摘要
目的探讨用于传统回归与零膨胀回归模型选择的Vuong检验(简称V检验)SAS实现方法及应用。方法介绍V检验的原理及方法并给出实现V检验的SAS宏程序,结合实例,将SAS宏的检验结果与Stata软件进行核对,证明宏程序的正确性。结果应用实例中ZINB模型-2Log Likelihood为44 623,回归模型选择标准(Akaike Information Criterion)为44 665,BIC(Bayesian Information Criterion)为44 819,相比其他模型拟合效果最好;SAS宏程序对ZINB与负二项回归模型V检验结果为30.326 9,>1.96,数据分析应采用ZINB模型;Stata软件的输出V值为30.33,和宏程序检验结果一致。结论 V检验用于判断传统回归和零膨胀回归模型拟合的优劣;SAS宏程序可以实现V检验。
Objective To explore the clinical application and SAS implementation method of Vuong test （V-test） in selecting a more suitable model between traditional used model and zero-inflated model. Methods Firstly, principle and method of Vuong test were introduced, and then SAS macro of Vuong test and its direction for use were presented. Fi- nally, we compared the results of SAS macro for Vuong test with result of Stata software demonstrating correctness of the SAS macro. Results The ZINB regression model had a better fitting effect in the clinical example, -2 Log Likelihood = 44 623, AIC（Akaike Information Criterion） =44 665, BIC（Bayesian Information Criterion） =44 819, which implied data was negative binomial distributed and zero inflated. Using SAS macros to have Vuong tests between ZINB and negative bi- nomial regression model, V value was 30. 326 9, which was larger than 1.96. It proved again that the clinical data exist ze- ro inflated and ZINB was more suitable to analyze it. V value from Stata software was 30. 33, it was equal to the result crea- ted by SAS macros. Conclusions V test can quantitatively judge which model is more suitable between the traditional used regression and zero - inflated regression. V test can be implemented both in SAS macro procedure and Stata software.
出处
《中华疾病控制杂志》
CAS
北大核心
2013年第7期613-616,共4页
Chinese Journal of Disease Control & Prevention
关键词
生物医学研究
数据说明
统计
软件
Biomedical research
Data interpretation, statistical
Software
