Biometrika 1970 57(1):224; doi:10.1093/biomet/57.1.224-a
© 1970 by Biometrika Trust
CORRECTIONS |
Corrections
The main result was that if x
N(0,C), then x'Ax has a chi-squared if and only if the eigenvalues of AC are all 0 or 1. The three corollaries were then intended to be in part contributions to matrix theory, but Dr D.N.Shanbhag and Dr G.P.H. Styan have kindy pointed out that the necessary parts of Corollaries (i) and (ii) are incorrect was incorrect. Also I misquoted one of Shanbhag's results by writing rant in place of trace. He proved that if AC is symmetric then the criteria of Corollary (i) are valid; a fact that I stated ambiguously. The error made in the proof of Corollary (ii) was in the assumption that the rank of a matrix B is equal to the number of its nonzero eigenvalues. In fact it is equal to the number of its nonzero singular values, which are the square roots of the nonzero eigenvalues of BB'. It is correctly, but only implicitly assumed that AC has real eigenvalues when A and C are real and symmetric and one of them is non-negative definite; see, for example, p.378 of Good [J. R. Statist. Soc. B 25 (1963), 37782; Corrigenda 28 (1966), 584].