Greetings,
This is a question that seems to come up often. Users need to be very CAUTIOUS when scaling regressors after waver.
Recall that regressors constitute models of what you expect to have in your FMRI timeseries in response to a succession of stimuli.
Assume that for one subject (S1) the stimuli were more bunched up in time, say, than for subject 2 (S2). As a result the FMRI response Y1 for S1 will reach higher values than Y2 for S2 (same reasoning as Giorgio's here). This stimulus-timing-induced difference in response amplitudes between S1 and S2 is normally reflected in the regressors R1 and R2 created for these two subjects, respectively. In this case, assuming S1 and S2 processed the stimuli in the same way, then the coefficients b1 and b2 from the linear regression would be the same. However, if you scaled R1 and R2 to have the same peak, then b1 will be higher than b2. This would then be erroneously interpreted as 'subject 1 responded more to the task than subject 2'. However, this difference in b is merely reflecting a difference in stimulus delivery timing and nothing else.
The moral of the story is that one should rarely scale the regressors AFTER they come out of waver. I say rarely, because there are always exceptions lurking in someone's unorthodox designs.
cheers,
-ziad