Stop 04 · Statistics for AnalystsTopics · 5/11

Correlation vs Causation

In short

Two numbers rising together is exciting. It's also the most common trap in all of data work. Before you say 'X causes Y', check what else might be pushing both.

Two variables moving together doesn't mean one causes the other — a distinction that's easy to state and surprisingly easy to forget under stakeholder pressure for a clean story.

The jaadu in the rain holds chai, looking puzzled at piles of umbrellas and cups.

Think of it like this

Umbrellas and chai in the monsoon

In July, umbrella sales and hot chai sales both shoot up. Do umbrellas make people thirsty for chai? No. The rain drives both. That hidden third factor is called a confounder, and it's behind many misleading correlations.

Watch it happen

The hidden third factor

Monsoon rain

Umbrella sales

Chai sales

1/4You see umbrella and chai sales move together. Correlation looks strong.

The big ideas

What you need to know

Correlation measures moving together

A coefficient near +1 or -1 means a strong linear link. It says nothing about why.

Look for confounders

Ask what could drive both variables: season, city size, income, time.

Reverse causation

Maybe Y causes X. Happy users use the app more, rather than usage making them happy.

Experiments show cause

Randomised tests like A/B tests are the most reliable way to show one thing causes another.

Save yourself the pain

Common mistakes

Oops

Writing 'users who use feature X spend 2x more, so feature X increases spend'.

Do this instead

Write 'are associated with'. Heavy spenders may simply explore more features. Test it to know.

Oops

Ignoring timing and season.

Do this instead

Check if both variables simply follow the calendar, like festive season or exam months.

Your mission

Find a spurious link

~20 min

Quick check

Cities with more hospitals have more deaths. What's the most likely explanation?

Go deeper

1Find (or recall) a classic spurious correlation example (ice cream sales and drownings).
2Practice proposing 2-3 alternative explanations for a correlation you observe.
3Learn what would actually be needed to establish causation (e.g. a controlled experiment).