From:"GORDI, TOUFIGH [R&D/0437]" 
Subject:[NMusers] A simple question
Date:Tue, 18 Jun 2002 02:29:57 -0400


I have noticed that a common practice to validate a model is to apply the
model based on a previous data set to a new set of data, fixing all model
parameters and see how the model describes the new data. This is specially
done when the model does not estimate the parameters properly on the new
data set.

Could someone explain for me how it is possible that the model does not
generate good estimates from a data set but still can describe the same data
set fairly well when its parameters are fixed based on previous findings? To
me it sounds as that there are several equally well estimations of model
parameters, choice of the "correct estimates" being dependent on how
reliable the estimates are in relation to known physiological values.

Thank you.

Toufigh Gordi,


From:"LSEN (Lars Erichsen)" 
Subject: RE: [NMusers] A simple question
Date:Tue, 18 Jun 2002 10:23:40 +0200

Hi Toufigh,

The situation you describe may occur when the new data do not contain as
much information
as the previous data on which parameter estimates are based. With the new
data only you may not be able to 
estimate all the model parameters because they lack some information, e.g.
if new data are from a smaller dose range compared to previous data. 


Lars Erichsen