Perbandingan Metode Least Trimmed Squares dan Penaksir M dalam Mengatasi Permasalahan Data Pencilan

Abstract

Regression analysis is used to determine the relationship between variables. One of methods for estimating the parameters in model analysis is ordinary least square (OLS). If there are outliers, OLS is not efficient again so the suitable method for problems of outliers is robust regression method. Outlier is data that inconsistent with the pattern and located away from the data center, can be detected with graphical method and determine the leverage value, DfFITS and Cook’s Distance. Least trimmed squares (LTS) is an estimating method of robust regression that using a fitting concept of OLS to minimize the sum square error. M estimator is a method to overcome the outliers and can use Huber function in estimating the regression parameter. The purpose of this study is comparing two methods of robust regression, those are LTS and M estimator with ordinary least squares method in overcoming the problems of outlier. The conclutions of it are LTS is the best method because it can overcome the outliers and give a good estimation in coeficient of regression, and so produce the smallest mean square error. Then, M estimator also gives a good estimation and produce smaller mean square error than OLS.