Estimating perceptual distributions for endpoint stimuli

Abstract

Rather than to assume that all stimuli on a continuum give rise to perceptual distributions that are Gaussian and equal, it is sometimes desirable to determine these distributions more directly, for example when the stimuli are speech sounds. This can be done by means of (absolute) identification or (non‐numerical) magnitude estimation. A problem with such methods is that the apparent distributions obtained with them become skewed as the stimuli approach either end of the continuum. An attempt was made to recover the underlying distributions by extending the response range by three stimulus steps beyond the continuum endpoints, both for intensity and for stop consonants. The resulting distributions were fitted by means of Gaussian functions, so that a direct measure of d’ in these tasks was obtained. It is assumed that in tasks other than identification or magnitude estimation, the widths of the distributions change, but that their means are unaffected.