Mixture Distribution and Latent Class Models

Here's an example growth mixture model from the NIDA workshop this week. I've copied the two functions I wrote and placed them in the text of this post for further discussion. Everything else in the code has been added to the GMM demo and documentation pages for the next binary release.

The first function calculates posterior class probabilities for a mixture model based on the following equation:

P(class=k|i, p_k, L_k) = p_k * L_ik / sum_k=1^m (p_k * L_ik)

Hi Gang,

I'm playing with a growth mixture model and am hitting an optimizer snag. My code continually ends with an error indicating one row of my data producing a non-positive definite expected covariance matrix between 30 and 60 iterations in, leading to an NPSOL status 6. Removing the offending row simply gives another row the opportunity to take center stage and become non-positive definite.

Questions, both for me and for future users with similar problems:
-What's the next thing I should look at to solve the problem, and relatedly:

Hi all,
I am starting to learn OpenMX and wondering if there are any scripts available for fitting growth mixture models, specified in a path-oriented notation. Any help would be greatly appreciated,
Thank you,
Mariya

Hi All,

On my way to doing other models, I found a way to estimate independent mixture distribution models in OpenMx by using the undocumented, unsupported, but oh so useful mxRowObjective function. I've attached the R script, its plain text output, and the data. The data is from Zucchini and MacDonald's (2009) text on Hidden Markov Models, and is freely available on the book's website .

Hi all,

Are example scripts available for latent class models and/or finite mixture models? Or are we waiting for the user guide to be updated?

Will Open Mx be able to deal with such problem.

I am in the begining of a PHD and I apm looking for a software to test such problem, being familiar with R and on mac os X I was wondering if OpenMx was an option ?

Thank you for this work.

Guillaume Avenin