A bilevel model for multivariate risk analysis of pedestrians' crossing behavior at signalized intersections

Pedestrians who cross streets during the red-man phase of traffic light signals expose themselves to safety and health hazards and hence are considered to be at risk. Pedestrians' street-crossing behavior is in general the outcome of interaction between pedestrians and vehicles: the gaps between vehicles provide an opportunity for pedestrians to cross the street, and pedestrians may or may not accept the street-crossing risk during the red-man phase. In this paper, we propose a multivariate method to investigate pedestrians' risk exposure associated with unsafe crossings. The proposed method consists of two hierarchically interconnected generalized linear models that characterize two different facets of the unsafe crossing behavior. It gauges pedestrians' attitudes toward risk-taking and also measures the impact of potential risk factors on pedestrians' intended waiting times during the red-man phase of the traffic lights. A Bayesian approach with the data augmentation method is used to draw statistical inference for the parameters associated with risk exposure. The proposed method is illustrated using field traffic data.