Thesis Type: Doctorate
Institution Of The Thesis: Gazi University, Fen Bilimleri Enstitüsü, İSTATİSTİK ANABİLİM DALI, Turkey
Approval Date: 2015
Thesis Language: Turkish
Student: IRMAK ACARLAR
Supervisor: HAMZA GAMGAM
Open Archive Collection: AVESIS Open Access Collection
Abstract:Influential observations in data cause some significant differentiation on Least Square Estimations of model parameters in multiple linear regression. These observations would also cause increasing in the variances of estimations. Therefore, results of statistical inferences would be insignificant. Since these notable differentiations decrease the explicable of the model, the detection of the influential observation or the influential groups in data is important for the efficiency of regression analysis. The importance of detecting influential observations and some studies about these topics in literature are referred in the first section of this study. Some basic concepts as multiple linear regression model, leverage value, some types of residuals, influence curve and its sample versions are given in the second section of the study. Then some diagnostic methods in order to detect influential observations in literature are demonstrated in the third section. In the fourth and fifth sections a graphical method based on sample influence curve and a graphical method based on Biplot method are proposed to detect influential observations respectively. Two applications of classical methods and the two graphical methods proposed in this study are included in the sixth section. In the seventh section a simulation study is given in order to compare the classical diagnostics and the two graphical methods in this study. At the end of this study, conclusion and recommendations are presented.