CHAPTER 1: Light Vehicle - Heavy Vehicle Interactions Collected in the 100-CAR Study

Introduction

Overview of the Light Vehicle-Heavy Vehicle Problem

Truck crashes represent a significant problem on our highways. In 2002, 434,000 large trucks (gross weight > 10,000 lbs) were involved in vehicle crashes; 4,542 of these crashes resulted in a fatality. A total of 4,897 people died and an additional 210,000 were injured. Large trucks accounted for 4% of all registered vehicles in 2002, yet represented 8% of all vehicles involved in fatal crashes (National Highway Traffic Safety Administration, NHTSA, 2003). Truck crashes and their associated injuries and fatalities cost an estimated 24.4 billion in direct and indirect costs in 2002 (FMCSA, 2002).

The disproportionate number of vehicles to fatalities among large trucks is likely to contribute to the perception that truck drivers are irresponsible. However, these data do not signify that truck drivers are necessarily the problem. In fact, truck drivers have lower non-fatal crash rates per million vehicle miles traveled than light vehicles (NHTSA, 2003). However, light vehicles are extremely vulnerable when they interact with trucks because trucks often weigh 20-30 times as much as light vehicles (Insurance Institute for Highway Safety, 2002), and trucks take 20-40% farther to stop than light vehicles (NHTSA, 1987). This is best illustrated by the fact that over three-fourths of multiple vehicle fatal truck crashes resulted in the occupant(s) of the other vehicle being killed (NHTSA, 2004).

To combat this problem, proposals have been made to separate light and heavy vehicles on high-volume roads. For example, STAR Solutions has proposed to separate heavy trucks from passenger traffic on Interstate 81 (http://www.virginiadot.org/infoservice/resources/is-I-81-Star-exec.pdf), thereby reducing the likelihood of light vehicle-heavy vehicle (LV-HV) interactions. However, the enormous cost and logistical difficulties associated with new and modified road construction suggest that, in most cases, these vehicles will have to share the road for the foreseeable future. Thus, a better understanding of LV-HV interactions is needed to develop alternative interventions and countermeasures directed at reducing and/or eliminating the problem.

Prior Research on Light Vehicle-Heavy Vehicle Interactions

When Hanowski, Wiewille, Gellatly, Early, and Dingus (1998) conducted focus groups with local/short-haul truck drivers, they found that participants ranked "problems with light vehicles" as the most important safety issue. In fact, this was the only safety issue that was consistently cited in all 11 focus groups that were conducted. Similarly, Neale et al. (1998) found that LV-HV interactions were a significant safety concern among a sample of long-haul truck drivers. Only recently has empirical evidence supported truck drivers' claims.

Blower (1998) analyzed the University of Michigan Transportation Research Institute's "Trucks Involved in Fatal Accidents" database for all two-vehicle, truck-passenger vehicle fatal crashes in 1994 and 1995 (n = 5,453). He found that truck drivers were cited with a driver-related factor in 26.5% of the fatal crashes, while passenger vehicle drivers were cited in over 80% of the fatal crashes. The passenger vehicle driver was the only driver cited in 70.3% of the fatal crashes, while truck drivers were the only driver cited in 16.2% of the fatal crashes.

Stuster (1999) found similar results when he reviewed the U.S. Department of Transportation's Fatality Analysis Reporting System. He found that truck driver-related factors were cited in 29% of fatal truck crashes involving a passenger vehicle, while 67% of these same interactions were cited as passenger vehicle-related. Moreover, Wang, Knipling, and Blincoe (1999) found that LVs were the initiators in LV-HV fatal crashes by a ratio of approximately 3:1. Thus, it appears the actions of LVs are responsible for a substantial amount of the fatal LV-HV interactions.

Council, Harkey, Nabors, Khattack, and Mohamedshah (2003) took a different approach when they analyzed 16,264 LV-HV interactions from the North Carolina Highway Safety Information System. Rather than examining the police reports from fatal LV-HV interactions, they examined the police reports of both crashes and fatal crashes. While the prior studies assessed the most severe crashes, the Council et al. (2003) study assessed the overall LV-HV crash picture. Contrary to the other studies, however, Council et al. (2003) found that the truck driver was assigned fault in 48% of the crashes, while the passenger vehicle driver was assigned fault in 40.2% of the crashes (8.9% of the crashes were assigned fault to both drivers, while 2.9% were assigned fault to neither driver). The Council et al. (2003) data suggests that HV drivers were responsible for the majority of the LV-HV interactions (for all types of crashes). Thus, there appears to be some inconsistencies in the literature regarding assigned culpability in LV-HV interactions.

These prior studies assessed LV-HV interactions by examining vehicle crash databases that rely on police reports and crash reconstruction. These approaches are generally reliable, but they do have limitations, including witnesses and crash participants can be biased and report conflicting stories; police officers, while often experienced, generally do not receive extensive training in crash reconstruction; and witnesses or crash participants that were severely injured or killed in the crash are unlikely or are unable to effectively persuade the police officer about their side of the crash (also referred to as the "surviving driver" hypothesis).

Blower (1998) acknowledged the surviving driver limitation in his discussion and compared fatal LV-HV interactions with respect to driver survivability. When only the truck driver survived the fatal LV-HV interaction, the LV driver was cited with at least one driver-related factor in 81.9% of the fatal crashes, while the HV driver was cited in only 24.1% of the fatal LV-HV crashes. Conversely, when only the LV driver survived the fatal LV-HV interactions, the LV drivers were cited with at least one driver-related factor in 46.7% of the crashes, while HV drivers were cited in 57.7% of the crashes.

It would appear that driver survivability does affect which driver is cited in fatal LV-HV crashes. This makes intuitive sense because the surviving driver is able to report their biased account of the event. However, when both LV and HV drivers survive the crash, the LV driver was cited in 74.1% of the crashes while the HV driver was cited in only 35.5% of the crashes (Blower, 1998).

The crash database approach does not necessarily shed light on the full variety of LV-HV interactions because they rely solely on crashes and fatal crashes. While LV drivers have been shown to be culpable in a significant proportion of LV-HV interactions (Wang, Knipling, and Blincoe, 1999), we do not know why. An alternative approach, and the method used in the current study, is to study the pre-event behaviors of all LV-HV interaction critical incidents, including crashes, near-crashes, and crash-relevant conflicts.

Because the focus of this research is on analyzing critical incidents, it is important to define the three categories that are of most interest: crashes, near-crashes, and crash-relevant conflicts. In the 100-Car Study, Dingus et al. (2004) defined crashes, near-crashes, and crash relevant conflicts as follows:

Crash: Any contact with an object, either moving or fixed, at any speed in which kinetic energy is measurably transferred or dissipated, and includes other vehicles, roadside barriers, objects on or off of the roadway, pedestrians, cyclists, or animals.

Near-Crash: Any circumstance that requires a rapid, evasive maneuver by the subject vehicle, or any other vehicle, pedestrian, cyclist, or animal to avoid a crash. A rapid, evasive maneuver is defined as a steering, braking, accelerating, or any combination of control inputs that approaches the limits of the vehicle capabilities. As a guide: Subject vehicle braking >0.5 g or steering input that results in a lateral acceleration >0.4 g to avoid a crash constitutes a rapid maneuver.

Crash-Relevant Conflict (Incident): Any circumstance that requires a crash avoidance response on the part of the subject vehicle, any other vehicle, pedestrian, cyclist, or animal that is less severe than a rapid evasive maneuver (as defined above), but greater in severity than a "normal maneuver" to avoid a crash. A crash avoidance response can include braking, steering, accelerating, or any combination of control inputs. A "crash avoidance response" for the subject vehicle is defined as a control input that falls outside of the 99% confidence limit for control input as measured for the same subject.

Hanowski, Keisler, and Wierwille (2004) assessed two on-road in situ data collection efforts, one involving local/short-haul (L/SH) trucking and the other long-haul trucking with drivers who used sleeper berths, to examine critical incidents that occurred between LVs and HVs. In this study, critical incidents were defined as crashes and near-crashes. Near-crashes were events resulting in a close call or requiring rapid action by a driver to avoid a crash. Video and non-video data collected during the two studies were used to characterize 210 critical incidents involving LV-HV interactions. Of the 210 critical incidents analyzed in the Hanowski, Keisler, and Wierwille (2004) study, 78% were assessed to have been initiated by the LV driver, while the remaining 22% were initiated by the HV driver. It should be noted that in Hanowski, Keisler, and Wierwille (2004), "initiate" is synonymous with "at-fault." Thus, a vehicle that initiated an incident is meant to reflect the vehicle that was at-fault or responsible for the incident.

The benefits of the naturalistic data collection approach used in Hanowski, Keisler, and Wierwille (2004) are three-fold: (1) video and other supporting data are collected before, during, and after the event occurs, thereby providing a complete picture of the incident as it unfolds; (2) various types of non-crash LV-HV interactions can be analyzed; and (3) the use of video and non-video data allowed one to make objective assessments on the critical reason(s) for the incident (rather than incomplete, subjective police reports).

However, one limitation of this approach was that the video cameras were only installed in the HVs and not the LVs. Therefore, Hanowski, Keisler, and Wierwille (2004) were only able to assess LV-HV interactions from the HV driver's perspective. Thus, it is possible they missed critical incidents that were only apparent from the LV driver's perspective. Furthermore, the lack of instrumentation in LVs limits the understanding of the LV driver's behavior during the incident.

These limitations were addressed in the 100-Car Study by installing video cameras on LVs (Dingus et al., 2004). All identified LV-HV interactions from the 100-Car data set were included in the current analyses. Together, results from the current study and the Hanowski, Keisler, and Wierwille (2004) study may provide a more complete picture of the LV-HV interaction problem.

The current study used two classification methodologies to assess all LV-HV interactions: the classification methodology used in Hanowski, Keisler, and Wierwille (2004) (originally developed by Wierwille et al., 2001), and the methodology and terminology from the Large Truck Crash Causation Study (LTCCS) (Thieriez, Radja, and Toth, 2002). Thus, all LV-HV interactions in the current data set were coded with two similar, yet distinct, classification approaches. The primary advantage of this method is that reliable and valid comparisons can be made with both prior and future research studies using either approach.

Research Goals

The data from the 100-Car Study (Dingus et al., 2004) were used in the current project to assess the LV-HV interaction problem from the LV drivers' perspective. There were four primary aims in the current effort:

• Gain a better understanding of LV-HV interactions on our nation's roadways.
• Continue to develop the classification scheme and corresponding Contributing Factors list for LV-HV interactions used in Hanowski, Keisler, and Wierwille (2004) and use the terminology and methodology described in the LTCCS (Thieriez, Radja, and Toth, 2002).
• Compare the current data to the data obtained in the Hanowski, Keisler, and Wierwille (2004) study for a more complete picture of the LV-HV interaction problem.
• Provide background information that would serve as a necessary prerequisite to the development of countermeasures for LV-HV interactions.

Method

Participants and Setting

One hundred participants who commute to and from the Washington, DC metro area were initially recruited as drivers in the 100-Car Study (Dingus et al., 2004). As some participants had to be replaced for various reasons (e.g., dropped out of the study because they moved from the area), the final number of participants was 109. Age and number of miles driven annually were used to select a subject population that would increase the probability that rear-end collisions would occur in 13 months of data collection. High mileage drivers were selected to increase the number of vehicle miles traveled per year (increase exposure). Greater number of younger drivers (ages 18 through 25) were recruited as they are overly represented in rear-end collisions as compared to other age groups. Also, more males than females were recruited since males are overly represented in rear-end crashes (Knipling, Wang, and Yin, 1993). It should be noted that participants were recruited from all age groups and that the target average annual mileage per year was approximately 27,000 miles/year. However, the actual mileage driven by participants in the 100-Car Study did not match their self-reported annual mileage prior to the study. The actual mileage of participants in the 100-Car Study can be found in Dingus et al. (2004). Table 1, shown below, displays the age and gender distribution of participants.

Table 1. Participant Age and Gender Distributions.

Light Vehicle Types

The data that were collected in the 100-Car Study came from six makes/models/years of LVs, including Toyota Camry (1997-2001), Toyota Corolla (1993-2002), Ford Explorer (1995-2000), Ford Taurus (2000-2002), Chevrolet Malibu (2002), and Chevrolet Cavalier (2002). The Toyota and Ford models were chosen based on recent sales figures and on the number of vehicles available in the Washington, DC area. Figures 1 and 2 show examples of the LVs used in the 100-Car Study.

Figure 1 . Toyota Camry and Toyota Corolla Used in 100-Car Study.

Figure 2 . Ford Explorer and Ford Taurus Used in 100-Car Study.

A total of 20 Chevrolet vehicles (10 Malibus and 10 Cavaliers) were leased from the Virginia Tech Motor Pool and were instrumented with data collection equipment. Twenty participants were given leased vehicles to drive for one year. The additional 80 vehicles (comprised of the aforementioned Toyota and Ford models) were the participants' personal vehicles. These vehicles were instrumented with the identical data collection systems as the leased vehicles.

Data Collection Methodology for the 100-Car Study

A full description of the research methodology used in the on-road portion of the 100-Car Study can be found in Dingus et al. (2004). Because the data used in the current effort consisted of the video recordings of critical incidents, the primary methodological considerations to be described in this report are those related to the video systems.

Video Camera Systems

As shown in Figure 3, five video cameras were used in the video recording system: (1) a forward-looking camera that captured the forward roadway scene, traffic situation, and possible incidents; (2) a driver's face camera that was used to record facial expressions, eyelid closure, glance position, and head turns; (3) a right-side camera was mounted on the A-pillar of the passenger side and faced outward; (4) a dome camera was mounted from inside the vehicle and faced over the driver's shoulder towards the steering wheel, hands, and feet; and (5) a rear camera that was intended to capture the situation behind the vehicle. Infrared lighting was used to illuminate the vehicle cab so that the driver's face as well as their hands could be viewed by the camera during nighttime driving.

Figure 3 . The Five Camera Views Recorded in the Instrumented Vehicle.

The video camera arrangement shown in Figure 3 had several advantages. First, it provided good coverage around the vehicle so that incidents could be captured as they developed. Second, the driver's facial expression, approximate glance direction, and approximate level of eye closure were also captured. Third, the arrangement provided appropriate views, whether moving forward or backward.

The five camera images were multiplexed into a single image as shown in Figures 4 and 5. Note that the right side camera and the rear camera were presented in the lower left quadrant in a split arrangement.

Figure 4 . Diagram of the Multiplexed Camera Views.

Figure 5 . Split-Screen Presentation of the Five Camera Views.

Video Recording Operation

Video recording was tied to the booting/powering system and it began to operate 2 minutes after the ignition was on. It also shut down in an orderly manner when the ignition was turned off. It was desired for the recording system to record for as long as possible without requiring technician/researcher attention. Therefore, multiple recorders designed to operate in sequence were used. The video continuously recorded while the ignition was on, thereby allowing laboratory review and selection of the video without losses of any kind.

The videotaped episodes/incidents were selected and keyed to digitally recorded data. In some cases, the videotape timestamp was used to access the corresponding digital data. In other cases, the incident flags (described later) in the digital data were used to access the corresponding video. Therefore, there was a straightforward keying procedure that allowed both kinds of access to take place efficiently.

Data Collection and Storage

"Chase" vehicles drove to pre-determined locations (e.g., parking lots) and downloaded the data from the experimental vehicles via a data transfer cable that connected to an outlet located near the rear license plate. Each chase vehicle had a laptop computer with a large hard drive to store all vehicle data. After each download from the experimental vehicles, the success of the duplication procedure was verified. Once 2.3 GB of data were downloaded from experimental vehicles, the data were copied to a DVD and verified. This DVD was duplicated; one copy was sent to VTTI and the other copy was kept in Northern Virginia.

As the data arrived at VTTI, they were downloaded to VTTI's network attached storage (NAS) and saved. Once the data was safely copied to the networked attached storage at VTTI and quality checks were performed, the data were then remotely deleted from the experimental vehicle hard drive.

Incident Flag

A critical incident involves an unexpected event resulting in a close call or requiring rapid action (evasive maneuver) on the part of a driver to avoid a crash. Critical incidents were detected by one of three methods. The first method involved flagging events where the car sensors exceeded a specified value. An example of this is a braking response of >0.6 g would be recorded as a potential incident where the driver may have braked in a panic.

Table 2 lists all of the triggers and levels that were used in this first method. The second incident flagging method occurred when the driver pressed an incident pushbutton located on the dashboard. Drivers were instructed to depress a button on the dashboard (after the event, not during the event) if they witnessed an incident or were involved in an incident. The third method of detecting incidents was through analysts' judgments when reviewing the video. Note that the video systems were operational as long as the ignition was turned on. In identifying incidents, analysts looked through epochs flagged from either of the first two methods and could flag additional events within the epoch (termed "user triggered") if an incident was detected visually. Only those events that involved a LV-HV interaction are described in the current analyses. The results of other project analyses can be found in Dingus et al. (2004).

Table 2. Triggers and Their Levels Used to Identify Critical Incidents in the 100-Car Study Database.

The incident flags (associated with the first and second triggering methods) were computed and detected on-line (as well as stored) with the flag appearing in the video. Since the entire video recording was reviewed, the presence of flags served as an indicator to the analyst of the high likelihood, but not certainty, of an incident occurrence. However, the analyst was also mindful of the possibility of incidents without flags and reviewed the tapes accordingly. The data analysts watched 90 s epochs (1 min prior and 30 s post incident) of each driving incident and recorded the information shown in Table 3.

Table 3. Information Recorded During 90 s Epoch Analysis.

Data Reduction Reliability

Given that data analysts were asked to perform subjective judgments on the video and driving data, training procedures were implemented to improve both inter- and intra-rater reliability. Reliability testing was then conducted to measure the resultant inter- and intra-rater reliability. First, data analyst managers performed spot checks of the data analysts' work, monitoring both event validity judgments as well as recording all database variables. All data analysts also performed 30 mins of spot-checking of their own or other data analysts' work per week.

To determine how successful these techniques were, an inter- and intra-rater reliability test was conducted during the last three months of data reduction. Three reliability tests were developed (each containing 20 events) for which the data analyst was required to make validity judgments. In each of the three reliability tests, three of the 20 events were also fully reduced by the data analysts. Three of the test events on Test 1 were repeated on Test 2 and three other events were duplicated between Tests 2 and 3, to obtain a measure of intra-rater reliability.

The Kappa statistic was also used to calculate inter-rater reliability (Cicchetti and Feinstein, 1990). The Kappa coefficient (K = 0.65, p < 0.0001) indicated that the association among raters was significant. The average of the pair-wise correlation coefficients for the inter-rater analysis was 0.86. The coefficients for the intra-rater analysis were extremely high with nine raters achieving a correlation of 1.0 among the three reliability tests and five raters achieving a correlation of 0.99. Given these three methods of calculating inter-rater reliability, it appears that the data analysts training coupled with spot-checking and weekly meetings proved to be an effective method for achieving high inter-rater and intra-rater reliability.

Strengths and Limitations of the Methodology Used

All research approaches have strengths and limitations. Listed below are the strengths and limitations of the approach used in the current study.

Strengths

The primary strength of the approach used in this study was that all driver behaviors, visible by way of the video camera, were recorded whenever the vehicle was on and in motion. This information is vital in developing an understanding of the incident, the events leading up to the incident, and the aftermath of the incident from the LV driver's viewpoint. The video camera arrangement described allowed researchers to watch the critical incidents unfold from multiple camera views. The video camera system that was used not only afforded an opportunity to understand what happened, but in many cases why it happened. A second advantage of this approach was that multiple cameras views helped ensure that any critical incidents involving the LV driver were captured and available for analysis.

Possible Limitations

There were two possible limitations of the approach used in the current research. First, because the video cameras were installed in the LVs and not the HVs, critical incidents could only be captured from the LV driver's perspective. It was possible that LV-HV interactions, which may have only been apparent from the HV driver's viewpoint, were not recorded. However, because there was fairly complete video recording coverage around the entire LV, it was likely that most LV-HV interactions that occurred were recorded.

Second, because there were no cameras mounted in any HV, it is difficult to have a complete understanding of the HV driver's behavior during the incident. The video camera that was directed at the LV driver's face, along with the verbal utterances of the driver, provided the researchers with a fairly complete understanding of the LV driver's behavior before, during, and after each incident. However, this was not the case regarding the behavior of the HV driver. The absence of video footage of the HV driver's face meant that the HV driver's behavior had to be surmised based on the video of the HV collected from the LVs and the comments and facial expressions made by the LV driver.

These limitations were also raised in Hanowski, Keisler, and Wierwille (2004) for the LV-HV analyses that were conducted with both L/SH and sleeper berth (SB) trucks. Considering the previous Hanowski, Keisler, and Wierwille (2004) work in conjunction with the current research, a more complete assessment of LV-HV interactions from both the HV and LV driver's perspective was expected. Assumptions regarding driver behavior were required for each of these research efforts, particularly for the driver of the non-instrumented vehicle.

Results

Light Vehicle-Heavy Vehicle Interaction Data Set

The 100-Car Study (Dingus et al., 2004) captured 9,125 incidents. These 9,125 incidents were divided into four categories: LV-LV Interactions, LV-HV Interactions; Single Vehicle Conflicts, and Other. Table 4 provides a description of the different vehicle types in each category.

Table 4. Vehicle Types Captured in the 100-Car Study.

The data set used in the current effort was comprised of a subset of incidents from the 9,125 incidents described above. The 9,125 incidents were reviewed and only those that involved a LV-HV interaction were included in the present analysis.

Figure 6 shows a pie chart of the 9,125 events as a function of the vehicles involved and whether or not the incident was an interaction between vehicles. As can be seen, of the 9,125 events, 246 (2.7%) involved a LV-HV interaction. In 2003, there were a total of 6,328,000 crashes in the U.S. (NHTSA, 2004). Of these crashes, 313,663 (5%) were classified as a LV-HV interaction. Thus, the present data set has fewer LV-HV interactions than the national crash statistics.

Of the 246 LV-HV recorded incidents, 219 (89%) were crash-relevant conflicts, 25 (10.2%) were near crashes, 1 (.4%) was a crash, and 1 (.4%) was undetermined. For the 79 incidents where the HV driver was judged to have been at-fault, 66 (83.5%) were crash-relevant conflicts and 13 (16.5%) were near crashes. For the 138 incidents where the LV driver was judged to have been at-fault, 128 (92.8%) were crash-relevant conflicts, 8 (5.8%) were near crashes, 1 (.7%) was a crash, and 1 (.7%) was undetermined. For the 29 incidents where it was unknown if the LV or HV driver was at-fault, 29 (100%) were crash-relevant conflicts.

Figure 6 . Distribution of the 9,125 Incidents Captured in 100-Car Study.

Incident Types

Given that the 100-Car data set was comprised of 246 LV-HV interactions, the next step in the analysis was to determine the vehicles' actions for each incident. To this end, the video and relevant data for each incident were carefully reviewed and then classified as an "Incident Type." Twenty-seven different Incident Types were identified (a detailed description of each is presented in Table 5). It should be noted that the 27 Incident Types listed do not necessarily comprise the entire universe of all types of LV-HV interaction incidents. Rather, the 27 Incident Types listed comprise those that were identified in this data set (N = 246). The Incident Types are written in such a way as to be interchangeable regarding LVs and HVs. Note that this is the same classification strategy outlined in the Hanowski, Keisler, and Wierwille (2004) study. However, in the Hanowski, Keisler, and Wierwille (2004) study, only 20 Incident Types were identified in their data set.

This was a unique event that was not identified from any of the triggering methods. This event involved the driver reporting that he had received a ticket for illegally passing a stopped school bus. This self-report was then confirmed by reviewing the video.

Table 5. Description of the Incident Types that were Identified in the Current Research.

Table 6 shows the frequency, percentage, and rank ordering for the Incident Types across the entire 100-Car data set. The rank ordering highlights the frequency of Incident Types from most frequently occurring (ranked as a low number, "1") to least frequently occurring (ranked as a high number, "23.5"). Incident Types that had an equal number of occurrences were ranked as a "tie" and the mean of the rankings was assigned. For example, "Approaches Traffic Quickly," and "Roadway Entrance Without Clearance" occurred equally with a frequency of "6." Because their order in the ranking would consist of the ninth and tenth positions, a mean ranking of "9.5" was assigned to both Incident Types.

As can be seen from the data presented in Table 6, the most frequent Incident Type involving a LV-HV interaction was Late Braking for Stopped/Stopping Traffic. Across all 246 incidents, this particular Incident Type occurred 66 times and accounted for 26.8% of the incidents captured. The bar graph shown in Figure 7 illustrates the frequency and percentage of each Incident Type across the entire data set. As can be seen from Figure 7, the majority of the incidents (48.8%) involved one of two different Incident Types: Late Braking for Stopped/Stopping Traffic, and Lane Change Without Sufficient Gap.

Table 6. Frequency, Percentage, and Rank Ordering of the Incident Types Across all LV-HV Incidents (N Total = 246).

Figure 7 . Frequency of Incident Types Across all LV-HV Incidents (N Total = 246).

Descriptive statistics for the Incident Types were also calculated for incidents as a function of the at-fault driver. The at-fault driver is the driver that was assessed, by the analyst, to have been responsible for causing the event. Of the 246 LV-HV interaction incidents recorded, 138 (56%) were judged to have been the fault of the LV driver, while 79 (32%) were attributed to the HV driver. For the remaining 29 incidents (12%), it was unclear which vehicle driver was at-fault. By removing the "unknown" cases from the LV-HV driver at-fault analyses, it was found that the LV driver was at-fault in 64% (138/217) of the LV-HV interaction incidents while the HV driver was at-fault in 36% (79/217) of the incidents.

Table 7 shows the frequency, percentage, and rank ordering for the Incident Types where the HV driver was judged to be at-fault. As can be seen, the most frequent Incident Type for HV driver at-fault incidents was Lane Change Without Sufficient Gap (26.6%), followed by Lateral Deviation of Through Vehicle (21.5%), and Left Turn Without Clearance (13.9%). Figure 8 shows a bar graph of the 79 HV driver at-fault incidents as a function of the Incident Type.

Table 7. Frequency, Percentage, and Rank Ordering of the Incident Types for HV Driver At-Fault Incidents (n HV = 79).

Figure 8 . Frequency of Incident Types for HV Driver At-Fault Incidents (n HV = 79).

Table 8 shows the frequency, percentage, and rank ordering for the Incident Types where the LV driver was at-fault. The most frequent Incident Type for LV driver at-fault incidents was Late Braking for Stopped/Stopping Traffic (41.3%) and Lane Change Without Sufficient Gap (21.7%). Figure 9 shows a bar graph of the 138 LV driver at-fault incidents as a function of the Incident Type.

Table 8. Frequency, Percentage, and Rank Ordering of the Incident Types for LV Driver At-Fault Incidents (n LV = 138).

Figure 9 . Frequency of Incident Types for LV Driver At-Fault Incidents (n LV = 138).

Table 9 shows the frequency, percentage, and rank ordering for the Incident Types when the at-fault driver was unknown. The most frequent Incident Type for Unknown at-fault incidents was Late Braking for Stopped/Stopping Traffic (27.6%), followed by Conflict With Oncoming Traffic (13.8%), Lane Change Without Sufficient Gap (10.3%), and Unable to Determine (10.3%). Figure 10 shows a bar graph of the 29 Unknown at-fault incidents as a function of the Incident Type.

Figure 10 illustrates the Incident Types, with respect to the driver assessed to be at-fault, by group (HV, LV, and Unknown). The figure shows that the Incident Types differed depending on whether the HV or LV driver was at-fault. Across all at-fault incidents, the most frequent Incident Type were Late Braking for Stopped/Stopping Traffic, Lane Change Without Sufficient Gap, and Lateral Deviation of Through Vehicle.

Table 9. Frequency, Percentage, and Rank Ordering of the Incident Types for Unknown At-Fault Incidents (n Un = 29).

Figure 10 . Frequency of Incident Types for Unknown At-Fault Incidents (n Un = 29).

Figure 11 . Frequency of Incident Types for HV, LV, and Unknown At-Fault Incidents (n HV = 79, n LV = 138, and n Un = 29).

Summary of Incident Type

Overall, the most common Incident Types were Late Braking for Stopped/Stopping Traffic (26.8%), Lane Change Without Sufficient Gap (22%), and Lateral Deviation of Through Vehicle (8.1%). These three Incident Types represented 56.9% of the LV-HV incidents.

A substantial number of LV-HV interactions were judged to have been the fault of the LV driver. Of the 246 LV-HV interaction incidents, 56.1% (63.6%, excluding the incident where it was unknown if the HV or LV driver was at-fault) of the LV drivers were at-fault, 32.1% (46.4% excluding the Unknown at-fault incidents) of the HV drivers were at-fault, while in the remaining 11.8%, it was unknown if the HV or LV driver was at-fault.

As can be seen in Figure 11, the Incident Types differed depending on whether the HV or LV driver was at-fault. The most prevalent Incident Types for HV driver at-fault incidents were Lane Change Without Sufficient Gap (26.6%), Lateral Deviation of Through Vehicle (21.5%), and Left Turn Without Clearance (13.9%). These three incident types accounted for 51.7% of the HV driver at-fault incidents. The most frequent Incident Types for LV drivers were Late Braking for Stopped/Stopping Traffic (41.3%) and Lane Change Without Sufficient Gap (21.7%). These two incident types accounted for 63.0% of the LV driver at-fault incidents. Note that the most prevalent Incident Type for at-fault LV drivers, Lane Change Without Sufficient Gap, was similar to the truck driver focus groups' reports in Hanowski et al. (1998) who indicated that "being cut-off by LV drivers" was a frequently occurring problem in local/short-haul trucking.

Primary Maneuvers, Secondary Maneuvers and Conflict Types

After each of the 246 incidents was classified by Incident Type, the next step in the analysis was to identify the "Primary Maneuvers" and "Secondary Maneuvers" involved in each incident. The Primary Maneuver refers to the maneuver of the driver who initiated the incident (not necessarily at-fault). Table 10 shows each Primary Maneuver and its corresponding definition. Across the 246 interaction incidents, 19 different Primary Maneuvers were identified.

Table 10. List and Definition of Each Primary Maneuver Types.

In addition to identifying the Primary Maneuver for each incident, the Secondary Maneuver, or the maneuver of the responding driver (i.e., driver of the second vehicle involved in the interaction), was also classified. Considering the maneuvers of both vehicles involved in the incident, a clear picture of the conflict, or Conflict Type, could be determined. Table 11 shows the Conflict Types that were identified in the 246 interaction incidents that were analyzed. As can be seen, Table 11 consists of 66 different Conflict Types (i.e., Primary Maneuver and Secondary Maneuver combinations).

Table 11 . The 66 Different Conflict Types Identified Across all LV-HV Incidents.