Auditory Signal Detection and Amplification in a Neural Transmission Network
William J. McGill
Malvin C. Teich
A topic as large and as intricate as signal detection poses inevitable frustrations for both reader and writer. The general reader is entitled to clear, unambiguous prose. What he or she gets is an argument sometimes requiring word-by-word translation as though it were written in some long-lost, hieroglyphic script. The reader is often as baffled by what is omitted as by what is said.
The writer, conscious of all the limitations of research, tries to say what is true, but ends up with sentences that are either clear, yet subtly wrong, or so carefully hedged as to be barely decipherable. Such dilemmas, we say, are practically unavoidable. They ensure that the communication of knowledge in complex areas will always be painful. Each of us must do the utmost to minimize that pain. Accordingly, at the very beginning, we owe our readers a simple account of what we attempted, what we omitted, and why we chose as we did.
Detection theory has become a fixture of modern psychology because it provides useful quantitative tools and an easily understood vocabulary for describing relations that typically occur when a stimulus emerges in some distinctive way from a background. The ROC curve portrays expected tradeoffs between false alarms and missed signals when outcomes, stimulus probabilities, or decision criteria are varied. The detectability measure d′, obtained from the ROC curve, adds quantitative precision to the meaning of discriminability.
The idea that there are tradeoffs rather than fixed thresholds lies at the heart of signal detection theory. Moreover, the tradeoffs are not limited to