Abstract:

When measuring hierarchical latent trait, the standard IRT models have low efficiency in estimating item parameters and ability parameters. Without considering hierarchical structure, MIRT models have a high efficiency in estimating the first-order latent traits. It is not suitable for dealing with hierarchical latent trait. The HO-IRM not only can produce more accurate item parameter and ability parameters estimates, but also can obtain the first-order and high-order latent traits. The currently-existing HO-IRM includes high-order DINA model, high-order two parameter normal ogive hierarchical model, high-order logistic model, high-order polytomous items model and higher-order testlet model. In the future HO-IRM study, attention must be paid to the multi-level HO-IRM, HO-IRM within an item in the multidimensional situation and high-order cognitive diagnostic models.

Key words: IRT, HO-IRM, MIRM, first-order latent traits, high-order latent traits