 |
 |
Dear Rob, For very brief stimuli the time to peak is something like 5 s, the undershot starts at about 10 s and baseline is reached again after ~ 20 s. However, intervals like 5 - 7 s are indeed reasonable, as the temporal overlap occurs where BOLD response is already decaying / still starting to rise. For shorter intervals it's not just that the overlap becomes more evident, but also that activations from subsequent trials (or components) do no longer add up linearly (which would conflict with assumptions when constructing the predictors). There have been several early studies on the issue like that by Dale & Buckner (1997), but usually they rely on visual stimuli / activations in early sensory cortices, which might not necessarily transform to higher-order regions with more diffuse and variable activations. The shortest interval might have been the 500 ms employed in Buckner et al. (1998, Neuroreport), although this was also combined with a TR of 1 s and the interval between trials of the same type (stimulation left / right / full visual field) was longer. However, mean ITIs with something like 2 s are common in fast event-related designs, it seems activations still add up almost linearly then. Given your design I would definitely implement an interval between outcome and cue which is noticeably longer than the interval within the trial, so that the temporal structure of a trial remains evident for the subject. Besides, as subjects probably have to concentrate during the interval within the trial, and as it might take a little for the subject to settle from the response, the longer interval could serve as a little break. So maybe something like 2 - 4 s and 5 - 7 s. However, I don't really understand why you have to separate the cue from the response to the outcome (partial trials would be another option for that purpose, but this would also increase no. of trials and required scanning time). If you're not really interested in the cueing effect then you could decrease that interval maybe to 1 - 3 s or even less variable (but still some jitter so that subjects can't exactly predict stimulus occurrence) and just model the outcome. Outcome would thus be confounded by cue, but as long as this holds for all your different conditions to the same extent (same jitter, same cue length) this shouldn't really be an issue. Best Helmut