Ð35Ð
Solving zÈRb for b
¥ Number of equations = number of time points
H 100s per run, but perhaps 1000s per subject
¥ Number of unknowns usually in range 5Ð50
¥ Least squares solution:
H     denotes an estimate of the true (unknown)
H From    , calculate            as the fitted model



o            is the residual time series = noise (we hope)
¥ Collinearity: when matrix         canÕt be inverted
H Near collinearity: when inverse exists but is huge