XCALIBRE 프로그램의 가교-고정 IRT 추정 기능 분석

초록

In item response theory (IRT), fixed-anchor parameter estimation (FPE) is used to maintain an established ability scale by estimating the parameters of non-anchor items on the established scale of the fixed items whose parameters are known. The first purpose of this study was to verify that the FPE facility provided by a recent version (4.2) of the XCALIBRE program is different from that of an earlier version (1.0 or 1.1) of the program, which was actually the “separate estimation and scale linking” approach. The second purpose was to compare the performance in FPE of XCALIBRE with that of IRT Command Language (ICL) and examine whether the former would be properly functioned in its own, and better functioned than the latter. For these purposes, computer simulations, based on the Rasch model and the 3PL model, were carried out in a variety of FPE conditions in which the XCALIBRE Fix, scale-linking, and ICL Fix methods were conducted. The simulation results showed that the recent version of XCALIBRE uses a unique, EM algorithm-based FPE method, which does not equally perform as a scale linking method. However, unlike the scale-linking and ICL Fix methods, the XCALIBRE Fix method substantially under- or over-estimated the difficulty parameters of non-anchor items except when the test data were generated using the standard normal [N(0, 1)] ability distribution. It was reasoned that such biased FPE results should be related to the point that during the EM iterations for FPE, XCALIBRE does not update the prior distribution for ability as the one estimated from the previous iteration but keeps the prior distribution as the N(0, 1) distribution.

키워드