Detectability Measures in Vigilance: Comment on a Paper by Wiener, Poock, and Steele

Summary.-Results of a study of rime sharing and vigilance reported by Wiener, et al. (1964) have been reanalysed from the viewpoint of signal detection theory. The reported decline In probability of detecting a signal is shown to be due entirely to a progressive change in the observers' response criteria and not to a change in the detectability of the signal. In contrast to the original report rhat tlme sharing had a non-significant effect on the probability of detecting a signal, the effect on the detectability of the signal is large. The importance of considering detectability rather than detection measures in vigilance studies is stressed. Wiener, Poock, and Steele (1964) studied the monitoring behaviour of observers required simultaneously to do simple mental arithmetic. As is common in studies of vigilance, they reported their findings in terms of the percentage of signals detected during each quarter of the run. In terms of this measure, the time-sharing required by simultaneous monitoring and mental arithmetic had no significant effect on the level of detection in the vigilance task. The experimental group and two control groups all showed the usual decline in detection probability as the run progressed. Reports of vigilance studies that use detection probability as the primary measure usually do not present data on what Wiener, et al. call "commissive errors." The fact that such data were given by Wiener, et al. permits reanalysis of their results in terms of statistical decision theory (e.g., Swets, Tanner, & Birdsall, 1961 ) . The present note reports this reanalysis, which changes the apparent implications of the results from the study, and stresses the importance of such analytical procedures in the evaluation of vigilance data.