Human Error: Causality and the Confusion of Normative and Descriptive Accounts of Human Performance

Abstract

That human errors can cause accidents is a core assumption of classical safety management systems, which aims to reduce the occurrence and consequences of human errors. This paper will present two interconnected lines of arguments that are contrary to the classical assumptions of causality and focus on human error. First it identifies challenges to the assumed causal connection between human error and accidents by discussing selection bias, lack of statistical contrasts between accidents and non-accidents with respect to human error and accidents, as well as the challenge of causal inferences for N=1 cases. The second line of arguments involves the conflation of normative and descriptive accounts of performance and how this hinders a structured and scientific approach to system performance. The paper ends with three principles that require a pre-hoc identification of behaviours that deviate from performance standards. This approach hopefully allows for practitioner acceptance of performance standards, a just evaluation of performance, as well as enabling proactive safety management by requiring that errors be identified prior to performance.