Using a signal detection safety model to simulate managerial expectations and supervisory feedback

The present work studied a basic discrimination task that underlies many safety problems (such as hazard identification and supervisory inspection). The experimental task required decision-makers to discriminate between two classes of stimuli representing hazardous and secure cues in the environment. Errors in which decision-makers failed to identify hazardous cues were defined as risky errors. The payoff for risky errors was probabilistic. Some risky errors went unnoticed (representing lucky outcome of near accidents). Other risky errors resulted in penalty (representing damage incurred by an accident). The discrimination task was modeled utilizing Signal Detection Theory. A Cutoff Reinforcement Learning model provided predictions for choice behavior. Two controlled experiments are reported here. Experiment 1 manipulated the payoff for unnoticed risky errors. This manipulation was suggested as an analogy to different levels of the conflict between safety and productivity. Experiment 2 tested the effect of outcome feedback and cognitive feedback for risky errors. This manipulation was suggested as an analogy to supervisory feedback. The results of the two experiments showed that the simplified experimental conditions were sufficient to induce risky behavior. The findings also suggested some ways to reduce risk taking. The Cutoff Reinforcement Learning model's predictions captured most of the findings. The paper discusses the theoretical implications of the findings and their relevance for safety research.