Date of Award


Document Type


Degree Name

Master of Science


Department of Mathematics and Statistics

First Advisor

John S. Crown, PhD

Second Advisor

Albert H. Moore, PhD


The Weibull distribution finds wide applicability across a broad spectrum of disciplines and is very prevalent in reliability theory. Consequently, numerous statistical tests have been developed to determine whether sample data can be adequately modeled with this distribution. Unfortunately, the majority of these goodness-of-fit tests involve a substantial degree of computational complexity. The study presented here develops and evaluates a new sequential goodness-of-fit test for the three-parameter Weibull distribution with a known shape that delivers power comparable to popular procedures while dramatically reducing computational requirements. The new procedure consists of two distinct tests, using only the sample skewness and sample kurtosis as test statistics. Critical values are derived using large Monte Carlo simulations for known shapes k=0.5(0.5)4 and sample sizes n=5(5)50. Attained significance levels for all combinations of the two tests between alpha=0.01(0.01)0.20 are also approximated with Monte Carlo simulations and presented in a useful contour plot format. Extensive power studies against numerous alternate distributions demonstrate the test's excellent performance compared to popular EDF test statistics such as the Anderson-Darling and Cramer-von Mises tests. Recommendations are included on techniques to choose significance levels of the two component tests in a manner that should optimize power while maintaining the overall significance level.

AFIT Designator


DTIC Accession Number