Cyclists’ crossing behaviour at roundabouts: A Generalized Spatially Correlated Nested Logit model

Cyclists navigating roundabout crosswalks face heightened crash risk due to ambiguous right-of-way rules and the absence of signal control. Understanding their operational behaviour at these conflict points is essential for designing safer facilities. Existing behavioural models often overlook the way cyclists perceive vehicle motion when deciding to cross and spatial dependence among feasible trajectories. This study develops a Generalized Spatially Correlated Nested Logit (GSCNL) model to capture cyclists' decision processes when crossing in the presence of motorized vehicles. The model extends the spatial Generalized Extreme Value (GEV) framework to represent correlation among spatially adjacent movement alternatives. Each alternative corresponds to a potential future position that a cyclist may occupy at the next time step. Spatial grouping is used to model dependencies among neighbouring choices. Two perception assumptions were investigated: (1) cyclists perceive interacting vehicles as moving in constant velocity, and (2) cyclists perceive vehicles as following curved paths. Results show that the constant-velocity formulation provides a superior fit, suggesting that cyclists may not exhibit strong perceptual sensitivity to vehicle acceleration. The proposed GSCNL and the conventional Spatially Correlated Nested Logit (SCNL) model achieved a similar overall fit. However, the GSCNL offered enhanced interpretability of non-proportional substitution patterns through its spatial correlation structure. The findings suggest potential advancements in spatial choice modelling of vulnerable road users. They may also inform the development of safer roundabout designs that better reflect the behavioural mechanisms underlying cyclists’ crossing decisions.