Number Mistakes — Statistical & Probability Errors

The Big Total Tells A Different Story Than The Smaller Groups

Simpson's Paradox

One-line definition: The overall numbers point one way, but the pattern changes or reverses when you break the data into meaningful groups.

In Plain English

Simpson's Paradox happens when a big total tells one story, but the story changes when you split the data into smaller groups that matter. The overall numbers are not fake, but they can hide how the groups were mixed together. Maybe one option looks worse overall even though it performs better inside each subgroup. This bug matters because people often trust the headline number and stop there. The fix is to ask whether important groups are being blended in a way that hides the real pattern.

Featured Example

Class average confusion

One teaching method looks worse across the whole school, but inside both beginner and advanced classes it actually performs better.

Classrooms

What This Sounds Like in Classrooms

  • A schoolwide average hides the fact that different class levels faced very different conditions.
  • One program looks weaker overall until the results are split by starting skill level.
  • A headline number hides the subgroup story.
Business

What This Sounds Like in Business

  • One team seems less productive overall, but inside each project type it actually performs better.
  • A hiring funnel looks unfair or effective until the data is split by role, region, or applicant mix.
  • The total makes a policy look bad while the subgroup view changes the story.
Real Life

What This Sounds Like in Real Life

  • A health result looks clear overall, but age groups tell a more careful story.
  • A neighborhood total hides the fact that very different blocks have very different patterns.
  • One app feature seems popular overall until usage is split by user type.
Fiction

Examples from Literature or Fiction

Detective and courtroom stories

A broad summary points one way, but the case changes when the details are sorted into the right groups.

The overall picture hides the structure underneath.

School or war novels with class or rank divisions

A general claim about performance or fairness shifts once results are broken apart by role.

Mixed groups hide the real comparison.

Social dramas about status and opportunity

Big averages hide the fact that different groups faced different chances from the start.

The total can tell the wrong lesson when important groups are blended together.

Why People Fall for It

Summary numbers feel clean and powerful. People like the big total because it is easier to read than a more careful grouped analysis.

How to Spot It

  • The headline number is doing all the work.
  • Important subgroup differences are missing.
  • The pattern changes when the data is split more carefully.
  • The summary feels simple, but the situation is clearly mixed.

What to say instead

  • What happens if we break this into meaningful groups?
  • Are different populations getting blended together here?
  • Does the overall number hide the real comparison?
  • Big totals are useful, but they are not always the full story.

Common Confusion

Compare Nearby Ideas

Mini Practice

Question: A schoolwide average says one method did worse, but inside both beginner and advanced classes it did better. What is the bug?

Answer: Simpson's Paradox.

The overall number hid the subgroup pattern and gave a misleading summary.

Remember This

The big total can hide the real story inside the groups.

Related Brain Bugs

Base Rate Neglect

Ignoring The Big Background Numbers

Number Mistakes

A test flags a rare condition, and someone assumes the condition is now very likely without looking at how rare it is in the first place.

Learn this bug

Correlation vs. Causation

Together Does Not Mean Caused

Number Mistakes

Ice cream sales rise when beach rescues rise. That does not mean ice cream causes the rescues. Hot weather drives both.

Learn this bug

Texas Sharpshooter Fallacy

Pick The Pattern After The Fact

Number Mistakes

A team looks at many product metrics, finds the two that moved together, and presents that pair as if the trend had been clear from the s...

Learn this bug