Human reliability is usually investigated either from a technical perspective, in terms of probability models, or from a general psychological perspective using information processing or decision models. Both approaches view human actions at the level of the individual. Based on a re-analysis of 100 accidents at sea (Wagenaar & Groeneweg, 1987), it is shown that focusing solely on the individual, or an aggregation of several individuals, has severe shortcomings. The statistical analysis reveals that the individuals' contributions are not mutually independent; rather, they seem to be affected by social interactions manifested in the two-person case. These results may be explained by theories of social influence.
1999 Academic Press
1. Reliability and error
''Human reliability'' is usually defined as the probability that a person appropriately completes a task, under certain conditions, within a given period of time (cf. Carnino, Nicolet & Wanner, 1990). The term ''error'' is used if the task is not successfully completed, or completion is inadequate or incorrect (cf. Timpe, 1 992, p. 257).
The concepts of reliability and error include two aspects: utility and damage. ΒΊtility refers to the positive value associated with reaching a goal. The damage aspect is not simply a diminution of utility due to the incomplete or incorrect goal attainment but is, rather, an additional negative value resulting from the occurrence of unpredicted accidents. This damage aspect is particularly important for human-machine interactions in modern high technology, such as in chemical or nuclear technical plants, as well as in the areas of air, water and street traffic. Also its relevance is growing steadily as manual control of machines is increasingly replaced by computer-based automated control systems without direct feedback to the human operator.
The damage aspect is usually investigated either from a technical viewpoint in terms of probability models or from a general psychological perspective in terms of information processing and decision models. Both approaches view human actions at the level of the individual. This paper shows that focusing only on the individual, or an aggregation of several individuals, has severe shortcomings. A new statistical analysis of 100 accidents at sea (Wagenaar & Groeneweg, 1987) reveals that the individuals' contributions are not mutually independent. Rather, they seem to be affected by social interactions manifesting in the two-person case.