This paper by Judea Pearl addresses the distinction between total, direct, and indirect effects in causal analysis, particularly in the context of nonlinear models. The author proposes a new method to define and measure direct and indirect effects without the need to hold certain variables constant, which is a challenge in nonlinear models. This approach allows for a more natural interpretation of these effects and their policy implications. The paper establishes conditions under which these effects can be consistently estimated from both experimental and nonexperimental data, extending path-analytic techniques to nonlinear and nonparametric models. Key concepts include the "descriptive" and "prescriptive" interpretations of effects, with the descriptive interpretation being more relevant for policy analysis. The paper also introduces the notion of "path-specific effects," which involve deactivating specific causal paths in the model. The results are illustrated through various examples and theoretical proofs, providing a comprehensive framework for understanding and applying these concepts in causal inference.This paper by Judea Pearl addresses the distinction between total, direct, and indirect effects in causal analysis, particularly in the context of nonlinear models. The author proposes a new method to define and measure direct and indirect effects without the need to hold certain variables constant, which is a challenge in nonlinear models. This approach allows for a more natural interpretation of these effects and their policy implications. The paper establishes conditions under which these effects can be consistently estimated from both experimental and nonexperimental data, extending path-analytic techniques to nonlinear and nonparametric models. Key concepts include the "descriptive" and "prescriptive" interpretations of effects, with the descriptive interpretation being more relevant for policy analysis. The paper also introduces the notion of "path-specific effects," which involve deactivating specific causal paths in the model. The results are illustrated through various examples and theoretical proofs, providing a comprehensive framework for understanding and applying these concepts in causal inference.