Not causal chains, but interactions and adaptations

Summary

I’ve been a bit of an anti-root-cause-analysis (RCA) tear lately. On LinkedIn, health and safety expert Vincent Theobald-Vega left a thoughtful, detailed comment on my last post. In his comment, he noted that RCA done poorly leads to bad results, and he pointed me to what he described as a good guide to using the RCA approach: Investigating accidents and incidents. This is a free book published by the UK Health and Safety Executive. However, after reading this guide, my perception of RCA has not improved. I still believe that RCA is based on a fundamentally incorrect model of complex systems failure. To clarify my thinking, I’ve tried to explain where I see its flaws in this post. A quick note on terminology: while the guide uses the term accident, I’m going to use the term incident instead, to remain consistent with the usage in the software domain. The guide uses incident to refer to near misses. Some content in the guide that I appreciated While I disagree with RCA as described in the guide, I wanted to start by pointing out areas of agreement I had with the guide. Not just a single cause The guide does note that there are multiple causes involved in incidents. It notes that adverse events have many causes (p6), and it also mentions that Very often, a number of chance occurrences and coincidences combine to create the circumstances in which an adverse event can happen. All these factors should be recorded here in chronological order, if possible. (p10). While I disagree with the causal language, I do at least appreciate that it points out there are multiple factors. Examine how work is really done The guide does talk about examining the work and the context under which it takes place. Under “information and insights gained from an investigation”, one of the bullet points is A true snapshot of what really happens and how work is really done (p7). Under the “Gathering detailed information: How and what? section, the guide asks What activities were being carried out...