Regression with race-modifiers: towards equity and interpretability.

Area of Science:

• Quantitative social science
• Health equity research
• Statistical methodology

Background:

• Structural racism and racial discrimination demonstrably affect health and life outcomes, with effects varying by race.
• Standard statistical regression analyses often introduce racial biases when estimating race-modified effects.
• Existing methods compromise parameter interpretability, equitability, and statistical efficiency.

Purpose of the Study:

• To introduce abundance-based constraints (ABCs) as a novel method to eliminate racial biases in statistical regression.
• To demonstrate that ABCs provide invariance, ensuring main effect estimates are unaffected by race modifiers.
• To enable the estimation of race-specific effects without sacrificing interpretability, equitability, or efficiency.

Main Methods:

• Advocacy and theoretical exposition of abundance-based constraints (ABCs).
• Application of ABCs integrated with statistical learning techniques (regularization and selection).
• Estimation of joint effects on 4th-grade reading scores using North Carolina student data (n=27,638).

Main Results:

• ABCs were shown to possess a remarkable invariance property, preserving main effect estimates regardless of race modifiers.
• The method facilitates the estimation of race-specific effects without compromising statistical rigor or fairness.
• Identified race-modified effects of residential isolation, PM2.5 exposure, and maternal age on reading scores.

Conclusions:

• ABCs offer a powerful solution to mitigate racial bias in quantitative research.
• The method enhances the ability to study health disparities and social determinants of health equitably.
• This approach supports more accurate and equitable understanding of factors influencing educational outcomes.