A variable sensitivity theory of signal detection.

Introduction This paper deals with an analysis of some simple detection experiments in terms of a theory that incorporates two separate but interdependent processes: an activation process and a decision process. The activation process specifies the relation between external stimulus events and hypothesized sensory states of the subject. The decision process specifies the subject's observable response in terms of his sensory state and information acquired during the course of an experiment. Both processes are dynamic. The activation process defines the subject's level of sensitivity to external stimuli, and we postulate that sensitivity may fluctuate (within certain limits) .from trial to trial as a function of past events. The decisipn process is similarly dynamic, for it may change from trial to trial as information accrues to the subject. The processes interact in that the momentary state of one process operates in a reciprocal fashion to determine·;the state of the other, As will be indicated later, most theories of signal detection view the subject's sensitivity level as fixed (or at most fluctuating ina strictly random fashion over time) and account for variations in his performance to a fixed intensity signal by postulating changes in the decision rule. In contr&st, for the present theory changes in performance to a fixed intensity sign&l may &rise in sever&l W&ys: (1) manipul&ting aspects of the experimental situations th&t affect the subject's sensitivity level but le&ve the decision process unchanged, (2) manipulating v&ri&bles th&t &ffect the decision process but le&ve the sensitivity level unch&nged, or (3) manipl1-lating p&r&meters that &ffect ch&nges in both processes. The theory th&t we present gener&tes predictions for &11 &spects of the subject's response protocol (me&n response pro-b&bilities, &ssoci&ted v&ri&nces, sequenti&l st&tistics such as &utocorrel&tion functions on both responses &nd stimuli, &nd so forth) &nd thereby permits & det&iled tre&tment of individu&l tri&l.,by_trial d&ta. Some predictions arep&r&meter free, but by &nd l&rge the predictions depend on estim&tes of p&rameters th&t describe the stimulus situ&tion &nd the hypothesized detection process. Some readers may feel th&t we have been too liber&l in postul&tingparametersj however, for most &pplic&tions, restrictions &re &ppropri&te that markedly reduce the number of ! parameters that need to be estimated. For example, predictions regarding receiver operating characteristic curves and certain first-order sequential phenomena may require that only two parameters be estimated. In contrast, autocorrelation predictions in complex detection experiments may requi.re that as m&ny as six parameters be estimated.-<3-The type of psychophysical …