缺失数据下广义非线性回归的经验似然及诊断∗

摘要

研究了数据缺失情况下广义非线性回归模型的统计诊断问题；在响应变量随机缺失的情况下，先利用经验似然方法进行参数估计，得到其渐近置信区间，并通过随机模拟比较出经验似然方法比一般方法求置信区间的优越性；对模型进行影响分析，提出经验似然距离、经验Cook距离以及标准化残差等诊断统计量，最后通过实例验证统计诊断方法的有效性和可行性。%This paper studies the diagnosis problems of generalized nonlinear regression model under data missing, under random missing of response variables, firstly uses empirical likelihood method to conduct parameter estimation, obtains its asymptotic confidence interval, then through random simulation and comparison, gets that empirical likelihood method is more superior than general methods in solving the asymptotic confidence interval, based on the analysis of the impact of the model, proposes the diagnosis statistical data such as empirical likelihood distance, empirical Cook distance, and standardized pseudo⁃residuals and finally uses examples to verify the effectiveness and feasibility of the statistical diagnosis method.