[Fwd: CLIN PHAR STAT: Mixed Vs Fixed]

Date: Sat, 05 Aug 2000 09:21:59 +1200 From: Nick Holford <n.holford@auckland.ac.nz> Subject: [Fwd: CLIN PHAR STAT: Mixed Vs Fixed] Stephen Senn contributed these comments to the cps@topica.com list. I thought they might be of interest to NMUSERs. Would anyone like to comment on the biased SE issue as it might apply to NONMEM? Is it possible to put "a random effect on the variances themselves" in NONMEM? stephens@public-health.ucl.ac.uk wrote: > > Of course, the mixed (random effects) analysis makes an > assumption of Normality that the fixed effects model does not > make. There is a further technical problem. Ideally, the between > and within patient information should be weighted by the true > variances. In practice they are weighted by the observed variances > and this leads to a downward bias in the estimated overall standard > error and a consequent bias towards significance in the p-values. > (A similar problem occurs in meta-analysis whether fixed or > random.) This can be corrected by putting a random effect on the > variances themselves but I would guess (I don't know) that this is > not incorporated currently in PROC MIXED in SAS. > > Regards > > Stephen -- Nick Holford, Dept Pharmacology & Clinical Pharmacology University of Auckland, Private Bag 92019, Auckland, New Zealand email:n.holford@auckland.ac.nz tel:+64(9)373-7599x6730 fax:373-7556 http://www.phm.auckland.ac.nz/Staff/NHolford/nholford.htm