Likelihood ratio testing on partial multinormal data (Q1058243)

The paper is mainly concerned with the problem of missing data in a multinormal distribution. Of particular interest is the question: When does the incorporation of partial records prove beneficial (in terms of reducing variance of the estimators)? This depends on the significance of the information provided by the partial records. Since the methods involved in the inclusion of such information into estimation procedures are quite complex, the authors suggest an a priori evaluation of this significance. The results of the paper show that the likelihood ratio procedure is appropriate for testing for the significance of the information provided by the partial records. The test statistic suggested here is: \[ Q = \ell n \left({\tilde\mu},{\tilde\Sigma}\right) - \ell n L\left({\hat\mu},{\hat\Sigma}\right), \] where \(L\) is the likelihood of \(({\underline\mu},\Sigma)\). Here \(\underline{\tilde\mu}\), \({\tilde \Sigma}\) are the \textit{R. R. Hocking} and \textit{D. L. Marx} estimates [Commun. Stat., Theory Methods A8, 1155--1181 (1979; Zbl 0412.62031)] while \(\underline{\hat\mu}\), \({\hat\Sigma}\) refer to the usual \(ML\) estimates. The statistic \(Q\) is appropriate for testing the hypothesis of ''no added information in the partial data vectors''. It is pointed out that the Hocking-Marx procedure and the present procedure for testing can be tied up together, stepwise. A remark on the sloppiness of the paper will be apropos. There are several statements that need clarification. The reference work is also rather poor. The \textit{J. N. Srivastava} and \textit{M. K. Zaatar} paper, Ann. Inst. Stat. Math. 26, 299-313 (1974; Zbl 0382.62068), is missing in the bibliography, as is the full reference for the \textit{T. Orchard} and \textit{M. A. Woodbury} paper, Proc. 6th Berkeley Sympos. math. Statist. Probab., Univ. Calif. 1970, 1, 697-715 (1972; Zbl 0263.62023). Some important references in this area have been overlooked, for example \textit{R. C. Dahiya} and \textit{R. M. Korwar}, Ann. Stat. 8, 687-692 (1980; Zbl 0435.62032).