复杂抽样下截取因变量回归系数方差估计的模拟研究

摘要

目的 探讨复杂抽样下截取因变量数据拟合回归模型后其回归系数的方差估计.方法 模拟复杂抽样下分别从左右方向发生截取的数据,按照是否考虑抽样特征分别拟合参数与半参数回归模型,给出两种情况下模型中回归系数的标准误,比较这两种情况所得结果的异同.结果 在样本量固定的前提下拟合截取回归模型,考虑复杂抽样特征后估计所得的回归系数与假设完全随机抽样一致,但其回归系数的标准误却不同于复杂抽样的情形.如果群内异质性高,群内相关系数很小,在复杂抽样条件下回归系数的标准误要低于不考虑复杂抽样特征的情形.结论 对于抽样框完整的复杂抽样截取数据,进行数据处理时应尽可能地将抽样特征考虑在内,运用复杂抽样数据方差估计得到的结果更接近于实际情况,统计推断结果更加真实可靠.%Objective To study the method of variance estimation by using complex censored data to fit tobit regression model.Methods To simulate complex deigned censored data, then fit tobit regression model with and without taking sampling characteristics into account respectively,compute the mean and standard error of regression coefficient and compare the differences of these two results. Results With a fixed sample size, the regression coefficient of tobit model taking into account of the sampling characteristics is more closer to the ture value, and the strandard error of it is different from that of ignoring the sampling characteristics. If the units in the cluster are highly heterogeneity,the strandard error of regression coefficient is lower than that of ignoring the sampling characteristics. Conclusion We should take into account the sampling characteristics when dealing with complex survey data, and estimate relevant variance using appropriate techniques so as to get more reliable results.