一类零过度数据的建模和诊断分析.pdf

综上所述，本文比较深入系统地研究了零过度偏差泊松回归模型和基于重复测量数据的 混合效应模型的参数估计、统计诊断以及模型误判检验等问题，并得到了一系列新的诊断统计 量．大量的数值实例和随机模拟结果表明，这些统计量都是很有效的． 关键词：零过度数据；广义泊松分布；双泊松分布；EM算法；最佳线性无偏预测型对数似然； 约束极大似然；Score检验；梯度检验；散度参数；ZI参数；数据删除；局部影响；误判； Carlo模拟． Bayes估计；Gibbs抽样；MH算法；K—L距离；Monte 中图分类号：0212．2 分类号AMS(2000)：62J02；62F25 II Abstract Countdataare encounteredawide as in of commonly varietydisciplines，such and publichealth，biomedical，econometrics，road safety,manufacturingagriculture， etc．Classicaldiscrete linearmodelsareusedtofitsuchdata．However．real generalized countdataoftenshowmorezerocountsthanwhatcanbe a model predictedbysimple toas standarddiscretedistribution (referedzero-inflation，denotedbyZI)，therefore，a failtofit discretemodelshave moreand become may them．Recently,zero-inflated more for suchzero-inflateddataandhave an role popularmodelling playedimportant in and thesis studiesstatistical for theorypractice．This diagnostics systematically zero-inflated and Poisson models mixedmodels． dispersionregression corresponding 2devotesto statistical forzero-inflated Chapter studying diagnostics dispersion Poisson models．Wefirst discusstheestimateof basedonordi- regression