On estimation of matrix of normal mean (Q1820529): Difference between revisions

X\({}_ 1,...,X_ n\) are independent p-dimensional random vectors, \(X_ i\) having a p-variate normal distribution with unknown mean vector \(\theta_ i\) and covariance matrix the p by p identity matrix, \(n\geq p+1\). Denote by X the random matrix \((X_ 1,...,X_ n)\), and by \(\theta\) the corresponding matrix of means \((\theta_ 1,...,\theta_ n)\). Let \(L_ 1\geq L_ 2\geq...\geq L_ p\) denote the characteristic roots of the matrix \(XX^ T\). The problem is to estimate the matrix \(\theta\). Estimators of the form \[ X+(\partial \log f(L_ 1,...,L_ p)/\partial X_{jk}) \] are considered. Conditions on f are given which guarantee that the resulting estimator is minimax. These conditions generalize a result of \textit{C. Stein} [Proc. Prague Symp. Asympt. Stat., Vol. II, Prague 1973, 345-381 (1974; Zbl 0357.62020)]. Then estimators which dominate the minimax estimator are constructed.