IDEA FOUND // IDEA 07
Correlation and Causation
“When ice-cream sales rise, drownings also increase. What might cause both?”
01 / PLAINLY
What it means, plainly
A correlation says that two variables change together; by itself it does not say why. The pattern may come from direct causation, reverse causation, a common cause, selection, or chance.
02 / CONTEXT
A little more
An association may arise through direct causation, reverse causation, a common cause, selection, or chance. Causal inference requires designs and assumptions capable of ruling out rival explanations; correlation alone offers no automatic test.
03 / WHY IT MATTERS
Why it matters
Mistaking association for cause can lead us to intervene on the wrong variable. A correlation can still be a useful clue when paired with design, assumptions, and additional evidence.
02 / SEQUENCE
What happens
- 01
You observe two variables changing together.
- 02
You list possible causal directions, common causes, and selection effects.
- 03
You seek a design or additional evidence that can distinguish them.
04 / EXAMPLE
A familiar example
Hot weather may raise both ice-cream consumption and time spent swimming. Temperature is a plausible common cause.
05 / LIMIT
What it does not mean
“Correlation does not imply causation” does not mean correlations are useless or only randomized experiments provide causal evidence.
06 / NOTICE
Notice it in your day
Draw ice cream and drownings as two nodes. Add temperature as a possible common cause, then write what further evidence could distinguish that story from direct causation.
FINAL NOTE
The idea worth keeping
When you see an association, ask: what earlier variable might move both, and how were the observations selected?
QUESTIONS / 02
Questions people still have
What explanations can produce a correlation?
It may arise from direct or reverse causation, a common cause, selection, or chance; correlation alone cannot choose among them.
Does ‘correlation is not causation’ make correlations useless?
No. A correlation can supply evidence and a question, but causal inference also needs design, assumptions, and rival comparisons.
CHECK / 01