MODELING GEOGRAPHICALLY WEIGHTED BETA REGRESSION ON POVERTY DATA IN INDONESIA

Abstract

Poverty is a major development issue influenced by socioeconomic factors and regional disparities. Previous studies on poverty in Indonesia have largely employed global regression approaches that assume homogeneous relationships between explanatory variables and poverty across regions. This assumption may overlook spatial heterogeneity, resulting in less accurate estimates and limited relevance for local policymaking. Therefore, this study applies Geographically Weighted Beta Regression (GWBR) to model the proportion of the poor population in Indonesia and compares its performance with global beta regression. The independent variables are the Human Development Index, Open Unemployment Rate, Mean Years of Schooling, Life Expectancy, Literacy Rate, and Access to Safe Drinking Water. Parameter estimation was performed using Maximum Likelihood Estimation (MLE), while statistical inference was conducted through simultaneous and partial significance tests. Spatial dependence was assessed using Moran’s I statistic. The main contribution of this study is the identification of spatially varying determinants of poverty using the GWBR approach, providing localized insights for more targeted poverty alleviation policies. The results indicate significant spatial autocorrelation in the residuals of the global beta regression model, suggesting that spatial effects should be considered in poverty analysis. Compared with the global beta regression model, GWBR demonstrated superior performance, yielding lower values of AIC (-909.462), AICc (-905.439), and MAAPE (0.218). Local parameter estimates revealed substantial spatial heterogeneity in the determinants of poverty. Life Expectancy was the most consistent factor, exhibiting a significant negative effect in 34 provinces, while Mean Years of Schooling showed a significant negative effect in 15 provinces. The effects of the remaining variables varied across regions, indicating that the determinants of poverty differ spatially across Indonesia. These findings suggest that GWBR more effectively captures local variations in the determinants of poverty and supports the development of region-specific poverty alleviation policies in Indonesia.