RE: $OMEGA blocks and log-likelihood profiling

From: Leonid Gibiansky lgibiansky@emmes.com Subject: RE:[NMusers] $OMEGA blocks and log-likelihood profiling Date: Thu, June 10, 2004 5:47 pm Nick Every rule has exceptions. I will not write to the editor, of course (why you included this suggestion to the message?) but this is an open forum, and everybody can express their opinions even the opinions that are not based on the data. My opinion is that the results based on the strongly failed run should be taken with great suspicion, even if they are confirmed by 1000s more strongly failed bootstrap runs. I would agree with you that you can accept the model (treated as black box) and model predictions based on the model diagnostics (who cares how the model was obtained if it describes the data well enough?) but it is more difficult to argue that the parameters and CI obtained by this procedure reflect true parameters and CI. It cannot be only NONMEM problem. NONMEM was able to fit thousands of models to thousands of data sets successfully. The failure of the program on one particular data set means that this particular model and this particular data set have some intrinsic problems so that general and stable nonlinear model fitting algorithm fails on this particular problem. In any case, this means that experience gained on this problem is unique and cannot be (or should not be) generalized to other problems. Leonid