Designing Conjoint Choice Experiments Using Managers' Prior Beliefs

TLDR

The authors propose eliciting managers’ prior information and demonstrate via Monte Carlo studies that this approach yields more efficient conjoint designs than existing methods. They employ a Bayesian design that optimizes over a prior distribution of parameter values and apply it empirically by eliciting managers’ prior information and uncertainty. The Bayesian design achieves 30–50% lower standard errors and about 20% higher predictive validity, producing more efficient conjoint choice experiments than current procedures.

Abstract

The authors provide more efficient designs for conjoint choice experiments based on prior information elicited from managers about the parameters and their associated uncertainty. The authors use a Bayesian design procedure that assumes a prior distribution of likely parameter values and optimizes the design over that distribution. The authors propose a way to elicit prior information from managers and show in Monte Carlo studies that the procedure provides more efficient designs than the current procedures. The authors provide an empirical application, in which they elicit prior information on the parameter values and the associated uncertainty from managers. Here, the Bayesian design provides 30%-50% lower standard errors of the estimates and an expected predictive validity that is approximately 20% higher.