Experimental design criteria and their behavioural efficiency : an evaluation in the field.

Abstract

Comparative results from an evaluation of inferred attribute non-attendance are provided for experimental designs optimised for three commonly employed statistical criteria, namely: orthogonality, Bayesian D-efficiency and optimal orthogonality in the difference. Survey data are from a choice experiment used to value the conservation of threatened native species in New Zealand’s production forests. In line with recent literature, we argue that attribute non-attendance can be taken as one of the important measures of behavioural efficiency. We focus on how this varies when alternative design criteria are used. Attribute non-attendance is inferred using an approach based on constrained latent classes. Given our proposed criterion to evaluate behavioural efficiency, our data indicate that the Bayesian D-efficiency criterion provides behaviourally more efficient choice tasks compared to the other two criteria.