Hi Gang,
Thanks for your help. It slipped my mind to thank you last time I posted as well. It turns out looking at the individual model fit for each subject revealed that we had some issues with the SENSE coil on the last part of the first run for one subject. The variance attributed to this was being eaten up by the 2nd and 3rd derivatives in the SPMG3 basis function but my polort was too low for the GAM function to work -- hence the wacky results when comparing the GAM to the SPMG3 basis functions.
Thanks for all your help!!!
Aaron