@ Donald: Thanks for clarifying! I just checked the updates.txt, what I had been thinking of was actually related to scaling of the TM depending on whether onsets /durations were specified in units of seconds vs. scans.
@ Colin and others: It depends. Going with condition mean (or trial-specific) RTs instead of an overall duration allows to control for amplitude differences that would emerge if the conditions (trials) result in the same processes in principle, which last longer/shorter according to RT though ("time on task" effect). These differences could be considered trivial. It will introduce bias if conditions result in (additional) processes that are RT-independent and/or if they depend on RT but in the sense of a time-shifted response of the same length instead of a longer response (or in a more complex way). When going with fixed durations across conditions one inevitably introduces bias as well, it's just the type of bias that changes.
Notably, the model-based approaches are very dominant despite very limited evidence that the predictors used are reasonable approximations of the time courses. But if they are poor then the estimates will be so as well. Maybe the limited temporal resolution of fMRI has led to the impression that timing aspects of the processes involved are not very important. Maybe it's because it's too much about getting colourful blobs no matter what they mean. Often, there is only a vague idea of the processes involved, which again results from lack of theories. But without good concepts you will not be able to come up with reasonable predictors, as you don't know whether conditions differ in durations / onsets / magnitude / combination of those.
Best regards
Helmut
|