摘要
为了探测含零较多的计数数据对ZIP回归模型的影响,基于EM算法和完全数据似然函数,笔者利用数据删除方法研究了模型的全局影响,得到了参数估计的一步近似表达式以及相应的广义Cook距离和Q-距离,并研究了在EM框架下的广义杠杆值。最后,通过实际问题说明了所得统计量的有效性。
To detect the effect of count data with many zeros on the zeroinflated poisson regression model, the authors studied the global diagnostics analysis and using the completedata log-likelihood function and the EM algorithm. The one-step approximation of parameter estimates and some statistics, such as generalized cook distance, Q - distance, are obtained.Meantime, the generalized leverage was discussed based on the EM algorithm. Finally, a numerical example was given to illustrate our results.
出处
《南京林业大学学报（自然科学版）》
CAS
CSCD
北大核心
2007年第6期109-112,共4页
Journal of Nanjing Forestry University：Natural Sciences Edition
基金
国家自然科学基金资助项目(10671032)
南京农业大学青年科技创新基金资助项目(KJ06026)
关键词
ZIP回归
EM算法
数据删除
广义COOK距离
Q-距离
广义杠杆值
Zero inflated poisson regression model
EM algorithm
Case-deletion
Generalized Cook distance
Q - distance
Generalized leverage
