How do I evaluate class homogeneity in parameters allowed to vary across classes in growth mixture models?
I've heard Katherine Masyn stress the importance of documenting this (along with separation) in a few (excellent!) mixture model courses, but I can't find guidance on how to do it outside of LPA. In that case, it's relatively straightforward (ratio of within class variance to overall population variance for an indicator), but it is not clear to me how to do that with the latent variables that are the indicators of class (intercept, slope, and quadratic, in my case). The SE of the class estimates for these values don't appear to be the answer, because depending on the model specification (LCGA versus diagonal/class invariant, etc.), residual variances of the observed indicators become bloated while SE remain small.