Distinguishing Trait And Method Variance In Multitrait-Multimethod Matrices: A Reply To Golding

It was suggested that methods for evaluating trait and method variance need not await a fully developed nomological network for application, but might proceed through an interplay between inductive generalization and theory.

Abstract

Some observations are provided regarding certain techniques for the evaluation of the problem posed by Campbell and Fiske, that of convergent and discriminant validation. A recent proposal by Golding and Seidman (1974) for the evaluation of multitrait-multimethod matrices is found to be inadequate for this problem because it (a) is inconsistent with the Campbell-Fiske requirements of testing traits as separate units; (b) confounds trait and method variance; (c) offers no empirically defensible definition or technique for isolating method variance; and (d) yields absurd results when applied to a matrix highly consistent with Campbell-Fiske criteria for convergent and discriminant validity. Golding's (1976) conjecture that factors designated as trait factors contained method variance in Jackson's (1975) reformulated multimethod factor analysis logically was found to be based on an implicit redefinition of terms by Golding, and could not be verified empirically. It was suggested that methods for evaluating trait and method variance need not await a fully developed nomological network for application, but might proceed through an interplay between inductive generalization and theory.