Unidentifiable Relationships in Conceptual Marketing Models

All conceptual models published in the Journal of Marketing from 1995 to 1999 are analyzed using the traditional diagnostic methods for identification, finding two-thirds of the published conceptual models contain relationships that are unidentifiable.

Abstract

For fifty years methods have existed to diagnose whether a conceptual model is unidentifiable, but it appears marketing scholars do not regularly check identification before estimation. To confirm this, all conceptual models published in the Journal of Marketing from 1995 to 1999 are analyzed using the traditional diagnostic methods for identification. Two-thirds of the published conceptual models contain relationships that are unidentifiable. These relationships have been empirically estimated, although it is impossible to measure their parameters validly. The published empirical estimates are spurious and cannot be trusted to represent the behavior they claim to measure until the identification problem has been corrected. The theory, not the statistics, must change to validate the measurements, so the paper concludes with suggestions that can help avoid unidentifiable conceptual theories.