Dear everyone,
just ran into that recently: What are the current opinions with regard to autoregressive modeling (if there are any new ones at all) for task-based fMRI? There's not much literature on that topic in the context of fMRI, and looking at these papers it sounds as if there's no final agreement. For example Della-Maggiore et al. (2002) concluded:
Finally, given that both the low pass filter and the first-order autoregressive function decrease power, it is only recommended to use them for fMRI datasets with short ISI (8 s), which are more susceptible to inferential bias. In those cases, AR1 appears to be more efficient than HRF in that it controls for the incidence of false positives while maintaining a relatively high power.
There are a few messages on that issue at SPM mailing list, for example https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=spm;7f532b94.1205 by Torben Lund (original thread was about autoregressive modeling in the context of resting state) and https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=spm;eb8bf667.0803 with answers by Will Penny and Thomas Nichols (short message "check models with and without, if there are large differences auto-correlation seems to be an issue). In some of the papers nowadays AR(1) is explicitely mentioned, others don't state anything related, which might also mean that they went with the default settings = AR(1) though. So, are there any instances in which one does not have to rely on autoregressive models, for example longer ISIs? What about FIR models?
Sorry for bringing up this topic :-)
Helmut
|