Geographically Weighted Logistic Regression Model (Case Study on Poverty Modelling in East Java Province).
Model Regresi Logistik Terboboti Geografis (Studi Kasus : Pemodelan Kemiskinan di Provinsi Jawa Timur)
Abstract
Geographically Weighted Logistic Regression (GWLR) model is locally logistic regression model. The data in this model is assumed following Binomial distribution and the geographical factor is considered. The geographical factor is used to analyze spatial data from nonstationary processes. The basic idea of this model consider the geography or location as the weight in parameter estimation. The parameter estimator is obtained from Iteratively Reweighted Least Square method by giving different weight for different location. Model for determining the poverty level with a global logistic regression was not suitable to be applied in all districts of East Java Province, because it could be a predictor effect on the poverty level in the region but in other regions the predictor is not significant. The data in this research is from National Social Economy Survey 2008. This research will determine the factors that affect the poverty level in the East Java Province using logistic regression model and GWLR model with a weighting adaptive bi-square kernel function. The results showed that the classification accuracy of logistic regression model was 78.90% and the classification accuracy of GWLR model was 89.47%. GWLR model with a weighting adaptive bi-square kernel function was better than logistic regression model because it had the high classification accuracy and small AIC value.