Department of Quantitative Economics, Maastricht University, PO Box 616, NL-6200 MD Maastricht, The Netherlands. e-mail: firstname.lastname@example.org
NONPARAMETRIC META-ANALYSIS FOR MINIMAL-REPAIR SYSTEMS
Article first published online: 16 DEC 2010
© 2010 Australian Statistical Publishing Association Inc.
Australian & New Zealand Journal of Statistics
Volume 52, Issue 4, pages 383–401, December 2010
How to Cite
Beutner, E. and Cramer, E. (2010), NONPARAMETRIC META-ANALYSIS FOR MINIMAL-REPAIR SYSTEMS. Australian & New Zealand Journal of Statistics, 52: 383–401. doi: 10.1111/j.1467-842X.2010.00591.x
- Issue published online: 27 DEC 2010
- Article first published online: 16 DEC 2010
- nonhomogeneous Poisson process;
- nonparametric estimation;
- nonparametric prediction intervals
We consider the situation that repair times of several identically structured technical systems are observed. As an example of such data we discuss the Boeing air conditioner data, consisting of successive failures of the air conditioning system of each member of a fleet of Boeing jet airplanes. The repairing process is assumed to be performed according to a minimal-repair strategy. This reflects the idea that only those operations are accomplished that are absolutely necessary to restart the system after a failure. The ‘after-repair-state’ of the system is the same as it was shortly before the failure. Clearly, the observed repair times contain valuable information about the repair times of an identically structured system put into operation in the future. Thus, for statistical analysis and prediction, it is certainly favourable to take into account all repair times from each system. The resulting pooled sample is used to construct nonparametric prediction intervals for repair times of a future minimal-repair system. To illustrate our results we apply them to the above-mentioned data set. As expected, the maximum coverage probabilities of prediction intervals based on two samples exceed those based on one sample. We show that the relative gain for a two-sample prediction over a one-sample prediction can be substantial. One of the advantages of the present approach is that it allows nonparametric prediction intervals to be constructed directly. This provides a beneficial alternative to existing nonparametric methods for minimal-repair systems that construct prediction intervals via the asymptotic distribution of quantile estimators. Moreover, the prediction intervals presented here are exact regardless of the sample size.