| Crash frequency estimation and injury severity analyses have long been a focus in the safety research field. One of the common assumptions in safety analyses is that associations between dependent variables and explanatory variables are stationary across space. However, due to spatial heterogeneity, this assumption may not fully hold across the entire research region at jurisdiction level. The magnitude and significance of coefficients might vary spatially. Failure to capture the spatial heterogeneity could cause instability in model estimates and biased or invalid inferences. Therefore, the objectives of this study are to incorporate spatial heterogeneity into Safety Performance Function (SPF) development and modeling of crash injury severity, explore spatial varying patterns in associations between the dependent variable and explanatory variables, and provide empirical results for safety planning. For these purposes, police-reported crash data in New York City (NYC) was used. Geographically Weighted Regression method and several statistical methods are applied in this study. An effort in developing spatial SPFs was made for signalized four-leg intersections. Moreover, injury severity of bicyclist-involved crashes across NYC was investigated. In all analyses, global models for the research areas and local models for center intersections or crashes were estimated and compared. In addition, tests were conducted to identify factors having non-stationary associations with crash frequency or injury severity and these coefficients were mapped to visualize their spatial patterns.;The results show that associations between Annual Average Daily Traffic (AADT) on major road and crash frequency at intersections vary substantially across space. Besides AADTs, the number of lanes on major roads and minor roads are significantly correlated to crash frequency. But, at a borough level (i.e., Manhattan), some local models did not outperform the global models. Part of the reason could be the relatively small sample size and subject area. Turning to injury severity at intersections, bike lanes were found to have significant associations with probabilities of casualties in some areas not constrained by administrative boundaries. The other spatially varying correlates of injury severity include heavy duty vehicle involvement, bad light conditions, inclement weather, and multi vehicle involvement. Implications of the results were discussed in detail. |
