ABSTRACT
This chapter focuses on fitting simple models to data using the method of maximum likelihood. In some cases there is evidence that the method of maximum-likelihood produces estimators that are more biased than the method of minimum logit chi-square, but the result is not uniformly true. When a model such as the logit is fitted to quantal assay data the model may be used to summarize the data, through the pair of parameter estimates, and the model may form the basis for comparisons between different sets of data. Fitting the logit or a similar model by maximum likelihood is therefore a straightforward problem of numerical optimization in a small number of variables. Some numerical analysis procedures may be better suited to a particular optimization problem than others, which may even fail to converge. For example, Chambers found the general optimization procedures of Fletcher converged more rapidly than iterated linear least squares in a number of logistic regression examples.
