Simpson's Paradox

Group 1 Size (Easy cases): 80 Group 2 Size (Hard cases): 80 Treatment A Boost (Easy): 0.85 Treatment A Boost (Hard): 0.45 Treatment B Boost (Easy): 0.80 Treatment B Boost (Hard): 0.40 A patients in Easy Group: 70%

---

Animate Merge UC Berkeley Data Reset to Default

Simpson's Paradox

A trend that appears in several groups of data reverses when the groups are combined.

Treatment A beats B in Group 1 AND Group 2. But combined, B appears better! The trick: Treatment A is disproportionately given to the easy group, inflating its combined rate relative to B which tackles the hard group.

Real example: UC Berkeley 1973 admissions seemed to favor men overall, but department-by-department, women were admitted at equal or higher rates.