Stochastic comparisons of \(m\)-spacings (Q2581797)

Let \(X_{1:n}\leq X_{2:n}\leq\cdots\leq X_{n:n}\) be the order statistics from a sample of \(n\) independent random variables \(X_1,X_2,\ldots,X_n\), all identically distributed with support \((0,\infty)\). For \(1\leq m\leq n\), let \(D_{k:n}^{(m)}=X_{m+k-1:n}-X_{k-1:n}\), \(k=1,2,\ldots,n-m+1\), denote the \(m\)-spacings; here \(X_{0:n}\equiv0\). The authors show that if \(X_1\) has a logconvex [logconcave] density then, for \(k=1,2,\ldots,n-m+1\), it holds that \(D_{k:n}^{(m)}\leq_{\text{ lr}}[\geq_{\text{ lr}}]\;D_{k+1:n+1}^{(m)}\), where \(\leq_{\text{ lr}}\) denotes the likelihood ratio order. Furthermore, if \(X_1\) has a logconcave density then \(D_{k:n}^{(m)}\leq_{\text{ lr}}D_{k-1:n+1}^{(m+1)}\). The authors also show that if \(X_1\) has increasing failure rate and decreasing reversed hazard rate then \(D_{k:n}^{(m)}\leq_{\text{ hr}}D_{k-1:n+1}^{(m+1)}\), where \(\leq_{\text{ hr}}\) denotes the hazard rate order.