A class of asymptotic tests for principal component vectors (Q796934): Difference between revisions
From MaRDI portal
Added link to MaRDI item. |
Set profile property. |
||
Property / MaRDI profile type | |||
Property / MaRDI profile type: MaRDI publication profile / rank | |||
Normal rank |
Revision as of 01:14, 5 March 2024
scientific article
Language | Label | Description | Also known as |
---|---|---|---|
English | A class of asymptotic tests for principal component vectors |
scientific article |
Statements
A class of asymptotic tests for principal component vectors (English)
0 references
1983
0 references
Let \(Y_ 1,...,Y_ n\) be a random sample from a p-variate normal distribution \(N_ p(\mu,\Sigma)\) with \(\Sigma\) nonsingular and let \(S_ n\) be the sample covariance matrix. The eigenvalues of \(\Sigma\) and \(S_ n\) are denoted by \(\lambda_ 1\geq...\geq \lambda_ p>0\) and \({\hat \lambda}{}_ 1\geq...\geq {\hat \lambda}_ p\geq 0\), respectively. Assume that \(\lambda_{i-1}>\lambda_ i\) (if \(i>1)\) and \(\lambda_{i+m- 1}>\lambda_{i+m}\) (if \(i+m\leq p)\). Let \(P_ 0\) be the orthogonal projection onto the eigenspace spanned by the eigenvectors of \(\Sigma\) associated to the roots \(\lambda_ i,...,\lambda_{i+m-1}.\) For a \(p\times r\) matrix A of rank \(r\leq m\), the null hypothesis to be tested is \(H_ 0:P_ 0A=A\). For \({\hat \Omega}=\{{\hat \lambda}_ 1,...,{\hat \lambda}_ p\}\) and \({\hat \omega}=\{{\hat \lambda}_ i,...,{\hat \lambda}_{i+m-1}\}\), define \(U_ n(\nu,A)=A^ t[\sum_{\lambda \in {\hat \omega}}\{\lambda \nu /(\lambda -\nu)^ 2\}\hat P_{\lambda}]A\) where \(\hat P_{\lambda}\) is obtained from the spectral representation of \(S_ n=\sum_{\lambda \in {\hat \Omega}}\lambda \hat P_{\lambda}\). It is shown that \[ W_ n(A)=n\sum_{\nu \in {\hat \Omega}-{\hat \omega}}\{U_ n(\nu,A)\}^{- {1\over2}}A^ t\hat P_{\nu}A\{U_ n(\nu,A)\}^{-{1\over2}} \] has asymptotically a Wishart distribution \(W_ r(p-m,I)\) under the null hypothesis and a noncentral Wishart distribution with the same degrees of freedom and scale matrix under local alternatives. A class of invariant tests for the transformation \(Y_ j\to QY_ j+b\), \(A\to QAB\) for orthogonal matrix Q, \(p\times 1\) vector b and nonsingular matrix B is constructed from the latent roots of \(W_ n(A(A^ tA)^{- {1\over2}})\) which are asymptotically level \(\alpha\) and consistent. Their local powers are obtained. The case when the population distribution is elliptically contoured is also discussed. The results are an extension of an earlier author's work, ibid. 9, 725-736 (1981; Zbl 0474.62051).
0 references
principal component vectors
0 references
invariant asymptotic tests
0 references
spectral decomposition
0 references
tests for eigenvectors
0 references
normal distribution
0 references
orthogonal projection
0 references
eigenspace
0 references
Wishart distribution
0 references
local alternatives
0 references
latent roots
0 references
local powers
0 references