The sum of squares Q for any null hypothesis can always be computed as
a difference in residual sums of squares. For null hypotheses where m = 0,
the same result can be obtained, sometimes more conveniently, by taking
the difference in the model sums of squares; that is,