A Diagnostic Framework for Selecting Spatial Regression Models in Urban Accessibility Studies

Objective: This study develops a route-based spatial modeling framework to investigate travel time inequity (TTI) in access to urban parks across Tehran, compares alternative spatial regression models, and introduces a diagnostic Commercial Intensity Index (CII) for interpreting model performance under different land-use conditions.
Method: TTI was defined as the normalized average travel time from each Traffic Analysis Zone (TAZ) to its nearest park using segment-level speeds derived from real-time traffic data. Area Ratio (AR) and Count-Area Ratio (CAR) indicators were calculated for residential, commercial, and educational land uses along shortest travel paths. Ordinary Least Squares (OLS), Geographically Weighted Regression (GWR), and Multiscale GWR (MGWR) models were compared. A Commercial Intensity Index (CII) was also developed as a diagnostic indicator to support the interpretation of spatial model performance under different land-use configurations.
Results: OLS achieved an adjusted R² of 0.904 but showed strong residual spatial autocorrelation and systematic overestimation. GWR improved local model fit (mean R² = 0.972) and reduced residual clustering, while MGWR produced the best overall performance (R² = 0.974; AICc = –171.0) by accounting for variable-specific spatial scales. Residential AR is generally associated with lower TTI, whereas commercial AR exhibited spatially heterogeneous effects. Within the analyzed dataset, CII-based clustering identified three spatial regimes—peripheral, transitional, and core zones—that were associated with the optimal performance of OLS, MGWR, and GWR, respectively.
Conclusions: The results highlight the importance of accounting for spatial heterogeneity and land-use intensity in modeling park accessibility inequities. The study contributes a route-based land-use modeling framework together with a Commercial Intensity Index that facilitates the interpretation of spatial regression model performance under different urban land-use configurations.