Simpson's Paradox is a trend appears in different groups but disappears or reverses when groups are combined. It sits in the Cognition dimension (COG) of the Human Behavior Taxonomy™ as element HBT-COG-0618, within the Probability family. The core principle: a trend appears in different groups but disappears or reverses when groups are combined. In incentive terms, it matters because it changes the payoff people perceive before they choose — which means it can be designed for, or exploited.
Scientific Definition
A trend appears in different groups but disappears or reverses when groups are combined.
Plain-English Definition
A trend appears in different groups but disappears or reverses when groups are combined.
Feynman Explanation
The aggregate can lie about the parts. The parts can lie about the aggregate.
Core Principle
A trend appears in different groups but disappears or reverses when groups are combined.
Mechanisms
Pending editorial review.
A trend appears in different groups but disappears or reverses when groups are combined.
Pending editorial review.
Pending editorial review.
Pending editorial review.
Pending editorial review.
Aggregated metrics can hide important subgroup effects.
Inputs (Triggers)
Pending editorial review.
Outputs (Behaviors)
Pending editorial review.
Behavioral Signature
The aggregate can lie about the parts. The parts can lie about the aggregate.
Examples
- A treatment helps men and women separately but appears harmful overall.
- Aggregated metrics can hide important subgroup effects.
Pending editorial review.
Original analysis from The Incentives Lab — how this element behaves inside real payoff structures.
Why this element matters to incentive design
The mistake with this element is treating it as irrationality. It is almost always a rational response to a payoff nobody wrote down. The mechanism underneath it is straightforward: a trend appears in different groups but disappears or reverses when groups are combined. You can recognize it in the field by its signature: the aggregate can lie about the parts. The parts can lie about the aggregate. Every element in the Cognition dimension changes the perceived payoff of an action before the action happens, which is exactly where incentive design has leverage.
How it gets exploited
Left undesigned, aggregated metrics can hide important subgroup effects. It is amplified whenever aggregated metrics can hide important subgroup effects. Inside organizations that shows up as aggregated metrics can hide important subgroup effects. The pattern is the same one Goodhart's Law describes: the measurable proxy attracts the effort, and the purpose behind it quietly loses funding.
How the Lab designs around it
The redesign move is to analyze data at multiple levels of aggregation. The leverage is not in explaining the behavior to people. It is in changing what the behavior earns.
Famous Experiments
Pending editorial review.
Design Principles
- Analyze data at multiple levels of aggregation.
Measurement Approaches
Pending editorial review.
Evidence
Pending editorial review.
Pending editorial review.
The Perverse Incentive Lens™
How this behavior is exploited — and how to redesign around it.
- Analyze data at multiple levels of aggregation.
Pending editorial review.
Pending editorial review.
Interactive Mini Network
Click any neighbor to re-center the graph and follow the threads of connection.
Knowledge Graph Neighbors
Auto-linked to the rest of the Human Behavior Taxonomy by family, domain, dimension, and shared keywords.
Ignoring general statistics in favor of specific, vivid information.
Posterior = (likelihood × prior) / evidence.
Update beliefs in proportion to the strength of new evidence.
Start with a prior; update with new evidence.
Novices experiencing early success, often due to variance and small samples.
High-impact, hard-to-predict, retrospectively explainable events.
Striking pattern that is statistically expected in large samples.
Reasoning in distributions — ranges and probabilities — rather than points.
Average outcomes across the population differ from outcomes across time for one person.
A rough calculation using order-of-magnitude reasoning.
In quantum mechanics, certain pairs of properties cannot both be precisely known.
We ignore sample size when judging probability.
Where Simpson's Paradox is cited in the corpus
Essays, field guides, and diagnostics from The Incentives Lab that apply this element.
- EssayGoodhart's Law in the Real World
How measurable proxies capture judgment.
- EssayThe Perverse Incentives Hiding in Your KPIs
Cognitive shortcuts turned into scorecards.
- EssayAI Agents Inherit Your Incentives
How this element propagates into automated systems.
- CourseIncentives 101
The free ten-part primer on reading a payoff structure.
- ReferenceThe incentive glossary
Definitions for every mental model, bias, and fallacy in the corpus.
- ReferenceThe Periodic Table of Human Behavior
The full 1,267-element map this page belongs to.
Questions about Simpson's Paradox
- What is Simpson's Paradox?
- Simpson's Paradox is a trend appears in different groups but disappears or reverses when groups are combined. It sits in the Cognition dimension (COG) of the Human Behavior Taxonomy™ as element HBT-COG-0618, within the Probability family. The core principle: a trend appears in different groups but disappears or reverses when groups are combined. In incentive terms, it matters because it changes the payoff people perceive before they choose — which means it can be designed for, or exploited.
- What is an example of Simpson's Paradox?
- Aggregated metrics can hide important subgroup effects. The Incentives Lab catalogs everyday, organizational, and historical instances of this element on its Human Behavior Taxonomy™ page (HBT-COG-0618).
- How is Simpson's Paradox exploited?
- Aggregated metrics can hide important subgroup effects.
- How do you design around Simpson's Paradox?
- Analyze data at multiple levels of aggregation.
- Which behavioral dimension does Simpson's Paradox belong to?
- Simpson's Paradox is classified in the Cognition dimension (COG) of the Human Behavior Taxonomy™, family "Probability", class "Mental Model". Its permanent identifier is HBT-COG-0618 and its evidence grade is B.