mlsanityData to 7 Oct 2026

Path-specific harm decomposition: A partial identification framework

arXiv 2609.29938 ↗First posted 24 Sep 2026stat.ML cs.LGcited by 0

Not on the map. Why?

Abstract (arXiv)

A central goal when designing treatment policies is often to "do no harm", that is, to avoid interventions that improve average outcomes while worsening outcomes for some individuals. A widely used notion for harm is the fraction of negatively affected (FNA), defined as the probability that an intervention decreases an individual's outcome. However, in many applications, treatments operate through mediators, and a single "total" FNA can obscure whether harm arises primarily through direct pathways or indirect (mediator-induced) pathways. In this work, we introduce a path-specific analogue of the FNA. For this, we disentangle total harm into direct and indirect harm in causal mediation settings. However, these quantities depend on joint distributions of potential outcomes that are not point-identified even in randomised controlled trials. As a remedy, we develop a novel partial identification framework for direct and indirect FNA. In our framework, we (i) derive sharp Makarov bounds for the FNA, and (ii) propose a semiparametrically efficient estimator with valid confidence intervals for these bounds under mild margin conditions. We demonstrate our framework across various numerical experiments. To the best of our knowledge, we are the first to study path-specific decomposition of causal harm and to develop an orthogonal inference framework for its analysis.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Builds on 21 of its 59 references that are in the map. First posted on arXiv 24 Sep 2026.

Builds on

21 of 59 references.

References · 21

  1. 2017Counterfactual Fairnesscited by 802
  2. 2013Direct and Indirect Effectscited by 276
  3. 2017Avoiding Discrimination through Causal Reasoningcited by 191
  4. 2018Path-Specific Counterfactual Fairnesscited by 179
  5. 2016A causal framework for discovering and removing direct and indirect discriminationcited by 71
  6. 2019PC-Fairness: A Unified Framework for Measuring Causality-based Fairnesscited by 67
  7. 2013Robust Inference on Average Treatment Effects with Possibly More Covariates than Observationscited by 38
  8. 2021Causes of Effects: Learning individual responses from population datacited by 36
  9. 2022Counterfactual harmcited by 29
  10. 2020Principal Fairness for Human and Algorithmic Decision-Makingcited by 20
  11. 2022A Causal Analysis of Harmcited by 15
  12. 2022What's the Harm? Sharp Bounds on the Fraction Negatively Affected by Treatmentcited by 14
9 more
  1. 2022Quantifying Harmcited by 10
  2. 2023Causal Fairness under Unobserved Confounding: A Neural Sensitivity Frameworkcited by 10
  3. 2024Quantifying Aleatoric Uncertainty of the Treatment Effect: A Novel Orthogonal Learnercited by 10
  4. 2022Policy Learning with Asymmetric Counterfactual Utilitiescited by 9
  5. 2024Controlling Counterfactual Harm in Decision Support Systems Based on Prediction Setscited by 9
  6. 2023Path-Specific Counterfactual Fairness for Recommender Systemscited by 7
  7. 2023Bounding the Probabilities of Benefit and Harm Through Sensitivity Parameters and Proxiescited by 2
  8. 2024Do No Harm: A Counterfactual Approach to Safe Reinforcement Learningcited by 1
  9. 2026Quadruply robust methods for causal mediation analysiscited by 1

Built on by

Not cited yet.