Dear everyone, I'm trying to fit a 3-class growth-mixture model with clustered data (families) and varying times of observations (age in years). I also want to predict the I and S factors using several binary variables (within each class).
It turns out that I get two error messages.
1. WARNING: THE SAMPLE COVARIANCE OF THE INDEPENDENT VARIABLES IS SINGULAR.
This appears as soon as I include any single covariate and at the same time TYPE is set to RANDOM (even if I don't use tscores in the model). It does not matter which covariate I use. I don't see any unit correlations or linear dependecies - especially since I only use one covariate, which also shows variance in all the classes.
2. WARNING: THE MODEL ESTIMATION HAS REACHED A SADDLE POINT OR A POINT WHERE THE OBSERVED AND THE EXPECTED INFORMATION MATRICES DO NOT MATCH.
This appears as soon as I model the clustered structure and thus cannot use the MLF-estimator. I have tried lowering the MITER and MCONV values and given starting values - this didn't help. MLF estimation unfortunately is not available with COMPLEX data. I do however get standard errors in the output.
I don't have any problems to run the RANDOM model with tscores using the MLF-estimator (without COMPLEX structure and predictors).