Causation, First Pass

Strike a match and it lights. Everyone knows what happened, and nobody has to be taught the concept. Saying what the striking did to the lighting, in terms that distinguish it from mere sequence, has resisted three centuries of effort.

The vocabulary

The first definition is circular, and openly so. "Brings about" and "is responsible for" are causal words, so nothing has been analysed. That is the honest starting point. Everyone recognises causal relations without difficulty; the enquiry is into what recognising them consists in, and beginning with a false show of analysis would only conceal the problem.

Two features of the relation can be fixed before any theory. Causation has a direction, so that if the striking caused the lighting, the lighting did not cause the striking. And it supports claims about what would have happened otherwise, which is why establishing a cause is what makes intervention possible. Any account has to earn both.

Conditions

Before the theories, two notions that get used constantly and confused almost as often.

Neither notion is causation, and the gap is instructive. Oxygen is necessary for the lighting and nobody says the oxygen lit the match. Being a widow is sufficient for having been married and nothing was caused. Conditions are patterns of co-occurrence, and causation is something more than a pattern, which is exactly what has to be explained.

Ordinary causal talk also selects. A great many conditions were jointly required for the fire, including the presence of oxygen, the dryness of the wood, and the absence of a sprinkler. Calling the dropped cigarette the cause picks one out, and the selection is governed by what is unusual, what was under someone's control, and what the enquiry is for. That selection is real and it is not metaphysics. It is worth separating from the question of what the causal relation is.

Hume's challenge

David Hume put the problem in the form everyone since has worked from. Watch the collision as closely as you like and you observe the first ball moving, the contact, and the second ball moving. You do not observe a further ingredient, the causing. What is available to observation is one thing following another, together with the fact that things of the first sort have regularly been followed by things of the second.

Hume's first definition of a cause, in the Enquiry Concerning Human Understanding, is built from exactly that material: an object followed by another, and where all the objects similar to the first are followed by objects similar to the second.

The account is admirably austere and it fails on cases that are not hard to find.

Night follows day with perfect regularity and day does not cause night. Regular succession is present and causation is absent, so the account is too permissive. Worse, it gets the direction wrong in both senses at once, since night is also regularly followed by day.

The falling barometer is regularly followed by the storm, and the barometer does not cause the storm. Both are effects of the falling pressure. Regularity cannot distinguish a cause from a shared symptom, and this is not a marginal difficulty: most of the misleading correlations that trouble the sciences have this shape.

The other definition

In the same section Hume offers a second formulation, and it points somewhere quite different: where, if the first object had not been, the second never had existed.

That is not a claim about regular succession. It is a claim about what would have been the case had things gone otherwise, and it handles the counterexamples immediately. Had today not dawned, night would still have fallen, so day does not cause night. Had the barometer been stuck, the storm would have come anyway, so the barometer does not cause the storm. Had the match not been struck, it would not have lit, so the striking caused the lighting.

David Lewis built the modern theory on this foundation. The core proposal is that one occurrence depends on another when, had the first not occurred, the second would not have occurred either. It gets the direction right without further stipulation, because the dependence runs one way: remove the striking and the lighting goes, while removing the lighting does not remove the striking.

Chains

Dependence on its own is not yet causation, and the reason is that some causes act through intermediaries that break the direct dependence.

Building the chain into the account has an immediate consequence: causation comes out transitive. If the first causes the second and the second causes the third, the first causes the third. That result is wanted in the ordinary cases, since the striking caused the fire that burned the house, and it has been challenged by cases where a cause prevents something that would otherwise have prevented the effect. Those counterexamples are taken seriously and they are not settled.

What to take forward

Causation is not observed as an extra ingredient in a sequence, and it is not captured by regularity, and the counterfactual test does better while facing serious difficulties of its own. Those three points are the state of the question at the level of this tier.

Two further approaches deserve naming, since the intermediate tier develops both. One takes causation to be the transfer of a conserved quantity, so that the cause and effect are linked by something physical passing between them. Another treats causal claims as claims about what interventions would produce, and builds them into structured models of a system.

The next article turns to the vocabulary the counterfactual test has been using without licence, since "had the match not been struck" is a claim about how things could have been rather than about how they are.