The big idea
🔗
The error with the highest cost
Correlation is where a causal investigation starts, never where it ends.
Scientific Thinking · Lesson 2
Correlation, Causation and Confounding
beginner⏱ 18 min · 7 cards
The hidden third factor
What a confounder does
Associated with the exposure, and independently affecting the outcome.
Try it
Name the Confounder
“Coffee drinking correlated with lung cancer.”
What explains the pattern without one causing the other?
Diagnose what produces each pattern without causation.
Key term
When the groups lie
Can a pattern reverse when data is pooled?
tap to reveal ↻
Simpson’s paradox
Yes. In the 1973 Berkeley admissions analysis, most departments favoured women slightly while the aggregate favoured men — because applications clustered in harder departments. Only disaggregating fixes it.
Without a trial
How smoking was settled
Nobody could randomise people into forty years of smoking. Causation was built by accumulation instead.
Strength, consistency across populations, dose-response, correct timing, a plausible mechanism, and the failure of every proposed confounder to account for it.
Quick check
When is a truncated y-axis legitimate?
You did it
Nice work
Mark this lesson complete to track your progress.