Accident Analysis and Prevention 40 (2008) 1637–1643 Contents lists available at ScienceDirect Accident Analysis and Prevention journal homepage: www.elsevier.com/locate/aap Driving performance assessment: Effects of traffic accident location and alarm content Shun-Hui Chang a , Chih-Yung Lin b , Chin-Ping Fung c , Jiun-Ren Hwang a , Ji-Liang Doong d,∗ a Department of Mechanical Engineering, National Central University, Chung-Li 32054, Taiwan, ROC Department of Mechanical Engineering, Lunghwa University of Science and Technology, Guai-Shan 33306, Taiwan, ROC c Department of Mechanical Engineering, Oriental Institute of Technology, Pan-Chiao 22061, Taiwan, ROC d Department of Industrial Design, Tatung University, Tai-Pei 10452, Taiwan, ROC b a r t i c l e i n f o Article history: Received 8 October 2007 Received in revised form 8 May 2008 Accepted 9 May 2008 Keywords: Traffic accident location Alarm content Driving simulator Driving performance Logistic regression analysis a b s t r a c t According to accident statistics for Taiwan, the two most common traffic accident locations in urban areas are roadway segments and intersections. On roadway segments, most collisions are due to drivers not noticing the status of leading vehicle. At intersections, most collisions are due to the other driver failing to obey traffic signs. Using a driving simulator equipped with a collision warning system, this study investigated driving performance at different accident locations and between different alarm contents, and identified the relationship between crash occurrences and driving performance. Thirty participants, aged 20–29 years, were recruited in this study. Driving performance measures were perception-reaction time, movement-reaction time, speed and a crash. Experimental results indicated that due to different demands for processing information under different traffic conditions, driving performance differed at the two traffic accident locations. On a roadway segment, perception-reaction time for a beep was shorter than the time for a speech message. Nevertheless, at an intersection, a speech message was a great help to drivers and, thus, perception-reaction time was effectively reduced. In addition, logistic regression analysis indicates that perception-movement time had the greatest influence on crash occurrence. © 2008 Elsevier Ltd. All rights reserved. 1. Introduction Traffic accidents were in the top ten causes of death in Taiwan. According to traffic accident statistics (MOTC, 2006), the main reasons for traffic accidents were overlooking the status of the leading vehicle, drunk driving, failing to yield right-of-way, and disobeying traffic signs. Most traffic accidents occurred on urban roads, accounting for roughly 27.46% of all traffic accidents. Most accidents occurred on straight roadway segments and at intersections. On roadway segments, most accidents occurred because a driver failed to pay attention to the vehicle in front, whereas accidents at intersections were most often caused by a vehicle running a red light or not paying attention to traffic signage. Driving safety is related to driving performance. Many studies investigated the effects of different traffic environments on driving performance. Jahn et al. (2005) analyzed the effects of road types on peripheral detection while driving. The roads used were classified based on traffic complexity. Roads in city centers with complex intersections were considered to have high traffic com- ∗ Corresponding author. Tel.: +886 2 25925252x2967; fax: +886 2 25867348. E-mail address: jldo@mail2000.com.tw (J.-L. Doong). 0001-4575/$ – see front matter © 2008 Elsevier Ltd. All rights reserved. doi:10.1016/j.aap.2008.05.003 plexity, while roads in urban and rural areas and on highways without interactions were considered to have low traffic complexity. Liu and Lee (2006) examined the effects of cellular phone communication while driving on driving performance, with particular emphasis on driving in different traffic situations (urban roads or highways). Their study revealed that mean response time was increased markedly (11.9%) when driving on urban roads compared to highways. These studies, however, did not consider the accident variable, and, thus, driving performance at different accident locations were not clearly understood. To improve traffic safety and transportation efficiency, a collision warning system (CWS) and other driving assistant systems have been developed and investigated. Most studies of CWSs used driving simulators due to their safety, low equipment cost and ability to control experiments (Liu and Lee, 2006). Maltz and Shinar (2007), based on experimental results from a fixed-base driving simulator, pointed out that drivers may misuse warning systems. A study by Abe and Richardson (2006), who focused on alarm timing and its impact on driver response, concluded that drivers typically expect alarms to be activated before braking is initiated. Lee et al. (2002) indicated that early warning helped distracted drivers react more quickly and avoid more collisions than did late warning. A.W.L. Ho et al. (2006) studied driver reaction times and 1638 S.-H. Chang et al. / Accident Analysis and Prevention 40 (2008) 1637–1643 the accuracy of responses under a single master alarm and multiple alarms. Suzuki and Jansson (2003) evaluated driver steering in response to auditory and haptic warnings for the design of a lane-departure warning system. Experimental results showed that beeps were effective in reducing steering reaction time. Ho et al. (2005) investigated whether speedy responses to potential emergency driving situations could be facilitated by vibrotactile cues. Kolisetty et al. (2006) analyzed the effects of different message signs on driving speed using a fixed-base driving simulator with three wide screens. Their analytical findings indicated that different message signs caused drivers to reduce their speed. With a driving simulator that provided typical driving vibration, C. Ho et al. (2006) investigated the possibility that driver responses to front-to-rear end collisions could be facilitated by using vibrotactile warning signals. All these studies demonstrated that an alarm system is important to driving safety. However, those studies simply underlined the effects of alarm timing (Abe and Richardson, 2006; Lee et al., 2002) or alarm type (Ho et al., 2005; A.W.L. Ho et al., 2006; C. Ho et al., 2006; Kolisetty et al., 2006; Suzuki and Jansson, 2003) on drivers; that is, the effects of auditory alarm content on driving performance were rarely mentioned. Some novel systems and methodologies developed to improve safety are in the experimental stage. Sengupta et al. (2007) developed the first prototype of a cooperative collision warning system and presented experimental results tested on a fleet of five real vehicles. This cooperative collision warning system provided warnings to drivers based on information for the motion of neighboring vehicles obtained by wireless communications from those vehicles, without use of range sensors. Clement and Taylor (2006) developed the simple platoon advancement model to increase vehicle throughput at intersections with traffic signals. This model described a conceptual system that used several intelligent transport system technologies, including automatic cruise control, lane-departure avoidance, and collision avoidance, to control vehicle progression through intersections. Additionally, this model allowed vehicles arriving in a queue at the stopline to move automatically. Analytical results demonstrated that the model can move nearly twice as many vehicles past the stopline as can in today’s road network. The information obtained from studies examining the difference in driving performance at different accident locations and with different auditory alarm contents indicates that few studies have addressed these factors. Therefore, this study adopted the experimental environment of an urban road and a driving simulator to investigate the two most common urban traffic accident locations, i.e., roadway segments and intersections. Most collisions on roadway segments are due to drivers failing to pay attention to the vehicle in front, while those at intersections are due to drivers not following traffic signage. The effects of the alarm content from CWSs on driving performance were also investigated. This study analyzed further the relationship between crash occurrence and driving performance by applying logistic regression analysis, and investigated the influence of driving responses on crash occurrence. 2. Methods 2.1. The IOT driving simulator This study used a driving simulator developed by the Institute of Transportation (IOT), Taiwan (Lin et al., 2004a,b). The simulator integrates a real vehicle (NISSAN Sentra 180), a six degree-offreedom Stewart motion platform, a virtual reality-based visual and audio system, vehicle motion simulation software, and a host computer system to simulate the virtual environment of an urban Fig. 1. The configuration of the IOT driving simulator. road. The real vehicle is mounted on a hydraulic Stewart motion platform that generates motion experienced under normal acceleration, braking and steering. Inside the driving cabin, the driving control mechanisms, such as an accelerator pedal, brake pedal and steering wheel, were implemented with force feedback. A driver’s behavior in manipulating driving control mechanisms was recorded by the monitoring system at a rate of 30 Hz. The platform provides longitudinal, lateral, heave, yaw, pitch, and roll motion with displacements of ±20◦ , ±19◦ , ±32◦ , ±228, ±240 and ±130 mm, respectively. The visual system consists of three screens that provide a 135◦ (horizontal) × 36.87◦ (vertical) field of view. The scene is updated at rates of 25–35 Hz. The audio system provides simulated noises from the engine, road tires and street. Fig. 1 shows the configuration of the IOT driving simulator. 2.2. Experimental design The experiment was a 2 × 3 factorial design that compared experimental results for traffic accident locations (roadway segment, intersection) and auditory alarm contents of the CWS (null, beep sound, and speech message). Each subject encountered an emergency traffic event in each condition during the experimental drive. The different conditions were randomly presented so subjects could not predict the location of an emergency event. The driving environment and scenario were as follows. The driving environment was a straight roadway with intersections—a hypothesized road in an urban area. The road has two lanes, each 3.5 m wide, with 1-m wide sidewalks. This road consisted of an acceleration section (300 m), an experimental section (5100 m), and a braking section (900 m). Eleven intersections were located every 400–600 m in the experimental section. The first traffic accident location was a roadway segment. A vehicle on a roadway segment suddenly appeared such that the scenario could not be expected by the subject. This vehicle first appeared at lateral side, and then overtook the host vehicle and cut into the lane used by the host vehicle. A traffic event was triggered at a time gap of 2 s (Abe and Richardson, 2005) between the two vehicles. At the time a traffic event was triggered, the leading vehicle braked suddenly and its brake lights were activated. This time was designed to allow the host vehicle to respond to the leading vehicle braking at a deceleration rate of 8.5 m/s2 as the host vehicle drove at a steady speed of 50 km/h, about 28 m behind the leading vehicle. In this design, the host vehicle only had about 3 s to react and avoid a crash. 1639 S.-H. Chang et al. / Accident Analysis and Prevention 40 (2008) 1637–1643 Table 1 The participants’ demographic and driving information (N = 30) Age (year) Driving experience (year) Frequency of driving (day per week) An alpha level of 0.05 was the minimum level of significance. All statistical analyses utilized SPSS for Windows. Average S.D. min max 23.2 3.6 2.4 2.2 2.5 1.8 20 1 1 29 10 6 The second traffic accident location was an intersection. The intersection at which a violator’s vehicle rushed into traffic was randomly chosen to avoid subject expectation. An emergency traffic event was triggered 3 s (Chen et al., 2005) before the host vehicle entered the intersection. This time gap was designed such that a crash occurs when the host driver driving at 50 km/h did not properly react when the violator’s vehicle traveling at 70 km/h entered a blind intersection from the left or right. As buildings at the intersection blocked the view, the host driver was only aware of the violator’s vehicle after it entered the intersection. The host vehicle (i.e., driving simulator) was equipped with a CWS. This system automatically generated a beep sound or speech message when an urgent traffic event was triggered, i.e., the leading vehicle braked suddenly on the roadway segment or a violator’s vehicle entered an intersection from the left or right. Participants were told the CWS system would generate not signal to simulate the condition of driving without a CWS. The signal beep was a pure tone at 2 kHz at 70 dB. The statistical data for traffic accidents published by National Police Agency, Taiwan, indicated that young drivers are a highrisk group. Therefore, this study recruited 30 young drivers, ranged 20–29 years, to participate this driving experiment. All participants held a valid driving license and had driving experience of 1–10 years (average, 3.6 years). Table 1 lists the participants’ demographic and driving information. 2.3. Experimental procedure Subjects first provided their personal information—sex, age, and driving experience. Experimental instructions were given for the driving task and subjects were instructed by assistants in how to operate the simulator. After practice driving for 10–15 min with a CWS, the formal experiment, which took about 7–10 min, was conducted. Each participant was asked to maintain a speed of 50 km/h and stay in the inner lane. At experiment end, participants were de-briefed, paid US$10 and thanked for their participation. 2.4. Data collection and analysis Driver performance refers to a driver’s perceptual and motor skills. The ability to judge speed, keep a constant speed, and react to hazards are all associated with driver performance (Evans, 1991). The driving performance measures in this study are perceptionreaction time, movement-reaction time, speed and a crash. Driver performance was measured under an emergency—not normal performance under normal traffic conditions. Perception-reaction time was measured from the time when a traffic event was triggered to the time when the driver released the accelerator. Movement-reaction time was measured when the accelerator was released to when the brakes were activated. Speed was measured from the time when a traffic event was triggered to the time when a crash occurred, or the leading vehicle or violator vehicle drove away from the host vehicle. The paired t-test was used to determine whether driving performance at the two traffic accident locations are significantly different; the level of significance was p < 0.05. Analysis of variance (ANOVA) was utilized to test for differences among driving performance induced by different alarm contents. 3. Results This study assessed the effects of different traffic accident locations and CWS alarm contents on driving performance. Furthermore, the relationship between a crash and driving performance was modeled by logistic regression to investigate the influence of driving performance on the crash occurrence. Experimental results can be seen in the following sections. 3.1. Influence of traffic accident location Table 2 lists the driving performance at different traffic accident locations. Experimental results show that perception-reaction time for the traffic accident at an intersection (1.47 s) was longer than that for the traffic accident on a straight roadway segment (0.77 s). The perception-reaction time at different locations was statistically significant (p = 0.000), indicating that the accident location had a significant effect on driver perception of a traffic event. This study also measured driver movement-reaction time to identify the speed at which a driver hit the brakes to avoid a crash. Experimental results show that it took longer for a driver to react to the traffic accident at an intersection than on a straight roadway segment; the difference in movementreaction time was statistically significant (p = 0.056). The sum of perception-reaction time and movement-reaction time, defined as perception-movement time, is the time from when a peripheral stimulus appeared to a driver braking. The perception-movement time in the intersection traffic accident (1.89 s) was longer than that for the straight roadway segment accident, and the difference between different traffic accident locations was statistically significant (p = 0.000). The average driving speeds from the time when a traffic event was triggered to the time when the leading vehicle or violator vehicle drove away were almost the same at different traffic accident locations. Average speed was 41.71 km/h for the accident on the straight roadway segment, and 41.04 km/h for the intersection accident. On the other hand, the difference in average speed reductions induced by braking was statistically significant (p = 0.030). The average speed reduction in the intersection traffic accident (26.45 km/h) was less than that for the accident on the straight roadway segment (30.34 km/h). 3.2. Influence of alarm content 3.2.1. Perception-reaction time Table 3 lists driving performance with different types of alarm content. Experimental results show that the accident on the straight roadway segment without any audio signals had the longest perception-reaction time, i.e., the CWS was not workTable 2 Average driving performance at different traffic accident locations Variables Roadway segment Intersection p-Value Perception-reaction time (s) Movement-reaction time (s) Perception-movement time (s) Average speed (km/h) Average speed reduction (km/h) 0.77 (0.32) 0.48 (0.21) 1.27 (0.30) 41.71 (4.92) 30.34 (6.58) 1.47 (0.82) 0.58 (0.54) 1.89 (0.72) 41.04 (9.60) 26.45 (15.63) 0.001a 0.056 0.001a 0.182 0.030b Values enclosed in parentheses represent standard deviation. a p < 0.001. b p < 0.05. 1640 S.-H. Chang et al. / Accident Analysis and Prevention 40 (2008) 1637–1643 Table 3 Average driving performance with different types of alarm content Accident location Alarm content PRT (s) MRT (s) Speed (km/h) Speed reduction (km/h) Roadway segment Null Beep sound Speech message 0.83(0.42) 0.69(0.27) 0.76(0.2) 0.48(0.21) 0.47(0.19) 0.51(0.26) 41.67(4.59) 41.06(5.12) 42.70(4.57) 30.01(6.41) 30.18(7.22) 30.97(6.37) Intersection Null Beep sound Speech message 1.58(0.71) 1.10(0.74)* 0.97(0.85)* 0.45(0.34) 0.58(0.54) 0.84(0.74)* 42.16(8.81) 37.74(10.6) 39.26(9.86) 22.82(17.4) 28.63(14.3) 28.20(16.3) Crash (%) – – – 39.39 12.12* 14.29* PRT: Perception-reaction time. MRT: movement-reaction time. Values enclosed in parentheses represent standard deviation. * Significantly different from null signal (p < 0.05). ing. Perception-reaction time for a beep sound was shorter than that for speech message. However, no significant difference existed for perception-reaction time among the three alarm contents. For the traffic accident at an intersection, the experimental result indicates that perception-reaction time was the longest with no warning. The host vehicle with a speech warning had the shortest perception-reaction time. The difference (F = 5.260, d.f. = 2, p = 0.007) in perception-reaction time was significant among the three alarm contents. Tukey’s honesty significant difference (HSD) was further used for post hoc testing. Statistical analysis shows that differences reached statistical significance between the null signal and beep sound, and between the null signal and speech message. 3.2.2. Movement-reaction time For the accident on the straight roadway segment, the statistical analysis shows that no significant difference existed in movement-reaction time among the three alarm contents. For the traffic accident at an intersection, experimental the results show that movement-reaction time was the longest for the speech alarm. This long time span could be attributed to a driver needing additional time to understand the spoken message before applying the brakes. The ANOVA for movement-reaction time indicates that the difference among the three kinds of alarm content was statistically significant (F = 3.132, d.f. = 2, p = 0.050). The Tukey post hoc tests indicate that the difference of movementreaction time between the null signal and speech message was significant. On the other hand, no difference existed for movement-reaction time between the null signal and beep sound. 3.2.3. Speed For the accident on the straight roadway segment, experimental results (Table 3) show that the difference in average speed, and average speed reduction, of the host vehicle was small among the three types of alarm content. On the other hand, for the traffic accident at an intersection, the average speed of the host vehicle with warning of a beep sound or speech message was slower than the speed of the host vehicle without a warning signal. In addition, average speed reduction for the host vehicle with warning of a beep sound or speech message (28.63 km/h, 28.20 km/h) was significantly larger than the speed reduction of host vehicle without any warning signal (22.82 km/h). Experimental results indicate that driving speed decreased for traffic accident for vehicles equipped with a CWS. 3.2.4. Crash Only one accident occurred on a straight roadway segment. It was because the driver wheeled too much and then crashed into a vehicle in the next lane. For the traffic accident at an intersection, experimental result indicates that the crash rate was 39.39% for vehicles without warn- ing signals. On the other hand, the crash rate was 12.12% and 14.29% for vehicles with warning content of beep sounds and speech messages, respectively. This finding shows an effective crash reduction rate of around 25% for an intersection accident when a vehicle has a CWS. Furthermore, the Chi-squared test was employed to evaluate the statistical difference between crash rates for different conditions of alarm content. This difference between crash rates with a null signal and beep sound was statistically significant (2 = 6.398, p = 0.014). The difference between crash rates for driving with a null signal and speech message was also statistically significant (2 = 4.716, p = 0.049). 3.3. The relationship between driving performances and crash occurrence All crashes except for one occurred at intersections. Thus, the crash on the straight roadway segment is excluded from discussion in this study. For the traffic accident at an intersection, driver perceptionreaction time versus average driving speed was illustrated as a scatter plot, shown in Fig. 2(a). The plot shows that crashed occurred when drivers had long perception-reaction times—this was expected. However, experimental results show that a crash still occurred when drivers perception-reaction time was short. Some crashes even occurred when perception-reaction time less than 1 s. This study also observed driver movement-reaction time and calculated perception-movement time by summing perceptionreaction time and movement-reaction time. Fig. 2(b) shows driver perception-movement time versus average driving speed for each intersection crash. All crashes happened when perceptionmovement time was at least 2 s. According to driving performance at an intersection, this study selected three variables, i.e., perception-movement time, average speed, and average speed reduction, to investigate their level of influence on a crash. Logistic regression analysis was further employed to analyze the relationship between these three driving performance variables and intersection crashes. Logistic regression parameters were calculated and listed in Table 4. The logistic regression model reveals that the possibility of a crash was positively correlated with perception-movement time (OR = 1.402, p < 0.05) and average speed (OR = 0.163, p < 0.05) whereas no strong evidence exists that the possibility of a crash was correlated with average speed reduction (OR = −0.020, p > 0.05). Of the three driving performance variables, perception-movement time had the greatest influence on a crash. The model indicates that an increase of 1 s in perception-movement time induces a fourfold increase in crash risk. On the other hand, one km/h speed reduction decrease crash risk by 2%. In the logistic regression model, variables perceptionmovement time (p = 0.001) and average speed (p = 0.047) were statistically significant. Overall, the fit of the logistic regression model to data was satisfactory; that is, 81.6% of total 120 traffic events were correctly classified into crash or non-crash groups. 1641 S.-H. Chang et al. / Accident Analysis and Prevention 40 (2008) 1637–1643 Fig. 2. (a) Driver perception-reaction time versus average driving speed for the intersection crashes. (b) Driver perception-movement time versus average driving speed for the intersection crashes. 4. Discussion 4.2. Influence of alarm content 4.1. Influence of traffic accident location Due to the CWS, driver perception-reaction time was reduced, providing the driver with additional time to manipulate a vehicle in an emergency. The advantages of a CWS were identified by Lee et al. (2002), who demonstrated that with a collision avoidance system, drivers respond and redirect their attention to the road. Additional, a collision avoidance system reduces the number and severity of collisions, and increases the safety margin. However, drivers displayed different levels of driving performance with different alarm contents at different traffic accident locations in this study. Experimental results for perception-reaction time (Table 3) indicate that a driver did not have the same performance with different alarm systems and in different driving scenarios. On a straight roadway segment, perception-reaction time for a beep sound was shorter than that for a speech message. The likely reason is that drivers did not have to discern the direction of a traffic event on a straight roadway segment. Thus, a speech message did not have a better effect of reducing driver perception-reaction time than a beep sound. Nevertheless, this situation is different at intersections. Driver did not know the direction from which a violator’s vehicle was coming. In this situation, an alarm content is extremely important in advising a driver of the direction of a danger, and a speech message will be better than a beep sound. A study by Suetomi and Niibe (2001) also obtained a similar conclusion indicating that when a sound that drivers experienced in a dangerous situation is generated from the direction in which a dangerous situation occurs, the driver can easily pay attention to the dangerous direction. Experimental results in this study indicate that the result of a traffic event did not depend entirely on perception-reaction time. The result of an emergent traffic event, i.e., a crash or no crash, may differ for drivers even when their perception-reaction times were the same, as shown in Fig. 2(a). Experimental results show This study investigated driving performance at different traffic accident locations in an urban area. Experimental results show that on a straight roadway segment, the vehicle driver, when realizing that the leading vehicle was braking, spent less time to make a decision and take action than at an intersection when faced with a violator vehicle. The likely reason is that at an intersection, a driver must pay attention to vehicles not respecting traffic signage and driving into the intersection from the left or right. Conversely, on a straight roadway segment, a driver only needs to focus on the vehicles in front. This indicates that the traffic conditions at an intersection are more complex than that on a straight roadway segment. Liu and Lee (2005) also indicated that when approaching an intersection, in addition to the high degree of visual and spatial processing necessary for safe navigation, dynamic decision making with respect to accelerating, proceeding or braking is principally governed by approach speed, perceived duration of the signal, and consideration of traffic density at, and immediately prior to, the light changing. In addition, driving in different traffic environments generate different driving performances. A traffic environment can be classified into two categories, as suggested by Liu and Lee (2006), based on a taxonomic approach to information-processing demand: (a) high demand for information processing, such as driving within city centers and complex intersections with road signage; and, (b) low demand for information processing, such as driving on motorways without intersections. Experimental results in this study show that due to different information-processing demands in different traffic conditions, driving performance differed in the two traffic accident locations in an urban area. Table 4 Logistic regression analysis for driving performance and crash Variables Estimated coefficient Perception-movement time Average speed Average speed reduction Constant Goodness of fit 1.402 0.415 0.163 0.082 −0.020 0.859 −12.038 4.054 2 *  = 37.858, p = 0.000 , Hosmer–Lemeshow = 1.802, p = 0.986 * p < 0.05. Estimated standard error p-Value Odd ratio 0.001* 0.047* 0.354 0.003 4.062 1.177 0.981 – 1642 S.-H. Chang et al. / Accident Analysis and Prevention 40 (2008) 1637–1643 that a short perception-reaction time did not prevent a crash. A crash could still occur with short perception-reaction time and long movement-reaction time, as shown in Fig. 2(b). A crash can occur when movement-reaction time is long, as this indicates that a driver is hesitating in braking vehicle. Therefore, perception-movement time is a more suitable indictor than perception-reaction time for predicting crash occurrence. Additionally, driving speed was also a crucial factor. A slow driving speed with fast braking allowed sufficient time to avoid a dangerous traffic event. Average speed and perception-movement time had a great influence on the occurrence of crashes in this study. Average speed and speed reduction data indicate that drivers reduced their speed when they encountered an emergent traffic event. The difference in speed reduction between those with and without a CWS, listed in Table 3, was statistically significant at an intersection. The warning system alerted drivers to an emergency condition, and drivers then responded by releasing the accelerator and applying the brakes. This experimental finding is in agreement with the experimental result obtained by Penney (1999), who found that low speed when approaching an intersection followed installation and operation of a CWS. The experimental results in this study demonstrate that a CWS for a vehicle at an intersection is effective in alerting a driver to a dangerous vehicle early, and reducing driver perception-reaction time. Evans (1991) also indicated that reduced reaction time reduces the probability and severity of crashes. In addition, the purpose of the ICWS, as indicated by Penney (1999), is to enhance driver awareness of the traffic situations at intersections by providing timely and easily understood warnings of vehicles entering an intersection. The experimental results of this study also suggest that driver can take appropriate actions to avoid a crash with an effective warning signal, indicating that a speech message had a better effect in reducing driver perception-reaction time than did a beep sound. In city streets, one cannot have faith in other drivers obeying stop signs, or adhering to right-of-way rules. Evans (1991) confirmed this point and indicated that a vehicle proceeding after a traffic light has turned red is not rare. Many drivers seem to attack stop signs at high speeds, and brake at the last moment. A traffic accident at an intersection was designed in this study to simulate typical road conditions. This study found that the crash rate was significantly reduced, around 25%, for an intersection accident when a vehicle had a CWS, and demonstrated the effectiveness of an ICSW in alerting driver entering an intersection. 4.3. The relationship between driving performance and crash occurrence For a traffic event at an intersection, average speed and perception-reaction time (r = 0.658, p = 0.000) were positively correlated by Pearson correlation analysis. The positive correlation indicates that as average speed increases, the perception-reaction time of a driver increases. When a driver was unaware of an emergent traffic event and did not take any braking action his driving speed was high. Long perception-reaction time was thus induced after a driver noticed an emergency traffic event. On the other hand, a long perception-reaction time meant that a driver noticed a traffic event late, and, thus, driving speed stayed high. A negative correlation (r = −0.702, p = 0.000) depicted an inverse relationship between perception-reaction time and speed reduction in this study. This analytical result indicates as perception-reaction time decreased, the amount of speed reduction increased. It showed when a driver had reduced perceptionreaction time, that driver had ample time to apply the brakes, and thereby reduce driving speed. Driver performances were correlated with each other, and a crash can be always attributed to such performance. Through logistic regression analysis, the relationship between driving performance and crash occurrence was determined (listed in Table 4). Among the performance variables, perception-movement time had the greatest influence on crash occurrence. When perception-movement time increases by 1 s, the possibility of crash increases four times. In addition, a decrease in driving speed reduces the odds ratio of a crash. We expected that a decrease in speed via braking directly impact the likelihood of a crash. Typically, a slow reaction time increases the possibility of a crash occurring when a driver does not identify that a risk is present. A long perception-movement time and small speed reduction result from a slow reaction to a traffic event. Therefore, the results of regression analysis of perception-movement time and speed reduction are understood easily. 4.4. Limitations and comparison of different driving simulators A driving simulator was employed in this study for driving experiments as real traffic environments are extremely complex and some factors are uncontrollable in on-road tests. Previous investigations using different driving simulators mainly focused on different applications of CWSs. The driving simulators employed in the studies by Lee et al. (2002) and Suzuki and Jansson (2003) incorporated with a motion platform, whereas only the simulators used in the studies by Kolisetty et al. (2006), Charlton (2007), and C. Ho et al. (2006) had a multi-screen. However, to study driving performance when a driver encounters an emergency event, an IOT driving simulator is suitable because it involves a real vehicle and has six degrees-of-freedom. The Stewart motion platform gives a subject an experience close to that in reality while driving. Furthermore, an IOT driving simulator provides a wide front view using a multi-screen to present traffic conditions, such as that when a violator’s vehicle drives from left or right into an intersection. These were the advantages of the IOT driving simulator used for the specific topic. Based on their experience driving on roads, subjects in this study had no difficulty controlling the simulator, and responded to the simulator with normal driving behaviors. The experimental data recorded by the simulator also generated reasonable results and logical trends. However, the driving environment in the simulator differs from that in the real world. Drivers may drive sparingly and tend to underestimate the consequence of a crash in a simulated environment in which road dangers are excluded. Therefore, simulator-based experiment requires extensive on-road tests for validation. Part of the validation, including vehicle dynamics simulation and a visual system, was done (Lin et al., 2004a,b; Ting et al., 2008), however, it is worthy of further investigation. The experimental scenario in this study involved a designed driving task on a straight, urban, two-lane road with intersections. Subjects experienced emergency events both on straight roadway segment and at intersection. However, serious accidents, such as rollover, were not involved as the driving simulator cannot simulate such damage. Only front-end collisions on a roadway segment and side/front-to-side collisions at an intersection occurred and were discussed in this study according to the designed scenario. The experimental scenario used for further research can involve relatively more complex collision types, such as an oblique collision or front-end/rear-end collision on a straight roadway segment, large curve and at an intersection, to extend the investigation of the relationship between crash occurrence and driving performance. S.-H. Chang et al. / Accident Analysis and Prevention 40 (2008) 1637–1643 5. Conclusion This study investigated the two most common traffic accident locations in urban areas, i.e., a straight roadway segment and an intersection, using a driving simulator equipped with a CWS. Experimental environments were established to assess the influence of traffic accident location and alarm content on driving performance, and identify the relationship between crash occurrence and driving performance. Experimental results indicate that because of different information-processing demands in different traffic conditions, driving performance differed at the two different road locations in an urban area. On a straight roadway segment, perception-reaction time for a warning beep sound was shorter than that for a speech message warning. Nevertheless, at an intersection, a driver was unaware of the direction of an oncoming vehicle. In this situation, a speech message was of greater assistance to drivers than a beep sound and, thus, perception-reaction time was effectively reduced. In addition, as driver performances were correlated, the relationship between driving performance and crash occurrence was determined using logistic regression analysis. The analytical result indicates that perception-movement time has the greatest influence on crash occurrence. Acknowledgements This study was commissioned by the Institute of Transportation (Transportation Research Centre). Fu-Chuan Wu is appreciated for his data processing and contributions to this study. References Abe, G., Richardson, J., 2005. The influence of alarm timing on braking response and driver trust in low speed driving. Safety Science 43, 639–654. Abe, G., Richardson, J., 2006. Alarm timing, trust and driver expectation for forward collision warning systems. Applied Ergonomics 37, 577–586. Charlton, S.G., 2007. The role of attention in horizontal curves: a comparison of advance warning, delineation, and road marking treatments. Accident Analysis and Prevention 39, 873–885. Chen, W.H., Lin, C.Y., Doong, J.L., 2005. Effects of interface workload of in-vehicle information systems on driving safety. Transportation Research Record 1937, 73–78. Clement, S., Taylor, M., 2006. The simple platoon advancement model of ITS technologies applied to vehicle control at signalized intersections. Journal of Advanced Transportation 40, 1–21. 1643 Evans, L., 1991. Traffic Safety and The Driver. Van Nostrand Reinhold, New York, USA. Ho, C., Tan, H.Z., Spence, C., 2005. Using spatial vibrotactile cues to direct visual attention in driving scenes. Transportation Research Part F 8, 397–412. Ho, A.W.L., Cummings, M.L., Wang, E., Tijerina, L., Kochhar, D.S., 2006. Integrating intelligent driver warning systems: effects of multiple alarms and distraction on driver performance. In: TRB 2006 Annual Meeting CD-ROM. Ho, C., Reed, N., Spence, C., 2006. Assessing the effectiveness of “intuitive” vibrotactile warning signals in preventing front-to-rear-end collisions in a driving simulator. Accident Analysis and Prevention 38, 988–996. Jahn, G., Oehem, A., Krems, J.F., Gelau, C., 2005. Peripheral detection as a workload measure in driving: effects of traffic complexity and route guidance system use in a driving study. Transportation Research Part F 8, 255–275. Kolisetty, V.G.B., Iryo, T., Asakura, Y., Kuroda, K., 2006. Effect of variable message signs on driver speed behavior on a section of expressway under adverse fog conditions—a driving simulator approach. Journal of Advanced Transportation 40, 47–74. Lee, J.D., McGehee, D.V., Brown, T.L., Reyes, M.L., 2002. Collision warning timing, driver distraction, and driver response to imminent rear-end collisions in a highfidelity driving simulator. Human Factors 44, 314–334. Lin, F.F., Chen, W.H., Hwang, J.R., Chang, K.K., 2004a. Driving simulatior development at IOT and future application targets. In: The Seventh Sino-French Symposium on Ergonomic Design and Research Transportation & Workplace Safety, pp. 37–42. Lin, F.F., Chang, K.K., Chang, C.C., Huang, C.H., Doong, J.L., 2004b. Validation of a real-time vehicle dynamics simulation program. The Cross-Strait Forum on the Intelligent Transport System, 278–283. Liu, B.S., Lee, Y.H., 2005. Effects of car-phone use and aggressive disposition during critical driving maneuvers. Transportation Research Part F 8, 369–382. Liu, B.S., Lee, Y.H., 2006. In-vehicle workload assessment: effects of traffic situations and cellular telephone use. Journal of Safety Research 37, 99–105. Maltz, M., Shinar, D., 2007. Imperfect in-vehicle collision avoidance warning systems can aid distracted drivers. Transportation Research Part F 10, 345–357. Ministry of Transportation and Communication (MOTC). Traffic accident statistics, January 2006–December 2006, Taiwan. http://www.motc.gov.tw/motchypage/ view95/d4150.wdl. Penney, T., 1999. Intersection Collision Warning System. Report No. FHWARD-99-103. Federal Highway Administration, United States Department of Transportation, Washington, D.C. Sengupta, R., Rezaei, S., Shladover, S.E., Cody, D., Dickey, S., Krishnan, H., 2007. Cooperative collision warning systems: concept definition and experimental implementation. Journal of Intelligent Transportation Systems 11, 143–155. Suetomi, T., Niibe, T., 2001. A human interface design of multiple collision warning system. In: International Driving Symposium on Human Factors in Driver Assessment, Training, and Vehicle Design, pp. 14–17. Suzuki, K., Jansson, H., 2003. An analysis of driver’s steering behaviour during auditory or haptic warnings for the designing of lane departure warning system. JSAE Review 24, 65–70. Ting, P.H., Hwang, J.R., Fung, C.P., Doong, J.L., Jeng, M.C., 2008. Rectification of legibility distance in a driving simulator. Applied Ergonomics 39, 379–384.