in the contrast you show in the matrix you would ask for voxels in which
the correlation is more positive in the first group than the second.
Seems alright to me.
Note that "more positive" cannot be distinguished from "less negative" of
course - however, the individual contrasts you performed before or contrast
plots from the combined model will answer this question.
Hope this helps...
(Only minor point: you need to decide if for the other covariates also
group-dependent correlations may exist,
e. g. age. So maybe setting up a second model in which all covariates are
modelled separately for the groups would be more conservative - the
contrast of interest stays the same. Ideally, differences to the first
model should not be too strong.)
Max Planck Institute of Psychiatry
NMR Research Group
Mail: [log in to unmask]