Detection of influential observations for the regression model in the presence of multicollinearity: Theory and methods
İssam Dawoud, Hussein Eledum
Abstract
Influential observations constitute a significant threat that affects the regression model’s performance. So, many influential statistics, such as Cook’s Distance and DFFITS have been provided using ordinary least squares. Another problem is multicollinearity, which affects the efficiency of these measures. However, these problems can coexist in the regression model. Therefore, in this study, new diagnostic measures based on the Dawoud-Kibria estimator (DKE) are proposed as alternatives to the available ones. The Cook’s distance, as well as the DFFITS, are introduced for the DKE. Additionally, the above measures and approximate formulas are proposed for the DKE. Two real-world applications and a simulation study are conducted to evaluate the performance of the methodologies presented in this study.