Notice that the default F-tests for each factor shown here are correct only for the model in which
all effects except residuals are fixed. The default F-tests use the residual error variance as the
denominator in all cases. Since we have assumed that all effects are random, the correct form of
the F-test depends on the expected mean squares. So, for example, to test the null hypothesis that
the variation among RILs is zero, the correct F-test is: