Influence function analysis applied to partial least squares (Q964607)

The authors consider partial least squares (PLS) components as solutions to the corresponding eigenvalue problem. Empirical influence functions (EIFs) are constructed for the estimates of eigenvalues and eigenvectors. These EIFs enable one to identify observations that directly influence the eigenstructure of PLS and hence the performance of the PLS model. Applications to simulated and real life data are considered.