The Field Guide to Human Error Investigations
More on systems accidents
Of a piece with Charles Perrow’s Normal Accidents, Sidney Dekker’s book is compelling in rooting the cause of accidents in poor system design and unnecessary complexity, overlaying safety features and compliance measures which only make the problem worse — that is, at the door of management and not poor benighted subject matter experts who are expected to make sense of the Rube Goldberg machine that management expect them to operate.
There are two ways of looking at system accidents:
- It’s the meatware: Complex (and complicated) systems would be fine were it not for the meatware screwing things up. Human error is the main contributor to most system accidents, introduce unexpected failures into an essentially robust mechanism. Here, system design and management direction and oversight are otherwise effective strategies that are let down by unreliable human operators.
- It’s the system: Accidents are an inevitable by-product of operators doing the best they can within complex systems that contain unpredictable vulnerabilities, where risks shift and change over time and priorities are unclear, conflicting and variable. Here, human operators are let down by shortcomings in system design and conflicting management pressure.
Blame the meatware
Those investigating accidents are motivated in ways which will favour the “meatware” theory:
- Resource constraints: the wish for a simple narrative leading to a quick and inexpensive means of remediation
- Failure reaction: an instinctive reaction that since an accident has happened there must have been a failure
- Hindsight bias: the benefit of hindsight in that (i) the accident has happened (ii) the causal chain is now clear (thanks to the simple narrative) (iii) there has been a failure.
- Personal responsibility: In an ironic self-hack those who are paid to be safety practitioners assume great responsibility, with some pride, notwithstanding that they don’t have the power or authority (what Daniel Pink would call “mastery” and “autonomy”) to effectively discharge that responsibility. But if they too are motivated to fall on their sword, as a matter of pride, it is an easy conclusion for an accident investigator to draw.
- Political imperative: the “best” outcome for the organisation — and its management — is that there is no deep, systemic root cause requiring extensive recalibration of the system's fundamental architecture (much less criticism of management for its design or supervision), and that superficial action (rooting out/remediating “bad apples”) is all that is needed.
Recommendations thus tend to double down on the system design, further constraining employees from effectively managing situations and setting them up for blame should future accidents occur:
- Tightening procedures: Further removing autonomy from employees, obliging them to follow even more detailed instructions, rules and policies
- Introducing safety mechanisms: Adding more fail-safes, warning lights, system breakers and other mechanical second-order derivatives designed to eliminate meatware screw-ups but which actually further complicate the system and bury opportunities to directly observe its operation
- Downgrading employees: removing subject matter experts and replacing them with lower calibre (i.e., cheaper) employees with even less autonomy to follow the even more complicated rules and processes now introduced.
But blaming the meatware is to ignore history and condemn yourself to repeat it. Changing the make-up of your workforce won’t help if the basic conditions under which they are obliged to operate aren’t fixed. Simply adding more, increasingly detailed, policies — “codified over-reactions to situations that are unlikely to happen again” in Jason Fried’s elegant words — will only make the gap between theory and practice wider.
The work to rule as falsification of policy
There is an often-stated but still wildly optimistic idea that all policies are complied with. Not only are they not, but they are disregarded explicitly. All concerned understand that optimal — even basically acceptable performance requires turning a blind eye to the rules. There is no better example than the work-to-rule: a form of industrial action adopted by those who are, by regulation, not permitted to go out on strike. The work-to-rule involves, literally, insisting rigorously on complying with every aspect of every prescribed policy as a means of frustrating the commercial objectives of the organisation.
What does it mean? It means that if people don’t want to or cannot go on strike they say to one another: “let’s follow all the rules for a change!” Systems come to a grinding halt. Gridlock is the result. Follow the letter of the law, and the work will not get done. It is as good as, or better than, going on strike.
The vibe is: “Oh, I see, Mr. Employer, is that it? Are we being dicks about out employment relationship? Well, two can play at that game.”
Reacting to failure
Reactions to accidents tend:
- Retrospective: To be made with the benefit of hindsight and full knowledge of inputs and outputs, and with ample time to construct a narrative that neatly links events into a causal chain that was not at all clear to the actors at the time. This may be an impressive feat of imagination from a middle manager not normally known for their creative skills, but it is a feat of imagination.
- Proximal: To blame the meatware at the sharp end, closest to the accident — the operators, traders, negotiators — and not so much on those at the “blunt end” — the executive, its goals, target end-states, its strategic management of the process, how it balances risk and reward, what tools tools and equipment it provides, its rules and policies and the constraints and pressures it imposes on those subject matter experts to get the job done.
- Counterfactuals: To construct alternative sequences of events — where operators “zigged”, but could have “zagged” — which might have avoided the incident. “Forks in the road stand out so clearly to you, looking back. But when inside the tunnel, when looking forward and being pushed ahead by unfolding events, these forks were shrouded in the uncertainty and complexity of many possible options and demands; they were surrounded by time constraints and other pressures.”
- Judgmental: To explain failure by seeking failure: incorrect analyses, mistaken perceptions, misjudged actions. Again, hindsight is king. In each case, if you presented the operator with the facts as they were available to the investigator, in the same unpressurised environment, you might expect the “correct” outcome.
- Cause-consequence equivalence: The assumption that a bad outcome must have had equally bad causes and, seeing as management-mandated and properly governed processes are unlikely to have been the process of really bad governance —we bureaucratised the shit out of that, after all — therefore the malignant cause must be the fault of a bad apple somewhere.
- But bad outcomes are not necessarily caused by equally bad inputs: the Three Mile Island disaster was a concatenation of seemingly insignificant and benign, but unusual, events. Indeed, that is the conclusions of normal accidents theory: You can’t prevent unexpected non-linear interactions, and these can often quickly spiral out of control.