Dear Josh,
The F numbers can be positive or negative.
F is an approximation to the log of the model evidence, log p(y|m), with data y and model m where p(y|m) is a probability *density*. p(y|m) can be smaller or bigger than 1. Hence log p(y|m) can be negative or positive.
What's important is the relative values of F for different models (within subject). We can then look for consistency in the differences, over subject.
The Bayes factor for model 1 versus 2 is p(y|m_1)/p(y|m_2).
The log of this is approximated by F_1 - F_2 with (for the case of two models) a difference of 3 corresponding to a posterior probability in favour of model 1 of 95%.
Best, Will
> -----Original Message-----
> From: SPM (Statistical Parametric Mapping) [mailto:[log in to unmask]]
> On Behalf Of Josh
> Sent: 01 March 2012 11:33
> To: [log in to unmask]
> Subject: [SPM] BMS F matrix (log evidences for all subjects and all
> models)
>
> Dear SPMers,
>
> Are the values in the BMS.DCM.ffx.F matrix supposed to negative?
>
> What does it mean if they are?
>
> Many thanks,
>
> Josh
|