A response time theory of separability and integrality in speeded classification (Q1343130)

A fundamental problem in perception is to determine how different stimulus components combine and interact in perceptual processing. Although a variety of perceptual interactions have been studied, it is especially popular to characterize perceptual dimensions according to whether they are separable or integral. A pair of dimensions are said to be separable if it is easy to attend to one and ignore the other. Consequently, the perception of a value on one dimension is unaffected by variation along the other. Prototypical separable dimensions are hue and shape. On the other hand, with integral dimensions, it is extremely difficult, if not impossible, to attend to one dimension and ignore the other. Thus, the perception of one dimension is affected when the level of the other is varied. Prototypical integral dimensions are brightness and saturation. A variety of methods for testing between separability and integrality have been suggested, but two of the most popular involve speeded classification tasks in which the subject is told to name the level of one stimulus component. A pair of dimensions are assumed to be separable if response time (RT) is unimproved when the stimulus dimensions are correlated and if no RT interference is found when the dimensions are varied orthogonally. If RT decreases when the dimensions are correlated and an RT interference is found with orthogonal variation, then the dimensions are assumed to be integral. Unfortunately despite their intuitive appeal, these speeded classification tests are operational and have no rigorous theoretical justification. \textit{F. G. Ashby} and \textit{J. T. Townsend} [Psychol. Rev. 93, 154-179 (1986)]\ proposed rigorous theoretical definitions of separability and integrality and they developed several simple methods for testing between the two. One limitation of these recent tests, however, is that they are all stated in terms of response probability and therefore they ignore response time (RT). This article develops an RT theory of separability and integrality. Such a theory is important, first, because it will allow to examine the validity of the speeded classification tests and, second, because it will unite two large and growing literatures. The development begins, in the second section, with an overview of the speeded classification tests of separability and integrality. The third section briefly reviews general recognition theory and the definitions of separability and integrality that it suggests. The fourth section introduces the RT-distance hypothesis and considers its empirical validity. In the fifth section the two most popular speeded classification tests of separability are examined critically through analytical investigation and Monte Carlo simulation. Tables 5 and 6 summarize our theoretical findings. Finally, we close with some conclusions and general comments.