Bayesian and classical estimation of mixed logit: An application to genetic testing

Area of Science:

• Health Economics
• Econometrics
• Behavioral Economics

Background:

• Discrete choice experiments (DCEs) commonly employ mixed logit (MXL) models to analyze preference heterogeneity.
• Classical estimation methods and normal distribution assumptions in MXL can lead to interpretation errors and convergence problems.
• Hierarchical Bayes (HB) offers an alternative estimation approach for MXL models.

Purpose of the Study:

• To compare the performance of Bayesian (HB) and classical estimation approaches for MXL models.
• To investigate potential convergence issues and result validity in health economic DCEs.
• To assess the reliability of willingness to pay estimates derived from different MXL estimation methods.

Main Methods:

• Employed a discrete choice experiment (DCE) to elicit preferences for a genetic technology.
• Applied both classical and Hierarchical Bayes (HB) estimation techniques to the mixed logit (MXL) model.
• Evaluated the convergence properties and economic realism of the results from each estimation method.

Main Results:

• The classical estimation approach yielded unrealistic results and an erroneous willingness to pay estimate in one specification.
• The HB procedure produced consistent and reasonable results across both tested specifications.
• HB estimation identified that the classical approach had converged to a local maximum, not the global optimum.

Conclusions:

• Hierarchical Bayes (HB) is a more robust method for estimating mixed logit (MXL) models in health economics compared to classical approaches.
• HB estimation mitigates convergence issues and enhances the reliability of preference heterogeneity analysis and willingness to pay estimates.
• Researchers should consider HB for DCEs to ensure accurate econometric modeling and valid economic interpretations.