Small Area Estimation of Extreme Poverty Using Zero-Inflated Binomial GLMM: A District-Level Case Study in North Sumatra 2024

Abstract

Eradicating extreme poverty is a key objective of Sustainable Development Goal (SDG) 1, with a global benchmark of reducing the proportion of people living below the US$1.90 PPP poverty line. However, in 2024, Indonesia—particularly North Sumatra Province—continues to face persistent challenges in achieving this target. Direct estimation based on the Foster-Greer-Thorbecke (FGT) formula using SUSENAS microdata suffers from large sampling errors (RSE > 25 percent) and zero estimates in multiple districts due to small or absent samples, indicating serious issues of zero inflation and overdispersion. To overcome these limitations, this study applies a model-based Small Area Estimation (SAE) approach using the Zero-Inflated Binomial Generalized Linear Mixed Model (ZIB-GLMM). This method incorporates auxiliary variables from the 2024 PODES dataset and effectively addresses the dual complexities of excess zeros and inter-district variability. Simulation results show that ZIB-GLMM outperforms conventional SAE models in terms of predictive accuracy and model stability. The proposed method offers realistic and policy-relevant district-level estimates of extreme poverty, providing robust evidence to inform targeted interventions and strengthen Indonesia’s national agenda to eradicate extreme poverty.