An inter-method comparison of four Human Reliability Assessment models.

This paper presents a comparison of four common Human Reliability Assessment (HRA) models through a scoping literature review and sensitivity analysis. The scoping literature review identified 72 relevant studies which formed the basis of the comparison. Studies reported the four selected models have similarities in terms of the sector of origin, applied sectors, output calculation, and a lack of clear guidelines on Performance Influencing Factors (PIFs) selection and risk level allocation. The studied models have differences in the number and type of PIF inputs and Human Error Probability (HEP) calculation procedures. The One Factor At a Time (OFAT) and "combined" sensitivity analysis were conducted to examine the HRA models' responses to systematic risk level changes when each of 8 matching PIFs were systematically set to "high" and then "low" levels individually and simultaneously. The OFAT analysis showed coefficients of variation (CV) in HEP varying from 9% for skills/training up to 94% for work procedure when the PIFs are assigned to a "low" risk level individually. The combined analysis showed the median HEP value close to 97% and 1% when PIFs are assigned to" high" and "low" risk levels respectively. Although the selected HRA models were reported to be validated in high-risk domains there was no study found that validated these models in low-risk domains such as manual order picking, or manual assembly lines. The HRA models examined here are disconnected from specific system design elements which can inhibit design improvement efforts. The study outcome suggests the need for clear guidelines for PIFs selection and risk level allocation. Future research should address both the connection of error assessment to the design of the system and the features of new HRA models that affect its reliability and validity in a variety of industrial contexts.