Influence Diagnostics for a Class of Biased Estimators in Regression
Fractional Principal Components (FPC) is a large class of estimators that include, among others, least squares (LS), ridge, principal components, fractional rank, and Stein estimators. This paper addresses the measurement of influence for FPC estimators. It is shown that the residual and the leverage of any observation depends directly on the “fraction” of the principal components that is used. Generalizations of popular measures of influence for LS are proposed. Approximate deletion formulas are presented.