Hi,
I have a question relating to practical usage of F-contrast. It's very clear how I employ it with movements (manual example). However, consider the case that I have a simple functional localizer with blocks of faces and objects. I build a model with HRF derivatives (time and dispersion). So, at the end, I have 6 regressors (3 for faces and 3 for objects). Now,by using F-contrast for my derivative regressors I found some significant activations. How should I compute a contrast faces > objects? When I had a model without derivatives, I just made a t-contrast for two regressors. Now, the derrivatives are part of the HRF (in cotrnast to movemenets), so I have to take them in the contrast. But how build t-contrast with more than two regressors?
Thanks a lot,
John
On Tue, May 19, 2009 at 10:00 PM, Guillaume Flandin
<[log in to unmask]> wrote:
Dear Michael,
Yes, an F-test allows you to compare nested models and you can easily do
that in SPM. See slide 20:
http://www.fil.ion.ucl.ac.uk/spm/course/slides08/contrast08_fil.ppt
and Section 5.1, page 12:
http://www.fil.ion.ucl.ac.uk/spm/doc/books/hbf2/pdfs/Ch8.pdf
for a bit of theory and the SPM manual p.233 for a practical example:
http://www.fil.ion.ucl.ac.uk/spm/doc/manual.pdf
Note that SPM also allows to compare non-nested models with Bayesian
inference by comparing model evidence maps, see:
http://dx.doi.org/doi:10.1002/hbm.20327
This is available in SPM8 under:
SPM > Stats > Bayesian Model Selection > BMS: Maps
Best regards,
Guillaume.
Michael Froelich wrote:
> Hello,
>
> does anyone have an idea how I would go about comparing the model of a
> block-design fMRI dataset with a regressor to the same without regressor. In
> simple statistics one could do this with an F-test.
>
> Can SPM produce a contrast image comparing these two models?
>
> Michael
Guillaume Flandin, PhD
Wellcome Trust Centre for Neuroimaging
University College London
12 Queen Square
London WC1N 3BG
