Copula bivariate probit models: with an application to medical expenditures.

Area of Science:

• Econometrics
• Biostatistics
• Health Economics

Background:

• The bivariate probit model is standard for analyzing binary outcomes with endogenous binary regressors.
• Existing models often assume normal dependence, which may not reflect complex health-related behaviors.
• Accurate modeling is crucial for understanding factors like insurance status on healthcare utilization.

Purpose of the Study:

• To introduce a modified bivariate probit model incorporating copulas for non-normal dependence structures.
• To assess the performance of this copula-based approach in a health economics context.
• To compare the copula bivariate probit model against the standard bivariate probit model.

Main Methods:

• Developed a bivariate probit model framework allowing for flexible dependence structures via copulas.
• Utilized the Frank copula to model the dependence between insurance status and ambulatory healthcare expenditure.
• Applied the models to a dataset examining the effect of insurance on healthcare utilization.

Main Results:

• The copula bivariate probit model, specifically using the Frank copula, demonstrated superior performance.
• The proposed model outperformed the standard bivariate probit model in predicting healthcare expenditure patterns.
• This indicates that non-normal dependence structures are important in this application.

Conclusions:

• Copula-based bivariate probit models offer a valuable extension to standard methods for health economics research.
• Accounting for non-normal dependence improves the estimation of treatment effects on binary health outcomes.
• The Frank copula provides a suitable approach for modeling insurance status and healthcare expenditure relationships.