ABSTRACT
Nelder and Wedderburn (1972) introduced the term “generalized linear models”
(GLM) when they extended the scoring method for maximum likelihood estima-
tion (ML) for normal data to exponential family models. This allowed for a unified
approach for analysis that included linear, logistic, and Poisson regression, among
others, as special cases. Wedderburn (1974) further generalized GLM by noting that
key components of a GLM involved the first 2 moments of the distribution of the out-
come variable, which could be used to define a quasi-likelihood that could be used
to generalize the assumption of an exponential family for GLM.