Causal Identification from Counterfactual Data: Completeness and Bounding Results

2026-03-02 · Source: cs.AI updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Expert, quick

Summary

Arvind Raghavan and Elias Bareinboim introduce the CTFIDU+ algorithm, which addresses the identification of counterfactual queries using data from Layer 3 of Pearl's Causal Hierarchy. Previous work on counterfactual identification was limited to observational or interventional distributions (Layers 1 and 2), as Layer 3 data was considered unobtainable. However, recent research (Raghavan & Bareinboim, 2025) defined "counterfactual realizability," allowing direct estimation of certain counterfactual distributions via experimental methods. The CTFIDU+ algorithm is proven complete for identifying counterfactual queries from an arbitrary set of Layer 3 distributions. This work also establishes the theoretical limits of counterfactuals identifiable from physically realizable distributions, defining a fundamental limit to exact causal inference in non-parametric settings. For non-identifiable counterfactuals, the authors derive novel analytic bounds using realizable counterfactual data, demonstrating in simulations that this data tightens these bounds.

Key takeaway

For AI Researchers and Causal Inference Scientists working with non-parametric models, this work fundamentally shifts the landscape of what is possible in counterfactual identification. You should explore integrating the CTFIDU+ algorithm to leverage newly accessible Layer 3 data, potentially achieving more precise causal inferences and tighter bounds for previously non-identifiable quantities in your experimental designs.

Key insights

The CTFIDU+ algorithm enables counterfactual identification using newly characterized realizable Layer 3 data.

Principles

• Counterfactual realizability allows direct Layer 3 data estimation.
• CTFIDU+ is complete for identifying counterfactual queries.

Method

The CTFIDU+ algorithm identifies counterfactual queries from arbitrary Layer 3 distributions, establishing theoretical limits for exact causal inference and deriving analytic bounds for non-identifiable quantities.

In practice

• Utilize CTFIDU+ for Layer 3 counterfactual identification.
• Apply realizable counterfactual data to tighten bounds.

Topics

• Causal Identification
• Counterfactuals
• Causal Inference
• Non-parametric Causal Models
• CTFIDU+ Algorithm

Best for: AI Researcher, AI Scientist, Research Scientist

Related on AIssential

• Causal Learning with the Invariance Principle — Two auxiliary environments suffice for causal graph and counterfactual inference in acyclic, invertible SCMs.
• Optimal Experiments for Partial Causal Effect Identification — Optimal experimental design for partially identifiable causal queries is NP-hard but can be significantly pruned using graphical criteria.
• Improving RCT-Based Treatment Effect Estimation Under Covariate Mismatch via Calibrated Alignment — CALM improves treatment effect estimation by aligning RCT and observational data in a shared embedding space.
• DeepSWIP: Quotient-WMC Counterfactuals for Neural Probabilistic Logic Programs — DeepSWIP enables exact single-world counterfactual reasoning in neurosymbolic DeepProbLog programs via a novel WMC transformation.
• Debiased Counterfactual Generation via Flow Matching from Observations — Deconfounding flows efficiently estimate counterfactual distributions by transforming observational data, leveraging inherent structural similarities.
• Partial Causal Structure Learning for Valid Selective Conformal Inference under Interventions — Selective conformal prediction under interventions can be robustly maintained by learning partial causal structures.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.AI updates on arXiv.org.