RE: $OMEGA blocks and log-likelihood profiling

From: jeffrey.a.wald@gsk.com Subject: RE:[NMusers] $OMEGA blocks and log-likelihood profiling Date: Wed, June 2, 2004 3:04 pm I am with Ken on this one. I am not too troubled by the lack of the $COV, but the low fraction of convergence is bothersome. Whatever the cause, I have a hard time reasoning what the bootstrap actually means in this case. The sample of parameter estimates is effectively censored by a presumably non-random filter (vis. NONMEM). Marc's suggestion sounds pragmatic, but I still have to think that something more systematic is occurring with the structural model. If some parameter/s is/are not identifiable in a large fraction of the runs then my feeling is that the 'correct' bootstrap distribution for one or more of the parameters would have a large fraction of its values (up to 72%?) sitting on 1 or 0 or whatever its null value would be. Jeff Jeff Wald, PhD jeffrey.a.wald@gsk.com Clinical Pharmacokinetics/Modeling and Simulation Neurology and GI RTP, NC