Academic journal article Academy of Marketing Studies Journal

A Comparative Study of Two Approaches to Modeling Consumer Preferences

Academic journal article Academy of Marketing Studies Journal

A Comparative Study of Two Approaches to Modeling Consumer Preferences

Article excerpt

INTRODUCTION

Consumer researchers have proposed various approaches to coping with large numbers of product attributes and scenarios in multiattribute choice modeling (e.g., Green & Srinivasan, 1990; Oppewal, Louviere & Timmermans, 1994). The most notable development is the hybrid conjoint analysis (HC), which reduces task sizes for a compensatory approach to decision making (Green & Srinivasan, 1990). This hybrid approach plays an important role in designing new products entailing large amounts of product information (e.g., Wind et al., 1989).

In contrast, consumer behaviorists have suggested that decision makers tend to use simplifying tactics when considering large amounts of product information (e.g., Johnson, 1988; Johnson & Meyer, 1984; Meyer & Kahn, 1991; Wright, 1975). Consequently, researchers have suggested using non-compensatory hierarchical models purported to be better representations of consumer decision processes (Currim, Meyer & Le, 1988; Fader & McAlister, 1990; Gensch & Ghose, 1992; Gensch & Javalgi, 1987). A study shows that consumers use both approaches in making choice decisions (Lee & Geistfeld, 1998). The issue: Which of these two approaches provides a more accurate prediction of consumer preferences?

The two approaches to preference modeling differ in assumptions, theories of decision-making strategies, and choice processes. This paper discusses these differences and reports an empirical test of their predictive performance. We examine the predictive accuracy of selected models in both approaches at the individual consumer level. We compare model performance in predicting product versus service choices of the same individuals for two different choice sets.

MAXIMUM LIKELIHOOD HIERARCHICAL MODEL

Applications of most hierarchical models have been limited because of their assumptions about product attributes. For instance, applications of the Elimination-By-Aspects (EBA) model (Tversky, 1972) have been inhibited because the model assumes only a few attributes common to all alternatives. The EBA model creates further difficulties by requiring a large number of parameters common to only subsets of the choice set (hence, for N alternatives, EBA requires 2n -3 parameters or attributes)(Batsell & Polking, 1985). A recent successful application of EBA handled only a single product attribute common to all alternatives (Fader & McAlister, 1990).

Similarly, the Preference Tree (also known as PRETREE or Elimination-By-Tree) model reduces the number of parameters required by the EBA by imposing a strict inclusion rule on product attributes (Tversky & Sattath, 1982). Another tree-structured model proposed by Currim, Meyer, and Le (1988) does not accommodate monotonicity of attribute levels and forces product attributes into dichotomous variables (Green, Kim & Shandler, 1988).

On the other hand, the Maximum Likelihood Hierarchical Model(MLH)is a probabilistic version of its predecessor HIARC, which is also a multiattribute disaggregate lexicographic model (Gensch & Svestka, 1979). As a probabilistic model, the MLH derives maximum likelihood estimates of attribute cut-offs in order to reduce the sensitivity to sampling errors found in its predecessor (Gensch & Svestka, 1984).

Both models handle a large number of product attributes in their prototype applications (i.e., 10 attributes in HIARC [Gensch & Svestka, 1979]; 16, 21 and 24 attributes in MLH [Gensch & Svestka, 1984]). Both models simulate the hierarchical elimination approach to choice processes proposed by Tversky (1972); yet, they differ from Tversky's models such as Elimination-By-Aspects (Tversky, 1972) and Preference Tree (Tversky & Sattath, 1979) by assuming common attributes for alternatives in a choice set. This assumption of the MLH is shared by the hybrid conjoint models, thus enabling inter-class model comparisons based on the same data. …

Search by... Author
Show... All Results Primary Sources Peer-reviewed

Oops!

An unknown error has occurred. Please click the button below to reload the page. If the problem persists, please try again in a little while.