NA样本下随机设计情形线性模型的经验似然

Empirical Likelihood for Linear Models with Random Designs under Negatively Associated Samples

在NA样本下,利用大小分块方法和矩不等式首次构造并证明了随机设计情形下的线性模型的回归系数的经验似然比统计量的极限分布近似服从卡方分布,并进而构造其经验似然方法下的置信域,最后,模拟比较得出经验似然法下的置信域优于正态逼近法下的置信域。

Under the samples of negative association, empirical likelihood of regression coefficient in linear models with random designs is first constructed and proved by blocking method and moment inequalities. And then the confidence regions based on empirical likelihood are constructed. Simulation studies show that confidence regions based on empirical likelihood is better than those based on normalapproximation.