Hi,
I am running a multi-level model with a random intercept (no other random effects) and keep encountering the error in SPSS that
"The final Hessian matrix is not positive definite although all convergence criteria are satisfied. The MIXED procedure continues despite this warning. Validity of subsequent results cannot be ascertained."
I have ensured the variables in the model are not highly correlated and have altered the convergence criteria (e.g. increasing MXITER, MXSTEP and SCORING) yet the problem still remains.
Any suggestions as to how to resolve the issue would be greatly appreciated.
Thanks!
This discussion might point you in the right direction (http://www.methodspace.com/forum/topics/hessian-matrix-error-in), and this is a helpful explanation (http://www.theanalysisfactor.com/wacky-hessian-matrix/). Try standardizing your predictor variables (eg, root mean square). If that doesn't fix the problem, consider respecifying the random effect since the error is basically saying that the best estimate for the variance of the random effect is zero.
This discussion might point you in the right direction (http://www.methodspace.com/forum/topics/hessian-matrix-error-in), and this is a helpful explanation (http://www.theanalysisfactor.com/wacky-hessian-matrix/). Try standardizing your predictor variables (eg, root mean square). If that doesn't fix the problem, consider respecifying the random effect since the error is basically saying that the best estimate for the variance of the random effect is zero.
You might ponder a step further and consider that, if the only random effect in your model has no variance, you may not be using the right method to model your data. The first step in good multi-level analysis is always to inspect the descriptive characteristics of your variables.
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS