Abstract
Consider experimental treatments with consequences so irreversible that baseline performance cannot be recovered. The conventional method of assessing the effects of such treatments by statistical means involves separate experimental and control groups. An alternative proposed here is to administer the experimental treatment to each subject, one subject at a time and in a random order; whenever any subject receives the experimental treatment, those subjects which have not yet received it receive a control treatment. This procedure permits results significant at the one-tailed 0.05 level to be obtained with four subjects; if a two-group procedure evaluated by means of the U test is used, a minimum of six subjects is needed for the same significance level. More generally, the procedure permits equal sensitivity to any experimental effect with over 30% fewer subjects than a two-group procedure. Extensions of the basic method are made to a variety of levels of the experimental treatment and to treatments without irreversible effects, and limitations of the method are discussed.