Accidents and Nonrandom Error Propagation

The use of random elements in accident analyses is practical in that it avoids the need to enumerate all possible failure paths and allows the application of probability analyses to some elements of a complex system. Sometimes, however, the concept of randomness is used only as a residual category to label unexplained features of accident analysis. When common cause failures or system interactions are involved, such an approach can be misleading. Analyses focusing upon nonrandom elements can thus be important in understanding failures of both technical and organizational systems, and some of the problems of such an approach are explored. A detailed analysis of a fatal fire in a railway sleeping car at Taunton, England, in 1978, demonstrates how initial errors can interact with an existing sociotechnical structure to produce new orderly patterns as an accident develops. A simple model to understand this nonrandom error propagation requires a description of the initial system structure in social and technical terms, specifying features such as the task and the sentient boundaries of subsystems. When an error or a set of errors is introduced into this system, the consequent system interventions are structured by the constraints of the preexisting system which they do not destroy. Rather than offering randomness as an account of such phenomena, the analysis encourages a search for regularities in the apparently unstructured events surrounding large‐scale accidents or system failures.