Failure Rate Analysis Using GLIMMIX

摘要

GLIMMIX is a SAS macro that provides analysis of data modeled by a generalized linear mixed model, GLMM, based on an approach in Wolfinger and O'Connell (1993) for maximizing the associated likelihood. We use this computing tool in the special case of Poisson failure counts for data on catastrophic time-dependent pump failures obtained from Martz and Whiteman (1984). The original analysis of this data, assuming a GLMM, was performed using Failure Rate Analysis code (FRAC), a code developed at Los Alamos National Laboratory to perform an analysis of variance of failure rates as described in Martz, Beckman and McInteer (1982). The FRAC approach to estimating the effects in a GLMM is basically fitting a linear mixed model to a transformed response variable. The GLIMMIX and FRAC approaches to GLMM estimation are described , delineating differences. The analysis results of these two computational approaches are compared for the pump failure data.