scientific article; zbMATH DE number 1058146

From MaRDI portal
Publication:4353265





A recurrent theme in reliability research is the analysis of distributions with unknown endpoints. Specific examples are the three-parameter Weibull, lognormal, gamma and inverse Gaussian distributions. Each of these distributions has the property of being concentrated on an interval \((\gamma,\infty)\), where \(\gamma\), typically called an endpoint or threshold parameter, is unknown. In most cases this is non-negative and is interpreted as the minimum possible lifetime of the system, a parameter of obvious importance in reliability analysis. In such families it has been known for a long time that there are difficulties with maximum likelihood estimation. A. C. Cohen was one of the first to consider maximum likelihood estimation (MLE) for distributions of this type, and over the years has made many contributions. However, he has also been one of the pioneers in considering alternative methods. He has developed a number of variants of a method which he called modified maximum likelihood estimation (MMLE), and in recent years has also worked on modified moments estimators (MME). The main motivation for developing these alternative methods has apparently been ease of numerical implementation. However, it has also been suggested in a number of simulation studies that they may have superior statistical properties as well, in the sense of having smaller bias or variance in small or moderate-sized samples. In this paper I want to explore these aspects further, with particular attention to a theoretical comparison of the MLE and MMLE techniques. For the entire collection see [Zbl 0847.00014].











This page was built for publication:

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4353265)