Simpson’s paradox
Yule-Simpson effect, Paradoxe de Simpson
A trend that appears in each of several groups disappears or reverses when the data are combined.
The essentials
Mechanism
The groups are not the same size, and a hidden variable influences both group membership and the outcome. Intuition wrongly assumes that an overall average reflects the averages of each group.
In practice
Example
A treatment can be better for mild patients and for severe patients, yet look worse overall, if it is mostly given to severe patients.
Famous case
University of California, Berkeley, 1973: 44% of male applicants to graduate programmes are admitted, against 35% of female applicants. Bickel, Hammel and O’Connell (1975) show that, department by department, women are not disadvantaged: they applied more to the most selective departments.
Defending yourself
How to protect yourself
Before drawing a conclusion from an overall average, break it down by subgroup. Look for variables that could influence both the distribution and the outcome.
Don’t confuse
Not to be confused with: Correlation mistaken for causation (fallacy)
Correlation mistaken for causation overlooks that a common factor can create an apparent link; Simpson’s paradox shows that such a factor can even reverse the direction of a comparison.
Going further
Source
Edward Simpson (1951). The phenomenon had been described before him by Karl Pearson (1899) and George Udny Yule (1903).
Practise
In the app, each entry offers ways to practise recognising the concept. No sign-up: your progress stays on your device.