Human Performance in Low-Signal-Probability Tasks

Abstract

An extension of the theory of signal detection (TSD) to psychophysical tasks involving low-probability signals and free-response data is developed and evaluated. Emphasis is placed on tasks for which the observer is asynchronous[long dash]that is, the observer cannot perform optimally by making independent decisions on nonoverlapping intervals of time. A mathematical model of asynchronous observation in a class of temporally unstructured tasks with Neyman-Pearson solutions for optimal fixed-response rate is used to describe detection performance by human observers. Data from an experiment show (1) a conservative fixed-response rate, (2) a constant hit rate, and (3) interresponse distributions for false alarms with a general exponential shape showing periodic modes. Detection efficiency in the temporally unstructured task was approximately 1/10 of alerted-detection efficiency for two observers and 1/2 of alerted-detection efficiency for a third observer. Points on the obtained receiver-operating-characteristic (ROC) curve are fit better by an inefficient asynchronous observer than by synchronous power-law observers. A post hoc analysis of the effect of training showed an effect for distribution of responses in time but showed no effect of an improvement in memory for the signal. It is concluded that highly trained observers detecting important signals show constant efficiency over observation periods of 30-45 min. The TSD psychophysical model of asynchronous observation seems to be an adequate description of human performance in the low-probability free-response task used in this study.