“…Crash injury severity is usually represented by discrete categories such as fatal, incapacitating injury, capacitating injury, possible injuries, and property damage only (PDO). Due to the discrete nature of injury severity classes, it is often analyzed by discrete outcome statistical models such as binary, multinomial, probit, and logit models [ 26 , 27 , 28 , 29 ]. It is widely agreed that crash data may exhibit unobserved heterogeneity, which may be tackled by adopting other advanced statistical models such as ordered logit models [ 26 , 30 , 31 , 32 ], bivariate/multivariate models [ 33 , 34 , 35 , 36 ], random parameter model, [ 37 , 38 ], nested logit model [ 39 , 40 ], and Bayesian hierarchical models [ 41 , 42 ].…”