Maximum Entropy Dirichlet Modeling of Categorical Data With Application to Consumer Choice

We use the Maximum Entropy Dirichlet (MED) procedure to model consumer choice of long distance provider based on the perceived attributes of the companies. The MED is a computer-intensive method that uses Dirichlet prior and various attribute constraints as inputs and provides maximum entropy models that are in loglinear and logit forms. The MED generates prior and posterior distributions for the parameters of each model and for a Kullback-Leibler information function that measures the fit of the model. The MED also provides posterior distribution for inference about a normalized Kullback-Leibler information index of fit.

Keywords:

Contingency tables , Kullback-Leibler information , Logit , Logistic regression , Loglinear , Noninformative prior.