Ways To Eliminate The Outliers Effect On The Linear Regression Model

ABSTRACT

The research dealt with the elimination of the outliers effect on results

of fitting the linear regression models. Also the research aims to arrive at

a reliable detecting and treating ways which described as competency,

unbiased, and usability. So a proposed new ways, which included a way

for detecting outliers observation followed by a graphical measure so as

to determine the outliers size, and a way to treat the outliers values

substituted them with new ones. These ways has been applied on 35

simple linear regression models, and 31 multi linear regression models

for testing the hypothesis concerning in the proficiency ratio and the

biasness naught of the proposed ways in eliminating the outliers effect on

results of fitting the linear regression models, and the hypothesis

concerning of estimating the outliers numbers in the linear regression

model. The research depended on the inference method to estimate the

rational proficiency means, and the comparing method for the parameters

means that witch obtained from among the practice of model fitted by

SPSS, and the Microsoft Exile Program in calculating the mathematical

functions of the proposed ways. The important results in this research

showed the proposed ways netted a high proficiency in detecting and

treating outliers, noted the enhancement of the linear model parameters

as R & F 2 value raised and MSE value diminishes, with addition to

that proposed ways appears more efficiency and more unbiased than the

other previous ways. The important recommendation of this research is to

see if it is available to include the proposed ways in SPSS program to be

as an updating for the part concerning in the linear regression analysis method.