关于多水平删失正态回归模型的模拟研究

A Simulation Study on the Multilevel Censored Normal Regression Model

目的探讨不同条件下的数据对于多水平删失正态回归模型参数点估计及其标准误的影响。方法模拟二水平删失正态回归数据,分别拟合二水平正态回归模型和二水平删失正态回归模型,并给出不同的一、二水平样本量及内相关系数条件,比较其参数点估计及参数标准误估计结果。结果二水平样本量越大,固定系数、一水平残差及残差间协方差的标准误估计越准确,而方差成分标准误估计的准确性对于一水平样本量要求较弱,内相关系数对参数标准误估计的准确性影响不大。结论存在删失因变量的多水平结构数据应使用多水平删失正态回归模型进行参数估计,其标准误与合适的各水平样本含量及内相关系数有关。

Objective To study the bias and standard error of parameter estimation for the multilevel censored normal regression model under different data settings. Methods To simulate data from 2-level censored normal regression fit them with different 1-level/2-level sample size and intra-class correlation coefficients. Results The estimation for standard error of fixed parameter, 1 -level residual and between residuals covariance will become more accurate with the increasing of 2-level sample size,while the estimation for variance components demands less 1-level sample size, the intra-class correlation coefficient has tittle influence on the accuracy of parameter standard error estimation. Conclusion When we come to multilevel structure data with censored dependent variable, using multilevel censored normal regression model to estimate the parameters will be more appropriate, the standard error depends on sample sizes in each level and intra-class correlation coefficient.