[Psychometric function of stereo disparity in normal persons]

The accuracy and efficiency of threshold estimation depend on a priori knowledge of the shape of the psychometric function. Such knowledge is available for contrast detection and visual acuity, but not for disparity detection.

We studied the psychometric function for disparity detection in 26 visually normal, untrained observers by measuring detection rates over a wide range of disparities. In a two-alternative forced-choice (2AFC) task the subject had to decide whether the right or left of two vertical bars, differing in depth, was closer to them. An S-shaped Weibull function was fitted to each subjects's data using the maximum-likelihood procedure.

In 21 of the 26 subjects the psychometric function rose with increasing disparities up to a level of about 100%. Five of the 26 subjects only reached hit rates clearly below 100% (66% to 82%), even at disparities well above the threshold. The mean slpe of the S-shaped psychometric function of disparity detection was much lower (1.2 Weibull units) than that known for visual acuity and contrast detection (about 3.0 Weibull units). The slope differed widely among subjects, but was not related to stereo acuity.

For accurate and efficient estimation of stereo acuity, the shallow slope of the psychometric function requires more than 100 single trials around the threshold if a 2AFC procedure is used. In addition, several disparities well above the threshold must be presented in order to detect subjects who do not reach a hit rate of 100% at any disparity. Otherwise, stereo acuity would be underestimated in these subjects.