Hier finden Sie wissenschaftliche Publikationen aus den Fraunhofer-Instituten.

A Bayesian logistic regression model for binomial failure data

: Kempf, Michael

Fulltext urn:nbn:de:0011-n-2196882 (77 KByte PDF)
MD5 Fingerprint: 46195a96c25208247fa03efa3c110669
Created on: 22.11.2012

Santos, Daryl L. (Ed.) ; Society for Industrial and Systems Engineering -SISE-:
1st Annual World Conference of the Society for Industrial and Systems Engineering 2012. Proceedings : Washington, D.C., USA, September 16-18, 2012
Binghamton/NY, 2012
ISBN: 9781938496004
Society for Industrial and Systems Engineering (SISE Annual World Conference) <1, 2012, Washington/DC>
Conference Paper, Electronic Publication
Fraunhofer IPA ()
Fehlerdaten; Bayesian statistic; regression; reliability; Zuverlässigkeit; statistisches Verfahren; Fehleranalyse

The reliability assessment of systems which are relevant to security is an extremely important task in managing public health risk. We want to consider the failure probability p of such a critical system and find out whether there is a trend in this failure probability p over time. There are data available about unplanned demands for maintenance because of hazardous incidents in terms of incident times. Furthermore it has been recorded, whether or not a failure occurred shortly after the demand. As the failure probability obviously depends on these data it seems reasonable to find a statistical model covering these dependencies. Since we have binary outcomes, we use the logistic regression model. Here the logit function is used, which maps the odds ratio p/(1-p) of p via log(p/(1-p)) onto the real line. This logit function will be related to the times t , where the hazardous incidents occurred via log it( p ) = ß01t . If there really exists a bias for the failure probability p, the parameter ß1 should be nonzero.