非参数高斯核平滑法估计能力值的精度  

The Accuracy of Nonparametric Kernel Smoothing Approaches for Estimating Ability

作　　者：张军[1] 

机构地区：[1]北京语言大学

出　　处：《中国考试》2015年第5期47-52,共6页journal of China Examinations

摘　　要：单维非参数项目反应理论(NIRT)假设数据满足单维性、单调性和局部独立性三个假设,不要求数据拟合某种特殊的函数形态。非参数高斯核平滑法能根据数据本身的特点,构拟NIRT的项目反应函数,估计被试的潜在能力值。实验模拟了42套拟合单维双参数逻辑斯蒂模型的数据,比较了非参数高斯核平滑法在题量和被试样本数2个因素不同水平下估计被试能力值的精度。结果表明:样本数对方法的估计精度没有显著影响;题量的增加能较好改善估计的精度。在测试实践中,应采用不同的题量设计,以满足估计精度的要求。The unidimensional nonparametric item response theory（NIRT） only assumes that the datum satisfy three assumptions： unidimensionality, monotonicity, and local independence, does not require the datum fitting some special function. Nonparametric Gauss kernel smoothing approach can construct the item response function and estimate the ability, according to the characteristics of the datum itself. The experiment simulates 42 sets of datum fitting unidimentional two parameters logistic model, comparing the accuracy in various conditions. The results show that the factor“sample number”did not significantly affect the estimation precision; the increase of item quantity can improve the precision of estimation. In the practice, we should design different size of items in the test, in accordance with the requirements of accuracy.

关 键 词：核平滑 精度 

分 类 号：G405[文化科学—教育学原理]