Incorporating demographics into discrete choice analyses

Robert E. Carter

University of Louisville

Introduction

Discrete choice studies have been utilised extensively over the past three decades and applied to a broad range of experimental situations (McFadden 1974; Louviere & Woodworth 1983; Louviere et al. 2000; Sapede & Girod 2002). More specifically, Haaijer et al. (2001, p. 93) indicate that the popularity of discrete choice experiments is due to their ‘ability to mimic real market decisions’ on the part of consumers since the dependent variable for these types of research studies is consumer choice, as opposed to an overall rating or other evaluation measure. While discrete choice experiments have shown their value in terms of market share prediction, these approaches have not proved particularly helpful in terms of identifying or highlighting target audiences. That is, these methods have not typically been employed for identifying target segments for products and services, based on demographics. This is a real weakness given the nature of contemporary marketing (i.e. evolving from ‘broadcasting’ to ‘narrowcasting’) and the corresponding proliferation of cable channels and websites targeting relatively small audiences. Discrete choice experiments usually do not include demographics variables, or characteristics of individual respondents, because these components do not vary across the different alternatives in each choice scenario (Haaijer et al. 2001; Montgomery 2002; Train 2003). The primary objective of the current research is to address this methodological shortcoming and share a practical procedure that is designed to incorporate demographic variables in discrete choice experiments and related multinomial logit models. The prior research provides some guidance on this issue in theoretical terms (Hoffman & Duncan 1988), or is somewhat out of date (So & Kuhfeld 1995); however, the goal of the current study is to provide practical and up-to-date guidance on this important issue. In support of this objective, we use a classic transportation example (Daganzo 1979; SAS Version 9.1) that has been modified to include gender as a demographic variable.