LBI tests for multivariate normality in exponential power distributions (Q1182751)

Given a sample \(x_ 1,\dots,x_ n\) from \[ c(\theta)|\Sigma|^{-1/2} \exp[-2^{-1}\{(\underline x- \underline\mu)'\Sigma^{-1}(\underline x-\underline\mu)\}^ \theta] \] the authors test the hypothesis \(H: \theta=1\) vs \(A_ 0: 0<\theta<1\) or \(A_ 1: \theta>1\). In the case \(\underline\mu\) is known but \(\Sigma\) is unknown, they derive a unique locally best invariant (LBI) test and obtain the asymptotic null and non-null distributions of the test statistic. However, in the case where \(\underline\mu\) and \(\Sigma\) are both unknown, they simply derive an LBI test. They are unable to derive the null distribution of the test statistic, but do obtain the asymptotic null distribution of a modified test. A Monte Carlo study of the null distribution shows the adequacy of the asymptotic approximation.