Simon Davies posted on Tuesday, September 19, 2017 - 4:07 pm
I am testing a 3-factor model with 19 ordinal indicators for longitudinal measurement invariance. I have tried using CFA and ESEM but I keep running into problems with the residual covariance, seemingly because several of my indicators don't change much over time. I kept getting the error message saying RESIDUAL COVARIANCE MATRIX IS NOT POSITIVE DEFINITE, and the error message pointed to variables where the covariance with the same variable at another time point is greater than 1. I tried including a model constraint limiting residual covariance to no greater than 1 - this seemed to work when I had 2 time points in my models, but when I included a third time point, the NOT POSITIVE DEFINITE warning returned, and now I can't see the problem with the variable pointed to in the error message because the covariance is not greater than 1 for that variable.
Two questions: 1. Is it acceptable to use that model constraint to deal with problems with the residual covariance matrix? 2. If I have included the constraint, where should I be looking to find the cause of the error?
Simon Davies posted on Tuesday, September 19, 2017 - 7:41 pm
OK thank you. I will try posting it there.
Simon Davies posted on Thursday, September 21, 2017 - 9:47 pm
If I do include the model constraint, this seems to prevent me from using DIFFTEST - an error comes up saying difftest can't be used in conjunction with nonlinear constraints through the use of Model Constraint.
Is there an alternative available in Mplus for testing the change in chi-square when a model constraint is included?