Hello Jess,
this is to the Nyquist sampling theorem, see http://en.wikipedia.org/wiki/Nyquist%E2%80%93Shannon_sampling_theorem . http://spm.martinpyka.de/?p=51 might also give you an idea what happens if the high-pass filter is "too short".
You can determine an appropriate high-pass filter by calculating the mean interval between subsequent onsets of one regressor. For "HighA" in Design_1_high-1 the onsets seem to be approximately 40 seconds, 160 seconds, and 240 seconds (in case your TR = 2s). The mean difference of subsequent trials is then (120 + 80)/2 = 100 seconds. For "LowA", the onsets seem to correspond to 100 seconds, 300 seconds and 360 seconds, so you get (200 + 60)/2 = 120 seconds. For your high-pass filter, enter a value that is at least double the size of the mean difference, that is 240 seconds in this case, or maybe e.g. three times the size which would be 360 seconds then. The high-pass filter should be the same for different subjects, so if there are different onsets for different subjects calculate the mean intervals for all of them and take the "longest" .
You might also want to take a filter which is twice as large as the longest interval between trials of the same condition. This would be 400 seconds in your case. At least I have already read both options in papers. Also have a look at an older thread, which is a collection of various postings https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=ind06&L=spm&D=0&1=spm&9=A&J=on&K=4&X=35D0701792D24A3126&Y=oederland%40gmx.ch&d=No+Match%3BMatch%3BMatches&z=4&P=5990336
One problem might be that you pick up noise depending on whether you suffer from scanner drifts or not.
Hope this helps,
Gabor
|