Toward a generalization of signal detection theory to \(n\)-person games: The example of two-person safety problem (Q1916535)

The present paper represents a first step in an attempt to generalize the prescriptive and descriptive predictions of signal detection theory (SDT) to situations in which the strategic aspects of the situation are important. The major goal of this paper is to demonstrate that a generalization of SDT to \(n\)-person games can make this useful theory even more useful. No attempt is made here to provide a complete generalization as the task is too complex. Rather, the paper presents a formal description of a two-person signal detection problem, and then focuses on one example: a two-person safety problem -- a situation in which accidents can be avoided if at least one of two individuals detects the risky situation and takes the necessary safety measures. The paper proceeds as follows: The first section summarizes the predictions of traditional SDT. It is then demonstrated that when a second observer (we use the words observer, detector, decision maker, operator, and player interchangeably) is added, the detection task can be abstracted as a two-person game. The nature of the predicted behavior is described using game theoretical concepts. The game theoretical model is then used to explore the example of a two-person safety dilemma. It is shown that, in equilibrium, observers who share the responsibility of detecting warning signals are less likely to report signals. This predicted trend (criterion shift) is particularly strong for the less capable (less sensitive) observer. An experiment is then described that examines the game theoretical predictions. The results exhibit the general trends predicted by the game theoretical model. The psychological process by which behavior could converge toward the game theoretical equilibrium is discussed. It is demonstrated that a simple learning model that assumes that observers update their behavior in accordance with the law of effect is sufficient to reproduce the results. Thus, the suggested generalization of SDT can be based on the same psychological process that underlies classical SDT. Theoretical and practical implications of the results are discussed.