Bock의 명명반응 모형의 재모수화: 원형 모형과 개정 모형

초록

In item response theory (IRT), Bock’s (1972) nominal response (NR) model has been widely used for analyzing response data from various cognitive and non-cognitive measures. The original NR model has been newly parameterized by Thissen, Cai, and Bock (2010), who set primary goals that the reparameterization should facilitate the extension of the NR model to become a multidimensional IRT model and could make the model easier to explain, as compared with other IRT models. The present paper investigates how the original and new versions of the NR model relate to each other in terms of (1) modeling and estimable item parameters, (2) addressing the issues of reflection indeterminacy and highest-category specification, and (3) item parameter error variance-covariance matrices. The exposition of the relations is associated with a discussion on estimation settings of or outputs from the computer programs, MULTILOG, flexMIRT, and IRTPRO. Numerical examples are given to illustrate the relations. For researchers’ understanding of the new version of the NR model and for a successful transition from the original to the new version, some suggestions are provided for the flexMIRT program so that it may be used more easily and efficiently.

키워드