There are four deficiencies in the 3PL model of modern item response theory. First, localindependent assumption is made, that is, the relationship between items is ignored. It, hence,
can’t to use to estimation the structure between items. Second, it can’t apply to time seriescases directly. Third, missing data processing is neglected, so it is unsuitable for educational
measurement. Fourth, satisfactory parameters estimation depends on quite a few samples. Itcan be introduced to large scale entrance
examination only.
The main purpose of this study is to integrate a hybrid model of generalized hiddenmarkov model (GHMM) and kernel smoothing nonparametric IRT (KN-IRT) to conquer these four constraints. Such a model can be used for item analysis of both series correlation
andseries non correlation cases. Furthermore, this model can be
applied to item ordering theory(OT) or item relational structure (IRS) to enhance the structure between items more precisely.
Keywords: Generalized Hidden Markov Model, Item Response Theory, Item Relationship Structure, Ordering Theory, Kernel smoothing
Nonparametric IRT