Yes, exactly. Another aspect is the onsets. If onsets coincide with the TR, that is trials always start at the beginnig of the TR, always after x ms, ..., then everything's fine.
But if not: With the default FIR approach you cannot properly model trial onsets occuring within a TR, instead the onsets seem to be shifted to the nearest full TR (or something similar). You would have to go with separate conditions (one for trials in which the onset occurs at the beginning of the TR, one for onsets in the middle of the TR, ...) or possibly increase the sample resolution (not sure about that right now). For longer trials/epochs a shift +/- 1 s should be rather negligable though, thus the FIR should still be a better predictor than an (arbitrarily) adjusted HRF.
|