Dear Mike,
The shared variance for the TS and the WM regressors can be expected to be very high, which is why I don't think it is useful to separate those two components in the given design.
The scaling of the amplitude due to durations of 20 ms vs. 6 s is another issue, although a relevant one as well. Note that a certain stimulus duration does not ensure that the "on" periods of the neural processes correspond to that duration. As an arbitrary example, a stimulus of 20 ms duration might require 100 ms to be processed in a certain region, and the same might hold for a 40 ms stimulus (in that context, be aware of the fact that with the default microtime settings the entered 20 ms will actually be treated as repetition time in seconds / 16, so the effective duration is not 20 ms). However, a predictor based on a 40 ms duration would have a different amplitude of a 20 ms duration (at least if microtime resolution is increased accordingly, otherwise they might both end up with repetition time in seconds / 16 of course). In any case, depending on chosen effective durations the differences between TS and WM regressors would vary. As long as you don't have any a-priori information on the processes involved the comparison would be completely arbitrary even if results look very encouraging.
If you were interested in TS activation specifically you could go with a separate run in which stimuli are presented with no further requirements for the subjects. Your control condition might serve the purpose reasonably well though, as you seem to be able to ensure that subjects processed the stimulus in a certain manner. Going with the two 6 s conditions in GLM-2, looking at the control condition should reflect activations due to the stimulus/basic processing, looking at the experimental condition would be stimulus + WM, contrasting these two should reflect the WM component. In case you want to rely on the control condition as a localizer for ROIs for the comparison E vs. C it might be useful to go with a separate run though to rule out selection biases. If this is not possible, in case of a-priori hypotheses you could still easily define the ROIs via some structural labels or functional ones (e.g. from one of the available connectivity parcellation schemes), you could present contrast C for confirmatory purpose ("look, tactile stimulus results in activations in corresponding areas"), and then focus on E vs. C.
Best regards
Helmut
|