Estimating IFRA (Increasing Failure Rate Average) Based on Censored Data

摘要

This paper considers the problem of estimating a survival curve from randomly right censored data when it is known to have increasing Failure Rate Average (IFRA), or to be New Better than Used (NBU). Let F sub n(t) be the product-limit estimator (PL-estimator) of Kaplan and Meier for the life distribution. Since F sub n(t) never has the IFRA property and may not be NBU, we modify F sub n(t) to have the desired IFRA (NBU) properties. The modified estimators are easy to compute and, under mild conditions, are shown to be asymptotically n raised to the 1/2 power-equivalent to F sub n(t) on compact intervals. Thus the modified estimators share the asymptotic properties of the PL-estimator F sub n(t). (Author)