st: covariance matrix of multinomial logit

[Seamus Kent <Seamus.Kent@glasgow.ac.uk>]

Hi,

I have estimated a multinomial logit model in which the dependent variable has three categories. When I use the 'estat vce' command to extract the covariance matrix I receive an odd result that seems incorrect. The covariance matrix between regressors for outcomes 1 and 2 (base is 3) is non-symmetrical. Do you know why this is occurring?

The following contrived data demonstrates the problem. 

clear
input id income sex age_group outcome 
1 1000 1 1 3 
2 2000 1 2 1
3 2030 1 3 2
4 4000 1 4 3
5 6500 0 1 2
6 1000 0 2 3 
7 2000 0 3 1
8 2030 0 4 2
8 4000 1 4 3
9 6500 1 3 1
10 1000 1 2 3 
11 2000 1 1 1
12 2030 0 4 2
13 4000 0 3 3
14 6500 0 2 1
end

mlogit outcome income sex age_group
estat vce


Thanks,
Seamus

Seamus Kent
Health Economics & Health Technology Assessment 
Centre for Population & Health Sciences University of Glasgow
1 Lilybank Gardens
Glasgow G12 8RZ

Tel: +44 (0)141 330 3294 



*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/