Aims: Urban traffic safety was understood to be shaped by the spatial organization of street networks. For Baghdad, this study examined the relationships between network-topological metrics and crash risk, addressing the limited evidence base in the Middle East.
Methodology: Data from Baghdad City, Iraq, between 2024 and 2025 were assembled and quality-controlled. Crash rates were normalized by vehicle-kilometers traveled. A global ordinary least squares model with robust errors was estimated; When residual autocorrelation was detected, spatial lag/spatial error/spatial Durbin models were fitted. Geographically weighted regression with adaptive bandwidth was applied to capture spatial non-stationarity.
Findings: Significant clustering of crash rates was observed (Moran’s I=0.22; p<0.001). In ordinary least squares, positive effects were observed for annual average daily traffic, building area, openness, and meshedness, and a negative effect for betweenness. The selected spatial Durbin model confirmed spatial spillovers (ρ≈0.19) and large total effects of annual average daily traffic (~0.36) and building area (~0.26), while betweenness remained negative (~−0.12). Geographically weighted regression revealed broad spatial consistency: Betweenness was significantly negative across ~48% of cells, whereas annual average daily traffic and building area were significantly positive across ~68% and ~61% of cells, respectively. Crash rates peaked within 1-3km of hospitals (~310 per 100m vehicle-kilometers traveled) and declined beyond 5km (~170).
Conclusion: These patterns indicate that corridor-scale, network-oriented safety strategies are needed, with priority given to speed management, protected intersection design, and access control in high-exposure, highly meshed and more open sections, especially around hospitals.