Dear Feng,
yes, the smoothness is estimated on the residuals. In most cases, the residuals should be slightly smoother than the smoothing kernel applied during preprocessing. If this is not the case, there might be some problem in the data (caused by e.g. scanner or experiment hardware, uncorrectable subject head motion). In your case where the estimate is only slightly less smooth than expected, everything might be ok though. You may first want to check the resel per volume (RPV.nii) and residual mean squares (ResMS.nii) images in your statististics folder. They should look rather smooth, with little local variation. If they do show unexpected roughness (RPV) or regions with high error variance where your model should fit well, you probably need to go back and review the original data.
Volkmar
|