A Prior Distribution of Bayesian Nonparametrics Incorporating Distances of Various Types

• Activity Number: 349
• Type: Contributed
• Date/Time: Tuesday, August 3, 2010 : 10:30 AM to 12:20 PM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Brian M. Hartman*+ and David Dahl and Bani K. Mallick and Debabrata (Debu) Talukdar
• Companies: Texas A&M University and Texas A&M University and Texas A&M University and State University of New York at Buffalo
• Keywords: Bayesian Nonparametrics ; New Product Diffusion ; Dirichlet Process ; Clustering ; International Marketing

Abstract

We propose a prior distribution for Bayesian nonparametric modeling where, for each observation, the amount of information borrowed from other observations is influenced by several distance measures. We demonstrate various types of distance measures, including distances defined on covariates of mixed types. Our framework allows the model to learn about the important distance measures and eliminate unimportant measures. We demonstrate the proposed methodology in a Bass model for new product diffusion across many nations. We borrow diffusion information across nations, where the amount of borrowing is specific to each nation and depends on its similarity to other nations. The distances measures that we consider include the bilateral trade and tourism flow, the distance between population centroids, and covariates describing the country's culture.