For about the last decade plant performance and safety assessors have had a pressing need for an applicable human reliability data-base. Although many assessors would like to have fully quantified data, most would settle for any credible information, even if it were only a qualitative level. Fortunately the human factors literature now extends back in time some forty years or so. Data collected during the countless studies reported exhibit massive absolute variability but possess very much narrower relative variability. So narrow is the relative variability that it is possible (to a first approximation) to describe major performance modifying effects in a very simple, consistent mathematical terms, so that reliable predictions can be made out the probable extent of an error probability change, given that the situation can be modelled with some clarity. This paper describes the development of a technique designed to assist engineers not only to assess the likelihood and impact of human unreliability but to apply human factors technology to optimise overall system design. It explores the identity and magnitude of error-producing factors and presents a battery of defensive measures which can be applied to combat their effects. The method has been applied in a variety of industrial situations and assessed by several enterprises as an aid to cost-effective design and operational decision making. The current indications are that the method produces fairly consistent predictions which assessors have found helpful in both an absolute and relative sense. (A)
Abstract