Incident investigation techniques in all high-hazard industries still concentrate primarily on discovering the immediate causes, such as the technical failures and the more frequent human shortcomings that led directly to the incident. Issues that are identified at these superficial levels of investigation tend to be local and hard to relate to deeper underlying causes that may well be company-wide. Answers to more abstract questions, about organizational, cultural and regulatory failures, provide much more useful information that organizations can use to implement preventative programs with far-reaching benefits, rather than symptomatic remedies of underlying problems that may do little to prevent subsequent incidents.

Deeper analyses involve identifying causes that are much harder, or impossible, to place in strict logical sequences. The approach proposed here is to make a distinction between deterministic immediate causes and probabilistic underlying causes. The approach allows the aggregation of multiple causes, taking different points of view, rather than being based upon an essentially linear approach such as event trees. The different categories chosen can be aligned with the organization’s requirements for learning and improvement, as opposed to the primary aim of merely supplying an accurate description of what happened.

Many assignments to causal categories are hard to make unequivocal; often no amount of extra definition helps to make such assignments unique. The ‘statistical’ approach described in this paper relies upon the notion that lower accuracy of assignment to a cause category in any single accident analysis can be compensated by averaging over larger numbers of incidents. Investigators often remain close to the event because the evidence is more concrete and less ambiguous at the level of the immediate causes, resulting in a tendency to reinforce the idea that front-line operators are the primary causes of accidents – if only because no evidence is offered that there was a deeper-lying set of causes within the organization.