QUEST: a Bayesian adaptive psychometric method.

A psychometric function describes the relation between some physical measure of a stimulus and the probability of a particular psychophysical response. The physical measure is usually stimulus strength, and the response is "yes" (in a yes/no experiment) or "correct" (in a forced-choice experiment). More generally, there are several possible responses, each with its own psychometric function (e.g., in a rating scale experiment). Psychometric procedures are ways of testing the observer so as to gain information about the psychometric function. The advantages of adaptive procedures, which make use of previous responses to guide further testing, have been discussed by numerous authors (Cocnsweet, 1962; Levitt, 1971; Taylor & Creelman, 1967; Wetherill & Levitt, 1965). Because trials are more effective when· they are judiciously placed, an adaptive procedure will be more efficient the more it makes intelligent use of available information. This information is of two sorts: that derived from previous trials (data), and that drawn from the memory of the experimenter, published reports, and so on (prior knowledge). Prior knowledge may be further divided into information about the shape of the psychometric function and knowledge about threshold in the particular condition under study. Several recent methods make efficient use of the data and of knowledge about the shape of the psychometric function by calculating after each trial a maximum likelihood estimate of threshold (Hall, 1968; Klein, 1981; Pavel, 1981; Pentland, 1980; Watson & Pelli, 1979). Our procedure, which we have dubbed QUEST, also makes use of prior knowledge about the location of threshold.