Abstract
The paper discusses the use of injury severity distribution statistics as a proxy for the severity distribution of injury and non-injury incident statistics as described by Heinrich's iceberg. It examines those distributions that allow the estimation of the probability of more serious incidents than have already occurred. The organization involved has been anonymized.
A total of ±24,000 incident reports, over eight years in a high-hazard resource industry, were analyzed, of which ±12,000 had more than trivial incident severity. Of the selected reports the severities were assessed using expert knowledge, in combination with interviews with involved parties where possible to ensure assignment accuracy. The distributions over severities were compared both across the whole organization and for individual sites. After collation the distribution was applied to the current year's results, which had not been used to construct the distributions, to test the predictive efficacy of the model.
The incident distributions studied follow a power law, which can be described by a Pareto distribution, where the tail function describes the higher consequence levels. In this case study the general function followed the famous 80-20 rule. The distributions found, both within the sites and in the operation as a whole, were stable over the first seven years and predicted the eighth year. The stability of the distributions strongly suggests that these can be used to project into the future, thus allowing them to be used together with average cost profiles to project incident costs into the future for safety investment purposes.
The Power law distribution points to two factors:
The reduced probability of failure of consecutive barriers, with recovery controls having increasing weight
Increased management commitment and increased statutory requirements as incident potential increases
In addition the stability of the incident distribution across the years suggests that the distribution is a function of the safety culture.
This paper addresses and refines Heinrich's assertions about incident distribution, demonstrating with a large N size analysis that actual incident distribution is a power law function. In addition, it shows that this allows projection into the future.
