Bootstrap 방법을 이용한 Rasch 이분 문항 적합도 지수의 신뢰구간 탐색

초록

This study aims to examine distributional properties of five item fit statistics of the WINSTEPS program for the dichotomous Rasch model. Especially, this study investigates confidence intervals of five item fit statistics for identifying reasonable critical values using bootstrap method. Under the Rasch dichotomous model, Monte Carlo simulations are conducted to investigate these issues. The test lengths are set to 10, 20, 30, 40, 50 items with the sample size of 100, 200, 400, 600, 800 persons. Using bootstrap method, 1,000 replications are made under each condition. The results show that the Infit MNSQ, Outfit MNSQ, and point-measure correlation coefficient rely too heavily on sample size compared to standardized Infit and Outfit statistics. Therefore, the critical range of Infit and Outfit MNSQ should be adjusted according to sample sizes. Furthermore, standardized Infit and Outfit do not follow the standard normal distribution so the conventional critical values(for example, ±1.96) are problematic. The reason for the results are discussed and suggestions are made for future studies.

키워드