Transportation Research Part F 10 (2007) 229–241 www.elsevier.com/locate/trf

Larger size vehicles (LSVs) contribution to red light running, based on a driving simulator experiment Rami Harb

a,*

, Essam Radwan b, Xuedong Yan

a

a

b

Department of Civil and Environmental Engineering, CATSS, University of Central Florida, 4000 Central Florida Blvd., Orlando, FL 32816, United States Center for Advanced Transportation Systems Simulation, University of Central Florida, Orlando, FL 32816-2450, United States Received 17 January 2006; received in revised form 4 October 2006; accepted 20 October 2006

Abstract Larger size vehicles (LSVs), comprising semis and six-wheelers drive higher and wider than passenger cars which could affect the visibility of traffic lights for the following passenger car driver at signalized intersections. This paper investigates the contribution of LSVs to red light running resulting from vertical visibility blockage based on a driving simulator experiment. Three sub-scenarios were developed in the driving simulator. The first sub-scenario served as a base sub-scenario, where the simulator car follows a passenger car. The second sub-scenario served as the test sub-scenario, where the simulator car tracks an LSV. The results obtained by comparing the sub-scenarios confirmed that LSVs generate higher red light running rate at signalized intersections due to vertical visibility blockage. The third sub-scenario (or suggested solution sub-scenario) tested a suggested solution by the Federal Highway Administration for the vertical visibility blockage issue, which consists of installing an additional signal pole on the right side of the road. The comparison between the test sub-scenario and the suggested solution sub-scenario demonstrated that the addition of a signal pole on the right side of the road decreases red light running resulting from vertical visibility blockage significantly. Ó 2006 Elsevier Ltd. All rights reserved. Keywords: Larger size vehicle (LSV); Red light running; Vertical visibility blockage; Driving simulator; Brake response time; Deceleration rate

1. Introduction A Federal Highway Administration report titled ‘‘Stop Red Light Running Facts and Statistics’’ (FHWA, 2003a) stated that according to the General Estimates System of National Sampling System (GES), more than 1.8 million crashes occur at intersections each year. Of those, in 2003, about 206,000 were due to red light running resulting in 934 deaths and approximately 176,000 injuries. According to the Fatality Analysis Reporting System (FARS), on a national basis, fatal motor vehicle crashes at signalized intersections increased 13.2%

*

Corresponding author. Tel.: +1 407 9248954; fax: +1 407 8234676. E-mail addresses: [email protected] (R. Harb), [email protected] (E. Radwan), [email protected] (X. Yan).

1369-8478/$ - see front matter Ó 2006 Elsevier Ltd. All rights reserved. doi:10.1016/j.trf.2006.10.005

230

R. Harb et al. / Transportation Research Part F 10 (2007) 229–241

between 1993 and 2003, far outpacing the 6.6% mount in all other fatal crashes. Based on the same database, researchers in the Insurance Institute for Highway Safety reported that during this time period there were 9300 fatal red light running crashes, rising from 853 deaths in 1993 to 934 deaths in 2003, an increase of about 10%. Red light running is a highly dangerous driving act and the most frequent type of police-reported urban crashes. A study by Porter and England conducted at six traffic-controlled intersections in three US cities confirmed that 35.2% of the observed light cycles had at least one red light runner prior to the onset of opposing traffic. This rate represented approximately 10 violators per observation hour (Porter & England, 2000). Another study conducted by Retting, Williams, and Greene (1998) over several months at a busy intersection (30,000 vehicles per day) in Arlington, VA confirmed that there were one red light runner every 12 min during normal hours and one red light runner every 5 min during morning peak hours. A lower volume intersection (14,000 vehicles per day), also in Arlington, had an average of 1.3 violations per hour and 3.4 in the evening peak hour (Retting et al., 1998). Thus, based on previous research and accident data, red light running crashes constitute a significant safety problem that warrants attention. To diminish intentional violation of the signal, several states in the United States adopted red light cameras. A Federal Highway Administration report titled ‘‘Making Intersection Safer: A Toolbox of Engineering Countermeasures to Reduce Red Light Running’’ (FHWA, 2003b) stated that in Washington, DC, red light running fatalities were reduced from 16 to 2 in the first two years. In Fairfax, Virginia there was a 44% reduction in red light running crashes (Retting, Williams, Farmer, & Feldman, 1999b). In Oxnard, California there was a 22% reduction in red light crashes citywide (Retting, Williams, Farmer, & Feldman, 1999a). In New York City there was a 34% reduction in red light violations (New York City Department of Transportation). To reduce unintentional violation of the signal and to help drivers make their decision at the onset of yellow, some motorist information countermeasures are implemented such as enhancing the signal display or providing advance information to the driver about the signal ahead. With the additional information, the probability that a driver stops at a red signal may increase. Among them, the two most prevailing and controversial countermeasures are pre-yellow-signal indication and advance warning signs. Smith (2001) employed the Human Factors Research Lab’s driving simulator to investigate the effects of the Advance Warning Flashers (AWF) at signalized intersections on simulated driving performance. It was concluded that AWFs improve stopping behavior at suitable intersections to some extent (Smith, 2001). Sayed, Vahidi, and Rodriguez (1999) analyzed data from British Columbia to assess the benefits of using AWF on signalized intersections. The results indicated that intersections with AWFs have a lower frequency of accidents but this reduction was not statistically significant. Red light running is a complex problem that could result from various causes. In fact, a comprehensive report by the FHWA, 2003a, 2003b entailed a study of 306 police-reported red light running crashes that occurred at 31 signalized intersections located in three US states. The distribution of the reported predominant causes of red light running was as follows: (a) 40% did not see the signal or its indication; (b) 25% tried to beat the yellow-signal indication; (c) 12% mistook the signal indication and reported they had a green-signal indication; (d) 8% intentionally violated the signal; (e) 6% were unable to bring their vehicle to a stop in time due to vehicle defects or environmental conditions; (f) 4% followed another vehicle into the intersection and did not look at the signal indication; (g) 3% were confused by another signal at the intersection or at a closely spaced intersection; (h) 2% were varied in their cause. From the above analysis, 40% of the crashes, resulting from drivers’ disregard of the traffic light, may be possibly related to insufficient sight distance to the traffic light. The above research results showed that traffic signals’ degraded perception or invisibility could be one of the most important factors that cause red light running. Classical sight-distance problems usually concern the intersection geometric characteristics (i.e., the vertical and/or horizontal alignment of the highway) and stationary objects around the intersection that impede the motorist’s view. However, the lack of literature concerns the dynamic sight-distance situation at signalized intersections. In fact, during the yellow change interval at intersections, the following passenger car driver being in the shadow of the blindness caused by the leading LSV (moving/dynamic obstacle) may have insufficient response time to come to a complete stop, which could lead to red light running. Although the FHWA (2003a, 2003b) discussed the line of sight obstruction of traffic light due to trucks, there were no studies that quantify the red light running due to line of sight blockage. FHWA (2003a, 2003b) presented some countermeasures that are meant to reduce vertical visibility obstruction such as ‘‘more

R. Harb et al. / Transportation Research Part F 10 (2007) 229–241

231

than one primary overhead signal, plus pole-mounted secondary (left side) and tertiary signal heads (right side)’’ or ‘‘two overhead signals for a single lane approach and supplemental pole-mounted signal’’. Nevertheless, no studies were conducted to analyze the effect of the additional pole-mounted signal on reducing red light running resulting from vertical visibility blockage. Several studies of drivers’ performance employed driving simulation especially when a reproduction of dangerous conditions in a safe environment is needed. Knodler, Noyce, Kacir, and Brehmer (2005) evaluated the operational advantages and safety of various left turn controls at signalized intersection using a driving simulator. Mitchell, Schatter, and Datta (2005) analyzed speed zone reduction at work zones using a driving simulator. Hirata and Yai (2005) tested the deterioration of the awareness level while driving in a long expressway tunnel using a driving simulator. Lambert et al. measured the driving performance and eye glance behavior using a driving simulator. Harb, Radwan, Yan, and Abdel-Aty (2006) tested the behavior of passenger car drivers behind Light Truck Vehicles (LTVs) and the contribution of the latter to rear-end collisions using a driving simulator. The main objective of this paper is to identify whether LSVs increase the probability of red light running due to vertical view blockage. Another objective is to analyze whether the suggested solution by the FHWA which consists of installing an additional signal pole on the right side of the road decreases red light running resulting from the sight distance blockage. 2. Methodology 2.1. Apparatus The driving simulator acquired by the Center for Advanced Traffic System Simulation (CATSS) at the University of Central Florida, shown in Fig. 1, is able to generate real life driving conditions. A passenger car (Saturn sedan) is mounted on a motion base providing the drivers with the same real car motions on the roads. The simulator car includes five channels of image generation (1 forward, 2 side views, and 2 rear mirrors), an audio and vibration systems, and steering wheel feedback. The simulator allows simulations with different types of vehicles and has sophisticated vehicle dynamic models for different vehicle classes. The simulator also comprises a visual database such as rural, suburban and freeway roads plus an assortment of buildings and operational traffic control devices. Other features include the ability to implement vehicle system malfunctions, and to control the weather conditions (sunny, rain, snow). The scenario generation editor allows us to program the vehicles to follow specific routes, adhere to certain driving patterns, appear at specific points according to a predefined schedule or be a triggered based on other events within the simulation. Another class of vehicles can also be defined to serve as ambient traffic with random movements, making the overall driving experience in the simulator more realistic. Different types of vehicles such as passenger cars, buses, ambulance, police cars, and trucks are user selectable for scripted and random movements throughout the database (Klee, 2003). The simulator session is controlled from an operator’s console in an adjacent control room. The five video channels are monitored on computer screens in the control room. A road map of the database is viewable on the operator’s console showing the movement of the simulator vehicle and other vehicles that are present. Scenarios are created with the scenario editing software on a screen showing the location of roads, buildings, traffic control devices, and pedestrians. In addition to the five video channels and the real-time map, a camera was installed inside the simulator car to supervise the driver’s responses in the car. An emergency stop button is provided in the control room to immediately discontinue the experiment in case the drivers suffer a motion sickness. 2.2. Simulation scenario design The vertical view blockage scenario consists of three sub-scenarios: a base or control sub-scenario where the simulator car follows a regular passenger car; a test sub-scenario where the simulator car follows an LSV (school bus); and the suggested FHWA solution sub-scenario where the simulator car follows an LSV with the additional traffic signal pole.

232

R. Harb et al. / Transportation Research Part F 10 (2007) 229–241

Fig. 1. UCF driving simulator cab.

The whole experiment course can be described in three stages as shown in Fig. 2. In the first stage, the subject drives the simulator car to a T-intersection where he/she is instructed to make a left turn. At the second stage, the simulator car approaches the signalized intersection, where the signal phase has just turned green and where a leading vehicle (a school bus or a passenger car) just started making a right turn slowly. The subject is assigned to make a right turn behind the leading vehicle at that intersection. The speed limit in the second stage is 35 mph and the route consists of one lane per direction to inhibit the following car from any attempt to pass the leading vehicle. Finally, in the third stage, at the time T0, the traffic signal turns yellow. At that time, the leading vehicle is at a safe distance to cross the intersection. However, the following vehicle is not at a safe distance to clear the intersection. The hypothesis in this study is that the driver’s vertical sight view in the following car is more likely to be restricted by a leading LSV compared to a passenger car therefore, he/she may not be aware of the traffic signal change when following the LSV. At the time T1, 3.5 s (yellow change interval) after T0, the traffic signal turns red. At time T 2, this could be prior or subsequent to T1, the following vehicle’s driver reacts and is faced with two alternatives. The driver can either suddenly stop or run the red light. Data measurements of the subjects including cruising velocities, reaction delay time, deceleration rates, and gaps between vehicles were captured and analyzed.

R. Harb et al. / Transportation Research Part F 10 (2007) 229–241

233

Fig. 2. Experimental design stages.

2.3. Intersection at stage #3 As shown in Fig. 2, at stage #3 the main rural road consisted of 1 lane per direction to avoid any passing as explained in the scenario design sub-section. The lanes’ widths are 12 ft and the pavement markings and signings are implemented based on the MUTCD standards. The intersection width is 40 ft and the height of the traffic light is 18 ft based on AASHTO standards. The signal setting at stage #3 of the experiment was preset to be triggered by the leading vehicle. In fact, when the leading vehicle reaches a distance of 179 ft from the intersection the signal phase is set to change from green to yellow. The yellow time interval is 3.5 s. 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. 2.4. Participants Three sub-scenarios were designed in the driving simulator to test the vertical visibility blockage contribution to red light running and to analyze the suggested solution by the FHWA. The reader should be cautious that in all three aforementioned sub-scenarios the subjects follow the same path, drive through the same three stages, and experience the same sequence of events. In fact, the only difference between the base sub-scenario and the test sub-scenario is the leading vehicle type (PC and LSV respectively). The only difference between the test sub-scenario and the suggested solution sub-scenario is the additional signal pole. The reason for analogous scenario designs was to eliminate any possible additional factors resulting from a change of environmental characteristics or sequence of events that may bias the experiment. Thus, a subject cannot drive more than one sub-scenario because he/she may expect the same event (or sequence of events) in another sub-scenario

234

R. Harb et al. / Transportation Research Part F 10 (2007) 229–241

and therefore be prepared for it. To overcome any possible bias, three matching groups A, B and C were recruited. The matching criteria were age and gender. In order to make the selected subjects closely duplicate the actual Florida drivers’ population, the quasiinduced exposure method (Yan, Radwan, & Birriel, 2005) was utilized. The distribution of the age and gender of the subjects were generated from the Florida Crash Database where males represent 60% versus females 40%, and middle age represents 60% versus young 40% of the population. The young age group varies between the ages of 18 and 25 and the middle age group varies between 25 and 55. Table 1 shows the final gender and age breakdown of the subjects that completed the experiment. Each subject had at least two years of driving experience and held a valid Florida driver’s license. All recruited subjects were full-time students or employees at the University of Central Florida (UCF). The first 10 subjects from the base sub-scenario and the first 10 subjects from the test-sub-scenario were used as a pilot study to calculate the minimum required sample size for the formal study. Four out of 10 subjects driving behind an LSV ran the red light and 1 out of 10 subjects driving behind the PC ran the red light. The minimum required sample size is calculated using Eq. (1) 2

n¼

ðZ a þ Z b Þ ðp1 q1 þ p2 q2 Þ ðp1  p2 Þ

ð1Þ

where n Za Zb p1 p2 q1 q2

estimated necessary sample size Z-coefficient (Type I) error with 95% confidence interval Za = 1.96 Z-coefficient (Type II) error with 95% confidence interval, Zb = 1.64 the total number of red light running behind LSV over the total number of trials (4/10 = 0.4) the total number of red light running behind PC over the total number of trials (1/10 = 0.1) the total number of ‘‘no red light running’’ behind LSV over the total number of trials (1  p1) the total number of ‘‘no red light running’’ behind LSV over the total number of trials (1  p2)

From the above equation the minimum required sample size for each sub-scenario at 95% significance level is seven subjects and 13 subjects for 99% significance level. To be more conservative, we recruited 20 subjects for each sub-scenario. Each subject was trained on the driving simulator for 5 min and then he/she completed a sub-scenario without knowing its purpose and characteristics. At the end of the experiment, 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 completing the experiment. Moreover, the 60 subjects that completed the experiment, out of the 62 recruited subjects, confirmed that they did not experience any type of motion sickness. 2.5. Sight distance analysis In this section, sight distance models were completed to compute the minimum gap X1, shown in Fig. 3, at which the traffic light is visible for the following passenger car driver. In these calculations, the height of the Table 1 Age and gender distribution of recruited subjects Group

Age

Male

Female

Total

Sub-scenario driven per group

A

Young Middle age

5 7

3 5

20

SIM-PC sub-scenario

B

Young Middle age

5 7

3 5

20

SIM-LSV sub-scenario

C

Young Middle age

5 7

3 5

20

Additional signal pole sub-scenario

R. Harb et al. / Transportation Research Part F 10 (2007) 229–241

235

Fig. 3. Theoretical calculations for minimum gap.

LSV, the eye height of the following driver, and the height of the traffic light were standard values borrowed from AASHTO standards (AASHTO, 2001). From the trigonometry in Fig. 1 the following equations were computed: H2  H1 H3  H2 ¼ X1 X2

ð2Þ

H 3 ¼ XX 21 ðH 2  H 1Þ þ H 2. We would like: v2 X 2 ¼ vt þ þ w þ L þ D 2a where H2 H3 H1 W L D t a X1 X2 V

X2 ðH 2 X1

 H 1Þ þ H 2 to be

LSV height and an average value of 8.5 ft is used in the experiment signal head height and an average value of 21 ft is used in the experiment (AASHTO) eye elevation equal to 3.75 ft (AASHTO) width of the intersection 40 ft (AASHTO) length of the vehicle taken 30 ft (AASHTO) set back of the stop bar from the intersection, which is 10 ft standard response time which is 1.0 s (AASHTO) acceleration rate taken 10 ft/s2 (AASHTO) distance from the center of the car to the back of the leading vehicle in ft distance from the back of the leading vehicle to the traffic signal in ft (stopping sight distance, SSD) velocity of the vehicle taken 35 mph or 51.33 ft/s

Table 2 shows the minimum required distance X1, which is the distance between the rear-bumper of the leading and the center of the following vehicle, with the variation of the traffic light height H3 and the LSV height H2 using the equations listed above. From Table 2, the values of X1 are proportional to H2 Table 2 Gap for following PC and LTV V

T

w

A

L

D

H1

H2

H3

X2

X1

51.33 51.33 51.33 51.33 51.33 51.33

1 1 1 1 1 1

40 40 40 40 40 40

10 10 10 10 10 10

30 30 30 30 30 30

10 10 10 10 10 10

3.75 3.75 3.75 3.75 3.75 3.75

9 9 9 10 10 10

18 20 22 18 20 22

320 320 320 320 320 320

187 153 129 250 200 167

236

R. Harb et al. / Transportation Research Part F 10 (2007) 229–241

and H3. Indeed, the bigger H3 the larger X1 must be in order for the following vehicle driver to see the traffic light. In the formal experiment we used H2 = 8.5 ft and H3 = 21 ft. For instance, when H2 = 9 ft and H3 = 18 ft the minimum distance X1 must be 187 ft in order for the following vehicle driver to see the traffic light. Otherwise, there will be a potential sight distance problem during the signal change interval. 2.6. Location of the additional signal pole The signal pole position (X and Y coordinates) is calculated below following the trigonometry in Fig. 4. In triangles ABC, AKJ, and BNX the following ratios apply: AC AK ¼ BC KJ AK NX ¼ KJ BX

ð4Þ ð5Þ

AC is the stopping sight distance for the following passenger car and is calculated as follows: AC ¼ SSD ¼ vt þ

v2 2a

SSD ¼ 51:33 ðft=sÞ  1 ðsÞ þ

ð51:33 ðft=sÞÞ

ð6Þ

2

2  10 ðft=s2 Þ

¼ 183:1 ft

According to AASHTO the width of a typical school bus is 8 ft. Assuming that vehicles drive in the center of the lane, KJ is equal to 4 ft. The average time gap of all subjects in the experiment was 0.8 s and was utilized for the calculations. The assigned speed limit in the scenario was 35 mph at this stage. Therefore, the leading vehicle is consigned to drive at 35 mph, which will make the following vehicle adhere to the 35 mph speed. Thus, the distance AK is equal to 41 ft (0.8 s * 51.33 ft/s). Substituting all the variables in Eq. (1) gives BC = 18 ft. As shown in Fig. 4, BX = (BC  MN) = 18  (6 + Y). Replacing AK, KJ, and BX in Eq. (4), the following relationship between X and Y is obtained:

Fig. 4. Theoretical calculations for signal pole location.

R. Harb et al. / Transportation Research Part F 10 (2007) 229–241

Y ¼ 0:098X þ 12

237

ð7Þ

For instance, if the additional signal pole were to be placed somewhere at Y = 15 ft from the edge of pavement, the minimum required distance X, from the perpendicular edge of pavement, must be 31 ft or larger, as shown in Fig. 4. 3. Results 3.1. Red light running rate From the collected data, in group A, two subjects out of 20 subjects driving the simulator behind the PC ran the red light. However, from group B, 10 subjects out of the 20 subjects driving the simulator car behind the LSV ran the red light. A chi-square test showed a significant statistical difference (P-value = 0.006) between the red light running rates (0.1 for following a PC vs. 0.5 for following an LSV). The reader may consult Eq. (7) for the calculation of the chi-square statistic. X2 ¼

X ðExpected  ObserveredÞ2 Expected

ð8Þ

Group C drove the driving simulator behind the LSV with the additional traffic signal pole and four subjects out of 20 subjects ran the red light. A chi-square test showed that light running ratios when following an LSV with and without an additional signal pole at 0.05 significance level are significantly different (P-value = 0.047). The additional signal pole reduced red light running from 50% to 20%. Therefore, the additional signal pole is a potential solution for vertical visibility blockage due to LSVs. It should be cautioned that the probabilities of red light running based on the experiment results are higher than those in a real life. These probabilities are conditional probabilities which resulted from the critical traffic event. 3.2. Brake response time and deceleration rates As mentioned before, two subjects from group A and 10 subjects from group B ran the red light. It should be mentioned that when the simulator car runs the red light, its deceleration rate and response time would be null since it did not stop. Therefore, the deceleration rates and the response times of 10 subjects who did not run the red light when they were driving behind the school bus (Group B) were compared to the deceleration rates and response times of the 18 subjects who did not run the red light behind the PC (Group A). The deceleration rate was measured for the following vehicle’s speeds ranging from the speed at the instant when the simulator starts braking, after the onset of the yellow phase, to a speed of 5 mph. Zero km/h (0 mph) was not used because a large number of drivers maintained a crawling speed until the stop bar, which would have biased the experiment results. The response time is the time difference between the instant that the traffic light turns yellow and the time at which the simulator car starts baking. A two-sample t-test was completed to compare the deceleration rates and response time means for groups A and B at 0.05 significance level. The results showed no significant statistical difference (P-value = 0.97) between the deceleration rates for following a PC and following an LSV. The average deceleration rate for following an LSV (7.73 ft/s/s) is very close to the deceleration rate for following a PC (7.66 ft/s/s). The deceleration rates were also calculated for group C where subjects drove the suggested solution sub-scenario and were compared with the deceleration rates from group B. The result showed no significant statistical difference (P-value = 0.408) between them. However, the deceleration mean decreased from 7.73 ft/s/s in group B to 6.30 ft/s/s in group C with the addition of the signal pole. The reader may consult Eqs. (9) and (10) for the calculation of the t-test statistic y1  y2 t ¼ rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi  ffi Sp2 n11 þ n12

ð9Þ

238

R. Harb et al. / Transportation Research Part F 10 (2007) 229–241

where Sp2 ¼

ðn1  1ÞS 21 þ ðn2  1ÞS 22 n1 þ n2  1

ð10Þ

The collected data demonstrated that the brake response times of following a school bus are greater than the brake response times of following a PC. In fact, the mean response time for following a school bus is 3.45 s and the mean response delay time for following a PC is 2.02 s. However, a two-sample t-test confirmed that there is no significant statistical difference (P-value = 0.073) between the brake response time means for following a PC and following a school bus. The brake response times were also calculated for Group C and were compared with brake response times of group B. The conducted two-sample t-test corroborates that there is no significant difference (P-value = 0.065) between the response time means of groups B and C (3.45 s vs. 2.45 s). The P-values for the brake response time could have been affected by the sample size reduction as explained in the first paragraph of this section. In other words there may be a statistically significant difference if the sample size was not reduced. 3.3. Gaps and cruising velocities Gaps and cruising velocities of 20 subjects from group A, 20 subjects from group B, and 20 subjects from group C were calculated and compared. The gap is the distance between the leading and the following vehicle at the onset of yellow phase. The cruising velocity is also measured just at the onset of yellow phase. Fig. 5 shows that the gaps for group A (following PC), group B (following an LSV), and group C (following the school bus with additional signal pole). The calculated gap mean for group A (154 ft) is smaller than the calculated gap mean for group B (185 ft). However, a 2 sample t-test showed no statistical difference (P-value = 0.398) between the gap means of groups A and B. The calculated gap mean for group C (167 ft) is smaller than the calculated gap mean for group B. Conversely, a two-sample t-test completed at 0.05 significance level showed no statistical significance difference (P-value = 0.213) between the gap means for following an LSV with and without an additional signal pole. The purpose of calculating the cruising velocities is to study the behavior of subjects driving behind LSVs and to analyze the effect of this behavior on the red light running rate. Indeed, if the subjects are frustrated 300 Following PC Following LSV 250

Additional Pole

Gap (ft)

200

150

100

50

0 1

2

3

4

5

6

7

8

9

10 11 12 13 14 15 16 17 18 19 20

Subject Fig. 5. Gaps for groups A, B, and C.

R. Harb et al. / Transportation Research Part F 10 (2007) 229–241

239

because they are driving blindly behind the bus, they may drive faster to pass the latter. A two-sample t-test was completed to compare the velocity means of groups A and B at 0.05 significance level. The results showed no statistical significant difference (P-value = 0.403) between the group A and B. The mean velocity of following a school bus (34.95 mph) is slightly grater than the mean velocity of following PC (34.00 mph). This outcome, according to the above discussed hypothesis, suggests that drivers behind LSVs may be aware that it is not safe to pass large vehicles. Therefore, they drive at ordinary speed behind LSVs awaiting for a safe opportunity to pass it. A two-sample t-test was also computed to compare the velocity means of groups B and C, following a LSV and following an LSV with an additional signal pole at 0.05 significance level. The results confirmed that there is a statistically significance difference (P-value = 0.019) between velocity means. The velocity mean for following LSV (34.95 mph) is larger than the mean velocity of following an LSV with additional signal pole (32.61 mph). Therefore, one can conclude that adding an additional signal pole alerted the drivers of the traffic light change which may have resulted to slower velocities. 3.4. Survey analysis The participants took a survey after they completed the experiment. The subjects were asked if they saw the traffic light in both following a PC and following a school bus sub-scenario. As shown in Fig. 6, 10 subjects who drove behind the school bus reported that they did not see the traffic light and the 10 other subjects driving behind the school bus reported that they saw the traffic light. The subjects that reported that they did not see the traffic signal ran the red light. The same subjects were asked if it was too late for them to stop when they saw the traffic light. The 10 subjects who ran the red lights reported that they saw the traffic signal at some point but it was too late for them to stop. However, the two subjects driving behind the passenger car who ran the red light reported that they saw the traffic signal but they still ran the red light because they just decided to ‘‘beat the signal’’. The 20 subjects in the group A and the other 20 subjects in the group B were also asked if they drive closely to passenger cars and buses respectively in the real life. Ten subjects who drove behind the school bus reported that they drive close behind LSVs in daily life and the other 10 subjects who drove behind the school bus reported that they do not drive close behind large vehicle. However, eight subjects driving behind the

12 10 8

YES

6

NO

4 2

Number of Subjects

Number of Subjects

14

Driving too Close to a School Bus in Real Life?

0

10 8 6 4 2 0

YES

NO

NO

10 8 6 4 2 0

12

Number of Subjects

Number of Subjects

12

Driving too Close to a Passenger Car in Real Life?

20 18 16 14 12

Too Late to Stop Behind a Passenger Car

YES Too Late to Stop Behind a School Bus

10 8 6 4 2 0

Fig. 6. Survey analysis results.

YES

NO

240

R. Harb et al. / Transportation Research Part F 10 (2007) 229–241

passenger car reported that they drive close to passenger cars in daily life and the remaining 12 subjects driving behind a passenger car reported that they keep a large distance when they drive behind a passenger car in daily life and in similar circumstances. The 20 subjects driving behind a school bus were asked if they encounter this visibility problem in their daily life. Eleven subjects said that they come across this visibility issue at least once a week and the nine remaining subjects said that they rarely come upon this visibility issue. The subjects (from group C) were asked if they observed the actual traffic light first or the additional signal pole first. Seventy percent of the subjects said that they saw the additional signal pole mounted on the side of the road before they saw the originally installed traffic light. Moreover, 30% of the subject said that they saw them at the same time. Therefore, 100% of the subject from group C perceived the additional signal pole. The subjects were also asked if they think that the traffic signal pole addition would be profitable for the drivers’ safety in real life. Sixty five percent of the subjects said that it is profitable and the remaining subjects said that it is not profitable. However, the 35% remaining subjects said that the additional signal pole is not profitable due to its high cost. One must note that the additional signal pole would be added on intersections with high red light rates. 4. Conclusions and discussions One of the objectives of this research was to study whether LSVs augment the red light running rate for following passenger car drivers at signalized intersections due to vertical visibility blockage. The results confirmed that there is a significant statistical difference between the rates of red light running for following a passenger car and for following an LSV. Seven out of the 10 drivers who ran the red light behind the LSV (from group B) were younger than 30 yearsold and one out the two drivers who ran the red light behind the PC (from group A) was younger than 30 yearsold. Fisher’s exact test showed that younger age drivers are significantly more prone (P-value = 0.001) to run the red light. These results are homogeneous with the results of Retting et al. (1999a, 1999b) who confirmed that drivers younger than 30 years-old are more likely to run the red light than other age groups. Another objective of this experiment was to study the behavior of the subjects driving behind larger size vehicles. Based on the experiment results, it was confirmed that there is no statistical difference between the velocities of group A and group B and the gaps of group A and group B (driving behind a passenger car and driving behind an LSV). Therefore, one can conclude that subjects driving behind an LSV and a passenger car behave similarly as far as speeding and gaps. The efficacy of the proposed addition of the traffic signal pole on the right side of the road was also tested. From the above analysis, the additional signal pole decreased the red light running rate significantly from 50% to 20%. With the additional signal pole, the subjects drove behind the school bus at significantly larger gaps which suggests that the additional signal pole alerts the drivers. Moreover, 65% of the subjects that completed the experiment said that the traffic signal pole would be profitable in real life. As a typical study based on a driving simulator, some limitations may exist since a validation study by a field test was not conducted. In general driving simulator data are valid for relative comparison instead of absolute comparison. For instance, the deceleration rates, velocities, gap time and distance may not be exactly 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 (4th ed.). American Association of State Highway and Transportation Officials. Federal Highway Administration (FHWA) (2003a). Stop red light running facts and statistics. (Access date, December 2005). Federal Highway Administration (FHWA) (2003b). Making intersection safer: a toolbox of engineering countermeasures to reduce red-light running. . Harb, R., Radwan, E., Yan, X., & Abdel-Aty, M. (2006). Contribution of light truck vehicles to rear-end collisions. In Proceeding TRB 2006 annual meeting, Washington DC.

R. Harb et al. / Transportation Research Part F 10 (2007) 229–241

241

Hirata, T., & Yai, T. (2005). An analysis of driver’s awareness level and support system while driving in long expressway tunnel. TRB 2005 annual meeting, Washington DC. 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. Knodler, J., Noyce, D., Kacir, K., & Brehmer, C. (2005). An evaluation of the flashing yellow arrow permissive indication for use in simultaneous indications. In TRB Annual Meeting, Washington DC. Mitchell, G., Schatter, K., & Datta, T. (2005). Use of a driving simulator for evaluation of safety measures in highway work zones. In TRB annual meeting, Washington DC. New York City Department of Transportation (xxxx). Traffic signal and sign information. Frequently asked questions. Retrieved November 12, 2004. . Porter, B. E., & England, K. J. (2000). Predicting red-light running behavior: a traffic safety study in three urban settings. Journal of Safety Research, 31(1), 1–8. Retting, R. A., Williams, A. F., & Greene, M. A. (1998). Red-light running and sensible countermeasures: summary of research findings. Transportation Research Record, 1640, 23–26. Retting, R. A., Williams, A. F., Farmer, C. M., & Feldman, A. F. (1999a). Evaluation of red light camera enforcement in Oxnard California. Accident Analysis and Prevention, 31, 169–174. Retting, R. A., Williams, A. F., Farmer, C. M., & Feldman, A. F. (1999b). Evaluation of red light camera enforcement in Fairfax, VA., USA. ITE Journal, 69(8), 30–35. Sayed, T., Vahidi, H., & Rodriguez, F. (1999). Advanced warning flashers: do they improve safety? Transportation Research Record, 1692, 30–38. Smith, T. (2001). Effects of advanced warning flashers at signalized intersections on simulated driving performance. (Access date, November 2004). Yan, X., Radwan, E., & Birriel, E. (2005). Analysis of red light running crashes based on quasi-induced exposure concept and multiple logistic regression method. Journal of the Transportation Research Board, 1908, 70–79.