Determination of the best geographic weighted function and estimation of spatio temporal model – Geographically weighted panel regression using weighted least square

Abstract

This study proposes the development of a spatio-temporal model with geographic weights containing elements of location, time and the correlation between the two. The spatio-temporal model is a spatial regression model that combines geographic information and time series simultaneously. The model can overcome the problem of spatial heterogeneity and spatial effects. The spatial temporal model used is the Geographically Weighted Panel Regression (GWPR) model with a within estimator. Therefore, it is necessary to determine the best geographic weighting with the optimal bandwidth value and the lowest Cross Validation (CV). The geographic weights used were the Gaussian kernel function, the Bisquare kernel function and the exponential kernel function. Estimation of spatio-temporal model parameters using Weighted Least Square (WLS). The GWPR model was applied to food security index data in 34 Indonesian provinces. The problem of food security is an important problem to be solved in Indonesia, one way is to find the factors that influence the food security index through spatio-temporal modeling. This study consists of data exploration, descriptive statistics, spatial mapping distribution, selection of geographic weights and GWPR modeling. The results showed that the spatio temporal statistical model of GWPR was more accurate with a good model of 92.78 % and a Root mean Square Error value of 3.41. Some highlights of the proposed approach are:

Keywords

• Geographically Weighted Panel Regression