VIGILANCE CONSIDERED AS A STATISTICAL DECISION

A number of experiments are described in which untrained subjects kept watch during long periods for inconspicuous signals. When they reported such a signal, they had to indicate their degree of confidence that the report was correct. By analysing the accuracy of reports in each category of confidence, it can be shown that they are quite inconsistent with the concept that stimuli are either definitely perceived or definitely fail to be detected, with no intermediate category. The results are more consistent with a model of perception as the outcome of a statistical decision made with more or less caution upon evidence contaminated by randomness. Analysing the empirical data according to this model, neither the duration of a watch nor the presence of loud noise during visual experiments affect the degree of randomness in the evidence used for decision. Both factors do, however, affect the degree of caution with which the decision is made, and this may explain the failures of perception previously found in experiments upon prolonged watch and upon noise. Changes during the watch appear primarily, however, when the decision is being made cautiously (with few false detections) while the changes with noise appear rather as a lessening of the difference between cautious and risky behaviour. This divergence between the effects of noise and of prolonged work may explain the fact that experimental conditions which reveal effects of noise have not always in the past been those which are sensitive to effects of prolonged work.