No theory forbids me to say "Ah!" or "Ugh!", but it forbids me the bogus theorization of my "Ah!" and "Ugh!" - the value judgments. - Theodor Julius Geiger (1960)

Accidents and Non-random error propagation

"The impact and effects of accidents are essentially not random in nature. Errors generate chains of structured events within the sociotechnical system in which they occur, as they interact with the ordered social properties of organizations and social groups and with the ordered technical properties of physical arrangements."

Many people tend to view accidents as random occurrences, but this perspective may simply mask our lack of understanding. Defining randomness, generating randomness, and identifying when random conditions occur are all challenging tasks. Nonrandom, structuring elements influence accidents, particularly organizational structures and their interaction with technical arrangements.
Conventional risk assessment techniques, like Probabilistic Risk Assessment, often rely on assumptions of randomness to separate deterministic and random aspects of accident propagation. This approach helps simplify causation analysis and enables the use of probabilistic models. Randomness assumptions break down when common cause failures or system interactions occur, rendering the categorization of phenomena as random or deterministic incorrect or inadequate.
Near Taunton, England, a fire developed in a railway sleeping car while it was in motion, killing 12 of the passengers. A series of errors, combined with the sociotechnical system's structure, led to a tragic accident resulting in passenger deaths. The sociotechnical system that existed before the accident had been in place for many years, and the staff operating it were not aware of the risks this posed. The sleeping car, where the fire occurred, was a self-contained technical entity with specific boundaries that shaped events within it. These boundaries included the task boundary for the attendant, the boundaries for passengers within their sleeping compartments, and the technical boundary of the carriage. The presence of locked doors further reinforced these boundaries, but the focus was on preventing intrusion rather than facilitating escape.
Accidents typically result from multiple starting points or errors. These errors can be related to training staff, the design of the sleeping car, economic decisions, and regulatory shortcomings, among others. The consequences of errors introduced into the system are not random but constrained by the existing system's features. Unintended events highlight aspects of the preexisting system that they don’t destroy. These phases, from error to system intervention, extend from the incubation period into the onset stage of a large-scale accident. It’s important to recognize nonrandom influences of human and organizational factors on accidents.

The Taunton sleeping car fire serves as an example of how various elements with partial common mode characteristics can overlap, leading to specific constraining factors. Researchers need to seek regularities in seemingly unstructured events related to accidents and need to move away from the assumption that accidents are purely random occurrences.
Turner, B.A. (1989), Accidents and Nonrandom Error Propagation, in: Risk Analysis, Vol. 9, No. 4, 1989, pp. 437-444.
Picture: a sleeping car like the one in the accident at Taunton - Wikipedia