Dear Philippe,
> In a post-hoc random analysis of PET data, I am interested to perform a
> "conjunction" between a contrast [A-B] performed in Group 1 and a
> contrast [C-D] performed in Group 2. So, I have two different contrasts
> in two different populations of normal subjects, assuming that [A-B]
> and [C-D] share in common a cognitive component of interest. All
> individuals contrasts were computed, and the con*.img were forwarded to
> the 2nd Ievel, then a one-sample t-test computed (usual procedure). I
> would be grateful for your comments on the following points :
>
> 1) Since [A-B] and [C-D] are computed in different subjects, a single
> contrast per subject is forwarded to the second level analysis, and the
> sphericity assumption is not violated. Is this correct ?
That is correct.
> 2) Is this true that a one-sample t-test on all con*.img (i.e., [A-B]
> and [C-D]) is equivalent to a conjunction analysis, since the aim of
> this 2nd level analysis is to highlight the areas where a common
> activation was detected in individual contrasts?
Not quite. First you are assuming homogeniety of variance over the
contrasts from the two groups, but I think this is very tenable. Your
one-sample T test simply tells you where the activation (averaged over
both groups in significant. To get the conjunction you would have to
enter the two groups contrasts as different effects at the 2nd level
and do a conjunction of contrasts testing for each alone. This is best
done (I think) by using 'simple regression' in 'basic designs' and
entering [1 ... 1, -1 ... -1] as the regressor. The conjunction of
contrasts [1 1] and [-1 1] should give you what you want (this gets around
the fact the SPM always includes a constant term in the design matrix).
> 3) Is there a caveat with the (non) orthogonalization between [A-B] and
> [C-D] ?
No.
I hope this helps - Karl
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|