Quest: A Bayesian adaptive psychometric method

Abstract

An adaptive psychometric procedure that places each trial at the current most probable Bayesian estimate of threshold is described. The procedure takes advantage of the common finding that the human psychometric function is invariant in form when expressed as a function of log intensity. The procedure is simple, fast, and efficient, and may be easily implemented on any computer. The QUEST procedure was developed at the Kenneth Craik Laboratory of Cambridge University. A.B.W. was supported by an NIH postdoctoral fellowship, F32 EYO'219. D.G.P. was supported by a British Ministry of Defence grant, "Spatial noise spectra and target detection/recognition," to F. W. Campbell. We thank, Stanley Klein, Daniel Kersten, and Joseph Hall for helpful comments.A.B.W.'5 mailing addressis:

Adaptive psychophysical procedures

“…[...] When psychometric functions are measured in experiments in which the subject is asked to report wether or not the stimulus was seen (or heard), lbr example, the dependent variable is the hit rate. Several authors (Naehmias, 1981;Watson & Pelli, 1983) have suggested that these data may be fit with logistic or Weibull functions setting y [the guessing rate] equal to the false alarm rate. The problem with this approach is that the false alarm rate is not constant: there is a different false alarm rate tbr each point on the psychometric function.…”

“…Therefore, for a desired confidence level of y, solving the following equations leads to a condition where the best estimate of the threshold value lies in the interval (0t, 0u) with probability y In most cases equation (23) In a different approach, Watson and Pelli (1983) and Harvey (1986), referring back to Wiiks (1962), advised a likelihood-ratio test for determining a confidence interval. The arguments given by these authors for applying the test are not fully developed and three questions arise: first, the test does not account for the sequential nature of the estimation problem.…”

A deafening flash! Visual interference of auditory signal detection

“…Participants heard a single click if their response was correct, and two clicks if incorrect. On each trial, target intensity was chosen depending on responses to the previous trials, using the Quest algorithm (Watson & Pelli, 1983). Each block contained 40 trials, and there were 6 blocks in total.…”

Going, going, gone: Localizing abrupt offsets of moving objects.

“…Trial misalignments were centered around the observer's current point of subjective alignment (PSA), as calculated from the QUEST threshold estimation algorithm (Watson & Pelli, 1983). QUEST estimates the current most likely PSA from the performance on all previous trials.…”