Detection of Multiple Influential Cases in Principal Component Analysis: A Graphical Technique

摘要

The detection of multiple influential cases in principal component analysis through case-deletion necessitates the investigation of a large number of combinations of observation cases to locate the influential cases. In this article, a graphical technique is proposed for the detection of multiple influential cases in principal component analysis based on the empirical influence curve. Detection of multiple influential cases is undertaken by visually inspecting outlying points in a series of two-dimensional diagnostic plots. It is shown that the proposed graphical method is easily interpretable and incurs very low computational costs. A practical example based on soil composition data that has been used in previous studies into multiple influential cases is used to illustrate the proposed graphical technique.