Injuries are rare, random, and late. A safety scorecard built only on injury rates tells you that the system failed after someone was hurt, and it moves with chance as much as with safety. A good set of measures balances the lagging rates that show results with leading indicators that show whether the work of prevention is being done.
This guide covers leading and lagging indicators, the common rates (TRIR, DART, LTIR, LTIFR, and severity), how to choose leading indicators, and the trap of small numbers: why a fall in the rate may be only chance, and how to show exact intervals, compare rates, and chart a u chart. A worked year at an illustrative plant puts it all together.
Figures in the example are illustrative. Use the definitions and recording rules that apply to you.
Before You Start
Why Measure Safety
What gets measured gets managed, but measuring safety by injuries alone is like steering by the rear-view mirror. Injuries are rare, random, and late: they tell you that the system failed after someone was hurt. A good set of safety measures also tells you whether the system that prevents injuries is working today.
See What Is Happening
Rates and trends show whether safety is getting better or worse, and where.
Steer Before Someone Is Hurt
Leading indicators show whether the activities that prevent injuries are being done.
Compare Fairly
Rates adjust for hours worked, so you can compare periods, sites, and benchmarks.
Drive Action
A measure that nobody acts on is decoration. Each one should have an owner and a response.
Leading and Lagging Indicators
| Lagging indicators | Leading indicators | |
|---|---|---|
| What they measure | Outcomes: harm that has already happened | Inputs and activities: the work that prevents harm |
| Examples | Recordable cases, lost-time cases, days lost, fatalities, workers’ compensation cost | Hazards reported and closed, inspections done, training completed, audit scores, actions closed on time, near misses reported |
| Strength | Clear, comparable, and what regulators and customers ask for | Timely, frequent, and can be acted on before someone is hurt |
| Weakness | Late, rare, noisy, and can be hidden by under-reporting | Can become a count of activities that does not mean anything is safer |
| Use | To judge results and compare | To steer and to give early warning |
The Common Lagging Rates
Rates convert a count of cases into a number that accounts for how much work was done, so that you can compare periods and sites of different size. Definitions vary by country, so use the one that applies to you.
| Measure | Formula | Notes |
|---|---|---|
| TRIR (total recordable incident rate) | Recordable cases × 200,000 ÷ hours worked | US OSHA convention. 200,000 hours is 100 full-time employees for a year |
| DART rate (days away, restricted, or transferred) | DART cases × 200,000 ÷ hours worked | Cases serious enough to need time away, restricted work, or a job transfer |
| LTIR (lost-time incident rate) | Lost-time cases × 200,000 ÷ hours worked | Cases with at least one day away from work |
| LTIFR (lost-time injury frequency rate) | Lost-time injuries × 1,000,000 ÷ hours worked | Used in many countries and in ISO-style reporting |
| Severity rate | Days lost × 200,000 ÷ hours worked | Days of absence per 100 employees. A few long absences can dominate it |
| Near-miss rate | Near misses × 200,000 ÷ hours worked, or per 100 employees | A leading indicator: a higher rate usually means better reporting |
Check the recording and reporting rules where you operate. In the US, OSHA 29 CFR 1904 defines what is recordable and how to calculate the rates. Whatever you use, define each measure in writing so that it is calculated the same way each time.
Choosing Leading Indicators
| Area | Example indicator | What it shows |
|---|---|---|
| Participation | Hazard and near-miss reports per 100 employees per month | Whether people trust and use the system |
| Fixing hazards | Percentage of hazard reports and actions closed on time; average age of open actions | Whether found problems are fixed |
| Inspections and audits | Planned inspections completed; audit score; critical findings open | Whether controls are checked and holding |
| Risk assessment | High-risk tasks with a current assessment | Whether the main hazards are understood |
| Training | Required training completed on time | Whether people are competent |
| Leadership | Leader safety walks completed; committee actions closed on time | Whether management is visible and acts |
| Maintenance of controls | Guard, interlock, and emergency equipment checks done on time | Whether the physical controls work |
- Keep the set small. Five to eight indicators that you review and act on beat thirty that nobody reads.
- Define each one: what is counted, how, by whom, how often, and what level triggers action.
- Balance them. Mix participation, fixing, checking, and leadership, so that a number cannot be improved by gaming one of them.
- Beware of targets on reporting. Rewarding low injury counts teaches people to hide injuries, so pair any outcome target with leading indicators and check for under-reporting.
The Trap of Small Numbers
Injury counts at a single site are small, and small counts are very noisy. A plant with six recordable cases could easily have had three or ten in a similar year, with no change in how safe it is. If you react to every move in the rate, you will reward luck, punish bad luck, and miss real changes.
- Treat a case count as a Poisson count. A count of cases in a given exposure has about as much random spread as the square root of the count.
- Put an interval on the rate. An exact 95% confidence interval for a count of 6 runs from about 2.2 to 13.1 cases, so the true rate could be well over twice, or well under half, what you observed.
- Compare with a test. A difference between two periods is only a signal if it is larger than chance would produce. A conditional binomial test compares two rates.
- Chart it over time. A u chart of monthly rates shows whether any month is out of the normal range, and separates special causes from the normal noise.
- Lean on leading indicators and severity. They have much higher counts, so they give a signal sooner.
The method is the same as the one used for defect counts. See the Stat Dojo pages on distributions (including the Poisson), confidence intervals, and control chart theory.
Showing Results Honestly
- Plot over time, with the average and limits, not as a table of this month against last month.
- Show the interval on any rate that you compare, or say how few cases are behind it.
- Do not rank sites by rate when their hours differ greatly, because small sites will swing to both ends of the list.
- Separate the types of case. One serious injury and ten minor ones are not the same.
- Keep serious injury and fatality (SIF) potential in view, because the events with the most potential to kill are not always the ones that are counted as the most serious.
Worked Example: A Year of Data at Riverside Plant
Riverside Plant, an illustrative 140-employee metal fabrication and assembly plant, worked 262,000 hours last year. It recorded 6 recordable cases, of which 2 involved days away, restricted work, or transfer. One was a lost-time case with 12 days lost. It also recorded 23 first-aid cases and 47 near misses. All figures are illustrative.
| Measure | Calculation | Result |
|---|---|---|
| TRIR | 6 × 200,000 ÷ 262,000 | 4.58 (95% interval 1.7 to 10.0) |
| DART rate | 2 × 200,000 ÷ 262,000 | 1.53 |
| LTIR | 1 × 200,000 ÷ 262,000 | 0.76 |
| LTIFR | 1 × 1,000,000 ÷ 262,000 | 3.82 |
| Severity rate | 12 × 200,000 ÷ 262,000 | 9.16 days per 100 employees |
Is it better than last year? The year before, the plant had 9 cases in 255,000 hours, a rate of 7.06. The rate fell by 35%, which looks like good news. But a conditional exact test of the two rates gives p = 0.45, and the intervals overlap widely. With so few cases, a fall this size is quite consistent with chance. The honest conclusion is “encouraging, not proven”.
Compared with the benchmark? Against an illustrative industry rate of 3.5, the plant would expect 4.6 cases for its hours, and saw 6. A Poisson test gives p = 0.48. The plant is not distinguishable from the benchmark.
Enter your own figures in the Safety KPI Calculator for the rates, exact intervals, comparison, and monthly u chart, or use the Safety KPI Dashboard workbook.
Building the Dashboard
- Choose the set. A few lagging rates and five to eight leading indicators, each with a written definition.
- Assign owners to each measure, and decide what level triggers action.
- Collect consistently. Same definitions, same hours source, same cut-off date every month.
- Display it where the team can see it, as trends with the average and limits.
- Review monthly in the safety committee. For each measure out of range, decide an action and an owner.
- Review the set yearly. Retire indicators that no longer help, and add new ones when risks change.
Common Mistakes
Only Counting Injuries
A rare, late, noisy measure is the only one watched.
Reacting to Noise
Celebrating a good month and investigating a bad one with no regard to chance.
Rewarding Low Numbers
Targets and bonuses on injury rates encourage hiding injuries.
Too Many Indicators
Dozens of measures that no one reviews or acts on.
Changing Definitions
The calculation changes, so trends cannot be trusted.
No Response
Indicators are published, and nothing happens when they go off track.
Self-Assessment Questions
- Do we balance leading and lagging indicators, with a written definition for each?
- Do we show uncertainty when we compare rates based on small counts?
- Do our targets encourage reporting, or could they encourage hiding?
- Does every indicator have an owner and a defined response?
- Do we review the set at least once a year?
Safety KPIs: Leading and Lagging Indicators: Frequently Asked Questions
What is the difference between leading and lagging indicators?
Lagging indicators measure outcomes after the fact, such as recordable injuries, lost-time cases, and days lost. Leading indicators measure the inputs and activities that prevent harm, such as hazards reported and closed, inspections completed, training done, and audit scores. Use both: lagging indicators show results, and leading indicators let you steer.
How is TRIR calculated?
TRIR is the number of OSHA-recordable cases multiplied by 200,000 and divided by the hours worked. The 200,000 is the hours that 100 full-time employees work in a year, so the result is the number of cases per 100 full-time employees. LTIFR uses a base of 1,000,000 hours and counts lost-time injuries only.
Is a lower injury rate always better?
Not necessarily. A low rate may reflect good safety, good luck, or under-reporting. With small counts, year-to-year changes are mostly chance. Look at intervals, trends, near-miss reporting, and leading indicators. Be careful with targets and rewards on injury rates, which can encourage people to hide injuries.
How do I know if a change in the injury rate is real?
Compare it with the random variation you would expect. For counts of cases, treat them as Poisson counts, put an exact confidence interval on each rate, and use a test of two rates, or chart monthly rates on a u chart. A difference inside the interval or limits is consistent with chance.
How many safety KPIs should we track?
A small number: a few lagging rates and five to eight leading indicators that you review and act on. Define each one in writing, give it an owner and a trigger level, and review the set at least once a year.
Sources and Further Reading
- US Occupational Safety and Health Administration, 29 CFR Part 1904, Recording and Reporting Occupational Injuries and Illnesses; OSHA Injury and Illness Recordkeeping materials.
- ISO 45001:2018, clause 9.1 on monitoring, measurement, analysis, and performance evaluation (check current edition).
- ANSI/ASSP Z16.1, Records and Measurement of Occupational Safety and Health Performance.
- US Bureau of Labor Statistics, Survey of Occupational Injuries and Illnesses, incidence rate definitions.
- Wheeler, D. J., Understanding Variation, SPC Press; Montgomery, D. C., Introduction to Statistical Quality Control, on count data and u charts.
- UK Health and Safety Executive, Developing process safety indicators (HSG254), on leading and lagging indicators.