Spatial Determinants of Stunting in East Java: A Comparative Assessment between OLS and Spatial Models Approach

Abstract

Stunting remains a critical challenge in Indonesia, aligning with the 2030 SDGs. This study examines the spatial patterns of stunting prevalence across 29 districts and 9 cities in East Java Province using 2022 data. Classic Ordinary Least Squares (OLS), Spatial Autoregressive (SAR), and Spatial Error Models (SEM) were deployed. Baseline OLS shows that Gender Development Index, Poverty, GRDP per Capita, Access to Adequate Sanitation, and nurse density simultaneously exert a significant joint effect on stunting. Although Moran's I indicated marginal evidence of positive spatial autocorrelation in the OLS residuals (p = 0.0556), the SAR and SEM specifications yielded non-significant spatial parameters (ρ and λ) and no meaningful AIC improvement over OLS. Local Indicators of Spatial Association (LISA) further show that, unlike sanitation access, stunting itself does not form a statistically significant High-High cluster, suggesting that the observed residual spatial dependence may be partly accounted for by the included structural covariates, although the study's small sample size (n = 38) may also limit the power to detect spatial effects directly. Consequently, the classical OLS specification was retained as the preferred model for inference, with SAR and SEM results reported as spatial diagnostics.