零膨胀有序logit模型和传统模型在零膨胀等级资料分析中的比较研究  

Comparison on the Effects of the Zero-inflated Ordered Logit Model and Traditional Models for Zero Inflation Hierarchical Data

作　　者：黎怡 张欣 樊虹 徐艺耘 吴声 张铁军[1] Li Yi;Zhang Xin;Fan Hong(Department of Epidemiology,School of Public Health,Fudan University,Shanghai 200032)

机构地区：[1]复旦大学公共卫生学院流行病学教研室,公共卫生安全教育部重点实验室,200032

出　　处：《中国卫生统计》2024年第5期671-675,共5页Chinese Journal of Health Statistics

基　　金：国家自然科学基金(81772170);国家重点研发计划重点专项(2017YFC0211704)。

摘　　要：目的对零膨胀有序logit模型展开应用并和传统模型进行比较,探讨其优势和局限性。方法对各自变量进行单因素分析并计算相应的对数似然值、AIC及BIC等模型评价指标,根据这些指标结果依次纳入各自变量并组成不同的自变量集。分别使用有序logit模型、广义有序logit模型和零膨胀有序logit模型基于不同自变量集进行数据拟合,通过比较不同情况下各模型的评价指标及参数估计结果对模型进行评价。结果根据模型评价指标,零膨胀有序logit模型对数据的拟合效果优于其他两种模型。零膨胀有序logit模型对不同自变量集进行拟合时,其参数估计结果高度一致。不同模型的参数估计结果存在一定的差异。结论零膨胀有序logit模型在应用时有着较好的性能。单因素分析的模型评价指标可以用来帮助筛选使得模型性能最优的自变量集。Objective Apply the Zero-inflated ordered logit model and explore its advantages and limitations by comparing it with the traditional models.Methods Univariate analysis of each independent variable was performed,and the corresponding model evaluation indicators such as log likelihood,AIC,and BIC were calculated.According to the results of these indicators,the respective variables were incorporated in turn and formed different sets of independent variables.The ordered logit model,generalized ordered logit model,and Zero-inflated ordered logit model were used to fit the data based on these different sets of independent variables,respectively.The models were evaluated by comparing the model evaluation indicators and parameter estimation results.Results According to the model evaluation indicators,the Zero-inflated ordered logit model performed better than the traditional models.The parameter estimation results of the Zero-inflated ordered logit model based on different sets of independent variables are highly consistent.There are differences in the parameter estimation results of different models.Conclusion The Zero-inflated ordered logit model has good performance in application.Model evaluation indicators from univariate analyses can be used to help filter the set of independent variables that make the model′s performance optimal.

关 键 词：零膨胀有序logit模型 等级资料 STATA软件 

分 类 号：R181.2[医药卫生—流行病学]