Relation between signal detectability theory and the traditional procedures for measuring sensory thresholds: Estimating d' from results given by the method of constant simuli.

Abstract

The theory of detectability assumes that the central effect of a stimulus varies because of physical and neural noise; consequently, the detection of a requires a central statistical decision procedure. Similar assumptions have been made by psychophysicists to explain the results of traditional threshold measurement procedures. The interrelations between detectability and threshold measures are discussed in relation to psychophysica l statistical decision theory, and is shown that (a) the false positive rate should be related to the Crozier ratio C = Al/o-ai, and (b) should be possible to use responses given in the method of constant stimuli to predict the value of d' that will be assigned to a given stimulus by a detectability procedure. Evidence supporting both predictions is reported, and the relation between threshold measures and personality tests is discussed. The theory of detectability has had considerable success in accounting for a number of features of sensory performance (Swets, 1964; Tanner & Swets, 1954). It rests on two main assumptions. The first is that a presented for detection by a human observer has a central neural effect which varies from presentation to presentation because of the presence of physical and neural noise. If no stimulus is presented, a central effect which varies from time to time will be produced by the alone. The second assumption is that the resulting detection problem is solved as statistical decision theory would require: a criterion is selected and any given input receives the response Yes, I detect it if exceeds the criterion but not if fails to do so. Figure 1 shows this account applied to a common procedure used in studying detection. On each trial either or signal + noise is presented to the subject. The is a background level of stimulation which can be measured on a physical dimension, 7. Its magnitude, /#, may be zero or