探究MIMIC方法在多级计分项目中的DIF检验效能——基于蒙特卡洛模拟实验

Exploring the Efficacy of the MIMIC Method in Conducting DIF Analysis for Polytomous Items:Based on the Monte Carlo Simulation Experiment

随着多级计分在心理和教育领域中日益广泛的应用,对检验项目功能差异(DIF)的方法提出新的挑战。已有研究表明,在检验DIF的方法中,MIMIC是一种经济有效的检验方法,然而还没有研究系统地分析MIMIC方法在多级计分项目中的有效性。本研究通过蒙特卡洛实验,探讨参照组与目标组的样本容量、DIF类别、项目区分度、组间能力差异和在锚题中存在的DIF题量5个因素,并在这些因素不同情况的组合中分析MIMIC方法的第一类错误率和检验力。研究发现:1)MIMIC是一种能够灵敏地检验一致性DIF的方法,即使在目标组样本容量较小或明显小于参照组的情况下,它仍然能很好地控制第一类错误率;2)纯化步骤对MIMIC方法控制第一类错误率、提高检验力是有必要的,但MIMIC方法对污染程度又有一定的容忍性;3)检验力受到低区分度的严重影响,但太高的区分度又会导致第一类错误率的增加;4)MIMIC方法对一致性DIF的检验力随着样本容量的增大而增大。

With an increasing use of multi-level scoring in the field of psychology and education, new challenges have been raised in methods of testing differential item functioning(DIF). Studies have shown that MIMIC is a cost-effective method to test DIF, but there has been no systematic study to analyze the effectiveness of MIMIC method in polytomous items. This study considers five influencing factors through a Monte Carlo experiment: sample size of a reference group and a target group, DIF category, item discrimination, difference in inter-group ability, and DIF contamination in anchor items. Combined with the interaction effects of the five factors, type I error rates and test power of the MIMIC method are discussed. The study finds that: 1)MIMIC is a method that could sensitively identify consistent DIF. Even if a target group sample size is small, or significantly smaller than a reference group, it could still control the type I error rate well;2)the purification step is necessary for the MIMIC method to control the type I error to improve test power. The MIMIC method is yet tolerant to DIF contamination to some degree;3)the test power is seriously affected by low discrimination, but too high discrimination will lead an increase in the type I error rate;4)the MIMIC method’s test power for consistent DIF increases as the sample size increases.