signal detection theory
A psychophysical theory of the detectability of stimuli developed in 1952–4 by a number of US researchers led by John A(rthur) Swets (born 1928), based on the assumptions that there is a normal probability distribution N of activation of the sensory system by noise (2) generated externally or by internal random neural activity in the absence of any signal, that there is a different normal probability distribution (SN) of activation by signal plus noise, and that the difference between the means (or equivalently the separation between the peaks) of the SN and N distributions, divided by the standard deviation of the N distribution, is an index of detectability called d prime or d′ that can be estimated experimentally from the relative frequencies of correct detections of the signal when it is present (hits) and incorrect detections of the signal when it is absent (false alarms), the performance of an observer usually being depicted by means of a receiver operating characteristic (ROC). According to the theory, an observer sets a criterion level of sensory activation below which the signal is reported as absent and above which it is reported as present, but if the two probability distributions overlap, then stimuli below the criterion must include not only instances of noise (correct rejections) but also instances of signal (misses), and stimuli above the threshold must include not only instances of signal (hits) but also instances of noise (false alarms). Also called sensory decision theory. See also ideal observer, likelihood ratio (2), memory operating characteristic, psychometric function, signal-to-noise ratio, two-alternative forced-choice task. SDT abbrev.