Hi Carmen,
The free energy, according to which models are scored, is the accuracy minus the complexity. In principle, additional driving inputs will be penalised for the complexity they add, and so will only be favoured in the BMS if they add to the accuracy over and above this. So you should be fine to have driving inputs coming into two locations, based on your comparison. You can also look at the parameter estimates in the C matrix - you may find one is estimated to be stronger than the other.
Best,
Peter.
-----Original Message-----
From: SPM (Statistical Parametric Mapping) [mailto:[log in to unmask]] On Behalf Of Carmen
Sent: 20 January 2015 07:48
To: [log in to unmask]
Subject: [SPM] DCM driving input
Dear SPM- and DCM-experts,
by means of DCM I am trying to explore how brain activity induced by experimental pain is influenced by medication. To do this, I defined a model including 4 regions influenced by the medication as evident from GLM and PPI analysis. The driving input consists of the experimental pain stimuli; modulator is the medication (experimental pain stimuli applied under influence of medication).
My aim is to compare modulation of intrinsic connections. However, I am not sure, whether only one or two regions receive driving pain input as both would make sense physiologically.
If I understood correctly, the driving input is needed to “inject activation” into the model system. Bayesian model comparison shows that models that include two regions receiving direct input perform consistently better than models where only one region is receiving driving input. However, I wonder whether this is because these models are more “realistic” or simply because models become better with increasing activation that is put into the system?
Thanks a lot,
carmen
|