The purpose of the exploratory analysis in this section is to summarize the collected SHRP2 NDS curve dataset. This initial curve dataset contained 132 crashes and 220 near-crashes on all types of curves. Additional 2,373 baseline events were also included in the analysis as the control group. It should be noted that the events this section included all crashes on all types of curves from the SHRP2 NDS project. In general, 45% percent of the all curve-related crashes occurred on left turn curves, while the other 55% of crashes occurred on right turn curves. The majority (84%) of crashes and near-crashes events were happened in rural area. Almost half of the crashes (46%) occurred under free flow conditions. It was found 93% of the drivers in safety critical events worn safety belt properly. Three quarters of the crash events occurred in day time. In order to better understand the requested data, the events were further summarized in the following paragraphs.
The crashes and near-crashes on all curves were further classified into four categories as shown in Figure 4.4. There are four levels of severities, including the most severe crashes (level 1), police-reportable crashes (level 2), minor crashes (level 3) and near-crashes (level 4). It was found the most severe type of crashes had the smallest number of events. The near-crash had the largest number of events. This phenomenon was referred to as Heinrich’s law in the book
Industrial Accident Prevention, A scientific Approach (Heinrich, 1931). Heinrich’s law found the
event frequency decreases as event severity increases. In this curve dataset, for every 7 severe crashes, there were 12 police-reportable crashes, 113 minor crashes, and 220 near-crashes as shown in Figure 4.4. In this dissertation, both crashes and near-crashes are referred as one event type as safety critical event. Guo et al. (2010) evaluated the causes of crashes and near-crashes
and concluded that the near-crashes could be used as crash surrogate for crashes in naturalistic driving study. This evidence supports the combination of crashes and near-crashes as one event category.
Figure 4.4 Crashes and near-crashes by severities
Although all 352 curve-related crashes were physically occurred on curves, the crashes were caused by a variety of reasons. The incident types of the curve-related crashes are shown in Figure 4.5. The most frequent incident type was roadway departure event. The rear-ended events were the second frequent type of events on curves. However, the rear-ended crashes were
probably caused by traffic conditions rather than the curve geometry, so these events might not be included in the final analysis. Similarly, curve design also had little influence for animal- related crashes. Therefore, they were excluded from the final analysis. The pedestrian-related events were also irrelevant to the curve geometries, and they were excluded from the analysis.
Figure 4.5 Distribution of curve-related crashes and near-crashes by incident types
Figure 4.6 Distribution of curve-related crashes and near-crashes by junction types
The curve-related crashes were further plotted by junction types as shown in Figure 4.6. Sixty percent of the curve-related crashes did not related to any junctions. However, the rest 40%
1% 3% 3% 5% 7% 8% 9% 28% 38% 0% 5% 10% 15% 20% 25% 30% 35% 40% Pedestrian-related Turn across path Head-on or sideswipe…
Other Animal-related Turn into path Sideswipe… Rear-end, striking Road departure 60% 11% 7% 6% 5% 4% 1% 0% 10% 20% 30% 40% 50% 60% 70%
Non-junction Intersection Entrance/Exit ramp Parking lot entrance/exit Driveway, alley access, etc. Interchange area Other
of curve-related events were related to certain types of junctions, such as intersection (11%), entrance or exit ramps (7%), parking lot (6%), and driveway (5%). The forward view video was manually checked to confirm the relation of the crashes to roadway junctions. Since the presence of junction could have significant impact on divers’ curve driving behavior, all the junction- related crashes were excluded from the final analysis.
Figure 4.7 Subset the SHRP2 curve-related events
The requested dataset found some types of crashes had little relationship with the curve geometry. For example, rear-ended crashes and animal-related crashes had minimum relationship with curve geometry. Those types of crashes should be treated separately. Hence, it was
determined that only the roadway departure, heads-on collision, and sideswipe events were included in the final analysis. Additionally, the events near intersections, roundabout,
construction zone and parking lots were excluded from the analysis. Only the roadway departure crashes on rural two-lane (one lane in each direction) curves were kept in the analysis. The forward view videos were carefully reviewed to make sure the selected events followed the definition of roadway departure crashes on rural two-lane roads in this study. The reduced rural two-lane curve event dataset is illustrated in Figure 4.7. The dataset contains 67 safety critical events on rural two-lane curves. For baseline events, the goal was to sample twice the number of
SHRP2 Roadwway Departure Events on Rural Two-Lane Curves 67 Safety Critical Events 136 Baseline Events
SHRP2 Curve-related Events
baseline events than the safety critical event, otherwise, the small ratio of safety critical events and baseline events could create instability issue for the logistic regression model. Hence, 136 baseline events were sampled from the 2,373 balanced-sample baseline events.