Stability and change of unidimensionality: The sample case of deductive reasoning

In most longitudinal studies, researchers take the stability of a latent dimension for granted. This article shows that items assumed to measure a given construct can change their meaning during developmental progress or across cohorts. The authors suggest that Rasch models offer a viable approach to assessing such aspects of dimensionality. First, they outline the general ideas of the Rasch model. Then, they discuss how mixed Rasch models, an exploratory approach for identifying latent classes of participants within which unidimensionality holds, can be used to assess the dimensionality of a given item set. To illustrate their ideas, the authors present data from a study on adolescents' deductive reasoning that clearly show that the same test items can be measuring qualitatively different latent traits in different subsamples or at different time points.