Access
You are not currently logged in.
Access JSTOR through your library or other institution:
Journal Article
Robust Tests of Mean Vector in Symmetrical Multivariate Distributions
N. C. Giri and B. K. Sinha
Sankhyā: The Indian Journal of Statistics, Series A (19612002)
Vol. 49, No. 2 (Jun., 1987), pp. 254263
Published by: Indian Statistical Institute
Stable URL: http://www.jstor.org/stable/25050647
Page Count: 10
You can always find the topics here!
Topics: Minimax, Statism, Matrices, Mathematical theorems, Density distributions, Mathematical vectors, Applied statistics, Mathematical robustness
Were these topics helpful?
See something inaccurate? Let us know!
Select the topics that are inaccurate.
 Item Type
 Article
 Thumbnails
 References
Abstract
Let $X=(X_{ij})=(\mathbf{\mathit{X}}_{1},\ldots,\mathbf{\mathit{X}}_{n})\text{'}$, $\mathbf{\mathit{X}}_{i}^{\prime}=({\rm X}_{i1},\ldots ,X_{ip})$ be a n×p random matrix with probability density function $f_{X}(x)=\Sigma^{n/2}q({\rm tr}\ \Sigma ^{1}(x\mathbf{\mathit{eu}}^{\prime})^{\prime}(x\mathbf{\mathit{eu}}^{\prime}))$ where x ε X = {x:n×p matrix rank of x = p}, $\boldsymbol{\mu}=(\mu _{1},..,\mu _{p})^{\prime}\epsilon R^{p},\mathbf{\mathit{e}}=(1,..,1)^{\prime}$ nvector and Σ > 0 (positive definite). Set $Q_{1}=[q\colon [0,\infty)\rightarrow [0,\infty)$, $\underset R^{np}\to{\int }q({\rm tr}\ u^{\prime }u)du=1$, q thrice continuously differentiable, $\underset R^{np}\to{\int }q^{(2)}({\rm tr}\ u^{\prime }u)\ du<\infty $, $\underset R^{np}\to{\int }q^{(3)}({\rm tr}\ u^{\prime }u+\varepsilon )du<\infty $ for some ε > 0}, $Q_{2}=[q\colon [0,\infty)\rightarrow [0,\infty)$, $\underset R^{np}\to{\int }q({\rm tr}\ u^{\prime }u)du=1$, q convex}. Assume that n > p so that $S\equiv X^{\prime}(I_{n}\mathbf{\mathit{ee}}!n)X>0$ with probility one. It is proved that for testing $H_{10}\colon \boldsymbol{\mu}={\bf 0}$ versus the alternative $H_{{\bf 11}}\colon \boldsymbol{\mu}\neq {\bf 0}$, the Hotelling's $T^{{\bf 2}}\text{}\text{test}$ is locally minimax for $q\epsilon Q_{1}$, and for testing $H_{2{\bf 0}}\colon \boldsymbol{\mu}_{(1)}={\bf 0}$ versus the alternative $H_{2{\bf 1}}\colon \boldsymbol{\mu}_{(1)}\neq {\bf 0}$, the appropriate Hotelling's $T^{2}\text{test}$ is UMPI for $q\epsilon Q_{2}$ and locally minimax for $q\epsilon Q_{1}$. In the second case $\boldsymbol{\mu}_{(1)}=(\mu _{1},..,\mu _{p_{1}})^{\prime},p_{1}
Page Thumbnails

[254]

255

256

257

258

259

260

261

262

263
Sankhyā: The Indian Journal of Statistics, Series A (19612002) © 1987 Indian Statistical Institute