Computationally Efficient Comparison of Experimental Designs for System Reliability Studies With Binomial Data

A common practical problem in experimental design is that of quantitatively determining how to best use a fixed amount of resources to supplement an existing analysis with additional data. We address this problem in the context of the second stage in Bayesian system reliability studies; these second...

Full description

Saved in:
Bibliographic Details
Published in:Technometrics 2012-11, Vol.54 (4), p.410-424
Main Authors: Chapman, Jessica L., Morris, Max D., Anderson-Cook, Christine M.
Format: Article
Language:English
Subjects:
Algorithms
Applied sciences
Bayesian analysis
Binomials
Complex systems
Design of experiments
Estimate reliability
Estimation methods
Exact sciences and technology
Experiment design
Experimental design
Experimentation
Interval estimators
Markov chain Monte Carlo
Mathematics
Missile systems
Operational research and scientific management
Operational research. Management science
Parametric inference
Precision
Probability and statistics
Reliability
Reliability theory. Replacement problems
Resource allocation
Sciences and techniques of general use
Statistics
System reliability
Test data
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:A common practical problem in experimental design is that of quantitatively determining how to best use a fixed amount of resources to supplement an existing analysis with additional data. We address this problem in the context of the second stage in Bayesian system reliability studies; these second-stage data are aimed at obtaining a more precise estimate of the system's reliability. The current strategy for comparing potential experimental designs is computationally intensive and time-consuming. We present new, more computationally efficient methodology that can be used to quickly assess the anticipated improvements for candidate allocations, and demonstrate its effectiveness with a missile system application. While we find that there are some situations in which our methodology may provide a more optimistic estimate of the expected improvement to be gained from a candidate experiment than the current approach, the results of the two methods tend to closely match when the rankings of candidate experiments are considered. Our implementation of the algorithm (in C), along with a brief description and an example input set, is available online as supplementary materials.
ISSN:0040-1706
1537-2723
DOI:10.1080/00401706.2012.694784