半变系数再生散度混合效应模型的影响分析

Influence Analysis of Semivarying Coefficient Reproductive Dispersion Mixed Models

本文把随机效应看作缺失数据并利用P-样条拟合非参数部分,应用Monte Carlo EM 加速算法得到半变系数再生散度混合效应模型的未知参数的估计,同时利用Q函数,得到了模型的广义Cook距离。此外,本文还研究了三种不同扰动情形的局部影响分析,得到了相应的影响矩阵。最后,通过一个实际例子验证了所提出的诊断统计量的有效性。

This paper proposes several case-deletion as well as local influence measures for assessing the influence of an observation for semivarying coefficient reproductive dispersion mixed models. The essential idea is to treat the latent random effects in the model as missing data and estimate unknown parameters by acceleration of Monte Carlo EM algorithm. Onthe basis of the Q-function which is associated with the conditional expectation of the complete-data log-likelihood, we generate generalized Cook Distance. Moreover, three different perturbation schemes are discussed. Finally,one real illustrative example is presented to prove the methodology.