Rasch 모형을 사용한 가교-고정 문항 모수 추정에서 WINSTEPS와 ICL 프로그램의 기능 비교

초록

In item response theory (IRT), two standard approaches to estimating item parameters are the joint maximum likelihood estimation (JMLE) and the marginal maximum likelihood estimation (MMLE). The purpose of this study was to investigate the properties of JMLE-based fixed-anchor IRT parameter estimation (FPE) and MMLE-based FPE methods using the Rasch model. For this purpose, this study used two computer programs, WINSTEPS, which is based on JMLE, and ICL, which is based on MMLE, to examine their relative performances in FPE. Computer simulations were conducted in various calibration conditions that were formed by crossing the levels of four factors: underlying ability distribution, test type (in configuration and size), sample size, and number of anchored items. Simulation results suggested that the two programs worked well for FPE in all conditions. However, ICL yielded Rasch difficulty parameter estimates more accurately and more consistently than did WINSTEPS. Overall, the simulation results showed that both JMLE and MMLE approaches are feasible for FPE but the latter is preferred to the former in practice. Possible explanations for such relative performances of the two programs were presented in discussion.

키워드