DESIGNING CHOICE EXPERIMENTS WITH MANY ATTRIBUTES. AN APPLICATION TO SETTING PRIORITIES FOR ORTHOPAEDIC WAITING LISTS

Abstract

The aim of this paper is to undertake a discrete choice experiment using a 'blocked attribute' design. To date in the health economics literature, most discrete choice experiments have used only a relatively small number of attributes due to concerns about task complexity, non-compensatory decision rules, simplicity of experimental designs, and the costs of surveys. This may lead to omitted variable bias and reduced explanatory power when attributes have been pre-selected from a longer list. There may be situations where it is desirable to include a longer list of attributes, such as attaching weights to quality-of-life instruments to obtain single index scores. We examine this issue in the context of attaching weights to a disease-specific quality-of-life instrument used to prioritise patients on orthopaedic waiting lists in Victorian hospitals. Eleven attributes are allocated across three separate experimental designs and the data pooled for analysis. Pooling is justified given the specific context of the study, including attempts to minimise the effect of unobserved heterogeneity across the three models when designing the study and collecting data. Blocked attribute designs may offer flexibility to researchers when it is not possible or desirable to reduce the number of attributes.