Hi,
I am trying to run a growth mixture model using a bayesian estimator with unknown a priori class membership in Mplus. Comparing 2, 3, and 4-class solutions, it seems that the analyses always end up with solutions with classes of equal (or almost equal) size. Is it a coincidence or a specific feature of Bayesian GMM in Mplus? Of note, I don't get similar results when running the same analysis with an ML estimator.
My first questions would be why did you use a Bayesian approach and what kind of priors did you use (in particular, priors for the class size parameters)? It seems somewhat suspicious that Bayesian and ML estimation would lead to very different results.
Christian Geiser I used a Bayesian approach because I have a fair amount of missing data and not a big sample. I used Mplus' default settings for priors.
I would check with the Mplus support team. They could probably help you best to figure out what's going on.
I'm not an expert with Bayesian but did you check convergence, trace plots etc.? Maybe the results are unstable due to the small sample.
You could try and post your question on the SEMNET listserv. You may get a better answer there. I think there are at least some people on the listserv with expertise in Bayesian SEM.
- Badges
- Science method