomial and it was ok. Now I would like to run a panel multinomial fixed effects. The estimation is as follows:

xtmlogit cooking scaled_pcc net_monthly_income incomesq employment_stat birth_yr i.education race, fe

The error I get says "outcome does not vary in any group". Does anyone know what the problem may be?

Thanks in advance.

Regards,
Phindile

[Andrew Musau]

What is your panel identifier (i.e., the variable that you used to xtset the data)? If your data is on individuals, fixed effects (FE) panel multinomial logit supposes that an individual is faced with a series of choices over time and at each time, she has to select one of the choices (from the same menu). As the FE estimator is the within-groups estimator, you require variation over time (the individual chooses a different option at least once over the sample period) for identification. If it happens that each individual never chooses a different option, you cannot estimate an FE model.

[Phindile Nkosi]

Dear Andrew,

Thanks for the response and the explanation. I am using a household dataset and the identifier is the household ID. I actually get the same error for both fe and re. The data is for 5 years.

[Andrew Musau]

You should get some estimates with RE if you have variation in choices between households. Can you show the result of

xtset hhid
xttab cooking

where you replace "hhid" with the name of your household identifier and "cooking" with the name of your outcome if different.