An alternative analysis of geometric models of proximity data, based on a feature-matching model, leads to the coincidence hypothesis that the dissimilarity between objects that differ on 2 separable dimensions is larger than predicted from their unidimensional differences on the basis of the triangle inequality and segmental additivity. A series of studies of 2-dimensional stimuli with separable attributes (including house plants, parallelograms, schematic faces, and histograms), using judgments of similarity and dissimilarity, classification, inference, and recognition errors, all support the coincidence hypothesis. The size of the effect is determined by the separability of the stimuli, the transparency of the dimensional structure, and the discriminability of the levels within each dimension. Applications of the coincidence effect to inductive inference are investigated, and its relations to selective attention and spatial density are discussed. It is concluded that the triangle inequality and segmental additivity cannot be jointly satisfied for separable attributes. The implications of this result for multidimensional scaling are discussed. (57 ref) (PsycINFO Database Record (c) 2016 APA, all rights reserved)
このアブストラクトは、Tversky & Gati (1982)による「coincidence仮説」について述べたものであり、従来の幾何学的モデルが前提とする三角不等式や部分的加法性(1次元ごとの差を足し合わせれば2次元間の差が説明できるという考え)を、人間の類似・非類似判断が必ずしも満たさないことを示唆している。