ABSTRACT
Generalized linear models (GLMs), originally introduced by Nelder and Wedderburn (1972), provide a unifying family of models that is widely used for regression analysis. These models are intended to describe non-normal responses. In particular, they avoid having to select a single transformation of the data to achieve the possibly conflicting objectives of normality, linearity and homogeneity of variance. Important examples include the binary and the count data. Over the years, GLMs have expanded much in scope and usage, and are currently applied to a very broad range of problems which include analysis of multicategory data, dynamic or statespace extensions of non-normal time series and longitudinal data, discrete time survival data, and non-Gaussian spatial processes.
