Testing Senders' Visual Occlusion Model: Do Operators (Drivers) Really Predict During Visual Occlusion?

Abstract

The present study tests the hypothesis that humans are capable of predicting the state of a system during visual occlusion, an assumption often made in models of sampling behaviour, but seldom tested. In 1967, John Senders introduced the visual occlusion paradigm to evaluate attentional demand of tasks such as automobile driving. Despite multiple studies employing this paradigm, the concept of operators actually being able to resolve uncertainty during occlusion by predicting system output has remained unvalidated. A self-paced visual occlusion monitoring task was contrived, involving a randomly rotating basin with a ball at the bottom. Participants were required to detect critical events (ball falling off the edge) while looking only as often as subjectively deemed necessary. Assuming the need to resolve uncertainty imposed by the random rotations, we examined relations between occlusion durations and system states preceding occlusion, for different glance durations, to infer whether predicting may have taken place. Results suggested that glance requests were consistent with the use of simple first order predictions. This pertained not only for longer (300 and 500ms) glances, but even for 100ms glances whenever critical events were imminent. The presumption that human operators are capable, under certain circumstances, of predicting system state in the absence of visual information appears feasible; however, glance duration plays an important role. By providing support for some of its basic premises, the use of Senders' visual occlusion paradigm as a potential tool for evaluating human monitoring performance has been strengthened.