Experimental choice analysis: Introduction and overview

Abstract

Experimental choice analysis is a natural extension to econometric and psychometric analyses of observational choice data. Real markets frequently exhibit limited ranges of variation in behaviorally or managerially important variables. As well, new products or services may contain features or enhancements not available in past choice data. Some new products or services pioneer new categories and offer new benefits not previously available in past market data. All too frequently explanatory variables in real market data are highly collinear, measured imprecisely (or even incorrectly), or subject to a variety of statistical nasties that lurk in real data, which set statistical traps to catch even the most alert and adept analysts unaware and unprepared. Thus, despite the obvious and immediate appeal of external validity exhibited by observational choice data, there often are compelling reasons why analysts might want to collect experimental choice data. Towards this end, the papers in this special issue illustrate some, but by no means all, of the ways that one might go about conducting, modeling, and applying the results of choice experiments. The experimental analysis of choice behavior is not new. For example, Thurstone (1927) was the first of many to model individual and group choice behavior. What is new are relatively recent developments in behavioral theory and methods for modeling mulfinomial, as distinct from binomial, choices. Paired comparison experiments and their numerous derivatives often serve (d) as frameworks for collecting choice data. Yet, only recently have optimal designs for such experiments been derived (van Berkum, 1987). Although optimal designs for multiple choice experiments remain elusive, progress has been made (Louviere and Woodworth, 1983; Louviere and Gaeth, 1988; Louviere, 1988a,b; Batsell and Louviere, 1991; Wiley and Anderson, 1990), and there is reason to be optimistic that practical, relatively efficient designs soon will be available for a wide variety of experimental conditions. Although popular and relatively easy to implement, paired comparison choice experiments have limited use for modeling and understanding multiple choice behavior. Their primary limitation is theoretical: it is difficult to generalize to multiple