Dear Luke,
If I understand correctly what you describe, you create two models
that differ only in which input (A or B) is driving the system in
order to investigate how the connection strengths (DCM.A matrix) in
the system change as a function of this drive. This would be an
unusual application of DCM - and suboptimal, if your main interest is
to investigate context-dependent changes in connectivity. For that
purpose, you should use conditions A and B as modulatory inputs
(DCM.B matrix), and apply a second-test on the associated parameter
estimates. Have a look at some of the published DCM papers, they
almost unanimously use this approach.
Best wishes,
Klaas
At 21:00 09/07/2007, you wrote:
>Klaas:
>
>I sent the message below earlier, but you seemed to be away on
>holiday. I am a graduate student working with Dr. Barry Horwitz for
>the summer. Hopefully, you can provide a bit of insight to the
>situation I describe below. Thanks so much again!!
>
>I have a simple passive-viewing block design study with three
>different image categories (A, B, and C-control stimuli) displayed
>in a pseudorandom order within each run across six different runs
>for each subject. I have a patient group and a control group. I want
>to test the hypothesis that for the A > B contrast, the patient
>group shows stronger activation within a given circuit than the
>controls. The circuit consists of 3 regions for each hemisphere
>(total # of regions = 6), which showed group differences in our
>ROI-based random effects analysis. For each subject, I would like to
>create a DCM with stimulus category A entering the model at one of
>the 3 regions to determine the path values among all the connections
>in the DCM (in the A matrix in DCM, I assume). I would then do the
>same thing having stimulus category B entering the model at the same
>region for the same DCM within the same subject. Using a standard
>2X2 mixed ANOVA (within-subject factor: path values for the regions
>in response to 1) stimulus A and 2) stimulus B for each subject;
>between-group factor: patients vs. controls), I would test whether
>this circuit showed stronger activation in response to stimulus
>category A vs. B in patients compared to controls. Does this seem
>like a reasonable approach. I have already tested various models in
>an individual subject and used the Bayesian model comparison
>approach to determine the best model (which was my "working
>hypothesis" model from the beginning, so that is nice:). At this
>point, given my design and the questions I am hoping to answer, I am
>planning on running a connectivity analysis using SEM, but if DCM
>has any obvious advantages I have overlooked that are specific to my
>design and questions, I would appreciate your insight on that point
>as well. Any input would be appreciated....thanks in advance for
>your time and your feedback.
>
>Luke
|