Hello,
I conducted a DCM study on 14 subjects. Among 20 plausible models, BMS found one model to be the winning model (RFX analysis).
Individual posterior means of the B matric (modulatory effect on coupling, Ep.B) were used for correlation with behaviour. Also, the mean across the Ep.B's in all subjects was calculated to report mean coupling strengths and tested against 0 (student's T) to look for statistical significance.
Now I wonder why I get, at least, in part different results in BPA of the same winning model across the subjects.
For instance, while I found a positive significant coupling for one connection (mean Ep.B = 1.09, p<0.00001), BPA calculates -0.18 at a posterior probability of 0.95 for the same parameter.
Is it because I had to use RFX in the initial BMS (since patients) and BPA is only considered to work in the FFX framework. If so, would BMA the way to follow to report averaged posterior means instead? For this case, how can I perform BMA on one model across different suibjects?
Any help and advise in this issue is highly appreciated.
Thank you in advance,
Kind regards,
Robert
|