calculatorgate.com
Home
/Ice cream and shark attacks: when correlation lies

Ice cream and shark attacks: when correlation lies

Two things rising together doesn't mean one causes the other. Here's how to read correlation correctly and avoid the most common traps, with simple examples.

Published: · 3 min read · Written by the CalculatorGate team

Ice cream sales go up in summer. So do shark attacks. Put the two on a chart and they move together beautifully. Does ice cream attract sharks? Of course not. Both have the same cause: hot weather sends more people to the ice cream stand and into the sea.

That example sums up the most important rule in statistics: correlation is not causation.

What correlation tells you

The correlation coefficient (r) measures how closely two things move together, on a scale from −1 to +1:

  • Close to +1: when one goes up, so does the other.
  • Close to −1: when one goes up, the other goes down.
  • Close to 0: no straight-line relationship.

Say five students studied 1, 2, 3, 4 and 5 hours and scored 52, 58, 61, 70 and 74. The correlation is r = 0.99, a very strong link. Here we have good reason to believe studying raises scores, but that belief comes from what we know about studying, not from the number itself.

Four common traps

  • A hidden cause: like heat behind ice cream and shark attacks, a third factor drives both.
  • Reverse direction: people in hospitals are sicker than average, but hospitals don't make them sick; they're there because they're ill.
  • Pure chance: with little data, striking links appear by luck. Cheese consumption and the number of people in some job can rise together for years with no connection at all.
  • A narrow range: looking at only part of the data can make a link look stronger or weaker than it really is.

Questions to ask when you read a headline

  • Could something else be driving both?
  • Could cause and effect run the other way?
  • How many people or data points were studied?
  • Was there an experiment, or only observation?

Headlines like "coffee drinkers live longer" usually come from observation. Coffee drinkers may differ in other habits too. The strongest way to show cause and effect is a controlled experiment where people are randomly split into groups.

So is correlation useless?

Far from it. Spotting a relationship is the first step toward finding its cause, and a regression line can make reasonable predictions from past data. Just keep the difference between "rise together" and "one causes the other" in mind.

How to read a correlation coefficient

Exact cut-offs differ by field, but a rough guide is:

  • 0 to 0.3 (plus or minus): weak relationship
  • 0.3 to 0.7: moderate relationship
  • 0.7 to 1: strong relationship

There's also R², the square of the correlation coefficient. In the study example r = 0.99 and R² = 0.98, meaning about 98% of the variation in scores lines up with study time. Real-life numbers are rarely that high; in the social sciences, correlations of 0.3–0.5 are often considered meaningful.

One outlier can change everything

Add a sixth student to the same group: they studied 6 hours but were ill on exam day and scored 40. That single value drags the correlation from 0.99 down to almost zero, about −0.01. With small data sets, one or two unusual values can flip the result. That's why it's always worth plotting your data on a chart and checking for odd values before trusting a correlation.

The correlation and regression calculator shows how strongly two data sets move together and draws the regression line. The statistics calculator summarizes your data, and the sample size calculator tells you how much data you need for reliable results.