Efficient procedure for selecting among five reliability models. [In FORTRAN for IBM 360/91]
Several parametric families of probability distributions are commonly used to model reliability and life testing data. Efficient techniques are needed for the selection of the family of distributions which best describes a given set of experimental observations. A brief survey of methods used for discriminating among competing families of distributions is given. An efficient technique for selecting the best fitting family of distributions is derived. This technique uses scale-invariant statistics that minimize the probabilities of misclassification, and is extended beyond pairwise selections to selecting from a class of families of competing distributions. These statistics are not used in the hypothesis testing framework, but rather as measures of likelihood that the observed data have a distribution from a particular family of probability distributions. The class of families considered are the Uniform, Exponential, Gamma, Weibull, and Lognormal distributions, for the cases of shape parameters both known and unknown. Results of computer simulations are given for pairwise and multiple family selection procedures, showing the misclassification error rates achieved for selected sample sizes. Illustrations based on actual data are presented, along with a user program for convenient implementation of this procedure. 16 figures, 17 tables.
- Research Organization:
- Oak Ridge National Lab., TN (USA)
- DOE Contract Number:
- W-7405-ENG-26
- OSTI ID:
- 5968182
- Report Number(s):
- ORNL/CSD-43
- Country of Publication:
- United States
- Language:
- English
Similar Records
Can Data Recognize Its Parent Distribution?
Frequency distributions of trace metal concentrations in five freshwater fishes