微核试验数据的Poisson和负二项回归模型拟合效果比较

Comparison of Fitting Results of Poisson Regression and Negative Binomial Regression Models for Data of Cytokinesis-block Micronucleus Test

目的比较Poisson和负二项回归模型对微核试验数据(每1 000个双核淋巴细胞中具有微核的淋巴细胞数)的拟合效果,为微核试验数据的模型拟合提供依据。方法运用微核试验数据,拟合Poisson分布和负二项分布回归模型,采用对数似然函数、偏差统计量、Pearsonχ2统计量和聚集性指数等指标比较2种回归模型对实例数据的拟合效果。结果负二项回归模型对数似然函数值与自由度的比值(-2.51)大于Poisson回归模型(-3.52);负二项回归模型拟合优度统计量-偏差统计量和Pearsonχ2统计量与对应的自由度比值(1.16和1.07)小于Poisson回归模型;聚集性指数的似然比检验(H0:k=0)显示,聚集性指数不等于0具有统计学意义(χ2=1 160.42,P<0.001)。结论对于微核试验数据,拟合负二项回归模型要优于Poisson回归模型。

Objective To compare the fitting results of the Poisson regression model and negative binomial regression model for data of cytokinesis-block micronucleus test,and to provide a basis for statistical analysis of data of cytokinesis-block micronucleus test.Methods By using the log likelihood function,the deviance,Pearson χ2 and cluster index,the fitting results of Poisson regression model and the negative binomial regression model for data of cytokinesis-block micronucleus test were evaluated.Result The ratio of log lielihood function to degree of freedom for negative binomial regression was greater than that for Poisson regression.The ratio of deviance to degree of freedom and the ratio of Pearson χ2 to degree of freedom for negative binomial regression were less than those for Poisson regression.There was a significant difference in cluster index that was not equal to zero for negative binomial regression model（χ2=1 160.42,P0.001）.Conclusion The negative binomial regression model was superior to Poisson regression model for data of cytokinesis-block micronucleus test.