Goodhart's law: rewarded metrics get gamed
This is a named principle rather than a measured effect. The logic is sound, and the documented cases keep bearing it out. Trust the direction, and treat the boundaries as open.
When you reward a number, people learn to produce the number, not the thing you actually wanted. The clearest documented case comes from French colonial Hanoi in 1902. The city had a rat problem, so the administration paid a bounty for each rat killed, asking people to hand in a rat's tail as proof. It seemed clever. But tails started arriving from rats that were still alive and released to breed again, and enterprising residents began farming rats specifically to harvest tails for the bounty. The programme aimed at fewer rats and, by paying for tails, produced more rats. A widely told cousin of this story, about a cobra bounty in colonial India, captures the same idea and gave the effect its popular name, though that version has no solid primary record and is best treated as a fable. The Hanoi rats are the case that actually happened.
The moment you attach a reward to a measure, clever people optimise the measure in ways you did not intend, and the goal behind it stops getting attention. You wanted fewer rats and paid for tails, so you got tail farms. In your world it is the sales target hit with bad deals, the hiring quota filled with whoever is available, the response-time metric gamed while real problems fester. Before you reward a number, ask how a smart, self-interested person could hit that number without doing the thing you care about. They will find the route you did not imagine, because the reward pays them to. The wider version of this failure, where what cannot be measured stops counting at all, is the McNamara fallacy.
Goodhart says that once a measure becomes a target it stops being a good measure, because people optimise the number instead of the goal. Moneyball says a broke team that trusted the numbers over a century of scouts' eyes won twenty straight. Both are right, and the difference is what the number is for. The Oakland statistics were a diagnostic nobody was being paid to hit, so they still described the thing. The rat tails were a bounty, so they described only the bounty. The test before trusting any number is whether anyone's reward depends on it. If nobody's does, it can beat the eye. If somebody's does, the eye is looking at the wrong thing already.
Source: The principle: Charles Goodhart, Problems of Monetary Management: The U.K. Experience, presented at a Reserve Bank of Australia conference in Sydney in 1975 and reprinted in his Monetary Theory and Practice, 1984. The phrasing quoted everywhere, when a measure becomes a target it ceases to be a good measure, is Marilyn Strathern's, from 'Improving ratings': audit in the British University system, European Review, 1997. The Hanoi rat bounty of 1902, one cent per tail, is the documented case: the historian Michael G. Vann found it in a French colonial archive file on the destruction of rats, first wrote it up in a 2003 journal article, Of Rats, Rice, and Race, and tells it in full in The Great Hanoi Rat Hunt: Empire, Disease, and Modernity in French Colonial Vietnam, with Liz Clarke, Oxford University Press, 2018. The popular cobra version from colonial India is a widely circulated tale with no solid primary source.
Jerry Muller, 2018
Case after case of the measure becoming the target, and the target being lost.
The Saturday letters
Every Saturday, one idea worth your time, with an honest account of how far the evidence really goes and where it stops. Free, a few minutes to read, and it leaves you with questions rather than instructions.
Draw your own card. It does not take long, and it rewards taking your time.