Item Parameter Calibration in the Multidimensional Graded Response Model with High Dimensional Tests
Kenneth E. McClure, Ross Jacobucci · 2023
Multidimensional item response models, such as the multidimensional graded response model (MGRM), are becoming increasingly common in educational and psychological research in part due to the associated advantages of multidimensional computerized adaptive testing. These methods, however, require known or calibrated item parameters. Existing research on item parameter calibration for the MGRM is largely restricted to tests with simple structure and few dimensions. Given the increasing complexity and dimensionality of psychological measures, the current paper examines MGRM item parameter calibration for higher dimensional settings using Metropolis-Hastings Robbins-Monro (MH-RM) estimation. Complex test structure and non-normal latent traits are also examined. Results suggest that test and trait characteristics impact item parameter recovery though discrimination and boundary parameters are differentially influenced. Calibration samples of 1000 respondents appear sufficient for many research settings. MH-RM estimation facilitates time-efficient item calibration for high dimensional tests. Findings support the feasibility of calibrating multidimensional item banks for psychological research.