Goodhart's Law

You have almost certainly seen this sentence somewhere:

"When a measure becomes a target, it ceases to be a good measure."
Once a measure becomes a target, it stops being a good measure.Usually attributed to Goodhart. He never said it.

Behind that sentence is a mechanism: why a perfectly good measure stops working the moment it gets used to grade people. That mechanism is the bedrock of this whole family. It also drags along a small scandal: the most widely quoted version of that line was not, in fact, written by Goodhart himself; tracing the real source takes three people passing it hand to hand before it can even be pieced together. Which produces this irony: a law about how measures degrade had its own famous line degrade right along with it, in circulation.

Start with where the law was born. It did not come out of an armchair. It is the autopsy report on a monetary policy accident.

In the 1970s the British government wanted to control inflation. The method it picked was called monetary targeting: watch a number called £M3 (one way of counting how much "broad money" is circulating in the economy), hold its growth rate inside a target band, and hope that this keeps prices in line. The logic sounded clean, because in the historical data the relationship between that number and the economy had been stable.

Then things got strange. In September 1971 the Bank of England loosened credit controls, and the once stable relationship between that number and the economy immediately came apart. In 1972 to 73 broad money growth briefly passed 25%, completely off the leash. At the end of 1973 the central bank reversed into hard controls, introducing a scheme nicknamed "the corset" that clamped down directly on the size of bank liabilities. Banks immediately invented substitute liabilities that fell outside the controlled definition, going around it that way. The jargon for this is disintermediation, and over time, getting around the regulator went from an emergency measure to standard craft. The ugliest scene came under the Thatcher government. The 1980 Medium Term Financial Strategy set the £M3 annual growth target at 7% to 11%. The actual figure hit 18% to 19%, close to double. If that ruler still tracked the real economy, money running this far out of control should have brought overheating and rising inflation right along with it. Instead the economy was in deep recession at that very same time, and inflation was falling. What does that combination tell you? The number the government was watching had fully decoupled from the real economy it wanted to manage. The number ran out of control and the economy cooled at the same time. The ruler and the world went their separate ways. Treasury ministers later cited "Goodhart's law" themselves to explain the overshoot, and broad money targeting was effectively abandoned in 1985 to 86.

Goodhart was then a monetary policy adviser at the Bank of England. In July 1975 he went to Sydney for a Reserve Bank of Australia conference and submitted two papers (one subtitled "a view from Threadneedle Street," Threadneedle Street being where the Bank of England sits). He compressed the accident above into a half joking sentence, called it a "law" in the mock solemn register of Murphy's law, and tucked it into a footnote: any observed statistical regularity will tend to collapse once pressure is placed upon it for control purposes. He later said of himself that it does feel slightly odd to have one's public reputation largely based on a minor footnote.

The mechanism takes one sentence: correlation is not identity.

When nobody is watching a measure, the reading is a by-product of people doing their jobs normally, and it correlates naturally with the real goal. The moment you announce "pay is set by this ruler," everyone graded by it gets a new motive: go hunting specifically for the space where the reading is high and the real goal has not improved. Effort drifts from "do the thing well" toward "make the number bigger," and the cheapest path to a bigger number is almost never actually doing the thing well. Optimization pressure pulls the correlation apart until it snaps. The collapse Goodhart named is exactly that snap.

Note one thing that is easy to miss: none of this requires anyone to act in bad faith. The student grinding practice problems may sincerely believe that solution templates are knowledge. The engineer watching click-through may sincerely feel that raising clicks is creating value for users. The frightening part of a measure is not that it tempts you to cheat. It is that it quietly replaces the goal inside your head. That quiet replacement has its own name, surrogation, and B07 is devoted to it.

Back to the scandal. The popular sentence was born twenty-two years after Goodhart's footnote, with three hands in the relay.

The first hand is Goodhart himself. In 1975 those two papers held only the footnote joke, no clean single sentence. The most reliable polished wording only appears in his 1984 collection; by then he was already referring to it in the third person as "Goodhart's law," which shows the naming had been accepted by him by that point. In 1997 he also left behind a restatement set in a government context. The second hand is the management accounting scholar Hoskin. In 1996 he lifted the law out of its monetary setting and rewrote it as "every measure which becomes a target becomes a bad measure," under Goodhart's name. The third hand is the anthropologist Strathern. In 1997 she read Hoskin's version and forged it into the best cadenced maxim we have today, in the context of British university audit culture. The accurate way to remember it: the name belongs to Goodhart, the generalization to Hoskin, the famous line to Strathern.

The detective work also turns up an older layer of bedrock. Long before it carried anyone's name, "measures deform once you grade people by them" was already field common sense in organizational sociology. In 1956 Ridgway published an 8 page article in the founding volume of Administrative Science Quarterly, reviewing the fieldwork of the day: employment agency clerks graded by the number of placements simply picked the applicants who were easy to place (Blau's records); Soviet factories graded by output sacrificed everything other than output (Berliner's research); plus Argyris on budgets. He laid out the distortion mechanics of single, multiple and composite measures in full, 19 years before Goodhart, and left the family's earliest aphorism: the cure is sometimes worse than the disease. His criticism of composite measures was especially ahead of its time. Weighting several dimensions into one total score does not solve the problem, because the people being graded will only optimize along whichever single dimension carries the highest weight, and externally set weights displace each person's own private judgment of what matters. It is the same move, compressing many dimensions into one number: training today's models by compressing many objectives into a single scalar reward replays that same trap.

Finally, priority. Economics has a twin proposition born almost simultaneously, the Lucas critique (roughly: the economic regularities you observe already contain people's expectations about the old policy, so change the policy, expectations change, and the regularity collapses; B03 is devoted to it). Chrystal and Mizen's authoritative 2003 review starts with full honours: very few people get an economic law named after themselves, and Goodhart belongs to a very small club alongside Gresham, Walras and Say. Then comes the sharpest ruling in the whole piece: it can be argued that Goodhart's law and the Lucas critique are essentially the same thing, and if so, Lucas almost certainly said it first. The two later divided the labour: the Lucas critique rewrote the methodology of macroeconometrics, Goodhart's law rewrote the design of monetary policy, with money targets leaving the stage and inflation targets taking it. So the accurate version of "several laws discovered independently at the same time" is a layered process spanning organizational sociology (1956), social science (1969 to 79, Campbell's line in B02) and economics (1975 to 76): Ridgway came first, and Goodhart got the largest posthumous fame for the naming.

By now the law has grown from a maxim into a science with a scale, able to answer when it strikes and how badly.

El-Mhamdi and Hoang proved in 2024 that what matters is the shape of the distribution of the difference between the true objective and the optimized measure, and split the outcomes into two regimes on that basis. When the difference is broadly bounded you get weak Goodhart: over-optimizing the measure is merely wasted effort and stops helping the true objective. When the difference is heavy tailed (the probability of extreme deviations is not negligible) you get strong Goodhart: pushing the measure further does real damage to the true objective. The dividing line is set by the shape of the tail.

The same scale marks out the law's boundary. It is not a claim that measurement is useless. It is a failure law for one specific combination, proxy measure plus high pressure optimization. Remove the pressure, shorten the proxy chain, keep one independent audit channel, and measurement remains the strongest management technology known. That is exactly what Campbell's own "experimenting society" position is saying. When measures are benign, B14 lays out the counter-evidence and the conditions.

One open question: next time someone uses a famous quote, a measure or a clean conclusion to persuade you, do two things. Check where it came from, and ask how long it has been optimized. Quotes work like measures. The more useful they are and the longer they get used, the more likely the core has already been quietly swapped out in use. Plus one small unresolved matter: Goodhart read his papers at the conference in 1975, Lucas published formally in 1976, so "who said it first" depends on which act you count as the debut, and to this day nobody has verified the 1975 conference papers against the original copies. A law whose whole business is measurement still has its own birth certificate sitting in the Reserve Bank of Australia's archive. That is probably the joke it played on itself.

The one-line takeaway: a measure is honest only while it is not the target; once pay follows it, what it measures is no longer the world but people's reaction to it.