Track 9 · Correlation and Regression · Lesson 3 of 15
Correlation is not causation
Explain an observed association by chance, reverse causation, confounding or selection before considering a causal story, and see how adjusting for a confounder can change — even reverse — a comparison.
A school district notices that students who attended after-school tutoring scored lower on the final exam than students who did not. Should it cancel the programme? Before acting on any correlation, an analyst has to ask what else could produce it — here, the answer turns the conclusion upside down.
A correlation (or any association) tells you that two variables move together in the data. A causal claim says something stronger: if we changed , would change. Observational data — where nobody assigned the values of — rarely support that step on their own, because several other mechanisms produce exactly the same pattern.
Included with a Formula Quest membership
It has 8 graded exercises with feedback, hints and worked solutions. After it you will be able to:
- List chance, reverse causation, confounding and selection as alternatives to a causal explanation
- Show how adjusting for a confounder changes, and can reverse, a comparison
- Distinguish a confounder from a mediator before deciding what to adjust for
- Explain why an adjusted pay gap depends on the chosen controls and cannot by itself prove or rule out discrimination
- Write conclusions from observational data without causal over-claims