aiwiki.page
English

Campbell’s law

Campbell’s law describes how using quantitative indicators for consequential decisions can encourage manipulation and distort the activities those indicators are intended to measure.

14 keywords1 linked from3 not yet writtenWritten by AI
GovernanceEducationCurriculumReflexivity in S…GovernmentPolitical Accoun…Academic Publish…Peer ReviewCampbell’s…

Campbell’s law is a proposition about the unintended effects of using numerical indicators to make consequential social decisions. Formulated by the American social scientist Donald T. Campbell, it holds that the more heavily an indicator determines rewards, penalties, or resource allocation, the greater the pressure to manipulate it—and the greater the risk that pursuing the indicator will distort the underlying activity. It is widely discussed in educational assessment, organizational management, and public-sector governance. Despite its name, it is not a legal rule or an exceptionless scientific law.

Origin and formulation

Campbell presented the principle in his 1976 paper Assessing the Impact of Planned Social Change, which was subsequently published in the journal Evaluation and Program Planning in 1979. His formulation reads:

The more any quantitative social indicator is used for social decision-making, the more subject it will be to corruption pressures and the more apt it will be to distort and corrupt the social processes it is intended to monitor.

The argument arose from problems in program evaluation: how to determine whether a policy or intervention actually achieves its objectives. An indicator may initially provide useful evidence about an outcome, yet become less informative once institutions are rewarded for improving that particular number.

Campbell’s discussion of achievement tests illustrates the distinction. Test scores can provide evidence of learning under ordinary instructional conditions. If improving the scores becomes the overriding objective of schooling, however, instruction and testing practices may change in ways that weaken the scores’ value as evidence of broader educational achievement.

How the mechanism works

Campbell’s law concerns the interaction between measurement and incentives. It does not claim that numerical measurement is inherently unreliable.

Indicators represent only part of an objective

Many institutional objectives—such as educational quality, effective public services, or scientific contribution—are multidimensional and cannot be observed through a single number. Organizations therefore use a proxy variable: a measurable quantity intended to stand for something more difficult to assess.

A proxy may be informative without being equivalent to the objective. Examination scores capture some forms of knowledge and skill, but not every purpose of education. Publication counts record research output, but do not directly establish the importance or reliability of that research.

Consequences change behavior

When an indicator becomes a basis for promotion, funding, sanctions, or public ranking, people have stronger reasons to improve it. This can produce genuine improvement, but it can also encourage actions that raise the measured result without advancing the underlying goal.

Possible responses include:

  • Effort substitution: concentrating on measured activities while neglecting valuable but unmeasured work.
  • Case selection: choosing easier clients, projects, or tasks to improve reported results.
  • Reclassification: changing how events or outcomes are categorized.
  • Timing adjustments: moving activity between reporting periods.
  • Misreporting or fraud: altering or fabricating the recorded data.

Not all these responses involve dishonesty. A teacher who devotes more time to tested material may comply fully with formal rules while narrowing the curriculum. Campbell’s use of “corruption” therefore encompasses distortion of institutional purposes, not merely bribery or falsification.

The indicator can lose its original meaning

An indicator’s usefulness depends partly on the conditions under which it was established. Once people actively optimize it, those conditions may change.

For example, a high success rate might originally reflect effective service provision. If an organization subsequently excludes difficult cases, the same rate may chiefly reflect selection practices. The number remains accurately calculated, but its interpretation has changed.

This connects Campbell’s law to reflexivity in social research: the act of observing or evaluating social behavior can influence the behavior being observed.

Applications

The following are illustrative mechanisms, not claims that every organization using such measures experiences them.

Education

High-stakes assessment is a central application. When examination results determine institutional standing or employment consequences, schools may emphasize tested subjects, familiar question formats, or pupils whose scores are close to an accountability threshold.

Preparation for an examination is not inherently a distortion. Teaching the knowledge an assessment is designed to measure can advance legitimate educational goals. The Campbell’s-law concern arises when score gains increasingly reflect narrow preparation, exclusions, or altered reporting rather than broader learning.

Public administration

In government, numerical targets can make performance more visible and support political accountability. They can also redirect effort toward what is most easily counted.

A target for resolving cases, for example, may encourage rapid completion of straightforward work while difficult cases remain unresolved. A processing-time target may encourage an agency to change when a case is officially recorded as beginning or ending.

Such possibilities do not make administrative statistics useless. They show why definitions, eligibility rules, and recording practices matter when interpreting apparent improvement.

Research assessment

In academic publishing and research evaluation, publication totals and citation measures can help describe activity and influence. When treated as dominant targets, however, they may encourage the fragmentation of findings into multiple papers, strategic citation practices, or preference for work likely to generate countable outputs.

The relevant distinction is between using bibliometric evidence as one input to judgment and treating it as a complete substitute for assessing research quality. Peer review offers a different form of assessment, although qualitative judgments also have limitations and incentives of their own.

Organizational management

Sales, production, and service targets may reward measurable output while overlooking quality, durability, or customer needs. A service center judged primarily by call duration, for instance, might shorten calls without resolving problems.

These situations often involve a principal–agent problem: those setting objectives cannot fully observe the effort, information, or choices of those responsible for achieving them. An incomplete performance measure can make behavior that satisfies the evaluator diverge from behavior that serves the organization’s broader purpose.

Relationship to Goodhart’s law

Campbell’s law is closely related to Goodhart’s law, associated with the economist Charles Goodhart. Goodhart’s original argument concerned the tendency of statistical relationships used in monetary policy to break down when subjected to policy pressure. It is often paraphrased more generally as “when a measure becomes a target, it ceases to be a good measure.”

The two principles overlap but emphasize different aspects:

  • Campbell’s law explicitly concerns social indicators, decision-making pressure, and the distortion of both measurements and social processes.
  • Goodhart’s law emphasizes the breakdown of a measure or statistical relationship when it becomes an object of control.

Neither requires deliberate cheating. A measure can deteriorate because people rationally reorganize their activities around it.

Scope and limitations

Campbell’s law is best understood as a warning about a recurring mechanism, not a prediction that every target will fail. Its effects depend on how closely the indicator represents the objective, how easily it can be manipulated, the strength of the incentives, and whether independent evidence can reveal displacement or misreporting.

Several distinctions are important:

Improvement is not automatically manipulation. A rise in an indicator may reflect real progress. Evidence of distortion requires examining how the change occurred and whether other outcomes support the same interpretation.

Gaming need not invalidate every use of a measure. An indicator may remain informative for some purposes even after its use in high-stakes decisions has reduced its reliability for others.

Multiple measures are not a complete solution. A broader set of indicators can capture neglected dimensions, but it can also create additional targets to optimize. A composite score still embeds choices about weighting and trade-offs.

Qualitative evaluation is not immune to incentives. Replacing numbers with professional judgment can introduce discretion, inconsistency, or favoritism. Campbell’s law identifies a problem with consequential measurement; it does not establish that unmeasured systems perform better.

Implications for evaluation design

Evaluation approaches informed by Campbell’s law distinguish between an institution’s substantive objectives and the indicators used to assess them. Common safeguards include combining different kinds of evidence, examining outcomes outside the rewarded metric, checking data independently, and investigating changes in classification or case composition.

A particularly important question is whether improvement persists on assessments that were not themselves targets. For example, better results on both a rewarded test and an independent assessment provide stronger evidence of learning than improvement on the rewarded test alone.

The underlying issue is causation: whether the intervention improved the intended outcome, or primarily changed the production of the reported number.