---
title: Regularity Theory of Causation (Hume)
type: concept
tags: [philosophy-of-science, causation, hume, empiricism]
sources: ["[[sources/2023-anjum-rocca-phi403-causation-in-science]]", "[[sources/2023-anjum-rocca-phi403-lecture-04-whats-in-a-correlation]]"]
created: 2026-04-20
updated: 2026-04-20
---

# Regularity Theory of Causation

Attributed to **David Hume** (*A Treatise of Human Nature*, 1739): there is nothing more to causation than a **constant conjunction** of two events — the regular, repeated succession of one kind of event by another. Causal necessity is not observable; only regularity is ([[sources/2023-anjum-rocca-phi403-lecture-04-whats-in-a-correlation]]).

## Core commitments
- Causation is **universal** — a covering-law model: *C* causes *E* iff *all* instances of *C* are followed by *E* ([[sources/2023-anjum-rocca-phi403-lecture-11-is-more-data-better]]).
- Causation is **observable only via its regularities** — epistemological parsimony.
- **Same cause, same effect** — "like causes always produce like effects" (Hume 1739) ([[sources/2023-anjum-rocca-phi403-lecture-05-same-cause-same-effect]]).
- Implies: more data is better; representative sampling; [[concepts/evidence-hierarchy|statistical methods]] are the natural scientific method.

## Strengths
- Fits the empiricist ideal: only observable features enter the analysis.
- Motivates correlation-driven science: statistics, RCTs, large-N studies.
- Easy to operationalise.

## Problems the course raises
- **Correlation is not causation** — 5 possible conclusions from any correlation (C→E, E→C, common cause, mutual causation, spurious); spurious correlations (Tyler Vigen's *divorce in Maine × margarine consumption in US*) exhibit near-perfect regularities ([[sources/2023-anjum-rocca-phi403-lecture-04-whats-in-a-correlation]]).
- **Real causal connections are rarely perfect regularities** — exceptions, outliers, noise, non-responders, interferers ([[sources/2023-anjum-rocca-phi403-lecture-05-same-cause-same-effect]]).
- **Additive interference** breaks constant conjunction: A can cause B while an intervention *I* prevents B from following A ([[sources/2023-anjum-rocca-phi403-lecture-08-have-your-cause-and-beat-it]]).
- **The problem of induction** (Hume's own): any finite data set underdetermines the regularity.
- Modern statisticians work hard to separate genuine from accidental regularities — which implies they *already* take causation to be *more* than mere regularity.

## Modern remnants
The theory is "alive and well" in the evidence hierarchy of **evidence-based medicine**, in the "correlations first" norm of quantitative science, and in the assumption that more data yields better causal claims. The course treats this as a **philosophical bias** scientists adopt implicitly.

## Related
[[concepts/causation]] · [[concepts/probabilistic-causation]] · [[concepts/dispositionalism]] (the main rival) · [[concepts/rct-limitations]] · [[concepts/philosophical-bias]]