A data-informed dependency assessment of human reliability

Dependency assessment is an aspect of human reliability analysis that identifies the causal relationship between two human events and quantifies the conditional probability of the successor event when two or more events exist in an accident sequence. Despite broad recognition of the impact of dependency on the overall system risk, many experts have been concerned that most current methods are rooted in the THERP method without a sufficient theoretical and empirical basis for dependency models. In this study, we propose a method that calculates the conditional failure probability of a successor event based on quantitative evidence of the dependency between two human events. Quantitative assessment is performed by evaluating six features and integrating the failure probabilities due to the features into the assessment based on an arithmetic equation. The estimates obtained from this empirical data analysis, a statistical function for time insufficiency, and a sequence alignment algorithm were employed to support the basis of the calculation with several assumptions. Two case studies are presented to show the feasibility of the study and the result differences between the proposed and existing methods. Since this study presents a new approach to dependency assessment, additional issues to be tackled regarding the assumptions and technical bases used are discussed with further research directions.