Accident versus near miss causation: a critical review of the literature, an empirical test in the UK railway domain, and their implications for other sectors
Introduction
Decision making about investing in safety improvements is usually based upon the relative importance of root causes in accidents and failures. However, such decisions can only be reached reliably by referring to statistics from large databases. As accidents themselves are (fortunately) too few in number to aid such decision making processes the use of near misses to dramatically increase the number of data in databases is one way to counteract this problem. This use of near misses as causal predictors of later, more serious, accidents is based upon the assumption that these near misses and accidents have the same relative causal patterns (the so-called common cause hypothesis). Such a causal relationship is also a vital argument to motivate employees to contribute to near miss reporting schemes on a voluntary basis.
When the common cause hypothesis is discussed there is inevitably discussion of the ratio data studies performed by Heinrich [1], Bird [2] and Salminen et al. [3]. This very starting point is central to the way thinking about the common cause hypothesis has become focussed. Heinrich’s original triangle was not intended to convince the reader of the commonality of causes between different accident outcomes, but to illustrate the fact that prevention need not wait until an accident occurred, and that prevention should not only be aimed at the most severe consequences but also to events at the lower levels of triangle. In this endeavour, Heinrich was successful. The ratio triangles or icebergs are used profusely in industry today. However, Heinrich did not base the common cause hypothesis upon the ratio relationship between major accidents, minor accidents and no injury accidents, although the proposed ratio relationship seemed (to him) to substantiate the idea of a common causal pathway. Today, the common cause hypothesis has come to imply a ratio relationship of consequences (and not of causes). How did then this confusion arise, and where did the common cause hypothesis spring from?.
The validity (or refuting) of the common cause hypothesis has major implications for accident prevention and analysis. If the different levels of severity really do have completely different patterns of causes, then industry has been concentrating on levels of severity (near misses, small failures) which may have little impact on the frequency of accidents which cause the greatest injuries. On the other hand, if common causal pathways can be demonstrated then a concerted effort is required to collect appropriate data (i.e. via voluntary near miss reporting schemes) and to ensure that causal analysis techniques become more widespread.
Section snippets
The confounding of the ratio model versus the common cause model
How did then these two separate models become so interdependent? Heinrich [1] discusses the relevance of the triangle model as providing some evidence for similarity of cause of frequency and severity (i.e. reduction of events at the bottom of the triangle should lead to a reduction in the number of events at the top of the triangle). However, the common cause hypothesis itself emerged from data analysis rather then from a deduction based on the ratio model, although Heinrich did not further
Literature review
The table below summarises the limited literature in the area.Reference Type of data used Confounded view of the iceberg model? Conclusions [8] Frequency data for major and minor accidents in manufacturing and construction industries Yes. Confuses ratio of minor to major incidents as being the same as causal mechanisms of major and minor incidents As ratios not in agreement with original iceberg theory, concluded that different causal mechanisms present between major and minor accidents [9] Frequency
Conclusions from the literature review
It is apparent that frequency, severity and causal mechanism have become inextricably linked (Heinrich, op cit; Petersen, op cit). It appears that researchers have not differentiated between the causes of severity and frequency and the causes of accidents and incidents. Thus, if a ratio is established and the data follow the pattern of the ratio found by Heinrich or Bird, it is suggested that the similar cause hypothesis is validated. Where the ratio is invalidated i.e. severe incidents do not
A proper test defined
There are three possible ways in which the common cause hypothesis can be tested:
- •
comparing the actual occurrence of causal codes based on a dichotomy of causal codes being either present or absent;
- •
comparing the actual frequency of causal codes contributing to the different incident outcomes; and
- •
comparing the relative proportions of causal codes contributing to the various incident outcomes.
Testing the common cause hypothesis using causal codes as either present or absent is the weakest method
Data collection
The three different investigation methods used by the UK railway industry and providing the data for this study are described below.
CIRAS analysis
The data were analysed according to the University of Strathclyde CIRAS human factors model which is hierarchical (see [17] for a full description of the system). According to this model individual causal codes are subsumed under one of four top-level categories: ‘technical’, ‘proximal’, ‘intermediate’ and ‘distal’ which we called the ‘macro’ codes. These macro codes each comprise an exclusive set of individual causal codes, which we termed ‘micro’ codes. Thus the common cause hypothesis can be
Inter-rater reliability
Inter-rater reliability is a vital (and often neglected) part of any analysis system. Data analysed via the CIRAS system are subject to periodic inter-rater reliability trials. Index of concordance was above 80% for each trial. To ensure the data used in this study were also reliably coded two independent raters (experienced in using the coding scheme) coded a total of 14 incidents from various classes of event used in this study. This resulted in an index of concordance of 78.4%.
Results and discussion
Fig. 1 shows how the four macro causal codes are distributed over the three levels of severity. A Chi-square test for proportions showed non-significant differences.
At the level of macro codes (i.e. the superordinate categories of technical, proximal, intermediate and distal) no significant differences were found in the proportion of causal codes between the three severity outcomes (injury, damage and near miss). However, despite the fact that these results are supportive of the common cause
References (17)
- et al.
Fatal and non-fatal occupational accidents: identical versus differential causation
Safety Sci.
(1992) - et al.
Industry differences in accident causation
Safety Sci.
(1996) - H.W. Heinrich, Industrial Accident Prevention, McGraw-Hill, New York,...
- F.E. Bird, Damage Control, Insurance Company of North America, Philadelphia,...
- D.C. Petersen, Techniques of Safety Management, McGraw-Hill, New York,...
- D.C. Petersen, Techniques of Safety Management, second ed., McGraw-Hill, New York,...
The accident evolution and barrier function (AEB) model applied to incident analysis in the processing industries
Risk Anal.
(1991)- T.W. van der Schaaf, L. Kanse, Errors and error recovery, in: P.F. Elzer, R.H. Kluwe, B. Boussoffara, (Eds.), Human...
Cited by (102)
International benchmarking of railroad safety data systems and performance – a cross-continental case study
2023, Journal of Rail Transport Planning and ManagementExploring spatial associations between near-miss and police-reported crashes: The Heinrich's law in traffic safety
2023, Transportation Research Interdisciplinary PerspectivesCorrelation between events with different safety significance in nuclear power plants
2022, Nuclear Engineering and TechnologyArtificial intelligence-aided railroad trespassing detection and data analytics: Methodology and a case study
2022, Accident Analysis and PreventionPromoting farming sustainability: The effects of age, training, history of accidents and social-psychological variables on the adoption of on-farm safety behaviors
2022, Journal of Safety ResearchCitation Excerpt :Near misses occur more frequently and are smaller in scale compared to severe accidents, and each major accident is usually preceded by several near misses (Phimister, Oktem, Kleindorfer, & Kunreuther, 2000). Near misses have been investigated in different sectors: road and rail traffic (Wright & Schaaf, 2004), plant engineering (Uth & Wiese, 2004), building safety (Wu, Gibb &Li, 2010), home safety (Marsh & Kendrick, 2000) and health care systems (Lyons, Arora & Farnan, 2016) and, less frequently, agriculture (Caffaro et al., 2017). Wright and Schaaf (2004) showed that near misses and accidents substantially share the same determinants, confirming the idea that near misses may be considered a proxy of being exposed to the risk of suffering a more severe accident.
Applying AcciMap to test the common cause hypothesis using aviation near misses
2020, Applied Ergonomics