PRM98 Optimizing Discrete Choice Experiment Designs Using Simulations

Abstract

Discrete choice experiments (DCE) range prominently among the applied methods to elicit preferences in the field of health economics. With ongoing methodological learning, best practice remains a moving target. It is seldom plausible to implement a full factorial design and while there have been developed software tools to improve design efficiency there remains “design error”. The aim of this study was to illustrate how simulation studies can inform the designing of DCE's and minimize design error given study size constraints and prior knowledge on preferences. We specified a hypothetical set of attributes and levels for a DCE game as well as an expected linear additive utility function for individuals. We used Monte Carlo simulations - programmed in SAS 9.2 - to examine how different design decisions affected design error given the specified utility function, attributes and levels. Using blocking to increase choice sets minimizes design error. Maximizing the number of respondents may improve estimation but will not markedly improve design error unless used to include more choice sets. Prior knowledge – either theoretical or from prior studies – can be used to deselect choice sets that are implausible or with none or limited informational gain improving design efficiency. Simulations can provide a tool for optimizing design choices. We illustrate how it can supplement software design routines and provide an intuitive understanding of design properties and how it will likely affect the design efficiency to e.g. include more respondents, blocks or questions or use prior knowledge. Simulations are not reality – the “respondents” behave the way they are specified to behave. However, this methodology enables the researcher to isolate effects and we believe that our simulation framework can be a useful tool for practitioners to think systematically about DCE design decisions given actual study characteristics and constraints.