Leveraging the Second Strategic Highway Research Program Naturalistic Driving Study: Examining Driver Behavior When Entering Rural High-Speed Intersections

CONCLUSION AND DISCUSSION

Naturalistic data of drivers approaching rural high-speed, stop-controlled intersections were analyzed for the purposes of modelling stopping and scanning behaviors. Older and younger drivers were found to apply the brakes at different distances from the intersection. This may contribute to rear-end collisions when younger drivers follow older drivers traveling at comparable speeds. However, given that drivers in general were found to focus predominantly on the forward ROI, they would likely use other visual cues related to braking to avoid potential collisions. Local transportation departments are powerless to affect age distribution, but the knowledge of this discrepancy could help explain or prevent such collisions in communities with higher proportions of elderly residents. This difference may also be due to the use of engine braking or driver comfort.

The probability of coming to a complete stop was found to correlate positively with reported AAM, which suggested an experience effect: drivers learned the high value (and negligible cost) of making a complete stop as they drive longer distances. The non-significant relationship with age group suggests that mileage is more reflective of experience than time.

Drivers approaching rural high-speed, stop-controlled intersections spent the last 98 ft of the approach principally engaging in scanning behaviors. This behavior could be leveraged for a variety of applications such as the placement of intersection conflict warning systems (ICWSs) and guide signs. A 2015 assessment of ICWSs examined the effects of sign wording and placement on comprehension using a driving simulator.⁽²⁴⁾ Signs were placed across the intersection for some participants and to the left for others (along the intersecting major route, visible to the driver only after arriving at the stop sign, and visible to the driver when looking approximately 90 degrees to the left). No statistically significant difference in comprehension was detected based on placement. Because the present study found that a significant proportion of drivers did not come to a complete stop prior to entering an intersection, such drivers would likely miss the left-placed sign despite pre-arrival scanning, thus rendering it ineffective. ICWS signs placed across the intersection maximize the probability of detection and thus the probability of conflict avoidance, regardless of scan time allocation.

This study was also meant to assess the SHRP2 database in its capacity to address further questions regarding driver safety. Due to the nature of any NDS, no aspect of driving can be controlled by the researcher; instead, observations must be filtered out to arrive at comparable situations. Though numerous data elements were captured—with varying rates of completeness—few observations may exist across comparable situations. Narrowly defined research topics may find sample sizes insufficient for analysis. For those topics with sufficient sample sizes, the SHRP2 database can be a valuable complement to other research tools such as driving simulators and instrumented vehicles.

This research analyzed 411 crossings of four Pennsylvania intersections. Though this research represents a small fraction of all of the data captured during the NDS, this will likely be the case for future narrowly defined research endeavors. Based on the results of this study, the SHRP2 database should be considered a useful resource for the exploration of driver safety issues. The RID provides ample infrastructure features to explore a wide range of settings (e.g., intersections, curves of specific radii, specific road types, etc.), and vehicular kinematics were recorded at sufficiently high frequency for temporally detailed analysis. However, studies requiring precise eyeglance vectors (e.g., to specific roadside signs) should not rely on head movement-based video reduction. This research did not use any of the radar-enabled data (e.g., headway, time-to-collision, etc.) and therefore cannot comment on the usefulness of such data.