The Structure of Ellipsoidal Distributions I. Canonical Analysis (Q3680089)
From MaRDI portal
!
This is the item page for this Wikibase entity, intended for internal use and editing purposes. Please use the normal view instead:
scientific article; zbMATH DE number 3901831
| Language | Label | Description | Also known as |
|---|---|---|---|
| default for all languages | No label defined |
||
| English | The Structure of Ellipsoidal Distributions I. Canonical Analysis |
scientific article; zbMATH DE number 3901831 |
Statements
The Structure of Ellipsoidal Distributions I. Canonical Analysis (English)
0 references
1984
0 references
robustness
0 references
large sample approximations
0 references
domains of attraction to ellipsoidal stable laws
0 references
joint distribution of the squared sample canonical correlations
0 references
distribution of the standardized canonical variables
0 references
departures from ellipsoidal symmetry
0 references
0.8352190256118774
0 references
0.8311418294906616
0 references
0.8208178281784058
0 references
0.8157216906547546
0 references
This paper is an important contribution in the area of multivariate robustness studies. While most of the work in this connection is concerned with the performance of the tests developed under normality, the author shows that the actual concerns should be directed towards lack of ellipsoidal symmetry. It is pointed out that the large sample approximations for the distributions attracted to the normal law under the central limit theorem are equally applicable to distributions ''lying in the domains of attraction to ellipsoidal stable laws''. Two main theorems are established here: (i) The joint distribution of the squared sample canonical correlations \((r^ 2_ 1,..., r^ 2_ p)\) is the same for all distributions in the class \(\{S_{n,m}(I_ n\times \Sigma): (M, \Sigma)\in M\times S^+_ m\}\). They depend on the parameters only through the population canonical correlations \(\rho^ 2_ 1, \rho^ 2_ 2,..., \rho^ 2_ p.\) (ii) The distribution of the standardized canonical variables is the same for all distributions in this class; this depends on the \(\Sigma\) only through \((\rho^ 2_ 1,..., \rho^ 2_ p).\) A consequence of the first result is that the distribution of \((r^ 2_ 1,..., r^ 2_ p)\) is exactly as that given by Fisher and others (for the null case) and by Bartlett and others (for the non-null case). It is noted by the author that several of the results using the normal distribution for testing hypothesis are exact in this class of distributions. Questions of robustness in the canonical analysis and related testing problems should therefore be concerned with the departures from ellipsoidal symmetry rather than from normality.
0 references