dewant2 posted on Thursday, March 24, 2005 - 1:09 pm
HI, I'm pretty new to mplus and have been trying to model growth of a continuous variable over time in a mixture analysis. Orgiinally I set each of the 4 time periods to 0,1,3,and6 (the data collection periods) and the model ran but gave me the warning: WARNING: THE RESIDUAL COVARIANCE MATRIX (PSI) IN CLASS 1IS NOT POSITIVE DEFINITE. PROBLEM INVOLVING VARIABLE S. I tried freeing up the parameters and changing the number of latent classes (against my judgement and the indices fit) but no matter what, I get the same warnings about the slope in each latent class. example with 2 free parameters MODEL: %OVERALL% i s | CESDM0@0CESDM1@1 CESDM3* CESDM6*;
THE MODEL ESTIMATION TERMINATED NORMALLY
WARNING: THE RESIDUAL COVARIANCE MATRIX (PSI) IN CLASS 1 (...2 3 and 4) IS NOT POSITIVE DEFINITE. PROBLEM INVOLVING VARIABLE S.
Any suggestions to get correct the model to remove the warning? Thank you so much
bmuthen posted on Thursday, March 24, 2005 - 1:37 pm
You should update to version 3.12 where this message is explicated further to suggest reasons for the problem. E.g. you may have a zero variance for s, or s may be perfectly correlated with i for some reason (see Tech4). Often when doing growth mixtures, the within-class variation of s is rather small because the classes capture a majority of the variation.