What Your Safety Numbers Can (and Can’t) Prove

Ask most companies how they know their safety training is working, and the answer usually comes back to one number: their injury rate. If it went down, the training worked. If it went up, something needs to change. 

The third principle in The Three Principles, says that logic doesn't hold up, and that what you choose to measure decides what you can honestly claim. 

Two Kinds of Safety Numbers 

Safety data falls into two broad groups. Leading indicators measure what happens before an incident: what people know, whether they can spot a hazard, whether they follow safe work practices. Lagging indicators measure what happens after: recordable injuries, lost-time incidents, fatalities. 

The study finds that the research is clear on which ones training actually moves. Knowledge, hazard recognition and safe behaviour improve in study after study, by large margins. Injury rates at a single company are a different story. 

Why One Company's Injury Rate Is Mostly Noise 

Recordable injuries are, thankfully, rare. The study gives the example of a typical construction company with 2,000 workers, which might record 30 to 60 recordable injuries a year. With numbers that small, a few incidents either way can swing the rate dramatically. 

That goes both ways. A company can genuinely improve how ready its people are and see no meaningful change in its injury rate over a year. Another company can change nothing at all and see a 30 percent drop by chance. As the study puts it, “The statistics do not care what the press release says.” 

Trends across several years and several sites are meaningful. One company's single-year rate, according to the study, is “statistically, almost noise.” Yet that's the number that often decides bids, supplier audits and internal performance reviews. 

What to Measure Instead 

The study doesn't say to stop tracking injuries. It says to track them honestly, over enough time and enough sites for the numbers to mean something. Alongside them, it recommends rewarding the leading indicators the research backs, including: 

• Knowledge gains and comprehension rates 

• How well people identify hazards 

• How many near misses get reported 

• How often toolbox talks happen, and how good they are 

• How long it takes to get someone cleared to work 

• Whether readiness checks are completed before work starts 

The same logic applies when choosing a training provider. The study is direct about it: a vendor claiming a 30 or 40 percent drop in injury rates from one customer's first year is “either misreading the statistics, getting lucky on a small sample, or making claims they cannot defend if pressed.” A provider that improves leading indicators and reports on injury rates honestly, limits included, is the more useful partner. 

Measuring What Training Changes 

LUMA1 is built around the numbers the training can move. Every training touchpoint records not just whether someone finished, but whether they understood it, so comprehension is tracked worker by worker. The Workforce Readiness Dashboard shows who is cleared to work right now, who isn't and why, which certifications are missing or expiring, and how readiness compares across trades and sites. 

Those are numbers you can act on this week, rather than waiting a year to see whether an injury rate moved for reasons nobody can prove. 

To see how it works in practice, book a demo at luma1.com. You can read the full study here.