Gaming the Measure
When improving the reported number is easier than improving the work.
A target can focus attention, but it can also reward a better-looking number. Check what changed for people, what changed in the work, and whether delay or harm moved elsewhere.
Why targets can distort behaviour
Deming’s Point 11 warns against quotas and management by numerical goals, calling for leadership instead. A number does not explain the system that produces it or provide the means to improve it.
The concern often associated with Goodhart’s law is that a measure can become less useful when people are rewarded or penalised for hitting it. Gaming is a risk, not an inevitable response to every target. Pressure, definitions and incentives matter.
Start by asking whether the measure still represents its intended purpose. A shorter reported wait is useful only if people receive appropriate help sooner—not simply if the clock stops earlier.
How the number can improve without the service
- Retiming: move work across a reporting boundary or change when a clock starts or stops.
- Reclassification: move cases into a category excluded from the measure.
- Selection: prioritise easier cases while people with greater needs wait longer.
- Incomplete recording: leave difficult contacts, delays or adverse outcomes out of the data.
These are possible mechanisms to investigate, not accusations about a particular service. Involve staff and people using the service; check definitions and records before concluding that anyone has deliberately manipulated results.
A corridor-care example
Suppose a hospital targets the number of patients in corridors at one daily census. That count could fall even if corridor use grows at other times or patients wait in a differently labelled space.
Keep the snapshot if it is useful, but pair it with the duration and scale of the problem: for example, patient-hours in unsuitable spaces, long waits and safety concerns. Keep definitions consistent and record exclusions. The corridor-care worked example shows a proposed evaluation—not evidence of a completed improvement.
Use a small, connected set of measures
Outcome, process and balancing measures answer different questions:
- Outcome: did the result that matters improve—for example, less time waiting for appropriate care?
- Process: did the intended change happen—for example, were agreed discharge tasks completed when needed?
- Balancing: did the change create problems elsewhere—for example, more repeat contacts, workload or unequal access?
This makes misleading improvement easier to challenge. It does not make the measures impossible to game, or prove that the intervention caused an outcome.
Agree the checks before making the change
- State the intended benefit and who should experience it.
- Define each measure, its denominator, exclusions and data source.
- Record a baseline and predict what should change, by when.
- Check implementation, data quality, differences between groups and possible harm.
- Review results with staff and service users, including evidence that challenges the preferred story.
Read the pattern, not just the score
Use suitable time-ordered data rather than two selected before-and-after figures. Choose a chart that fits the question and data. A fitted change in a retrospective StepChange chart can guide investigation; it does not establish the cause or show that every measure improved.
Avoid tampering with the process after routine fluctuations. Equally, do not ignore poor performance or a safety concern while waiting for a statistical signal.
How to take this further
Start with Right Measures in the improvement guide. Then use the other two questions to connect the explanation, the change and its evaluation.
Return to Why Improvement Programmes Fail for the complete framework.
Read your StepChange chart and decide what to check next → Use a chart alongside the intervention history and other evidence, not as a substitute for an explanation.
Source and scope
The Deming Institute: the 14 Points, particularly Points 10 and 11. The gaming mechanisms and corridor scenario above are explanatory examples, not findings about a named organisation.