Simpson's Paradox: A Trend Can Completely Reverse When Subgroups Are Combined
A treatment can outperform another inside every separate patient subgroup—yet look worse overall once all rows are pooled together.
🔄 Data Paradox
Stanford Encyclopedia of Philosophy
If Treatment A beats B in both mild cases AND severe cases, must A also win in the combined total?
Imbalanced subgroup sizes mixed with a lurking confounder can flip the aggregate percentage upside down.
📊 Statistics
Data Paradox
Why / How?
Scientific Mechanism
In the famous 1973 UC Berkeley admissions study, overall acceptance looked skewed toward men—but department-by-department data showed women were accepted at equal or higher rates. The hidden confounding factor was that women disproportionately applied to the most competitive departments with tiny acceptance rates.
Why It Matters
Imbalanced subgroup sizes mixed with a lurking confounder can flip the aggregate percentage upside down.