Why correlation isn’t enough
A treatment can look better in the data and be worse for every kind of patient. Confounding, Simpson’s paradox on real surgical data, collider bias, and why randomised experiments are the gold standard.
40 min
After this lesson you can
- Explain the difference between a predictive question and an interventional one
- Reproduce Simpson’s paradox from the kidney-stone data and say exactly why the ranking flips
- Recognise confounding, and explain why it biases a comparison
- Explain collider (selection) bias and generate it from independent variables
- Say what randomisation guarantees, and what it does not
In this lesson
- Two different questions
- Why the ranking flips
- Confounding
- Should you always split the data?
- Selection: correlation from nothing
- Two more ways to fool yourself
- The fix that always works: randomise
This lesson is part of Descent Pro
Descent Pro unlocks every lesson on the route, from base camp to the summit. Every plan starts with a free trial.