Communications in statistics-theory and methods vol:28 issue:10 pages:2487-2495
Local influence on the eigenvalues of sample covariance matrices in principal components analysis is examined for a reasonable modification of Shi's (1997) perturbation scheme. The modification is suggested for samples from populations with both unknown mean vector and covariance matrix. While Shi's detection indexes (1997) consist of only quadratic terms, the modified perturbation scheme leads to detection indexes constituted by both linear and quadratic terms associated with centralized observations. These linear and quadratic terms reflect local influences on the first two sample moments. Examples are investigated based on the two detection indexes.