Human pattern recognition: individually different strategies in analyzing complex stimuli.

Multidimensional scaling experiments were performed with "size" and "brightness" as parameters. Subjects (ss) had to rate the dissimilarity of pairs of discs differing in size or in brightness only, or in both parameters. It was then attempted to find a rule, by which the dissimilarity judgements of two-dimensional differences can be predicted from judgements given for the one-dimensional components. Minkowski-metrics were chosen as models, since for most of the ss the prerequisites for applying these metrics seem to be fulfilled. A simple procedure is described by which the one metric out of all Minkowski-metrics can be found which gives the best description for the data of a subject (Fig. 4). Although the task and the parameters used were the same for all ss, individually best metrics differed considerably, indicating individually different modes of perceptual analysis: Most ss adhered, roughly, to either the City-block or the Euclidean metric (cf. Fig. 10). (Only City-block metric was to be expected, since "size" and "brightness" are generally thought to be "separable" parameters.) In addition, there were three ss with metrics clearly different from both the City-block and the Euclidean metric (best exponents around 1.4). In view of those substantial differences between individuals one has to be cautious against the practice of averaging data from different ss.