Dear Mike and others,
Thanks to Sophia Rabe-Hesketh, I have found the answer to my question. The
question also applies to two related threads posted earlier
(http://www.stata.com/statalist/archive/2008-11/msg00767.html and
http://www.stata.com/statalist/archive/2008-12/msg00623.html).
Sorry Mike for not responding on your answer before now. I am aware of the
work by Hill and his colleagues. I was, however, interested in using Full
Information Maximum Likelihood (FIML) to estimate my model, because I have
to model the membership of the latent classes simultaneously with the
"structural part" of the model. -gllamm- is excellent for this kind of
modeling and, as far as I remember, Hill separates the estimation into
several "stages".
For the interested Stata users, I have created an example that can be used
as a point of departure for those, who want to estimate their own
multinomial transition models with unobserved heterogeneity. Familiarity
with -gllamm- is a prerequisite (see www.gllamm.org). The example contains a
do-file and a dataset. Both files are zipped and can be downloaded from:
www.webcite.dk/mix2mlogit.zip.
The basic clue to the solution of my problem was that of first separating
the two processes into two datasets and manipulating the data for each
process. Second, the two datasets had to be recombined and some further
manipulations needed to be done on this combined dataset. Third, the model
could now be estimated with -gllamm- with the specification of the "random
effects" as desired. I owe special thanks to Sophia Rabe-Hesketh for
providing the basic do-file. The do-file in the downloadable example
contains only minor changes compared to that of Rabe-Hesketh, but my do-file
specifically estimates a multinomial transition model with two transitions
and with unobserved heterogeneity from the transition processes captured by
two latent classes.
All the best
Kristian
-----Oprindelig meddelelse-----
Fra: [email protected]
[mailto:[email protected]] På vegne af Mike Lacy
Sendt: 29. november 2008 23:19
Til: [email protected]
Emne: Re: st: Multinomial transition model with unobserved heterogeneity
>Date: Fri, 28 Nov 2008 19:30:16 +0100
>From: "Kristian Karlson" <[email protected]>
>Subject: st: Multinomial transition model with unobserved heterogeneity
>
>Dear expert users,
>
>Does anyone have suggestions about how to estimate a multinomial
>transition model with unobserved heterogeneity (captured by latent
>classes) for educational transitions in a multitier system? The first
>choice is among three educational tracks (in secondary education), and
>the second choice--contingent on a specific first choice¬¬?is among
>four tracks (in higher education). Maybe someone out there are
>familiar with this model and have tried to estimate it in Stata.
>
>A similar model can be found in Breen & Jonsson 2000: "Analyzing
>Educational Careers: A Multinomial Transition Model", American
>Sociological Review, vol. 65 no. 5. Breen & Jonsson, however, estimate
>the model in LEM, but this is not an option for me (because of server
>software limitations).
What you describe sounds like an application for
a nested logit model, with one equation to model
the first level of choice, and another to model
the second level. For an application in the
context of transition data analysis (event history), see:
Competing Hazards with Shared Unmeasured Risk Factors
Daniel H. Hill; William G. Axinn; Arland Thornton
Sociological Methodology, Vol. 23. (1993), pp. 245-277.
If I remember correctly, what this article
describes as "Shared Unmeasured Risk Factors" is
the same as "unobserved heterogeneity." The
solution is not via latent classes, but I suspect
it gives similar results. I have tried to apply
this in a situation similar to yours, and I would
think it could work for you. Let's hope someone
with deeper knowledge than me will chime in here to offer some more insight.
Regards,
=-=-=-=-=-=-=-=-=-=-=-=-=
Mike Lacy
Fort Collins CO USA
(970) 491-6721 office
*
* 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/
*
* 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/
