Dear Russ
> Hi. We have an fmri dataset in which data were collected for three
> conditions (B=baseline,X=condition X, Y=condition Y) in the same scan,
> during both memory encoding and memory retrieval. let's call the
> encoding conditions eB, eX, and eY, and the recognition conditions rB,
> rX, and rY. What we'd like to do is a conjuction analysis in which we
> look for regions that were active for a certain condition (X or Y)
> across encoding and recognition, and another where we look for regions
> that are active for either encoding or recognition across both
> conditions (X and Y). this is in the context of a random effects
> analysis, for which I've created adjusted mean images of each of the 6
> conditions for each subject.
>
> Thus the conjunctions that we'd like to do are: common encoding: (eX -
> eB) && (eY - eB) common recognition: (rX - rB) && (rY - rB)
>
> common X: (eX - eB) && (rX - rB) common Y: (eY - eB) && (rY - rB)
>
> The question that I have (ignoring for the moment the other recently
> discussed issues about conjuctions and random effects) is: How
> inappropriate is the first set of conjuctions, since the baseline is
> common across the two contrasts and the contrasts are thus not
> independent?
In SPM96 this issue is dealt with in the sense that colinearity of the
contrasts specifying the conjunction enter into the statistics.
Therefore it is not inappropriate but reduces to simply identifying
voxels that show an effect of (eX + eY - 2eB) and do not show
significant (eX - eY).
In SPM98, using recent advances from Keith Worsley, the contrasts will
be orthogonalized before the conputation. The interpretation is then
in relation to the orthogonalized contrasts.
I hope this helps - Karl
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|