For Better or for Worse: A Hierarchical Bayes Model for Partial Preference Rankings from Discrete Choice Experiments

• Activity Number: 38
• Type: Contributed
• Date/Time: Sunday, August 4, 2013 : 2:00 PM to 3:50 PM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Anna Liza Antonio*+ and Catherine Crespi and Robert E Weiss and Christopher Saigal
• Companies: UCLA and UCLA and University of California, Los Angeles and UCLA
• Keywords: Bayesian estimation ; choice modeling ; best-worst choice experiments ; multiattribute options ; conjoint analysis

Abstract

In health care, discrete choice experiments (DCEs) have been used to study how individuals value specific attributes of health states such as expected lifespan and physical functioning. Individuals are presented with sets of health states ("items") with various attributes and asked to make choices; these choice data are used to estimate subject specific regression coefficients (called marginal utilities or partworths) of the attributes. In many DCEs, a full ranking of a set of items is obtained by eliciting best and worst choices, first in the entire set and then in each successive subset of unranked items. In the best-worst DCE we consider, only best and worst choices are elicited for sets of at least four items, which reduces respondent burden but yields only a partial ranking of items. We discuss a hierarchical Bayes random-effects model in which we account for missing ranking information by marginalizing over all possible permutations of unranked items. The model is applied to a study of preferences for health state attributes among patients biopsied for prostate cancer.