Regression with race-modifiers: towards equity and interpretability