Semi-parametric Geographically Weighted Regression Modelling using Linear Model of Coregionalization

Abstract

Geographically Weighted Regression is a weighted analysis regression for local or spatially varying parameters, therefore each location has different regression parameters. In its application, one often finds a condition that needs some global parameters. Geographically Weighted Regression that has local and global parameters is called Semi-parametric Geographically Weighted Regression. This study modelled Semi-parametric Geographically Weighted Regression using Linear Model of Coregionalization to assist spesification of local and global parameters. Linear Model of Coregionalization represented spatial variability proportion at different spatial distances and spatial dependence of parameters. High spatial dependence variables were as local parameters while the other variables were as global parameters. The data used was poverty data in North Sulawesi Province. The results of Geographically Weighted Regression and Semi-parametric Geographically Weighted Regression models were compared based on Akaike Information Criterion Corrected and Mean Square Prediction Error. It showed that Semi-parametric Geographically Weighted Regression model was better than Geographically Weighted Regression.