Nelder and McCullagh (1972) describe a class of Generalized Linear Models (GLMs) that extends linear regression to permit non-normal stochastic and non-linear systematic components. GLMs encompass a broad and empirically useful range of specifications that includes linear regression, logistic and probit analysis, and Poisson models.

GLMs offer a common framework in which we may place all of these specification, facilitating development of broadly applicable tools for estimation and inference. In addition, the GLM framework encourages the relaxation of distributional assumptions associated with these models, motivating development of robust quasi-maximum likelihood (QML) estimators and robust covariance estimators for use in these settings.

The following discussion offers an overview of GLMs and describes the basics of estimating and working with GLMs in EViews. Those wishing additional background and technical information are encouraged to consult one of the many excellent summaries that are available (McCullagh and Nelder 1989, Hardin and Hilbe 2007, Agresti 1990).