Dear all,

I would like to estimate a multinomial logit model where the explanatory variables vary only by alternative (compared to the case where they vary by both alternative and individual). It seems that the asclogit command should do the job. However, asclogit tells me that the resulting likelihood function is not concave. I am thinking that perhaps asclogit only supports regressors that vary by both alternative and individual. Is there any other command that estimates multinomial logit models where regerssors vary only by alternative? Please see my code below:

Code:
 clear
 
 use http://www.stata-press.com/data/r13/choice
 
 list id car choice dealer sex income in 1/12, sepby(id)
 
 * The dealer variable varies both by individual and alternative
 
 asclogit choice dealer, case(id) alternatives(car)
 
 * The dealer variable varies only by alternative.

 replace dealer = 18 if car == 1
 replace dealer = 8 if car == 2
 replace dealer = 12 if car == 3
 
 list id car choice dealer sex income in 1/12, sepby(id)
 
 asclogit choice dealer, case(id) alternatives(car)
Many thanks for any help you can provide!

Best wishes,
Renata