Comment on Influential Observations, High Leverage Points, and Outliers in Linear Regression

摘要

The authors thank Chatterjee and Hadi for their review of diagnostics for influence of individual cases in regression. By describing connections and distinctions among a large number of techniques, they have provided a springboard for a timely discussion of specific methods, general approaches, and the contribution of diagnostics to the practice of regression analysis. This document offers some criticism of notation, considers cutoffs, rules of thumb, and their role in identifying influential cases, proposes simple residual plots which display high leverage, outlying, and influential cases simultaneously, comment on the selection of carrier subsets, discusses approaches to uncovering influential groups of observations, urges more comprehensive presentation of examples, and sketches a step by step diagnostic strategy that should be useful in practice.