Dear all,
since I did not receive any reply to my question, I decided to refresh it.
Dear all,
I would like to search for the best model (at the moment I’m interested in the parameters of the A matrix) in a group of 21 subjects. I’m using DCM for CSD with EEG data.
I briefly resume the steps I made since now:
- I specified a fully connected DCM model for each subject ( only forward connections ).
- Following the instruction provided at SPM/Parametric Empirical Bayes (PEB) - Wikibooks, open books for an open world I specified only the mean-group connectivity regressor in the design matrix
- I run spm_dcm_peb(GCM,M,field), where field is ‘A{1}’, and subsequently I run spm_dcm_peb_bmc(PEB) in order to compare all the possible reduced models nested with the full PEB model.
- I pruned away all the group-level parameters having probability<0.95
My questions are:
(1)I read in the documentation of spm_dcm_peb_bmc that "NB for EEG models the absence of a connection means it is equal to its prior mean, not that is is zero.”.
What does this mean? Can I use PEB for a post-hoc search using EEG models? I am a bit confused..
(2) Since we don't know anything about the hierarchical organization between our sources, how can I search for F and B connections at the group level? Should I specify a full model with A{1} = ones(N,N) and A{2} = ones(N,N). I need some help to clarify this point.
Thanks in advance,
Gianluca Rho.
|