Dear Robert,
Please could you send me your BMS.mat relating to question 1, so that I can see what you mean? Please zip the .mat file, or it'll be removed by my virus checker.
Thanks,
P
-----Original Message-----
From: SPM (Statistical Parametric Mapping) [mailto:[log in to unmask]] On Behalf Of Robert Schulz
Sent: 03 October 2015 09:42
To: [log in to unmask]
Subject: [SPM] Odd BMA results in DCM12 SPM v6470
Dear all,
I have realized a problem with BMA in DCM12 (SPM12, v6470).
I conducted a DCM study, my model space comprised 36 models (varying B matrices) x 8 families (varying C matrices) = 288 models for each subject.
Applying RFX BMS I found that no family showed superiority compared to the other. Hence, I used RFX BMA over "all families" to calculate subject specific posterior means "mEPs" over all models in the OCCAMS window for each subject.
I realized the following problems:
1. Some of my subjects had only one model in the occams window. However, in the BMA_mEPs I have only an empty B matrix (all zeros) for that subject. In contrast, the model specific Ep.B (the model in the occams window) showed reasonable connectivity values. This doesn't make sense to me.
2. Some of the subjects had two models in the occams window, for instance model 1 with 4 and model 2 with 8 non-zero connectivity values in the B matrix. After BMA I realized that the posterior means mEPs.B only showed 2 non-zero connectivity values. Also in this case I wonder whether there would be an error during BMA. As I understand BMA, the Ep.B of the different models in the occams window should be averaged, weighthed by the model probabilities. In my example I wonder whether this should give me a BMA mEPs of 8 non-zero connectivity values (the higher number of connections in model 2).
Any help on this issue would be highly appreciated.
Thank you very much,
Best, Robert
|