Key points are not available for this paper at this time.
Abstract A method is proposed in which a statistic, Θ, is used as a probabilistic measure of strength of association in the multivariate prediction of a nominal scale by nominal and/or ordinal scales. The relationship of theta to the bivariate Goodman and Kruskal's Lambda statistic is given. An algorithm using the theta statistic as criterion in an exhaustive search and optimal stepwise binary split procedure is described. A computer program and sample results are discussed.
Messenger et al. (Fri,) studied this question.