Some useful characteristics of performance models

This paper examines the demands placed upon models of human cognitive decision processes in application to Probabilistic Risk Assessment. Successful models, for this purpose, should, 1) be based on proven or plausible psychological knowledge, e.g., Rasmussen's mental schematic, 2) incorporate opportunities for slips, 3) take account of the recursive nature, in time, of corrections to mistaken actions, and 4) depend on the crew's predominant mental states that accompany such recursions. The latter is equivalent to an explicit coupling between input and output of Rasmussen's mental schematic. A family of such models is proposed with observable rate processes mediating the (conscious) mental states involved. It is expected that the cumulative probability distributions corresponding to the individual rate processes can be identified with probability-time correlations of the HCR Human Cognitive Reliability type discussed elsewhere in this session. The functional forms of the conditional rates are intuitively shown to have simple characteristics that lead to a strongly recursive stochastic process with significant predictive capability. Models of the type proposed have few parts and form a representation that is intentionally far short of a fully transparent exposition of the mental process in order to avoid making impossible demands on data