Spatial heterogeneity analysis of macro-level crashes using geographically weighted Poisson quantile regression

Area of Science:

• Traffic Safety
• Spatial Statistics
• Econometrics

Background:

• Traditional models like quantile regression and geographically weighted regression have limitations in capturing spatial variations in crash factor effects.
• The influence of crash-related factors on crash frequency distribution is known to vary spatially and across different quantiles.

Purpose of the Study:

• To introduce and validate a geographically weighted Poisson quantile regression (GWPQR) model for analyzing spatial heterogeneity in crash factor effects.
• To investigate how exposure, socio-economic, transportation, network, and land use factors influence crash frequency distribution across different spatial regions and quantiles.

Main Methods:

• Development and application of the geographically weighted Poisson quantile regression (GWPQR) model.
• Case study using vehicle-related crash data from New York City.
• Comparison of GWPQR performance against traditional models (NB, QR, GWNBR).

Main Results:

• The GWPQR model significantly outperforms NB, QR, and GWNBR in modeling skewed crash distributions and capturing spatial heterogeneity.
• Identified key, important, and general crash-influencing variables based on significant coefficients.
• Demonstrated that influencing factors have varying effects across different quantiles and spatial regions.

Conclusions:

• The GWPQR model provides a robust framework for understanding complex spatial variations in crash factor impacts.
• Findings support the development of targeted safety countermeasures and policies at the urban regional level.
• Acknowledges the dynamic nature of crash factor influence across space and distribution quantiles.