I am trying to run an LCGA on 6 waves of self-report data of 2 subscales of the same questionnaire. The scales are known to be correlated, their growth trajectories are known to be curvilinear, and there are gender differences in mean levels, slope, and intercept.
In order to incorporate all of this, I have modelled two seperate curvilinear growth curves, I have correlated their indicators (observed scores of the questionnaire subscales) within time, and regressed the intercept, slope, and quadratic term of both growth models, as well as the latent class variable, on sex. The optimal model includes three classes.
I'm not sure whether this is the corrrect way to model the data. I've tried to follow Jung & Wickrama's steps. What causes me anxiety, is the fact that the graph of estimated means is very different from the means I obtain from the Model Results. The growth trajectories displayed in the Estimated Means graph are RADICALLY different from the growth trajectories implied by the means in the output.