Survey-design and analytical strategies for better healthcare stated-choice studies
Stated-choice (SC) surveys, such as conjoint analysis, present some interesting problems for researchers that are not addressed in the traditional survey-development literature. While the constraints imposed by preference theory, the experimental design of the choice sets, and the statistical methods used to analyze choice data all pose challenges for researchers new to SC methods, they also direct such researchers towards techniques that are not possible with more traditional survey methods. In this article, we focus on issues of preference heterogeneity (variation in preferences across subjects by observable and non-observable co-variates) and attribute dominance to illustrate the synergistic roles that survey-design and analytical strategies play in SC research. In this article, we show how advanced analytical techniques are likely to be more important than survey design solutions when addressing preference heterogeneity. Good practice supports the use of mixed-logit and similar modeling approaches to mitigate the problem of unobserved preference or variance heterogeneity. However, if the sample size is not large enough or the survey instrument does not contain questions about important subject characteristics, then the source of heterogeneity cannot be identified and the problems caused by heterogeneity will be magnified.
In contrast, minimizing attribute dominance and testing for attribute dominance relies on careful survey design, rather than more complex analysis. In general, survey design needs careful attention from researchers. No amount of complex analysis can compensate for a poor survey design that can generate only flawed SC data.