Hi Andreas,
Thanks for your answer!
If we understand correctly, we could use cope images as inputs for the higher level analyses, define EVs for each cope and subject, and then define the f-tests (see attached for a simple example with 3 subjects). Is this correct?
We have one more question, though: We would also like to include a covariate (a personality characteristic) and test the interaction between stimulus type and the covariate. Would it be possible to add a covariate to this model as well?
Thanks,
Esther and Renske
-----Original Message-----
From: FSL - FMRIB's Software Library [mailto:[log in to unmask]] On Behalf Of Andreas Bartsch
Sent: donderdag 30 juli 2015 21:03
To: [log in to unmask]
Subject: Re: [FSL] F-test
Hi,
no - you can't pass F-contrasts to higher levels directly. Passing VARCOPEs up to a group level is based on t-test contrasts - the group stats framework cannot deal with passing up of f-test contrasts as not all of the assumptions would stand.
However, you use the COPEs of the lower level F-test (i.e. the COPEs that go into the lower lvel F-test) and then do an F-test at the higher level.
You may also consider using a type of summary stats of your COPEs of interest, i.e. the root-mean-squared (RMS) of the COPEs, and pass them up to the higher level using randomise.
Cheers,
Andreas
Am 30.07.15 16:19 schrieb "FSL - FMRIB's Software Library on behalf of Esther Heckendorf" unter <[log in to unmask] on behalf of
[log in to unmask]>:
>Hi,
>
>We have a small question about using an F-test in FEAT. We've shown
>participants three different types of stimuli (S1, S2 and S3) and would
>like to test the main effect of stimulus type. We have no specific
>hypotheses regarding the direction of effects. In the lower level
>analysis we have included contrasts comparing each pair of stimuli
>(i.e.,
>S1 vs S2, S1 vs S3, and S2 vs S3), as well as an F-test across two of
>these contrasts (S1 vs S2 and S2 vs S3) to test the 'entire' main
>effect in one test. When setting up a higher level analysis to test the
>significance of the effects in the entire group of subjects we could
>only include the contrasts comparing each pair, but not the F-test.
>That's of course a bit of a problem, as we would like to check the
>significance of the 'entire' main effect before performing individual
>comparisons. Is it possible to include the F-test in the higher level analysis somehow?
>
>Thanks!
>Esther
|