Hello,
I've previously asked about high pass filter cutoffs but I want to make sure
I understand. For event related designs, how long should this be? In the
paradigm I am using there are 36 trials, each 11 seconds. They are
randomized so we can't treat them as a cycle of a fixed number of events.
Here is the answer I got before:
2. I read in the FMRI book that the high pass filter cutoff should be
> set to 2-3 times the time for 1 trial/block. When should it be 2 and
> when should it be 3? Does this make a big difference?
The filter used in FEAT has quite a slow rolloff in frequency, so for
example in a block design it is actually safe to set the cutoff to the
same as a complete single cycle length of the paradigm. For event-related
designs you can reduce this (ie do more aggressive filtering) but there is
no hard rule on how far to take it. You want to check that the resulting
model (which is also filtered by the same filter by default) still shows
reasonable temporal variance.
So for event related designs, "a complete single cycle length of the
paradigm" means something like ABC if the paradigm were ABCABC...? Or is it
something else? What do we do in the case of randomized trials?
Thanks a million,
Bo
|