GLMs provide a valuable alternative to the general linear model for dealing with data that do not adhere to the assumption of normal errors and constant variance. In this chapter, we have illustrated the use of GLMs for data that include response variables that:
follow a Binomial distribution (represented as either relative counts or proportions), or
follow a Poisson distribution, which is usually encountered when the response is a count.
Such data are common in the biological sciences, and GLMs, relying on maximum likelihood estimation, provide a flexible framework for modeling such data, estimating effect sizes, and testing hypotheses. When the response variable appears to be over-dispersed, care should be taken to ensure that important predictors have not been overlooked. Once this has been addressed, quasi-likelihood models provide a mechanism for modeling even overdispersed data.