Field Failure Data Analysis With Discard Rate

Product units in service may be discarded even before they have failed. This paper discusses the need to consider such discarded units in analyzing eld failure data. Approaches based on maximum likelihood (ML) and conventional cumulative hazard (CH) methods are described and evaluated for their accuracy in estimating the parameters of failure time distributions. Simulation results showed that the ML method performs better than the CH method, and indicated that the CH method has a considerable bias. We also propose some nonparametric estimators that provide a way of improving the precision of the CH method.