I'm running a GMM with 4 time indicators (ppT1-ppT4). I have 2 covariates (x1T1-x2T1) that I'd like to use to inform the model when creating the classes. In other words, I want the model to take into account these variables while creating the classes, not after the classes are created (such as if x1T1 and x2T1 were predictors). What's the best way to do this (correlate the covariates with ppT1, use them in an auxiliary statement, etc)? I'd like to be able to draw conclusions similar to, "Group A is high on X1T1 and low on X2T1, starts high and decreases over time, whereas Group B is high on X1T1 and X2T2, starts high and is stable." I also have a distal outcome that I'm comparing means on using the BCH method. Syntax below:
I don't think so. Conceptually, what I'm trying to do is have the model simultaneously cluster individuals by both their T1 behavioral characteristics and their over time trajectories. The covariates (x1T1 and x2T1) have known direct effects on ppT1, so I'm not sure if regressing the latent class variable on them is the best route.