CONFTYPE = LOGLOG or LINEAR in PROC LIFETEST?

[Chaupak]

Hi, I am using PROC LIFETEST to generate the KM curve and its confidence band. I noticed that there is an option called CONFTYPE= (default is LOGLOG) which can let us choose the type of transformation to calculate the confidence band. I also tried CONFTYPE=LINEAR, and the confidence band is different. To me, the results from LINEAR makes more sense, but I am not sure if I can use it. My question is, is there any rule to choose the type of transformation? Is it related to the sample size? 

[OsoGris · Accepted solution]

Linear limits can sometimes go outside the interval (0,1). The CONFTYPE=loglog generally gives better coverage than linear limits and has been the default for a long time. It is not related to sample size.