Dear Ge,
> We are analyzing data from a bloc design fMRI study (2 condtitions A,B and baseline C). We had 13 subjects. We showed the stimuli as [C A B C
> A B C A B] for 7 subjects and [C B A C B A C B A] for the other 6 subjects.
> Contrasts at the first level analysis were computed as follows:
> Con_0003.img was computed as A-C for both subject specification [-1 1 0] for 7 subjects and [-1 0 1] for the other 6 subjects.
> Con_0004.img was computed as B-C for both subject specification [-1 0 1] for 7 subjects and [-1 1 0] for the other 6 subjects.
> Con_0005.img was computed as A-B for both subject specification [ 0 1 -1] for 7 subjects and [0 -1 1] for the other 6 subjects.
>
> We are interested in the contrasts: (A-C)>(B-C)
> So we do the contrasts at the second level analysis as follows:
> 1. We select 13 Con_0003.img files of all the subjects and then 13 Con_0004.img files for all the subjects.
> 2. We specify the "group? " input as 1 1 1 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 2 2 2 2
>
> Our question is to compute the (A-C)>(B-C) what should we contrast specify?
> Is it [ 1 -1 0] ?
>
First regarding your first level contrast. The question (A-C)>(B-C) as probed by the contrast (A-C)-(B-C) actually reduces to A-B, i.e. the
parameter estimates for the baseline doesn't enter into it. So, the first level contrasts should be [0 1 -1] for the group of seven and [0 -1 1]
for the group of six.
This leaves you with thirteen con_* images that you enter into a one-sample t-test and use the contrast [1]. See, much simpler than you thought.
Good luck Jesper
|