SF Wang posted on Thursday, September 12, 2013 - 10:15 am
We usually need to do further analyses based on the class membership obtained from GMM, such as identifying predictors for class membership.
I have heard some people saying that including covariates in GMM and then in further analyses seems like circular - they are used to form the classes and then were examined whether they are predictors of the classes, so they advocate not to include covariates in GMM.
Personally, I don't agree. If those variables are significant factors for the class membership, we should include them in GMM, so that GMM could provide a better classification.
If we should include covariates for this purpose, should we have the covariates for both the random effects and the latent classes, or just the latent classes?