The Yes-No Experiment:
In dealing with other people, “bias” is the tendency to respond on some basis other than merit, showing a degree of favoritism. In a correspondence experiment, response bias measures the participant's tilt toward one response or the other.
The sensitivity measure d′ depends on stimulus parameters, but is untainted by response bias: To a good approximation, it remains constant in the face of changes in response popularity. We now adopt the complementary perspective, seeking an index of response bias that is uncolored by sensitivity. Conceptually, d′ corresponds to a fixed aspect of the observer's decision space, the difference between the means of underlying distributions; a measure of bias should also reflect an appropriate characteristic of the perceptual representation. How can we assign a value to the participant's preference for one of the two responses?
Consider again the face-recognition experiment of chapter 1, in which viewers discriminated Old from New faces. Suppose the investigator now repeats the experiment, this time hypnotizing the participants in an effort to improve their memory, and obtains the following results from a representative observer: