Spectral mapping, a personal and historical account of an adventure in multivariate data analysis

摘要

Spectral mapping is a factorial method of multivariate data analysis that was developed around 1975 in the pharmacological research laboratory of Janssen Pharmaceutica. It has been applied in diverse fields of biology and economics where tabulated data are obtained which are defined on ratio scales. The approach has provided a few instances in which useful information was obtained that would be hard to get from bivariate analysis or from "eyeballing" of the raw data. Basically, the approach consists of logarithmic transformation and double centering of the data, factorization of the resulting cross-products matrix and a biplot of the resulting factor scores and loadings. Weighting by the marginal row and column sums of the original data can be applied in order to reduce the leverage of less important row and column items. This paper describes how the approach has fared over the past 30 years from a personal perspective.