Dear Fabrice,
> Using SPM 99b we tried to study correlations between an anatomic
> variable named A and the amount of activation during two conditions T
> and R (we have only two conditions and only one covariate)
> We have n subjects and we entered the scans as follows:
> T1 R1 T2 R2 ......Tn Rn
> the covariate was :
> A1 -A1 A2 -A2 ......An -An
> We computed two contrasts in the result section. The first was the
> activation T - R (weight 1 -1 0) and was named TR, the second was the
> positive correlation between A and (T - R) (weight 0 0 1) and was named
> corrA.
> We then tried to get the correlation coefficient and the graphic plot
> corresponding to each maximum displayed on the correlation map corrA
> using the Plot function and then the Fitted and adjusted response
> option. The question we have is about the next step which is to choose
> the appropriate contrast.
> In our case, we have two possibilities which are to either select TR or
> corrA.
> We tried both and obtained slightly different results. We don't know
> which one is the one we are looking for. We are guessing the
> appropriate solution is to select the file TR corresponding to the
> difference between T and R rather than the one corresponding to the
> correlation map corrA but we are not sure of it and would also like to
> understand better how the software works in both cases.
When plotting fitted and adjusted data you are given the option to
specify a contrast. This contrast specifies the variance components
you are interested in and treats the null space of the contrast as
confounds (confounds are removed from both the fitted and adjusted
data). In your case you might want to remove the average difference
between T and R to emphasize the subject to subject covariation in
activation with A (i.e. select contrast corrA). Generally however you
want simply to adjust for subject and global effects and would use the
F contrast 'all effects of interest'. On plotting this against
covariate A uou should see a positive correlation between the adjusted
activity and A. This correlation should be enhanced using corrA as the
contrast against which to adjust because the mean activation is treated
as a confound and is removed.
I hope this helps - Karl
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|