MEASUREMENT OF SIMILARITY IN SOCIAL PROFILES HAVING FEW COMPONENTS.

This article focuses on the measurement of similarity in social profiles. The comparison of states with respect to a series of characteristics simultaneously, when interest lies in the similarity between profiles of the series values rather than between composite indexes representative of each series, introduces the need for other measures of similarity. By similarity in profile is meant here similarity in the relative sizes of all values in a series. The method which proved most satisfactory in measuring similarity in profiles was based on the well-known fact that the process of squaring values in a series increases the relative weight of the larger values. Accordingly, two profiles with a given total of differences between corresponding component values in their series would be considered most similar when this total difference was distributed more evenly among the individual component value differences rather than occurring as a lump difference between corresponding values of only two of the components. Distributed evenly, the total difference would be broken down into smaller differences, the sum of whose squares would be less than the square of the few lumped differences would be if the total difference were concentrated in two or three pairs of components instead of being distributed in small amounts among all the component pairs. The chi square measure of dissimilarity between an observed and expected distribution seemed to offer a possibly appropriate inverse measure of similarity.