TWIMConcern: when trying to remove the effects of membership in a dichotomous group using SPM, is it better to model membership as one confound or two? that is, if half my subjects belong to one group, half to another, and I am testing for the effect of time or stimulus or whatever, I could enter one confound or two. Each scan, or each pair of scans for each subject, could then be indexed as subject condition global_mean grpA --where the entry for grpA would be 0 or 1 or is it better to include BOTH groups, as in subject condition global_mean grpA grpB where each scan is either [0 1] or [1 0] for group. I realize that the latter eats up an additional degree of freedom, but I am not clear which is a preferable model. Thanks. %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%