贝叶斯理论框架下的2种纵向缺失数据处理方法的比较——以潜在变量增长曲线模型为例

The Comparision of Two Approaches to Bayesian Method for Missing Data in Longitudinal Model——Growth Curve Model for Example

在贝叶斯估计框架下,通过模拟研究比较完全贝叶斯和部分贝叶斯方法对参数估计的影响.研究结果表明:随着缺失比例的增加,2种方法得到的均方误差(RMSE)都会增大;完全贝叶斯方法和部分贝叶斯方法在缺失比例较小时几乎相同,只在缺失比例为0.5时,前者明显优于后者.

The research explored the relative performance of Fully Bayesian method and Partially method in the estimation of growth curve model parameters.Only Simulation studies were used in the compassion in which four missing rates（0,0.10,0.30,and 0.50） were investigated.In each situation,50 matrixes with missing response were generated and the index RMSE（root mean square error） were compared the two approaches.The results showed that：（1） the accuracy of parameter estimations of the two approaches were both affected by the missing rate,and as the increasing of missing rate,the bigger of RMSE.（2）When the missing rate is small,the RMSEs of the two approaches were almost same,however,Fully Bayesian method got better than Partially method when missing rate came to 0.50.