ABSTRACT
Generalized linear models are essentially an extension of classical linear models and this chapter presents these classical models in a way that makes the extension appear natural. The emphasis on second-moment assumptions over fully specified distributional assumptions extends to all generalized linear models. The Normal distribution is useful primarily as a model for measurements of continuous quantities, though it can also be used as an approximation for discrete measurements. It is frequently used to model data, such as weights, lengths and time, which, though continuous, are essentially positive, although the distribution itself covers the entire real line. Models containing only terms with continuous covariates are often called regression models, to be contrasted with analysis-of-variance models, which have only terms involving qualitative factors. Factors occurring in a model may be of primary interest, meaning that a principal purpose of the study is to measure their effect.
