Re: st: Conditional expectations for each latent class with gllapred

[Partha Deb <partha.deb@hunter.cuny.edu>]

Stas and Kristian,

-fmm- does not support models for binomial outcomes because finite 
mixture distributions are not generally identified for binary outcomes.  
The setup in -glamm- imposes restrictions as Stas points out - only the 
intercept varies across classes and thus is, in principle identified 
(although if one estimated a model with no X variables, even this would 
not be identified).  -fmm- estimates "unconstrained" mixtures by 
default, so I chose not to include models for binomial outcomes.

Hope this helps.

Happy New Year to all.

cheers.

Partha

On 12/30/2010 10:37 AM, Stas Kolenikov wrote:
> On Thu, Dec 30, 2010 at 3:06 AM, Kristian Karlson
> <kristian.karlson@gmail.com>  wrote:
> > I have run the following -gllamm- model in Stata. It is a finite mixture
> > binary logit model with two latent classes:
> >
> > use http://www.ats.ucla.edu/stat/paperexamples/singer/hsb12.dta, clear
> > sample 25
> > gllamm union age grade year, i(idcode) ip(f) nip(2) l(logit) f(binom)
> >
> > I am interested in the conditional expectation for latent class: Pr(Union =
> > 1 | u_1) and Pr(Union = 1 | u_2), where u_1 is latent class 1 and u_2 is
> > latent class 2.
> >
> > I have looked at -gllapred-, but haven't been able to compute these. My idea
> > was to use options mu and marg, but these probabilities are the mixed
> > probabilities, not the ones from each component.
> Your example does not run:
>
> . gllamm union age grade year, i(idcode) ip(f) nip(2) l(logit) f(binom)
> variable union not found
> r(111);
>
> You probably meant a different data set.
>
> Note that this is a pretty restrictive model in which the effects of
> the predictors are kept constant across classes, and the difference is
> only via a shift.
>
> You might be able to manipulate -gllapred- using -from()- option. For
> that, you can create two matrices with the estimated point masses and
> their weights fixed to one and the other class. Just a suggestion, I
> never worked with mixture models using -gllamm-.
>
> Your other option is to use -fmm- package that might provide
> additional flexibility if it supports -logit- link (or an analogue
> of).

--
Partha Deb
Professor of Economics
Director of Graduate Studies
Hunter College
ph:  (212) 772-5435
fax: (212) 772-5398
http://urban.hunter.cuny.edu/~deb/

Emancipate yourselves from mental slavery
None but ourselves can free our minds.
	- Bob Marley

*
*   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/