ENBIS: European Network for Business and Industrial Statistics

The usefulness of optimal designs for conjoint choice experiments

22 September 2009, 09:00 – 09:45

Abstract

Submitted by

Thomas Svensson

Authors

Martina Vandebroek

Abstract

In marketing, transportation and many other areas, conjoint choice experiments are used for quantifying consumers' preferences for certain characteristics of products or services. Conjoint choice experiments require respondents to choose the alternative they like most in each choice set that is shown to them. Each choice set consists of several hypothetical alternatives that are described as a combination of attribute levels.
The proper design of these experiments is key to an efficient measurement of the partworths which represent the relative importance that consumers attach to each attribute level. Traditionally, researchers have relied upon the use of orthogonal experimental designs to populate the hypothetical choice situations. These orthogonal designs are also called utility-neutral designs in this context because they are optimal under the unrealistic assumption that people have no preference for any of the attribute levels.
Using an extensive case study it is shown that incorporating the limited available information about people's preferences for various product attributes in the design results in experiments that are much more informative for fitting the choice model than the utility-neutral designs. We touch upon several related issues such as the correct specification of the prior knowledge, the selection of the optimality criterion, the effect of choice complexity on design efficiency, the inclusion of a no-choice alternative and the estimation of the corresponding willingness-to-pay.
With the standard choice model, one can model homogeneous markets in which all respondents are assumed to have the same partworths. If one aims at modeling the heterogeneity in the market, the mixed logit model should be used. We conclude this presentation with the results of a recent study in which choice sets are generated for each respondent separately based on previous answers of that particular respondent.

This is joint work with Vishva Danthurebandara, Bradley Jones, Peter Goos, Roselinde Kessels, Bart Vermeulen and Jie Yu.

Return to programme