Gareth posted on Wednesday, August 04, 2010 - 12:15 am
I have a growth mixture model with complex survey data, which means I cannot get the BLRT test with TECH14. The BIC values are not helpful in choosing the number of latent classes, because I don't see the expected fall and then rise across different numbers of classes.
Are there any other suggestions (other than using prior knowledge and theory), such as running without the type=complex to get the BLRT test first, and then switching to type=complex afterwards?
BIC usually works pretty well. Perhaps the issue is that the model is not correct for the data. I would not select the number of classes without TYPE=COMPLEX as you would be making the decision treating the data as though they are independent when they are not.
Hi Linda, I actually read the paper by Nylund et al. (2007): Subtypes, Severity, and Structural Stability of Peer Victimization. They recommend usind the BLRT (instead of the LMR) in deciding on number of classes in nested data. But how can I obtain the BLRT when using type= mixture complex? Thanks a lot!