Dear Sebastien,
Yes, the large number of regressors can become problematic, also with regard to second level statistics. It's not uncommon to focus on selected time bins then (e.g. for brief epochs those where one would expect the largest activations) to reduce the number of necessary tests. This might also be the reason why people often just go with a FIR set for one of the components. Of course, if you have good reasons to assume the transient (or sustained) activation to follow the canonical HRF then there's no need for a FIR set for that component.
> included FIR regressors for all three experimental conditions
In your case it's worth a note that there might be large correlations between FIR time bins for dichotic listening and those for the two tasks. In an extreme case think of the task onsets to be jittered by a small amount only, thus e.g. always occuring ~ 20 s after "dichotic listening" onsets. It might be difficult, possibly even impossible, to properly differentiate between the sustained component and the transient one, as e.g. the first task time bin would be aligned to the 11th dichotic time bin and so on. You have two different tasks, but they might share some (transient) activation patterns, and these might be similar to the sustained activation for those time bins. Thus ideally, one would probably have to go with a mixture of dichotic listening blocks = blocks with sustained activation only and ones with the additional task somewhere in between - in case the distinction between sustained and transient components is relevant, as possibly, the sustained one is just treated as a "confound". Anyway :-)
Best
Helmut
|