从项目特征曲线的估算比较参数型及非参数型项目反应理论模型  

Recovery of Parametric Item Characteristic Curves with Parametric and Nonparametric IRT Models

作　　者：吴琼[1] Pui-wa Lei 

机构地区：[1]北京大学中国社会科学调查中心 [2]美国宾州州立大学

出　　处：《考试研究》2015年第6期46-55,共10页Examinations Research

摘　　要：项目反应理论模型的参数估计一般需要较大样本量,小样本量条件下参数型与非参数型项目反应理论模型的相对优势并无定论。通过计算机模拟数据比较两类模型在小样本量时(n<=200)估计项目特征曲线所产生的偏误及均方根误差。当模拟数据基于3PL模型生成时,参数型与非参数型模型在样本量低于200时估值偏误方面无差别,但前者均方根误差较小。在样本量为200时,两模型估算值类似。当真实数据基于3PL模型且样本量小于200时,参数型Rasch模型比非参数核平滑模型更值得推荐。This study investigates the recovery of parametric item characteristic curves （ICC） based on parametric and nonparametric item response theory （IRT） models at small sample sizes （N 〈 = 200）. Estimation method, sample size, test length and ability group were studied with simulated data generated based on a 3 - parameter logistic model. Both bias and root mean square error （RMSE） of ICC estimates were evaluated. When true response curves follow 3 - parametric logistic functions, none of the studied factors has a large effect on the bias of ICC estimates at the studied sample sizes. However, the Rasch estimates are consistently associated with smaller RMSE than estimates from nonparametric regression model with kernel smoothing. RMSE decreases as sample sizes increase in both models, and between- model difference shrinks as sample sizes increase. The nonparametric regression model with kernel smoothing yields comparable ICC estimates with the Rasch model at a sample size of 200.

关 键 词：项目反应理论 核平滑 项目特征曲线 非参数 模拟数据 

分 类 号：G424.74[文化科学—课程与教学论]