Dear Mikkel,
I think the text in the manual is outdated and the problem has been
fixed since. See http://www.ncbi.nlm.nih.gov/pubmed/22037420 . So it
should be OK now to use F-tests and the result should be very close to
doing two t-tests for positive and negative version of the contrast
with alpha divided by 2 and combining the results. I'm CCing Ged who
is a better expert on this issue in case and might want to add
something.
Best,
Vladimir
On Wed, Jun 20, 2012 at 11:10 AM, Mikkel C. Vinding <[log in to unmask]> wrote:
> Hi
>
> I have a question regarding the use of F- or T-contrasts when performing
> statistical tests on source reconstructions of EEG/MEG data (as far as I can
> see, this haven't been posted on the mailing-list before) .
>
> I am comparing the reconstruction of a target condition against a baseline
> condition using a paired t-test. I have no a priori hypothesis regarding the
> direction of difference (two-tailed hypothesis) so naturally I would expect
> to use a F-contrast . However the SPM manual states:
> "Note that the images exported from the source reconstruction are a little
> peculiar because of smoothing from a 2D cortical sheet into 3D volume. SPM
> statistical machinery has been optimized to deal with these peculiarities
> and get sensible results. This presently only works for T-contrasts but not
> for F"
> My question is what exactly the consequences of using T- or F-contrast in
> the analysis of inverted data? I have tried using both contrasts and they
> give (almost) the same results, so I am a bit confused about how and which
> assumptions I am violating? Is it possible to use the T-contrast in a way
> that is compatible with a two-tailed hypothesis?
> The source reconstruction I use in inverted from EEG obtained from 22
> subjects and inverted using group inversion.
>
> Thanks
>
> --
> Mikkel C. Vinding
> Email: [log in to unmask]
>
> Cognitive Neuroscience Research Unit (CNRU)
> Center for functionally Integrative Neuroscience (CFIN)
> Aarhus University
|