Abstract
A compound Poisson distribution, the Ehrenberg NBD model, has proven useful in the past for analyzing consumer purchase behavior encountered in dealing with frequently purchased low-cost consumer items. The present article proposes the use of a generalized negative binomial distribution in which the NBD model is used as a purchase occasion component and is combined explicitly with the conditional distributions of the single-purchase volume variables associated with the purchase occasions. Conventional and logit regression methods are illustrated for estimating parameters of the single-purchase quantity probability function. This kind of formulation can be useful for conditional trend analyses and similar segmentation studies.