A hierarchical Bayesian multivariate ordered model of distracted drivers' decision to initiate risk-compensating behaviour

Mobile phone distracted drivers have been reported to initiate risk-compensating behaviour depending on a multitude of factors such as roadway environment and traffic characteristics, personal demographics and psychological attributes, and mobile phone task characteristics. However, the complexities of drivers’ decisions in engaging in such behaviour are not well known. This study aims to fill this gap by developing a comprehensive multivariate ordered model in Bayesian framework for risk-compensating behaviour of distracted drivers. The multivariate setting captures the common unobserved factors between multiple types of risk-compensating behaviour. In addition, an instrumental variable is employed to account for the endogeneity between crash risk and driving behaviour. To capture the varying effects of exogenous factors as well as varying propensity of initiating risk-compensating behaviour, the model is specified with grouped random parameters and random thresholds. This model is then empirically tested by data from a survey, which was specifically designed to understand the risk-compensating behaviour of mobile phone distracted drivers in Queensland, Australia. Results indicate that the grouped random parameters random thresholds ordered model has a substantially improved fit compared to its fixed parameters/fixed thresholds counterparts, indicating that the unobserved heterogeneity is significant, both in the effects of exogenous factors and in the propensity of initiating risk-compensating behaviour. It is found that drivers’ decisions to engage in different types of risk-compensating behaviour are correlated, indicating that they generally initiate different types of risk-compensating strategies simultaneously. Overall, the perceived crash risk has been found to increase the likelihood of risk-compensating behaviour among distracted drivers. Demanding secondary tasks and complex road traffic environment are also found to initiate risk-compensating behaviours such as increasing headway, reducing driving speed and visual scanning of the surrounding environment.