Evaluation of relations between scales in an IRT framework

This report studies the application of a new procedure for measuring across-scales relationship in a multi-dimensional IRT test. A multi-dimensional test is one in which a test is divided into sub-sets and latent variables are measured separately for each scale. The latent variables are assumed to correlate and the new procedure called ‘limited information maximum likelihood estimation’ proposed by Rubin and Thomas (2001) is used to estimate this correlation. The authors did not indicate that they had empirically tested this new procedure and in this report empirical tests are carried out to observe the effects of different models and values of item and person parameters.