← News

Models & toolsSources: arXiv · 16 Nov 2026

Researchers derive sharp bounds for probabilities of causation in stochastic causal models

An arXiv paper presents an optimization-based method for bounding counterfactual quantities when treatment effects stay random even after latent variables are fixed.

In plain words

In causal statistics, researchers often want to know how likely a treatment actually caused an outcome for a given person. The authors study cases where outcomes remain random by nature, not just because some factor is unobserved. They show how to compute the tightest possible ranges for these causal probabilities from data.

What was announced

The paper, posted to arXiv by its authors, addresses probabilities of causation in nondeterministic causal models, where mechanisms may remain intrinsically stochastic even after all latent variables are fixed. The work focuses on binary treatments and outcomes and develops what the authors call a generic identification program: observational and interventional data are represented as moment constraints on latent stochastic mechanism types, and minimal and maximal values of counterfactual quantities are computed through constrained optimization.

Within this framework, the authors report deriving sharp analytic bounds on probabilities of causation under several assumptions tailored to nondeterministic models. They state that the resulting identification regions are often more informative than those attainable under deterministic alternatives, and they illustrate the results with a case study on personalized treatment decisions and experiments on semi-synthetic datasets. According to the abstract, the results clarify how classical bounds relate to nondeterministic causal semantics, generalize previous works, and provide a unified route from qualitative structural restrictions to computable partial identification regions.

What it does not show

  • The evidence is the paper's abstract only; the full derivations, proofs, and experimental details are not available for verification.
  • The claimed informativeness of the bounds rests on the authors' own statement and the semi-synthetic experiments described; no independent evaluation is included in the supplied source.

Sources

Source-support review completed 17 Nov 2026. This story was drafted with AI assistance and checked against retained source passages. Automated checking is not independent verification.

Coverage: arXiv: abstract.

Updates and corrections

No corrections so far.

First published 17 Nov 2026

New to the subject? AI from the Ground Up starts from the beginning.