Correlation != Causation

We’ve all seen the headlines:

• “Drinking red wine makes you live longer!”

• “People who make their beds are more successful!”

• “Eating chocolate improves cognitive function!”

These articles are usually based on a real statistical observation: two things are happening at the same time. But they almost always fall into the oldest trap in data science—assuming that because two trends move together, one must be causing the other.

Welcome to the world of Correlation vs. Causation, and the invisible puppeteer that tricks us all: Confounding Theory.

1. The Basics: Correlation vs. Causation

Before we look at the trap, we need to define our terms.

Correlation is simply a relationship between two variables. When Variable A goes up, Variable B also goes up (positive correlation). Or, when Variable A goes up, Variable B goes down (negative correlation).

Causation is a step further. It means Variable A directly causes the change in Variable B.

The Golden Rule of Statistics: Just because two things correlate, does not mean one causes the other. They might be completely coincidental.

For example, if you look at the data from 1999 to 2009, the number of people who drowned by falling into a pool is highly correlated with the number of films Nicolas Cage appeared in. Unless Nic Cage movies are somehow physically pushing people into water, this is what we call a Spurious Correlation—a mathematical coincidence.

2. Enter “Confounding Theory” (The Invisible Puppeteer)

Coincidences are funny, but they aren’t the real danger in data analysis. The real danger is Confounding.

A Confounding Variable (often called a confounder or a lurking variable) is a third, unmeasured variable that influences both the supposed cause and the supposed effect. It creates a powerful, mathematically real correlation that tricks you into believing a direct causal link exists when it doesn’t. The Classic Example: Ice Cream and Shark Attacks

Imagine you are analyzing summer data for a coastal town. You notice a terrifying trend: As ice cream sales increase, shark attacks increase. The correlation is incredibly strong.

If you don’t understand confounding, you might conclude:

• Theory A: Eating ice cream makes humans taste better to sharks.

• Theory B: Sugary ice cream makes people swim erratically, attracting predators.

Therefore, to save lives, the mayor bans ice cream.

Obviously, this is absurd. Banning ice cream won’t stop shark attacks. Why? Because there is a Confounding Variable: Temperature (The Weather).

When the weather gets hot:

• People buy more ice cream.

• People go swimming in the ocean more often (exposing themselves to sharks).

Ice cream and shark attacks have zero causal relationship. They are completely independent. But because they are both caused by the same hidden third variable (heat), they move up and down together perfectly.

1. How Do We Actually Prove Causation?

If data can be manipulated by hidden variables, how do scientists ever prove that smoking causes cancer, or that a drug cures a disease?

Randomized Controlled Trials (RCTs)

This is the gold standard. You take a large group of people and randomly split them in half. Group A gets the drug, Group B gets a placebo. Because the split is entirely random, all potential confounders (age, diet, genetics) should be equally distributed between the two groups. If Group A gets better, you can confidently say the drug caused it.

Controlling for Variables

If you can’t run an experiment (you can’t lock people in a room and force them to smoke for 20 years), you have to use observational data. Statisticians use math to “control” or “adjust” for known confounders. They might look at smokers and non-smokers who have the exact same age, diet, and exercise habits to isolate the effect of the smoke.

Temporal Precedence:

The cause must happen before the effect.

Plausible Mechanism

There must be a logical, scientific explanation for how A causes B.

Conclusion: Be a Data Skeptic

The next time you read a headline claiming that doing X will result in Y, pause. Ask yourself: “Is there a third variable driving both of these?”

In a world driven by data, understanding the difference between correlation and causation is your best defense against bad science, misleading marketing, and false promises.