A Bayesian method to investigate group of observations which is in outlier in multivariate-multiple linear regression analysis

A BAYESIAN METHOD TO INVESTIGATE GROUP OF OBSERVATIONS WHICH IS AN OUTLIER IN MULTIVARIATE-MULTTPLE LINEAR REGRESSION ANALYSIS (Dr.Thesis) Ufuk EKİZ GAZİ UNIVERSITY INSTUTE OF SCIENCE AND TECHNOLOGY May 2003 ABSTRACT In multivariate-muLtiple regression analysis, there are many methods in the literature to search whether there are outliers or not in the sample. Bayesian methods are used to and summarized in three groups. Varbanov(1998) proposed a bayesian method to investigate outlier observation individually. He proposed a posterior distribution for the mean squared error from of unobserved but realized error. To handle observations individually it is not satisfactory investigate masking and swamping problems. To show the presence of this problems, it is necessary to handle the observations as group to see whether the groups are outliers or not. Here, Varbanov's method is expanded to the multivariate-multiple regression and the expanded method is applied to the masking and swamping problems. Science Code : 406.01.01 Key Words ; Realizeted but unobserved error, Masking and Swamping Problems, outlier observation Page number : 85 Adviser : Prof.Dr.Müslîm EKNÎ