Dear FSL experts and users,
I have a simple design with 2 conditions A and B for each subject, and I'd
like to use randomise to get the group statistical maps. I have a question -
sorry if it's trivial, I couldn't find the answer in the archives -
regarding how to best set up my analysis.
At the first level (within subject), I have set up 3 contrasts: A>baseline,
B>baseline, and A>B. I am ultimately interested in the A>B contrast at the
group level, and was wondering what would be the best way to set up my
randomise analysis at the 2nd level (between subjects):
- set up a paired t-test using first level outputs for A>bl and B>bl,
entering the subjects in the same order in the 4D data file, and using the
following command:
randomise -i <4Dinput> -o <output> -d <matrix> -t <contrasts> -m <mask> -C 0.5
OR
- use a simple t-test to compute the group average for the A>B first level
copes, using the following command:
randomise -i <4Dinput> -o <output> -1 -v -5 -m <mask> -C 0.5
I also had a few additional questions:
- the manual example doesn't specify a mask for the simple t-test: is there
a specific reason for this, or should one use a mask in all cases?
- does the spatial smoothing of variance relate in any way to the spatial
smoothing used in the pre-stats step in FEAT, i.e. should I set it to a
specific value here depending on what I used in the first level analysis (7mm)?
- the randomise manual advises to use variance smoothing if one has less
than 20 subjects for a simple t-test. Should one use it for other tests as
well (e.g. paired t-test) when having < 20 subjects?
Thanks again for the great support on the list!
Best wishes,
Stephane
|