Using Check-all-that-apply Surveys to Support Product Development

• Activity Number: 203
• Type: Invited
• Date/Time: Tuesday, August 10, 2004 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics and Marketing
• Author(s): Mark Beltramo*+ and R. Jean Ruth
• Companies: General Motors Corporation and General Motors Corporation
• Keywords: Markov chain Monte Carlo ; censored data ; consumer surveys ; hierarchical Bayes

Abstract

Information for tracking purchasing behavior and consumers' product experiences is often collected using "check all that apply" survey batteries. Examples include sources of information, reasons for making (or not making) a purchase, and important product attributes. In product development, we are often interested in both the frequency with which each item is checked and the importance of each item to consumers. While estimating frequencies for the items is trivial, estimating each item's importance can be problematic if the list is long and the percentage of consumers checking any given item is relatively small. In addition, the importance of each item varies among consumers, so it is preferable to estimate the distribution of importance placed on each item. We describe an approach to this problem that uses two additional pieces of information from each consumer: which item is most important and a rating of how important it is (measured on an appropriate scale). We present an MCMC sampler based on a hierarchical Bayes formulation of the problem. The formulation allows the estimated importance of each item to borrow strength from data on different, but related, products.