Spatial Regression Analysis using Queen Contiguity Weight Matrix and PCA Dimensionality Reduction

Abstract

Conventional linear regression often falls short in poverty analysis, as it fails to account for spatial interdependence between neighboring regions and frequently encounters multicollinearity among socioeconomic variables. This study investigates the presence and nature of spatial effects in poverty data across regencies and cities in Central Java Province, Indonesia, and assesses the performance of an enhanced spatial regression model. We employ a Spatial Autoregressive Model (SAR) integrated with a queen contiguity spatial weight matrix and apply Principal Component Analysis (PCA) to reduce dimensionality and mitigate multicollinearity. The results demonstrate a strong model fit, with a pseudo R² of 0.94311, and reveal a statistically significant negative spatial lag coefficient (ρ = -0.2039, p-value = 0.04420), indicating that areas of lower poverty are often surrounded by higher poverty neighbors. This integrated approach provides a more accurate framework for spatial poverty mapping, offering actionable insights for designing regionally targeted development policies.