ABSTRACT
Classification experiments, in which several stimuli are partitioned by the observer into a smaller number of categories, can be used to study perceptual independence versus interaction and to measure attention. Uncertainty about which element of a stimulus subset is to be presented can reduce performance compared to a corresponding fixed-discrimination condition. Stimuli that differ perceptually in more than one way can be represented as distributions in a multidimensional space. Sensitivity measures, such as d′, are distances in such a space, and multiple experimental conditions can allow the geometric arrangement of the distributions to be determined. Many multidimensional tasks are susceptible to either integration or independent-observation decision strategies. These can sometimes be distinguished by predicted accuracy, but methods such as the Garner paradigm examine more detailed aspects of the data. Selective and divided attention are usually intrinsically more difficult than the corresponding baseline tasks. The loss due to attention depends on whether the dimensions on which the stimuli vary are independent or interacting.
