Step 1 of 5 · Look at the data
Two lines that move together.
Here's real municipal data from one city, month by month. Ice cream sales and swimming-pool accidents rise and fall almost perfectly in step with each other.
“The pattern is real — I'm not tricking you with the data. The question is what you're allowed to conclude from it.”
Correlation coefficient: 0.94. Almost a perfect match — and almost meaningless on its own.
Step 2 of 5 · The concept
The hidden third thing.
The variable driving both is temperature. Hot weather sends people to buy ice cream and sends them into pools, where more swimming means more accidents. Neither trend causes the other; both are downstream of the weather.
Correlation
Cheap to find
A measurable statistical relationship. Everywhere in the news.
Causation
Hard to prove
Needs a mechanism, a control group, and the confounders ruled out.
Four tests before you accept a causal claim: plausible mechanism, correct time order, confounders ruled out, and it replicates.
Step 3 of 5 · Apply it
Judge these three claims.
For each one, decide whether the evidence described supports a causal claim or only a correlation. Then read why.
“Two of these are correlation, one is genuinely causal. Read the source line carefully — that's where the answer hides.”
“Students who eat breakfast get better grades — so schools should serve breakfast.”
Based on a survey of 4,000 students comparing self-reported breakfast habits with end-of-year grades.
Correlation only. Families with steady routines, more money and quieter mornings tend to produce both breakfast and better grades.
“This new vaccine reduces infections by 62%.”
Based on a randomised controlled trial: 20,000 volunteers, half given the vaccine and half a placebo, assigned by lottery.
Causal. Random assignment is what breaks the link to confounders — that is exactly what a controlled trial buys you.
“Cities that spent more on parks saw obesity rates fall.”
Based on comparing city budgets against public health records over ten years.
Correlation only. Wealthier cities spend more on everything — parks, clinics, food programmes — so the spending is a marker, not the cause.
Step 4 of 5 · Written response
Your turn to make the argument.
A news site publishes: “Towns with more libraries have lower crime rates. Building libraries reduces crime.”
In a short paragraph, name at least one confounding variable that could explain this pattern, and describe what evidence would be needed before you'd accept the causal claim.
0 characters · 60 minimum
Step 5 of 5 · Study & revise
Keep the framework.
The four tests are worth memorising — they'll serve you in science class, history essays and every argument you have online. Print the guide and keep it in your folder.
“Take the dataset and chart it yourself if you like. Checking someone else's numbers is a habit worth building early.”
Lesson complete!
That's every step of “Correlation is not causation” finished. Nova is impressed.