2015 QPRC

Speaker: Yaohua Zhang, University of Connecticut; yaohua.zhang@uconn.edu

Title: Statistical Assessment of Highway Transportation Safety

Abstract: Pedestrian safety is a serious concern in the United States, as in many countries. On average, a pedestrian is killed once every 2 hours and injured once every 7 minutes in traffic crashes (NHTSA, 2014). Thus, providing a safer environment for pedestrians remains a major concern for traffic safety professionals. Traffic engineers usually talk about quality of roads related to safety by using pedestrian-vehicle interactions (or conflicts) as proxy for crashes. Is this reasonable? It will be very useful to investigate where there is significant association between conflicts and crashes, and whether this association varies with location and with time periods. One traditional way to answer this question might be via model-based methods such as log-linear regression models. We describe an alternate semi-parametric approach for quantifying the statistical association between count random variables and for ordering the locations by decreasing magnitude of the association. Specifically, our method enables us to study association between two observed count data vectors or between an observed and a predicted vector. For instance, under certain road characteristics at some locations, the relationship between conflicts and crashes may be strong, while at other locations with different road characteristics, the relationship may be weak. Identifying the subgroup within which a strong association exists can help us have a better assessment of the quality of safety on our roads. This approach is also relevant to more general data mining settings, where there is often a need to identify relationships between any pair of variables when the relationship may hold only over an unspecified subpopulation that is represented by a subsample of the data we have available. This is joint work with Nalini Ravishanker and John Ivan, University of Connecticut.