Asymptotic and finite sample behavior of the time on test estimator under random censorship when lifetimes are not Exponetial |
| |
Authors: | Blossom H Patterson Paul J Smith |
| |
Institution: | 1. Biometry Branch , National Cancer Institute , Bethesda, Maryland;2. Department of Mathematics , University of Maryland , College Park, Maryland |
| |
Abstract: | The robustness of the time on test estimator of mean life is studied in both asymptotic and finite sample situations under random censorship. The estimator is shown t o be asymptotically normal and generally in consistent , unless the life time sare exponential . The limiting value of the estimator depends on both the life time and censorship distributions . A simulations tudy of finite sample behavior shows that biases a reslight under exponentiality and serious if exponentia lity is viol at ed . The finite sample behavior is not well described by the limiting normal distribution . Jackknifing produces a useful variance estimate, but is of little value in bias correction. |
| |
Keywords: | asymptotic normality consistency jackknifing life testing mean lifetime robustness simulation variance estimation |
|
|