A set of validity criteria for modeling job-based compensation systems

In their classic text, Cook and Campbell (1979) presented a set of principles for evaluating the validity of field research in social science that is designed to test hypotheses about cause-effect relations. In the present article, we develop a parallel set of criteria for the evaluation of the validity of predictive models. Our four criteria are: statistical conclusion validity, latent variable validity, structural validity, and external validity. After drawing a distinction between the measurement and structural level of models, we present a framework for assessing model validity and apply it to job-based compensation systems as an illustrative case. The article includes a summary of the advantages of structural equations modeling over traditional regression approaches for assessing wage structure within organizations.