Detection of auditory stimuli : sensitivity and optimal performance

Abstract

The sensory functioning and decision processes of animals were investigated by training rats to detect a signal consisting of an Increment in the intensity of a random noise. The procedure was analogous to the yes-no method of human psychophysics in that one response was defined as correct and reinforced with brain stimulation if the signal was presented, and another response was correct and reinforced if the noise alone was presented. Incorrect responses produced periods of timeout. By varying the decibel-level of the signal relative to the noise (which was kept constant at 69 dB SPL), while holding the signal probability and the outcomes for correct and incorrect responses constant, bias functions were obtained showing how the animals* response biasses depended on the signal intensity when there was some constant bias in the signal probability or the payoff matrix. Conversely, by holding the decibel-level of the signal constant, while varying either the signal probability or the symmetry of the payoff matrix, receiver-operating-characteristics (ROC curves) were obtained, showing how the animals’ response biasses depended upon the bias of the signal probability or of the payoff matrix. When response biasses were small, the percentage of correct responses was related to the decibel-level of the signal between 3 and 10 dB by a cumulative normal probability function. A signal intensity of approximately 6 dB was required to maintain 75% correct responses. When response biasses varied, levels of detection were measured by d*, the bias-free index of sensitivity of detection theory. This index was related to the decibel-level of the signal between 0 and 20 dB by an ogival function. In addition, levels of detection remained constant when the absolute duration of timeout consequent upon incorrect responses was varied between 1 and 10 sec. Responses biasses varied systematically with changes in the signal probability between 0,1 and 0.9* A change as small as 0.05 in the signal probability was sufficient to produce different response biasses. Changes in the ratio of brain stimulations for correct responses between 1:3 and 3»1 also produced different response biasses. Biasses were virtually independent of changes in the relative duration of timeout for incorrect responses between 1:30 sec and 30:1 sec. In terms of statistical decision theory, the animals adopted response criteria under each signal probability so as to optimize the expected value of their responses, namely the percentage of correct responses, and hence the number of brain stimulations obtained. They did not match their response probabilities to the signal and noise probabilities at any signal intensity. Under each ratio of brain stimulations they compromised between optimizing the expected value (the overall number of brain stimulations) and optimizing the percentage of correct responses. Under each relative duration of timeout they did not respond so as to optimize the expected value (to minimize the time spent in timeout), but optimized the percentage of correct responses. The animals responded to optimize the probability of reinforcement in preference to optimizing the amount of reinforcement, which in turn, they optimized in preference to either the relative or the absolute rate of reinforcement in time. This result stands in contrast with those demonstrations from other procedures which show that relative rates or proportions of responses are determined by relative rates or amounts of reinforcement. The optimal response criteria exhibited by the rats when the signal probability was varied also stand in contrast to the findings that humans in similar psychophysical procedures tend to adopt sub-optimal criteria, and to match their probability of reporting the signal to the signal probability. The rats did not exhibit matching strategies either when the signal intensity exceeded 0 dB, or when the signal intensity was 0 dB, and the procedure was thus similar to that commonly used in probability-learning studies.