MENGATASI HETEROSKEDASTISITAS PADA REGRESI DENGAN MENGGUNAKAN WEIGHTED LEAST SQUARE

In the regression analysis we need a method to estimate parameters to fulfill the BLUE characteristic. There are assumptions that must be fulfilled homoscedasticity one of which is a condition in which the assumption of error variance is constant (same), infraction from the assumptions homoskedastic...

Full description

Saved in:

Bibliographic Details
Published in:	E-jurnal matematika Vol. 4; no. 1; pp. 20 - 25
Main Authors:	MAZIYYA, PUTU AYU, SUKARSA, I KOMANG GDE, ASIH, NI MADE
Format:	Journal Article
Language:	English
Published:	Universitas Udayana 30-01-2015
Subjects:	heteroscedasticity OLS WLS
Online Access:	Get full text
Tags:	Add Tag No Tags, Be the first to tag this record!

Description
Summary:	In the regression analysis we need a method to estimate parameters to fulfill the BLUE characteristic. There are assumptions that must be fulfilled homoscedasticity one of which is a condition in which the assumption of error variance is constant (same), infraction from the assumptions homoskedasticity called heteroscedasticity. The Consequence of going heteroscedasticity can impact OLS estimators still fulfill the requirements of not biased, but the variant obtained becomes inefficient. So we need a method to solve these problems either by Weighted Least Square (WLS). The purpose of this study is to find out how to overcome heteroscedasticity in regression with WLS. Step of this research was do with the OLS analysis, then testing to see whether there heteroscedasticity problem with BPG method, the next step is to repair the beginning model by way of weighting the data an exact multiplier factor, then re-using the OLS procedure to the data that have been weighted, the last stage is back with a heteroscedasticity test BPG method, so we obtained the model fulfill the assumptions of homoskedasicity. Estimates indicate that the WLS method can resolve the heteroscedasticity, with exact weighting factors based on the distribution pattern of data seen.
ISSN:	2303-1751 2303-1751
DOI:	10.24843/MTK.2015.v04.i01.p083