Gareth posted on Tuesday, August 11, 2009 - 4:16 pm
I am running a LCGA, progressively increasing the number of latent classes from 2 to 8. The BIC value continues to decrease, until I get "ONE OR MORE PARAMETERS WERE FIXED TO AVOID SINGULARITY OF THE INFORMATION MATRIX" with the 8-class solution.
When Mplus fixes a parameter to avoid a singular, can the BIC values and other parts of the model be trusted? Is this an indication that there are too many classes, even though the BIC has not increased yet?
With LCGA, BIC can continue to decrease without hitting a minimum. You may find that GMM is better for your data in this case.
See also the following paper which is available on the website. It gives suggestions for determining the number of classes using measures other than BIC:
Nylund, K.L., Asparouhov, T., & Muthen, 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.
Results can be trusted when parameters are fixed unless this is done because of non-identification.