Correlation > 1 between slope and qua...
Message/Author
 Nathan Smith posted on Tuesday, June 28, 2016 - 1:15 pm
I am trying to estimate a simple growth curve at 4 time points.

i s q | tot_pm1@0 tot_pm2@1 tot_pm3@2 tot_pm4@3;

This initial model ran the following two warnings:

WARNING: THE RESIDUAL COVARIANCE MATRIX (THETA) IS NOT POSITIVE DEFINITE.

WARNING: THE LATENT VARIABLE COVARIANCE MATRIX (PSI) IS NOT POSITIVE
DEFINITE.

For the first warning, I could see that the residual variance for the variable tot_pm1 was negative, but non-significant. So I constrained the residual variance to be 0.

This solved the first warning, but I still have the second warning.

Looking through the output I can see that the correlation between the slope and quadratic slope are correlated below -1.

I have read through a number of other threads to figure out how I can determine what might be causing this and how to solve it. Any advice you might have would be appreciated.
 Linda K. Muthen posted on Tuesday, June 28, 2016 - 1:52 pm
You can try correlating residuals of the outcome at adjacent timepoints, for example,

tot_pm1 WITH tot_pm2;
tot_pm2 WITH tot_pm3;
tot_pm3 WITH tot_pm4;
 Nathan Smith posted on Tuesday, June 28, 2016 - 2:04 pm
Thank you Dr. Muthen. The issue is that I only have 2 df. How can I correlate the residuals in this way? Do you suggest freeing up df in some other way?
 Linda K. Muthen posted on Tuesday, June 28, 2016 - 2:40 pm
Then you can free only two.