Citation

Siddiqui C, Abdel-Aty MA. Transp. Res. Rec. 2016; 2601: 59-71.

Copyright

(Copyright © 2016, Transportation Research Board, National Research Council, National Academy of Sciences USA, Publisher SAGE Publishing)

DOI

10.3141/2601-08

PMID

unavailable

Abstract

This study investigated two methodologies for allocating crashes within a zone and compared them in light of crash prediction models. The assumption that crashes within a zone were influenced only by the characteristics of that zone was examined with motor vehicle crash data. Models were specified by a hierarchical structure with submodels that provided separate estimates of covariate sets for crashes that occurred at or near boundaries and that occurred within a zone away from zonal boundaries. The proposed model structure was compared with spatial and nonspatial statistical models fitted for a wide array of independent variables. Two layers of zonal influences were investigated: (a) the influence of immediate neighbors and (b) the additional influence of neighbors of immediate neighbors. The roadway network hierarchy was considered in model sublevel specifications. Comparison of the candidate models showed that the complex nested model structure specifying interior and boundary crashes did not necessarily provide the best fit. Spatial models with submodels based on roadway network hierarchy were found to have the best goodness-of-fit measures for both total and severe crashes. Significant variables were common between total and severe crashes. However, only on-system crashes were significantly associated with the exposure variable, indicating that socioeconomic characteristics and land use types play an important role in off-system crash propensity.

Language: en