Predictive models of safety based on audit findings: Part 2: Measurement of model validity

• Published in: Applied ergonomics Vol. 44; no. 4; pp. 659 - 666
• Main Authors: Hsiao, Yu-Lin, Drury, Colin, Wu, Changxu, Paquet, Victor
• Format: Journal Article
• Language: English
• Published: Kidlington Elsevier Ltd 01-07-2013; Elsevier; Elsevier Science Ltd
• Subjects: Accidents, Occupational - statistics & numerical data; Applied physiology; Audits; Aviation - standards; Aviation maintenance; Biological and medical sciences; Ergonomics; Ergonomics. Work place. Occupational physiology; HFACS-MA; Human error; Human performance; Human physiology applied to population studies and life conditions. Human ecophysiology; Humans; Medical sciences; Models, Theoretical; Neural network; Neural Networks (Computer); Occupational Health; Organizational Culture; Poisson Distribution; Predictive control; Predictive Value of Tests; Reproducibility of Results; Risk Management; Safety; Safety Management - methods; Safety prediction; Series & special reports; Human error; Aviation maintenance; Neural network; HFACS-MA; Safety prediction; Audit; Validity; Prediction; Models; Maintenance; Safety; Predictive factor
• ISSN: 0003-6870; 1872-9126
• DOI: 10.1016/j.apergo.2013.01.003

Part 1 of this study sequence developed a human factors/ergonomics (HF/E) based classification system (termed HFACS-MA) for safety audit findings and proved its measurement reliability. In Part 2, we used the human error categories of HFACS-MA as predictors of future safety performance. Audit records and monthly safety incident reports from two airlines submitted to their regulatory authority were available for analysis, covering over 6.5 years. Two participants derived consensus results of HF/E errors from the audit reports using HFACS-MA. We adopted Neural Network and Poisson regression methods to establish nonlinear and linear prediction models respectively. These models were tested for the validity of prediction of the safety data, and only Neural Network method resulted in substantially significant predictive ability for each airline. Alternative predictions from counting of audit findings and from time sequence of safety data produced some significant results, but of much smaller magnitude than HFACS-MA. The use of HF/E analysis of audit findings provided proactive predictors of future safety performance in the aviation maintenance field.
► In order to prove the causality between human error and safety performance. ► We used human error categories to predict future incident rate via neural network. ► Prediction validity was examined with substantial performance (r around 0.6). ► The result significantly proved the causality between human error and flight safety. ► This is the first validation of safety audit reports against accepted safety measure.