| The unique contribution of this work comes from the intersection of the Bayesian network literature and the marketing modelling literature. In spite of their apparently attractive features for solving various marketing problems, Bayesian networks are still not a well-recognized technique within the marketing community. This lack of recognition can be attributed to the fact that the methodology is still in the early stages of its maturity with respect to specific requirements of marketing research, and there is a lack of a thorough discussion of basic features and potential added value of the Bayesian network technology as a tool in the arsenal a marketing researcher. Moreover, little attention has been paid to date on adapting or evaluating Bayesian networks as a potential technique for conducting theoretical research, let alone marketing research.; Motivated with these observations, the overall goal of this thesis is to provide a critical evaluation of the application of Bayesian networks in theoretical and practical Customer Satisfaction and Loyalty (CS&L) research, and propose new methods and developments within the Bayesian network modelling to improve its current abilities with respect to specific requirements existing in the CS&L research. The critical evaluation that the author undertakes should be regarded as internal validation rather than external one; consequently, the Bayesian network approach is considered in this thesis merely as another approach that can help understand and research the CS&L phenomenon.; More specifically, the thesis has several sub-objectives. Firstly, it is investigated how marketing theories can be discovered (developed) by means of the Bayesian networks approach taking in two routes: inductivist and deductivist. The author proposes and evaluates new methods for handling structural and measurement models, in particular aiming at accounting for the measurement model, latent construct validation, and finding the best cardinality of latent constructs. Furthermore, specific issues in theory development, such as the ability for modelling of moderating effects, and the issue of accounting for mediating variables are examined and discussed. In the context of scientific discovery the author considers to what extent purported marketing theories discovered with Bayesian networks are subject to scientific justification, and how they can be scientifically justified (validated); hence, he evaluates descriptive, predictive and explanatory potential of Bayesian networks on the example of the e-satisfaction and loyalty domain. The author investigates what could be the added value of modelling marketing problems with Bayesian networks. He proposes in this respect the ability of performing probabilistic reasoning (forward, backward, inter-causal) in the domain, the potential of performing what-if simulations, and the potential of combination of prior knowledge with data. All these issues are illustrated with examples. With regard to the practical CS research, it is investigated how can Bayesian networks be applied in service feature/dimension importance/performance study. Finally, the author discusses the strengths and weaknesses of Bayesian networks in terms of specific technical and statistical modelling issues, such as data distributional assumptions, missing data handling, etc. |
