단일집단 및 다집단 검사자료의 IRT 문항모수 추정을 위한 BILOG-MG, ICL, PARSCALE 프로그램의 기능 비교

초록

The three computer programs BILOG-MG, ICL, and PARSCALE have been developed to estimate item response theory (IRT) item parameters and ability distributions using the marginal maximum likelihood and Bayes modal methods. With simulated single-group and multiple-group (common-item nonequivalent groups) test data, relative performances of the three programs were investigated on the degree of accuracy in estimation of the item parameters of the two- and three-parameter logistic (2PL & 3PL) models. As methodological bases, the estimation principles each program is based on and the detailed command syntax for running each program were presented. It was noted that PARSCALE could conduct multiple-group IRT estimation using the DIF model but deal with the maximum number of 23 items for a test form. The simulation results showed that the relative performances of the three programs should differ by the test data structure (single-group vs. multiple-group data). For the 2PL- and 3PL-model fitted single-group test data that were generated under combinatory conditions of test length, difficulty, and sample size, ICL overall performed best but the three programs performed almost equally in a practical sense. For the multiple-group test data that were generated under combinatory conditions of degree of nonequivalence between examinee groups and sample size, BILOG-MG and ICL performed almost equally. For this multiple-group IRT estimation, PARSCALE worked for 23-item test forms but performed much worse than BILOG-MG and ICL.

키워드