Dear all,
-gllamm- offers the opportunity to mix distributions; that is,
-gllamm- can jointly estimate switching and sample selection models,
i.e., modeling (1) the selection process and (2) the outcome process.
For example, (1) a binary selection process into high school and (2)
social class destination as an ordinal process. This model would mix a
binary logit (or probit) with an ordered logit (or probit). -gllamm-
comes with a wrapper, -ssm-, which does much of the work for you. At
www.gllamm.org worked examples can be found, see especially
http://www.gllamm.org/books/kenkel.html, which comes from Skrondal &
Rabe-Hesketh 2004: Generalized Latent Variable Modeling, chap. 14.
However, I have to mix two multinomial logit models and capture the
unobserved heterogeneity affecting both processes with a finite number
of latent classes. Thus I cannot rely on -ssm- and have to do the
programming myself. My problem is that I cannot make out the meaning
of how to expand my data in order to mix the two distributions!
Expanding the data is a necessary condition for mixing two multinomial
logits in -gllamm-. My selection process has three alternatives and my
outcome process has four alternatives. Hence, I need to expand data
for both processes. I have tried to do so, but -gllamm- reports
missing values.
I have previously worked with the expansion option in modeling
single-process multinomial logits. This model worked just fine.
Nonetheless, I have not been able to figure out how to mix two
multinomial logits, simply because of necessity of expanding the data.
Thus can anyone help me with how to expand my data in order to utilize
the powerful -gllamm- routine? I would be grateful if anyone has a
worked example. This would be a great help.
Thank you,
Kristian
*
* 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/
