Estimation of Multivariate Distributions under Stochastic Ordering

Abstract

Abstract Let F and G be the cdf's of two p-dimensional multivariate distributions, such that F is stochastically larger than G. A straightforward derivation is given of the generalized maximum likelihood estimators of F and G, based on random samples from each population. An algorithmic approach to computing these estimators is described and motivating numerical examples are discussed. The special case when F and G correspond to multivariate ordinal contingency tables is also presented. The relationship of these results to those of Robertson and Wright (1974) is considered.