期刊名称：Brazilian Journal of Probability and Statistics
印刷版ISSN：0103-0752
出版年度：2001
卷号：15
期号：2
页码：107-119
出版社：Brazilian Statistical Association
摘要：An important generalization of the proportional hazards mo del is the stratifiedmo del arising when the data for some specified reason are divided into strata.The stratified model is used in situations where an important covariate or covari-ates do not satisfy the proportional hazards assumption. This paper presents asimulation study showing the impact on the maximum partial likelihood estima-tors when the strata are misspecified or an unstratified analysis is performed ina nonproportional hazards situation. The results of this simulation show that ingeneral there is a large bias in the estimate of the regression parameter for thecovariate asso ciated with the nonproportionality when stratification points aremisspecified in the stratified analysis. Also the estimate of this parameter in theunstratified analysis has a large bias when the stratified analysis is the correctprocedure. These results suggest the necessity of guidelines for the division ofthe data into strata when there is an indication in favor of the stratified analysis.A criterion is proposed as guideline for such division
关键词：Cox model; mo del misspecification; stratification criterion; strati-;fied model
