Comparing Designs Constructed With and Without Priors for Choice Experiments: A Case Study

Abstract

This article describes the second stage of an empirical comparison of the performance of designs for a discrete choice experiment. Six designs were chosen to represent the range of construction techniques that are currently popular for choice experiments, with some of the designs incorporating into the design generation process prior knowledge of the parameters gained from the previous stage of this experiment. Each design had 320 respondents, each of whom completed 16 choice sets. The results indicate that efficient designs constructed using several different strategies all identify various types of heterogeneity with similar levels of precision. Specifying the right model to best describe the underlying preferences of respondents in each sample may then become the limiting factor in the estimation of more complex generalized multinomial models, rather than the design per se.

AMS Subject Classification:

• 62K05
• 62P20

Keywords:

• MNL model
• G-MNL model
• Mixed logit
• Stated preference experiments

Additional information

Funding

The work described in this article was partially supported by the Australian Research Council, grant DP0774142.