Simpson’s Paradox Explained: 3 Real Examples

Quick answer: Simpson’s paradox happens when a trend that appears in every subgroup of data reverses, or disappears, when the groups are combined. In a 1986 kidney stone study, open surgery (A) had a higher success rate than a keyhole procedure (B) for both small and large stones, yet a lower success rate overall (78.0% vs 82.6%), because A was used mostly on the harder, large-stone cases. The fix is to look for a hidden lurking variable and compare like with like.

A treatment works better for small cases and better for large cases, yet looks worse overall. A baseball player hits better than a rival in two different seasons, but worse across both seasons combined. These are not arithmetic errors. They are examples of Simpson’s paradox, one of the most important traps in statistics.

Example 1: Kidney Stones

A 1986 study by Charig and colleagues compared open surgery (A) with percutaneous nephrolithotomy, a keyhole procedure (B). Success meant stones eliminated or reduced to under 2 mm after three months:

GroupTreatment ATreatment B
Small stones93.1% (81 of 87)86.7% (234 of 270)
Large stones73.0% (192 of 263)68.8% (55 of 80)
All patients78.0% (273 of 350)82.6% (289 of 350)

A wins in both groups but loses overall. Why? Treatment A was given mostly to large stones, the harder cases (263 of its 350 patients), while B was given mostly to small, easy stones (270 of 350). The overall rate is a weighted average, and the weights are very different. (These are historical, non-randomized data, so they illustrate the paradox rather than prove which treatment is better.)

Example 2: Berkeley Admissions

In 1973, the University of California, Berkeley graduate school admitted 44.3% of male applicants (3,738 of 8,442) but only 34.6% of female applicants (1,494 of 4,321). It looked like bias against women. But when statisticians Peter Bickel and colleagues (1975) looked department by department, most departments admitted women at similar or even higher rates; there was not a clean reversal in every department. Women had applied more often to highly competitive departments with low admission rates for everyone. The overall gap came largely from where people applied. (This analysis did not prove the whole system was free of discrimination; it showed the headline numbers were misleading.)

Example 3: Batting Averages

Player19951996Combined
Derek Jeter.250 (12 of 48).314 (183 of 582).310 (195 of 630)
David Justice.253 (104 of 411).321 (45 of 140).270 (149 of 551)

Justice had the higher average in each year; Jeter had the higher average overall. Both players hit better in 1996, but 92.4% of Jeter’s at-bats came in 1996, compared with only 25.4% of Justice’s. The combined average is total hits divided by total at-bats, not the simple average of the two seasons.

Why It Happens

Each overall rate is a weighted average of the group rates:

\[ \text{overall rate} = w_1 \cdot \text{rate}_1 + w_2 \cdot \text{rate}_2, \qquad w_1 + w_2 = 1. \]

If two treatments use very different weights, the one stuck with more hard cases can lose overall even while winning every group. The hidden factor (stone size, department, season) is called a lurking or confounding variable.

Why It Still Matters

The paradox is not just a classroom curiosity. An April 2026 Stanford study in Nature Communications, using brain-imaging and behavior data from more than 4,000 people, reported that some brain-behavior links measured across people reversed when measured within the same person over time. Whenever data are pooled across groups, the same trap can appear.

Which Answer Is Right?

It depends on the causal story, not on the numbers alone. In the kidney stone study, stone size affects both which treatment doctors chose and the outcome, so the group-by-group comparison is the fair one: A is better. In other situations, the grouping variable is itself caused by the treatment, and the combined data are more meaningful. Judea Pearl and others developed causal diagrams to decide which comparison to trust.

Common Misconceptions

“The data must contain an error.” No. Every number can be correct; the reversal comes from unequal group sizes.

“Always trust the combined data, because it is bigger.” No. Combining groups can hide a confounder. Ask what drives both the grouping and the outcome.

Watch: Simpson’s Paradox in Video

If you learn better by watching, these videos are worth your time.

Simpson’s Paradox (minutephysics). A fast, clear explanation with simple visuals.
Simpson’s Paradox: When Better Doctors Look Worse (Mirko von Hein). A medical example of how combined data can mislead.

Check Your Understanding

Five quick questions to see how much of this has stuck. Pick an answer and press Submit; you will see a short explanation either way.

Answer each question and press Submit to see your result right away.

1. In the kidney stone study, which treatment did better for both small and large stones?
Found a bug? Let us know!
2. Why did treatment B look better overall?
Found a bug? Let us know!
3. What is a lurking (confounding) variable?
Found a bug? Let us know!
4. What was the Berkeley 1973 overall admission rate for men, roughly?
Found a bug? Let us know!
5. Who had the higher combined batting average in 1995 to 1996?
Found a bug? Let us know!

Frequently Asked Questions

What is Simpson’s paradox in simple terms?

It is when a pattern that holds in every group of data reverses or disappears when the groups are combined.

What causes Simpson’s paradox?

Unequal group sizes combined with a lurking variable. Overall rates are weighted averages, and different weights can flip the comparison.

What is a famous example of Simpson’s paradox?

UC Berkeley’s 1973 graduate admissions: men were admitted at a higher overall rate, but most departments admitted women at similar or higher rates.

How do you avoid being fooled by Simpson’s paradox?

Look for confounding variables, compare like with like, and use a causal argument to decide whether the grouped or combined data answer your question.

Keep Exploring

Sources and Further Reading

Related to This Article

What people say about "Simpson’s Paradox Explained: 3 Real Examples - Effortless Math"?

No one replied yet.

Leave a Reply