Geographically and Temporally Weighted Log-Logistic 3-Parameter Regression Model for Poverty Severity Index : A Case Study on East Java Province

Abstract

This study proposes the Geographically and Temporally Weighted Log Logistic 3 Parameter Regression (GTWLL3R) model as a novel extension of LL3R that simultaneously captures spatial and temporal heterogeneity in poverty severity index. Using the poverty severity index of East Java Province for 2022–2024, local parameters were estimated through an fixed Gaussian kernel weighting matrix based on spatial and temporal distances, with optimization using the Newton–Raphson algorithm. Model performance was evaluated using the corrected Akaike Information Criterion (AICc). The results show that GTWLL3R outperformed the LL3R and GWLL3R models, achieving the lowest AICc value of 18.311, which indicates substantially better model fit and stronger explanatory capability. The estimated coefficients vary across districts/cities and time periods, revealing different patterns of predictor effects on poverty severity index. Based on significant predictor variables, the districts/cities were classified into three clusters. These findings demonstrate that integrating LL3R into the GTWLL3R framework provides a more flexible and accurate approach for analyzing spatiotemporal poverty dynamics and offers stronger evidence for targeted poverty alleviation policies.