Hi everyone,
I am reaching out due to a curious problem with DCM to seek any insights you might offer. Our analysis involves a comprehensive first-level full model that includes 10 regions of interest (ROIs). To streamline the model's complexity, we strategically reduced the connections from an the initial pool of 10x10 = 100 down to 57. (This reduction was based on the posterior probability (>0.95) obtained by inverting the A matrix in a preparatory step for the actual analysis.)
Consequently, in the single-subject full models, we set these 57 connections as active (value of 1), while the remaining connections were inactivated / pruned (value of 0) in our A and B matrices.
Following Zeidman et al. 2019, we then ran BMA+PEB on the inverted single subject model for both, the A and the B matrix. The results from the B matrix's PEB model appeared sensible and only included those 57 connections. However, the A matrix presented an unexpected outcome: 21 connections, which were previously pruned and set to 0 in the original single-subject DCMs, were part of the A matrix PEB model. I'm confused why PEB would have different parameters for A and B matrices because it is done on the same single-subject models. If I understand correctly, it should only calculate the Bayesian averages of those parameters.
I double checked the specifications etc. and I am not sure whether something has gone wrong. I would greatly appreciate any input, advice, or suggestions. Thank you very much in advance for your time.
Best,
Sabrina
|