基于数据删除的Poisson-Gamma模型的影响评价  

Influence Analysis for Poisson-Gamma Models Based on Case-deletion

作　　者：李爱萍[1] 解锋昌[1] 

机构地区：[1]南京农业大学数学系,江苏南京210095

出　　处：《南京理工大学学报》2006年第6期797-800,共4页Journal of Nanjing University of Science and Technology

基　　金：国家自然科学基金(10371016);南京农业大学青年科技创新基金(KJ06026)

摘　　要：为了研究记数数据对于Poisson-Gamma模型的影响,该文通过引入EM算法,利用基于完全数据似然函数的条件期望进行影响诊断分析,并且进一步基于数据删除模型研究全局影响,得到了数据删除模型中参数估计的一步近似和相应的广义Cook距离和Q-距离。最后,利用所得统计量获得一个实际记数数据中的影响点,结果说明该文提出的方法是有效的。To study the effect of the count data on the Poisson-Gamma models, this paper develops the diagnostics analysis based on the conditional expectation of the complete-data log-likelihood function in the EM algorithm. The global sensitivity based on the case-deletion model is discussed, and some statistics, such as generalized Cook distance, Q-distance, are obtained. The influence cases in a real count data are obtained based on the statistics derived in the paper, and the result demonstrates that the method is effective.

关 键 词：Poisson-Gamma回归 EM算法 数据删除模型 广义COOK距离 

分 类 号：O212.2[理学—概率论与数理统计]