Influential observations in anova and manova models

Abstract

A large number of diagnostic statistics have been proposed for identifying influential observations for the general linear model. Although regression, Anova, and Manova models are three special cases of the general linear model, many of the proposed diagnostics for regression are not suitable for Anova and Manova. In this paper several statistics are used for identifying influential observations in both Anova and Manova models. The results show that Welsch-Kuh distance along with likelihood displacements can be used for identifying influential observations. A few examples are considered to illustrate the methods.

Keywords:

• Diagnostic statistics
• influential observations
• ANOVA
• MANOVA