Fitting blockmodels to data

Whether the important conceptual advantages of blockmodelling can be practically utilized depends crucially upon the development of a satisfactory method for fitting blockmodes to data. Existing procedures suffer from several important limitations, principal among which is the lack of consensus on a measure of the goodness of fit of a blockmodel to the data it represents. In this paper I present a measure of goodness of fit, and an algorithm for finding blockmodels with maximal fit to data, that together render blockmodeling mathematically equivalent to regression analysis. To facilitate more detailed analyses, I also present measures of the degree to which individuals and their relations deviate from the overall pattern of the blockmodel of the sharpness with which each cluster is defined, and of the distances between clusters. The algorithm and measures are applied to data concerning international trade in the Western Hemisphere in two different decades.