Negative dependence in the balls and bins experiment with applications to order statistics (Q2507759): Difference between revisions

Let \(X_{1:n}\leq X_{2:n}\leq\cdots\leq X_{n:n}\) be the order statistics of \(n\) independent random variables \(X_1,X_2,\ldots,X_n\) with possibly different distribution functions. The authors show, for \(1\leq i< j_1<j_2<\cdots<j_r\leq n\) and fixed \(x_1,x_2,\ldots,x_r\), that \[ P(X_{j_1:n}>x_1,X_{j_2:n}>x_2,\ldots,X_{j_r:n}>x_r\,|\,X_{i:n}>s) \] is nondecreasing in \(s\). They also show that if \(A_{i,s}\) denotes either \(\{X_{i:n}>s\}\) or \(\{X_{i:n}\leq s\}\) then \[ P(X_{j_1:n}>x_1,X_{j_2:n}>x_2,\ldots,X_{j_r:n}>x_r\,|\, A_{i,s}) \] is nonincreasing in \(i\) for fixed \(s\). Furthermore, if each \(X_k\) has a continuous distribution function and if \(A_{i,s}\) denotes either \(\{X_{i-1:n}<s<X_{i:n}\}\) or \(\{X_{i:n}=s\}\) then \[ P(X_{j_1:n}>x_1,X_{j_2:n}>x_2,\ldots,X_{j_r:n}>x_r\,|\, A_{i,s}) \] is nonincreasing in \(i\) for fixed \(s\). These results complement and generalize results of \textit{P. J. Boland, M. Hollander, K. Joag-Dev} and \textit{S. Kochar} [J. Multivariate Anal. 56, No.1, 75--89 (1996; Zbl 0863.62044)].