Causal Probabilities Are Not Properties of the World

Read full text Summarise with VIRAT
Authors: Daniel Diamond Reidpath
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Causal claims in science are routinely expressed as probabilities, yet the question of what those probabilities mean is often unasked. The obvious answer---that the world is uncertain--defers rather than resolves the question. Identifying the source of the uncertainty matters, because a probability carried from one setting to another means something different depending on whether it describes how the world behaves or what an investigator knows about it. This paper argues that causal probabilities are epistemic. That conclusion follows from the structure of the material world and of how we investigate causation within it. We use "causal containers" as a device--circumscribed material configurations whose components collectively sustain pathways to outcomes. Fully specified, causal containers are deterministic. The pathway either exists or it does not. Observed probabilistic variation therefore does not reflect stochastic behaviour within any single container. It reflects variation across the heterogeneous, partially specified containers assembled in a study. Four sources of underspecification account for the variation: incomplete specification, container heterogeneity, temporal drift, and self-referential dissolution. Frequentist probability fails as a foundational account because it presupposes a stable reference class existing independently of any investigator---precisely what the container framework shows to be an artefact of circumscription rather than a feature of the world. Bayesian probability is the appropriate framework because its foundations are consistent with what causal probabilities have been shown to be: degrees of belief over partially specified containers, updated as evidence accumulates. A probability estimate is therefore always local, always indexed to a container, and always subject to revision as that container changes.

Fields of Study & Subjects

Bayesian Modeling and Causal Inference Philosophy of Biology and Scientific Explanation Probability and Statistical Research

Publication details

Journal / source
Zenodo (CERN European Organization for Nuclear Research)
Publication date
18 March 2026
Language
EN
Document type
Preprint