Research map / Probabilistic and scientific modeling / Causal discovery

PE-SHAP: Causally Interpretable Path-Wise Shapley Explanations

NeurIPS 2026 · Probabilistic and scientific modeling › Causal discovery · approach: Counterfactual identification and bounds

Read the paperCheck your idea against 108,967 papers

AI-extracted summary. Written by AI from the paper's abstract only. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Proposes PE-SHAP, a causally informed path-wise Shapley explanation method that attributes chain-mediated effects and aligns per-path values with classical mediation estimands.

Problem

Standard Shapley explanations capture associative rather than causal relationships; existing causally informed path-wise Shapley methods assign zero attribution to chain-mediated paths and their per-path values do not match classical mediation estimands.

Approach

Introduces a path-wise Shapley decomposition (PE-SHAP) that assigns nonzero attribution to chain-mediated paths and whose per-path values align with classical mediation estimands (e.g., path-specific effects), addressing the limitations of prior causal Shapley methods.

Experimental setup

Baselines: standard Shapley (SHAP) causal Shapley methods with path-wise decomposition

Builds on: Shapley values causal Shapley values path-specific effects

Closest papers by meaning

• PWSHAP: A Path-Wise Explanation Model for Targeted Variables (ICML 2023 · similarity 0.89)
• Practical do-Shapley Explanations with Estimand-Agnostic Causal Inference (NeurIPS 2025 · similarity 0.88)
• Incorporating Information into Shapley Values: Reweighting via a Maximum Entropy Approach (ICML 2024 · similarity 0.86)
• Structural Causal Bottleneck Models (NeurIPS 2026 · similarity 0.84)
• Local Causal Discovery for Structural Evidence of Direct Discrimination (AAAI 2025 · similarity 0.83)
• Learning Structural Causal Models from Ordering: Identifiable Flow Models (AAAI 2025 · similarity 0.83)