How far from home do crashes occur? A network based analysis.

• Crash spot–residence location distances were examined spatially and statistically. • Crash spot–residence distances were used to assess risky perimeters for analyzed groups. • Clear differences among the groups with respect to varying factors were identified. • Spatial crash distribution of individuals can be integrated into crash prediction. This paper investigates the proximity of crashes to the residential locations of the crash occupants. To this end, two years of crash data was disaggregated by the crash occupants' ZIP codes for a study area in Southwest Florida in order to calculate the roadway network distances between their residential ZIP code area centroids (origins) and crash spots (destinations). These distances are then used to create multiple O-D vectors, so that several different groups can be analyzed controlling for non-motorist types (e.g. pedestrians, cyclists), rural vs. urban origin ZIP codes, different levels of crash severity, DUI involvement, and different age groups. Then, the best-fitting statistical distributions were identified for each group to assess the proximity of crash spots to the residences of crash occupants. Finally, a selection model was implemented to identify the effects of several factors on the distance between the crash spots and the residence locations. Results indicate clear differences in crash involvement among the groups with respect to varying urban densities, people's ages and modes of travel. These findings can help in the development of more accurate crash prediction methods, as most current approaches only implement variables associated with traffic and roadway geometry. [ABSTRACT FROM AUTHOR]