Improving the Efficiency of Individualized Designs for the Mixed Logit Choice Model by Including Covariates

Abstract

Recent research shows that the inclusion of choice related demo- and sociographics in discrete choice models aids in modeling the choice behavior of consumers substantially. However, the increase in efficiency gained by accounting for covariates in the design of a choice experiment has thus far only been demonstrated for the conditional logit model. Previous findings are extended by using covariates in the construction of individualized Bayesian D-efficient designs for the mixed logit choice model. A simulation study illustrates how incorporating covariates affecting choice behavior yields more efficient designs and more accurate estimates and predictions at the individual level. Moreover, it is shown that the possible loss in design efficiency and therefore in estimation and prediction accuracy from including choice unrelated respondent characteristics is negligible.

Keywords: covariate, discrete choice experiment, mixed logit choice model, individual efficient design, hierarchical Bayes estimation