What the Plumber Is Actually Tracing

Opening Scene

Two pipes can run parallel to each other for the entire length of a house without ever being connected. They might even carry water at correlated times — both busiest during the morning rush, both quiet at night — purely because they’re both responding to the same household schedule, not because one feeds the other. A plumber who mistakes this shared schedule for an actual physical connection will trace the wrong pipe entirely.

In Plain English

Correlation measures whether two variables tend to move together. Causation means one variable’s change actually produces a change in the other. Correlation is necessary but nowhere near sufficient for causation — two variables can be strongly correlated while one has zero actual causal effect on the other, most commonly because a third factor (covered directly in Article 4) is driving both.

The Old Way

Before this distinction had formal statistical language, people already recognized it intuitively, if inconsistently:

“Just because it rained after the rain dance doesn’t mean the dance caused the rain” — a folk recognition of the correlation-causation gap, ancient and widespread across cultures.
A doctor recognizing that two symptoms occurring together doesn’t mean one causes the other — both might stem from the same underlying condition.
An economist recognizing that two markets moving together doesn’t mean one drives the other — both might respond to the same broader economic conditions.

The intuition that correlation alone isn’t proof long predates any formal statistical framework for actually testing it.

What’s Changing (and Why AI Is the Reason)

A mature, rigorous toolkit — covered throughout this series, from causal graphs in Article 5 to randomized trials in Article 6 to a range of observational methods beyond — now lets analysts move past the old, purely intuitive recognition of this gap toward actually closing it in specific cases.
As machine learning models increasingly learn from purely correlational patterns in data, the risk of a model confidently exploiting a correlation with no real causal basis has become a well-documented, practical concern, distinct from the model’s raw predictive accuracy.
Growing computational power has made formerly impractical causal inference methods, like matching across many variables or estimating complex causal graphs, genuinely usable at real organizational scale.

The Metaphor, Fully Extended

Behind the Wall	Correlation vs. Causation Concept
Two pipes running parallel, never actually connected	Two variables that correlate without one causing the other
Both pipes busiest at the same time, due to the household’s shared schedule	Two variables moving together due to a shared underlying cause
A plumber who traces the actual physical connection, not just the shared timing	An analyst who establishes actual causation, not just correlation
Confidently repairing the wrong pipe because the timing looked connected	Confidently acting on a correlation that isn’t actually causal

For Beginners: What to Actually Do

Practice generating alternative explanations for any correlation you encounter — a shared cause, reverse causation, or pure coincidence are all worth considering before assuming direct causation.
Learn to recognize “correlation without established causation” as an incomplete finding, not a final answer.
Get comfortable holding a correlational finding as a hypothesis worth testing further, rather than a conclusion ready to act on.

For Practitioners and Leaders: The Deeper Layer

Require an explicit causal justification, not just a correlational one, before major decisions are made from observational data.
Train teams to distinguish “we found a correlation” from “we established causation” in how findings are communicated internally.
Recognize that closing the correlation-causation gap rigorously, rather than just acknowledging it exists, is exactly what the rest of this series covers.

Quick Recap

Correlation measures whether two variables move together; causation means one actually produces a change in the other.
Correlation is necessary but not sufficient for causation, most commonly because a shared third factor drives both.
A mature toolkit of causal inference methods now allows this gap to be closed rigorously in many real situations.
Machine learning models that learn from pure correlation risk confidently exploiting relationships with no real causal basis.

Where This Fits in the Series

Article 2 drew the line between two pipes merely running near each other and one actually feeding the other. Article 3 looks at how people reasoned about cause and effect before any formal method existed.