April posted on Thursday, January 26, 2006 - 2:29 pm
Dear Drs. Muthen,
I have a cross-sectional data on injection risk behavior. I'm trying to examine the distribution of those risk behaviors in unknown classes and what covariates may affect that. Before adding any covariate into the model, I ran an analysis with 6 binary indicators (which were combined from 18 risk items) and 3 classes.
The input is attached below. First, the warning says "All variables are uncorrelated with all other variables within class." Then, it says, unperturbed, as well as 2 perturded, starting value runs did not converge, the model estimation did not terminate normally due to a change in the class counts during the last E step.
I'm wondering if this could be due to a sample size problem, or LCA is not an appropriate model by the nature of the data, and if any solutions?
I appologize if this kind of problem had already been discussed before. I searched but just didn't find it. Thank you very much.