From Regularity to Intervention: How Interventionist Theories Solve Hume's Causation Problem

The Allure of Spurious Correlation

As a statistician and product manager, you're undoubtedly familiar with the mantra "correlation does not imply causation." But what, precisely, is missing? Let's start with a humorous take on this very problem.

The joke works because we intuitively know it's absurd. Instructing people to shave their heads to boost sales is nonsensical. But why? A strong correlation exists in the data. What gives us the confidence to reject this as a causal claim? To answer that, we first need to understand the philosophical tradition that tried to equate the two.

1. Hume's View: Causation as Mere Regularity

The modern philosophical debate about causation begins with David Hume (1711-1776). As an empiricist, Hume was skeptical of anything that couldn't be traced back to direct experience. When he looked at the world, he noted that we never actually see "causation" itself. We don't perceive a mysterious force or "necessary connection" passing from a cause to an effect.

So, what do we see? We see one event, then another. We see a billiard ball strike another, and then we see the second ball move. If we see this happen over and over, we form an expectation. Hume argued that this is all causation is: a pattern of regular succession.

The following video lecture provides an excellent overview of Hume's "positive" account of causation.

To summarize, for Hume, "C causes E" means nothing more than:

1. Temporal Priority: C happens before E.
2. Spatiotemporal Contiguity: C and E are close to each other in space and time.
3. Constant Conjunction: Events like C are always, or at least regularly, followed by events like E.

In this view, causation is reduced to a specific kind of correlation. The "causal" force we feel is just a psychological habit, an expectation built from observing this constant conjunction. It's an austere and elegant theory, but as the "shaved heads" cartoon suggests, it has profound problems.

2. The Limits of a Regularity View

The regularity theory fails because it cannot meet several basic conditions we expect a theory of causation to satisfy. The following text provides a concise framework for analyzing these failures.

As the article explains, Hume's theory is neither sufficient nor necessary for causation.

Problem 1: Regularity is Not Sufficient
There can be perfect regularity without causation.

• Accidental Regularities (or Spurious Correlations): The sales and shaved heads example fits here. It's a coincidence, not a causal law.
• Common Cause Structures: This is a critical failure. The classic example is a barometer falling and a storm arriving. There is a constant conjunction between the barometer's fall and the storm, but the falling barometer doesn't cause the storm. Instead, a third factor—a drop in atmospheric pressure—causes both. Hume's theory has no way to distinguish this from a genuine causal relationship.

Problem 2: Regularity is Not Necessary
There can be causation without perfect regularity.

• Imperfect Regularities: We say "smoking causes cancer" even though not all smokers get cancer. If causation required invariable succession, this claim would be false. This challenges the "constant conjunction" requirement.

These problems show that a passive, observational approach is insufficient. Simply watching the world and recording patterns isn't enough to disentangle causal structure from mere correlation.

3. The Interventionist Turn: Causation as Manipulation

The failures of the regularity view led to a major philosophical shift. Instead of trying to reduce causation to non-causal patterns, modern theories often treat causality as a fundamental concept related to manipulation and control. This is the core idea of interventionism, most famously developed by philosopher James Woodward.

First, let's clarify the philosophical move being made. Hume's was a reductive analysis. The Arcadia article you read earlier draws this distinction well.

Woodward's theory is non-reductive. It doesn't try to define causation using other terms; instead, it clarifies what causation is by linking it to the concept of an intervention.

The following notes give a wonderfully clear and concise statement of the theory.

The core idea is simple and powerful: "X causes Y" means that if you could wiggle X, you would see a change in Y.

Let's see how this solves the problems that plagued Hume's theory:

• Common Cause (Barometer): The barometer reading is not a cause of the storm because if we intervene to manually change the barometer's needle, it has no effect on the weather. However, if we could (hypothetically) intervene to change the atmospheric pressure, this would affect both the barometer and the storm. The interventionist criterion correctly separates the cause from the correlated side-effect.
• Accidental Regularity (Shaved Heads): The theory correctly rules this out. If a CEO intervenes by ordering a price cut to increase sales, this action will not result in employees shaving their heads.

The manipulability account aligns with the experimental mindset of science. It reframes the question from "What patterns do I see?" to "What can I control?". The "intervention" doesn't have to be physically possible; it can be a hypothetical, counterfactual manipulation. This gives the theory broad applicability.

Conclusion

In this lesson, we traced a critical evolution in the philosophy of causation. We moved from a passive, observational theory to an active, experimental one.

Key Takeaways:

• Hume's Regularity Theory: Defines causation as constant conjunction, based on the empiricist idea that we only observe patterns, not causal powers.
• Limitations of Regularity: The theory fails because regularity is neither sufficient for causation (e.g., common causes, coincidences) nor necessary (e.g., probabilistic causes).
• Woodward's Interventionist Theory: Redefines causation in terms of manipulation and control. A variable X is a cause of Y if an intervention on X would lead to a change in Y.
• A Philosophical Shift: This represents a move from a reductive analysis (trying to define causation away) to a non-reductive one that embraces causality as a fundamental concept for explaining how we can manipulate the world.

This concept of intervention is central to modern causal inference. As we will see in our next lesson, it forms the second rung on Judea Pearl's influential "Ladder of Causation," a framework that organizes causal reasoning into three distinct levels: seeing, doing, and imagining. We will explore this ladder and see how it distinguishes observation from the powerful new idea of intervention.