Goodhart's law — when a measure becomes a target, it ceases to be a good measure — is usually quoted as a wry epigram. In government operations it is a working description. The documented cases span every service the public touches: veterans' health facilities that kept unofficial wait lists while reporting official appointment targets met, as Veterans Affairs inspector-general investigations found in 2014; police statistics downgrading offenses to improve clearance rates, the subject of successive audits and journalistic investigations in major departments; border and immigration agencies re-categorizing queues; and, in the era of 311 dashboards, service requests closed without resolution, which municipal re-inspection programs catch at measurable rates. The pattern is not fraud by villains. It is what happens when a measure is used as a target without a verification design.
Why gaming is rational
Three conditions produce it reliably. The measure proxies the outcome: waiting-time statistics proxy care received; clearance rates proxy safety — and proxies can be improved without the outcome moving. The target attaches consequences: budgets, performance ratings, careers. And verification is absent or self-reported. When all three hold, the cheapest path to the target is the metric, not the mission — an observation public-administration research has replicated from street-level bureaucracy studies onward. The Government Accountability Office's program-evaluation work keeps re-encountering the same triad: GPRA-era performance plans full of measures agencies could hit by definition, documented in GAO's implementation reviews.
The design fixes that work
Gaming is a design failure with design answers. Independent verification: sample the underlying reality — re-inspect closed 311 requests, audit sampled case files, verify wait lists against appointment-system logs, the exact mechanism VA reforms adopted after 2014, with third-party scheduling audits. Paired measures: pair every throughput metric with a quality metric that the throughput producer cannot control — speed with accuracy, closures with verified resolution — so improving one at the other's expense shows. Outcome measurement over output where feasible: count recidivism, not program completions; count verified repairs, not requests closed. And rotation of measurement attention: metrics audited intermittently and unpredictably raise gaming's cost, an approach from the audit-sampling tradition.
The honest limits
Two limits deserve stating. First, the alternative to imperfect metrics is not no metrics — it is anecdote and politics allocating resources instead; the GPRA record shows even unbinding measurement builds useful infrastructure for oversight. Second, some gaming is actually useful adaptation: when street-level staff meet a target by reorganizing work in ways headquarters had not imagined, that is improvement wearing gaming's clothes — which is why the verification layer matters more than the target layer. Distinguish the two by asking whether the outcome moved when the metric did. That test, applied routinely by auditors and inspectors general, is the whole discipline in one question.
FAQ
What is Goodhart's law?
The observation that when a measure becomes a target, it ceases to be a good measure — because people optimize the measure rather than the outcome it proxies.
What is the largest documented government gaming case?
The 2014 Veterans Affairs wait-list scandal: facilities maintained unofficial lists while official appointment-time targets were reported as met, per VA inspector-general investigations.
How do you stop metric gaming?
Design, not exhortation: independent verification of a sample of cases, paired throughput-and-quality measures, outcomes over outputs, and unpredictable audit rotation.
For more context, read The Evidence Act, Seven Years On: Slow Progress, Real Infrastructure.
For more context, read trust in government.
For more context, read Why State Unemployment Systems Can't Finish Modernizing.