MODELS FOR ACCIDENT INVESTIGATION

This report represents the second half of a two part literature review project commissioned by the Alberta Occupational Health and Safety Division under the direction of a project team consisting of Judith Evans, Lynn Hewitt and John McDermott. The project was initiated by Dr. Herbert Buchwald, Managing Director, who recognized the need for critically examining the various approaches to understanding accidents. Such an analysis represents an important prerequisite to refining the division’s strategies for collecting, interpreting and using accident data in the service of effective accident prevention programs.

Dr. Harvey’s first literature review (Theories of Accident Causation December. 1984) traced the historical progression ofaccident causation theories and modelsfrom the original single factor theories to the more recent systems theory approach.

In this report, Dr. Harvey discusses a cross-section ofaccident investigation modelsin terms of their ability to satisfy five evaluative criteria. These criteria, derived from the occupational health and safety literature, represent major purposes of accident investigation. On the basis of his analysis, Dr. Harvey recommends the “best’ current approach. In addition, he discusses factors which limit the usefulness of data from accident investigations

The project team gratefully acknowledges the advice and assistance of colleagues from Occupational Health Services and Work Site Services both in developing the terms of reference for this project and in providing commentary on draft versions of the reports.

Four general models for accident investigation are reviewed and evaluated with respect to five purposes which comprise the goals of accident investigation. The investigation models chosen for review are: The Heinrich model, with its focus on unsafe acts and unsafe conditions; the epidemiological model, which considers the three broad factors of host, agent, and environment; fault tree models (specifically, the MORT system); and the multilinear events sequencing model recently proposed by Benner.

The evaluative criteria developed for this review consist of five purposes served by accident investigations. These purposes arelegal(does the model consider safety code violations?),descriptive(can use of the model provide a detailed description of the accident?),causal(can accident causes be determined by the model?),prevention(does use of the model lead to recommendations for improved safety?), andresearch(will use of the model provide reliable and comprehensive data useful for accident research?).

The Heinrich Model. This model seems to be most clearly associated with the legal purpose of investigation, with its emphasis upon unsafe acts and conditions. The major criticism of. the model is its potential for introducing bias into the investigation procedure, since the investigator’s attention is focussed not upon the facts per se, but upon the unsafe aspects of the accident event only. The identification of unsafe aspects after the accident has occurred is deceptively easy (the hindsight bias), and can lead to conclusions that

Epidemiology. Epidemiology is a methodology applied to accident events that seeks to identify the factors associated with the host, agent, and environment that are correlated with various categories of accidents. This model avoids the potential for bias noted above, and can potentially lead to an investigation report that describes the accident event completely. However, epidemiology is deficient in two respects; first, it needs guidance from a theory of accidents, and second, it needs an efficient and theoretically based scheme for the classification of accidents.

Fault Tree Models. The general fault tree approach to accident investigation is attractive because it advocates a description of all the necessary and sufficient conditions for an accident within the work system in question. However, a specific adaptation of the model (MORT) has failed to attend closely to the accident event itself, and instead focuses the investigation largely toward management oversights. The model is also criticized for facilitating bias and because it could easily lead to broad recommendations for prevention (eg., more training; more supervision) rather than specific ones.

Multilinear Events Sequencing. This model, proposed by Benner, is similar in many respects to a general fault tree model, but unlike MORT, it does limit its focus to the accident event itself. Benner advocates close attention to the sequence of events leading up to the accident, with special status given to the temporal relations between events. This investigative model is very compatible with the systems theory approach

to accident causation. The model encourages a complete description of the accident event and successfully avoids introducing investigator bias. The multilinear events sequencing model is judged to be the best investigative model currently available.

(1) It is important to consider what the content of an on—site investigation should be. An investigation report could consist of facts alone, or could include inferences, conclusions, and recommendations as well. (2) It is important that the potential for the biased gathering of facts be recognized and minimized so far as possible. (3) There is a need, especially if research purposes are to he achieved, for a relatively simple yet theoretically guided system for the classification of accidents.

`FOREWORD`	`i`
`EXECUTIVE SUMMARY`	`iii`
`MODELS FOR ACCIDENT INVESTIGATION`	`1`
`A. Introduction`	`2`
`B. Purpose of accident investigation`	`4`
`C. Accident investigation models` `Heinrich’s domino model` `Epidemiology` `Fault tree models (e.g., MORT)` `Multilinear events sequencing` `Benner’s (1983) evaluation criteria`	`8` `8` `12` `17` `21` `25`
`D. Issues relevant to accident investigation` `Gathering facts` `Avoiding bias` `Regulations` `Investigator conclusions` `Accident classification` `E. Summary` `References`	`27` `27` `27` `28` `29` `29` `31` `32`