Dear List,
Is the test for proportional hazards valid when the model contains a cluster
variable? The output looks strange with the cluster variable.
My intervals are based on calendar time and the clustering variable is
related to the season the event occurs in.
model1<-coxph(Surv(Start,Stop,Event)~LagAOO+I(LagAOO^2)+
FirstSeen+TSLE+strata(CumPOO.rc)+cluster(quarter),data=data8, x=T)
>Cox.zph(model1)
LagAOO 0.209 5.61 0.0179
I(LagAOO^2) -0.209 5.61 0.0179
FirstSeen 0.209 5.61 0.0179
TSLE 0.209 5.61 0.0179
GLOBAL NA 5.61 0.2304
In this example there is strong evidence for non-proportional hazards for
each of my covariates.
By defining my strata differently I can remove the non-proportional hazards.
A quick comment is appreciated.
Alex
[[alternative HTML version deleted]]
