Simpson's paradox
When a pattern that holds in every group reverses once the groups are combined, because the groups differ in size and baseline rate and people are spread unevenly across them.
In plain English
It isn't an error in the data: both the group figures and the total are correct, and they answer different questions. The group figures compare like with like; the total also includes the effect of who ends up in which group. When the two disagree, the variable that defines the groups is a confounder, and the interesting question is usually about it.
Uncle Gary explains
In our company example, women are promoted more often than men in both departments, yet overall men are promoted at 43% and women at 22.5%. At Berkeley in 1973, about 44% of male and 35% of female graduate applicants were admitted, while most departments admitted women at the same rate or higher.
Why it matters for your money
A fund can beat a rival in every year and still have a lower total return if more money was invested in the bad years. Check comparisons within each period, not just the total.
Where we tested it
Related terms
All glossary terms. Educational content, not financial advice.