A concept of negative dependence using stochastic ordering (Q1067292)

The random variables \(T_ 1,...,T_ n\) (in the random vector T) are said to be negatively dependent through stochastic ordering (NDS) if \((T_ 1,T_ 2,...,T_{i-1}T_{i+1},...,T_ n)\downarrow^{st}T_ i\), \(i=1,2,...,n\). For the uninitiated, a stochastic ordering is introduced by considering the random variables X to be stochastically smaller than the random variable Y if \(P\{X>x\}\leq P\{Y>x\}\) for every real x. In this case we write \((X\leq^{st}Y)\). The concept is extended to vector valued random variables by requiring that \(g(X)\leq^{st}g(Y)\) for every g in the class of Borel measurable increasing functions on \({\mathbb{R}}^ n\). In this case X is said to be stochastically smaller than Y or \(X\leq^{st}Y.\) The definition and nomenclature of negative dependence are justified by showing that it implies \(Cov(T_ i,T_ j)\leq 0\), \(1\leq i<j\leq n\) (provided the second moments exist.) Several ''preservation'' results are proved. Models that yield such dependence are obviously of prime interest. The authors provide three such conditions. It is shown that some of the familiar multivariate distributions satisfy the conditions set down in the models. The authors provide stronger bases for some of the results proved earlier; for example, \textit{K. Jogdeo} and \textit{G. P. Patil} [Sankhyā, Ser. B 37, 158-164 (1975; Zbl 0343.62040)] and \textit{C. L. Mallows'} claim [Biometrika 55, 422-424 (1968; Zbl 0162.507)]. The results in this paper are considered analogously for positive dependence.