An experimental analysis of human monitoring behavior in multivariable failure detection tasks

Abstract

This report deals with the results of an experimental program designed to validate a model of the human observer and decision maker formulated in terms of linear estimation and classical sequential decision theory. The experiment comprised a variety of monitoring tasks in which the subjects had to detect and diagnose the occurrence of ramp failures which were superimposed upon zero mean stochastic Gaussian processes. The independent variables were signal bandwidth, number of displays, correlation among displays, failure magnitude, failure type (single vs correlated failures), and prior knowledge about failure type. The dependent variables were detection times, display deviations at the moments of response and falso alarm rates. In addition, the experiment included measurements of heart rate, skin resistance and eye point of regard. A good overall agreement was found- between the experimental results and the corresponding model predictions. Furthermore, an analysis of the display deviations indicated a constant detection strategy. In addition, the eye point of regard measurements rendered a useful insight into certain scanning characteristics. At last, the physiological measures appeared to be sensitive only to the number of displays involved, and prior knowledge about failure type.