Let x be an n-dimensional random vector for which there are (constant) nonzero vectors a and b so that a'x and b'x are independent in every coordinate system. The paper proves that this condition implies that either the \(\{x_ j\}\) are degenerate or \(a\perp b\) and the coordinates of x are mutually independent normal variables with a common variance. (Less general earlier results of this kind are due to Maxwell, Bartlett, M. Kac, etc.) The paper also considers the linear model \(x=Vb+\epsilon\), where V is a known \(n\times p\) matrix, b is an unknown \(p\times 1\) vector of regression coefficients, and \(\epsilon\) is spherically distributed in \(R^ n\) with known distribution. The least-squares estimator \(\hat b\) of b is a sufficient statistic iff the \(\{\epsilon_ j\}\) are iid normal variables with mean zero.
This page was built for publication: Sphericity and the normal law
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q1077795)
