Carnegie Mellon University

Bayesian Statistics

Course Number: 47747

Bayesian methods are widely recognized for the unified approach that they offer to statistical analysis. Additionally, these methods are readily amenable to decision problems which are frequently found in Marketing problems. The growth of Bayesian methods in the last ten years is largely due to a new class of computational methods to solve previously intractable problems. These methods rely upon simulation techniques, particularly the use of the Markov Chain Monte Carlo (MCMC). These new techniques have allowed applied researchers and practitioners to focus more on solving new problems instead of the methodology. This has opened up new areas applied research to Bayesian methods. The focus of this course is applications of Bayesian statistical analysis to one of these areas, Marketing. We begin with a discussion of the Bayesian approach and follow this with the general linear model with special emphasis on hierarchical models. Another major focus is discrete choice modeling.

Degree: PhD
Concentration: Marketing
Academic Year: 2022-2023
Semester(s): Mini 1
Required/Elective: Elective
Units: 6


Lecture: 100min/wk and Recitation: 50min/wk

Learning Objectives

The goals of this course are: Expose students to the Bayesian approach. The Bayesian method can be characterized by the understanding of probabilities as subjective measures, priors that summarize existing knowledge, and the use of Bayes rule to compute the posterior distribution from the observed data and prior. Show how MCMC methods can be applied to compute the marginal posterior distribution. Specifically we will use R functions to construct MCMC algorithms. Moreover to improve the application of these techniques we illustrate how to make inferences using JAGS/Nimble/Stan. Illustrate the application of Bayesian methods to applied problems, especially those in marketing but more generally to problems in business and social sciences. Specifically, we will consider hierarchical linear models, truncated normals, discrete choice models, decision theoretic problems, and non-parametric models.