The fact that estimates of treatment effects using change scores, where (posttest pretest) is regressed on treatment, differ from analysis of covariance, where the posttest is regressed on both treatment and pretest.