Constructing Efficient Choice Experiments

Abstract

Research on the construction of efficient designs for stated choice (SC) experiments has been limited to either unlabeled experiments with generic parameter estimates or labeled experiments with alternative specific parameter estimates. Designs combining both generic and alternative specific parameters have not yet been addressed. In this paper, by deriving the asymptotic (co)variance matrix for the most general MNL model, the authors are able to demonstrate how efficient experiments that allow for the estimation of both types of estimates may be generated. The authors go onto show how estimation of the asymptotic (co)variance matrix may also be used to determine sample size requirements for SC experiments.