缺失数据比率和处理方法对非随机缺失数据能力参数估计准确性的影响

探讨了IRT背景下非随机缺失数据的合适处理方法.采用IRTLAB模拟产生50批500个被试在20个0-1记分项目上的反应数据,产生了不同比率的MNAR; 再用IN、NP、FR、CM、MI和EM共6种方法分别处理MNAR,使用BILOG-MG软件估计被试的能力参数,并计算在不同条件下各种方法的BIAS、BIAS_abs、R_(θ,θ^∧)和RMSE.研究发现:随着缺失比率的增加,参数误差越来越大; FR会导致IRT参数估计产生较大的误差,且不稳定,而MI与EM算法则相对稳定; 综合BIAS和RMAE等几个指标,NP在处理MNAR时产生的误差较小也更稳定.因此,在IRT背景下估计被试能力参数时,应选择NP、MI或EM方法处理缺失数据

The Effects of Missing not at Random Data to the Accuracy of Ability Parameter Estimation in IRT

A simulation study is conducted to explore proper methods of handling missing not at random data in IRT context.First,generate response data of 500 subjects on 20 items,every item is scored by 0 or 1.Then,different percent of missing data were simulated,next,6 kinds of methods were used to deal with the missing data.Estimate the subjects’ ability via the BILOG-MG software,comparing different methods with the following four criterias:BIAS、BIAS_abs and RMSE_abs and is shown that these methods exhibit varying degrees of effectiveness in dealing with MNAR.It is advisable for us to us the NP,MI,EM methods to handle with MNAR in IRT context.