Accession Number : ADA466621


Title :   Change-Point Methods for Overdispersed Count Data


Descriptive Note : Master's thesis


Corporate Author : AIR FORCE INST OF TECH WRIGHT-PATTERSON AFB OH GRADUATE SCHOOL OF ENGINEERING AND MANAGEMENT


Personal Author(s) : Wilken, Brian A.


Full Text : http://www.dtic.mil/get-tr-doc/pdf?AD=ADA466621


Report Date : MAR 2007


Pagination or Media Count : 111


Abstract : A control chart is often used to detect a change in a process. Following a control chart signal, knowledge of the time and magnitude of the change would simplify the search for and identification of the assignable cause. In this research, emphasis is placed on count processes where overdispersion has occurred. Overdispersion is common in practice and occurs when the observed variance is larger than the theoretical variance of the assumed model. Although the Poisson model is often used to model count data, the two-parameter gamma-Poisson mixture parameterization of the negative binomial distribution is often a more adequate model for overdispersed count data. In this research effort, maximum likelihood estimators for the time of a step change in each of the parameters of the gamma-Poisson mixture model are derived. Monte Carlo simulation is used to evaluate the root mean square error performance of these estimators to determine their utility in estimating the change point, following a control chart signal. Results show that the estimators provide process engineers with accurate and useful estimates for the time of step change. In addition, an approach for estimating a confidence set for the process change point will be presented.


Descriptors :   *QUALITY CONTROL , *STATISTICAL PROCESSES , *CHANGE DETECTION , MATHEMATICAL MODELS , SIMULATION , PARAMETRIC ANALYSIS , MAXIMUM LIKELIHOOD ESTIMATION , THESES , MONTE CARLO METHOD


Subject Categories : STATISTICS AND PROBABILITY
      MISCELLANEOUS DETECTION AND DETECTORS


Distribution Statement : APPROVED FOR PUBLIC RELEASE