The Third Pipe

Opening Scene

Two fixtures on opposite sides of a house both lose water pressure at the same time, every single day, around six in the evening. It looks like one must be causing the other. Trace the actual plumbing, though, and the real explanation is a single, shared supply pipe feeding both — busy elsewhere in the house at that hour, dropping pressure to both fixtures simultaneously, with neither fixture affecting the other at all.

In Plain English

A confounding variable is a third factor that influences both the supposed cause and the supposed effect, creating a correlation between them even when neither one actually causes the other. Confounding is the single most common reason a strong, genuine correlation turns out not to reflect any direct causal relationship at all — and identifying the likely confounder is often the very first step in any real causal investigation.

The Old Way

Before this had formal statistical language, people already recognized this pattern in specific, memorable cases:

The classic example: ice cream sales and drowning deaths both rise in summer, correlated with each other, but both actually driven by a shared cause — warm weather — with no direct causal link between them.
A city’s crime rate and its ice cream sales might rise together, both driven by the same warmer weather that brings more people outdoors.
A company’s revenue and its employee headcount might rise together, both driven by the same underlying business growth, without headcount directly causing revenue in a simple, linear way.

Once named, this pattern becomes easy to spot in hindsight — the real challenge is identifying it before it leads to a wrong conclusion, not after.

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

Formal methods for identifying and adjusting for confounders — covered directly in the matching techniques of Article 9 and the graphical methods of Article 5 — now let analysts systematically account for known confounders rather than relying on catching them by accident.
As organizations collect richer data with more measured variables, it’s become increasingly possible to identify and control for plausible confounders statistically, closing gaps that once could only be addressed through careful experimental design alone.
Growing awareness that a correlation “surviving” a naive check isn’t the same as surviving a rigorous confounder analysis has pushed serious causal investigation toward explicitly listing and testing for plausible confounders as standard practice.

The Metaphor, Fully Extended

Behind the Wall	Confounding Concept
Two fixtures losing pressure together, with no direct connection between them	Two variables correlated with each other, with no direct causal link
A single shared supply pipe feeding both fixtures	A single confounding variable influencing both correlated factors
Tracing the plumbing back to the real, shared source	Identifying and adjusting for the actual confounding variable
Realizing neither fixture affects the other at all	Realizing the correlation reflects a shared cause, not direct causation

For Beginners: What to Actually Do

Practice the ice-cream-and-drowning example until identifying a plausible confounder becomes a genuine reflex whenever you see a correlation.
Before accepting a causal claim, explicitly brainstorm what third factor might be driving both variables.
Learn the basic idea of statistically “controlling for” a confounder, even before diving into the specific methods covered later in this series.

For Practitioners and Leaders: The Deeper Layer

Require an explicit discussion of plausible confounders before any causal claim drawn from observational data is acted on.
Invest in richer data collection specifically to make confounder identification and adjustment more feasible.
Recognize that a correlation that “survives” an informal sanity check hasn’t necessarily survived a rigorous confounder analysis — the two are genuinely different bars.

Quick Recap

A confounding variable influences both a supposed cause and a supposed effect, creating a correlation without direct causation.
Confounding is the single most common reason a real, strong correlation isn’t actually causal.
Formal methods now let analysts systematically identify and adjust for known confounders.
Explicitly brainstorming plausible confounders should be a standard step before accepting any causal claim.

Where This Fits in the Series

Article 4 covered the hidden third pipe that explains a false connection. Article 5 introduces a formal way to map out the entire plumbing system at once.