Dear Jewgeni Kegelses, I would use the posterior probability here, g_post. The situation here is analgous to the expression of individual preferences within a group. F is akin to ones preference before knowing what the group think, and g_post afterwards. eg. before finding out what everyone else thinks, you might tentatively go for option A. But upon finding that 90% of others have gone for option B you may switch. The posterior probabilitiy embodies this - weighing up both the strength of evidence observed individually and the proportion of others in disagreement. Best, Will. ________________________________________ From: SPM (Statistical Parametric Mapping) <[log in to unmask]> on behalf of Jewgeni Kegeles <[log in to unmask]> Sent: 23 September 2013 16:02 To: [log in to unmask] Subject: [SPM] Question regarding BMS: BMS.DCM.rfx.F vs. BMS.DCM.rfx.model.g_post Dear DCM Experts, after performing BMS I'm going to look at the individual results (which model won individually for each subject). In the BMS.mat file there are 2 matricies comprising individual model evidences: (1) BMS.DCM.rfx.F (2) BMS.DCM.rfx.model.g_post Since the model ranking between (1) and (2) is not always the same (e.g. according to g_post the individual winner model for Subject 1 is the model #2, but the minimal value in BMS.DCM.rfx.F for this subject got the model #3). So I'm wondering what should be used to identify individual winner model: BMS.DCM.rfx.F or BMS.DCM.rfx.model.g_post? Any kind of help would be greatly appreciated!