Mixture analyses using Akaike Information Criterion

#1
Hi guys,
Just wondered if you could shed some light on interpreting results of a mixture analyses using Akaike Information Criterion.
I am trying to fit a model to my data to determine whether within the sample there are 2, 3 or 4 potential groupings.
I have been using PAST software for this which uses the AICc formulae. How do I determine which is the greatest fitting model, is it by using the Log l.hood value or the AIC value.
And what am I looking for in terms of value to demonstrate greatest fit?
I have attached a sample of my results to demonstrate both the readout AIC and l.lhood values of which I am dealing with.
Cheers all
Sam
 

Ktau

New Member
#2
It looks like PAST offers very little in terms of fit statistics. Log likelihood is not appropriate for this sort of comparison. The AIC can be used (lower = better), but it doesn't function well under all circumstances. Theory and group meaningfulness/utility should play major roles in determining how many groups you're working with. Aside from that, you'd use a variety of fit statistics if they were at your disposal, but if you're stuck with PAST, then the AIC is it.

A very relevant paper that discusses these things:

Nylund, K. L., Asparouhov, T., & Muthén, B. (2007). Deciding on the number of classes in latent class analysis and growth mixture modeling: A Monte Carlo simulation study. Structural Equation Modeling, 14, 535-569.