In one portion of a fixed‐interval‐observation experiment, observers indicated their certainty that a 500‐cps signal had been presented in a thermal‐noise background by positioning a slider after each interval. In a second portion of this experiment, the observers responded by making binary decisions. Slider positions were treated as typical confidence ratings; the conditional probability of a given rating or of one indicating greater confidence, given signal plus noise, was plotted against the probability of these ratings, given noise alone. Functions produced in this manner, for a rating scale divided into thirty‐six positions, were fit closely by the receiver‐operating characteristics (ROC's) of the theory of signal detectability. A psychophysical model using two straight‐line segments did not provide a good approximation to these data. Values of d′ were generally lower for the rating procedure than for the binary‐decision procedure. The use of a large number of rating categories did not result in large increases information transmitted by the observers, but it did provide the desired fine resolution of the ROC's.

This content is only available via PDF.
You do not currently have access to this content.