Modelling gap selection of mobile phone distracted drivers at roundabouts: A mixed multinomial logit with heterogeneity-in-means approach

Gap selection at roundabouts is a critical driving task whereby drivers simultaneously evaluate gaps in circulatory traffic and control their vehicle dynamics. Whilst this task is already challenging and increases workload, it becomes more difficult when drivers concurrently use mobile phones. A few studies suggest that the gap selection of distracted versus undistracted drivers is not different; however, the determinants of gap selection could vary, and heterogeneous gap selection can also be observed if appropriate modelling techniques are applied. To this end, this study finds that past studies either applied simple pairwise comparisons or developed fixed parameter models, which are incapable of deciphering distracted driver-level heterogeneity in gap selection. Motivated by this research gap, the current study focusses on understanding the relationship between mobile phone distraction and gap selection at roundabouts using a mixed multinomial logit model. Thirty-two young, licensed drivers, with equal male and female representation, were asked to select a gap for entry into the roundabout whilst considering giving way to oncoming traffic. In the driving simulation environment, participants faced a roundabout entry scenario thrice, with each corresponding to no phone, handheld, and hands-free driving conditions. To model drivers’ gap selection outcomes (small, medium, and large/no gap), a mixed multinomial logit model is developed, accounting for repeated measures of experiment design, capturing unobserved heterogeneity, and allowing the mean of a random parameter to vary. The model has two random parameters (hands-free and handheld driving conditions), whereas the fixed parameters include age, gender, acceleration noise, initial post encroachment time, and self-reported crash history. The model suggests heterogeneous effects of handheld and hands-free driving conditions on gap selection, implying that distracted drivers may select large and small gaps depending on the context. The model also indicates that heterogeneity-in-means for the handheld parameter is associated with driving experience, implying that experienced distracted drivers are more likely to select medium gaps. The findings of this study provide in-depth information about understanding gap selection mechanisms and forming policies for distracted drivers.