As this expectation does not equal 1 when the null hypothesis is true (RIL variance component = 0), the F-test is not correct. Instead, more complicated forms of the F-test are required, and Proc GLM computes these forms when the random statement is given: