Normal Accidents

Charles Perrow, a sociologist, wrote this after serving on the inquiry into the Three Mile Island accident, and it argues that some catastrophic failures are structural properties of systems rather than the results of error. His two variables are interactive complexity and tight coupling.

A system is interactively complex when components can affect each other in ways designers did not anticipate and operators cannot see, so that multiple small failures combine into a sequence nobody predicted. A system is tightly coupled when processes are time-dependent, sequences are invariant, and there is no slack that would allow recovery. Systems with both properties, which Perrow says include nuclear power, chemical plants, nuclear weapons and some aircraft and marine operations, will produce what he calls system accidents no matter how well they are managed. Adding safety devices increases complexity and can make things worse, which is why he treats redundancy as an ambiguous remedy.

He works through cases across many industries: Three Mile Island in detail, petrochemical plants, aircraft and air traffic control, marine collisions, dams, mines, space missions and nuclear weapons command and control. His policy conclusion is that some technologies should be abandoned because their accident potential cannot be engineered away, a position that made the book controversial. The main counterargument comes from the high reliability organisations school, which studied aircraft carriers and air traffic control and concluded that organisations can achieve near-error-free operation in complex tightly coupled systems. That debate remains open. The framework has been applied extensively to financial systems since 2008.