Is it possible, within the limits of mixture modeling, to determine the relation between two grouping structures. For example, what if we believe that there are several distinct classes of illness progression, and several distinct classes of life satisfaction change. Is there a way to derive both class membership AND the prediction of being in one class from membership in the other within a single model?
Just define 2 latent class variables, say c1 and c2, and use Model c1: and Model c2: to indicate which parameters vary across which latent class variable classes. Use Parameterization = loglinear and say c1 WITH c2 or c2 ON c1.