Dear Mark,
Due to the settings for microtime onset and microtime resolution a duration of 0 s will be modelled as a duration corresponding to one time bin. With the default settings 8 and 16 and a TR of 2 s one time bin corresponds to 2/12 s = 0.167 s. Thus it does not matter whether to specify the duration as 1 ms, 10 ms or 100 ms, as they all will be rounded to 0.125. Zero is treated in a special way, resulting in a different scaling of the predictor, which should be reflected by the scaling of the beta estimates. The results (T, p values) should be identical to e.g. 10 ms though (or at least very close, there might always be some rounding issues). The scaling issue is only relevant if you wanted to contrast conditions with durations of zero to non-zero durations.
> hemodynamic responses return to the baseline about several seconds after the cessation of a painful stimulation
Yes, in pain/stress/odor experiments it is likely that the measured BOLD response does not agree well with the predicted "canonical HRF plus convolution of some sort". For stimuli that build up or decease over time some groups try to model the plateau phase with an epoch regressor, but this is suboptimal, as it neglects the underlying time course. Instead of a simple on-off stimulus function ignoring the pattern at the beginning and post-stimulus it would be better to go with a stimulus function matrix that reflects the temporal course of the stimulus - if this is a good predictor for *BOLD response*. If it is not, which can well be the case, then it would be useful to turn to e.g. FIR models, which do not make any assumptions on the temporal pattern of the activation. Whether this is an option for your design is difficult to guess, in any case it would require several TRs between successive stimuli.
Best
Helmut
|