Dear all,
I am interesting in investigating the heterogeneity of the ordinal
outcomes in my sample. So far I've been running ordered probit
regressions with county dummies and, to avoid any problem related to
inconsistency of estimators in such ordered probit fixed effects, I
also use random effects using GLLAMM.
In other terms I am exogenously fixing the "classes" (i.e., counties)
in order to study (spatial) heterogeneity.
However it will be interesting to run some ordered latent class
(finite mixture) models to verify what are the number and
characteristics of classes endogenously determined by the model.
My question is a simple one. I used Stata for all the statistical
analysis and I'd like to know what you think about implementing GLLAMM
to run ordered latent class models and what are the advantages and
disadvantages. Would it be better to use a specific latent class
software instead? If I well understood, the choice of the number of
classes in such models is determined by using BIC and AIC. How I can
obtain such statistic if I run GLLAMM? I read the manual, but I think
they just compare LL ratio, which is not the best way.
To identify such classes studies usually implement Simulated Annealing
EM algorithm, what GLLAMM use is not perfectly clear to me (adaptive
quadrature using -adapt- command, I guess).
Any answer and discussion will be very appreciated.
Thanks to all of you for the amazing service of Statalist.
Best,
Mirko Moro
PhD Candidate
University College Dublin
Dublin
*
* 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/
