EM算法对缺失数据极大似然估计的实证研究

An Empirical Study on Maximum Likelihood Estimation of Missing Data by EM Algorithm

摘要: 用EM方法对缺失数据进行极大似然估计是处理缺失数据的一种基本方法。 本文在介绍基于EM的极大似然估计方法基础上，对一类随机缺失数据集用EM方法进行补齐，结果表明在10%、20%、30%三种不同缺失率下，EM方法进行缺失数据补齐的相对误差均小于0.1，呈现出低缺失率情况下的高准确性的特点。进一步将EM方法应用于对实际问卷调查中的缺失数据进行补齐，对补齐前后调查结果的影响进行了分析。

Abstract: The maximum likelihood estimation of missing data by EM is a basic method to deal with missing data. In this paper based on the maximum likelihood estimation method based on EM, was filled with the EM method for a class of random missing data sets; the results show that in 10%, 20%, 30%, three different loss rate, the relative error of EM method to fill up the missing data is less than 0.1, showing a high accuracy under the condition of low the loss rate. Further, the EM method is applied to the missing data in the actual questionnaire survey, and the influence of the survey results before and after the complement is analyzed.