Abstract
Many measures of predictive power have been recommended. These include Mean Cost Rating (MCR), P(A), (from Signal Detection Theory), Kendall's rank correlation coefficient tau, and Goodman and Kruskal's gamma. It is shown that these four measures are related as all are merely different transformations of the statistic S. By reference to S, whose distribution is known, hypothesis testing is made possible. Of the four measures considered gamma is generally to be preferred. Remarks made here apply equally to measures of association for ordered contingency tables.
Get full access to this article
View all access options for this article.
