Application of Geographically Weighted Regression Technique in Spatial Analysis of Fatal and Injury Crashes

Pirdavani, Ali; Bellemans, Tom; Brijs, Tom; Wets, Geert · 2014 · Crossref

DOI: 10.1061/(asce)te.1943-5436.0000680

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This study addresses the limitations of Generalized Linear Models (GLMs) in crash prediction, which assume fixed global relationships between explanatory variables and crash outcomes. Because crash occurrences are often spatially heterogeneous, ignoring spatial correlation can deteriorate predictive power. The research aims to develop Zonal Crash Prediction Models (ZCPMs) using Geographically Weighted Generalized Linear Models (GWGLMs), specifically Geographically Weighted Poisson Regression (GWPR), to capture spatial non-stationarity in the association between injury crashes and various predictors. The analysis covers 2,200 Traffic Analysis Zones (TAZs) in Flanders, Belgium, using crash data from 2004 to 2007. Predictors included exposure metrics derived from the FEATHERS activity-based transportation model (e.g., Vehicle Hours Traveled, Vehicle Kilometers Traveled, Number of Trips), network characteristics (e.g., capacity, intersection density), and socio-demographic variables (e.g., population, income). The authors first confirmed the necessity of spatial modeling by calculating Moran’s I statistics, which revealed significant spatial clustering for all dependent and explanatory variables. They then compared standard GLMs against GWPR models, utilizing adaptive bandwidths and Gaussian kernel functions selected via the corrected Akaike Information Criterion (AICc). The results demonstrate that GWPR models significantly outperform traditional GLMs. Model #4, a GWPR model using Vehicle Hours Traveled and Number of Trips with an adaptive bandwidth, achieved the lowest AICc (10,605) compared to the best GLM (16,918). Additionally, the GWPR models exhibited lower Mean Squared Prediction Error (MSPE) and higher Pearson Correlation Coefficients (PCC) than their GLM counterparts. While significant variables like "Number of Trips" maintained consistent positive signs across all local estimates, other coefficients varied spatially, sometimes exhibiting counterintuitive signs. The authors attribute this variation to local multicollinearity or the specific spatial context of each zone, highlighting that variable impacts are not uniform across the study area. The significance of this work lies in validating the superiority of spatially explicit modeling techniques for traffic safety analysis. By capturing spatial heterogeneity, GWGLMs provide more accurate predictions and nuanced insights into how different factors influence crash risks in specific locations. This approach allows for better integration of Travel Demand Management policies at the planning level, as it accounts for the varying effectiveness of safety measures across different geographical contexts.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• incidence prevalence

Information type

What kind of knowledge this paper contributes, grouped by family — independent of topic (what it is about) and method (how it was studied).

• Empirical Findings: crash risk outcomes