Philosophical Accounts of Causation

When we say smoking causes cancer, or that a new teaching method raises test scores, what are we actually claiming? For centuries, philosophers have tried to pin down what “causation” means using only what we can observe. This chapter walks you through the most famous attempt—regularity theories. We’ll start with David Hume’s simple idea of constant conjunction, then follow it through John Stuart Mill’s improvements, the logical positivists’ strict demands, and John Mackie’s clever INUS conditions. In the end, we’ll see why these regularity views, smart as they are, still leave policy makers wanting something more.

The Big Picture#

Every time we design a policy—a health campaign, an economic boost, a school reform—we lean on causal claims. We assume that changing one thing will bring about a desired result. But how do we know? Philosophical accounts of causation try to answer what it means for one event to cause another, without appealing to mysterious forces or invisible links. The regularity tradition says: causation is just a pattern of one type of event being followed by another. This chapter traces that tradition from its start in Hume’s empiricism (the view that knowledge comes only from what we can see, hear, or touch), through Mill’s refinements, the positivists’ dream of perfect laws, and Mackie’s complex set‑up. We’ll then see why regularity views, powerful as they are, stumble when we need to make a real‑world difference.

Hume’s Constant Conjunction and the Birth of Regularity Theory#

Picture yourself playing billiards. You strike the cue ball; it rolls and hits the eight ball, which then moves. You see the cue ball moving, the contact, and the eight ball moving. But do you ever see a thing called “causation” flowing from one ball to the other? David Hume, an 18th‑century Scottish philosopher, argued that you do not. All you observe is a sequence of events: first the cue ball moves, then the eight ball moves, and they touch. Repeat this a hundred times, and you come to expect the second event whenever the first happens. That expectation, Hume said, is the whole story of causation.

Hume boiled causation down to three ingredients we can actually observe:

Constant conjunction: Whenever an event of type A occurs, an event of type B follows.
Temporal priority: The cause (A) happens before the effect (B).
Contiguity: The cause and effect are close in space and time.

According to Hume, there is no “necessary connection” in the world itself—no hidden glue that forces B to happen when A does. The idea of necessity is just a feeling in our minds, born from habit. We see A and B constantly conjoined, so our imagination jumps from the impression of A to the idea of B. That mental leap is what we mistake for a real causal power.

This view is the foundation of regularity theories of causation. A cause is simply an event that is regularly followed by another event, with the cause coming before the effect. No hidden mechanisms, no metaphysical forces—just patterns in our experience.

Hume’s account, however, created a famous puzzle. If all our causal knowledge rests on past regularities, how can we justify the belief that the future will resemble the past? We can’t, Hume admitted, without circular reasoning. This is the problem of induction: we can’t prove the future will be like the past without assuming the very thing we’re trying to prove. So causal thinking, at its core, is a leap of faith guided by custom.

For policy, this is unsettling. If we can never truly prove that a programme that worked last year will work again, on what grounds do we act? Hume’s answer was practical: we have no choice but to rely on habit. But later thinkers wanted a firmer footing.

📝 Section Recap: Hume defined causation as constant conjunction, temporal priority, and contiguity, stripping away any idea of necessary connection. This starting point launched the regularity tradition but also exposed the deep puzzle of induction.

Mill’s Refinements: Distinguishing Causal from Accidental Regularities#

Suppose you notice that every time the rooster crows, the sun rises shortly after. That’s a constant conjunction, but we wouldn’t say the rooster’s crow causes the sunrise. John Stuart Mill, a 19th‑century philosopher, saw that Hume’s bare constant conjunction wasn’t enough—we need a way to separate genuine causal patterns from mere coincidences.

Mill introduced the idea of an invariable and unconditional sequence. A causal law isn’t just any repeated pattern; it’s one that holds no matter what else is going on, as long as the relevant circumstances stay the same. He developed a set of methods—the Method of Difference, the Method of Agreement, and others—to identify causes by comparing situations where the effect occurs and where it does not.

The Method of Difference is the most intuitive. Imagine two identical factories, except one introduces a new safety rule and the other does not. If the first factory sees a drop in accidents and the second does not, then the rule is likely the cause. Mill’s insight was that we need to control for background conditions: the regularity must be unconditional relative to a fixed set of circumstances. The rooster‑sunrise link fails because if you remove the rooster, the sun still rises; the sequence is not unconditional.

Mill also tried to ground the idea of a necessary connection more firmly. He argued that causation is not just a habit of the mind but a real relation in the world, discoverable through careful observation and experiment. Yet his account stayed within the regularity camp: the relation is still just an invariable sequence. He didn’t invoke mysterious forces, only a sharper idea of what counts as a genuine regularity.

This refinement matters for policy because it gives us a practical tool: compare similar groups, change only one factor, and see if the outcome differs. Mill’s methods are the ancestors of modern controlled experiments and difference‑in‑differences designs.

📝 Section Recap: Mill sharpened the regularity view by requiring causal sequences to be invariable and unconditional, not just accidentally paired. His Method of Difference gave early guidance for isolating causes through comparison.

Logical Positivism and the Ideal of Exceptionless Laws#

In the early 20th century, a group of philosophers known as the logical positivists pushed the regularity idea to its limit. They wanted to ground all meaningful statements in verifiable experience. For them, a causal law was simply a universal generalisation: “All A are followed by B,” with no exceptions. If a supposed law had any counterexample, it wasn’t a genuine law at all.

This view reduced causation to exceptionless regularities. The statement “Smoking causes lung cancer” would be read as “Whenever a person smokes, they will develop lung cancer”—a claim that is clearly false if taken literally, because many smokers never get the disease. To handle this, the positivists added an “all else being equal” clause: “All else being equal, smoking is followed by cancer.” But what does “all else being equal” mean? It means we hold fixed a host of background conditions. The trouble is, we can never fully list all those conditions. The law then becomes something like “If conditions C1, C2, …, Cn hold, then smoking is followed by cancer.” Since we can’t list all the C’s, the law is not strictly exceptionless.

This created a crack in the positivist programme. Real‑world causal claims rarely satisfy the ideal of a perfect, exceptionless regularity. Most of the regularities we rely on in social science, medicine, and policy are imperfect—they hold only most of the time, or under roughly normal conditions. The positivists’ strict demand for exceptionless laws clashed with the messy reality of causal inference.

Even so, the positivist emphasis on observables and logical structure left a lasting mark. It reinforced the idea that causation is fundamentally about patterns in data, and it encouraged the search for formal, mathematical ways to describe causal laws. But it also highlighted a problem: if we can’t state a law without a long list of “all else being equal” conditions, how do we ever know we have a genuine causal regularity rather than a mere coincidence?

📝 Section Recap: Logical positivists aimed to reduce causation to exceptionless regularities, but the need for “all else being equal” clauses revealed that most real‑world causal claims are imperfect. This tension set the stage for more detailed accounts.

Mackie’s INUS Condition: A More Nuanced Regularity#

By the mid‑20th century, it was clear that simple “A always causes B” statements rarely capture real causal complexity. John Mackie offered a clever refinement: a cause is typically an INUS condition—an Insufficient but Necessary part of an Unnecessary but Sufficient condition.

Let’s unpack that mouthful with a classic example. A house catches fire. The fire investigator says a short circuit caused the fire. But the short circuit alone wasn’t enough; it needed oxygen, flammable material, and the absence of a sprinkler system. Together, the short circuit plus these background factors formed a condition that was sufficient for the fire—whenever that whole cluster occurs, a fire follows. However, that cluster is not necessary for a fire; a fire could also be caused by a lit cigarette, arson, or a lightning strike. Within that particular sufficient cluster, though, the short circuit was an insufficient part (it couldn’t cause the fire alone) but a necessary part (without it, that specific cluster wouldn’t have produced the fire). So the short circuit is an INUS condition for the fire.

Mackie’s analysis stays within the regularity family. It doesn’t ask what would have happened if things were different; it just describes the logical shape of the constant conjunctions we observe. A cause is a factor that plays a special role in the complex web of conditions that are regularly followed by the effect. This handles many real‑world cases where several causal pathways can lead to the same outcome, and where no single factor is either necessary or sufficient on its own.

For policy, the INUS idea is helpful because it reminds us that interventions rarely work in isolation. A job‑training programme might be an INUS condition for employment: it’s insufficient by itself (the economy must also have jobs), but it’s a necessary part of one particular pathway to employment. Understanding this structure can guide us in designing complementary policies.

Yet Mackie’s account still relies on observed regularities. It doesn’t tell us what would happen if we stepped in and changed the short circuit; it only says that in the past, short‑circuit‑plus‑background was followed by fire. That limitation becomes pressing when we turn to imperfect regularities.

📝 Section Recap: Mackie’s INUS condition refines regularity theory by showing that causes are often insufficient but necessary parts of unnecessary but sufficient conditions. It captures complex causal webs without leaving the regularity framework.

Imperfect Regularities and the Difference‑Making Problem#

Real life is messy. Smoking doesn’t always cause cancer; education doesn’t always raise income; interest‑rate cuts don’t always boost investment. We live in a world of imperfect regularities—probabilistic, exception‑ridden, context‑dependent patterns. This creates a deep challenge for any purely regularity‑based account of causation: how do we tell a genuine cause apart from a mere statistical association?

Think about a classic puzzle: ice‑cream sales and drowning deaths both rise in the summer. They show a strong constant conjunction. But nobody thinks ice‑cream causes drowning. The regularity alone doesn’t tell us which factor makes the difference. This is a spurious correlation: two things that happen together because of a third factor (summer weather), not because one causes the other. The difference‑making problem is precisely this: we need a way to pick out, among all the regularities we see, the ones where changing the cause would change the effect, and the ones that are just by‑products of a common cause.

Regularity theories, even Mackie’s, struggle here. They are fundamentally about what actually happens, not about what would happen under hypothetical changes. If we only have data on ice‑cream sales and drownings, the regularity theorist can note the constant conjunction and perhaps look for unconditional sequences (Mill) or INUS structures. But without extra information about the underlying causal structure—that summer heat drives both—the regularity alone cannot flag the relationship as non‑causal.

That’s why philosophers and scientists began to look beyond regularities. The idea of difference‑making—that a cause is something which, if changed, would make a difference to the effect—points toward counterfactual (thinking about what would have happened if things were different) and interventionist theories. Those approaches ask: “If we were to step in and raise ice‑cream sales (say, by giving away free cones in winter), would drowning deaths go up?” The answer is no, revealing there’s no causal link. Regularity theories, tied to passive observation, have no built‑in notion of intervention.

The problem of imperfect regularities thus exposes a basic gap. Policy makers don’t just want to predict; they want to change outcomes. They need to know which levers to pull. A theory of causation that only catalogues observed patterns, without a way to model what happens when we pull a lever, leaves them guessing.

📝 Section Recap: Real‑world causal claims involve imperfect, probabilistic regularities. The difference‑making problem shows that regularity alone cannot separate genuine causes from spurious correlations, because it lacks a concept of intervention or hypothetical change.

The Policy Relevance Problem: Why Regularity Isn’t Enough#

Now we can see why regularity theories, for all their historical importance, are unsatisfying for policy analysis. Policy is about action. We ask: “If we implement this tax credit, will employment rise?” That is a question about what would happen under a deliberate change, not merely about what has regularly followed in the past.

Regularity theories give us correlations. They tell us that, in observed data, tax cuts have often been followed by employment gains. But they cannot rule out the possibility that both were caused by a third factor—say, a booming global economy—or that the observed regularity will break down when we actively intervene. The classic example is the barometer. A falling barometer reading is regularly followed by a storm, but we don’t think fixing the barometer will prevent the storm. The regularity is there, but the causal direction is wrong for policy.

Also, policies are often designed to change a system, not just to watch it. When we introduce a new policy, we may alter the very conditions that supported the past regularity. A job‑training programme that worked in a tight labour market might fail in a recession. Regularity theories, which lean on stable (or at least fairly reliable) patterns, offer no principled way to predict what happens when the underlying conditions shift.

This critique does not mean regularity theories are useless. They gave us the first rigorous attempt to define causation in purely empirical terms, and they laid the groundwork for modern causal discovery methods. But they miss a key ingredient: a notion of manipulation or intervention. To guide policy, we need a causal account that tells us what would happen if we do something, not just what we have seen. That insight motivated the development of counterfactual, interventionist, and structural‑equation frameworks that dominate causal inference today.

In short, regularity theories answer “What follows what?” Policy analysis demands “What happens if we change this?” The gap between those two questions is the central lesson of this philosophical journey.

📝 Section Recap: Regularity theories fail to guide policy because they cannot tell spurious correlations from genuine causal relationships and lack a concept of intervention. Policy requires knowing what would happen under a deliberate change, not just what has been observed.

Summary#

We’ve seen a long line of philosophers—from Hume to Mackie—trying to pin down causation using only patterns we can observe. Hume started with constant conjunction, Mill added strict conditions, the positivists demanded no exceptions, and Mackie offered the INUS framework. All these regularity views share the belief that causation is just a pattern in what we see. But real life is messier. Most causal claims are imperfect regularities. When we need to act—to pull a policy lever—merely observing patterns isn’t enough. That’s why modern approaches focus on interventions and counterfactuals. The philosophical groundwork, though, remains vital: it teaches us why simply spotting correlations is never enough.