Hello
I've been looking at DCM for resting state fMRI, and trying to understand the second level results. I've run a DCM model on about 100 subjects with 4 regions, all connections on as per the SPM manual.
Then code along these lines (adapted from https://en.wikibooks.org/wiki/SPM/Parametric_Empirical_Bayes_(PEB) )
M.Q = 'all';
M.X = ones(nsubjects,1);
M.X(:,2) = [1 1 1 1 1 ... 0 0 0 0]; % group membership. Approx 50/50 patient/normal
GCM = { list of first level DCMs }
PEB = spm_dcm_peb(GCM,M,{'A'});
BMA = spm_dcm_peb_bmc(PEB);
spm_dcm_peb_review(BMA,GCM);
If I select the group variable in the review panel, I get a pleasing number of PEB parameters with 'Very Strong evidence' which I presume means that there is good evidence those connections are stronger/weaker in the patient group.
However, I also ran the same analysis, but with random assignment
M.X(:,2) = (rand(length(M.X),1)>0.5);
And I get several PEB group parameters with very strong evidence. [for the example I just ran, and 5 for 'free energy' two of which also pass threshold for 'probability > 0'
So I don't know how much faith to put in my findings with the actual group membership.
Am I doing something wrong? Or is my interpretation wrong?
Michael
|