Accident Analysis and Prevention 39 (2007) 1026–1036

Light truck vehicles (LTVs) contribution to rear-end collisions Rami Harb a,∗ , Essam Radwan b , Xuedong Yan a , Mohamed Abdel-Aty a a

b

Department of Civil and Environmental Engineering, University of Central Florida, United States Center for Advanced Transportation Systems Simulation, University of Central Florida, Orlando, FL 32816-2450, United States Received 28 February 2006; received in revised form 8 January 2007; accepted 19 January 2007

Abstract Light truck vehicles (LTVs), comprising light-duty trucks, vans, and sport-utility vehicles (SUVs) drive higher and wider than passenger cars which could affect the visibility for the following passenger car driver. This paper investigates the contribution of LTVs to rear-end collisions resulting from horizontal visibility blockage using the University of Central Florida sophisticated reconfigurable driving simulator. Indeed, a sudden stop of a leading LTV, in the shadow of the blindness of the succeeding passenger car driver, may deprive the latter of a sufficient response time, which may lead to high probability of a rear-end collision. To investigate this issue, two scenarios were developed in the UCF driving simulator. The first scenario serves as a base scenario where the simulator car follows a passenger car, and the second scenario serves as a test scenario, where the simulator car follows an LTV. The results obtained by comparing the scenarios showed that LTVs produce more rear-end collisions at unsignalized intersections due to horizontal visibility blockage and due to the resulting drivers’ behavior when driving behind an LTV. © 2007 Elsevier Ltd. All rights reserved. Keywords: Light truck vehicle (LTV); Rear-end collision; Horizontal visibility blockage; Driving simulator; Brake response time; Deceleration rate

1. Introduction According to the National Center of Statistics and Analysis (NCSA), in year 2003 alone, rear-end collisions accounted for one-third of the 6 million reported crashes nationwide in the U.S. (NCSA, 2001), which is the most abundant crash category according to Wang et al. (1999). Wiacek and Najm (1999) reported that rear-end collisions constitute 28% of all policereported crashes and that 29% of these rear-end crashes are associated with light truck vehicles (LTVs). LTVs include pickups, vans, truck-based station wagons, and utility vehicles and can weigh up to 10,000 pounds. According to Graham (2000), a major concern often voiced by the US motorists is that LTVs make it impossible for the following drivers in smaller vehicles to see the traffic ahead. Rear-end collisions are partly attributed to the traffic conditions misperception ahead. In fact, the National Transportation Safety Board (NTSB, 2001) investigated nine rear-end collisions in which 20 people died and 181 were injured. Common

∗

Corresponding author. Tel.: +1 407 823 5810/924 8594. E-mail addresses: [email protected] (R. Harb), [email protected] (E. Radwan), [email protected] (X. Yan), [email protected] (M. Abdel-Aty). 0001-4575/$ – see front matter © 2007 Elsevier Ltd. All rights reserved. doi:10.1016/j.aap.2007.01.007

to all crashes was the following vehicle drivers’ degraded perception of the traffic ahead. Moreover, Yan et al. (2005) stated that LTVs are 62.5% more likely to be rear-ended than large trucks and 17.7% more likely to be rear-ended than passenger cars. This issue is becoming a significant concern in the U.S. especially with the significant increase of LTVs on the US highways nowadays. In fact, Polk (2001) stated that for year 2000, motor vehicle registrations show 77.8 million light trucks in the U.S., a 63.8% increase from 1990. During the same period, there was a 1% decrease in the number of passenger cars (PCs). LTVs now represent 40% of all registered vehicles. Moreover, Acierno et al. (2004) related the mismatch in weight, stiffness, and height between the LTVs and the PCs to the increase of fatalities among PC drivers when their vehicles collide with LTVs. The degraded traffic perception caused by LTVs is defined in this study as horizontal view blockage. The horizontal view blockage occurs when a driver’s visibility is inhibited to his left or/and right at an intersection. This can occur when someone is driving a passenger car, which could be any sedan type car such as Saturn, Honda Accord, or Ford Taurus, closely behind a light truck vehicle (LTV), such as vans and SUVs. The driver’s sight view in the following passenger car may be restricted and the driver may not be able to see and know what is happening beyond the LTV at the intersection. For instance, if a pedestrian invisible to the driver in the following vehicle suddenly crosses

R. Harb et al. / Accident Analysis and Prevention 39 (2007) 1026–1036

the intersection from left to right, the LTV driver would be forced to slam on his brakes. This critical event may leave the following passenger car driver insufficient time to react appropriately and stop, which could lead to a rear-end collision with the leading LTV. Several studies examined the effect of LTVs on rear-end collisions. Abdel-Aty and Abdelwahab (2003) presented an analysis of the geometric incompatibility of LTVs on drivers’ visibility of other passenger cars involved in rear-end collisions with the objective to explore the effect of the size of the lead vehicle (i.e. height and width) on the rear-end crash configuration. The results based on the calibrated nested logit model suggested that LTVs block regular passenger car drivers’ visibility and that drivers behind LTVs are more susceptible to collide with the LTV in case of a sudden application on the brakes. Furthermore, Abdel-Aty and Abdelwahab (2004) concluded that there is a higher probability for a rear-end crash when a regular passenger car follows an LTV. Moreover, Sayer et al. (2000) examined the effect that the lead vehicle sizes such as height and width has on a passenger car driver’s gap maintenance under near optimal driving conditions characterized by daytime, dry weather, and free-flowing traffic. The data were obtained from a random sample of licensed drivers who drove an instrumented passenger car, unaccompanied, as their personal vehicle for 2–5 weeks. The results showed that passenger car drivers followed LTVs at shorter distance than they followed passenger cars. Also, the results of this study suggest that knowing the state of the traffic ahead of the lead vehicle, even by only one additional vehicle, affects gap length. Specifically, it appears that when dimensions of the lead vehicles permit the following drivers to see through, over, and around them, drivers maintain significantly longer distances. Previous studies entailing rear-end collisions analysis and modeling were typically conducted based on traffic crash databases (e.g. Wiacek and Najm, 1999; Sayer et al., 2000; Abdel-Aty and Abdelwahab, 2003, 2004; Yan et al., 2005). However, these databases do not provide behavioral information such as deceleration rates, response times, gap distance and time headway. To collect these behavioral measures that cannot be collected by the general databases (i.e. FARS, GES) and to replicate dangerous situations into a safe environment, several researchers employed driving simulators. For instance, Kumar et al. (2005) studied whether drivers of different ages can scan beyond the right edge and the left occluding edge of a leading truck using a driving simulator. Their results showed that up to 50% of the drivers scanned beyond the right edge of the leading truck and up to 70% scanned the left edge of the leading truck. Lee et al. (2002) studied the efficacy of their rear-end collision avoidance system (RECAS) using a high fidelity motion-based driving simulator. Similarly to Pradham et al. and Lee at al., we used a high fidelity driving simulator for this study. The main objective of this paper is to investigate whether LTVs contribute to rear-end collisions due to horizontal view blockage and to analyze drivers’ behavior model at intersections, including speeds and gaps using the UCF driving simulator. For that purpose, two scenarios were developed in the UCF driving simulator. The first scenario serves as a base scenario where the

1027

simulator car follows a passenger car, and the second scenario serves as the test scenario, where the simulator car follows an LTV. 2. Methodology 2.1. Participants A pilot study was conducted to determine the required sample size of the participants to be recruited and to test the adequacy of the data collection method. 2.1.1. Pilot study In the pilot study 10 subjects were recruited to complete the experiment. Five subjects drove the base scenario than the test scenario and the five remaining subjects drove the test scenario than the base scenario. Five out of 10 subjects driving behind the LTV were involved in a rear-end collision and 2 out of 10 subjects driving behind the passenger car were involved in a rear-end collision. A chi-square test conducted to compare the conditional probabilities showed that there was no statistically significant difference (P = 0.138) between the rear-end collision ratios at 0.05 significance level. The required sample size for the formal study was calculated as follows: n=

(Zα + Zβ )2 (p1 q1 + p2 q2 ) (p1 − p2 )

(1)

where n is the estimated necessary sample size, Zα the Zcoefficient (type I) error with 95% confidence interval, Zα the 1.96, Zβ the Z-coefficient (type II) error with 95% confidence interval, Zβ the 1.64, p1 the total number of collision with LTV over the total number of trials (5/10 the 0.5), p2 the total number of collision with PC over the total number of trials (2/10 the 0.2), q1 the total number of “no collision” with LTV over the total number of trials (1 − p1 ), and q2 is the total number of “no collision” with PC over the total number of trials (1 − p2 ). From the above equation the minimum required sample size to distinguish a difference for 95% significance level is 13 and 25 for 99% significance level. To be more conservative, we recruited 40 subjects to complete the experiment. 2.1.2. Recruited subjects As concluded from the pilot study, one subject cannot drive more than one scenario. In fact, the only difference between the base and the test scenarios is the leading vehicle type. If a subject drives both scenarios simultaneously, he/she may expect the same event and therefore be prepared for it. Therefore, we recruited two matching groups A and B with the gender and age as matching factors. Each group consisted of 20 subjects. Subjects of groups A drove the base scenario (following a passenger car) and subjects of group B drove the test scenario (following an LTV). The young age group varies between the ages of 18 and 25 and the middle age group varies between 26 and 55. Each group consisted of five young males and seven middle age males, three young females and five middle age females. The mean age of subjects from groups A and B were 30.8 and 32.75, the age stan-

1028

R. Harb et al. / Accident Analysis and Prevention 39 (2007) 1026–1036

• Scenario Editor: The software helps researchers to edit a tested traffic scenario. • APIs for reading real-time data: APIs (Application Programmer Interface) can read the real-time data from Simview. The sampling frequency is 60 Hz. 2.3. Scenarios

Fig. 1. UCF driving simulator cab.

dard deviations were 10.67 and 12.45, and the age ranges were 32 and 36, respectively. The recruited subjects were either fulltime students or employees at the University of Central Florida and held a valid Florida’s driver’s license with at least 2 years of driving experience. The experiment lasted for about 30 min in total and the participants were compensated 20 dollars for their participation. All participants had normal or corrected to normal eyesight. 2.2. Apparatus/equipment This study used the UCF (University of Central Florida) driving simulator (Klee, 2003) as a tool for data collection. The driving simulator, shown in Fig. 1, is an I-Sim Mark-II system with a motion base capable of operation with 6 degrees of freedom. It includes five channels (one forward, two side views and two rear view mirrors) of image generation, an audio and vibration system, and steering wheel feedback. The simulated environment is projected at 180◦ of field view as shown in Fig. 1 and at a resolution of 1280 × 1024 pixels. The driving simulation system is composed of the following components: • Simulator cab: Saturn Sedan, automatic transmission, air conditioning, the left back mirror and the back mirror inside the cab. • Simview: The software provides the graphical display based on the computation. • Motion base: It provides motion when driving.

In order to determine whether LTVs contribute to rear-end collisions, two scenarios were designed and analyzed in the driving simulator (that was between-subject). The first scenario serves as a base scenario where the simulator car drives behind a regular passenger car, and the second scenario serves as a test scenario where the simulator car drives behind an LTV. A typical rear-end collision due to horizontal view blockage occurs as the procedure illustrated in Fig. 2. Initially, the leading vehicle (vehicle 1 in Fig. 2) is traveling at a cruising speed followed by another vehicle (vehicle 2 in Fig. 2) keeping following-car headway. At time T0, an opposing vehicle (vehicle 3 in Fig. 2) unexpectedly and suddenly turns left in front of the leading vehicle. At time T1, the driver in the leading vehicle (vehicle 1) starts to sharply decelerate to avoid collision with vehicle 3. For the following vehicle (vehicle 2), there are two possibilities in response to this event. First, the driver of vehicle 2 could not see what happened beyond vehicle 1, and he/she decelerates at time T2, after response time (T2 − T1), to avoid collision with vehicle 1 once realizing the latter’s urgent deceleration. Fig. 2 illustrates only the first possibility. Second, the following driver could see vehicle 3 making a sudden left turn at time T0 and realizes the potential danger ahead, therefore he/she decelerates at T3 after his/her response time (T3 − T0). Generally, T3 is earlier than T2 and even maybe earlier than T1 because the following vehicle’s driver also makes a direct response to the first event that happened in front of the leading vehicle. Then, the time interval T3 − T1 (can be a negative value) is smaller than T2 − T1. This possibility and T3 are not illustrated in Fig. 2. The hypothesis in this study is that the driver’s horizontal sight view in the following car is more likely to be restricted by a leading LTV compared to a passenger car. Therefore, a rear-end collision is anticipated to occur when the time it takes the following vehicle to come to a complete stop is greater than the time headway between the two vehicles. The driving course of each scenario can be described in three stages as shown in Fig. 3. In the first stage, the driver in the simulator car cruises on a four-lane urban road with a 72.42 km/h (45 mph) posted speed limit. The traffic in the scene is assigned to flow at 72.42 km/h (45 mph). The purpose of this design is to make the simulator car drivers adapt to relatively higher speed traffic. At the second stage, the simulator car approaches the signalized intersection and stops at the red phase behind the LTV, which is assigned to be there. When the light turns green, the LTV is assigned to cruise at a 56.33 km/h (35 mph), following the speed limit, while the following vehicle (simulator car), accustomed to a higher speed limit follows the LTV at a speed tending to be greater than the speed limit. Moreover, the two-lane road in the direction of the simulator is dropped to one lane to inhibit any passing between vehicles. Therefore, the

R. Harb et al. / Accident Analysis and Prevention 39 (2007) 1026–1036

1029

Fig. 2. Rear-end collision process.

simulator car driver is forced to drive behind the LTV until the two-way stop intersection in the third stage. It should be mentioned that rear-end collisions would occur at stage #3 of the experiment subsequently to T0 and T1 and T2. For the design of the experiment, the width of the LTV was 1.88 m (6.17 ft) while the width of the passenger car was 1.70 m (5.58 ft) and the assigned deceleration rate for the leading vehicle is 0.8 g. It should be mentioned that many vehicles were set to drive on the pre-scheduled path uninterrupting the experiment in addition to pedestrian traffic, road signs, pavement markings, and parked vehicles in order to make the driving environment more realistic as shown in Fig. 3. The configurations of the roads (i.e. width, pavement markings) in addition to the unsignalized intersection and the road signs were designed per standards and specification/standards of the FDOT, MUTCD, and AASHTO. 2.4. Procedure Upon arrival, the subjects were asked to fill out and sign an informed consent form (per IRB). This form explained briefly how the driving simulator operates and its general purpose in

traffic engineering. The drivers were advised to drive and behave as they normally would and to adhere to traffic laws as in real life situations. The drivers were also notified that they could quit the experiment at any time in case of motion sickness or any kind of discomfort. Prior to the formal experiment, drivers were trained on the driving simulator for 5 min where the drove on different roads to familiarize with the deceleration and acceleration rates of the simulator car. Then, the participants were instructed to drive straight throughout the experiment and to follow the posted speed limits, as shown in Fig. 3, without knowing its research purpose or the type of leading vehicle and its rationale. The experiment (both scenarios) ends at stage #3 – after the induced sudden brake of the simulator car – whether a rear-end occurred or not. Subsequently, each subject took a survey regarding the experiment. It should be mentioned that two participants suffered light simulator motion sickness during the training session which disabled them from qualifying to the formal experiment. Moreover, the 40 subjects that completed the experiment, out of the 42 recruited subjects, confirmed that they did not experience any type of motion sickness.

1030

R. Harb et al. / Accident Analysis and Prevention 39 (2007) 1026–1036

Fig. 3. Experimental design stages.

2.5. Independent and dependent variables The independent variables are age and gender of the subjects, and the type of the leading vehicle; PC or LTV. The dependent variables are listed below. The first dependent variable is the deceleration rate of the simulator car. The deceleration rate was recorded for two cases at stage #3 of the experiment. In the first case where there were no rear-end collisions, the deceleration rate was measured for the following vehicle’s speeds ranging from the speed at the instant when the simulator driver starts braking to a speed of 8.05 km/h (5 mph). Zero km/h (0 mph) was not used because a large number of drivers maintained a crawling speed until they finished the experiment, which would have biased the experiment results. In the second case where there were rear-end accidents, the deceleration rate was measured for the following vehicle’s speeds ranging from the speed at the instant when the simulator driver starts braking to the impact speed. The second dependent variable is the reaction response time. The reaction response time is the difference between the instant the vehicle from the opposite direction makes a sudden left turn and the instant the simulator car starts braking (or sharply decelerates). A sharp deceleration rate was defined as when the brake input – recorded in percentage of the current position of the brake pedal relative to its maximum position – is equal to 20% or larger. The reader should be cautioned that the reaction response time is different than the drivers’ reaction time. The reaction response time (T0 − T2) is a surrogate measure of the ahead traffic degraded perception. In other words if T0 − T2 for following an LTV is larger than T0 − T2 for following a PC, it means that drivers behind LTVs

may have reacted to the sudden stop of the leading LTV instead of the opposing car making a sudden left turn. This fact implies that drivers behind LTVs did not perceive the opposing vehicle making a sudden left turn due to the horizontal visibility blockage caused by the LTV. The third dependent variable is the gap distance. The gap distance was measured as the headway between the leading vehicle and the following vehicle when the leading vehicle starts braking. The fourth dependent variable was the time headway which is defined by many researchers (e.g. Vogel, 2002) as the time that passes between two vehicles’ reaching the same location and as one of the indicators used to estimate the criticality of a certain traffic situation. The time headway was calculated by dividing the gap distance by the cruising speed. The fifth dependent variable is the cruising speed of the simulator. The cruising speed was measured as the cruising speed just before the simulator car starts braking. 2.6. Experimental design The design was a 2 (age groups) × 2 (gender) × 2 (type of leading vehicle) between-subject controlled experimental design. The test factor (between-subject factor) was the type of the leading vehicle (PC–LTV) with two levels: passenger car (PC) and light truck vehicle (LTV). It should be mentioned that the type of the following vehicle (simulator car) is always a passenger car (Saturn Sedan). The age and gender factors were used as matching criteria (control criteria) to select matching subjects for groups A and B. The dependent variables of interest were concerned with the drivers’ behavior and actions resulting from the horizontal visi-

R. Harb et al. / Accident Analysis and Prevention 39 (2007) 1026–1036

1031

6 − V), the distance between the eye of the driver and the center of the lane. From AASHTO (2001) the average distance between the eye of the driver to the outer edge of the car is 1.5 ft as shown in Fig. 4. Therefore V is equal to 1.70/2 m − 0.46 m = 0.39 m or ((5.58 ft/2) − 1.5 ft) = 2.79 ft − 1.5 ft = 1.29 ft). Therefore, MK = 1.83 − 0.39 m = 1.44 m or (6–1.29 = 4.71 ft); KL the distance between the center of the road and the center of the lane = 1.83 m or (6 ft); KE the distance between the center of the lane and the edge of the opposing vehicle = 1.70 m/2 = 0.85 m or (5.58/2 = 2.79 ft). Rearranging the above equation with ME = 4.11 m (13.5 ft): AM = Fig. 4. Theoretical calculations.

−4.11 × BM , BC − 4.11

BM = vt +

v2 + w + L + D (3) 2a

The theoretical sight distance analysis is essential for the design of the scenarios in the driving simulator and it provides an approximation of the distance at which the simulator car driver would not be able to perceive the left turning vehicle for following an LTV and following a PC. From Fig. 4, the minimum required distance, in order to see the opposing vehicle making a left turn, between the eye of the driver and the rear-bumper of the leading vehicle was calculated. In the calculations it is assumed that all vehicle drive in the center of the lane. In triangles ABC and AME the following ratios can be applied:

where w is the width of the intersection 7.32 m (24 ft) (each lane is assumed to be 3.66 m (12 ft)), L the length of the vehicle taken 9.14 m (30 ft) (AASHTO, 2001), D the set back of the stop bar from the intersection, which is 3.05 m (10 ft) (AASHTO, 2001), t the standard response time which is 1.0 s (AASHTO, 2001), a the acceleration rate taken 3.05 m/s2 (10 ft/s2 ) (AASHTO, 2001), AG the distance from the center of the car to the back of the front vehicle in m; BM the distance the leading vehicle needs to clear the unsignalized intersection and v is the speed of the vehicle taken 56.33 km/h (35 mph) or 15.65 m/s (51.33 ft/s). Table 1 shows the variation of AM with the variation of BC, the width of the leading vehicle. In the design the leading LTV width is 1.88 m (6.17 ft) therefore, the required distance AG in order for the following driver to see the vehicle making a left turn is 14.79 m (48.53 ft). Completing the same calculations for the base scenario where the leading vehicle is a passenger car with a width of 1.70 m (5.58 ft), the resulting minimum required distance AG is 12.01 m (39.4 ft).

BC AB AM − BM = = ME AM AM

3. Results

bility blockage due to LTVs. These can be analyzed by looking at the deceleration rates, reaction response time, cruising speed, gap distance, and time headway (dependent variables). 2.7. Theoretical analysis of sight distance model

and

(BC × AM) = ME, (AM − BM)

ME = MK + KL + KE

3.1. Rear-end collisions

= 1.44 m (4.71 ft) + 1.83 m (6 ft) + 0.85 m (2.79 ft) = 4.11 m (13.5 ft)

(2)

where MK is the distance between the eye of the driver and the center of the road (median) which is equal to the distance between the center lane and the center of the road minus V (or

As anticipated in the design of the experiment, all rear-end collisions whether with a leading PC or LTV occurred at stage #3 of the experiment past times T0, T1, and T2. Two subjects out of 20 subjects driving the simulator behind the PC were involved in a rear-end collision with the PC. However, 8 subjects out of

Table 1 Variation of AM and AG with the variation of BM v (m/s)

t (s)

A (m/s2 )

w (m)

L (m)

D (m)

2 × BC (m)

ME (m)

BM (m)

AM (m)

AG (m)

15.65 15.65 15.65 15.65 15.65 15.65 15.65 15.65 15.65 15.65 15.65

1 1 1 1 1 1 1 1 1 1 1

3.05 3.05 3.05 3.05 3.05 3.05 3.05 3.05 3.05 3.05 3.05

7.32 7.32 7.32 7.32 7.32 7.32 7.32 7.32 7.32 7.32 7.32

9.14 9.14 9.14 9.14 9.14 9.14 9.14 9.14 9.14 9.14 9.14

3.05 3.05 3.05 3.05 3.05 3.05 3.05 3.05 3.05 3.05 3.05

1.70 1.88 1.98 2.13 2.29 2.44 2.59 2.74 2.90 3.05 3.20

4.11 4.11 4.11 4.11 4.11 4.11 4.11 4.11 4.11 4.11 4.11

75.31 75.31 75.31 75.31 75.31 75.31 75.31 75.31 75.31 75.31 75.31

89.12 90.88 91.90 93.49 95.13 96.82 98.58 100.41 102.30 104.27 106.31

27.16 27.70 28.01 28.49 28.99 29.51 30.05 30.60 31.18 31.78 32.40

1032

R. Harb et al. / Accident Analysis and Prevention 39 (2007) 1026–1036

Table 2 Logistic regression analysis (binary dependent variable: rear-end/ no rear-end) Wald-χ2

Parameter Main effect (likelihood ratio Gender Age PC–LTV Interactions Age × gender Age × PC–LTV Gender × PC–LTV

χ2

P-value

Significant

3.2. Drivers’ behavior (objective measures)

= 6.5246; P-value = 0.0106) 1.5049 0.2199 0.0722 0.7882 5.4153 0.02

No No Yes

0.4652 0.3011 1.5958

No No No

0.4952 0.5832 0.2065

7.364 times more likely to be involved in a rear-end collision with a leading LTV compared to a leading PC (P = 0.0132).

the 20 subjects driving the simulator car behind an LTV were involved in a rear-end collision with the LTV. It should be cautioned that the probabilities of rear-end collisions based on the experiment results are higher than those in a real life because the probabilities are conditional probabilities which resulted from the critical traffic event. The probability of being involved in a rear-end collision with a PC is 0.1 (PPC = 2/20 = 0.1) and with an LTV is 0.4 (PLTV = 8/20 = 0.4). Since the dependent variable is binary (accident/no accident), a logistic regression analysis was conducted to assess the effect of the independent variables age, gender, and type of the leading vehicle (PC or LTV) on the rear-end occurrence. The reader may refer to Cox (1966), Day and Kerridge (1967), and Anderson (1972) for detailed information on logistic regression. As illustrated in Table 2, the type of the leading vehicle PC or LTV was the only significant variable (Wald-χ2 = 5.4153, P = 0.02). Moreover, according to the logistic regression, the odds ratio for LTV compared to PC was 7.364. In other words, passenger car drivers behind an LTV are

The correlation test results between the dependent variables (i.e. deceleration rates, response reaction time, speed, gap-distance, and gap-time) are summarized in Table 3. The reaction response time is negatively correlated with speed (Pearson correlation = −0.498, P = 0.001) and positively correlated with gap distance (Pearson correlation = 0.497, P < 0.001) and time headway (Pearson correlation = 0.583, P = 0.001). The deceleration rate is positively correlated with speed (Pearson correlation = 0.366, P = 0.02) and negatively correlated with gap distance (Pearson correlation = −0.604, P < 0.001) and time headway (Pearson correlation = −0.602, P < 0.001). These correlations between reaction response time and deceleration rates with speed, time headway and gap distance are inherent. The relationship between the driver behavior measurements (dependent variables) and the independent variables (age, gender, and type of leading vehicle) are analyzed in the following section. 3.2.1. Speed According to the summary statistic shown in Table 4 and the three-factor ANOVA analysis results, the mean speed for following a PC (52.37 km/h) is significantly lower (F(1,33) = 4.91, P = 0.034) than the mean speed for following an LTV (55.20 km/h) (see Fig. 5). This fact suggest that drivers behind LTVs may be uncomfortable due to the ahead traffic degraded perception and therefore drive at a higher speed (compared to PC) with the intention to pass the latter. Age

Table 3 Dependent variables correlation RTIME

DEC

SPEED

GAPDIST

GAPTIME

1.000

−0.077 0.637 40

−0.498** 0.001 40

0.497** 0.001 40

0.583** 0.000 40

0.366* 0.020 40

−0.604** 0.000 40

−0.602** 0.000 40

1.000

−0.431** 0.006 40

−0.608** 0.000 40

1.000

0.973** 0.000 40

RTIME Pearson correlation Significance (two-tailed) N

40

DEC Pearson correlation Significance (two-tailed) N

−0.077 0.637 40

40

SPEED Pearson correlation Significance (two-tailed) N

−0.498** 0.001 40

0.366* 0.020 40

40

GAPDIST Pearson correlation Significance (two-tailed) N

0.497** 0.001 40

−0.604** 0.000 40

−0.431** 0.006 40

40

GAPTIME Pearson correlation Significance (two-tailed) N

0.583** 0.000 40

−0.602** 0.000 40

−0.608** 0.000 40

0.973** 0.000 40

* **

Correlation is significant at the 0.05 level (two-tailed). Correlation is significant at the 0.01 level (two-tailed).

1.000

1.000 40

R. Harb et al. / Accident Analysis and Prevention 39 (2007) 1026–1036

1033

Table 4 Dependent variables vs. independent variables statistics summary (DEC × PC–LTV, RTIME × PC–LTV, SPEED × PC–LTV, GAPDIST × PC–LTV, GAPTIME × PC–LTV) Factor, PC–LTV

DEC (m/s s)

RTIME (s)

SPEED (km/h)

GAPDIST (m)

GAPTIME (s)

PC–LTV PC Mean N S.D.

5.42 20 1.51

2.40 20 1.22

52.37 20 4.11

34.94 20 16.95

2.46 20 1.33

LTV Mean N S.D.

6.78 20 1.04

2.23 20 0.50

55.20 20 2.52

23.03 20 9.46

1.51 20 0.60

Gender Female Mean N S.D.

6.44 16 1.40

2.50 16 1.16

53.76 16 4.79

29.89 16 16.32

2.07 16 1.31

Male Mean N S.D.

5.86 24 1.47

2.19 24 0.73

53.80 24 2.77

28.38 24 14.08

1.93 24 1.01

Age Young Mean N S.D.

6.24 16 1.58

2.27 16 0.81

54.20 16 2.58

27.93 16 15.65

1.87 16 1.06

Middle Mean N S.D.

6.00 24 1.40

2.34 24 1.01

53.51 24 4.26

29.69 24 14.56

2.06 24 1.19

6.10 40 1.46

2.31 40 0.92

53.79 40 3.66

28.98 40 14.83

1.98 40 1.13

Total Mean N S.D.

(F(1,33) = 0.38, P = 0.542), gender (F(1,33) < 0.1, P = 0.989), and interactions between these independent variables do not affect speed significantly. 3.2.2. Gap distance As shown in Table 4 and by the three-factor ANOVA analysis results, the mean gap distance for following a PC (34.94 m) is significantly larger (F(1,33) = 6.17, P = 0.018) than the mean gap distance for following an LTV (23.03 m) (see Fig. 6). This fact also proposes that drivers behind LTV may be uncomfortable due to the ahead traffic degraded perception and therefore keep a short gap awaiting a chance to pass the LTV. This ANOVA analysis also showed that age (F(1,33) = 0.09, P = 0.87), gender (F(1,33) = 0.73, P = 0.394), and interactions do not affect the gap distance significantly.

Fig. 5. Speed for following PC and LTV.

3.2.3. Reaction response time and ratio The three-factor analysis of variance demonstrated that age (F(1,33) = 0.089, P = 0.416), gender (F(1,33) = 0.678, P = 0.768), or type of the leading vehicle (F(1,33) = 0.117, P = 0.735) do not

1034

R. Harb et al. / Accident Analysis and Prevention 39 (2007) 1026–1036

Fig. 6. Gap-distance for following PC and LTV.

Fig. 7. Time headway for following PC and LTV.

affect the reaction response time significantly. From Table 4, the mean reaction response time for following an LTV and a PC (2.23 and 2.40 s, respectively) may seem counterintuitive. However, this analysis may have resulted in a biased conclusion because some subjects kept larger gaps behind PC than LTV, which affected the response reaction time. Generally, a larger gap accommodate driver to have more space and time to make a stop decision slowly. There were no significant interactions between age, gender, and type of leading vehicle. Therefore, the time headway (see Fig. 7) was used as exposure to calculate the ratio which is the reaction response time over the time headway (in seconds). According to the threefactor analysis of variance, gender (F(1,33) = 3.01, P = 0.092) and age (F(1,33) = 0.83, P = 0.369) were not significant. The mean ratio for following an LTV (0.110) is significantly (F(1,33) = 4.72, P = 0.037) larger than that for following PC (0.0784) at a 0.05 significance level. This trend suggests that the drivers following LTVs have significantly larger reaction response time to avoid the rear-end accident compared to those following PCs using time headway as exposure. The drivers following a PC may benefit from a better front sight view to perceive the potential

accident risk. No interactions between age, gender, and type of leading vehicle were found significant. 3.2.4. Deceleration rates The deceleration rate, as mentioned before, is correlated with speed, gap distance, and time headway. These variables may have a significant effect on the deceleration rate. The reaction response time is not correlated (see Table 4) with the deceleration rate. However, the ratio defined in the previous section is correlated with deceleration rate (Pearson correlation = 0.543, P < 0.001). Therefore, since the ratio combines all the dependent variables correlated with the deceleration rate (speed, gap distance, time headway) and is correlated with the deceleration rate, a three-factor analysis of variance (ANOVA) was conducted to test the effect the independent variables PC–LTV, gender, and age on the deceleration rate while using the ratio as a covariate. As shown in Tables 4 and 5, the ANOVA results show that the deceleration rate for following LTV (6.78 m/s s) is significantly larger (P = 0.038) than following a PC (5.42 m/s s) (see Fig. 8). Moreover, the results showed that a higher ratio (covariate) –

Table 5 ANOVA analysis for deceleration rate using ratio as covariate Source

Type III sum of squares

Tests of between-subjects effects (dependent variable: dec) Corrected model 41.441a Intercept 126.775 Ratio 16.035 PC–LTV 6.215 Gender 0.202 Age 0.007 PC–LTV × gender 0.182 PC–LTV × age 0.412 Gender × age 2.332 PC–LTV × gender × age 4.682 Error 41.148 Total 1568.804 Corrected total 82.589 a

R2 = 0.502 (adjusted R2 = 0.373).

d.f.

Mean square

F

Significant

8 1 1 1 1 1 1 1 1 1 31 40 39

5.180 126.775 16.035 6.215 0.202 0.007 0.182 0.412 2.332 4.682 1.327

3.903 95.509 12.080 4.682 0.152 0.006 0.137 0.310 1.757 3.527

0.003 0.000 0.002 0.038 0.699 0.941 0.714 0.581 0.195 0.070

R. Harb et al. / Accident Analysis and Prevention 39 (2007) 1026–1036

1035

opposite vehicle making a sudden left turn were involved in a rear-end collision. Answering the second question, 65% of the subjects from group B said that they encountered similar visibility problem at least once in real life and 35% said that they never encountered similar visibility problems in real life. 4. Conclusions and discussions

Fig. 8. Deceleration rates for following PC and LTV.

or in other words higher reaction response at same time headways – may result in higher deceleration rate (P = 0.002) and vice versa. 3.2.5. Impact speeds The impact speed was measured as the speed of the simulator car just before it collides with the leading vehicle. The mean impact speed (eight impacts with LTV) for following LTV (30.79 km/h) is higher than the mean impact speed (two impacts with PC) for following PC (22.25 km/h). Since the impact sample size with PC and LTV is small (two impacts with PC versus eight impacts with LTVs) statistical inference about the difference in impact speeds cannot be made. However, this trend suggests that rear-end collisions with LTVs may be more severe than rear-end collisions with PC. 3.3. Survey analysis (subjective measures) As mentioned before, participants from group A (driving behind PC) and B (driving behind a LTV) completed a questionnaire at the end of the experiment. In addition to simulator assessment questions, two questions were asked specifically about this experiment for all participants: (1) Did you see the car (vehicle #3 in Fig. 2) making a left turn from the opposite side? (same question for both groups) (2) Do you encounter similar visibility issues in real life? (for group B) Answering the first question, 10 subjects following LTV answered that they did not see the vehicle from the opposite direction making a left turn and 6 subjects following the passenger car answered that they did not see the vehicle from the opposite direction making a left turn. It should be noted that 7 subjects out of the 10 subjects (following LTV) confirming not seeing the opposite vehicle making a sudden left turn were involved in a rear-end collisions. Moreover, two subjects out of the six subjects (following PC) corroborating not seeing the

One of the paper’s objectives was to study whether driving behind an LTV increases the probability of rear-end collisions. From the conducted analyses, it was confirmed that LTVs contribute to the increase of rear-end collisions probability in the critical traffic event. This result based on the driving simulator experiment complied with the results obtained by previous researchers (Abdel-Aty and Abdelwahab, 2003; Abdelwahab and Abdel-Aty, 2004). Another objective was to investigate the behavior of drivers behind LTVs that contributed to the growth of rear-end risk. From the experiment’s results, the time headway and gap distances for following LTV are significantly smaller than those for following a PC and the speeds for following an LTV are slightly larger than those for following a PC. These results were consistent with Sayer’s (2000) results, which indicated that drivers follow LTVs at smaller gaps. Sayer (2000) relates the visibility degradation beyond the lead vehicle to headways. In fact, he states that when the dimensions of the lead vehicles permit the following drivers to see through, over, and around them, drivers maintain significantly longer distances. Moreover, Broughton et al. (2007) stated that drivers tend to drive faster and keep smaller headways during limited visibility conditions (i.e. foggy weather) beyond the lead vehicle. Evans (1991) discussed the two potential justifications for why drivers follow at unsafe headways. First, the drivers may have learned that the relative speed between themselves and the vehicle ahead rarely changes suddenly. Second, drivers may have learned to feel safe following too closely because they have repeatedly done so without unpleasant consequences. However, Evans (1991) did not distinguish between the type and the size of the lead vehicles and their effect on headways. Dingus (1997) reported that the headway maintenance can increase by 0.5 s if the appropriate in-vehicle visual display is used. The issue of smaller headways behind LTVs is still obscure; however, a possible reason could be related to the discomfort of PC drivers when they drive behind LTVs. In other words, PC drivers may feel uncomfortable behind PC and keep smaller headways awaiting a chance to pass them. The discomfort issue is not quantified in this study and could be subject for future investigations. The vehicle’s mean deceleration rate and the driver’s mean ratio (reaction response time standardized with time headway) for following an LTV are significantly larger than those for following a PC. The above results also indicated that a larger ratio may result in a higher deceleration rate. In other words, larger reaction response time at the same time headways may result in higher deceleration rate. This indicates that the horizontal visibility blockage which degraded the perception of the traffic ahead of the LTV may have increased the ratio (reaction response

1036

R. Harb et al. / Accident Analysis and Prevention 39 (2007) 1026–1036

time standardized with exposure) and subsequently resulted in higher deceleration rates to avoid the collision. Many researchers such as Kumar et al. (2005) segment their analysis by gender and age. For instance, Kumar et al. (2005) associate the age difference with the individual risk recognition. They state that older experienced drivers (60–75 years old) look more often at areas in which risky events may occur and respond accordingly more quickly than younger experienced drivers (19–29 years old) and novice drivers (16–17 years old in their first 6 months of driving experience). However, the results of our analyses showed insignificant difference in the age and gender as well as interactions. This fact may be due to the size of the sample which was calculated in the pilot study for significant difference between collision and no collision for following an LTV versus following a PC. The survey analysis was conducted with the intention of reinforcing the explanation of the results from objectives measures analysis. For instance, the initial hypothesis is that drivers behind LTVs were involved in rear-end collisions due to the degraded perception of the traffic ahead. In the survey analysis, seven out of the eight drivers involved in a rear-end collision with the LTV confirmed that they did not see the opposing vehicle making a sudden left turn due to the LTV. This key question validated our postulation and analysis. Also from the survey analysis, 65% of the subjects from group B (driving behind an LTV) confirmed that they encountered this kind of visibility issues in real life which underscores the importance of this issue. Some recommendations can be drawn based on the above findings. First, a possible solution for horizontal visibility blockage at unsignalized intersections could be the addition of a convex mirror on the right side of the road that would show what is happening beyond the LTV. This mirror would be slightly tinted to avoid the reflection of the sunlight and other vehicle lights. Convex mirrors are not deployed for such kind of line of sight obstruction and are subject to future investigation. Such mirrors are usually used on mountainous roads (in Europe and Tennessee) with significant sight distance problems as well as parking garages (Hazard safety, 2002). Another possible solution could include ITS technologies such as intervehicular warning that would be triggered when someone drives relatively too close to a leading larger vehicle at a relatively high speed. Finally another solution could be educating drivers to drive at relatively large distances behind leading vehicles especially LTVs. As a typical study based on a driving simulator, some limitations may exist since a validation study in the field was not conducted. In general, driving simulators are valid for relative comparison instead of absolute comparison. For instance, the deceleration rates, speeds, time headway and distance may not be equal in real-life and in the simulated conditions. However, the relative comparison and the conclusions are often reliable. References AASHTO, 2001. A policy on geometric design of highways and streets, Washington, DC.

Abdel-Aty, M.A., Abdelwahab, H.T., 2004. Analysis and prediction of traffic fatalities resulting from angle collisions including the effect of vehicle configuration and compatibility. Accid. Anal. Prev. 36, 457– 469. Abdel-Aty, M.A., Abdelwahab, H.T., 2003. Configuration analysis of twovehicle rear-end crashes. J. Transport. Res. Rec. 1840, 140–147. Abdelwahab, H.T, Abdel-Aty, M., 2004. Investigating the effect of light truck vehicle percentages on head-on fatal traffic crashes. J. Transport. Eng. (ASCE) 130 (4), 429–437. Acierno, S., Kaufman, Rivara, F.P., Grossman, D.C., Mock, C., 2004. Vehicle mismatch: injury patterns and severity. Accid. Anal. Prev. 36, 761– 772. Anderson, J.A., 1972. Separate sample logistic discrimination. Biometrika 59, 15–18. Broughton, K.L.M., Switzer, F., Scott, D., 2007. Car following decisions under three visibility conditions and two speeds tested with a driving simulator. Accid. Anal. Prev. 39, 106–116. Cox, D.R., 1966. Some procedure associated with the logistic qualitative response curves. In: David, D.F. (Ed.), Research Papers in Statistics. Wiley, New York, pp. 55–71. Day, N.E., Kerridge, D.F., 1967. A general maximum likelihood discriminant. Biometrics 23, 313–323. Dingus, T.A., 1997. Human factors field evaluation of automotive headway maintenance/collision warning devices. Hum. Factors 39 (2), 216–229. Evans, L., 1991. Traffic safety and the driver (book review). Ergonomics 36, 863–866. Graham, J., 2000. Civilizing the sport utility vehicle. Issue Sci. Technol. 17 (2), 57–62. Hazard safety, traffic calming and safety, access date October 2005, http://endat.co.uk/EWC/mainlisting1.cfm?ID=65522&AutoID=9570& Category=1060. Klee, H., 2003. Overview of driving simulator research capabilities at the University of Central Florida. In: Proceedings of 2003 Summer Computer Simulation Conference, Montreal, Canada, July 2003. Lee, J.D., McGehee, D., Brown, T.L., Reyes, M., 2002. Collision warning timing, driver distraction, and driver response to imminent rear end collision in a high fidelity driving simulator. Hum. Factors 44 (2), 314–334. National Center of Statistics and Analysis (NCSA), 2001. National Transportation Safety Board, 2001. Access date, May 2005, http://wwwnrd.nhtsa.dot.gov/departments/nrd-30/ncsa/. Vehicle and InfrastructureBased Technology for the Prevention of Rear-End Collisions, Special Investigation Report, Report number PB2001-917003, Washington, DC 20594. National Transportation Safety Board (NTSB), 2001. Vehicle and InfrastructureBased Technology for the Prevention of Rear-End Collisions, Special Investigation Report, Report number PB2001-917003, Washington DC 20594. Polk, 2001. Polk study shows SUV sales strong despite rising fuel costs. R.L. Polk & Company, January 1, 2001. http://www.polk.com/news/rleases/ 2001-0104.asp. Kumar, A.P., Hammel, K., DeRamus, R., Pollatsek, A., Noyce, D.A., Fisher, D.L., 2005. Using eye movements to evaluate effects of driver age on risk perception in a driving simulator. Hum. Factors, 840–852. Sayer, J.R., Mefford, M.L., Huang, R.W., 2000. The Effect of Lead-Vehicle Size on Driver Following Behavior (Report number UMTRI-2000-15). Ann Arbor, MI: The University of Michigan Transportation Research Institute. Vogel, K., 2002. A comparison of headway and time to collision as safety indicators. Accid. Anal. Prev. 35, 427–433. Wang, J., Knipling, R., Blincoe, L., 1999. The dimensions of motor vehicle crash risk. J. Transport. Stat. 2, 19–43. Wiacek, C.J., Najm, W.G. 1999. Driver/vehicle characteristics in rear-end pre crash scenarios based on the general estimates (GES). SAE 1999-01-0817, Society of Automotive Engineers, pp. 99–103. Yan, X., Radwan, E., Birriel, E., 2005. Analysis of rear-end accidents at signalized intersections crashes using multiple logistic regression method. Accid. Anal. Prev. 37, 983–995.