ALTERNATIVE OR DERED RESPONSE FRAMEWORKS FOREXAMINING PEDESTRIAN INJURY SEVERITY IN NEW YORK CITY

摘要

This paper focuses on identifying the appropriate ordered response structure for modelingpedestrian injury severity. The alternative ordered response approaches considered for theempirical analysis include: ordered logit model (OL), generalized ordered logit model (GOL)and latent segmentation based ordered logit model (LSOL). The GOL and LSOL modelsenhance the traditional ordered logit model in different ways. The GOL model relaxes therestrictive thresholds in the ordered logit model by allowing for individual level exogenousvariable impacts on the threshold parameters. On the other hand, the LSOL model allows fordifferential impact on the alternatives by segmenting the pedestrian crash population into varioussegments with segment specific ordered logit parameters. In our study, we focus on examiningthe performance of these two model structures relative to the traditional OL model in the contextof pedestrian injury severity. The performance of the formulated injury severity models aretested based on the “New York City (NYC) Pedestrian Research Data Base” for the year of 2002through 2006. To our knowledge, the study provides a first of its kind comparison exerciseamong OL, GOL and LSOL models for examining pedestrian injury severity. The modelestimation results clearly highlight the presence of segmentation based on the crash location ofpedestrian accidents. The crash location attributes that affect the allocation of pedestrians intothese segments include: regional county, functional classification of roadway, pedestrian locationon roadway, number of travel lanes and number of parking lanes in the roadway system. The keyfactors influencing pedestrian injury severity are weather condition, lighting condition, vehicletypes, pedestrian age and season. Overall, the results of the empirical analysis provide credenceto the hypothesis that LSOL model is a promising ordered framework to accommodatepopulation heterogeneity in the context of pedestrian injury severity.