POSSIBLE INTERPRETIVE PROBLEMS IN ANALYSES USING GROUP MEANS AS EXPERIMENTAL UNIT

FICTITIOUS DATA ARE PRESENTED FOR 5 6TH-GRADE SS IN EACH OF 2 GROUPS WITH ALL SS TESTED UNDER 4 TREATMENTS OF MONOTONICALLY INCREASING DIFFICUTY. ALL SCORES WERE RANDOMLY DETERMINED EXCEPT THOSE OF ONLY 1 DEVIANT S IN EACH GROUP WHOSE SCORES WERE ALLOWED TO REFLECT THE PREDICTED TREATMENT EFFECTS. WHEN APPLIED TO THE 2 SETS OF 4 TREATMENT MEANS, BOTH A NONPARAMETRIC TEST (PAGE'S L TEST) AND A PARAMETRIC TEST (F TEST FOLLOWING AN ANALYSIS OF VARIANCE) PERMITTED REJECTION OF THE NULL HYPOTHESIS-THAT THE OBTAINED SCORES REPRESENT RANDOM SAMPLES OF NUMBERS FROM THE SAME POPULATION-AT THE .01 LEVEL. HOWEVER, WHEN APPLIED TO ALL OF THE DATA, NEITHER TEST PERMITTED ITS REJECTION. IT IS SUGGESTED THAT ALL SCORES SHOULD ENTER ANY ANALYSIS SO THAT THE RESULTING STATISTICAL INFERENCE WILL BE MORE ANALOGOUS TO THE EXPERIMENTAL INFERENCE USUALLY OF INTEREST IN BEHAVIORAL RESEARCH. (PsycINFO Database Record (c) 2006 APA, all rights reserved). © 1968 American Psychological Association.