The Dirichlet distribution as a model of brand choice : further testing

This paper provides further evidence on the appropriateness of the Dirichlet model for modeling brand choice. A simple estimation procedure is used and the model is tested on consumer panel data for three product categories (margarine, regular coffee and instant coffee). The results show that a) the model tends to underpredict the proportion of light users of a brand, b) compared to the Hendry model and to the Switching law, it provides an acceptable fit to aggregate switching data over consecutive choice occasions, and c) it gives an adequate description of the correlation between relative frequencies of choice across consumers. Consistently with previous studies, the Hendry model tends to overestimate repeat purchase whereas the switching law, like the Dirichlet model, does not indicate any systematic deviation in its predictions of Switching behavior.