A REDESCENDING M-ESTIMATOR FOR DETECTION AND DELETION OF OUTLIERS IN REGRESSION ANALYSIS

SOURCE:

Faculty: Physical Sciences
Department: Statistics

CONTRIBUTORS:

Anekwe, S. E.
Onyeagu, S. I.

ABSTRACT:

M-estimators are robust estimators that give less weight to the observations that are outliers while Redescending M-estimators are those estimators that are built such that extreme outliers are completely rejected. Several researchers proposed different methods of M-estimator and Redescending M-estimators for detection and deletion of outliers as discussed in the literature. However, there is still need to have a Redescending M-estimator that will be more efficient and robust when outliers are in both two-dimensional space compared with the existing ones. In view of this, a Redescending M-estimator is proposed while its objective, influence and weight functions are established.The proposed method is applied to different examples (real-life data) to verify its effectiveness in detecting and deleting outliers. The Monte Carlo simulation method is used to investigate the performance of the newly proposed method. The results from the simulation study and the real life data indicate that the proposed method is very good for detecting and deleting outliers. Furthermore, the proposed method is particularly more efficient and robust when outliers are in both x- and y-directions compared to the existing ones.