ABSTRACT
This chapter examines the difficulties and limitations of linear models. Generalized linear models (GLMs), is not particularly broad from a mathematical point of view, but it is sufficiently flexible to include a large number of practical important subcases. The GLM class has the advantage of allowing a unified treatment for a set of relevant specific models, which in the past were introduced separately from each other, answering quite distinct practical needs, but which now appear as specific instances of the same approach. One of the reasons that contributed to the success of GLMs is the possibility of using a single algorithm for solving, with little adjustment to be made for the choice of the link function and the type of probability distribution. Since the logarithmic function is the canonical link for the Poisson distribution, it is natural that log-linear models are most commonly used for the analysis of frequency tables.
