The paper by Tyler J. VanderWeele discusses the principles of confounder selection in causal inference, emphasizing the importance of selecting appropriate confounders to ensure reliable causal estimates. The author reviews recent theoretical and methodological developments and proposes a practical approach to confounder selection when complete causal diagrams are not available. The key proposal is to control for covariates that are causes of the exposure, the outcome, or both, while excluding instrumental variables and including proxies for unmeasured common causes. The paper also relates these principles to statistical covariate selection methods, such as forward and backward selection, and discusses the limitations of these methods. Additionally, it introduces the "disjunctive cause criterion" as a more flexible and theoretically informed approach to confounder selection. The paper concludes with a discussion on the timing of covariate selection and the use of machine learning algorithms for more principled covariate selection.The paper by Tyler J. VanderWeele discusses the principles of confounder selection in causal inference, emphasizing the importance of selecting appropriate confounders to ensure reliable causal estimates. The author reviews recent theoretical and methodological developments and proposes a practical approach to confounder selection when complete causal diagrams are not available. The key proposal is to control for covariates that are causes of the exposure, the outcome, or both, while excluding instrumental variables and including proxies for unmeasured common causes. The paper also relates these principles to statistical covariate selection methods, such as forward and backward selection, and discusses the limitations of these methods. Additionally, it introduces the "disjunctive cause criterion" as a more flexible and theoretically informed approach to confounder selection. The paper concludes with a discussion on the timing of covariate selection and the use of machine learning algorithms for more principled covariate selection.