Predictive Validity of Alternative Aggregation Schemes in Conjoint Analysis

Publication Type

Working Paper

Publication Date

1980

Abstract

In a recent article, Green and Srinivasan reviewed the developments and the state of the art in conjoint analysis (Green and Srinivasan, 1978). The methodologies used in this area have considerable appeal and the various alternative approaches have found widespread use and application (Green and Wind, 1975). Yet relatively little is known about the extent to which the preference models obtained are capable of predicting actual future choices to be made between competing alternatives. A few studies have examined the predictive validity of conjoint analysis—based individualized preference models in the context of choices made before and during data collection (Parker and Srinivasan, 1978). Recently, Montgomery, Wittink and Glaze obtained measures of ax ante predictive validity for individualized job preference models for MBA students at the Graduate School of Business at Stanford University (1977). One of the key elements of this study is that it was possible to examine the behavioral (actual real world choice) predictive validity of conjoint analysis. Further, the students are highly involved in the choice of a first job after graduate school, which should have a favorable effect on their motivation to provide careful judgments about hypothetical job alternatives. The accuracy of the preference models was measured subsequently by predictive testing of future job choice, that is, choices among job alternatives which were not known at the time the student provided his/her preference data. To allow for individual differences in the preference models, the parameters have usually been estimated at the individual level. It is clear, however, that to describe the results to a manager at least some aggregation has to take place. In this paper, we examine the predictive validity for alternative segmentation schemes against some established benchmarks. These benchmarks consist of the greatest amount of individual differentiation possible (i.e., the individualized preference models) on one side, and the least amount of individual differentiation (i.e., parameter estimates averaged across all individuals) on the other side.

Discipline

Marketing

Research Areas

Marketing

Issue

545

First Page

1

Last Page

13

Publisher

Stanford University, Graduate School of Business

City or Country

Stanford, CA

This document is currently not available here.

Share

COinS