Dear all,
I am trying to code up a simple EM algorithm to estimate a latent
class model (a.k.a finite mixture on a higher level than the
observations). This should be doable if I can keep track of all my
parameters in the score and Hessian (of simple classical normals) in
the M-step (maximization of log-likelihood for a mixing taken as
given). I'll try to keep the M-step in -moptimize-. But it would be
also nice to embed the M-step within another moptimize call for the
whole procedure. Do you think it is feasible, or it is unnecessary, or
even unfeasible? It would be great to have some expert opinion on this
before I go down that path.
Thank you,
Laszlo
*
* 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/
