大尺度教育测评中IRT等值方法的比较研究

发布时间：2019-02-28 12:54

【摘要】：世界各国和国际组织都把大尺度教育测评作为关注教育质量的有效方式。有效发挥大尺度教育测评的作用离不开对其测评技术和方法的深入研究。等值是大尺度教育测评中数据处理的重要方法,有助于实现测验的公平。研究采用共同题非等组设计,利用实际数据,比较3种等值方法(同时性校准、StockingLord分别校准方法和Mean-Mean分别校准方法)与2种分别校准链接方案(集中式链接、链式链接)的题目参数等值精度。结果显示:多题本共同题设计下,StockingLord方法的估计精度要高于Mean-Mean方法;Mean-Mean和StockingLord两种分别校准等值方法在集中式链接方案和链式链接方案中的参数估计精度都比同时性校准的精度高;在链式链接方案中,随着链接题本数量的增加,参数等值精度在下降,在集中式链接方案中等值精度则没有这样明显的规律;等值方法对精度的影响要大于链接方案对等值精度的影响。其中,不同等值方法虽然在不同研究中分别被采用过,但同时考虑不同题本链接方案在国内外均没有实证研究。此研究结果,既弥补了国内外有关多题本共同题设计下不同等值方法以及不同题本链接方案的实证研究的不足,也为我国相关大尺度教育测评的等值设计提供参考和实证依据,进而更好地实现测验的公平。
[Abstract]:All countries and international organizations regard large-scale education evaluation as an effective way to pay attention to the quality of education. The effective use of large-scale education evaluation can not be separated from the in-depth study of its evaluation techniques and methods. Equivalence is an important method of data processing in large-scale education evaluation, which is helpful to realize the fairness of the test. In this study, three equivalent methods (synchronous calibration, StockingLord calibration method and Mean-Mean calibration method) and two separate calibration link schemes (centralized link, centralized link) were compared by using non-equal group design of common problem and using actual data. Link) the equivalent precision of the title parameters. The results show that the estimation accuracy of StockingLord method is higher than that of Mean-Mean method. The precision of parameter estimation of Mean-Mean and StockingLord is higher than that of synchronous calibration in centralized link scheme and chain link scheme respectively. In the link scheme, with the increase of the number of link items, the equivalent precision of the parameters is decreasing, but in the centralized link scheme, the equivalent accuracy is not so obvious. The effect of the equivalent method on the accuracy is greater than that of the link scheme. Among them, although different equivalence methods have been adopted in different studies, there is no empirical study at home and abroad to consider the link schemes of different questions at the same time. The results of this study not only make up for the deficiency of domestic and foreign empirical research on different equivalence methods and link schemes under the common design of multi-topic books, but also provide reference and empirical basis for the equivalent design of the related large-scale education evaluation in our country. And then better realize the fairness of the test.
【作者单位】：北京师范大学教育统计与测量研究所;北京师范大学发展心理研究所;北京师范大学认知神经科学与学习国家重点实验室;
【分类号】：G647

【参考文献】