• Sociology: research methods

What is a spurious correlation?

Written with AI (How we create content)Target level: Grade 11Last updated: Report an error
Quick answer

A spurious correlation is a link between two variables that is not cause and effect: both change together because a third variable drives them. Towns with more libraries record more thefts, for example, only because bigger towns have more of both. Sociologists test for a spurious correlation by controlling for the third variable.

The topic explained

What a correlation is

Two variables are correlated when they change together across many cases. In a positive correlation both rise together: across Israeli workers, more years of schooling tend to go with higher pay. In a negative correlation one rises as the other falls. When there is no pattern, there is no correlation. Researchers sum up the strength of a straight-line link with the correlation coefficient r, which runs from −1, a perfect negative link, through 0, none, to +1, a perfect positive link.

A correlation is the starting point of much sociological research: it tells a researcher where to look. The misreading to avoid: a correlation describes a pattern across many cases, not a rule for each case. Plenty of people with long schooling earn little, and the correlation is still real. And a correlation, however strong, does not yet say which variable, if either, is the cause.

A table of three kinds of correlation: positive, where the other variable rises and r is near plus 1; negative, where it falls and r is near minus 1; and none, with no pattern and r near 0.CorrelationAs one risesValue of rPositivethe other rises+1Negativethe other falls−1Noneno pattern0

A correlation that is not a cause

Here is a made-up example. A researcher collects figures from six towns, A to F, and finds that the more libraries a town has, the more thefts it records each year. Town A has 2 libraries and 220 thefts; town F has 12 libraries and 1,140 thefts. The correlation is very strong, r = 0.96. Do libraries cause theft?

No. The link is a spurious correlation: both variables depend on a third variable, the size of the town. A town of 120,000 has more libraries, more shops, more cars and more thefts than a town of 20,000, simply because it has more of everything. The two variables move together, but neither moves the other. Notice that the figures are accurate: the towns really do have those libraries and those thefts. What is wrong is the conclusion drawn from them. A spurious correlation is not a mistake in counting but a mistake in explaining.

A scatter graph of six made-up towns: libraries on the horizontal axis and thefts a year on the vertical axis. The points rise from town A, 2 libraries and 220 thefts, to town F, 12 libraries and 1,140 thefts.LibrariesThefts a year24681012200400600800100012000ABCDEF
Made-up figures. Raw counts: r = 0.96.

Controlling for the third variable

To test a suspicion like this, sociologists control for the third variable: they hold it constant or take it out of the figures. Here, dividing by population turns counts into rates. Town A has 220 thefts for 20,000 residents, 220÷20=11220 \div 20 = 11 thefts per 1,000 residents. Town F has 1,140÷120=9.51{,}140 \div 120 = 9.5 per 1,000. The biggest town is not the most theft-prone at all.

Computed for all six towns, the correlation between libraries per resident and thefts per resident falls to r = 0.03, close to nothing. The strong link was an effect of size. The same logic explains the textbook case of fire damage: fires with more firefighters do more damage, because bigger fires bring both. Controlling does not always make a link vanish. Sometimes a correlation survives the control, which makes a causal story more believable, and sometimes it weakens without disappearing, which means the third variable explains only part of it.

A diagram: town population at the top, with arrows down to libraries and to thefts. A line between libraries and thefts is marked as a spurious link.Town populationLibrariesTheftsspurious link
The third variable drives both; the link between them is spurious.

Three conditions for saying cause

Research methods textbooks set three conditions before a study may claim that one variable causes another. First, the two vary together. Second, the cause comes first in time. Third, the link is not spurious: no third variable explains it. Experiments meet the third condition by assigning people at random, but sociologists usually cannot assign people to a family, a town or a class, so they compare groups and control for the variables they can measure.

Take schooling and pay again. Part of that link may run through a third variable, family background, which can raise both a person's schooling and their pay, so researchers compare people from similar backgrounds before they speak of the effect of education. The most common mistake in a bagrut answer is to read a cause straight off a correlation. A better answer names a possible third variable and says what data would test it.

Four boxes from top to bottom joined by arrows: they vary together, the cause comes first, no third variable, and then a causal claim.1. They vary together2. The cause comes first3. No third variableA causal claim
Free printable worksheet

Practice: spurious correlation

For 11th grade students of sociology. All the figures on this sheet are made up for practice. Answer from what is printed, then check your work against the answers.

Free, no sign-up. The answer key prints on the second page. Regenerating a new worksheet on the same topic needs a free account.

Free printable worksheetlessoons.com
Practice: spurious correlationName:
  1. Positive, negative or no correlation? (a) the older a car, the lower its price; (b) the more hours students study, the higher their grades; (c) shoe size and the month of birth.

  2. Fires with more firefighters at the scene cause more damage. Write the third variable in the empty box.

    A diagram with a question mark at the top and arrows down to firefighters and to damage, with a spurious link between them.?FirefightersDamagespurious link
  3. Calculate the thefts per 1,000 residents for towns A and F.

    A table of two made-up towns: town A with 20,000 residents and 220 thefts, town F with 120,000 residents and 1,140 thefts, and the rate per 1,000 left empty.TownResidentsTheftsPer 1,000A20,000220?F120,0001,140?
  4. The graph shows the six towns again, this time as rates. What does it show about the link between libraries and thefts?

    A scatter graph of the six made-up towns as rates: libraries per 10,000 residents on the horizontal axis, between about 0.9 and 1.25, and thefts per 1,000 on the vertical axis, between 9 and 11, with no pattern.Libraries per 10kThefts per 1,0000.811.291011ABCDEF
  5. A survey finds that students who eat breakfast get higher grades. Which of the three conditions for a causal claim is most in doubt, and why?

  6. Write the missing condition.

    The four boxes of the conditions for a causal claim, with the second box marked with a question mark.1. They vary together?3. No third variableA causal claim
  7. Design a check: a researcher finds that neighborhoods with more police officers record more crime. Name a possible third variable and say what data you would collect to test it.

More topics and worksheets at lessoons.com© 2026 Lessons

Common questions

What is the difference between correlation and causation?

A correlation means two variables change together. Causation means a change in one produces a change in the other. A correlation is needed for causation but is not enough: the cause must also come first, and no third variable may explain the link.

What is a third variable?

A third variable is a factor that affects both variables in a correlation and so makes them move together. In the made-up towns, population size drives both the number of libraries and the number of thefts.

What does it mean to control for a variable?

To control for a variable is to remove its effect, by holding it constant, comparing like with like, or turning counts into rates. If the correlation disappears once the variable is controlled, the original link was spurious.

Can a correlation be negative?

Yes. In a negative correlation one variable rises as the other falls, as with the age of a car and its price. The correlation coefficient is then below 0, down to −1 for a perfect negative link.

Why can sociologists rarely run experiments?

Sociologists study families, towns, classes and religions, and they cannot assign people to them at random. They compare groups that already exist and control statistically for the differences they can measure, which leaves room for variables they did not measure.

Still have a question? Ask Lessi

Any question about this subject. Lessi answers right here, step by step.

Try asking

Asking needs a free account; your question is kept while you sign up. Answers are written by AI and can contain mistakes. Lessi doesn’t give medical or legal advice.

Learn it with a teacher

Learn this topic live with Lessi

Lessi, the AI teacher, explains it out loud on a board, with short quizzes, at your pace. The first 60 minutes are free.