Testing constant failure rate against NBAFR alternatives with randomly right-censored data

Reliability analysts and biometricians have found it useful to categorize life distributions by the properties of the failure rate. This paper considers the problem of testing exponentiality vs (nonexponential) new better than average failure rate (NBAFR) alternatives. Often, in practice, the data are incomplete because of: (a) withdrawals from the study; and (b) survivors at the time the data are analyzed. The authors propose a test statistic based on a function of the Kaplan-Meier estimator to accommodate randomly right censored data. The asymptotic efficacy of the test is derived and the efficiency loss due to censoring is studied. This test is applied to published survival data, and to simulated data