RE: $OMEGA blocks and log-likelihood profiling

From: "Kowalski, Ken" Ken.Kowalski@pfizer.com Subject: RE:[NMusers] $OMEGA blocks and log-likelihood profiling Date: Wed, June 2, 2004 11:19 am Marc, Nmusers, I just have a hard time believing that this is the primary reason when so large a fraction (78%) fail to converge, and for those that do converge the $COV step fails. Certainly there are a number of reasons why we may encounter minimization failures but I would first rule-out over-parameterization before I pursued other explanations when such a large fraction fail. It would be of interest to review the $COV step output for some of the bootstrap runs of the 7% that had successful $COV estimation to see if the diagnostics suggest that the model is over-parameterized for these bootstrap runs. For example, are there extremely large SEs for one or more of the parameters, are there any pairwise correlations near +/-1, etc? If so, then over-parameterization would remain high on my list of potential reasons for the large fraction of minimization failures. Ken