KAELAN.FYI READ / THINK / MAKE
Library
KAELAN.FYI / NOTE 0571 WORDS

understand / seed

Cause

What causal claims mean and what evidence they need

Seed · early Updated Jul 19, 2026 3 min read
epistemologyphilosophysciencethinking

Hume asked what we perceive when one billiard ball strikes another. We see motion followed by motion. We do not see a “necessary connection” that forces the second event. Repetition trains us to expect it. Hume develops this argument in Section VII of An Enquiry Concerning Human Understanding.1

That is an argument about knowledge and the idea of necessity, not proof that the world contains no causal relations. Philosophers still disagree about what a cause is: a regular pattern, a counterfactual difference, a physical process, a power, or something else.

In 1913, Bertrand Russell argued that the traditional law of causality did little work in advanced science and that mathematical laws related states without naming one event as the cause of another.2 His essay remains a critique, not a report that physicists stopped using causal ideas.

Modern causal inference does not settle the metaphysics. It makes a narrower task precise: estimate how an outcome would differ under another action or exposure.

Two main languages help:

  • Potential outcomes compare what would happen to the same unit under two conditions. We observe at most one outcome, so a study needs a design and assumptions to estimate the missing one.3
  • Causal graphs state which variables cause which others. Judea Pearl’s do-calculus gives rules for identifying intervention effects from such a graph.4

Neither method turns correlation into cause by notation alone. The answer depends on the graph, the comparison group, measurement, and assumptions about hidden common causes. Hernán and Robins put the point plainly: causal inference from observational data needs subject knowledge and cannot become a set of data-analysis recipes.3

Random assignment can make a comparison more credible, but a planned experiment is not the only useful design. Natural experiments use events or policy changes that assign different exposures without a researcher’s intervention. Their causal force still depends on the comparison and its assumptions. The 2021 economics prize recognized work on causal conclusions from natural experiments.5

U.S. tort law uses related but distinct tests. The but-for test asks whether the harm would have happened without the act. Courts often combine that test with proximate cause, which limits liability based on matters such as foreseeability and intervening events.6 These are rules for assigning legal responsibility. They do not solve every scientific or philosophical question about cause.

Everyday claims often compress a system into one cause: stress made me sick, the chief executive ruined the company, we split because we wanted different things. One factor may have mattered. The sentence may also hide timing, feedback, chance, and several contributing conditions. The narrative fallacy makes one clean account feel more certain than the evidence allows.

When you say that X caused Y, ask five questions:

  1. What would have happened without X?
  2. Which other factor could produce both X and Y?
  3. What design makes the comparison credible?
  4. Which assumptions does the result need?
  5. Does the claim apply to one case, an average in a group, or a different setting?

If no experiment is possible, state the other evidence and its limits. “X preceded Y” is weaker than “changing X changed Y.” “X likely contributed” is often more honest than “X was the cause.”

Go Deeper

Books

  • Causal Inference: What If by Miguel Hernán and James Robins — A free text on potential outcomes, study design, and observational evidence.
  • Causality by Judea Pearl — The mathematical account of causal graphs and interventions.

Sources

  1. David Hume, An Enquiry Concerning Human Understanding, Section VII, 1748. ↩

  2. Bertrand Russell, “On the Notion of Cause”, Proceedings of the Aristotelian Society 13 (1913), 1–26. ↩

  3. Miguel A. Hernán and James M. Robins, Causal Inference: What If, 2025. ↩ ↩2

  4. Judea Pearl, “A Causal Calculus for Statistical Research”, Proceedings of the Fifth International Workshop on Artificial Intelligence and Statistics (1995), 430–449. ↩

  5. Nobel Prize, “The Prize in Economic Sciences 2021”, 11 October 2021. ↩

  6. Cornell Legal Information Institute, “But-for test” and “Proximate cause”. ↩