The reason a safety bonus can hide accidents

A site introduces a bonus for completing a quarter with no recordable incidents. Recorded incidents fall by half within two quarters. Minor first-aid entries in the on-site log, which carry no bonus consequence, do not fall at all. Both numbers are accurate, and they describe different things.

What everyone sees

The falling incident count is presented as a safety improvement and celebrated as one. The bonus appears to have worked exactly as designed, and the design is unobjectionable: reward the outcome you want. The measure and the outcome are treated as the same object because the measure is the only view of the outcome available.

What is actually happening

When a metric depends on voluntary disclosure, an incentive attached to it acts on the disclosure as well as on the underlying event. Ruser and Butler and later work on injury reporting found evidence that safety incentive programmes and experience-rated premiums are associated with reduced reporting rather than reduced incidence alone. This is the reporting variant of Goodhart’s law: the measure ceases to be a good measure once it becomes a target.

Why it stays hidden

The suppression hides because non-reporting leaves no record by definition. There is no dataset of the incidents that were not written down, so the improvement is unfalsifiable from inside the system. Social pressure completes the concealment: reporting an incident now costs colleagues money, so silence is enforced horizontally rather than by management, and nothing in the paperwork shows it.