FITTING AND COMPARING PROBABILITY-DISTRIBUTIONS WITH LOG LINEAR-MODELS

Probability distributions in the exponential family can bc fitted directly as log linear models and the usual maximum likelihood estimates of parameters obtained. If a composite distribution is constructed from several competing candidates, these may also bc compared within a log linear model, and those not acceptable eliminated. The approach also applies to truncated distributions, as well as when independent variables are present (generalized linear models). Here, in the most general case, the different sub-populations may even have different distributions within the same model. With additional iterations. estimation may bc extended to certain models outside the generalized linear framework.