Quality, Safety, and Performance Management

Why this matters in public value

Every public and social-sector organisation of any size now runs on numbers: waiting times, caseloads closed, response minutes, pass rates, satisfaction scores, inspection ratings. Targets built from these numbers are not an imposition from outside public management; they are how legitimacy and support are demonstrated to funders, legislatures, boards, and citizens who cannot otherwise see inside daily operations (see Topic 1.2 — The Strategic Triangle: Legitimacy, Value, and Capacity). A leader who cannot point to a number that moved in the right direction struggles to defend a budget, a mandate, or a reform.

The difficulty is that a number built to represent quality is not the same thing as quality itself, and the two quietly separate the moment the number starts to matter more than the reality behind it. A police force can drive down recorded crime by changing how officers classify incidents rather than by reducing harm. A hospital can hit a waiting-time target by reclassifying which patients the clock applies to rather than by treating people faster. Neither change shows up in the published figure, which is exactly the problem this topic exists to address.

This topic is the home treatment of that problem in the public sector: performance indicators and the perverse incentives they create, quality assurance regimes built to keep standards consistent regardless of what any single target says, and independent inspection and ratings bodies that check the reality behind the numbers from outside the chain of command that produced them. Topic 2.1 covers where indicators should come from in the first place — outcomes frameworks and logic models that connect a measure back to a genuine causal claim about value (see Topic 2.1 — The Public Value Scorecard and Outcomes Frameworks). This topic covers what happens once those indicators are turned into live targets that staff are managed against, and how an organisation keeps its own measurement system honest.

The stakes rise with how publicly a number is used. A private internal metric that drifts from reality wastes a manager's time. A public performance target, a headline inspection rating, or a safety statistic that drifts from reality can cost lives, misallocate resources for years, and — when the gap between the reported number and the lived experience eventually becomes visible — destroy exactly the legitimacy the number was published to build (see Topic 3.4 — Accountability, Transparency, and Legitimacy).

Core concepts

A performance indicator is a measurable variable chosen to stand in for an aspect of an organisation's performance that cannot itself be observed directly and continuously (see performance indicator). Indicators are proxies by construction: nobody can measure 'quality of care' or 'community safety' directly, so an organisation measures something narrower and more countable — a waiting time, a recorded crime rate — and treats movement in it as evidence of movement in the broader thing it stands for. That gap between proxy and reality never closes; this topic is about managing it rather than forgetting it exists.

Goodhart's law states, in its most common public-sector formulation, that when a measure becomes a target, it ceases to be a good measure. The law is named for the British economist Charles Goodhart, who formulated the underlying insight in 1975 in the context of monetary policy: once a central bank starts targeting a particular measure of the money supply, the statistical relationships that made that measure useful begin to break down, because people change their behaviour in response to the target itself. The anthropologist Marilyn Strathern restated the idea for a public-sector audience in a widely cited 1997 study of British university audit culture, giving it the compact phrasing now most commonly quoted: 'when a measure becomes a target, it ceases to be a good measure.' Topic 3.1 introduced this dynamic briefly in the context of New Public Management's reliance on explicit targets; this topic is its full home treatment (see Topic 3.1 — Public Administration Systems and Models). Goodhart's law is not an argument against measurement — it is an argument that measurement and reward cannot be fused carelessly without changing what is being measured.

Perverse incentives are the mechanism through which Goodhart's law does its damage: an incentive structure that produces results contrary to what its designers intended, because the people responding to it are optimising for the letter of the reward rather than the purpose behind it (see perverse incentive). A target attached to funding, reputation, or a career creates a perverse incentive whenever there exists a way to improve the reported number without improving the underlying outcome — and in a service delivered through thousands of individual judgement calls, such a way almost always exists once people start looking for it.

A typology of dysfunctional responses to targets. The economist Peter Smith, in an influential 1995 study of public-sector performance data, catalogued the ways published targets distort behaviour, several of which remain the standard vocabulary for diagnosing a struggling indicator system. Tunnel vision is concentrating effort on what is measured at the expense of what is not, even when the unmeasured aspects matter as much or more. Measure fixation is treating success on the indicator as success on the underlying objective, so that the proxy quietly replaces the thing it was meant to represent in everyone's attention — the pattern behind teaching to the test in education systems, where instruction narrows to whatever a standardised assessment covers rather than the wider learning the assessment was meant to signal. Misrepresentation is the deliberate manipulation of reported data — reclassifying a case, adjusting a clock, choosing a more favourable denominator — so the figures improve without the reality changing at all. Ossification is a target's slow decay into irrelevance as it is left unrevised long after the operating environment it was designed for has changed. A closely related 2006 study of targets in the English public health care system by Gwyn Bevan and Christopher Hood showed how far these dynamics can run when a target is paired with severe consequences for missing it, coining the phrase 'targets and terror' for a regime that produces exactly the gaming behaviour it was designed to eliminate.

Quality assurance is the systematic, organisation-owned discipline of ensuring that a product or service consistently meets a defined standard, through defect prevention built into the process rather than inspection of the finished output alone (see quality assurance). Where a performance target measures an outcome after the fact, quality assurance is designed to make a poor outcome less likely to occur in the first place, typically through standardised procedures, checklists, peer review, and root-cause analysis of failures. Total quality management is the broader management philosophy built on this discipline, treating continuous improvement in quality as everyone's responsibility rather than a specialist function bolted onto the end of a process (see total quality management), and the ISO 9000 family of international standards published by the International Organization for Standardization (ISO), first issued in 1987, gives organisations a certifiable framework for demonstrating that such a management system exists. A body can be certified to a quality-management standard, and pass every internal quality-assurance check, while still gaming its externally reported performance targets; the two systems answer different questions and neither substitutes for the other.

Safety indicators and safety culture deserve separate treatment from ordinary performance indicators because the harm from a perverse incentive is often irreversible. In health care, patient safety practice has learned, often at high cost, that punishing staff for reporting an error suppresses exactly the information a safety system needs to prevent the next one — an insight that has spread from clinical settings into aviation, transport, and other high-consequence public services. A 'no-blame' or 'just culture' reporting environment treats an honestly reported near-miss as a valuable data point rather than an admission of guilt, which is the opposite of how most ordinary performance targets are designed to work: a performance regime rewards the good number, while a safety regime has to reward the honest number even when it is bad.

Independent inspection and ratings bodies provide a check on all of the above that comes from outside the chain of command being assessed. An inspectorate — such as England's Ofsted, which inspects education and children's-services providers, or the Care Quality Commission, which inspects and rates health and social care providers — sends assessors who observe practice directly, exercise professional judgement, and publish a rating that the inspected body does not control. This is structurally different from internal performance management: a self-reported waiting-time figure can be gamed by the people whose incentive is to game it, but an inspector who walks the wards or observes the classroom is looking at the reality the number was supposed to represent. Inspection is a form of horizontal accountability — scrutiny by a body of independent standing rather than by an electorate or a market (see Topic 3.4 — Accountability, Transparency, and Legitimacy) — and it earns that role only to the extent its independence, in funding, appointment, and publication, is genuinely protected.

Best practices

1. Derive indicators from the logic model, not from what is already easy to count. Choosing a measure because the data already exists, rather than because it tracks a genuine link in the causal chain, is how measure fixation and tunnel vision begin (see Topic 2.1 — The Public Value Scorecard and Outcomes Frameworks).
2. Pair every output or activity target with at least one balancing measure that would reveal the most likely gaming strategy. If a waiting-time target can be gamed by reclassifying cases, track the reclassification rate alongside the headline figure, not as an afterthought once a scandal forces the question.
3. Treat a target that has been hit for several consecutive periods as a prompt to investigate, not simply to celebrate. A number that stops discriminating between good and bad performance has either genuinely solved the problem or has been gamed into irrelevance, and the two look identical from the dashboard alone.
4. Separate who sets a target from who is measured against it from who independently checks the result. Collapsing these three roles into one part of the organisation removes the structural tension that keeps a measurement system honest.
5. Revise or retire a target on a fixed cycle, not only after visible failure. Ossification is a target's default fate; a scheduled review that asks whether the target still tracks anything meaningful catches the decay before a crisis does.
6. Build a no-blame reporting channel for safety-relevant near-misses and errors, kept structurally separate from performance-reward decisions. A safety system that punishes honest reporting will still have the same rate of underlying error — it will simply stop finding out about it until the harm is larger and harder to hide.
7. Give independent inspectors and ratings bodies genuine access, protected funding, and publication rights the inspected body cannot veto. An inspectorate that depends on the goodwill of the organisation it rates for its budget or its access is performing scrutiny, not providing it.
8. Use professional judgement to interpret a quantitative score, rather than letting the score stand alone. A single composite rating compresses a rich, uneven reality into one figure; publish the qualitative narrative behind an inspection rating alongside the rating itself, so a reader can see what the number is summarising.
9. Design targets with the frontline staff who will be measured against them, not only for them. People closest to the work usually know the easiest gaming strategy before any analyst does; involving them in design surfaces that risk early rather than after the target is already live.
10. Certify a systematic quality-management process, such as one built on the ISO 9000 family of standards, as a floor rather than a substitute for outcome measurement. Certification demonstrates a consistent process exists; it does not by itself demonstrate that the process produces a good outcome, and treating the certificate as sufficient evidence of quality is its own form of measure fixation.
11. Publish a target's known limitations alongside the target itself. Naming, in the same report as the figure, what the measure does not capture and how it could plausibly be gamed pre-empts exactly the discovery that otherwise damages legitimacy later (see Topic 3.4 — Accountability, Transparency, and Legitimacy).

Questions to discuss with your team

Which of our current published targets could be hit without the underlying outcome improving at all? This is the most direct application of Goodhart's law available to any team, and almost every organisation that has run numeric targets for more than a few years is carrying at least one live example. Walk through your headline indicators one at a time and ask, concretely, what behaviour each one rewards rather than what it was designed to measure. Discuss whether frontline staff already know an easier way to move the number than the intended one, since they usually do long before it reaches senior leaders. Treat a confident, unexamined 'nobody would do that here' as itself worth investigating rather than as reassurance.

Where have we let an easy-to-count output quietly stand in for a genuine quality or safety judgement? Outputs are almost always cheaper and faster to collect than outcomes, which creates a permanent temptation to report the former using the language of the latter. Identify one flagship indicator your organisation currently reports and ask, as a group, whether it actually measures the quality of what happened or only the volume of activity that took place. Discuss what a genuine quality measure for that same activity would require to collect, and what has stopped the organisation collecting it so far. This question often reveals that the barrier is resourcing rather than difficulty of definition.

If a member of staff reported their own near-miss or error honestly today, what would actually happen to them? The existence of a no-blame policy is not evidence that people believe it applies to them personally. Discuss a recent case, real or hypothetical, where an honest disclosure of an error would have tested the policy, and be specific about what the consequences were or would likely have been. Consider whether your safety-relevant reporting sits in the same system as your performance-reward decisions, since mixing the two is the most common way a nominally no-blame culture quietly punishes honesty anyway. This question is uncomfortable by design; a team that cannot discuss it openly has already answered it.

Does our inspection or ratings body have real independence from the organisation it assesses, or only formal independence? Independence in name is common; independence in funding, access, and the right to publish an unwelcome finding is rarer and easier to erode gradually. List the inspection or ratings relationships your organisation is subject to and rate each honestly on whether it could survive a determined attempt by the inspected body to shape the outcome. Discuss what has happened the last time an inspection produced a genuinely unwelcome result, and whether the response was to fix the underlying problem or to manage the relationship with the inspector.

Which of Smith's dysfunctional responses to targets — tunnel vision, measure fixation, misrepresentation, ossification — is currently most visible in our own reporting? Naming the specific failure mode, rather than discussing gaming in the abstract, makes the conversation concrete enough to act on. Go through the four patterns and ask which one the team recognises in its own recent experience, rather than assuming the organisation is exempt because no scandal has yet occurred. Discuss what a corrective step for that specific pattern would look like in the next reporting cycle, not as a general commitment to 'do better on measurement.'

If we stopped publicly reporting our current headline target for a year, what would we actually lose, and what might we gain? This question tests whether a target is genuinely steering decisions or has become an end in itself that nobody would defend if asked to justify it from first principles. Discuss honestly what decision, if any, has changed in the last year because of this specific number, as distinct from decisions that would have been made anyway. Consider what staff time and behavioural distortion the target currently costs, and whether a different, harder-to-game measure would serve the same accountability purpose better. This is not an argument for abandoning measurement — it is a discipline for keeping only the measurement that earns its place.

In practice: a public value example

The ambulance service of a mid-sized metropolitan region — fictional, illustrative of a pattern rather than a report on any real body — had operated for years under a published target requiring an emergency response within eight minutes for the most life-threatening calls. The target had been set with good intent, backed by clinical evidence that response speed genuinely affects survival for a defined category of emergencies, and the service's board treated the published percentage of calls meeting the target as its single most important indicator of organisational success.

Over three years, the reported figure improved steadily, from just above half of qualifying calls to nearly nine in ten, and senior leadership presented the trend at every board meeting as clear evidence that operational reforms were working. What the headline figure did not show was how the improvement had actually been achieved. Call handlers, aware their own performance was assessed partly on how many calls they classified into the target-bearing category, had begun — gradually, without any single decision to do so — downgrading a growing share of genuinely serious calls into a lower-priority category the eight-minute clock did not apply to, simply because a smaller, more tightly defined pool of calls was easier to hit the target against. Separately, crews had learned the clock stopped the moment a vehicle arrived at the scene, not when a patient was actually reached, which created a small but real incentive to park at the boundary rather than reach the patient fastest.

Neither practice was fraudulent in the sense of falsifying a record; each was a locally rational response to a target whose definition created room for exactly this drift — a textbook instance of measure fixation and misrepresentation operating together. The problem surfaced not through the service's own reporting, which continued to show improvement, but through an independent regional health inspectorate's routine inspection, which combined case-file review with direct observation of call-handling and found downgraded classifications inconsistent with the clinical detail recorded in the same calls. The inspectorate's rating dropped sharply, from a description equivalent to 'good' to 'requires improvement', despite the service's own published response-time figure being better than it had ever been.

The service's initial response — pointing to its own improving figures — made the credibility problem worse, because it answered a question the inspectorate had not asked; the inspection had never disputed the arithmetic, only what the arithmetic was measuring. Its more effective response was to redesign the indicator suite from the underlying logic model rather than patch the existing target: response time was redefined to the moment of patient contact rather than vehicle arrival, call classification was audited independently on a rolling sample rather than trusted to the handler's own record, and a parallel clinical-outcome measure — survival to hospital admission — was introduced because it was much harder to game by reclassification alone. The reported response-time percentage fell in the first year under the new definition, which the board had prepared its funders to expect; the clinical-outcome measure, tracked for the first time on a comparable basis, showed a smaller but genuine improvement the original target had never actually demonstrated one way or the other.

Four sector lenses

Local government

Local services such as fire and rescue, adult social care, and waste collection are usually measured against a blend of nationally mandated indicators the local authority did not choose and locally chosen measures reflecting a specific council's own priorities, so a single service can be judged a success against one indicator set and a concern against another. Inspection at local level is often delivered by a national or regional body with a statutory remit covering many local authorities at once, which gives useful comparative perspective but less granular local context than the organisation it is inspecting. Local elected members and scrutiny committees add a faster, more immediate layer of accountability, and are often the first to hear from residents when a published figure and lived experience diverge.

National government

National government typically runs the largest and most politically visible performance regimes, because a national target — a waiting-time standard, a crime-reduction goal, an exam-pass-rate objective — becomes a proxy in public debate for the competence of the government of the day, which raises the stakes on gaming far beyond the operational level. This is the setting where Bevan and Hood's 'targets and terror' dynamic is most fully realised, because national targets are frequently paired with visible political consequences for missing them, sharpening the incentive to manage the number rather than the reality behind it. National inspectorates, precisely because they operate at arm's length from the ministries whose political fortunes a rating can affect, carry a disproportionate share of the burden for surfacing the gap between reported and actual performance (see Topic 3.4 — Accountability, Transparency, and Legitimacy).

Social sector and nonprofit

Nonprofits and non-governmental organisations more often face quality assurance and accreditation as a voluntary or funder-imposed discipline than as a statutory inspection regime, since few jurisdictions run a dedicated independent inspectorate for the sector as a whole. This leaves social-sector quality assurance more fragmented — different funders imposing different requirements on the same organisation — but also more flexible, since a nonprofit can often design its own indicator suite around its actual theory of change rather than a set of measures imposed from outside. The absence of a mandatory external inspector raises the relative importance of an independent board and external audit as the sector's own horizontal-accountability substitute (see Topic 3.4 — Accountability, Transparency, and Legitimacy), since a nonprofit's own leadership is otherwise both the measurer and the measured.

Multilateral and international

Multilateral bodies and the comparative indices they publish or rely on — cross-country rankings of governance quality, ease of doing business, or public-service performance — face Goodhart's law at the scale of entire national governments, where a country can improve its ranking through reforms targeted narrowly at the measured criteria rather than the broader condition the index claims to represent. Independent inspection in the domestic sense used elsewhere in this topic has no real equivalent across sovereign borders, since no state accepts an external inspectorate with the coercive standing a national body has over an agency it oversees; multilateral quality assurance instead relies on peer review, self-reporting validated by technical missions, and the reputational cost of a published rating. This makes the independence and technical credibility of the body compiling any cross-country indicator especially important, since it is often the only check available.

Common failure modes

Tunnel vision. An organisation concentrates effort and resource on whatever is measured, at the direct expense of unmeasured aspects of the same service that matter just as much to the people it serves. The cost is a service that looks like it is improving in every published report while an unmeasured dimension — patient dignity, case complexity, long-term outcomes — quietly deteriorates, unnoticed until it becomes a crisis large enough to force new measurement.

Measure fixation. Success on the indicator comes to be treated, inside the organisation, as equivalent to success on the underlying objective, so the proxy displaces the goal in everyone's actual attention even while the formal mission statement still names the goal. The cost is a widening, self-reinforcing gap between reported performance and lived reality, because nobody inside the organisation is any longer watching the reality directly.

Misrepresentation. Data is adjusted — a clock stopped early, a case reclassified, a denominator chosen for favourable effect — so the reported figure improves without the underlying reality changing at all, as in the worked example above. The cost, once discovered, is disproportionate to the original distortion, because misrepresentation damages not only the specific figure but the credibility of every other number the same organisation has ever published (see Topic 3.4 — Accountability, Transparency, and Legitimacy).

Ossification. A target set for a specific operating context is left unrevised long after that context has changed, so it continues to be reported and managed against even though it no longer discriminates between genuinely good and genuinely poor performance. The cost is invisible for years, because an ossified target still produces a number; the number has simply stopped meaning anything, and the organisation's real management attention has usually already moved elsewhere without anyone formally retiring the measure.

Inspectorate capture. An independent inspection or ratings body becomes gradually dependent on the goodwill, access, or continued cooperation of the organisations it inspects, whether through funding arrangements, career paths that move staff between inspector and inspected, or informal relationships built over repeated inspections. The cost is a slow loss of real independence that a formal charter or statute does not by itself prevent, and it typically surfaces only when a major failure passes through a rating system that should have caught it.

Blame-driven safety reporting. A safety-relevant error or near-miss reporting system is nominally no-blame but is, in practice, connected closely enough to performance appraisal or disciplinary process that staff learn reporting an honest mistake carries personal risk. The cost is not fewer errors — it is fewer reports of the errors that continue to happen, which removes exactly the early-warning signal a safety system exists to provide before the next, more serious incident occurs.

Maturity model

Checklist

• Trace each headline performance indicator back to the logic model or outcomes framework it is meant to represent.
• Identify the most plausible way each published target could be gamed, and name a balancing measure that would reveal it.
• Confirm that whoever sets a target, whoever is measured against it, and whoever independently checks the result are not all the same part of the organisation.
• Schedule a fixed-cycle review of every live target, independent of any visible failure prompting it.
• Confirm the safety-relevant reporting channel is structurally separate from performance-reward and disciplinary decisions.
• Test whether staff believe an honestly reported error or near-miss would be treated without blame, not only whether a policy says so.
• Confirm any inspection or ratings body assessing the organisation has protected funding, access, and publication rights.
• Publish the known limitations of each headline target alongside the figure itself.
• Check whether any indicator has been hit consistently for several periods without a corresponding investigation into why.
• Involve frontline staff in the design of any new target before it goes live.
• Confirm a quality-management certification, where one exists, is treated as a floor and not cited as evidence of good outcomes on its own.

Key sources

• Charles Goodhart's 1975 formulation of what became known as Goodhart's law, developed in the context of monetary policy targeting.
• Marilyn Strathern, ''Improving ratings': audit in the British university system', European Review, 5(3) (1997), 305–321 — the widely cited public-sector restatement of Goodhart's law.
• Peter Smith, 'On the Unintended Consequences of Publishing Performance Data in the Public Sector', International Journal of Public Administration, 18(2–3) (1995), 277–310 — the source of the tunnel vision, measure fixation, misrepresentation, and ossification typology this topic draws on.
• Gwyn Bevan and Christopher Hood, 'What's Measured Is What Matters: Targets and Gaming in the English Public Health Care System', Public Administration, 84(3) (2006), 517–538 — the 'targets and terror' study of gaming under severe consequences.
• Jerry Z. Muller, The Tyranny of Metrics (Princeton University Press, 2018) — a wide-ranging account of measurement failure across education, health care, business, and government.
• UK National Audit Office, 'Performance measurement by regulators', and US Government Accountability Office guidance distinguishing performance measures from programme evaluation.
• The Care Quality Commission and Ofsted, as examples of independent inspection and ratings bodies operating at arm's length from the organisations they assess.

References

1. Goodhart's law — Wikipedia — https://en.wikipedia.org/wiki/Goodhart%27s_law
2. Charles Goodhart — Wikipedia — https://en.wikipedia.org/wiki/Charles_Goodhart
3. Marilyn Strathern — Wikipedia — https://en.wikipedia.org/wiki/Marilyn_Strathern
4. Performance indicator — Wikipedia — https://en.wikipedia.org/wiki/Performance_indicator
5. Perverse incentive — Wikipedia — https://en.wikipedia.org/wiki/Perverse_incentive
6. Quality assurance — Wikipedia — https://en.wikipedia.org/wiki/Quality_assurance
7. Total quality management — Wikipedia — https://en.wikipedia.org/wiki/Total_quality_management
8. ISO 9000 family — Wikipedia — https://en.wikipedia.org/wiki/ISO_9000
9. Patient safety — Wikipedia — https://en.wikipedia.org/wiki/Patient_safety
10. Teaching to the test — Wikipedia — https://en.wikipedia.org/wiki/Teaching_to_the_test
11. Care Quality Commission — Wikipedia — https://en.wikipedia.org/wiki/Care_Quality_Commission
12. Ofsted — Wikipedia — https://en.wikipedia.org/wiki/Ofsted
13. Marilyn Strathern, ''Improving ratings': audit in the British university system' — European Review / EconPapers record — https://econpapers.repec.org/article/cupeurrev/v_3a5_3ay_3a1997_3ai_3a03_3ap_3a305-321_5f00.htm
14. Peter Smith, 'On the Unintended Consequences of Publishing Performance Data in the Public Sector' — citation record — Scientific Research Publishing — https://www.scirp.org/reference/referencespapers?referenceid=1708685
15. Gwyn Bevan and Christopher Hood, 'What's Measured Is What Matters: Targets and Gaming in the English Public Health Care System' — LSE Research Online — https://researchonline.lse.ac.uk/id/eprint/16211/
16. Jerry Z. Muller, The Tyranny of Metrics — Princeton University Press — https://press.princeton.edu/books/hardcover/9780691174952/the-tyranny-of-metrics
17. Performance measurement by regulators — UK National Audit Office — https://www.nao.org.uk/insights/performance-measurement/
18. Performance Measurement and Evaluation: Definitions and Relationships — US Government Accountability Office — https://www.gao.gov/products/gao-11-646sp
19. Care Quality Commission — official site — https://www.cqc.org.uk/
20. Ofsted — GOV.UK — https://www.gov.uk/government/organisations/ofsted