Quantifying and reducing inequity in average treatment effect estimation.

Area of Science:

• Statistics
• Health Equity
• Epidemiology

Background:

• Systemic disparities in sample representation across studies can lead to inequitable generalization of average treatment effects.
• Underrepresented subgroups may experience biased or less reliable treatment effect estimates.

Purpose of the Study:

• To develop a framework for quantifying representation inequity in studies.
• To propose a data analysis method for mitigating disparities in sample representation.
• To improve the generalizability and equity of average treatment effect (ATE) estimates.

Main Methods:

• Developed a framework to quantify inequity from sample representation disparities.
• Proposed a method for estimating ATE in representation-adjusted samples, allowing subgroups to leverage full sample data.
• Offered two representation adjustment approaches: minimizing subgroup mean-squared error (MSE) and balancing MSE with equal representation.
• Conducted simulation studies comparing proposed estimators to subgroup-specific estimators.

Main Results:

• The proposed estimators demonstrated lower mean squared error (MSE) compared to existing methods.
• This improvement was particularly significant for smaller, underrepresented subgroups.
• A case study applied the method to a published subgroup analysis, validating its practical utility.

Conclusions:

• The proposed estimators effectively mitigate the impact of representation disparities on ATE estimates.
• While statistical methods can help, fundamental structural changes are ultimately necessary for true equity.
• Recommends adopting these estimators to improve research fairness and reduce bias.