基于边缘ZI回归模型的分组数据影响分析

为了研究含零较多的分组计数数据对边缘ZI(zero-inflated)回归模型的影响,在完全数据的对数似然基础上利用局部影响法对其进行诊断.同时,基于模型中参数的ES(expectation-solution)估计,分别研究了类内加权扰动、类音加权扰动、解释变量扰动和响应变量扰动下的局部影响分析,得到了相应的诊断统计量.最后,根据所得统计量,获得了园艺试验中一组计数数据的影响点,结果说明该文提出的方法是有效的.

Influence Analysis for Marginal Zero Inflated Regression with an Application to Clustered Data

To study effect of the clustered data on the marginal zero - inflated regression models, this paper develops the diagnostics analysis for it by using local influense method based on the completedata log - lidelihood. Meanwhile,baed on the expectation -solution estimates of parameters in the model, the local influence under the caseweights within clusters, case- weights among cluster, perturbation of explanatory variables and response are discussed, respectively. Finally, the influence cases in a real based on statistcs derived, and the result demonstrates hat count data from a horticultural experiment are obtained the method of the article is effective.