Confidence Intervals on Hazard Plots
On Sep 5, 2008, at 9:20 AM, Alan Cox wrote:
Hello, Is it possible to create confidence intervals for hazard rates? I'm creating two muhaz objects: haz1 <- muhaz(NumDaysCustomer[cRV=="true"],status[cRV=="true"]) haz2 <- muhaz(NumDaysCustomer[cRV=="false"],status[cRV=="false"]) and plotting them. There are many, many more observations in the cohort cRV=="false" than =="true". And, there are many more observations with lifetimes in the middle of the range than at the ends. I suspect that this is a common phenomenon. When I plot the two hazard rate curves, haz1 looks very different than haz2, but I'd like to see if it's different enough. How can I go about creating confidence intervals and plotting them?
?muhaz.object says that var.min and bias.min could be extracted from haz1 and haz2, since you have not changed the default for bw.method. The default haz.est vector is twice as long as the default var.min vector, so you will need to take care to only plot the haz.est values only for the the event times, or perhaps you could specify that n.est.grid be the same as n.min.grid at the time of creating your fits. Not sure if the second method might have any undesirable side- effects, however.
David Winsemius Heritage Laboratories