Modeling And Analysis Of Cyclists Crossing Violation Behavior Considering Heterogeneity

Many countries gradually advocate cycling for its green,healthy,environmental protection,and low-cost advantages.The illegal crossing behavior of cyclists is one of the main reasons for frequent conflicts between motor and nonmotor vehicles and has been a hot issue in road traffic safety research in recent years.Crossing violations are influenced by numerous factors such as people,vehicles,roads,and the traffic environment,and it isn’t easy to obtain relevant data comprehensively.The effect of unobserved heterogeneity is not fully considered in existing studies,which may easily lead to biased estimation of model parameters and may lead to wrong statistical inferences.To this end,based on field survey data of unsupervised signalized intersections,this paper constructs a model of two-wheeled cyclists’ violation behaviors to reveal the influence of unobserved heterogeneity and propose corresponding safety enhancement measures accordingly.The main work is as follows.(1)Data processing and fundamental feature analysis of cyclists’ crossing behavior.Three signalized intersections without traffic warden supervision in Beijing were selected for video filming of cyclists’ crossing behaviors,extracting data on their individual characteristics,behavioral characteristics,traffic environment,and crossing gaps,comparing the differences in cyclists’ red-light crossing behaviors under different attributes using cross-linked tables and chi-square tests,studying the distribution characteristics of cyclists’ crossing gaps using basic statistical methods,and fitting them with parameter models to explore cyclists’ distribution pattern of gap values for red-light crossing.The results show that: the accepted gap of the two types of bicyclists obeys the Weibull distribution,and the rejected gap obeys the Log-normal distribution;the proportion of red-light running behavior after waiting is higher than that of immediately red-light running(38.70% vs.21.15%),therefore,when carrying out traffic safety education,emphasis should be placed on the fact that red-light running behavior after waiting is still a high-risk behavior and is responsible for the primary accident responsibility.(2)Modeling of red-light running behavior of cyclists considering heterogeneity and analysis of influencing factors.The correlated random parameters Logit with the heterogeneity-in-means approach was selected to deeply reveal the effect of unobserved heterogeneity in the interaction of random parameters on red-light running behavior.The results show that the chosen model has the best fit compared with the traditional logit model;the number of violating individuals and the number of waiting individuals show significant random parameter effects,and the unobserved heterogeneity of the interaction between the two variables has a positive impact on red-light running behavior after waiting;the results of the heterogeneity-in-means of the random parameters indicate that the e-bike riders and riders approaching from the left are prone to run the red-lights,thus,the regulation of illegal crossing behavior of these two types of riders needs to be strengthened.(3)Modeling of riders’ gap acceptance behavior and analysis of influencing factors.An Accelerate Failure Time model(AFT model)considering Gamma shared heterogeneity was constructed based on survival analysis methods to investigate the unobserved heterogeneity effects among riders within different signal cycles and the impact of various risk factors on riders’ gap acceptance behaviors.The results show that gender,crossing stage,gap type,and conflict vehicle have significant effects on cyclists’ crossing gap values.Shared heterogeneity is observed in gap acceptance behaviors among cyclists with different signal cycles,and the Log-normal AFT model with Gamma shared heterogeneity has a better fit than the standard parametric AFT model.Ignoring the shared heterogeneity would result in underestimating the effects of each explanatory variable on the gap values and even lead to wrong statistical inference.