ABSTRACT
This chapter examines log-linear models, relate them to logistic regression models and shows how to use log-linear models to develop prediction models for factors that have more than two possible outcomes. Log-linear models are more often used for exploring relationships between factors than for prediction. Log-linear models are more often used to model independence relationships between factors. Consider a log-linear model for a two-dimensional table that involves the use of a continuous predictor variable x to model interaction. Another approach is to realize, as in the previous section, that logistic models are equivalent to log-linear models and to use log-linear models to deal with multiple response categories. The log-linear model can be used directly to fit multiple logit models that address specific issues related to the multinomial responses. Standard sampling schemes for count data are multinomial sampling and Poisson sampling.
