2024 Volume 12 Issue 2 Pages 181-200
In spatial econometrics, traditional spatial weight matrix (SWM) methods often fail to capture the complex spatial dynamics of large cities. This study optimizes SWM calculations within spatial econometric models, constructing Graph- Based Spatial Weight Matrices (GBSWM) through the Simple Shortest Path Algorithm by analyzing urban road networks, thereby capturing the intricate spatial relationships within the city. The methodology compares the performance of GBSWM with traditional Simple Distanced Spatial Weight Matrices (SDSWM) using Geographically Weighted Regression (GWR) models. The results show that GBSWM significantly outperforms SDSWM in predicting minor crime events (e.g., 'summonses') in New York City. Improved p-values, Pseudo R-squared values, and model accuracy matrices attest to the improved predictive accuracy of GBSWM. These findings demonstrate the superior capability of GBSWM in capturing complex spatial relationships and interactions within urban settings. The integration of graph theory into spatial econometrics represents a theoretical and methodological advancement. The findings of this study are essential for improving the calculation of spatial weigh matrices, providing a more accurate tool for prediction and analysis in spatial econometric models. This result emphasizes the potential of applying graph methods in spatial econometrics, paving the way for implementing more detailed and practical urban spatial analysis.
