Dear Gregor,
Log evidence can certainly be negative. Still a better model is the
one that has higher evidence (i.e. closer to +Inf). Note, however that
not all comparisons are meaningful, for instance comparisons with
different forward models are not (because of some implementation
details of the algorithm). You should look at the ID field in
D.inv{..}.inverse and only compare inversions with identical IDs.
Best,
Vladimir
On Mon, Nov 28, 2011 at 7:31 PM, Gregor Strobbe <[log in to unmask]> wrote:
> Hi everyone,
>
> I have a question regarding EEG source reconstruction in SPM8 with the
> imaging approach. I am doing reconstructions on ERP-data and to compare
> different models (e.g. forward and inversion models) I am comparing their
> log-evidences.
> Is it possible to get negative log-evidences for some reconstructions or is
> the reconstruction only meaningful if the log-evidence is positive? How
> should we interpret this?
> Can I compare also negative log-evidences or more specific the more negative
> the log-evidence the worse the model?
> thanks in advance
> --
>
> Gregor Strobbe
> Doctoral researcher
> IBiTech - Medisip (Ghent University)
> De Pintelaan 185, Blok B
> BE-9000 Gent (Belgium)
> Tel: +3293324322
> http://medisip.elis.ugent.be/
>
|