零膨胀计数资料模型选择与比较

The Selection and Comparison of Zero-inflated Count Data Model

目的阐明处理零膨胀计数资料的Hurdle模型和零膨胀模型原理,并探讨模型选择与比较。方法拟合心肌缺损节段数资料的基础计数模型、Hurdle模型和零膨胀模型,通过似然比检验、Vuong检验和AIC、BIC信息准则进行模型选择与比较。结果心肌缺损节段数资料拟合ZINB和NBH模型要优于其他计数模型;ZINB与NBH进行模型选择时,Vuong检验结果尚不能判断哪个模型更优,但由于数据中零计数包括结构零和抽样零,结合AIC、BIC和专业解释,可以认为ZINB模型更合适。结论 Hurdle模型和零膨胀模型是处理零膨胀计数资料的有效方法。模型选择时需要考虑数据结构和专业解释。

Objective To introduce the theorys of zero-inflation model and Hurdle model in the zero-inflated count data,and discuss model selection and comparison.Methods To fit basic count model、zero-inflation model and Hurdle models with the segment numbers of myocardial ischemic data.Likelihood ratio test,Vuong test and AIC,BIC information criterion are used to select the models.Results The ZINB and NBH model are superior to other count models;When selecting the better model between ZINB and NBH,although Vuong test results can not decide which model is better,it shows that ZINB model is the more appropriate for the data according to AIC,BIC,data structure and professional interpretation.Conclusion Both zero-inflation model and Hurdle model are able to deal with zero-inflation problem of count data.When model selection,data structure and professional interpretation should be considered.