Thanks for the quick, informative reply. I'm using DCM8, and probably not the newest version of SPM8 either (ie not 4010).
I'll look at the example you cite from the seminal DCM paper.
Stephen J. Fromm, PhD
From: Mohamed Seghier [[log in to unmask]]
Sent: Friday, February 25, 2011 11:44 AM
To: Fromm, Stephen (NIH/NIMH) [C]
Cc: [log in to unmask]
Subject: Re: [SPM] DCM: significance and graph connectivity
Are you using DCM8 or DCM10?
If your connection R1==>R2 is not significant at the group level, then
it is possible that (1) your modulatory parameter MI1 is significant,
which may suggest that the inter-regional interaction between R1 and R2
increased "specifically" during your context MI1; or (2) you have a
large inter-subject variability on this connection. For a similar
situation, see Figure 24 in Karl's 2003 paper (Friston 2003 p1298:
connection V1-to-V5 not significant but motion modulation was
I hope this helps,
On 25/02/2011 16:23, Stephen J. Fromm wrote:
> I'm working on a DCM project. So far the best fitting model looks like this:
> DI--->R1 ==> R2 ==> R3
> ^ ^
> | |
> MI1 MI2
> That is:
> * there are three regions, with intrinsic connections from R1 to R2 and R2 to R3
> * there is one driving input DI acting on R1
> * there are two modulatory connections MI1 and MI2 acting on the two between-regions intrinsic connections
> (The experimental variables for MI1 and MI2 are identical.)
> At the group level, the second connection (R2 ==> R3) is significant, but the first (R1 ==> R2) isn't. Conceptually, this doesn't make sense, insofar as the only way the system perturbations introduced by the DI can get to (R2 ==> R3) is through (R1 ==> R2).
> Does this reduce the credibility of the model? Or is it alone not enough to do that because of the possible vagaries of what is significant?