That makes sense--thank you! So erring on the side of excluding potentially problematic data may be better than erring on the side of not getting rid of data?
Regarding our SNR cutoff: A researcher who was performing some unfinished analyses of our dataset in AFNI calculated SNR using a command in AFNI (described in the following link:
http://www.personal.reading.ac.uk/~sxs07itj/web/AFNI_SNR.html). To check for the validity of these SNR results, we compared them against head motion results that we obtained using SPM, and found that there was a very high correlation between between bad SNR in AFNI and large head motions in SPM. So, we continued using the SNR cutoff from AFNI as a proxy for head motion.
I was wondering if I might be able to ask you one more thing, somewhat unrelated to my previous questions:
Do you have any thoughts on whether it would be advisable to use global normalisation when running first level analyses in SPM? (I've been having trouble getting advice on this issue, and your explanations have been so thorough I thought I should check to see if you might have some advice).