NONMEM Users Network Archive

Hosted by Cognigen

RE: Block versus diagonal omega

From: Elassaiss - Schaap, J. - <jeroen.elassaiss>
Date: Sun, 29 Aug 2010 21:47:07 +0200

Mark, Douglas,
The Akaike Information Criterion is more general, so is (should be?) =
applicable to $OMEGA changes as well.
The addition of off-diagonal elements may be treated more liberal as far =
as I am concerned. I personally focus more on the simulation properties =
than how much the ofv drops with them.
And Douglas, see =
fig 3 for the effects of shrinkage on those scatterplots: shrinkage may =
also hide correlations. And as a consequence, I rather try a full matrix =
regardless of scatterplots, retaining of course only those elements that =
are significant or have a large impact on the matrix. The maximum =
likelihood is not affected, so I completely agree with you: one indeed =
can trust the ofv drop / parameter estimate wrt to shrinkage.
Best regards,

Modeling & Simulation Expert
Pharmacokinetics, Pharmacodynamics & Pharmacometrics (P3) - DMPK
PO Box 20 - AP1112
5340 BH Oss
The Netherlands
T: +31 (0)412 66 9320
F: +31 (0)412 66 2506


From: owner-nmusers
On Behalf Of Mark Sale - Next Level Solutions
Sent: Friday, 27 August, 2010 12:21
To: Eleveld,DJ
Cc: nmusers
Subject: RE: [NMusers] Block versus diagonal omega

  But how large a drop? As I understand it, adding elements to OMEGA =
(diagonal or off diagonal) do not follow a chi-square distribtion, and =
therefore there is not any basis for determining how large a drop is =


        -------- Original Message --------
        Subject: RE: [NMusers] Block versus diagonal omega
        From: "Eleveld, DJ" <d.j.eleveld
        Date: Fri, August 27, 2010 4:52 am
        To: "Elassaiss - Schaap, J. (Jeroen)" <jeroen.elassaiss
        "yhb5442387" <yhb5442387
        Hi Jeroen,
        If shrinkage induces correlations (which arent "true") in the posthoc =
ETAs then the data isnt very informative for at least 1 of the =
parameters. If this (misleading) correlation causes the researcher to =
test a model with off-diagonal covariance, I would expect that they =
would not find a significant drop in objective function, and therefore =
they would reject the correlation from the model. So, in the end, no =
harm done (to the model). My thinking is that if the data is not =
informative about the value of some parameter, then it probably wont be =
informative about the relationship between that paramater with some =
other parameter.
        The concerns about how to handle shrinkage properly simply disappear if =
you treat the off-diagonal elements like any other parameter, i.e. you =
require some drop in objective function when you accept a parameter into =
the model.
        Best regards,
        Douglas Eleveld



This message and any attachments are solely for the intended recipient. =
If you are not the intended recipient, disclosure, copying, use or =
distribution of the information included in this message is prohibited =
--- Please immediately and permanently delete.
Received on Sun Aug 29 2010 - 15:47:07 EDT

The NONMEM Users Network is maintained by ICON plc. Requests to subscribe to the network should be sent to:

Once subscribed, you may contribute to the discussion by emailing: