Dear Mike,
To properly separate between subcomponents of a trial one would usually go with jitter between the subcomponents and/or half/partial trials. The jitter is meant to make the two time courses more dissimilar. Partial trials would consist of e.g. the two tactile stimuli only, with no decision required, thus providing an "unbiased" estimate for the first subcomponent.
Note that this only works properly in theory though, while one could think of confounds in practice. With regard to partial trials, subjects might always think about the decision (as there's nothing else to do, as you think about it automatically maybe, ...). It's probably less likely with a high percentage of partial trials, or in case the task were a very difficult one (as subjects might prefer to wait and see then). Partial trials also assume subjects not to respond on the "missing" part, but they might actually react in some manner ("did I miss the instruction"). With regard to jitter, technically you could insert an interval between stimulus offset and decision period onset, but it would require subjects to do "nothing" within the interval, which might again be unrealistic, as mentioned above. One could also think of an interaction between interval and decision activation, with subjects (starting) thinking about the decision well before the "official" decision period - with short intervals subjects might not yet have come to a conclusion, with longer intervals they might have decided and just wait for the "Go" signal, with much longer intervals they might shift attention / fall asleep, in any case, the activation after the decision onset might be a different one depending on their "progress" with the task on that particular trial, which might correlate with time.
Turning back to your design, and agreeing with Colin, don't try to model the stimulus period.
BTW, with the latest SPM versions it shouldn't make a difference whether to go with 0 or 0.05 s if you really wanted to go with a regressor like that, as both durations should result in an "on" periods with lengths of two time bins (2 * repetition time/microtime resolution, e.g. with a TR of 2 s and the default resolution of 16 the effective "on" period used for the predictors would be 250 ms). There always has to be *some* "on" period for the convolution to work, so at least one time bin is required, and due to some internal settings you get an addition time bin on top. In previous versions there were some differences still between "0 s durations" and "very-short non-zero duration", as the "0 s" were treated in a special manner, resulting in a differently scaled predictor (although the time course would have been identical to a 0.05 s regressor).
Best
Helmut
|