Correlation and Causation

Why It Matters

Your brain is a pattern-matching machine. It is extraordinarily good at noticing when two things happen together. You eat a certain food and feel sick — you blame the food. A team wins every game when you wear your lucky socks — you credit the socks. A country passes a new law and the economy improves — people credit the law. In each case, you’ve spotted a correlation: two things that seem to move together.

But spotting a correlation is easy. Figuring out whether one thing is actually causing the other is one of the hardest problems in science, medicine, economics, and policy. Get it wrong, and you waste money on programs that don’t work, take medicine that doesn’t help, or blame innocent people for problems they didn’t create.

This lesson will make you permanently suspicious of claims that one thing causes another. That suspicion isn’t cynicism — it’s clear thinking. You’re learning to ask the right question: not “Do these two things happen together?” but “Is one of them actually making the other happen — and how would I know?”

A Story

The Homework Hypothesis

Ms. Rivera, the eighth-grade science teacher, showed her class a graph. It plotted two lines: the average amount of homework assigned per night and the average test scores at fifty different schools. The lines moved together almost perfectly. Schools that assigned more homework had higher test scores. Schools that assigned less homework had lower test scores.

“The data is clear,” she said. “More homework equals higher test scores. Right?”

Most of the class nodded. The graph looked convincing. But a student named Kai raised his hand. “Not necessarily. What if the schools that assign more homework are also richer schools with better teachers, smaller classes, and more resources? Maybe the homework isn’t causing the higher scores. Maybe being a well-funded school causes both more homework and higher scores.”

Ms. Rivera smiled. “That’s exactly the question I was hoping someone would ask. What Kai just described is called a confounding variable — a third factor that might be causing both things you’re observing. The correlation between homework and test scores is real. But the causation is uncertain, because there’s a plausible third factor that explains both.”

She showed a second graph. This one compared ice cream sales and drowning deaths across twelve months. They moved together almost perfectly: both went up in summer, both went down in winter. “Does ice cream cause drowning?” she asked.

The class laughed. “Of course not,” said a student named Jaya. “Hot weather causes both. People buy more ice cream when it’s hot, and more people swim when it’s hot, and more people drown when more people swim.”

“Right,” said Ms. Rivera. “But notice that the graph alone can’t tell you that. The data looks identical whether ice cream causes drowning or hot weather causes both. The numbers can’t distinguish between correlation and causation. You need to think about the mechanism — the actual process by which one thing would cause another. Is there a plausible way that buying ice cream makes someone drown? No. Is there a plausible way that hot weather leads to both? Yes. That’s how you start to untangle correlation from causation.”

Then she showed a third graph. This one showed that students who eat breakfast every morning score better on tests than students who skip breakfast. “Here’s the harder question,” she said. “Is this one correlation, causation, or could it be both? Think carefully. This is not as obvious as the ice cream example.”

Vocabulary

Correlation

A statistical relationship between two things — they tend to move together, either in the same direction or in opposite directions. Correlation tells you that a pattern exists, but it does not tell you why.

Causation

A relationship where one thing actually makes another thing happen. Causation is much harder to prove than correlation. Establishing causation requires showing a mechanism, ruling out confounding variables, and often conducting controlled experiments.

Confounding variable

A third factor that influences both of the things you’re comparing, creating the illusion that one causes the other. Hot weather is a confounding variable in the relationship between ice cream sales and drowning — it drives both, but neither causes the other.

Mechanism

The specific process by which one thing causes another. Asking “What is the mechanism?” is one of the best ways to test a causal claim. If you can’t explain how A causes B, you should be skeptical that it does.

Spurious correlation

A correlation that exists in the data but has no real causal connection. The number of movies Nicolas Cage appears in per year correlates with the number of people who drown in swimming pools. This is a real statistical correlation that means absolutely nothing.

Guided Teaching

Let’s start with why this error is so common. Your brain evolved to find causes. When our ancestors heard rustling in the bushes, the ones who assumed “predator” and ran survived more often than the ones who assumed “wind” and stayed. Your brain is wired to see causation even where only correlation exists. That instinct kept your ancestors alive, but it leads you astray when you’re trying to reason carefully about complex issues.

The fundamental problem is this: correlation is observable, but causation is invisible. You can see two things moving together. You cannot see one thing making another happen. You have to infer it. And inference can go wrong. When you see a graph showing that homework and test scores move together, your brain screams “homework causes better scores!” because that’s the causal story your brain constructs. But the data is equally consistent with several other stories. Can you think of at least two alternative explanations for the homework-test score correlation, besides “homework causes higher scores”?

There are three main ways that two correlated things can be related. First: A causes B. Maybe homework really does cause higher scores. Second: B causes A. Maybe schools with higher-scoring students assign more homework because those students can handle it. Third: C causes both A and B. Maybe school funding causes both more homework and higher scores, as Kai suggested. Until you’ve ruled out the second and third possibilities, you can’t claim the first. Why is the third option — the confounding variable — the hardest to spot?

Ms. Rivera’s breakfast example is deliberately harder. Unlike ice cream and drowning, there actually is a plausible mechanism by which eating breakfast could improve test performance: your brain needs glucose to function, and a fed brain works better than a hungry one. But there’s also a plausible confounding variable: families where kids eat breakfast every morning might also be more stable, more structured, and more supportive of education generally. Both stories fit the data. How would you figure out which explanation is correct? What kind of evidence would you need?

This is why scientists use controlled experiments. If you want to know whether breakfast causes better test scores, you can’t just compare students who eat breakfast with students who don’t, because those groups differ in many ways. You’d need to randomly assign students to a breakfast group and a no-breakfast group, controlling for everything else, and then compare their scores. That’s the only reliable way to isolate causation from correlation. And for many real-world questions — Does social media cause depression? Does gun control reduce violence? Does homework help learning? — truly controlled experiments are difficult or impossible to run.

Here’s a real-world application. You’ll often hear claims like: “States with stricter gun laws have fewer gun deaths, so gun laws reduce gun deaths.” Or the opposite: “States with more guns have less crime, so guns deter crime.” Both claims rely on correlations. Both might be confounded by other variables: poverty, urbanization, policing, culture. The data alone cannot settle these debates. What you need is a plausible mechanism AND control for confounding variables. Be skeptical of anyone who jumps from “these two numbers move together” to “one causes the other.”

One last point: correlation isn’t worthless. Correlation is a clue. It tells you where to look. If two things are correlated, there might be a causal relationship, and it’s worth investigating. But correlation is the beginning of the investigation, not the end. Treat every correlation as a question: “Why do these two things move together?” The answer might be causation. It might be a confounding variable. It might be coincidence. But the correlation itself doesn’t tell you which.

For Discussion

1. 1.What is the difference between correlation and causation? Why is the difference so important?
2. 2.In the homework example, what was Kai’s alternative explanation? How does a confounding variable create the illusion of causation?
3. 3.Why does your brain naturally assume causation when it sees correlation? Is that instinct ever useful?
4. 4.Ms. Rivera’s breakfast example is harder than the ice cream example. Why? Is it possible that breakfast both correlates with and causes better test scores?
5. 5.Can you think of a correlation in your own life that you assumed was causation? What might the confounding variable be?
6. 6.Why is saying “correlation doesn’t equal causation” not always a good response? When is it a genuine insight and when is it a cop-out?
7. 7.What would you need to prove — not just suggest — that one thing causes another?

Practice

Correlation Detective

1. 1.Here are four correlations. For each one, propose at least two possible explanations: one where A causes B, and one involving a confounding variable. Then decide which explanation you think is most likely and explain why.
2. 2.1. Countries with more television sets per capita have higher life expectancy.
3. 3.2. Students who play musical instruments tend to have higher grades.
4. 4.3. People who own more books tend to earn higher salaries.
5. 5.4. Cities with more police officers have higher crime rates.
6. 6.For each one, ask: What is the mechanism? Is there a plausible confounding variable? What experiment would settle the question?
7. 7.Bonus: Find a real correlation claim in the news this week. Bring it to a discussion and analyze it using the tools from this lesson.

Memory Questions

1. 1.What is the difference between correlation and causation?
2. 2.What is a confounding variable? Give an example from the lesson.
3. 3.What are the three possible relationships between two correlated things?
4. 4.What is a mechanism, and why is it important for testing causal claims?
5. 5.Why can’t a graph alone tell you whether a correlation is also a causation?
6. 6.When is “correlation doesn’t equal causation” a genuine insight, and when is it a cop-out?