Paired comparisons and estimates of failure likelihood

Abstract For quantitative realibility assessment, there is a need for data. Determined efforts have been made to collect such data from a vast range of technological operations. Although there has been success with inanimate components — the hardware — little has been achieved on the human side. There is virtually no form of ‘human factors’ database despite many attempts. Reasons are put forward to explain this situation and it is argued that there will be no progress until an advance is made in defining a workable structure for such a database. As an interim measure it should be possible to use individual subjective experiences in a formalized way. The methods of paired comparisons is proposed for this and the procedure is described. The means by which a consensus is obtained, judged for quality, and used to provide absolute estimates is detailed and a trial run is reported. A well known railway accident is used as a test example. Possible applications of the paired comparison technique in other fields are suggested.