I've been curious about this as well. It certainly seems to be an oversight in the switchover from SPM99 to SPM2 (and perpetuated through SPM5 and SPM8). SPM99 first combines task and user regressors then detrends them (lines 315 and 361 of spm_fMRI_design.m for SPM99). SPM2 does the detrending when merging task and user regressors, but only detrends the user regressors (i.e. line 309 of spm_fMRI_design.m for SPM2)
Here, it'll make more sense if I grab the lines:
SPM99
315: X = [X D(:,i)];
361: Xx(etc.) = spm_detrend(X)
SPM2/5/8
309: X = [X spm_detrend(C)]
It certainly looks like an oversight (see http://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;Z42AyQ;20040224173212%2B0000;ind04), easy to fix, but a pain to remember to do every time one upgrades SPM. The danger would seem to be in relying on the design orthogonality check as a measure of the collinearity between conditions, as in SPM2+ this could lead one astray. Very astray according to this old message: https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;ZeV13w;20040303210036%2B0000;ind04)
|