V X posted on Thursday, December 20, 2007 - 6:35 pm
Hi,professor, I have a question about the warning message I got, when I was running a model using Monte Carlo integration.
At the beginning of the message, it said,
WARNING: THE SAMPLE COVARIANCE OF THE INDEPENDENT VARIABLES IS SINGULAR.
THE MODEL ESTIMATION DID NOT TERMINATE NORMALLY DUE TO A NON-POSITIVE DEFINITE FISHER INFORMATION MATRIX. CHANGE YOUR MODEL AND/OR STARTING VALUES.
THE MODEL ESTIMATION HAS REACHED A SADDLE POINT OR A POINT WHERE THE OBSERVED AND THE EXPECTED INFORMATION MATRICES DO NOT MATCH. THE CONDITION NUMBER IS -0.198D+00. THE PROBLEM MAY ALSO BE RESOLVED BY DECREASING THE VALUE OF THE MCONVERGENCE OPTION.
THE STANDARD ERRORS OF THE MODEL PARAMETER ESTIMATES COULD NOT BE COMPUTED. THIS IS OFTEN DUE TO THE STARTING VALUES BUT MAY ALSO BE AN INDICATION OF MODEL NONIDENTIFICATION. CHANGE YOUR MODEL AND/OR STARTING VALUES. PROBLEM INVOLVING PARAMETER 3.
RESULTS ARE PRESENTED FOR THE MLF ESTIMATOR.
Then, the last sentence it said,
THE MODEL ESTIMATION TERMINATED NORMALLY
I am wondering how should i deal with the message. Did my model run OK or I should be suspicious about the result I got esepcially the PARAMETER 3 (mean of latent growth rate)?
Your analysis went ok. When the MLF estimator goes through ok you can be confident that your model is identified.
Stine Hoj posted on Tuesday, February 03, 2015 - 11:10 pm
Dear Prof Muthen,
Could you please assist me in understanding the conditions under which it is safe to ignore this message?
"WARNING: THE MODEL ESTIMATION HAS REACHED A SADDLE POINT OR A POINT WHERE THE OBSERVED AND THE EXPECTED INFORMATION MATRICES DO NOT MATCH. AN ADJUSTMENT TO THE ESTIMATION OF THE INFORMATION MATRIX HAS BEEN MADE. THE CONDITION NUMBER IS -0.419D-02. THE PROBLEM MAY ALSO BE RESOLVED BY DECREASING THE VALUE OF THE MCONVERGENCE OR LOGCRITERION OPTIONS OR BY CHANGING THE STARTING VALUES OR BY USING THE MLF ESTIMATOR."
My model is a GMM with covariates predicting class membership & within-class intercepts/slopes. I am currently running this model using one imputed dataset to obtain starting values as input for a full TYPE=IMPUTATION analysis. I have decreased the mconvergence and logcriterion options and provided starting values, but this did not resolve the warning.
Does the warning mostly have implications for the accuracy of SE estimates, or should I be concerned about the accuracy of the beta estimates also? Specifically, given I am not interested in directly interpreting the output but only wish to export the beta estimates as starting values, is it safe to ignore the saddle point message?
Yes, the warning mostly has implications for the accuracy of SE estimates, not the parameter estimates.
This message occurs (1) when the likelihood is relatively flat (the data-model combination offers little information on what the parameter estimates should be), perhaps due to an over-parameterized model, and/or (2) when the numerical precision is low, for instance when not using enough points of integration. It is often accompanied with negative logL changes seen in TECH8. Solutions include a sharper convergence like
mconv = 0.000001;
or more integration points, or simplification of the model.