Model Robust Geographically Weighted Regression pada Data Kemiskinan di Sulawesi Selatan Tahun 2019

Abstract

Geographically Weighted Regression (GWR) is a method of spatial analysis that can be used to perform analysis by assigning weights based on the geographical distance of each observation location and the assumption of having spatial heterogenity. The result of this analysis is an equation model whose parameter values apply only to each observation location and are different from other observation locations. However, when there are outliers at the observation location, a more robust estimation method is needed. One of the robust methods that can be applied to the GWR model is the Least Absolute Deviation method. In this study, model estimation was carried out on the factors that affect poverty in South Sulawesi in 2019 using Robust Geographically Weighted Regression (RGWR) with the Least Absolute Deviation (LAD) method. Determination of weighting is done by using the adaptive kernel bisquare weighting function. The results obtained are RGWR models which are different and apply only to each district/city in South Sulawesi. In addition, it was also found that the RGWR model with the LAD method was the best model for data that experienced spatial heterogenity and contained outliers.