Theory and illustration of regression influence diagnostics

Abstract

Regression diagnostics are an important tool of the professional statistician, and they are neglected at one's peril Deletion and local influence diagnostics (see Cook 1977, 1986 for origins) relevant to Normal linear regression are considered here, as these are in most common use. The presentation extends beyond the traditional parameter estimate influence to indicate the possibilities for model test influence and model goodness of fit influence, both also based on familiar uses of the likelihood. Two real-life data sets are presented and analysed using some of the more traditional diagnostics.

Keywords:

• deletion influence
• regression diagnostics
• perturbation

Additional information

Notes on contributors

G. C. Brown