Every year, someone publishes a list. Federal agencies are ranked by employee engagement. Hospitals are sorted into star ratings. Police precincts are compared on crime reduction, and school districts are arranged in league tables. The premise is simple: put organisations side by side, and the laggards will feel pressure to catch up.

The results rarely match the premise. Some ranked agencies genuinely improve. Others improve only on paper. A few become so preoccupied with their position that the work the ranking was meant to measure quietly suffers. The same instrument, applied to broadly similar organisations, produces strikingly different behaviour.

Organisational analysis can help explain this puzzle. A ranking is not a neutral mirror. It is an intervention. It changes incentives, redistributes attention, and alters what managers believe they will be held accountable for. Knowing how agencies read and respond to that signal is the key to knowing when comparison helps and when it does harm.

Rankings differ from absolute performance targets in one crucial respect: they are relative. An agency can improve on every measure and still fall in the table because its peers improved faster. Performance management stops asking 'are we doing well?' and starts asking 'are we doing better than them?' Economists call this yardstick competition. Inside the agency, it often feels like a zero-sum contest.

The morale effects are asymmetric. Agencies near the top tend to treat a ranking as validation, and they use it in recruitment and budget negotiations. Agencies near the bottom experience it as public reproach, often delivered without context about mission difficulty or resource levels. Studies of naming-and-shaming regimes, such as the star ratings applied to English hospital trusts in the early 2000s, found that low ratings could prompt rapid effort. They could also prompt defensiveness, turnover among senior managers, and a sense that leaders were being judged on factors beyond their control.

Risk-taking shifts as well. Top-ranked agencies have a position to protect, which encourages caution. Experimentation that might temporarily depress scores becomes harder to justify. Bottom-ranked agencies face the opposite calculation. With little to lose, some pursue bold reforms, while others gamble on quick fixes that promise visible movement. The middle of the table, often the largest group, tends toward conformity and copies whatever the leaders appear to be doing.

Most consequentially, rankings reorder priorities. James Q. Wilson observed that agencies gravitate toward tasks whose results can be seen. A ranking amplifies this tendency by declaring, publicly, which results count. Unmeasured duties lose standing in internal resource battles. These include maintaining relationships with other levels of government, preserving institutional knowledge, and handling rare but serious cases. They lose standing not because anyone decided they matter less, but because no one is keeping score.

Takeaway

A ranking does not simply report an agency's priorities. It quietly rewrites them by signalling to managers which results will be seen and which will not.

Donald Campbell's warning has proved remarkably durable. The more a quantitative indicator is used for social decision-making, the more it will be subject to corruption pressures, and the more it will distort the processes it was meant to monitor. Rankings intensify that pressure because the stakes are both comparative and public. The responses that follow are not random. They fall into recognisable patterns that any analyst should expect.

The first is selection, sometimes called cream-skimming. Agencies judged on outcomes have reason to serve the clients most likely to succeed. Evaluations of US job training programmes found evidence that providers rewarded for placement rates favoured applicants who might well have found work anyway. The second is reclassification. When crime statistics drive precinct comparisons, there is steady pressure to downgrade serious offences or to discourage reports altogether. Retired officers described this pattern in surveys about CompStat-era policing.

The third is threshold behaviour. Where a system rewards meeting a cut-off, effort clusters just above the line. Under England's four-hour emergency department target, analysts observed sharp spikes in patients admitted or discharged in the minutes before the deadline. The fourth is the ratchet effect, familiar from Soviet planning. Managers who expect next year's benchmark to be set by this year's results have reason to avoid exceeding expectations too visibly.

None of this requires bad faith. Gaming typically emerges from ordinary people responding sensibly to the incentives in front of them, often with leadership's tacit approval. Gwyn Bevan and Christopher Hood characterised the English health regime as 'targets and terror' precisely because the pressure flowed from the top. The analytical lesson is that gaming is a property of the measurement design, not merely a failure of individual integrity.

Takeaway

When an agency games a ranking, it is usually revealing something precise about where the measure has drifted away from the mission.

The evidence does not suggest abandoning comparison. Researchers compared England's naming-and-shaming approach with the gentler regimes in Wales and Scotland and found that England's waiting times fell faster. Public comparison generated effort that private reporting did not. The design question is how to capture that energy while limiting the distortions that come with it.

Several principles recur in systems that work. First, compare like with like. Risk adjustment and peer grouping reduce the sense of injustice that fuels both demoralisation and gaming. That means placing rural agencies with rural agencies, or adjusting for caseload complexity. Second, use a portfolio of measures that pull in different directions. An emergency department judged on both speed and readmission rates finds it harder to win by rushing patients out the door.

Third, audit the numbers and rotate emphasis. Indicators that are periodically verified and occasionally revised are harder to optimise against. Fourth, separate learning comparisons from accountability comparisons. Municipal benchmarking networks often share detailed data privately among participants, who use it to find peers worth learning from, while publishing only broader results. Candid diagnosis thrives where a poor number is treated as a question rather than a verdict.

Finally, match the instrument to the agency. Wilson distinguished production organisations, whose outputs and outcomes are both observable, from coping organisations, where neither is. Rankings are reasonably informative for a passport office and far less so for a diplomatic service or a child protection unit. In those settings, comparison works better as a prompt for professional review than as a league table read off a spreadsheet. Professional review can mean inspection, case audit, or peer visits.

Takeaway

The most useful comparisons treat a low score as the start of an inquiry rather than the end of an argument.

Performance rankings are best understood as governance interventions rather than measurement tools. They redirect attention, reshape risk appetites, and create incentives that agencies will respond to. Sometimes they respond by improving, and sometimes by manoeuvring.

The pattern of response is largely predictable from the design. Rankings built on narrow, unadjusted, high-stakes indicators invite selection, reclassification, and threshold behaviour. Rankings built on balanced measures, fair peer groups, and verified data are more likely to turn competitive energy into genuine improvement.

The broader principle runs throughout public administration: institutions behave according to how they are observed. Designing that observation carefully is not a technical afterthought. It is part of the policy itself.