缺失数据的估计方法及应用

The Estimation Method for Missing Data and Its Applications

应用EM算法的思想在双参数逻辑斯蒂克模型下对存在缺失数据的参数估计方法和恢复缺失数据的统计插补方法进行研究.蒙特卡洛模拟和实证研究结果表明,在进行统计插补恢复缺失数据时,该方法使得估计结果比较理想:联合极大似然估计与EM算法相结合,先估计参数,再填补缺失值,再估计,再填补,直到似然函数值稳定.

The way to deal with the missing data in large-scale tests attracts the attention from national and international researchers.The present research tries to get the possible responses for the missing data of the candidates based on their partial responses through statistical imputation method,and to estimate their ability theta and item parameters.The parameter estimation method for missing data and the statistical imputation method of getting back missing data are studied under the two-parameter logistic model,with the Monte Carlo simulation applied,and the results on the influence from different percentages of missing data are compared.To investigate the feasibility of statistical imputation method and parameter estimation method,an experimental test has been carried out with the data of English subject from a project in year 2007.The results from the Monte Carlo simulation and the experimental test suggest that it is practical and acceptable to estimate parameters through the combined method of JMLE and EM,and to get back the missing data by applying the statistical imputation method.