VULNERABILITY OF CONTRAST TESTS TO SIMPLER INTERPRETATIONS:

Abstract— The recent article in this journal by Rosnow and Rosenthal (1995) amplified on previous pleas that in analyses of variance, priority be given single-degree-of-freedom contrasts focused on specific hypotheses. The present commentary supports this plea, but notes a problem in choosing appropriate contrasts when global predicted patterns link two or more local effect predictions. Misleading claims can arise in very simple experimental designs, because global patterns are often confounded with relatively uninteresting simple effects or main effects. This problem is illustrated with a 1 × 3 example and a discussion of the 2 × 2. With careful attention to what points are of most interest in the data, and an appreciation of the potential sensitivity of some contrasts to banal alternative interpretations, this problem can be avoided.