Variation recordable accident rate

As an example of the importance of knowledge of variation, let us consider two similar chemical plants, each with 50 employees working 2,000 hours each during the year. In 1994, the Riverton plant had two recordable accidents, for a recordable accident rate of 4. The Hillton plant had four recordables, for a rate of 8. Obviously, the Riverton plant has a better safety process, does it not The answer is that it may, but the evidence just provided is very weak. 

Differences in recording create a problem when companies seek to find benchmarks. When benchmarking on safety, the accident rate is often used to determine who has the best safety performance. These benchmarking efforts seldom consider the wide variation in aiteria discussed above. 

To test the validity of published accident ratios and triangles, I did the research resulting in the publication of a paper titled Injury Ratios. The most important conclusion drawn from that research is that variations on the inherent risk levels in industries and businesses—as indicated by the substantial differences in OSHA incidence recordable rates and the percent of incidents that result in lost workday cases—are so great that it is impossible to develop meaningful injury ratios which are universally applicable. 

The upper control limit is three standard deviations from the mean, or 3 + (3 X 2.74), or 11.22. This is based on one year of data. If the observation period shrinks to a month, then the units of exposure go from. 4 to. 033, and the upper control limit rises to 31.60. This looks high, but how many incidents can happen in a month for the measure to remain within control limits The answer is that one accident in a month will yield a rate of 30.3 for that month, just inside the control limits. At this site, if we had two recordables in the month, we would exceed control limits. The point here is that comparing either the recordable rates of small sites or rates of larger sites computed over short periods is not very useful. If you measure two plants with 40 employees for over a year and one has a rate of 8 and the other of 0, there is no statistical basis for concluding they are different. This difference is within the normal random variation of measure being used. The incident rate of 8 took only 3 recordable events to reach. 

---