The sirens’ call in psychometrics: The invariance of IRT models

Abstract

The idea that item response theory (IRT) models yield invariant parameter estimates is widely accepted among scholars interested in achieving truly scientific measurements in social and behavioral sciences. Starting from a conceptual and mathematical definition of invariance, this article presents a critical examination of the theoretical and empirical support for the property of invariance with regard to populations and samples of items and subjects by means of simulated data. The distinction between internal and external invariance is introduced to clarify the meaning and limitations of invariance in IRT models. Furthermore, the consequences of giving in to the sirens' call of achieving invariant measurements in behavioral sciences are also discussed.