The criterion method: Some analyses and remarks

The distribution of trials to criterion under the null hypothesis of no learning is obtained for 2 broad classes of criteria. Tables are provided which give the maximum permissible number of trials which may be run such that the probability of reaching criterion when learning does not occur remains less than a selected value. These results include Grant's analysis of the "runs" criterion as a special case. Illustrations of a number of problems which arise in the use of the criterion method also are presented to emphasize the often ignored dependence of method upon theory.