Index
Technical Report
Background
Crash Typing Results
Survey Results
Follow-Up Process Evaluation
Approach
Results
Discussion
Conclusions
Recommendations
Appendix A and B:
Crash
Typing Results
Each participant in the study received a copy of the software on CD, a user's manual, 25 crash reports (13 pedestrian and 12 bicycle collisions), and several evaluation and comment forms. The first phase of the test was for the participant to enter the data from the collision report forms and "type" the crashes using the software. The completed data sets were then analyzed and compared to the "correct" answers. A complete set of results is attached. They are presented in terms of the percentage of responses that were correct, within one level of error, and greater than one level of error. Those responses that were correct had the same crash type as previously determined by the project team. The responses that were within one level of being correct were coded correctly up until the last decision was made. For example, a bicyclist being struck by an overtaking vehicle may have been coded as an overtaking vehicle that "misjudged the distance required to pass" as opposed to an overtaking vehicle that "did not detect the bicyclist." In general, these errors are not considered to be major mistakes due to the level of subjectivity still required from interpreting the crash reports and due to the fact that most countermeasures are appropriate to that level of error. Those responses that were more than one level away from being correct are considered to be more major mistakes. These types of errors generally occurred for two or three specific crashes and are discussed below.
As shown in figure
1, 92 percent of the bicycle crashes and 89 percent of the pedestrian
crashes were correctly typed to within one level of error. In general,
the bicycle crashes had a higher percentage of correct answers (88 percent)
compared to the pedestrian crashes (76 percent). As shown in Appendix
A, only 1 of the 13 pedestrian crashes (#132105958) had an extremely high
percentage of incorrect responses that was greater than one level of error.
This particular collision was a very unique crash that involved a driverless
vehicle, which is a very small percentage of real-world crashes. With
respect to bicycle collisions, there was also one crash (#15185645) for
which 33 percent of the responses were incorrect by more than one level
of error. This collision involved a motorist making a right turn into
a driveway and striking the bicyclist traveling in the same direction.
From the crash report, however, it is
Figure 1. Summary of results from the pedestrian and bicycle crash typing beta test.
understandable how this crash could have been typed as an overtaking collision as opposed to a turning collision. This crash is a very good example of how the subjectivity in this methodology can never be totally removed.