Deletion diagnostics for the linear mixed model

Abstract

Modeling data is an integral element of modern statistical analysis. Methodological developments combined with the explosion in computing power over the past ten to fifteen years have greatly enhanced statisticians' ability to model situations and phenomena. The need to assess a model's validity and suitability is an integral element of the model building process. Model criticism is central to this thesis and specifically model criticism for one of the most frequently utilised models in statistical analyses, the Linear Mixed Model with normally distributed errors. Particular emphasis is given to deletion diagnostics for both the fixed and for the covariance structure parameters.