Disentangling Preferences, Inertia and Learning in Brand Choice Models

The forces that influence a consumer in brand choice: preferences, experiences, and marketing mix have been of great interest to marketing scientists. The vast literature based on static survey data points to consumers using multi-attribute utilities in making a choice. At the same time the vast literature of scanner panel data based studies points to the existence of state dependence (in various functional forms) in brand choices in frequently purchased product markets. Using a unique dataset that contains stated preferences (survey) and actual purchase data (scanner panel) for the same group of consumers we attempt to untangle the effects of preference heterogeneity and state dependence, and to determine the exact nature of the latter. We propose a hierarchical model in which consumers within the same product category are heterogeneous in the order of the brand choice process and its parameters as well as in their preferences and responsiveness to marketing mix. The proposed model is designed to encompass side by side three different types of consumer experience based behavior: zero-order, inertia and learning. Two sources of state dependence, inertia and learning, are operationalized using a dummy for lagged choice and adopting the Bayesian learning process, respectively. We apply a Reversible Jump MCMC sampling scheme to sample across component processes and a Metropolis-Hastings/Gibbs step within each component process. Our results are striking and suggest that all three processes exist but that the extent of state dependence is spuriously overestimated in the absence of preference information. Both the inertia and in particular the Bayesian learning components are overstated, and the nature of learning significantly changes. A substantial segment of consumers who do not exhibit state dependence is uncovered and its size increases when preferences are available. Using individual-level outcomes we illustrate why the lack of preference information leads to faulty inferences. Furthermore, we find that sensitivity to marketing mix variables is biased. We also conduct various counterfactual simulations to assess the managerial implications of our findings.