Correlation vs Causation: The Difference, With Everyday Examples

What is the difference between correlation and causation?

The difference between correlation vs causation comes down to one question: does one thing produce the other, or do the two just show up together? A correlation is a pattern. Causation is a mechanism: if you changed the first thing, the second would change too.

The Australian Bureau of Statistics makes the point plainly. Ice cream sales and sunscreen sales climb and fall together over the year, but neither one drives the other. The season does. A correlation measures how strongly two things move together; on its own it says nothing about cause and effect.

Why correlation is not causation: the usual suspects

When two things go together, there are several explanations besides 'A causes B'. Before you blame the obvious suspect, run through these.

A third factor (confounder)

Something else drives both. The JMP statistics guide describes data where people who exercise more also have more skin cancer; the likely link is sunshine, which brings both outdoor exercise and sun exposure. At home, a teen's screen time and slipping grades might both come from poor sleep, a brutal semester or a friendship falling apart.

Reverse causation

The arrow points the other way. Windmills spin faster when the wind is stronger, but the windmills don't make the wind. Maybe the phone isn't sinking the grades; maybe the phone is where a kid hides because school already feels hopeless.

Both directions

Each feeds the other. A couple who texts less may feel less close, and feeling less close may lead to fewer texts. Asking which one started it rarely helps.

Coincidence

With enough data, unrelated things line up by chance. Tyler Vigen's Spurious Correlations project shows Google searches for 'cat memes' tracking airbag recalls, found by comparing thousands of variables until some happened to match.

Correlation vs Causation examples

How it shows up in everyday arguments

Most of us don't argue with spreadsheets. We argue with patterns we've noticed: 'Every time you go out with them, you come home in a bad mood.' 'Whenever your sister calls, we end up fighting.' 'Since you started gaming again, you never help with dinner.' The pattern may be real. The leap is from 'these go together' to 'this one causes that one, so it has to stop.'

The trouble is that the conclusion usually points at a person or a thing to ban: the friends, the phone, the sister. If the real driver is something else, the ban costs a relationship and leaves the problem exactly where it was. A noticed pattern can also slide into a hasty generalization about someone's character, especially if you only remember the nights that fit and forget the ones that don't, which is cherry picking.

Correlation vs causation and the post hoc fallacy

Assuming causation from correlation has an old Latin name, cum hoc ergo propter hoc: 'with this, therefore because of this.' Its sibling is post hoc ergo propter hoc, 'after this, therefore because of this,' which treats the order of events as proof. Both are false cause errors, and the Fallacy Files notes that many writers treat them as one.

In real arguments they blend together. 'Ever since you started the new job, you've been short with me' leans on timing, so it's the post hoc version. 'You're always grumpy on the days you see your dad' leans on two things occurring together, so it's the correlation version. The test for both is the same: look for other explanations before you assign blame.

How to tell if one thing really causes another

Researchers lean on controlled studies: split people into comparable groups, change one thing, and compare what happens. The Australian Bureau of Statistics calls this the most effective way to establish cause, and when an experiment would be unethical, researchers follow groups over time instead. You can't randomize your family, but you can borrow the logic.

• Check the timing. Does the supposed cause come first, every time?
• Look for a third factor. What else is different on those days? Work deadlines, sleep, money, who else is around.
• Flip the arrow. Could the 'effect' be causing the 'cause'?
• Look for exceptions. Does the effect happen without the cause, or the cause without the effect?
• Ask for a mechanism. Is there a believable way A would produce B?
• Run a small, agreed test. Change one thing for a few weeks and see what happens.

When a correlation is a fair clue

Correlation isn't worthless. It's often the first hint that something is going on, and science starts there too. JMP describes how a link between exercise and heart health led researchers to look for a biological mechanism and then test it in a controlled experiment. The pattern was the clue; the test was the proof.

So 'every time you go out with them, you come home upset' is worth saying. Just say it as an observation and a question, not a verdict: 'I've noticed you're often down after those nights. Is something going on?' If you can't agree whether the evidence supports the claim, you can lay out both sides for our free argument judge and see which reading holds up.

What to say