基于Markov Chain Monte Carlo模型对医院调查资料中缺失数据的多重估算

The Multiple Imputation of Missing Values in Hospital Survey Data with Markov Chain Monte Carlo Method

目的探讨基于Markov Chain Monte Carlo(MCMC)模型的多重估算法在处理医院调查资料缺失数据中的应用。方法运用SAS9.2编写程序,在分析数据的分布类型和缺失机制的基础上,采用MCMC法对缺失数据进行多次填补和联合统计推断,分析多重估算法的优势。结果数据服从多元正态分布与随机缺失,采用MCMC法填补10次所得的结果最佳。结论多重估算既可反映缺失数据的不确定性,又可充分利用现有资料的信息、提高统计效率、对模型的估计结果更加可信,是处理缺失数据的有效方法。

Objective To explore the applicability of Markov Chain Monte Carlo method（MCMC） in hospital survey data with missing values. Methods Based on the analysis of data distribution and missing mechanism, SAS 9.2 was used for MCMC method to impute missing values and summarize the results. Results The data is missing at random and conform to multivariate normal distribution. The result is acceptable as the dataset was imputed l0 times. Conclusion Multiple imputation can not only reflect the uncertainty of the missing data hut also can make full use of information, improve statistical efficiency and estimate the results more credible. This method is a effective method to deal with missing data.