COMPARISON OF QUASI‐LIKELIHOOD MODELS FOR OVERDISPERSION

Methods for modelling overdispersed data are compared. These methods are considered to be of two kinds: a likelihood based approach and a method‐of‐moments based approach. The likelihood method facilitates computation of maximum likelihood estimates which can be obtained through the same algorithm as that of weighted least squares. The quasi‐likelihood or moment approaches seem to be appropriate when severe overdispersion may be present. The comparisons are made via analyses of the Ames Salmonella Reverse Mutagenicity Assay (Margolin et a/., 1981) and a seed dataset (Crow‐der, 1978).