Decision strategies in the ABX (matching-to-sample) psychophysical task

The ABX psychophysical task is very useful when the dimensions along which the stimulus varies cannot be readily described. However, there are a few problems associated with its use. First, the task has often been used with slight procedural variations, which hinders the extraction of a bias-free index of performance. Second, the most common index of sensitivity derived from this task is the ubiquitous proportion correct. Third, some researchers undervalue the importance that the decision strategy adopted by the observer has on estimates of sensitivity. We describe a standard methodology for the ABX task and outline models of performance based on two decision strategies. These models can be used to extract the detection-theoretic index of performance, d', from ABX data. The results of three ABX discrimination experiments, in which circle size, number value, and tone amplitude are used, are reported. Analysis of the decision spaces derived from observer responses indicates that these observers adopted a suboptimal difference decision strategy, even after they had received considerable practice with feedback. This suggests that, given no prior knowledge of the nature of the decision strategy adopted by an observer, the difference model of ABX discrimination could be a good choice when observer performance is estimated.