ML vs. MRR: Weibull Parameter Estimation for Making Decisions

• Activity Number: 649
• Type: Contributed
• Date/Time: Thursday, August 7, 2014 : 10:30 AM to 12:20 PM
• Sponsor: Section on Physical and Engineering Sciences
• Author(s): Andrew Robinson*+ and Nicholas Armstrong
• Companies: University of Melbourne and Defence Science and Technology Organisation
• Keywords: Weibull ; maximum likelihood ; median rank regression ; hazard

Abstract

We present a comparison of the performance of maximum likelihood estimation (ML) and median rank regression (MRR) for parameter estimation for the Weibull distribution, using innovative metrics and realistic scenarios. The performance of these estimators has been compared previously using simulation experiments, but comparisons have been based on the properties of the parameter estimators or quantile estimators. We compare the estimators from a decision point of view, using 10e-5 and 10e-6 hazard cutoffs as suggested by some industrial applications, and show that MRR is preferred under a wide range of circumstances.