中文摘要：

目的针对两地区村民骨关节调查资料组群结构且零过多问题,阐明多水平零膨胀计数回归模型。方法介绍多水平Poisson和多水平ZIP模型原理,进一步完成山西省农村地区居民风湿性骨关节疼痛部位数影响因素分析的多水平模型SAS软件实现。结果多水平ZIP回归部分的随机项表明,两调查地区不同村庄居民关节是否疼痛以及疼痛部位数差别有统计学意义。logistic回归估计参数表明年龄、婚姻状况和高血压是影响关节疼痛与否的主要因素。而Pois-son部分估计结果表明年龄、地区、性别和地区交互作用是居民骨关节疼痛部位数的主要因素。Vuong检验进一步证实多水平ZIP模型比多水平Poisson模型更优。结论多水平ZIP回归模型是解决调查研究中组群结构及零发生次数过多问题模型拟合的最佳选择。

英文摘要：

Objective To use the multilevel zero-inflated count regression model for the data of the joints pain in rural areas with both clustered structure and excess zeros.Methods To introduce the multilevel Poisson and ZIP model,and to conduct the multilevel model for the factor analysis of the joints pain numbers of residents from rural area in Shanxi province using the SAS software.Results Random effects of the multilevel ZIP model indicate that these are significant differences of the probability of the pain and the number of painful parts with different villages.The logistic regression shows that age,marital status and hypertension are the main factors that affect the probability of pain,while the Poisson regression depicts that age,region,and the interaction of region with sex are related with the number of painful parts.Vuong＇s test further confirms that the multilevel ZIP model fit the data significantly better than the multilevel Poisson model.Conclusion The multilevel ZIP model is appropriate for the analysis of clustered count data with extra-zeros.