​Chapter 5

However causation is approached, the necessity or production connecting cause and effect is never among the things given. Something else is recognized when an event is called causal, and what it is has to be found.

Two billiard balls present a motion, contact, and another motion. Across many such cases the sequence repeats, and that repetition is also given — the second motion has followed the first every time. Necessity is supplied afterward from the habit formed by seeing the sequence hold, then read back into the objects as a link between them. Hume's conclusion goes further than his observation by reducing causation to regularity. Causal necessity does not present itself to observation, and whatever grounds our causal claims, it is not the sight of a thing making another happen.

The same feature appears in all major methods of establishing causation. A law records that states of a given type, under specified conditions, are followed by states of another type — the content of the law is the type-level connection, and a single unrepeated instance supports no law. A mechanism opens the transition into stages, so that the blow becomes momentum transfer, lattice deformation, bond failure, and structural collapse. Each stage is a claim that this leads to that, and such a claim carries the requirement — the stage must lead reproducibly to the next, or it is not a stage in a mechanism but an anomaly waiting to be explained. Decomposition multiplies the reproducible links. Intervention approaches causation from a third direction and needs no law or mechanism: change this, and observe whether that changes. The finding is causal when the difference is stable across trials. A single intervention producing a result establishes little, since it is indistinguishable from coincidence; what makes it evidence is that the difference recurs with the manipulation.

The requirement does not weaken where outcomes are irregular. An atom of a given isotope decays or does not, and no state of it fixes which; a drug helps a patient and fails another with the same diagnosis; smoking causes cancer in a population where most smokers never develop it. A distribution is what makes such a claim admissible, and it speaks only at the level it holds. That smoking causes cancer is established across a population; what happened in one smoker is not in the finding. The decay distribution has a stable profile, while the finding does not say why this atom decayed at this moment. The distribution holds as firmly as any deterministic law. The demand for a reproducible form remains; reproduction shifts from the individual outcome to the profile of outcomes.

Reproducibility is the entry condition rather than the finding. Correlations repeat without being causal: a common cause makes two effects track each other faithfully while neither touches the other, and selection in how cases are gathered manufactures associations that survive repetition. What sorts them out are the same tests again — manipulate the supposed cause, open the transition into stages, look for a third thing producing both — and each is a demand for stability of a different kind. The sorting never leaves the reproducible.

Causation is not simple repetition of events. A meteorite strikes the earth once, leaves a crater, and never recurs; nobody hesitates over whether it caused the crater. Singular causation happens constantly. The unique event is not thereby uncaused. The question is how singular causation becomes known.

Cartwright locates the causal work in capacities. Laws and regularities, she argues, do not say what produces an effect — they record what accompanies what, and a regularity holding across a population does not explain what any member does. The real work is done by what things can do: aspirin may relieve headaches, and it retains that capacity in a case where it fails, and possesses it whether or not any statistical regularity in a given population reflects it. A capacity explains a single occurrence without a universal law behind it. This gives the singular case more than Humean regularity allows. But ask how a capacity is identified. Not by inspection — nothing in aspirin's appearance announces what it can do. It is identified through a stable profile of manifestations: what this kind of substance does across cases, under intervention, in varying circumstances. And the directedness that makes it a capacity for this rather than that is where her account stops explaining, exactly as the law did.

In Anscombe, singular causation is not inferred at all; it is seen. When a knife passes through cloth, what is given is not a moving object followed by a separated material, with a causal connection supplied by the mind. It is a cutting. No universal premise enters the perception, no law is consulted, and the demand for one misdescribes what happens when a person watches anything. However, seeing that motion as a cutting requires knowledge that blades divide what they pass through, that the material parts along the edge and not elsewhere, that the separation follows the movement and does not precede it. The particular case is not derived afresh from a general statement — Anscombe is right that no such derivation occurs — and it is recognized through a form acquired from repetition.

Salmon and Dowe find singular causation in a continuous physical process: a conserved quantity carried along a worldline, momentum transferred between bodies, and a mark transmitted through a spatiotemporal path. Causing consists in a physical continuity within the case, and the particular sequence may be unique in the history of the universe without its causal status being any less secure. It exhibits what connects the two events rather than reporting that events of those types tend to be paired. Yet the identification of a transfer of momentum takes place inside a theory that has fixed a great deal: what momentum is, that it is conserved, how it is measured, and what the typical structure of its transfer looks like. Strip that away and what remains is a body moving and then another body moving. The event may be unique. The causal form under which it is identified cannot be unique, since it is drawn from a physics established across countless cases. The theory takes the dynamics of transfer and the conservation laws as given.

Causation becomes available to knowledge through a repeatable type of action, a stable mechanism, a law, an invariant transfer of a quantity, or a reproducible intervention. A route depends on a form extending beyond the single case. To call an event causal is to place it under a reproducible form of transition. The producing that the form is supposed to name stays where Hume left it, unobserved and unaccounted for, while the form does the epistemic work by which the transition is identified.