I am having some difficulty getting the MPLUS code correct for a project I am currently working on. I am trying to do some Latent Class analysis to group subjects based on variables x1-x4. These variables are categorical with the variables equal to 0, 1, or 2. We would like to assume that all of these variables are correlated with each other, which we accomplished using the WITH statement. However, I get the following error message (for each 2x2 variable combination) when I run the code:
“Covariances for categorical, censored, count or nominal variables with other observed variables are not defined. Problem with the statement: x1 WITH x2”
With maximum-likelihood estimation - which is what you need for mixtures - Mplus does not allow you to correlate variables using WITH unless they are continuous variables. But, the Latent Class Analysis model makes the x's correlated by them being influenced by the same latent class variable so you don't need to correlate the x's further - that would not be LCA. Such correlations would be within-class correlations which if needed are best done by applying a factor as in factor mixture modeling - see e.g.
Muthén, B. (2008). Latent variable hybrids: Overview of old and new models. In Hancock, G. R., & Samuelsen, K. M. (Eds.), Advances in latent variable mixture models, pp. 1-24. Charlotte, NC: Information Age Publishing, Inc.