为深入了解影响高速公路交通事故严重程度的显著因素，以广深沿江高速东莞段2014—2019年事故数据为研究对象，将事故严重程度分为3类（即无伤亡事故、轻伤事故、重伤亡事故），并以事故严重程度为因变量、13个潜在影响因素为自变量，通过条件自回归先验解析相邻事故间的空间相关性，建立不同关联距离阈值的空间广义有序Probit模型。结果表明：事故数据间存在显著的空间关联；空间广义有序Probit模型优于广义有序Probit模型和多项Logit模型，基于250 m关联距离阈值的空间广义有序Probit模型表现最优。该模型参数估计显示：车辆类型和归属地、事故发生时间、事故地点曲率、桥梁路段和事故类型对高速公路事故严重程度均有显著影响。边际效应结果表明：相对于小汽车间的交通事故，涉及客车、货车和其他类型车辆的交通事故导致人员重伤亡的概率分别增加3.27%、1.53%和4.11%；外省车使得重伤亡事故的发生概率增加1.02%；相对于周末、春季和桥梁路段，工作日、夏季和非桥梁路段发生的交通事故导致人员重伤亡的概率分别提高0.87%、2.38%和0.08%；单车事故导致重伤亡的概率比多车事故低1.64%；事故地点曲率每增加1 km^-1，重伤亡事故的概率将降低1.54%。

To provide a deep insight into the significant factors that affect the severity of freeway crashes, this study took the crash data from the Dongguan section of the Guang-Shen Yanjiang Freeway in China from 2014 to 2019 as the research object. Crash severity levels were divided into three categories (i.e., no injury crash, minor injury crash, severe injury or fatality crash). Accounting for spatial correlation among adjacent crashes via conditional autoregressive priors, spatial generalized ordered Probit models with different correlation distance thresholds were developed, where the crash severity was used as the dependent variable and 13 potential influencing factors were used as independent variables. The research results show that there is significant spatial correlation among crashes; the spatial generalized ordered Probit models outperform the generalized ordered Probit model and multinomial Logit model; and the spatial model with 250-meters correlation distance threshold achieves the best performance. The results of model parameter estimation reveal that the type and registered province of vehicles, the time of crash occurrence, curvature of crash location, bridge section, and crash type have significant effects on freeway crash severity. The marginal effects of these factors indicate that: as compared with crashes with cars involved only, the involvement of bus, truck and other type vehicles will increase the probability of severe injury or fatality by 3.27%, 1.53%, and 4.11%, respectively; the involvement of vehicles from other provinces will increase the probability of severe injury or fatality by 1.02%; as compared with those occurring on weekend, spring, and bridge, crashes occurring on weekdays, summer, and non-bridge sections would increase the probability of severe injury or fatality by 0.87%, 2.38%, and 0.08%, respectively; the probability of heavy casualties caused by bicycle accidents is 1.64% lower than that of multi-vehicle accidents; the probability of severe injury or fatality will decrease by 1.54% for per 1 km^-1 increase in horizontal curvature of crash location.