Not sure which lesson to do first? Find your level and get a plan →

Statistics · Grade 6-8

Correlation vs Causation

Correlation shows that two variables move together in some pattern; causation means one variable actually makes the other change.


Quick Answer

Correlation shows that two variables move together in some pattern; causation means one variable actually makes the other change. Use the distinction whenever a claim jumps from "these two go together" to "this one makes that happen". A link in data can come from a hidden common cause (hot weather drives both ice cream sales and swimming accidents), from cause running backward, or from coincidence. A randomized experiment, where chance decides who gets the treatment, is the standard way to show that a cause is real.

Why This Matters

Health headlines, ads and policy arguments often leap from a link to a cause: "people who drink red wine live longer, so drink wine". Knowing the difference lets you ask the right follow-up: what else differs between those groups, and which way could the arrow point? Without it, you might shut down a tutoring program that helps, or pay for a product that does nothing.

Intuitive Explanation

Across a year, a beach town's ice cream sales and its number of lifeguard rescues rise and fall together: both peak in July and bottom out in January. The correlation is real. But selling more ice cream does not put swimmers in danger. Hot weather sends more people to the beach, where they buy ice cream and also swim. Heat is a hidden common cause, called a confounder, and it creates the link on its own.

A correlation can appear without the cause you suspect in three usual ways. A confounder drives both, like heat, or like age behind children's foot length and reading level. The cause runs backward: students who use a tutoring center have lower grades because struggling students are the ones who go for help. Or the link is coincidence, which is likely when someone searches through many variables and reports the one pair that happened to match.

Compare that with an experiment. Researchers flip a coin for each of 100 tomato plants to decide which ones get fertilizer. Chance, not the gardener, chooses, so sunny spots and good soil end up spread across both groups. If the fertilized plants then yield more, the fertilizer is the only systematic difference left. So ask of any study who decided which individuals got the treatment: chance, or the individuals themselves.

Think of it like this. Correlation versus causation is like seeing wet streets and open umbrellas together: they show up at the same time, but neither causes the other, because rain causes both. To learn what really causes what, you have to change one thing yourself and watch, which is what an experiment does.

Core idea. A correlation shows that two variables move together; it becomes evidence of cause only when the other explanations (a common cause, a reversed arrow, coincidence) are ruled out, which a randomized experiment does best.


When to Use

Apply this whenever a claim uses data on what people already do to say that something causes, leads to, boosts or prevents something else. Ask whether the groups chose themselves, what third variable could drive both, and whether the arrow could point backward. If the study randomly assigned the treatment, a causal reading is reasonable; if the question only asks for the direction or strength of a link, use correlation tools instead.

Tip. Ask: Does a claim say one variable makes another change, when the evidence is only data showing the two go together?

How to Recognize It

Before accepting or rejecting a cause, check how the data were gathered and list what else could explain the link.

  1. Did the researchers decide by chance who got the treatment, instead of recording what people already did?

    Yes means it was a randomized experiment, so a cause can be supported; the details belong to Experimental Design. No means the data are observational, and the link alone cannot show a cause.

  2. Could a third thing, like age, weather or family money, push both variables up or down together?

    Yes means a lurking variable may create the link; Confounding Variables explains how to spot and handle one. If nothing plausible drives both, still check which way the arrow could point.

  3. Could the second variable be causing the first instead?

    Yes means reverse causation is possible: happier people may exercise more, rather than exercise making people happier. If the question only asks whether the variables rise or fall together, describe it as a Correlation and stop there.

  4. Is the link based on only a few years of data or a handful of cases?

    Yes means coincidence is a real possibility, since two unrelated counts can match by luck over a short stretch. Data from other places and times help rule chance out.

Correlation vs Causation vs Correlation vs Confounding Variables vs Experimental Design

All four come up when two things happen together, and people slide from one to the next. Ask what you are being asked to do: describe the link, judge a cause, name a hidden driver, or plan a fair test.

Correlation vs Causation

Meaning
Correlation vs causation is the judgment of whether a link between two variables shows that one actually changes the other.
Key test
A cause is claimed, and you must decide whether the evidence supports it.
Example
People in countries with more TVs per person live longer on average, but wealth drives both, so TVs are not the cause.

Correlation

Meaning
Correlation is the plain description that two measured variables tend to move together or in opposite directions.
Key test
Only describes the link; makes no claim about why it exists.
Example
More hours of sunshine go with more ice cream sold.

Confounding Variables

Meaning
A confounding variable is a third factor that affects both variables and can create a link between them by itself.
Key test
Names the hidden third factor behind a link.
Example
Wealthier families can pay for more music lessons and more tutoring, so lessons can look like the cause of higher scores.

Experimental Design

Meaning
Experimental design plans a study in which the researcher assigns treatments, ideally at random, so that a cause can be tested fairly.
Key test
The researcher controls who gets the treatment, usually by chance.
Example
A coin flip decides which half of a class uses a new study app for a month.

Formula & Notation

Worked Examples

Example 1 — Foot length and reading level

Easy

In a school, children aged 6, 8 and 10 have average foot lengths of 17, 19 and 21 cm and average reading levels of 1, 3 and 5. Among the 8-year-olds alone, foot lengths of 18, 19 and 20 cm go with reading levels 3.2, 2.9 and 3.1. Do bigger feet cause better reading?
Show the steps and the answer
  1. Across all ages, foot length and reading level rise together: 17, 19, 21 cm with levels 1, 3, 5.

    The correlation is real and positive, so the question is what explains it.

  2. Look for a third variable linked to both: age.

    Older children have grown more and have also had more years of reading practice, so age can push both numbers up.

  3. Hold age fixed: among 8-year-olds, levels 3.2, 2.9 and 3.1 show no trend as feet go from 18 to 20 cm.

    If feet mattered, longer feet would still mean higher levels inside one age group; the link vanishes, so age was driving it.

Answer: No. The correlation comes from age, a confounder that drives both foot length and reading level.

Takeaway: Before believing a cause, name a variable that could drive both, then check the link inside groups where that variable is the same.

Example 2 — When a causal claim is justified

Standard

Researchers use a coin flip to assign each of 100 tomato plants to fertilizer or no fertilizer, 50 in each group, and otherwise treat all plants alike. Fertilized plants average 1.8 kg of tomatoes and unfertilized plants 1.2 kg. Can they say fertilizer caused the bigger harvest?
Show the steps and the answer
  1. Ask who chose the treatment: a coin flip, not the plants' location or the gardener.

    When chance assigns treatment, differences in sun, soil and seed quality are spread evenly across the groups on average.

  2. Find the difference: 1.8−1.2=0.6 kg per plant.

    With the groups alike apart from the treatment, the fertilizer is the only systematic reason for a gap.

  3. Allow for luck: the coin may have put a few more strong plants in one group, so a gap of 0.6 kg, half again the unfertilized yield, is more convincing than a gap of 0.05 kg would be.

    Random assignment removes systematic differences but leaves small chance ones; later, a hypothesis test measures how big a gap luck alone could make.

Answer: Yes, reasonably. This is a randomized experiment, so the 0.6 kg difference is evidence that the fertilizer caused the larger harvest.

Takeaway: The difference from a correlation is the design: chance, not choice, decided who got the treatment.

Example 3 — Find the mistake: the arrow points backward

Application

At one school, students who use the tutoring center average 72% on exams, while students who never use it average 81%. Sam writes: "Tutoring lowers grades by 9 points, so the school should close the center." Find the mistake and correct it.
Show the steps and the answer
  1. Find the gap Sam saw: 81−72=9 points.

    The difference is real; the question is what it means.

  2. Ask who chooses tutoring: mostly students who are already struggling.

    Low grades lead students to the center, so the arrow can run from grades to tutoring, the reverse of Sam's claim.

  3. Name a fair check: compare the same students' grades before and after they start tutoring, or randomly offer tutoring to some struggling students and not others.

    These designs compare like with like, so a change can be credited to tutoring itself.

Answer: The 9-point gap is an association that most likely runs the other way: weaker students seek tutoring. It does not show that tutoring lowers grades, and it is no reason to close the center.

Takeaway: When a group chose its treatment because of the outcome, suspect reverse causation.

Where You'll See It

  • Ice cream sales and shark attacks

    Both rise in summer, but ice cream doesn't summon sharks — hot weather (a confounder) drives beach crowds and both numbers, showing correlation without causation.

  • Bigger feet and better reading

    Among all kids, shoe size correlates with reading skill, but age is the hidden cause: older kids have bigger feet AND read better. The link is confounded, not causal.

  • Night light and nearsightedness claim

    An old study linked babies' night lights to needing glasses later, but nearsighted parents (who tend to use night lights) pass on the trait — a confounded link, not a direct cause.

Common Mistakes

  1. The slip: If two things rise and fall together, one must cause the other.

    Why it's tempting: A tight, repeatable pattern feels like proof of a mechanism.

    Instead: A confounding variable can drive both, or the link can be coincidence. Only a controlled experiment that isolates the variable can establish causation.

  2. The slip: If A is correlated with B and A came first, A causes B.

    Why it's tempting: Time order seems to settle the direction of cause.

    Instead: Time order rules out reverse causation but not a lurking common cause. You still need to control for confounders before claiming A causes B.

  3. The slip: Finding a possible confounder proves there is no real causal link.

    Why it's tempting: Once a third factor is named, the original link feels debunked.

    Instead: A confounder shows the correlation isn't sufficient proof of causation — it doesn't prove the absence of a causal effect either. The honest claim is 'undetermined from this data.'


Mini Practice

Try these on your own. Tap Reveal when you want to check.

  1. In a town, monthly ice cream sales rise from 200 in March to 900 in July, and swimming-pool injuries rise from 5 to 22. Name a hidden variable that could explain both.

    Hint: What changes between March and July that affects both activities?

  2. A study finds students with more than 50 books at home score 15 points higher on reading tests. Give one possible confounder, and one way the cause could run in reverse.

    Hint: Think about family background, and about what strong readers might ask for.

  3. City A has 50 firefighters and $2 million in yearly fire damage. City B has 200 firefighters and $9 million in damage. Does having more firefighters cause more fire damage?

    Hint: What is different about the two cities besides firefighters?

  4. Researchers randomly assign 100 volunteers to drink green tea or water daily. After 8 weeks the tea group's blood pressure fell 6 points more. Can this study show causation, or only correlation?

    Hint: What does random assignment do to other differences between the groups?

  5. Across 20 countries, chocolate eaten per person and Nobel prizes per person have a correlation of 0.79. Does eating chocolate make people win Nobel prizes?

    Hint: Look for something about a country that could raise both numbers.

  6. A survey finds that older adults who use walking canes fall more often than those who do not. Ryan writes, "Canes cause falls, so people should stop using them." Correct his reasoning.

    Hint: Could the cause run the other way?

Want the full set?

50 practice questions for this concept — free to try, every one with a complete worked solution showing the why, not just the answer.

Frequently Asked Questions

If correlation doesn't prove causation, how do we know smoking causes cancer?

Researchers could not ethically assign people to smoke, so they built the case from many kinds of evidence. The link showed up in study after study, in many countries. Heavier smokers got lung cancer more often than light smokers, quitting lowered the risk over time, smoking came before the disease, and lab work showed that chemicals in smoke damage cells. No single correlation proved it, but together these ruled out the other explanations.

Can one thing cause another even when the data show no correlation?

Yes, when something else hides the effect. Studying more does raise grades, but in some classes the students who are struggling study the most, so in the data, study time and grades can look unrelated or even negatively linked. The real effect is there, buried under who chooses to study. This is another reason a fair experiment, not a glance at the data, is needed to test a cause.

Can two things be linked just by coincidence?

Yes, especially over a short stretch of time. One well-known chart shows the number of films Nicolas Cage appeared in each year rising and falling with the number of people who drowned in U.S. swimming pools from 1999 to 2009. Nobody thinks one causes the other. With millions of yearly counts out there, some will match by luck, so a link that keeps showing up in new data is far more convincing than one that appeared once.

Does a randomized experiment always prove a cause?

It is the strongest evidence, but not a guarantee. With small groups, chance alone can make one group do better, which is why scientists repeat experiments and use statistics to judge whether a difference is too big to be luck. The result also applies to the kind of subjects tested: a study app that helped 8th graders in one school might not help adults learning a language.

Learning Path

Correlation vs Causation

You are here

You need Correlation (variables moving together) and Data Collection (how the data were gathered) first. This concept separates a link from a cause and names the usual alternative explanations: confounders, reverse causation and coincidence. Next, Experimental Design shows how random assignment makes a causal claim fair, and Confounding Variables looks closely at hidden third factors.

See Also