ABSTRACT
When responses are qualitatively different, contingency chi-square tests can be employed to test the null hypothesis of no dif ferential treatment effect. However, a contingency chi-square test is nondirectional, and the results of the test reflect only the consistency — and not the nature — of the association between types of responses and types of tr eatments. For instance, a contingency chi-square test would not use the specific pr ediction that treatment A would lead to the death of animals and tr eatment B to survival but only the prediction that the tr eatments would have dif ferent effects on mortality. In this case, there is a dichotomous dependent variable, so Fisher ’s exact test can be used as the desir ed one-tailed test, but ther e are no standard procedures for accommodating specificity of prediction when there are more than two categories of r esponse. Let us now consider an example in which we predict a specific type of response for each subject and determine the P-value of the number of corr ect predictions.
