Using Priors in Choice Designs

Abstract

In the previous chapter we saw that the efficiency in choice designs depends on an (unknown) parameter vector. Therefore, generating an efficient choice design requires bringing in prior estimates of that parameter vector. In this chapter, we develop methods incorporating prior estimates of ß in order to generate more efficient designs. Our prior β’s are point estimates of the true β’s and the resultant designs will be more efficient in the region of those priors. We show that our methods produce designs that are 10–50% more efficient than corresponding designs falsely assuming that β=0, and that these efficiency gains are quite robust against misspecifications of priors. The efficiency gains come from making alternatives within choice sets closer in utility, i.e., the choice becomes more difficult to make. We therefore call these designs utility-balanced choice designs (cf. Huber and Zwerina 1996).