Dear Drew,
the idea is to cover the time period in which you still expect some deviation from baseline, at least in theory, which you would like to control for. For short stimuli this results in a length similar to the canonical HRF, so something like 25 - 30 s. If you use longer events, this might take (much) longer. In that case you could look at the corresponding boxcar functions convolved with the canonical HRF to get an impression of the expected "default" time course. Depending on the outcome you might then go with a longer time period for the FIR models as well.
Hope this helps a little,
Helmut
|