Attention and Interaction
In a classification design, a number of stimuli are sorted into a smaller or equal number of categories. When introducing this type of experiment in chapter 5, we restricted the discussion to sets of stimuli that differed on a single internal dimension, but we now abandon that limitation and examine paradigms in which the stimuli lead to representations that differ multidimensionally. Proceeding gently, we consider apparently simple problems in which just three or four stimuli must be classified into only two categories. This project turns out to be sufficiently challenging for one chapter.
This set of problems has both methodological and substantive applications. Methodologically, there is a set of discrimination paradigms that can be thought of as classification tasks. Recall that the comparison designs of chapter 7 always lead to a representation with only two distributions. As a result, although they can be modeled in two dimensions, they can also be analyzed by projecting the bivariate distributions onto a single axis and conducting a unidimensional calculation. As long as there are only two stimulus classes, and thus only two distributions, the projection strategy always works. This simplification cannot be made for classification paradigms, and in chapter 9 we use the tools developed here to analyze them.
Substantively, classification designs are extensively used to study two important topics: (a) independence versus interaction between two aspects of a stimulus, and (b) attention. The independence question was the first, historically, to which multidimensional detection theory was applied (Tanner, 1956), but the idea of independence turns out to be multifaceted. In chapter 6, we encountered the concept of perceptual independence—a characteristic of the representation of a single stimulus or stimulus class. An analogous concept applies to stimulus sets', this was Tanner's focus, and