An Algorithm for Critical Values of the Two-Group Rank Distance Classification Statistic

摘要

The rank distance method is restated for discriminating between two groups using a single variable. A formula is derived for the probability of misclassifications in a training set of n observations from each group. Knowledge of this probability permits inference as to the appropriateness of the variable for discrimination purposes. The formula is exact only for values of U below a bound which is proportional to n. A table of numerical values indicates the formula to be useful up to about n = 25 but not beyond n = 40. (Author)