极大似然估计和拟极大似然估计模拟之比较

The Comparison between the Random Simulation of the Maximum Likelihood Estimators and the Quasi-maximum Liklihood Estimators

通过对单个协变量的带有测量误差的一维结构回归模型中总体平均处理效应的极大似然估计和拟极大似然估计的随机模拟结果进行比较,发现这两个公式都不受测量误差的影响,并且可以互换使用.当其它误差较小时用两个公式计算结果虽然相差不大,但相比较而言用拟极大似然估计较好,反之,当其它误差较大时用极大似然估计较好.

In this paper, the comparison is qiven between the random simulation of the maximum likelihood estimators and the quasi-maximum likelihood estimators of the population-averaged treatment effects in the one-dimensional structural regression models with one covariate and the measurement errors. It is found that one of the formulae can be replaced by the other, and they are not influenced by the measurement errors. When the other errors（ that is, they are different from the measurement errors）are little, the results are similar when using the two formulae,but the use of the quasi-maximum likelihood estimators is better than the use of the maximum likelihood estimators. On the contrary, when the other errors are large, the use of the maximum likelihood estimators is better than that of the quasi-maximum likelihood estimators.