The smoothness estimator, 
, needs to be measured from the residuals
of the linear model fit. This estimator explains the spatial
correlation structure of the SPM by assuming that it can be modeled as
being due to a convolution of the (white) real signal with a Gaussian
filter with an unknown, but determinable, width. Calculating the width
of this Gaussian allows us to determine the effective resolution of
the data.