Cobra Effect · Incentives and systems
Goodhart’s law
A measure turned into a target stops telling you the truth.
6 cards, read aloud in 1:48, with a test and sources.
When a measure becomes a target, it stops being a good measure.
That is the short form. It was coined by the anthropologist Marilyn Strathern, from a 1975 observation by the economist Charles Goodhart. One sentence, and it explains half of every dashboard.
A measure works because nobody is gaming it.
Average call time is a fair proxy for how well a call centre runs. As long as it is just a number someone looks at, it tells you something true.
Then it becomes the target.
Agents are ranked on it. Bonuses hang off it. Now the fastest way to a short call is to hang up. The number gets better. The thing it measured gets worse.
It isn’t that people are dishonest.
A target changes what people optimise for. The measure used to be a shadow of the goal. Once it becomes the goal, it casts no shadow.
It has a cousin, and a cautionary tale.
Donald Campbell said the same about social policy in 1976. The more a number is used to make decisions, the more pressure there is to corrupt it. And there is an old story of a city that paid a bounty for dead cobras, and got cobra farms. Same law, with cash attached.
What to do instead.
Measure several things that are hard to fake together. Rotate the measures. Keep some numbers private to the people who decide. And when a number improves suspiciously fast, look for the hang up.
Sources
- Problems of Monetary Management: The UK Experience, Charles Goodhart, 1975. Where the law was born, as a remark about the Bank of England targeting the money supply. Search the title with the year; the Reserve Bank of Australia hosts a copy.
- “Improving ratings”: audit in the British University system, Marilyn Strathern, 1997. The paper that gave the law its one sentence form. European Review, volume 5, issue 3.
- The Tyranny of Metrics, Jerry Muller, 2018. The law applied to hospitals, police and schools, with the evidence for each.
Nearby ideas
- The Cobra effect. Why paying for a result can breed more of the problem.
- What you measure is what you get. Why the number you chase quietly replaces the thing you wanted.
- Survivorship bias. The failures you never see can reverse the lesson you draw.
- Second order effects. Every fix has consequences, and those have consequences too.
- The tragedy of the commons. Why shared things get used up, and how communities stop it.
- Moral hazard. Shield people from a risk and they take more of it.
- The Jevons paradox. Why making something efficient can make us use more of it.
- Braess’s paradox. Why adding a road can make everyone’s journey slower.