ABSTRACT
Many of the models that we have considered in this book have been tailored to a particular situation and fitted using general methods of function optimisation. It is important, however, to realise that examples which appear initially to be quite different often share a common structure, which can be exploited to remove much of the labour of non-linear model-fitting. Two apparently different models considered so far are logit models for quantal response data, and multinomial models for contingency tables, which are rectangular tables of counts. These can both be fitted to data using a computerised procedure for fitting generalised linear models (GLMs). Procedures for fitting GLMs may also be used more widely — for example, to fit models to data on survival times.
