The Identification of Industrial Complexes

The objective of this paper is to reassess the value of principal components analysis as a technique for identifying regional industrial complexes, taking explicit account of the different kinds of relationships that are possible among industries. The empirical evidence offered here tends to confirm recent criticism of principal components analysis as a means for detecting vertical relationships, but it demonstrates the usefulness of this method with complementary relationships. In this study two principal components analyses were performed on a single regional input-output table for the state of Washington. Although initial inspection of component loadings did not always disclose industrial complexes, it was found that tracing the network of interactions, as indicated by the profiles of purchases and sales, revealed many critical links in a variety of industrial subsystems. A comparison of the results of this study with other analyses of Washington data suggests that the technique used here provides more information on the industrial structure than did the multivariate analysis conducted by Czamanski (1971) and the graph-theoretic approach of Campbell (1974). Also explored here are the links between principal components analysis and the general issue of aggregation in input-output analysis.