Re: st: multinomial logistic regression with RE using gllamm

[Nick Cox <njcoxstata@gmail.com>]

You asked three questions. I'll peel off the one I can answer, #1.

-year- could be a valid predictor to be added to your variable list.
That depends on whether you think that the response is varying with
-year-. That's a substantive question depending on what you think is
happening to generate the dataset. We can't answer that for you.
Naturally, the character of any dependence on -year- needs to be
thought through as well.

Nick
njcoxstata@gmail.com


On 25 April 2013 09:26, Andreas Schiffelholz
<Andreas.Schiffelholz@gmx.de> wrote:

> I'm working with a unbalanced panel dataset (x: companies; t:years) and two
> models with ordinal (2) / nominal (3) responses. While I have no problem in
> calculating the results of the ordinal models using a -logit, cluster
> (company)- approach for pooled and an -xtlogit, re vce(bootstrap)- approach
> for the random effects model, I strugle with the calculation of the
> multinomial logistic regression. I calculate the pooled model using the
> -mlogit, cluster (company)- command. For the multinomial logistic regression
> with random effects I uderstand that I need to use -gllamm-, because this is
> not implemented in the standard Stata program.
>
> I'm using the following command:
> -gllamm varlist, i(Company) base(0) link(mlogit) family(binom) nip(12) adapt
> trace-
>
> 1. Do I need to specify the "year" variable somehow? I already have the
> -xtset Company Year, yearly- command in my .do file, for the previous
> -xtlogit- command-, but I wonder if I need to specify it again just as i
> specified the Company variable by the -i(Company)- option
>
*
*   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/