Goodhart's Law
mental-model proven
Categories: economics-and-financesystems-thinkingorganizational-behavior
Transfers
“When a measure becomes a target, it ceases to be a good measure.” Charles Goodhart originally stated this as an observation about monetary policy in 1975: any statistical regularity that the Bank of England tried to exploit for policy purposes would collapse once market participants anticipated the policy. Marilyn Strathern generalized it in 1997 to its now-famous form. The law describes a feedback loop in which measurement, incentive, and behavior co-corrupt.
-
The proxy gap — all measurement involves choosing a proxy for an underlying quality. Student learning is proxied by test scores. Policing quality is proxied by arrest rates. Software quality is proxied by code coverage. As long as the proxy is only observed, the correlation between proxy and quality holds. The moment the proxy becomes a target — with rewards, punishments, or career consequences attached — agents begin optimizing the proxy directly, and the correlation degrades. The proxy and the reality drift apart precisely because the proxy was trusted.
-
Directional corruption — the degradation is not random noise. Agents find the lowest-cost path to improving the metric, which is systematically different from the highest-value path to improving the underlying quality. Teaching to the test is cheaper than teaching understanding. Juking crime stats (downgrading felonies to misdemeanors) is easier than reducing crime. Writing trivial unit tests to hit coverage targets is faster than writing meaningful tests. The law predicts that the cheapest-to-game dimension of the metric will be exploited first.
-
Escalation dynamics — once agents begin gaming a metric, the metric’s administrators typically respond by adding more metrics, creating composite scores, or tightening definitions. This provokes more sophisticated gaming. Schools that gamed standardized tests faced value-added models, which were then gamed in different ways. The measurement system and the gaming behavior co-evolve in an arms race that consumes increasing resources on both sides.
-
Campbell’s Law convergence — Donald Campbell stated a closely related principle in 1979: “The more any quantitative social indicator is used for social decision-making, the more subject it will be to corruption pressures and the more apt it will be to distort and corrupt the social processes it is intended to monitor.” Campbell emphasizes corruption of the social process; Goodhart emphasizes breakdown of the statistical relationship. Together they describe a system where measurement intended to improve performance ends up degrading both the measure and the measured.
Limits
-
Not all metrics are equally gameable — the law’s force varies with how tightly coupled the metric is to the underlying reality. A bridge’s load-bearing test cannot be gamed because the test is the thing. A school’s graduation rate can be gamed because graduation and learning are loosely coupled. Invoking Goodhart’s Law against all measurement indiscriminately — as sometimes happens in anti-metric rhetoric — ignores this crucial variation and can justify abandoning measurement entirely, which is worse.
-
Gaming vs. displacement — the law bundles two distinct phenomena. Gaming is deliberate: agents consciously exploit the gap between metric and reality. Displacement is unconscious: well-meaning agents narrow their attention to what is measured and neglect unmeasured dimensions, not because they are cheating but because measurement focuses attention. A teacher who teaches to the test out of cynical careerism and a teacher who does so because the test genuinely seems important are exhibiting different behaviors with different remedies. The law does not distinguish them.
-
The law is not an argument against measurement — it is an argument for humility about what measurement can do when coupled with incentives. Organizations that invoke Goodhart to abandon metrics entirely often replace them with subjective judgment, which introduces different biases (favoritism, recency, halo effect) without the transparency that metrics at least provide. The law’s prescription is not “don’t measure” but “don’t attach high stakes to any single metric without expecting distortion.”
-
Context collapse — the law was formulated for macroeconomic policy, where the relationship between central bank actions and market behavior is mediated by rational anticipating agents. Extending it to education, policing, and software development assumes that the gaming dynamic is structurally identical across these domains. In some cases it is; in others the mechanism differs enough that the prescription (“rotate metrics,” “use multiple indicators”) may not apply.
Expressions
- “When a measure becomes a target, it ceases to be a good measure” — Strathern’s formulation, the canonical version
- “Teaching to the test” — the education-sector instance, where standardized test scores drive curriculum narrowing
- “Juking the stats” — from The Wire: deliberately manipulating crime statistics to meet targets
- “You get what you measure” — the management cliche, which is Goodhart’s Law stated as if it were a feature rather than a bug
- “Campbell’s Law” — the sociological cousin, emphasizing corruption of the social process rather than the statistical relationship
- “Hitting the number” — corporate shorthand for optimizing the metric at the expense of the business
- “Metric fixation” — Jerry Muller’s term for the institutional syndrome of excessive reliance on quantitative indicators
Origin Story
Charles Goodhart, an advisor to the Bank of England, first articulated the principle in a 1975 paper on monetary policy in the UK. His original formulation was narrowly technical: “Any observed statistical regularity will tend to collapse once pressure is placed upon it for control purposes.” He was describing how the Bank’s attempts to target specific monetary aggregates caused those aggregates to lose their predictive relationship with inflation, because market participants adjusted their behavior in response to the policy.
The law remained obscure outside economics until anthropologist Marilyn Strathern restated it in 1997 in a broader, pithier form: “When a measure becomes a target, it ceases to be a good measure.” This generalization connected Goodhart’s monetary observation to Campbell’s Law (1979), the Lucas critique in macroeconomics (1976), and a growing body of evidence about perverse incentives in education, healthcare, and policing. The concept became a standard reference in behavioral economics, public policy, and technology management.
References
- Goodhart, C. “Problems of Monetary Management: The U.K. Experience” (1975) — the original formulation
- Campbell, D. “Assessing the Impact of Planned Social Change” (1979) — the parallel principle in sociology
- Strathern, M. “‘Improving Ratings’: Audit in the British University System” (1997) — the generalized formulation
- Muller, J.Z. The Tyranny of Metrics (2018) — book-length treatment of Goodhart/Campbell dynamics across institutions
Related Entries
Structural Neighbors
Entries from different domains that share structural shape. Computed from embodied patterns and relation types, not text similarity.
- Software Peter Principle (organizational-behavior/metaphor)
- Words Are Weapons (war/metaphor)
- Rubber Duck Solution (comedy-craft/pattern)
- Secure Base (exploration/metaphor)
- Software Development Is a Bazaar (marketplace/metaphor)
- Letter vs. Spirit of the Law (language/metaphor)
- Inflation Is an Entity (embodied-experience/metaphor)
- Predator-Prey (ecology/mental-model)
Structural Tags
Patterns: pathforcematching
Relations: cause/misfittransform/corruptioncompete
Structure: cycle Level: generic
Contributors: agent:metaphorex-miner