From: Patrick Zhou patrickmzhou@yahoo.com
Subject: [NMusers] Weighting in NONMEM
Date: Wed, 19 Apr 2006 13:03:20 -0700 (PDT)

Dear All,
 
In NONMEM, we can assign a weight during minimization by using  W=F (interatively reweighted least
square weighting) in the $ERROR. What if we do not provide weight in the $ERROR, what does NONMEM use?
OLS, ELS or something else. Can somebody comment on it? Thank you very much.
 
Pat
_______________________________________________________

From: "GIRARD PASCAL" PASCAL.GIRARD@adm.univ-lyon1.fr
Subject: RE: [NMusers] Weighting in NONMEM
Date: Thu, 20 Apr 2006 08:12:57 +0200

Dear Pat,
 
Writing W=whatever in $ERROR has no action at all on NONMEM weightings. The purpose of writing
this is simply to compute individual weighted residuals (IWRES) which are not part of NONMEM output,
as opposed to population WRES provided by NONMEM.   IWRES have to be computed according to individual
predictions F, evaluated according to POSTHOC ETAs which are evaluated either with POSTHOC option
if you use FO or are part of the computation with FOCE.
 
Best regards,
 
Pascal
 
Dr Pascal Girard
EA 3738, Ciblage Thérapeutique en Oncologie
Fac Médecine Lyon-Sud, BP12
69921 OULLINS Cedex
France
Tel  +33 (0)4 26 23 59 54 / Fax +33 (0)4 26 23 59 76
_______________________________________________________

From: "Bill Bachman" bachmanw@comcast.net
Subject: RE: [NMusers] Weighting in NONMEM
Date: Thu, 20 Apr 2006 08:02:06 -0400

As Pascal suggests, “weighting” is accomplished in NONMEM via the variance model that
you code in your model.  This is the real attraction of maximum likelihood methods, rather
than choosing an arbitrary empirical weighting scheme, the variance parameters are fitted in
the regression and the weighting is then done via the theoretically “correct” method – the
inverse of the variance.  (Of course assumes that you have used an appropriate variance
model via the modeling process).

 
_______________________________________________________