Print

Print

Dear Sahil,

 

This seems to be a case where model inference doesn’t tell you anything interesting.

 

Perhaps you could then average over the model space (using Bayesian model averaging – BMA) and make inferences about the parameters instead.

 

For example, maybe the modulatory parameters are significantly non-zero over your group of subjects.

 

Maybe these or other parameters are significantly different between control and patient populations ?

 

Best wishes,

 

Will.

 

From: SPM (Statistical Parametric Mapping) [mailto:[log in to unmask]] On Behalf Of Sahil Bajaj
Sent: 07 May 2012 01:58
To: [log in to unmask]
Subject: Re: [SPM] DCM BMS result

 

Dear SPM/DCM experts,

 

I am facing the problem. In my case I have defined 7 models, out of which 4 models have same posterior probability of 0.14 and rest 3 have nearly 0.15 (for FFX) and in case of RFX, all models have model exceedance probabilities lying between 0.13 and 0.15. So none of the models seems to be clearly winning model. Please find the attached figures (FFX & RFX) for two different conditions/cases.

 

Any help would be greatly appreciated !!

 

Thanks a lot !!
Regards,

Sahil

On Sun, May 6, 2012 at 9:17 AM, Ilana Podlipsky <[log in to unmask]> wrote:

 

Hello,

We are analyzing data using DCM in SPM8. We have encountered a situation where the BMS output consists of more than 1 model (3-4 models out of 5 possible models) in Occam's window. The posterior probabilities of these models are not so high (~0.3). More so, sometimes 2 models receive the exact same probability. What is the winning model in such case? Is it likely to have such low probabilities? Also, in such a case, which model is the mEP calculated on?

Thank you,

Tamar

 



 

--