Figure 1.
The reliability paradox. Top panels (A–C) illustrate the statistical tests under consideration: (A) test-retest correlation (same measure obtained twice), (B) one-sample test (mean of a single condition compared to zero), and (C) paired-sample test (mean difference between two conditions). Bottom panels (D–F) show how the observed outcomes of these tests depend on the relative contributions of error variance and between-subject variance. Test-retest reliability (D) increases when error variance is minimized and between-subject variance is maximized, whereas observed one-sample and paired-sample effect sizes (E, F) increase when both error and between-subject variances are minimized.
