Dear list,
Looking for insights on how others model nuisance events such as:
instructions, small number of foils, No Responses, movement, bad scans etc.
In particular I'm curious what to do in the case of a small number
events. For example: if an instruction screen appears twice for 1 TR in
a run is it best to let it fall into baseline? Or try and model it even
though two events isn't really enough to estimate the response.
The feeling around here is that if your regressor(s) are not
estimable due to insufficient events, it's probably best to let them
fall into baseline if they're events like no-responses or instructions,
rather than attempt to model them as nuisance variables.
However I've seen mention on this list and in textbooks that one can
create nuisance variables for infrequently occurring events such as
artifacts and movement spikes
(http://www.jiscmail.ac.uk/cgi-bin/wa.exe?A2=ind06&L=SPM&D=0&I=-3&P=374076)
or for instruction screens
(http://www.jiscmail.ac.uk/cgi-bin/wa.exe?A2=ind05&L=SPM&P=R463067) even
though oftentimes these events are two few to be properly estimated.
I understand AFNI has a way of excluding volumes from the GLM and
fMRISTAT has an exclude variable that allows you to specify frames to be
left out. Is there any similar hacks someone has implemented in spm or are
we stuck with manually hacking out bad volumes and replacing with the average?
Cheers!
|