A concentration study of principal components

Abstract

Influence functions are commonly used as diagnostic tools in order to investigate sensitivity aspects in principal component analysis. This paper suggests a practical alternative for the eigenvalues by introducing a sensitivity measure derived from the classical Lorenz curve and associated Gini index. The results are illustrated by analysing an example.

Keywords:

• Gini index of concentration
• influence function
• Lorenz curve
• principal component analysis
• sensitivity