Dear Shoichi
I'm not sure what you mean by a "special model." A model with no intrinsic and no modulatory
connections would not show anything (actually assuming you have driving/input connections then the
only thing you might see are the driving connection parameters (i.e., C matrix). I don't think
there have been any such models used in the literature but I could be wrong.
What is frequently done, however, is to include a fully connected model that all possible intrinsic
and modulatory connections in the model comparison Such a model may be appropriate, for example, if
there really is no a priori information or theory about possible connections.
Darren
==============Original message text===============
On Tue, 12 Dec 2006 3:26:37 am CST James Rowe wrote:
Dear Shoichi,
> Some DCM works built several models in their papers, and decided which
> model was optimal based on their Bayes factors. However, they did not
> always compare one model to the special model such as DCM_control (no
> intrinsic connection, no modulatory effects). Is it necessary that the
> Bayes factor of the "optimal" model is larger than that of the spcecial
> model?
I look forward to Klaas' formal reply, but I suggest that the DCM model
comparison be used to compare plausible models, helping to select the
'optimal' model from a set of plausible models. If the DCM_control is
simply unrealistic, it cannot be said to be 'optimal' whatever the
nominal evidence in its favour. It could however force you to look more
closely at your other models and data in case you have built in problems
inadvertently.
best wishes,
James
===========End of original message text===========
--------------------------------
Darren R. Gitelman, M.D.
Department of Neurology
Northwestern Univ., 320 E. Superior St., Searle 11-470, Chicago, IL 60611
Voice: (312) 908-9023 Fax: (312) 908-8789
--------------------------------
|