 |
 |
Dear Tali,
1. We have a group study, in which each individual has 4 sessions, and therefore 4 DCMs - (all with the same structure: fully connected), one for each session. What would be the correct way to use the post_hoc comparison? Should this be done in 2 steps: one optimization among all sessions within individual, and then taking the DCM_opt from each individual into a second optimization step across individuals?
If you are assuming the model is a fixed effects over subjects, then there is no need to distinguish between DCMs of the same subject and DCMs from different
subjects. In other words, simply enter all the DCMs into a single (multi-DCM)
post hoc optimisation. Alternatively, if you think the model is a random effect over subjects, then a two-step procedure might be more appropriate. In this case, you could use BPA to reduce the 4 DCMs for each subject into one summary DCM. However, I do
not think the post hoc routines entertain RFX pooling - and I am not sure the free energy (log evidence approximation) will be evaluated for the BPA – so I would not pursue this unless you knew what you were doing (better than I know!).
2. Assuming that the optimized model can include a bilinear effect (B) for which the intrinsic connection (A) has been reduced - (This situation is impossible in the standard way of specifying models in DCM) - what would be the interpretation of the bilinear effect in such case?
It is perfectly possibly for a B (bilinear) effect to be present in the absence of an A (average) connection strength. This is because (in DCM for fMRI) bilinear
effects are additive. This means your experimental effect is causing increases and decreases in effective connectivity, around an average of zero.
With very best wishes,
Karl
From: Tali Bitan [mailto:[log in to unmask]]
Sent: 21 February 2013 21:16
To: Friston, Karl
Subject: Group level -post hoc BMS in DCM
Dear Prof. Karl Friston