A simple non-parametric test for decreasing mean time to failure (Q2633416)

The authors propose and study a new test for the hypothesis of an exponentially distributed lifetime against the alternative of a decreasing mean time to failure. The test is based on the observation that the expectation of minimum of two exponential distributed random variables is half the expectation of the individual random variables. The test statistic measures the difference between the expectations and is expressed as a bivariate U-statistic. The authors provide the exact distribution of the test statistic under the null hypothesis for a finite sample for independent data as well as the asymptotic distribution. They also calculate the Pitman efficiency for different families of distributions under the alternative and compare it with the efficiency of tests previously proposed by other authors. Furthermore, the case of censored observations is discussed. The authors provide a Monte-Carlo study of the test performance and apply it to data set previously discussed in the literature.