• BAYESIAN AND NON-BAYESIAN ANALYSIS FOR RANDOM CHANGE POINT PROBLEM USING STANDARD COMPUTER PACKAGES

Abstract

Change point problems are used in many application fields. A general approach to Bayes, hierarchical Bayes, and duplicate the results using non-Bayes approach, for random change point problem are presented. In Bayesian analysis, Markov Chin Monte Carlo (MCMC) methods have become a ubiquitous tool as the computer is more powerful. Although Spiegelhalter et al. 1996 propose code software using WinBUGS (standard Bayesian software’s package freely available, Spiegelhalter et al. 2003) to detect the change point of normal datasets using WinBUGS, their results yield not investigated well. We argue that a more realistic description of random change point problem as will be resolved in this article in some details and duplicate these results from non-Bayesian perspective using some of R packages (R Development Core Team, 2011). Because of the statistical performances of the Bayes estimates for the random change points are so sensitive for the prior elicitation that we should duplicate our results using the non-Bayesian approach. Within these contexts we handle two textbook data examples to compare the results between two approaches, Bayesian and non-Bayesian statistics.