Operationalizing normal accident theory for safety-related computer systems

Computer-related accidents have caused injuries and fatalities in mining as well as other industries. Normal accident theory (NAT) explains that some accidents are inevitable because of system complexity. NAT is a classic argument in organizational sociology although it has been criticized as having imprecise definitions and lacking criteria for quantifying complexity. These limitations are addressed by a unique approach that recasts this organizational theory into an engineering-based methodology to quantify NAT complexities of computer-based systems. In this approach complexity is categorized as external or internal. External complexity is defined by the external behavior of a system, and is quantified by these dependent variables: system predictability, observability, and usability. Dependent variable data contain the perceptions of 32 subjects running simulations of a system. The system?s internal complexity is characterized by modeling system-level requirements with the software cost reduction (SCR) formal method. Model attributes are quantified using 15 graph-theoretical metrics?the independent variables. Five of 15 metrics are correlated with the dependent variables as evidenced by structure correlations exceeding 0.25, with standard errors <0.10 and a 95% confidence interval. The results also show that the system predictability, observability, and usability decreased as NAT complexities increased. This research takes a step forward in operationalizing NAT for computerized systems. The research benefits mining and other industries as well.