This dissertation analyses the statistical and theoretical basis of Goodhardt, Ehrenberg and Chatfield's (1984) NBD-Dirichlet model of purchase incidence and brand choice. In doing so it makes a contribution to the understanding and use of this model specifically and to gamma models in general. / Particular attention is devoted to the mixing distributions of the NBD-Dirichlet model. It takes the oft-mentioned and little understood random variables of the model and exploits these as the principal tool of understanding. / The value of this approach is demonstrated by the proposition that all of the NBD-Dirichlet model parameters (A, K, S and the g brand α) can be comprehended to such a degree that a reader will be able to anticipate values of these parameters for different categories prior to fitting the model, and further, that it is possible to develop empirical generalisations relating to these parameters. / A deeper understanding of this model is, therefore, achieved through examining the gamma distribution. By understanding this continuous distribution, its parameters and random variables, the NBD-Dirichlet model and its parameters may be understood. The starting point is the random variable. Properly understanding the random variable of the NBD-Dirichlet model's mixing distributions in the marketing context is integral to this dissertation. / This dissertation expands the conceptualisation of the random variable briefly mentioned in the literature for the negative binomial distribution and also extends it to the underlying random variable for the Dirichlet multinomial distribution, hence resulting in a conceptually identical random variable for both brand choice and purchase incidence. These random variables are termed selection rates, and are latent, with each shopper having a 'latent selection rate' for every brand in the category and an additional, and independent, latent selection rate for the category itself. Thus in a g brand category, each shopper has g + 1 latent selection rates. In the NBD-Dirichlet model, latent selection rates are distributed gamma over the population. / Nine propositions regarding the NBD-Dirichlet model parameters in fast moving consumer goods (hereafter FMCG) categories are derived from latent selection rate theory. A number of these propositions are supported either directly or indirectly by the literature, with a tenth proposition arising solely from this second source. Observations of parameters in 50 FMCG categories are obtained in order to test these propositions. The empirical results demonstrate the utility of the theory of latent selection rates and in addition develop a number of generalisations relating to the NBD-Dirichlet parameters. The principal result of this dissertation is that the theory of latent selection rates enables the parameters of the NBD-Dirichlet model to be anticipated for a given FMCG category. / The notion of latent selection rates and describing the underlying random variables of the NBD-Dirichlet model is not unique to this dissertation, however using them as a tool for understanding, and anticipating the parameters of the model is. This dissertation thus enhances the practical and theoretical usefulness of the NBD-Dirichlet model by focusing on its parameters which fully specify any stationary and non-partitioned category. / This is achieved by showing that the parameter values of the NBD-Dirichlet model for a given category may be anticipated simply through a basic knowledge of the category and a theoretical knowledge of the model; specifically the theory of latent selection rates. The theory thus enables a greater comprehension of the NBD-Dirichlet model and has significant implications for both academic research and everyday managerial decision making. / Thesis (PhD)--University of South Australia, 2005.
| Identifer | oai:union.ndltd.org:ADTP/267275 |
| Creators | Driesener, Carl Barrie. |
| Source Sets | Australiasian Digital Theses Program |
| Language | English |
| Detected Language | English |
| Rights | copyright under review |
Page generated in 0.0056 seconds