Donald Campbell was a social psychologist working on how to evaluate government programs, and his 1979 formulation is longer than the epigram it gets reduced to: 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.

Two claims, and the second is the one that gets dropped.

The difference from Goodhart

They are usually quoted as the same law and they are not, which matters because the remedies differ.

Goodhart's law is about the measure: once a measure becomes a target, it stops being a good measure, because the relationship between the indicator and the thing it indicated was incidental and pressure breaks it. The failure is epistemic — you can no longer read the number.

Campbell's second clause is about the activity: the process being measured is itself deformed by the measuring. The failure is not that you have lost information about teaching; it is that the teaching is now worse.

Campbell's own example was standardized testing, and he set out the sequence before it happened at national scale. Scores rise. Some of the rise is inflation — coaching, exclusion of weak candidates, outright fraud — which is the Goodhart part. And some of it is real, achieved by moving effort from untested subjects to tested ones, which is not cheating and is worse, because the system is functioning as designed and producing a narrower education.

The sequence

It runs the same way wherever an indicator carries a consequence.

The indicator is chosen because it correlates with something that matters and is easy to count.

It acquires a consequence — funding, a bonus, a ranking, a job.

Effort reallocates toward it, which is what incentives are for and is not yet a problem.

Effort reallocates away from everything not counted, which is the deformation, and which nobody decided.

The number improves and the thing it stood for does not, and the improvement is used as evidence that the scheme is working.

The fourth step is the one to watch, because it is invisible in the data by construction. What was not measured cannot show up as having declined.

What to do about it

Never let one indicator carry a consequence alone. A basket is harder to game and, more importantly, makes the reallocation visible — if one number rises while another falls, the trade has surfaced.

Measure the things you are afraid will be sacrificed. Not to target them, only to watch them. An unwatched dimension is the one that pays for the improvement.

Keep some judgment in the loop. Campbell's law applies most sharply where the indicator is applied mechanically, because a mechanical rule cannot notice that it is being satisfied in a way nobody intended.

Change the indicator before it matures. The corruption takes time to develop, since people have to learn how the measure responds. Rotating what is measured keeps the learning from paying off, at the cost of making trends harder to read.

Where it does not apply

Where the measure and the thing are the same. Revenue is not a proxy for revenue, and cash in the bank is not an indicator of cash in the bank. The law bites on proxies — a test standing in for learning, a response time standing in for service, an activity count standing in for work — and the degree of danger tracks how loosely the proxy is attached to what anybody actually wanted.