ASIDT: Adaptive Signal Detection Testing

Abstract

Two-alternative forced choice (2AFC) methods which are routinely used in psychophysics provide criterion-free sensitivity measures but at a high price: 2AFC procedures generally need 2 – 3 times the number of trials than a simple yes/no design to achieve the same level of threshold variability. Similarly, signal detection methods provide measures for criterion and sensitivity, but they require predetermined signal levels for the presented stimuli and also need a large number of trials.

ASIDT (adaptive signal detection testing) combines rating responses with an adaptive stimulus placement strategy according to the accelerated stochastic approximation method (see Treutwein, 1995 Vision Research 35 2503 – 2522). More specifically, during the course of the experiment the rating responses are collapsed into two groups (yes/no). The border between these two groups is set at the current median of the rating categories actually used by the subject. With the collapsed yes/no responses the stimuli are placed according to accelerated stochastic approximation. This placement strategy is based on very simple rules and distributes the stimulus values asymptotically according to a normal distribution around the median. After the experiment, the rating responses can be used to compute a nonparametric and criterion-free sensitivity measure, the area under the empirical ROC, which is equivalent to the percentage correct in a 2AFC experiment. The usage of ASIDT is demonstrated by psychophysical data.