Re: Confidence interval

From:Nick Holford Subject: Re: [NMusers] Confidence interval Date:Mon, 19 Aug 2002 23:42:11 +1200 Benito, You are quite correct to be worried about using the asymptotic standard errors produced by NONMEM to construct a confidence interval. It is not unusual to find examples were such a confidence interval is not possible e.g. negative $OMEGA. The problem arises because the SEs produced by NONMEM are only asymptotic estimates BUT more importantly you must assume a symmetrical distribution to construct a CI. When CIs are constructed using methods such as the log likelihood profile or by non-parametric bootstrapping they are often found to be assymetrical. If you really want to know the CI for your parameters then I suggest you consider using the log likelihood profile method (5-10 NONMEM runs per parameter) or bootstrapping (1000 NONMEM runs but this will let you construct CI on all parameters in the model). Nick -- Nick Holford, Divn Pharmacology & Clinical Pharmacology University of Auckland, 85 Park Rd, Private Bag 92019, Auckland, New Zealand email:n.holford@auckland.ac.nz tel:+64(9)373-7599x6730 fax:373-7556 http://www.health.auckland.ac.nz/pharmacology/staff/nholford/