Direct and Indirect Effects
By Judea Pearl. 2001
2,374 citations recorded by the source corpus.
The direct effect of one event on another can be defined and measured by holding constant all intermediate variables between the two. Indirect effects present conceptual and practical difficulties (in nonlinear models), because they cannot be isolated by holding certain variables constant. This paper presents a new way of defining the effect transmitted through a restricted set of paths, without controlling variables on the remaining paths. This permits the assessment of a more natural type of direct and indirect effects, one that is applicable in both linear and nonlinear models and that has broader policy-related interpretations. The paper establishes conditions under which such assessments can be estimated consistently from experimental and nonexperimental data, and thus extends path-analytic techniques to nonlinear and nonparametric models.
Subfields: Management Science and Operations Research, Statistics and Probability, Causal Inference & Discovery
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Management Science and Operations Research
- Stijn Vansteelandt
- Tyler J. VanderWeele
- Judea Pearl
- Bernhard Schölkopf
- Kosuke Imai
- Ilya Shpitser
- E. T. Tchetgen Tchetgen
- Silvia Chiappa
In Statistics and Probability
- Judea Pearl
- Tyler J. VanderWeele
- Stijn Vansteelandt
- Ilya Shpitser
- James M. Robins
- Martin Huber
- Kosuke Imai
- Bernhard Schölkopf
In Causal Inference & Discovery
- Judea Pearl
- Tyler J. VanderWeele
- Bernhard Schölkopf
- Ilya Shpitser
- Silvia Chiappa
- Stijn Vansteelandt
- E. T. Tchetgen Tchetgen
- James M. Robins
Citations in the corpus
- Locating and Editing Factual Associations in GPT (2022, methodology): “This modification is a way of probing path-specific effects (Pearl, 2001) for paths that avoid MLP computations.”
- Mass-Editing Memory in a Transformer (2022, methodology): “We then employ causal mediation analysis (Pearl, 2001; Vig et al., 2020; Meng et al., 2022), which implicates a specific range of layers in recalling factual knowledge.”
- Causal Analysis of Syntactic Agreement Mechanisms in Neural Language Models (2021, methodology): “Here, we do not change u, but set the attention head’s value equal to 0 and observe how this changes the effect; this has an interpretation as the controlled indirect effect from Pearl (2001).”
- Causal Mediation Analysis for Interpreting Neural NLP: The Case of Gender Bias (2020, methodology): “Our key contribution is the adaptation of causal mediation analysis (Pearl, 2001) for analyzing the mechanism by which information flows from input to output through different model components.”
- Causal Modeling-Based Discrimination Discovery and Removal: Criteria, Bounds, and Algorithms (2019, background): “In fact, as shown in [31], the above equality holds if”
- Learning Optimal Fair Policies (2018, background): “Under certain identification assumptions discussed by Pearl (2001), the distribution of Y (a,M(a′)) (and thereby direct and indirect effects) can be nonparametrically identified from observed data by the following formula: p(Y (a,M(a′)) = ∑ X,M p(Y | a,X,M)p(M | a′, X)p(X). More generally, when ther”
- Causal Reasoning for Algorithmic Fairness (2018, background): “(b) This diagram, inspired by [30], is a representation of how counterfactuals are propagated only through some edges.”
- Path-Specific Counterfactual Fairness (2018, methodology): “Pearl (2001; 2012) and Sh-pitser (2013) indicate how and when the PSE can be expressed through integrations over conditional distributions involving only observed variables.”
- Fair Inference on Outcomes (2017, methodology): “Effects along paths, such as an effect along the path A → W → Y , are known as path-specific effects (Pearl 2001).”
- Simpson's Paradox in COVID-19 Case Fatality Rates: A Mediation Analysis of Age-Related Causal Effects (2020, methodology): “[38] A. Forney, J. Pearl, and E. Bareinboim, “Counterfactual data-fusion for online reinforcement learners,” in Proc.”
- Fairness in Decision-Making - The Causal Explanation Formula (2018, methodology): “(Pearl 2001) introduced natural direct (NDE) and indirect (NIE) effects to measure the effect of, respectively, the direct and indirect causal paths on the total effect of X on Y .”
- Causal inference in statistics: An overview (2009, background): “Formally, this reads (Pearl, 2001): IEx,x′(Y ) ∆ = E((Yx,Z x′ ) −E(Yx)), (49) which is almost identical to the direct effect (Eq.”
- Theoretical Impediments to Machine Learning With Seven Sparks from the Causal Revolution (2018, positive): “The graphical representation of counterfactuals enables us to define direct and indirect effects and to decide when these effects are estimable from data, or experiments (Robins and Greenland, 1992; Pearl, 2001; VanderWeele, 2015).”
- Unbiased Scene Graph Generation From Biased Training (2020, positive): “We propose a novel unbiased SGG method based on the Total Direct Effect (TDE) analysis framework in causal inference [57, 39, 58].”
- Direct and indirect treatment effects–causal chains and mediation analysis with instrumental variables (2014, methodology): “When for instance evaluating the earnings effects of math education as in Rose and Betts (2004), math courses likely affect the decision to obtain further education and may also have a direct effect on earnings (conditional on total years of education).”
- Counterfactual VQA: A Cause-Effect Look at Language Bias (2020, methodology): “In this section, we formally introduce the key concepts of causal inference [53, 38, 54, 55] used in this paper.”
- Reward Tampering Problems and Solutions in Reinforcement Learning: A Causal Influence Diagram Perspective (2019, background): “…Sk → Rk, and not via paths of the form Ai → · · · → Dj → Rk, i j ≤ k. Fortunately, separation of path-specific effects has seen a long and careful study in mediation analysis, where Pearl (2001)22 have suggested a definition of natural direct effect based on counterfactual outcomes (Rubin, 2005).”
- Equality of Opportunity in Classification: A Causal Approach (2018, background): “1-3 can be seen as a generalization of the marginal counterfactual measures, including the counterfactual effects introduced in [28] and the natural effects in [17, 11, 15].”