xtmlogit with fixed-effects

[Carsten Waeltner]

Dear statalist members,

currently I´m trying to regress a multinominal logit with fixed-effects.
The dependent variable has 3 categories (1,2,3) and I have several other categorical and contiunous variables in the regression.

The problem however is, that random-effects can be computed but fixed-effects regression does not work.
The error stata gives when pulling the FE model is:

"infeasible number of permutations
r(459);"

Syntax:
xtset Panel identifier
panel identifier consists of 7 groups
xtmlogit DV x1 x2 b1.a ... , baseoutcome(2) rrr fe
with DV = dependent variable
x1, x2 continuous variable
b1.a categorical variable with base 1

I tried to find something via an ordinary google search, but there were no results of this error message combined with the xtmlogit command.

I sincerely hope, that one of you might have a solution for that problem.

With kind regards
Carsten

[Quy Pham]

I ran into the same problem. Do you know how to fix this issue now? If so, please share.

[Joerg Luedicke (StataCorp)]

Fixed-effects multinomial logit suffers from a dimensionality problem that is primarily due to T, the number of repeated observations. With large T, you would either have to find a way to reduce T or use the random-effects estimator. For further information about this problem, please have a look at the 'Remarks & Examples' section of the xtmlogit manual: https://www.stata.com/manuals/xtxtmlogit.pdf