Effect Decomposition in the Presence of Treatment-induced Confounding: A Regression-with-residuals Approach