Goodhart's Law

Goodhart's Law explains why reliable forecasts stop working once people start using them as targets to hit. When you turn a measure into a goal, people naturally change their behavior to optimize for that specific number, which breaks the very relationship that made the forecast useful in the first place. Think of it like standardized tests: when test scores became the main goal for schools, some started teaching specifically to the test rather than broader learning, making the scores less meaningful as measures of actual education quality. Notice this: whenever you see a metric being rewarded, check whether it still measures what it claims to measure.

Goodhart's Law captures a paradox at the heart of forecasting and measurement: statistical relationships that work well for observation tend to collapse when used for control or policy. The core mechanism is that when you create incentives tied to a forecast or metric, people adapt their behavior to optimize for that specific measure. This adaptation breaks the underlying statistical relationship that made the forecast accurate in the first place.

How It Works in Practice

In the 1970s, central banks discovered a stable relationship between money supply and inflation. When they tried to use this relationship for policy by targeting money supply growth, the correlation broke down because banks and businesses changed how they managed money in response to the new policy framework. The forecast that had worked for years became useless precisely because it was being used as a policy target.

Everyday Examples

In education, standardized tests were designed to measure learning quality. But when high test scores became the goal for teacher evaluation and school funding, many schools shifted resources to test preparation rather than broader education. The test scores became less reliable indicators of actual learning quality. In business, when sales teams are rewarded solely on units sold rather than profitability, they may offer excessive discounts or push products that customers don't actually need—the sales number becomes a bad measure of business health.

What Makes Forecasts Vulnerable

Forecasts based on historical patterns are especially vulnerable when they capture stable behavioral relationships that people can consciously modify. The more people understand how the metric works and the stronger the incentives to optimize it, the faster the relationship breaks down. This is why short-term performance targets often backfire and why purely quantitative measures can lead to worse outcomes when used naively.

Practical Implications

The solution isn't to abandon metrics entirely, but to use them more thoughtfully. Combine multiple measures rather than relying on single numbers. Rotate metrics periodically to prevent gaming. Focus on leading indicators that are harder to manipulate. Most importantly, remember that measurement should inform judgment rather than replace it—numbers work best as tools for human decision-making, not substitutes for it.

Goodhart's Law originated in monetary economics but has since been recognized as a fundamental principle affecting any system where quantitative measures are used for decision-making. The original formulation by British economist Charles Goodhart emerged from observing how monetary targeting strategies in the 1970s broke stable relationships between monetary aggregates and inflation [1]. Related work by social scientist Donald Campbell on educational testing demonstrated the same phenomenon: achievement tests are valuable indicators under normal teaching, but when test scores become the goal of teaching, they lose value as indicators and distort the educational process [2].

Key developments:

• Goodhart (1975): Original observation that any statistical regularity will tend to collapse when pressure is placed upon it for control purposes, based on UK monetary policy experience [1]
• Campbell (1976): Formalized Campbell's Law showing that quantitative social indicators become more subject to corruption pressures the more they are used for decision-making [2]
• Lucas (1976): The Lucas Critique demonstrated that econometric models cannot predict policy effects because their parameters change when policy changes alter expectations and behavior [3]
• Muller (2018): Documented widespread damage from metric fixation across education, medicine, business, government, and philanthropy, showing how "teaching to the test" and "gaming the stats" have become systemic problems [4]

Modern applications extend to artificial intelligence and machine learning, where "reward hacking" occurs when AI systems optimize poorly specified objectives without reaching intended outcomes—a direct analog of Goodhart's Law in algorithmic systems.

Goodhart's Law describes a real phenomenon but doesn't mean all metrics are useless or measurement is futile. Many organizations successfully use metrics when they combine quantitative measures with human judgment, use multiple indicators rather than single numbers, and design incentives carefully. The law applies most strongly when: (1) people understand how the metric works and have incentives to optimize it, (2) the metric captures a complex phenomenon that can't be fully reduced to a single number, and (3) there are significant rewards or penalties tied to performance on that metric. In cases where metrics are used purely for information rather than evaluation, or where they measure things that cannot be easily manipulated, Goodhart effects are much weaker. The critique also shouldn't be used to justify abandoning accountability entirely—some measurement is better than none, even if imperfect.