Asymptotic improvement of the usual confidence set in a multivariate normal distribution with unknown variance (Q1383916): Difference between revisions

This paper considers an improvement of the coverage probability of the usual confidence set of the mean vector of the \(p\)-dimensional normal distribution, where \(p\geq 3\). Suppose that \(X\) is a \(p\)-variate normal random vector with unknown mean vector \(\theta\) and unknown covariance matrix of the form \(\sigma^2I\), \(S\) is independent of \(X\) and is distributed as \(\sigma^2\) times a chi-squared random variable with \(n\) degrees of freedom. The usual confidence set of \(\theta\) is \[ C_0(X,S)= \bigl\{\theta; |\theta-X|^2 <cS \bigr\}, \] where \(|\theta -X|\) denotes the Euclidean distance between \(\theta\) and \(X\), and \(c\) is some positive constant determined such that \(P\{\theta\in C_0(X,S)\} =1-\alpha\). We consider a confidence set of the form \[ C(X,S)= \biggl\{\theta; \bigl| \theta-\delta (X,S)\bigr |^2 <cS \biggr\}, \] where \[ \delta (X,S)= \bigl(1- b/(a+F+b) \bigr)X \] with \(F=| X|^2 /S\) and positive constants \(a\) and \(b\). In Section 2 we get an asymptotic expansion of the coverage probability of \(C(X,S)\) as \(a\to\infty\), from which the asymptotic improvement is shown if \(0<b<2 (p-2)(c+1)/(n+p+2)\).