Outlier Detection in Simple Linear  Regression Models and Robust Regression  � A Case Study on Wheat Production Data

B. Vinoth; A. Rajarathinam

Volume : III, Issue : II, February - 2014

Outlier Detection in Simple Linear Regression Models and Robust Regression � A Case Study on Wheat Production Data

B. Vinoth, A. Rajarathinam

Abstract :

In this study, the presence of outliers in wheat production data based on residuals obtained from the fitted simple linear regression model have been studied. Outliers were detected based on the following: i) residual analyses using standardized residuals, studentised residuals, jackknife residuals and predicted residuals; ii) residuals plots such as the graph of predicted residuals, the Williams graph, and the Rankit Q-Q plot; and iii) scalar measures of influence statistics such as cook�s Di (measures the change in the estimates that results from deleting each observation), DFFITSi (measures the change in the predicted value of the dependent variable when the current case is omitted from the calculations), DFBETASj(i) (measures the influence on regression coefficients), Atkinson measures, and the Covariance ratio (measure of model performance). Robust regression was employed to avoid the influence of outliers, and the linear growth rate was calculated based on the fitted robust regression model.