|
Abstract:
|
Under random censoring model, tied observation supposedly can not happen. However, under general censoring models, or, when both failure time and censoring time are discrete, there can be a significant proportion of tied observations. In this talk, we discuss the approaches to dealing with tied observations in estimating survival function. For a noninformative model with physical background, we verify the validity of the standard failure-before-censoring method. In the case when both of failure and censoring time are discrete, we give out a necessary and sufficient condition on the identifiability of survival function and derive several corresponding estimators associated with different observing mechanisms.
|
