Modeling consumers' response sensitivities across categories
The overall objective of this dissertation is to understand the variation in consumers' response sensitivities to marketing mix variables across different product categories. Specifically, our aim is to: (a) attempt to resolve the contradictory evidence in literature on this subject, (b) show that price sensitivity across categories is influenced by the nature of the relationship between categories, and (c) propose a modeling approach that can provide a concise perspective on consumer behavior across a large number of categories.
For this purpose, we first compare several models that vary in their specifications for the covariation in response sensitivities. Specifically, we compare a multivariate probit model with an unrestricted structure for the covariance across utilities and response sensitivities to a variance components approach and an independent probit approach. Next, we impose a factor structure on the covariation in the response sensitivities as a tool to deduce consumer behavior across a large number of categories. This factor decomposition not only provides model based correlation estimates of response coefficients across categories, but also helps in understanding the factors that may be driving these correlations. The models are estimated using Markov Chain Monte Carlo (MCMC) methods such as Gibbs Sampling and Metropolis Hastings Algorithm. Scanner data for six categories are used for estimation. Results from estimation are presented and analyzed.
Our analysis shows that model specification influences how a household's price sensitivity is interpreted-as a household trait or otherwise. We show that incorporating cross-category effects in the structure of the utility functions and their covariation is crucial in deducing the nature of price sensitivity. A key result from our modeling effort is that the nature of correlation in sensitivities depends on the nature of the relationship between any two given categories. The results from our Bayesian factor analytic models indicate the households in our data set do not exhibit any “intrinsic” price sensitivity. Rather, their price sensitivity is driven by goal-derived groups of categories and the nature of the relationship between these categories. Our research confirms that this technique is useful as a data reduction tool.