Correlation vs. Causation — I can distinguish correlation...

SECTION 1

Historical Context & Motivation

Humans are natural pattern-seekers. For thousands of years, people noticed that events often occur together and assumed one must cause the other. Ancient civilizations believed that comets caused plagues, and for centuries doctors thought that "bad air" caused diseases like malaria (the name literally means "bad air" in Italian). These errors persisted because people confused correlation — two things happening together — with causation — one thing actually making the other happen. The history of science is full of breakthroughs that came when researchers finally teased these two ideas apart.

1747

Lind's Scurvy Trial

James Lind conducted one of the first controlled experiments by giving different treatments to groups of sailors with scurvy, establishing that citrus fruit actually caused recovery rather than merely being correlated with it.

1854

Snow's Cholera Map

John Snow mapped cholera cases in London and traced the disease to contaminated water pumps, disproving the popular "miasma" (bad air) theory by showing it was water — not air — that caused the illness.

1950

Smoking & Lung Cancer Studies

Researchers Doll and Hill published landmark observational studies linking smoking to lung cancer. The tobacco industry argued the link was merely correlational for decades, highlighting how important it is to distinguish the two concepts.

1965

Hill's Criteria for Causation

Sir Austin Bradford Hill published nine criteria for evaluating whether a correlation reflects a causal relationship, giving scientists a formal framework that is still used today.

2000s

The Age of Spurious Correlations

With massive data sets and the internet, researchers (and comedians) began publishing absurd correlations — like the correlation between U.S. cheese consumption and the number of people who died tangled in bedsheets — to remind the public that correlation does not imply causation.

The central question this lesson addresses is deceptively simple: When two variables move together, how do we know if one is actually causing the change in the other? Answering this question correctly is one of the most important skills in statistics and in everyday life — from evaluating medical claims to understanding headlines.

SECTION 2

Core Principles & Definitions

Before you can tell correlation and causation apart, you need rock-solid definitions of each term and the key ideas that surround them. The four foundational concepts below form the backbone of this entire lesson.

Correlation

A statistical relationship between two variables — when one changes, the other tends to change in a predictable direction. Correlation can be positive (both increase), negative (one increases while the other decreases), or zero (no pattern).

Causation

A relationship in which a change in one variable directly produces a change in the other. Causation implies correlation, but correlation alone does NOT prove causation.

Lurking (Confounding) Variable

A hidden third variable that influences both of the variables being studied, creating the illusion of a direct relationship between them. Also called a confounding variable.

Study Design Matters

Only a well-designed randomized controlled experiment can establish causation. Observational studies, surveys, and anecdotal evidence can only show correlation — no matter how strong the pattern appears.

✦ KEY TAKEAWAY

Think of correlation like seeing two friends always arrive at school at the same time. You might assume one drives the other, but maybe they just live near each other and both leave when the morning bell is about to ring. The bell is the lurking variable — it causes both of them to show up, creating the illusion that one friend's arrival causes the other's. Correlation shows you the pattern; only careful investigation reveals the cause.

SECTION 3

Visual Explanation — Correlation Is Not Causation

The diagram below illustrates the three most common relationship structures you will encounter when analyzing data. Understanding these structures is the key to deciding whether a claim of causation is justified.

The three panels show direct causation, a confounding (lurking) variable, and coincidence. The flowchart below them summarizes the decision process: causation claims require both a correlation and a randomized controlled experiment.

Notice the critical difference between panels 1 and 2 in the diagram. In both cases, you would observe the same correlation between Variable A and Variable B. From the data alone, the scatter plots would look identical. The only way to rule out a lurking variable is through the design of the study itself — specifically, whether subjects were randomly assigned to groups. The flowchart at the bottom captures this logic: first confirm a correlation exists, then ask whether the study was a randomized experiment before concluding causation.

SECTION 4

Mathematical Framework — The Correlation Coefficient

While the distinction between correlation and causation is primarily a logical one, the strength of a correlation is measured with a number. The correlation coefficient, denoted r, quantifies how tightly two variables follow a linear pattern. Understanding r helps you evaluate the strength of a claim, but remember: even a perfect r = 1 does not prove causation.

CORRELATION COEFFICIENT

r = (1 / (n − 1)) × Σ [(xᵢ − x̄) / sₓ] × [(yᵢ − ȳ) / sᵧ]

Where n = number of data pairs, x̄ and ȳ are the means of the x- and y-values, and sₓ and sᵧ are their standard deviations. The result always falls between −1 and +1.

INTERPRETING r

−1 ≤ r ≤ +1

r = +1 means a perfect positive linear relationship. r = −1 means a perfect negative linear relationship. r = 0 means no linear relationship. Values close to ±1 indicate strong correlation; values near 0 indicate weak or no correlation.

Strength of Correlation (|r|)

None / Very Weak

Weak

Moderate

Strong

Very Strong

0.0

0.2

0.4

0.6

0.8

1.0

r = 0r = ±1

⚠️ Important Reminder

A strong r value tells you the correlation is strong, but it tells you nothing about whether the relationship is causal. The correlation between ice cream sales and drowning deaths is about r = 0.85 — very strong! — but ice cream clearly doesn't cause drowning. The lurking variable is summer heat, which increases both.

SECTION 5

Study Types & Their Limits

Whether a study can support a causal claim depends entirely on how the study was designed. Two broad categories of studies exist: observational studies and experiments. Understanding the difference is crucial for justifying — or rejecting — causal claims.

In an observational study (left), participants self-select into groups, so lurking variables may differ between groups. In a randomized experiment (right), random assignment distributes lurking variables evenly across groups, isolating the effect of the treatment.

Key differences between observational studies and randomized experiments

Feature	Observational Study	Randomized Experiment
Group Assignment	Self-selected or pre-existing	Randomly assigned by researcher
Lurking Variables	NOT controlled — may differ between groups	Balanced across groups by randomization
Conclusion	Can show correlation only	Can support causation
Example	Survey: "People who exercise report less stress."	Half of participants are randomly assigned to an exercise program; stress levels are compared.

A helpful mnemonic is: "Random assignment → causation claims allowed. No random assignment → correlation claims only." This is the single most important rule for evaluating study descriptions on quizzes, standardized tests, and in real life.

SECTION 6

Worked Example — Evaluating a Study Claim

Let's walk through a realistic scenario step by step. This is the kind of question you'll see on tests and AP exams.

📰 Scenario

A school newspaper reports: "Students who eat breakfast score an average of 12 points higher on math tests. Therefore, eating breakfast causes better math performance." The data comes from a survey of 400 students who voluntarily reported whether or not they eat breakfast.

Is the Causal Claim Justified?

Step 1 — Identify the Study TypeThe data comes from a survey. Students were not randomly assigned to eat or skip breakfast — they chose for themselves. This makes it an observational study.

Study type: Observational (survey)

Step 2 — Check for Random AssignmentBecause students self-selected into the "breakfast" and "no breakfast" groups, there was no random assignment. Without random assignment, lurking variables are not controlled.

Random assignment: No

Step 3 — Identify Possible Lurking VariablesStudents who eat breakfast may also come from families with more resources, have more structured morning routines, get more sleep, and have parents who are more involved in their education. Any of these factors could be the real cause of higher test scores.

Possible lurking variables: socioeconomic status, sleep, parental involvement

Step 4 — State the Correct ConclusionThe data shows a correlation between eating breakfast and higher math scores. However, because this is an observational study with no random assignment, we cannot conclude that eating breakfast causes higher scores.

Verdict: The newspaper's causal claim is NOT justified. The correct statement is: "Eating breakfast is associated with higher math scores."

Step 5 — Suggest a Better Study DesignTo test causation, researchers could randomly assign 400 students into two groups: one group required to eat a provided breakfast and one group that skips breakfast. Both groups would then take the same math test. If the breakfast group scored significantly higher, a causal claim would be more justified.

A randomized controlled experiment would be needed to support causation.

SECTION 7

Common Traps & How to Avoid Them

Even after learning the definitions, students often fall into predictable traps when analyzing claims. The table below lists the most common mistakes and how to steer clear of them.

Five common traps when evaluating correlation and causation

Common Trap	Why It's Wrong	What to Do Instead
"The correlation is very strong (r = 0.95), so it must be causal."	Strength of correlation says nothing about causation. Spurious correlations can be very strong.	Ask about study design first. A strong r with no experiment is still just correlation.
"The study had thousands of participants, so it proves causation."	Sample size affects precision, not the type of conclusion you can draw. A big survey is still a survey.	Check for random assignment, regardless of sample size.
"They used a control group, so it's an experiment."	An observational study can compare groups without randomization. A control group alone doesn't make it an experiment.	Verify that participants were randomly ASSIGNED to groups by the researcher.
"Reverse causation isn't possible here, so it must be A → B."	Even if reverse causation is unlikely, a lurking variable could still explain the relationship.	Always consider BOTH reverse causation AND lurking variables before claiming causation.
"Correlation does not imply causation" means the variables are unrelated."	The phrase means you can't PROVE causation from correlation alone. The variables may still be causally related — you just can't confirm it without an experiment.	Say "correlation does not PROVE causation," not "correlation means no relationship."

✦ KEY TAKEAWAY

Think of it like a courtroom. Correlation is like circumstantial evidence — it raises suspicion, but it isn't enough to convict. A randomized experiment is like DNA evidence — it directly links the suspect (variable A) to the crime (change in variable B). You wouldn't convict someone on circumstantial evidence alone, and you shouldn't claim causation based on correlation alone.

SECTION 8

Connection to Advanced Topics

The correlation-vs.-causation distinction you're learning now is the foundation for more advanced statistical reasoning. In AP Statistics, college courses, and research careers, these ideas expand into formal frameworks for determining cause and effect.

How today's concepts connect to future coursework

What You Know Now (Math 1)	Where It Leads (Advanced)
Correlation coefficient r measures linear association.	In AP Statistics, you'll learn regression (ŷ = a + bx) to make predictions and r² to measure explained variation.
Lurking / confounding variables create misleading associations.	College statistics introduces Simpson's Paradox, where a trend reverses when data is separated by a lurking variable.
Randomized experiments can establish causation.	Advanced courses cover experimental design in detail: blocking, blinding, placebos, factorial designs, and statistical significance tests.
You evaluate claims by identifying study type.	Research methods courses teach meta-analysis, where scientists combine many studies to build stronger evidence for or against causation.

The reasoning skill you're building — asking "Was this a randomized experiment?" before accepting a causal claim — applies far beyond math class. It's essential for evaluating news articles, medical advice, advertisements, and social media posts. Mastering this now gives you a critical-thinking advantage that will serve you in every subject and in daily life.

SECTION 9

Practice Problems

Test your understanding with these five problems. Each one asks you to analyze a claim or scenario and decide whether causation is justified. Answers include detailed explanations.

PROBLEM 1 — CONCEPTUAL

Explain in your own words why "correlation does not imply causation." Include an example of two variables that are correlated but clearly not causally related.

PROBLEM 2 — BASIC CALCULATION

A data set of 8 cities shows that as the number of parks increases, the crime rate tends to decrease. The correlation coefficient is r = −0.72. A city council member says, "Building more parks reduces crime." (a) What does r = −0.72 tell you about the relationship? (b) Is the council member's causal claim justified? Why or why not?

PROBLEM 3 — INTERMEDIATE

A pharmaceutical company tests a new headache medication. They randomly assign 500 participants to receive either the new medication or a placebo (sugar pill). After two hours, 78% of the medication group reports headache relief compared to 42% of the placebo group. Can the company claim the medication caused the improvement? Justify your answer by referencing the study design.

PROBLEM 4 — APPLIED

A news headline reads: "Teens who spend more than 3 hours per day on social media are twice as likely to report anxiety, according to a new study of 10,000 teenagers." The study was based on an anonymous online questionnaire. (a) Identify the study type. (b) Name two possible lurking variables. (c) Rewrite the headline to accurately reflect the study's limitations.

PROBLEM 5 — CRITICAL THINKING

A researcher wants to determine whether listening to classical music while studying causes students to earn higher grades. Describe a study the researcher could design that would allow a causal conclusion, and explain why each element of your design is important. Then explain why an observational study on this question would be insufficient.

SUMMARY

Lesson Summary

Correlation means two variables share a statistical pattern — when one changes, the other tends to change. Causation means one variable directly produces a change in the other. A lurking (confounding) variable is a hidden third factor that can create a misleading correlation between two variables that aren't actually causally connected. The correlation coefficient (r) measures the strength and direction of a linear relationship on a scale from −1 to +1, but even a perfect r says nothing about causation.

The only study design that can support a causal claim is a randomized controlled experiment, in which participants are randomly assigned to treatment and control groups. Observational studies — including surveys and data comparisons — can only establish correlation because they do not control for lurking variables. When evaluating any claim, always ask: "Was there random assignment?" If the answer is no, the correct language is "is associated with" or "is correlated with" — never "causes."