Re: st: conditional and mixed logit

[Shehzad Ali <sia500@york.ac.uk>]

Thank you, Maarten, this is very helpful.

Regarding the demographic variables, would you use the same logic for the 
mixed logit models and leave these variables out? When I tried leaving them 
in, I keep getting conformability error with -mixlogit-.

Your suggestion of not making time-related variables random makes sense but 
I still get the conformability error when using demographic variable.

Regards,

Shehzad

On Jan 19 2009, Maarten buis wrote:

> --- Shehzad Ali <sia500@york.ac.uk> wrote:
> > I am using conditional and mixed logit models for discrete choice
> > data (-clogit- and -mixlogit-). The demographic characteristics like
> > age, sex, education etc obviously do not vary for the same
> > respondent. When I use them the following model, Stata drops the
> > demographic variables from the model.
> >
> > clogit choice length frequency benefit age sex education, group(n)
> >
> > Its understandable that since demographic variables do not vary
> > within each a choice set (n) the demographic variables will not be
> > allowed. But these variables do vary between individuals and I want
> > to account for that. Can I use mixed logit model for this purpose?
>
> If you just want to control for these demographics than you can just
> use -clogit- without the demographic variables, as information that
> remains constant within individuals is already absorbed in the fixed
> effects. Conceptually you can think of it as adding dummies for
> individuals. If you do that than you can't add a dummy for sex, as all
> the information in the sex dummy is already present in the person
> dummy.
>
> > When I use the following -mixlogit- command, I get "comformability
> > error". Can anyone please suggest why and whats the best way to
> > accommodate demographic variables in the model?
> >
> > mixlogit choice, group(n) id(id) rand(length frequency benefit age
> > sex education)
>
> It is usually a good idea to simplify the model by making less
> coefficients random. Mean centering the variables usually also helps. I
> am actually not sure whether it makes sense to make time constant
> variables random. I am willing to be confinced either way, but it looks
> odd to me. 
>
> -- Maarten
>
> -----------------------------------------
> Maarten L. Buis
> Department of Social Research Methodology
> Vrije Universiteit Amsterdam
> Boelelaan 1081
> 1081 HV Amsterdam
> The Netherlands
>
> visiting address:
> Buitenveldertselaan 3 (Metropolitan), room N515
>
> +31 20 5986715
>
> http://home.fsw.vu.nl/m.buis/
> -----------------------------------------
>
>
>      
> *
> *   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/

--
Shehzad I Ali
Department of Social Policy & Social Work
University of York
YO10 5NG
+44 (0) 773-813-0094


*
*   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/