大规模测评中IRT等值的影响因素研究

A Research on the Influence Factors of IRT Equating in Large Scale Assessments

通过模拟和实证研究探讨样本量、题本量以及锚题题型对大尺度测评中项目参数等值精度的影响,模拟研究和实证研究的结果均表明:(1)0/1计分项目参数的等值精度在大多数条件下均好于多级计分项目,相对而言,实证研究的差异不如模拟研究明显;(2)相对而言,样本容量的增加对于提高项目参数等值精度有着重要的作用,而增加题本数量的作用甚微;(3)无论是区分度参数还是难度参数,均表现为3个题本和2 000人的搭配已经可以达到较好的等值精度,如果进一步提高等值精度,只需将每一题本的样本容量增加到3 000人即可;在多级计分时,当选用5个题本时,每一个题本2 000人是最适宜的组合。

Simulation study and empirical research are conducted to investigate the influence of sample size, the number of booklets, the item format on IRT parameter equating. Findings show that： a）dichotomous items produce small parameter equating errors than polytomous items, relatively, the difference between empirical research is not as obvious as the simulation study; b）relatively, the increase in sample size plays an important role in improving the equivalent accuracy of the parameters, but the effect of increasing the number of booklets is negligible;c）whether it is the a-parameters or the b-parameters, the matching of three booklets and 2 000 sample size has been able to achieve better accuracy. To further improve the equating accuracy, only the 3 000 sample size is needed. When using polytomous items, five booklets and 2 000 sample size are the best combination.