Goodhart’s Law as a Feedback Problem: When Measures Become Targets

Goodhart’s law is commonly summarized as: when a measure becomes a target, it ceases to be a good measure. The deeper issue is feedback. A metric influences rewards and decisions; people adapt; their adaptation changes the relationship between the metric and the underlying goal.

The corruption loop

A performance gap increases target pressure. Pressure encourages activities that improve the reported number. Rewards reinforce those activities. If measurement is incomplete, effort shifts from the real outcome toward the proxy, weakening validity while the dashboard appears to improve.

Common mechanisms

• Gaming definitions or timing.
• Selecting easier cases.
• Shifting unmeasured costs elsewhere.
• Suppressing bad news.
• Teaching to the test.
• Replacing professional judgment with threshold compliance.

Better metric systems

Use a balanced set of outcome, process, quality, distribution, and harm indicators. Keep some measures diagnostic rather than tied directly to rewards. Audit boundary cases, invite qualitative evidence, rotate indicators when behavior adapts, and give people safe channels to report distortion.

Metrics should support inquiry: “Why is this changing?” A target treated as a complete definition of success becomes an invitation to optimize the representation instead of the system.

References

• Goodhart, C. A. E. (1975). “Problems of Monetary Management.” Reserve Bank of Australia.
• Strathern, M. (1997). “‘Improving ratings’: audit in the British University system.” European Review, 5(3), 305–321.
• Muller, J. Z. (2018). The Tyranny of Metrics. Princeton University Press.

Diagnostic questions for a target

1. What outcome was the metric originally intended to represent?
2. Which behaviors does the reward make locally rational?
3. What quality, equity, or long-term effects remain unmeasured?
4. Can people improve the number without improving the outcome?
5. How quickly will leaders learn when the proxy has failed?

Example: service response time

A target for closing requests quickly can improve responsiveness. It can also encourage premature closure, reclassification, or avoidance of complex cases. If reopened requests are counted as new work, the target hides the demand it creates.

A better measurement system pairs time with resolution quality, repeat contact, case complexity, user outcome, and distribution. Review samples qualitatively. Keep the metric open to revision instead of defending it because historical comparisons depend on it.

Metric humility

Every measure compresses reality. State what it excludes, how it can be manipulated, and when it should stop guiding decisions. This makes measurement a feedback instrument rather than a substitute for judgment.