Print

Print
Dear colleague,

The subject-specific log model evidences are the input to the random effects BMS procedure, not its output.  In other words, if you use the random effects BMS method, there is no need to look at any differences in log model evidence.  The results that you are interested in are the expected posterior model probabilities (exp_r) and the exceedance probability of all models (xp).  Both of them will give an identical ranking of models and tell you which model is the best. 

Best wishes,
Klaas



Von: Maz <[log in to unmask]>
An: [log in to unmask]
Gesendet: Mittwoch, den 8. Juli 2009, 06:11:23 Uhr
Betreff: [SPM] BMS DCM SPM8

Dear SPMers,

Really needs help regarding BMS in SPM8. After comparing 6 subject/ 4 models
each, and got all the BMS.DCM.rfx value (F, SF, alpha, exp_r and xp). How
should I conclude the result? In Stephan (2009) paper, I should calculate
log model evidence differences, this is the F value right? For example, the
values for subject 1, model 1 is -505.75 and model 2 is -510.76. There for
log model evidence differences (only the magnitude?) = 510.76 - 505.76 = 5.
Favour model 2. Is this the correct way? Is this method can only consider 2
models (Best and 2nd Best Models)?

Other than that I also need to consider the exceedance probability value.

Thank you in advanced.