Obtaining more information from conjoint experiments by best–worst choices

Conjoint choice experiments elicit individuals’ preferences for the attributes of a good by asking respondents to indicate repeatedly their most preferred alternative in a number of choice sets. However, conjoint choice experiments can be used to obtain more information than that revealed by the individuals’ single best choices. A way to obtain extra information is by means of best–worst choice experiments in which respondents are asked to indicate not only their most preferred alternative but also their least preferred one in each choice set. To create DD-optimal designs for these experiments, an expression for the Fisher information matrix for the maximum-difference model is developed. Semi-Bayesian DD-optimal best–worst choice designs are derived and compared with commonly used design strategies in marketing in terms of the DD-optimality criterion and prediction accuracy. Finally, it is shown that best–worst choice experiments yield considerably more information than choice experiments.