Granger meets Rasch: investigating Gran...

The investigation of educational processes and trajectories is among the im-
portant goals of educational research and reflects the main aims of longitudi-
nal educational studies like the National Educational Panel Study (NEPS) in
Germany. From a methodological perspective, two foci can be distinguished:
1) measurement of the same competence across the life span and 2) effects
of competencies or context variables on the modifiability of a target compe-
tence. The latter can be addressed by Granger causation analysis. Granger
causation approaches examine whether past observations of a dimension X
can be used to predict more recent observations of dimension Y in a series
of repeated measures. A common item response model for scaling educa-
tional tests is the unidimensional Rasch model and its extensions. However,
more recently, multidimensional item response models became popular for
the exploration of dimensionality and assessing the latent factor structure
of multiple traits. In contrast to unidimensional models, these models of-
fer the advantage to simultaneously model effects of multiple traits on an-
swer processes. In this study, we combine the concept of Granger causation
with multidimensional longitudinal item response models. After introduc-
ing multidimensional item response models and their longitudinal extensions
for testing hypotheses compatible with Granger causation, we illustrate this
approach in an empirical data example. Further possibilities for the investi-
gation of Granger causation using item response models are discussed.