Rachel posted on Sunday, March 05, 2017 - 12:36 pm
Dear Professor Muthén, In a latent growth curve model, I'm getting negative slope variance & errors about the theta or psi matrix. Are these causing issues: 1) The scores do not grow much. 2) I tried both S | WJAPWC1S@0 WJAPWCG3@.22 WJAPWCG5@.44 WJAPWCX5@1; and S | WJAPWC1S@0 WJAPWCG3@2 WJAPWCG5@4 WJAPWCX5@9; since the measurement occasions are 1st, 3rd, 5th, and 10th grade and I am trying to specify the uneven distances between the time-points. 3) The scale of the repeated measures is centered on 500 (at 5th grade). 4) The repeated measures are correlated at .6 - .88, but this is a longitudinal model, so I'd assume it is set up to handle such dependency. Do any of the above raise red flags alerting you as to what to fix and how? The way I coded things after "S|" is the only part that deviated from the textbook code I was following, and I am not confident I chose an appropriate numbers to assign to specify the distances between time-points.
If the residual variances for your 4 outcomes are negative at some time points you could try holding them equal. If the slope variance is negative (but not large) you could fix it to zero (everyone grows the same way).
Thank you very much for your response, Dr. Muthén.
When I try to hold the residual variances equal, I get the message: THE STANDARD ERRORS OF THE MODEL PARAMETER ESTIMATES COULD NOT BE COMPUTED. THE MODEL MAY NOT BE IDENTIFIED. CHECK YOUR MODEL. PROBLEM INVOLVING PARAMETER 2. THE CONDITION NUMBER IS -0.906D-13.
So I suspect I am probably not coding it properly.
What would be the proper code to do so?
Here is what my code looks like before trying to hold the residual variances equal: MODEL: I | WJAPWC1S@0 WJAPWCG3@0 WJAPWCG5@0 WJAPWCX5@0; S | WJAPWC1S@0 WJAPWCG3@2 WJAPWCG5@4 WJAPWCX5@9;
I get this message: WARNING: THE LATENT VARIABLE COVARIANCE MATRIX (PSI) IS NOT POSITIVE DEFINITE. THIS COULD INDICATE A NEGATIVE VARIANCE/RESIDUAL VARIANCE FOR A LATENT VARIABLE, A CORRELATION GREATER OR EQUAL TO ONE BETWEEN TWO LATENT VARIABLES, OR A LINEAR DEPENDENCY AMONG MORE THAN TWO LATENT VARIABLES.CHECK THE TECH4 OUTPUT FOR MORE INFORMATION.PROBLEM INVOLVING VARIABLE S.
Is there anything else I can do to help with the negative latent slope variance, other than fix it to zero, when it would not make sense theoretically to assume everyone grows the same way?