Incorporating self-reported health measures in risk equalization through constrained regression

Area of Science:

• Health economics
• Insurance market regulation
• Statistical modeling

Background:

• Health insurance markets use risk equalization to manage spending variations.
• Existing systems often undercompensate individuals with poor self-reported health, creating risk selection incentives.
• Self-reported health is typically excluded from risk adjustment due to feasibility concerns.

Purpose of the Study:

• To investigate constrained regression (CR) as an alternative method for incorporating self-reported health into risk equalization models.
• To assess the impact of CR on under/overcompensation for different population groups compared to traditional methods.
• To evaluate the effectiveness of CR in mitigating risk selection incentives in health insurance.

Main Methods:

• Utilized large-scale administrative (N=17 million) and health survey (N=380,000) data from the Netherlands.
• Estimated five constrained regression (CR) models with varying degrees of coefficient restriction.
• Compared CR models against the 2016 Dutch ordinary least squares (OLS) risk equalization model.

Main Results:

• Constrained regression improved outcomes for groups not explicitly adjusted for but worsened outcomes for explicitly adjusted groups.
• Lighter constraints in CR models demonstrated superior performance over OLS based on a novel metric summarizing under/overcompensation.
• The study quantified the reduction in under/overcompensation for self-reported general health groups across different CR constraint levels (20% to 100%).

Conclusions:

• Constrained regression presents a viable strategy for integrating self-reported health into risk equalization, addressing limitations of current models.
• The findings suggest that carefully calibrated constraints can enhance the fairness of risk equalization systems.
• CR offers a promising avenue for reducing insurer incentives for risk selection by better accounting for unobserved health status variations.