A graphical procedure for comparing the principal components of several covariance matrices
作者:
E. M. Keramidas,
S. J Devlin,
R Gnanadesikan,
期刊:
Communications in Statistics - Simulation and Computation
(Taylor Available online 1987)
卷期:
Volume 16,
issue 1
页码: 161-191
ISSN:0361-0918
年代: 1987
DOI:10.1080/03610918708812583
出版商: Marcel Dekker, Inc.
关键词: eigenvectors;dispersion matrices;direction of scatters;orientation of point clouds
数据来源: Taylor
摘要:
Principal components analysis is an extensively used tool for reduction of dimensionality in multivariate analyses. In many applications, however, little attempt is made to compare principal components solutions (i.e., eigenvectors) across many samples. Methods are needed for assessing the degree of similarity of corresponding eigenvectors, a problem that is meaningful in the presence of clearly separated eigenvalues. This paper proposes a gamma probability plotting procedure for a measure of the angle between a pair of eigenvectors, or equivalently, the distance between points on the unit sphere defined by such vectors. One of the vectors in the pair is the principal component of a sample and the other can be either a prespecified vector or a “typical” vector obtained from the corresponding eigenvectors in all samples. Simulations, as well as real-data examples, are presented
点击下载:
PDF (681KB)
返 回