Stockertim Dictionary · Artificial intelligence · Technology
Simpson's paradox
Also called Simpson paradox, Yule-Simpson effect, reversal paradox
In short
A pattern that appears within each of several groups but reverses or disappears when the groups are combined, usually because the groups differ in size or mix.
- Intermediate
- Concept
- Checked
In detail
Simpson's paradox is a statistical trap in which combining groups changes the conclusion. One option can do better than another within every group, yet worse when the groups are added together. Nothing is wrong with the arithmetic; the reversal comes from how the groups are mixed.
It happens when the groups have very different base rates and the options are spread unevenly across them. If one option was mostly used in an easy group, where results are high anyway, its combined rate looks strong even if it loses within each group. The group variable is acting as a confounding variable.
The paradox matters for AI-assisted analysis because a tool asked for a single headline figure will usually give the combined rate, and an AI-written summary may draw a confident conclusion from it. Breaking results down by the main groups, such as new and returning customers, regions or channels, is the standard check.
Which view is right depends on the question. When the grouping reflects something that affects the outcome and was not chosen at random, the within-group comparison is usually the fairer one. Reports that show both views, and explain the mix, let readers see why they differ.
Worked example
Harbour Foods sends two promotions, X and Y, to 1,000 customers each and measures how many redeem them. X went mostly to new customers and Y mostly to regulars.
- Regular customers: X 180 / 200 = 90%; Y 680 / 800 = 85%
- New customers: X 240 / 800 = 30%; Y 50 / 200 = 25%
- X combined: (180 + 240) / (200 + 800) = 42%
- Y combined: (680 + 50) / (800 + 200) = 73%
X does better in both groups, yet Y looks far better overall because Y was sent mostly to regulars, who redeem at high rates anyway.
In practice
Analysts check for the paradox whenever groups differ in size or make-up, such as comparing hospitals, sales teams, marketing channels or loan approvals. Reports show breakdowns for the main groups alongside the total.
Watch out
- A combined rate can point the opposite way from every group within it.
- Splitting by every available field until a reversal appears is data snooping, not insight.
Learn it properly
Official sources
- AI Risk Management Framework · National Institute of Standards and Technology
- NIST AI RMF Playbook · National Institute of Standards and Technology
Facts last checked on 3 November 2026.
Education, not advice: the Stockertim Dictionary explains what terms mean. It never recommends buying or selling anything.