Key takeaways
  • Incident rates are lagging, heavily influenced by luck, and can be reduced by discouraging reporting.
  • Leading indicators measure activity and conditions that precede incidents, and they can be influenced before anyone is hurt.
  • A falling near-miss report count is usually a warning, not an achievement.
The metric that can lie to you
A falling incident rate

Rates fall when things get safer, and they also fall when reporting becomes unwelcome, when definitions shift, or when the numbers are small enough that luck dominates. Distinguishing these requires looking at reporting behaviour alongside the rate.

Why lagging indicators mislead

  • Small numbers. A site with two recordable incidents a year cannot detect a real change in a year.
  • Severity is largely luck. The same event can be a near miss or a fatality depending on where someone stood.
  • They are influenceable by classification, and by pressure not to report.
  • They tell you about the past, and only after someone was already hurt.

Leading indicators worth tracking

  • Near miss and hazard reports per person, with the trend more important than the level.
  • Percentage of reported hazards closed within a target time.
  • Safety observations or behavioural conversations completed against plan.
  • Overdue safety actions, and their age.
  • Training and competence currency for the hazards people actually face.
  • Permit audits: not permits issued, but permits checked in the field against actual conditions.
  • Emergency drills completed and the findings acted upon.

Reading reporting volume correctly

Rising near-miss reports usually mean people trust the process and are engaged, not that the site became more dangerous. Falling reports in a stable operation is the signal to investigate: it commonly follows a disciplinary response to a report, a backlog of unactioned items, or a message that the site is chasing a target.

Qualitative signals

Ask a few direct questions in a walkaround: can you stop the job if you think it is unsafe, and has anyone ever done that here? Has anything you reported been fixed? What is the most dangerous thing about this job? The answers are more diagnostic than most survey instruments, and the pattern across people matters more than any single response.

What to avoid

Incentives tied to low incident numbers. They reliably reduce reporting rather than injury, and they make the resulting data unusable for the improvement it was supposed to support. If recognition is used at all, tie it to reporting and to closing actions.