Re: st: constraining gamma 0,1 in GMNL command

["Heidi Pitts" <hmpitts@salud.unm.edu>]

Thank you for following up, Arne.  I appreciate it the information.  I had estimated the GMNL-II model constraining gamma to zero and it had a hard time converging, even with starting values from the uncorrelated model.  Perhaps that is an indication that the GMNL-I model is more appropriate.
 
Thanks again.
Sincerely,
Heidi Pitts
 
 

From: Arne Risa Hole <arnehole@gmail.com>
Subject: Re: st: constraining gamma 0,1 in GMNL command
Date: Mon, 31 Mar 2014 17:49:08 +0100
Dear Heidi,

You cannot impose the restriction that gamma lies between 0 and 1 I'm
afraid. But given that gamma is estimated to be negative in your
model, an ad-hoc solution would be to simply constrain it to equal 0.

Arne (co-author of -gmnl-)

On 28 March 2014 22:33, Heidi Pitts <hmpitts@salud.unm.edu> wrote:
> Dear Statalist
> I have been running the gmnl code written by Dr. Arne Hole and the estimated gamma is negative (gamma= -0.259). Fiebig etal (2010) constrain gamma to take on a value between [0,1], but other papers (Keane and Wasi, 2009) feel that is not necessary. The gmnl code does not constrain the gamma to lie between 0,1. However, I'd like to constrain it for the full GMNL model for comparison. Has anyone tried this before?
>
> I'm wondering if the best way is to write a constraint, using the solution suggested in (http://www.stata.com/support/faqs/statistics/regression-with-interval-constraints/), so that it would appear
> constraint 1 [gamma]_cons = 0+1*invlogit(gamma)
>
> Or is it necessary to go into the .ado file?
>
> Any suggestions would be appreciated. Thank you for your consideration.
>
> Best,
> Heidi Pitts
> Dept. of Economics
> University of New Mexico


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