multiple and conditional deletion diagnostics for general linear models

Abstract

In this paper we develop multiple case deletion statistics for the general linear model so that a residual vector and a leverage matrix are identified which have roles analogous to residuals and leverage for ordinary least squares models. We extend the notion of the conditional deletion diagnostic to general linear models. The residuals, leverage and deletion diagnostics are illustrated with data modelled by a linear growth curve.

Keywords:

• Cook's distance
• correlated errors
• generalised least squares
• in¬fluence
• leverage
• one-step approximation
• residual