Spatial Modeling of Highly Dispersed Poverty Data in East Java Using the Geographically Weighted Negative Binomial Regression Method

Abstrak

Poverty in East Java Province is a complex issue characterized by diverse regional characteristics. This study aims to model the number of poor inhabitants in East Java while accounting for overdispersion and spatial heterogeneity using the Geographically Weighted Negative Binomial Regression (GWNBR) method with four kernel weighting functions. The results indicate that the Negative Binomial regression model performs well at the global level, with an AIC of 941.01. However, to capture local variation in poverty drivers, the GWNBR model with an Adaptive Bisquare kernel proved optimal, yielding the lowest AIC of 309.32. These findings confirm the diversity of predictor variable influences across regencies and cities, as evidenced by the significance of variables in Blitar Regency. The GWNBR approach provides more accurate local parameter estimates than the global model, serving as a strategic tool for the government to design more targeted and effective poverty alleviation policies for each region.