Re: st: Re: multinomial logit on grouped data

[Marcello Pagano <pagano@hsph.harvard.edu>]

The short answer to your question is, no.

A workaround is to do the marginal logistics and work back.  That is,
redo the problem three times, each time doing a logistic.  First time 
with y0
versus the rest, second time with y1 versus the rest, and the third time 
with
y2 versus the rest.  If your sample sizes are large and you are not close to
the boundaries (your ys are not close to zero), your answers should not be
that much different.  There is a paper to that effect,  Begg & Gray  (1984)
Calculation of polychotomous logistic regression parameters using
individualized regression, Biometrika 71: 11-18.

(I also submitted "begg gray" to Google and got this paper
www.mzes.uni-mannheim.de/publications/wp/wp-19.pdf
where an example is worked out.)

m.p.


Scott Merryman wrote:


> ----- Original Message ----- 
> From: "A. Mushfiq Mobarak" <Ahmed.Mobarak@Colorado.EDU>
> To: <statalist@hsphsun2.harvard.edu>
> Sent: Friday, July 11, 2003 3:25 PM
> Subject: st: multinomial logit on grouped data
>
>
>  
>
> > Dear Listers,
> >
> > Stata can perform logit/probit estimation on grouped data (see blogit,
> > bprobit), where the dependent variable is fraction of positive responses
> > in the group rather than an individual's 0-1 response.
> >
> > Can it perform multinomial logit estimation on grouped data?
> >
> > Specifically, the type of estimation I'm trying to do is the following:
> >
> > My dependent variable has 4 categories:
> >
> > y0: electricity usage by industrial groups
> > y1: electricity usage by agriculture
> > y2: electricity usage by consumers
> > y3: electricity usage by others
> >
> > I can express these as fraction of electricity used by each of the 4
> > groups.  I would like to explain this composition of electricity use as
> > a function of a number of independent variables (e.g. the size of each
> > group, political conditions in the country, etc.)
> >
> > Maybe multinomial logit is not even the right way to think about it, and
> > would like to hear other suggestions if that's the case. It seems that
> > LIMDEP allows you to do the multinomial logit on grouped data estimation
> > (see http://www.limdep.com/On_Line_Help/lim00320.htm). I would rather do
> > this on Stata.
> >
> > Thank you in advance for your help. Since I get the digest, copying your
> > response to mobarak@spot.colorado.edu would be appreciated.
> >
> > Best,
> > Mushfiq
> >
> >    
>
> There is a Stata FAQ: "How can I do logistic regression or multinomial
> logistic regression with grouped data?"
> (http://www.stata.com/support/faqs/stat/grouped.html) which may be of help.
>
> Scott
>
>
> *
> *   For searches and help try:
> *   http://www.stata.com/support/faqs/res/findit.html
> *   http://www.stata.com/support/statalist/faq
> *   http://www.ats.ucla.edu/stat/stata/
>  

*
*   For searches and help try:
*   http://www.stata.com/support/faqs/res/findit.html
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/