[Dimensions of similarity of faces--a cross validation].

Using a series of three experiments, a cross-validation study on the dimensions of similarity of faces was carried out. In experiment 1,100 subjects sorted two sets of facial photographs (61 men, 62 women; age range 20-60) into groups of faces similar in appearance. In experiment 2, a group of 864 subjects rated a subset of 36 male and 36 female faces on a personality differential consisting of 90 scales and a physiognomic differential consisting of 29 scales. Each judge rated a single photograph on all scales. In experiment 3, the complementary subsets of 25 male and 26 female faces were rated by a third group of 48 subjects. Each judge rated one set of photographs on eight different scales from the personality sphere. Nonmetric multidimensional scaling (ALSCAL) was carried out on the similarity judgements obtained in experiment 1. Using multiple regression procedures, the ratings from experiment 2 were regressed on the coordinates of the similarity spaces. About two-thirds of the personality scales and nearly half of the physiognomic scales could be fitted into the three-dimensional similarity space. By using factor values in place of scale values, the Big Five personality factors as well as physical attractiveness and age could be fitted into the three-dimensional space. The regression coefficients obtained in experiment 2 were used to predict the scale values of the complementary subsets of stimulus persons in experiment 3. With the exception of one scale ("unlogical--logical"), the cross-validation was extraordinarily successful.