Models for Heterogeneous Variable Selection

Abstract

Preference heterogeneity is a major research stream in marketing aimed at quantifying and understanding the diversity of demand for product attributes and attribute levels. In experimental settings, in which consumers are presented with simple descriptions of product offerings, continuous distributions of heterogeneity, such as the multivariate normal, provide a useful representation of preference. However, in more complex cases in which respondents have value for only a few of the benefits associated with an offering or cognitive constraints that result in selective attention to a subset of the information available, continuous distributions of heterogeneity do not reflect the possibility that a subset of the variables has nonzero effect sizes for different respondents. Identifying which attributes are used in a brand choice decision is closely related to the statistical procedure of variable selection. This article extends variable selection methods to accommodate heterogeneity across consumers and data contexts, conditions frequently encountered in marketing studies. The authors apply the methods to a discrete-choice conjoint study in which data are collected in both full-profile and partial-profile formats.