Hi Rick,

Could you explain the rational for why smoothing should be done before mean-normalization? Is there a model for the relation between noise magnitude and signal-baseline level that is driving this?
If I understand your point correctly, smoothing in the original value-range (i.e., before scaling) would help canceling out noise whose magnitude is baseline-independent. But does the scanner noise (and other types of noise) always have this property of baseline independence? Is noise magnitude never relative to a voxel's baseline? Intuitively, if noise magnitudes can/may be scaled relative to baseline, then the best way to cancel that noise out (e.g., in 2 nearby voxels that have very different baseline means) would be after baseline normalization. I haven't read any literature on this, so any pointers/experiences would be greatly appreciated.
Also, if this is an important point, might be good to add it to the soup_to_nuts walk-through. Till today I've never given the order of these two steps much thought, but it can clearly make a big difference.

Oori