Dear all,
I'm working on a post-hoc DCM analysis of a large network (~20 nodes). The estimated full models of all subjects were used as input of the 'spm_dcm_post_hoc' routine. However, the log-evidences are pooled in a fixed effects fashion as mentioned in the help document of this script. I wonder whether it is possible to perform RFX Post-hoc DCM or not, and how if possible.
Maybe I can save all the models from the final iteration of the post-hoc search for each subject, and then perform the RFX BMS using these reduced (or full) models. I don't know if this is a reasonable way. Any comments or suggestions would be appreciated.
I have another question, that is, which parameter should be reported in the paper to represent and test the significance of the connection. The posterior mean and variance could be found in the fields DCM.Ep and DCM.Vp of the 'DCM_BPA.mat'. For any connection survived from the pos-hoc search, should I just use the posterior mean and variance to test whether it is significant, or use the corresponding value in the field DCM.Ep of the 'DCM_opt_*.mat' file of each subject?
Thanks a lot.
Best regards,
Bingjiang
|