Paul R. Lohnes

Abstract

It is the purpose of this paper to describe and illustrate a number of interpretivedevices which have recently emerged and which have not before beenutilized in the geographic literature. These devices are the computation of thecanonical factor structure matrix, the variances extracted from each measurementdomain by the canonical factors, the redundancy of the canonical factorsof one set given those of the other, and the canonical factor scores. The canonicalfactor structure matrix provides the correlations of the variables (or measurementdomain) with the canonical factors and takes the place of the raw canonicalvectors in which the variances are uncontrolled. The variances extractedfrom a measurement domain by a canonical factor may shrink to insignificanceif its canonical correlation with the corresponding factor for the othermeasurement domain is low. A better measure of the interrelationships betweenthe two measurement domains being analyzed is the redundancy measure,which is the product of the variance extracted and the variance shared foreach pair of canonical factors. This paper also introduces the notion of canonicalfactor scores which correspond to the scores computed in principal componentsanalysis and which provide a mapping of the observation units into thecanonical factor space. The computation of these indexes is described in the mathematicalsection which follows. Two research examples are then providedto illustrate the application and interpretation of the technique.

D. Michael Ray, State University of New York, Buffalo
Paul R. Lohnes, State University of New York, Buffalo