Automated vehicle (AV) technology is expected to make roads safer. However, until recently only a handful of studies could test such hypotheses due to limited access to testing data. This study contributes to the literature by jointly analyzing the associated factors of three interrelated outcome variables—vehicle at fault, collision type, and injury outcome in AV-involved crashes. We use Bayesian networks to analyze the manually extracted data from reports of 333 AV-involved crashes that occurred in California between January 2017 and October 2021. The summary statistics indicate that rear-end collisions are the dominant (63.5%), while AVs are at fault for a small proportion of crashes (14.4%), and a majority of crashes (84.4%) are noninjury. The joint inferences of the Bayesian networks show that irrespective of the collision type, when the AV is at fault, the chance of the physical injury in a crash increases significantly. Further, the chance of an AV being at fault seems higher in parking locations, and during driving at wet pavements in unclear weather. The chances of AV rear-end collisions are lower in the parking lot, and when nonvehicular participants are involved but increase in high traffic flow roadways. We also find that the likelihood of physical injury is higher at high-speed locations, intersections, and wet pavements. These insights suggest specific areas (unsignalized intersections, less structured right-of-way rules, and wet pavements) where technological improvements could enhance the safety performance of AVs.

In recent years, nonmotorized transportation and micromobility have become increasingly popular in urban areas, and bike sharing and e-scooter usage are more affordable and accessible than ever before. However, users of these modes face several significant challenges, including risky riding behavior and the danger of being involved in collisions or near collisions with other motorists. Because of a lack of relevant data sets, it is difficult to comprehend the underlying causes of e-scooter collisions. This study collected narrative descriptions of 24 e-scooter collisions that occurred in Texas in 2021. The perceptual cycle model (PCM) was applied to these narrative crash reports to analyze the causation patterns of the crash occurrences. The PCM mechanism was applied independently to fatal and injury collisions, and the results were then compared with Haddon’s matrix to develop policy recommendations. The novel approach proposed in this study can be widely utilized by road safety professionals to understand the causal relationships behind other key road safety issues.