Improving Statistical Rigor in Defense Test and Evaluation: Use of Tolerance Intervals in Designed Experiments.

摘要

Leveraging the use of statistical methods is critical in providing defensible test data to the Department of Defense Test and Evaluation (T&E) enterprise. This article investigates statistical tolerance intervals in designed experiments for the T&E technical community. Tolerance intervals are scarcely discussed in extant literature as compared to confidence/prediction intervals. The lesser known tolerance intervals can ensure a proportion of the population is captured in the design space, and have the ability to map the design space where factors can be reliably tested. Further, the article investigates several two-sided approximate tolerance factors estimated by Monte Carlo simulation and compares them to the exact method. Finally, the applicability of tolerance intervals to the defense T&E community is presented using a simple case study.