On the local minimaxity of a test of independence in incomplete samples (Q1086950)

Let \(X=(X_ 1,X_ 2)\) have p-variate normal distribution with covariance matrix \(\Sigma =(\Sigma_{ij})\), \(i,j=1,2\). Suppose that we have n independent observations on X and that in addition, \(m_ i\) independent observations on \(X_ i\), \(i=1,2\) respectively. It is shown that the locally most powerful invariant test for testing the independence between \(X_ 1\) and \(X_ 2\), obtained by \textit{M. Eaton} and \textit{T. Kariya} [Ann. Stat. 11, 654-665 (1983; Zbl 0524.62051)], is locally minimax with respect to \(tr\Sigma_{12}\Sigma^{- 1}_{22}\Sigma_{21}\Sigma^{-1}_{11}\) within the class of all tests of the same level.