A Comparison of SEM DIF Detecting Methods for Multiple Groups

• Main Authors: Min-shu Syu, 徐敏書
• Other Authors: none
• Format: Others
• Language: zh-TW
• Published: 2010
• Online Access: http://ndltd.ncl.edu.tw/handle/07459696284436221800

碩士 === 國立臺南大學 === 測驗統計研究所碩士班 === 98 === The purpose of this study was to compare the efficiencies of SEM-based DIF detecting methods using simulated complete data sets and simulated data sets with 20% systematic missing values. The comparison was across two levels of medium and serious DIF effect sizes. Under complete data sets, the efficiencies of two SEM-based methods, Multiple Groups Analysis(MG) and Multiple Indicators and Multiple Causes(MIMIC), were compared to the efficiency of SIBTEST in detecting DIF. Under data sets with 20% systematic vales, the study also examined how the Multiple Imputation(MI) and Zero Imputation(ZI) affected MIMIC and how the MI, ZI and maximum likelihood imputation (MLI) affected MG in detecting DIF items. The results showed that MIMIC, MG and SIBTEST produced similar results under complete data sets. Furthermore, compared to baseline, MI, ZI, and MLI increased the probability of type I error in most situations. However, they did not affect the power in detecting DIF.