I am conducting factor mixture modeling, with covariances free across classes. In two of my classes (in a four-class solution), I get an error message THE LATENT VARIABLE COVARIANCE MATRIX (PSI) IN CLASS 2 IS NOT POSITIVE DEFINITE...and the same message for CLASS 2. It appears that this is occurring because of a correlation greater than 1 between two of my factors. Does this mean that there is a problem with my model, or might it mean that in two of my classes, these factors are actually not different?
As a follow-up, if it is that these factors are better-represented as a single factor in two of the classes, would it be appropriate to model that as such (i.e., creating a model where two classes include these separate factors as a single factor)?
Thanks for your quick response. I am attempting to create different factors by classes. I am not sure how to model it though. I want to restrict two factors to a single factor in one of my classes but allow the factors to be separate in the other classes. When I tried to model the factors independently in each class, without listing them in the %overall% model, Mplus doesn't seem to recognize the factors that I created. Is there another approach. I was thinking that I could set the correlation between the two factors to one in the class that I wanted them to be the same, and then set the correlations between those factors and other factors to be equal. Would this work, or is there another way that I could model different factors in different classes?