Implementation of Geographically Weighted Regression (GWR) and Mixed Geographically Weighted Regression (MGWR) in The Calculation of Indigent Population in Central Java Rahmi Novika Harahap, Achmad Fauzan
Statistics Department,
Faculty of Mathematics and Natural Science,
Universitas Islam Indonesia
Indonesia
Abstract
One of the provinces with the highest number of indigent people in Indonesia in 2020 is Central Java. The different characteristics of each region in Central Java Province give rise to spatial heterogeneity. The existence of spatial heterogeneity makes using multiple linear regression methods inappropriate. Therefore, the Geographically Weighted Regression (GWR) method will be used in this study. On the other hand, not all independent variables have an effect locally, but there are independent variables that have a global effect or a combination of independent variables that have a global and local effect. Mixed Geographically Weighted Regression (MGWR) is a method that generates global and local parameters. This study aims to obtain the best model from data on the number of poor people in Central Java Province in 2020 based on three different models: multiple linear regression modeling, GWR, and MGWR. The results of the MGWR analysis show that both global variables have a significant effect. At the same time, local variables with a significant effect differ in each region, and the model formed will also differ for each region. Nevertheless, the best model formed is the model with the GWR method of the fixed gaussian weighting function, which is shown from the goodness-of-fit model values, namely MAPE of 6.96, MSE of 88.74, RMSE of 9.42, AIC of -13.20, and adjusted-R2 of 97.54.
