A Break from the Norm? Parametric Representations of Preference Heterogeneity for Discrete Choice Models in Health

Area of Science:

• Health Economics
• Econometrics
• Behavioral Science

Background:

• Discrete choice models analyze individual preferences.
• Mixed logit models are standard in health economics for preference heterogeneity.
• Current practices often rely on normal distributions for these models.

Purpose of the Study:

• To evaluate the impact of alternative distributional assumptions in mixed logit models.
• To compare standard normal distributions with other parametric specifications.
• To assess the benefits of model averaging for preference heterogeneity.

Main Methods:

• A scoping review of mixed logit modeling practices in health.
• Comparison of seven alternative distributions and model averaging.
• Analysis across four datasets: stated preference, revealed preference, and simulated.
• Evaluation of model fit, preference distributions, willingness to pay, and forecasting.

Main Results:

• Alternative distributional assumptions significantly outperformed standard normal distributions.
• Preference distributions and willingness to pay estimates varied substantially across models.
• Model averaging provided greater flexibility, improved fit, and mitigated selection bias.
• Distributional assumptions did not impact model predictions.

Conclusions:

• The standard practice of using normal distributions is suboptimal for capturing preference heterogeneity.
• Researchers should explore alternative distributions beyond the normal assumption.
• Model averaging is a valuable approach for robust preference analysis in health economics.