Dependence in human reliability analysis and in operational events

This paper deals with dependencies in accident sequences with multiple Human Failure Events (HFEs). When these are present, the probability of some HFEs may require modification, given knowledge of preceding failures. HFE dependence can have significant impact on Probabilistic Safety Assessment results, because the joint probability of the HFEs may increase by orders of magnitude. State-of-the-art dependence analysis in Human Reliability Analysis (HRA) relies on simplistic approaches with a largely subjective basis. This paper addresses empirical evidence of HRA dependencies from operational events, contributing to ongoing research to strengthen the technical and empirical basis of dependence analysis. This first-of-a-kind analysis of six events identifies three groups of coupling factors: work practice-related, task-related, and knowledge-related. Task-related factors have the strongest influence on dependence because they manifest as specific performance drivers for the multiple tasks. The evaluation of a widely adopted dependence analysis Decision Tree (DT) suggests that the method is practical and adequate for most analysis cases. The analysis identifies some cases for which the DT assessment of independence does not seem appropriate based on the event narrative, e.g. when the DT assessment yields independence (zero dependence) because of an intervening success. As a result, a recommendation is formulated to analysts to question the plausibility of the assessment of independence. The evaluation methodology applied to the DT may also be used to assess more recent modelling developments, also addressing the important need for a larger set of events for broader coverage of situations of interest for HRA.