Modeling Poverty Levels in West Java Using Generalized Linear Models with Poisson and Negative Binomial Distributions

Rachel Keshia Lovianna

Ambidextrous : Journal of Innovation, Efficiency and Technology in Organization
Vol. 3 No. 02 (2025): Ambidextrous: Journal of Innovation, Efficiency and Technology in Organization

Poverty in West Java Province remains a major challenge with 3.67 million poor people in 2024. This study aims to model the determinants of poverty using the Generalized Linear Model (GLM). This quantitative study uses secondary data from BPS from 27 districts/cities in West Java in 2024 (population census). The instrument is a data extraction sheet with the dependent variable being the number of poor people and independent variables including unemployment, education, GRDP, sanitation, and infant mortality. Analysis techniques include multicollinearity tests, Poisson GLM, Negative Binomial GLM, and model selection based on AIC. The results show that the Negative Binomial model (AIC=305.80) is better than the Poisson (infinite AIC) due to overdispersion. Significant variables are the average length of schooling (β=-0.384, p<0.001) which reduces poverty and infant mortality (β=0.523, p<0.001) which increases poverty. Conclusion: Policy priorities on education and maternal-child health are effective in reducing structural poverty in West Java.