Dear Qunjun
If the full model has poor explained variance, this will not get higher for the reduced models. The reduced models will only decrease in complexity, not increase in accuracy.
I suggest you choose a favourite subject who shows large within-subject effects in the SPM analysis. Make sure the VOIs are located correctly for this subject, and see if you can build a model with higher explained variance. Then return to the group analysis.
Let us know if you need further advice.
Best
Peter
-----Original Message-----
From: SPM (Statistical Parametric Mapping) <[log in to unmask]> On Behalf Of Qunjun Liang
Sent: 25 August 2021 05:00
To: [log in to unmask]
Subject: [SPM] Auto-search method for optimizing DCM
Hi all,
I am using DCM to model the effective connectivity of my task fMRI data with 4 VOIs.
I set a full model following the previous theory. However, the explained variance in subject level DCM is out of tolerance (all less than 10%). I learnt that the automatic search in BMA may help to optimize the model that the redundant connectivity would be switch off. Thus, I want to know the decency to use the reduced model after auto search as the full model and rerun the DCM. If it would receive some criticism about its data-driven tendency?
Thanks,
Qunjun
|