Bayesian Analysis in MarketingA breakthrough in customer analytics

Published January 2010 Updated March 2010 23 lectures
Prof. Greg M. Allenby
Fisher College of Business, Ohio State University, USA
Prof. Peter E. Rossi
Joseph T. and Bernice S. Lewis Professor of Marketing and Statistics, University of Chicago, USA

Modern Bayesian analysis has rapidly diffused into the field of marketing over the last 15 years, and is a widely accepted tool for empirical research in academic and practitioner communities. Bayesian analysis can be successfully applied to a broader array of problems, and affords greater insights from data than other... read moremethods of analysis. Bayesian analysis has been used in the development of models of consideration set formation, models of heterogeneous advertising effects, and models of optimal marketing mix allocation involving simultaneous sets of equations. The behavior associated with these models is difficult to study using conventional methods. Bayesian analysis deals directly with complex aspects of realistic models, while delivering actionable insights from the data.

The biggest inroad made by Bayesian analysis in the practitioner marketing world has been in the area of conjoint analysis. Conjoint analysis is a technique for understanding the drivers of value of product offerings. Offerings are thought to be comprised of benefit bundles delivered through product attributes and attribute levels, and conjoint analysis is a statistical technique for estimating the portion of utility that can be assigned to each of the attributes. Utility portions are referred to as ‘part-worths’, and the challenge addressed by Bayesian analysis is the ability to estimate these part-worths at the respondent level. Once respondent level estimates are available for analysis, segmentation analysis and source of volume calculations from new product entry are made possible. The availability of individual level estimates in conjoint analysis is made possible only through the use of Bayes Theorem.

The goal of this series is to introduce the science and practice of Bayesian analysis in marketing. We begin by reviewing the fundamentals of Bayesian analysis, including its computational arm known as Markov chain Monte Carlo (MCMC) estimation. MCMC estimation is a numerical method by which we obtain random samples from the posterior distribution of a model. We show that this method of estimation is particularly well suited for analyzing hierarchical model structures, which are frequently encountered in marketing. We then explore a series of applications that demonstrate the usefulness of Bayesian analysis in marketing.

Applications (15 Lectures)