Probing the tenets of attention theories via geometric visual illusions.

Abstract

Although criticism of theoretical treatments of attention has accelerated recently, new formulations have not emerged. In this work, a list of attributes of attention that require theoretical treatment is presented. Attention theories and results in the empirical literature are categorized in terms of their contribution to the understanding of each of the attributes. In addition, the literature on visual illusions is discussed. Several theorists have proposed attentional explanations for visual illusions, including one account suggesting that attention increases the perceived size of objects. Four experiments tested this hypothesis. Three of the four experiments support the hypothesis that attention makes lines look longer, but do not support an attentional explanation for visual illusions. Experiment 3, in which a different attentional manipulation was used, produced no effects of attention on line-length judgments. The results of three of the four experiments demonstrate an unusual effect of attention on information processing. In these experiments, the information conveyed in both high- and low-attention conditions is the same, but the response in one attention condition is a recoding of the response in the other. A new stimulus-selection method for psychophysical experiments, the Smith-Cook Up-Down (SCUD) method, is developed. The SCUD method implements a formalization of the principle of placing trials to maximize the total information gathered about one or more parameters of the psychometric size-judgment function. Computer simulations show that the method is effective and relatively robust against poor prior assumptions about the parameters of the underlying psychometric function. It is also flexible in terms of the underlying distribution parameters about which it can maximize information. In the terms of psychophysical methodology, the research goal may emphasize the constant error (difference between the points of subjective and objective equality), the variable error (judgment consistency), or both. The new method and existing methods are discussed in terms of appropriateness for different experimental contexts. The SCUD method is successfully implemented in the fourth experiment in order to obtain better estimates of variability in line-length judgments while preserving the quality of estimates of the point of subjective equality (PSE).