Start With the Right Metric
Raw injury counts are misleading because they ignore how much work was done. A site with 20 employees and a site with 200 will naturally record different numbers. That's why safety teams use frequency rates, which standardise injuries against hours worked.
The most common is the Lost Time Injury Frequency Rate (LTIFR):
LTIFR = (Number of lost time injuries ÷ Total hours worked) × 1,000,000
Worked example: a warehouse records 3 lost time injuries over 450,000 hours worked. Its LTIFR is (3 ÷ 450,000) × 1,000,000 = 6.67 injuries per million hours.
Because the rate is standardised, you can compare it across sites, years or industries.
Leading vs Lagging Indicators
LTIFR is a lagging indicator. It tells you what has already gone wrong. To judge training, you also need leading indicators, which measure the activities that prevent incidents. Useful examples include:
- The percentage of operators holding current licences and certifications
- Near-miss reports per month (more reporting often signals a healthier safety culture)
- Completion rates for pre-start equipment checks
- Results of practical competency assessments
Tracking both types lets you see whether training is changing behaviour before it shows up in the injury numbers.
Beware of Small Numbers
This is where many safety reports go wrong. Injuries are relatively rare events, and rare events are noisy. If a site averages three incidents a year, it would not be unusual to see one in a good year and five in a bad one, purely by chance. Count data like this often follows a Poisson distribution, where the variance equals the mean, so year-to-year swings can look dramatic without meaning anything.
Before declaring that training has worked, ask whether the change is larger than normal random variation. A simple control chart, plotting the rate over time with upper and lower limits, makes this much easier to see.
Watch Out for Regression to the Mean
Organisations often roll out new training straight after a bad year. The trouble is that an unusually bad year is likely to be followed by a more typical one regardless of what you do. This is regression to the mean, and it can make almost any intervention look effective.
To avoid being fooled, compare against a longer baseline (three to five years if possible) rather than just the previous year, and where you can, compare trained teams with similar teams that haven't yet completed the course.
Design Your Training Rollout Like an Experiment
The best evidence comes from planning measurement before training begins. Record a baseline, roll training out in stages, and track the same indicators throughout.
For example, a Hunter Region manufacturer that books Newcastle site training for its forklift and elevated work platform operators could train one shift first and compare its near-miss reports and competency results with an untrained shift over the following months. Because on-site sessions take place in the team's real work environment, the conditions are consistent, which makes the comparison cleaner.
Report the Uncertainty
When presenting results to management, avoid single numbers stripped of context. Report rates alongside confidence intervals, show the trend over several periods, and be clear about sample sizes. A drop from 6.7 to 4.1 sounds impressive, but if it's based on a handful of incidents, the honest conclusion may be "promising, but keep watching."
The Bottom Line
Safety training protects people, and that alone makes it worthwhile. But measuring its impact properly helps you invest in the courses that work and refine the ones that don't. Choose standardised rates, track leading indicators, respect random variation and plan your measurement before the training starts. Good statistics won't replace good training, but they will tell you what your training is really achieving.