ABSTRACT
Many testing instruments in language assessment aim for the measurement of multiple domains instead of collapsing several facets into a single measure. Multidimensional Rasch modeling is a measurement approach to assess multiple latent trait levels in complex test data that are often found in language assessment. Multidimensional Rasch models offer advantages such as a comprehensible criterion-referenced interpretation of test scores, robust parameter estimates under suboptimal testing conditions (medium sample size, complex design), a direct estimation of correlations among measurement dimensions, and the usage of collateral information from these correlations to increase the reliability of each single dimension. In language assessment, the approach offers substantial promise to gain insights into the structure of receptive skills in first and foreign language assessment. In the present study, several multidimensional Rasch models with between-item and within-item multidimensionality were applied to investigate the underlying structure of a first language listening comprehension of the eighth-grade students and its relations to first language reading comprehension (as another receptive skill) and foreign language listening comprehension. These models were estimated in data from large-scale, statewide standardized assessments of learning. Dissimilar implications offered by these different models were discussed against the background of the results from this study.
