This article was downloaded by: [The Aga Khan University] On: 09 October 2014, At: 08:29 Publisher: Taylor & Francis Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK
Traffic Injury Prevention Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/gcpi20
Risk Factors Related to Fatal Truck Crashes on Korean Freeways a
a
Saerona Choi , Cheol Oh & Mijeong Kim
a
a
Department of Transportation and Logistics Engineering , Hanyang University , Ansan-si , Korea Accepted author version posted online: 28 Feb 2013.Published online: 26 Nov 2013.
Click for updates To cite this article: Saerona Choi , Cheol Oh & Mijeong Kim (2014) Risk Factors Related to Fatal Truck Crashes on Korean Freeways, Traffic Injury Prevention, 15:1, 73-80, DOI: 10.1080/15389588.2013.778989 To link to this article: http://dx.doi.org/10.1080/15389588.2013.778989
PLEASE SCROLL DOWN FOR ARTICLE Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and should be independently verified with primary sources of information. Taylor and Francis shall not be liable for any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of the Content. This article may be used for research, teaching, and private study purposes. Any substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http:// www.tandfonline.com/page/terms-and-conditions
Traffic Injury Prevention (2014) 15, 73–80 C Taylor & Francis Group, LLC Copyright ISSN: 1538-9588 print / 1538-957X online DOI: 10.1080/15389588.2013.778989
Risk Factors Related to Fatal Truck Crashes on Korean Freeways SAERONA CHOI, CHEOL OH, and MIJEONG KIM Department of Transportation and Logistics Engineering, Hanyang University, Ansan-si, Korea
Downloaded by [The Aga Khan University] at 08:29 09 October 2014
Received 14 November 2012, Accepted 18 February 2013
Objectives: The major purpose of this study was to identify risk factors affecting truck crashes on freeways and propose recommendations for safer truck traffic operations. Methods: Truck crashes were analyzed to identify how truck traffic safety is related to prevailing traffic and weather conditions. Prevailing traffic conditions were characterized by central tendencies and the spatiotemporal variation of traffic parameters collected from freeway traffic surveillance systems. A total of 377 truck crashes occurring on Korean freeways in a recent 3-year period, 2008–2010, were analyzed together with corresponding prevailing traffic conditions and weather conditions. Several statistical tests were conducted to understand the characteristics of prevailing traffic conditions before crash occurrence based on different weather conditions. In addition, a binary logistic regression technique was applied to identify causal factors affecting truck crash severity under normal and adverse weather conditions. Results: Major findings from the analyses are discussed with truck operations strategies including speed enforcement, variable speed limit, and truck lane restriction from the safety enhancement point of view. Speed-related variables representing prevailing traffic conditions before crash occurrences were found to be the most statistically significant factors affecting truck crash severity, compared to volume-related variables such as the volume-to-capacity ratio (v/c). It is inferred that speed management is an effective tool for safer truck traffic operations on freeways. The major findings can be further discussed to derive valuable insights into truck traffic operations based on different weather conditions, such as normal and adverse. Conclusions: Some recommendations for safer truck traffic operations were presented based on the results obtained. The outcomes of this study could be effectively utilized to support the development of various traffic operations strategies and policies for truck traffic safety. Supplemental materials are available for this article. Go to the publisher’s online edition of Traffic Injury Prevention to view the supplemental file. Keywords: traffic safety, truck crashes, risk factors, truck traffic operations, injury severity, environmental conditions
Introduction Interest in truck traffic has increased largely due to greater safety concerns regarding truck-related crashes, in addition to the substantial economic impact of truck traffic on freight logistics. Recent Korean crash statistics (Traffic Accident Analysis System) revealed that the total number of fatalities caused by truck-involved crashes was 1266 in 2010, which makes up approximately 24 percent of total traffic fatalities. Safety issues regarding trucks can be described by several characteristics. First, the truck mass is generally greater than that of other vehicle types; thus, the resulting crash outcomes of truck crashes are more severe (Evans and Frick 1993). Sec-
Address correspondence to Saerona Choi, Department of Transportation and Logistics Engineering, Hanyang University, Sa 3-dong, Sangrok-gu, Gyeonggi-do, Ansan-si, Korea. E-mail: [email protected]
ond, when a truck is the leading vehicle in a car-following situation, the following smaller vehicle will struggle with ensuring a larger safety distance because the follower is not able to readily identify unexpected upcoming hazards. This situation is highly associated with an increased potential of rear-end collisions (Abdel-Aty and Abdelwahab 2004). Third, commercial truck drivers travel longer distances compared to the general public and tend to have increased nighttime driving times to avoid traffic congestion. Due to these circumstances, truck drivers are exposed to more hazardous situations caused by fatigued driving (Cantor et al. 2010). How to prevent truck-related crashes and reduce injury severity to minimize economic loss is an important consideration. One of the promising approaches to this issue is to devise effective truck traffic operations strategies that not only enhance safety but also alleviate congestion. The purpose of this study is to derive useful insights from analyses of truck crashes that can be further used for effective truck traffic operations for safety enhancement.
Downloaded by [The Aga Khan University] at 08:29 09 October 2014
74 To determine a research opportunity in accordance with the study purpose, a literature review was conducted from the traffic engineering point of view that a better understanding of the characteristics of truck crashes is necessary for safer traffic operations. The relevant literature can be broadly categorized into 2 areas. One area concerns truck traffic operations and the other concerns truck crash analysis. One of the countermeasures to alleviate safety concerns for truck traffic is to develop effective truck traffic operation strategies such as lane operations for trucks and differential speed limits by vehicle type. Significant efforts have been made to separate trucks from other vehicle types to provide more homogeneous and stable traffic streams (El-Tantawy et al. 2009; Fontaine 2008; Fontaine et al. 2009). Various lane operation scenarios have been evaluated based on field tests and simulation evaluations. Another approach is to apply differential speed limits to take the performance of truck maneuvering into account (Korkut et al. 2010). In addition, various existing studies have been conducted to analyze truck-related crashes. Their major purposes are to identify the dominant factors affecting crash severity (Chang and Mannering 1999; Duncan et al. 1998; Golob et al. 1987; Khattak et al. 2003; Khorashadi et al. 2005; Lemp et al. 2011; Zhu and Srinivasan 2011a). Among them, Golob et al. (1987) analyzed incident duration as a risk factor associated with truck-related crashes. Truck driver behavior, such as cell phone use and human factor characteristics, has also been explored to understand the characteristics of truck crashes (Cantor et al. 2010; Gander et al. 2006; H¨akk¨anen and Summala 2001; Hickman and Hanowski 2012; Zhu and Srinivasan 2011b). Truck crashes by lanes (Hallmark et al. 2009) and overturning crashes were also analyzed to capture more dominant factors (Young and Liesman 2007). Crash compatibility between different vehicle types was investigated by Gabler and Hollowell (2000). A focus of this study was to derive useful variables from prevailing traffic conditions that can be associated with truck safety. An important characteristic of truck traffic is that the flexibility of determining travel schedules for truck drivers is significantly limited, unlike for passenger vehicles, because the travel schedule is highly associated with commercial values. As a result, previous studies (National Institute for Occupational Safety and Health 2007; U.S. Department of Transportation 2005) have reported that truck drivers are often exposed to worse driving conditions compared to other drivers, such as adverse weather conditions. It is also known that adverse weather conditions tend to deteriorate traffic performance (Keay and Simmonds 2005; Knapp and Smithson 2000; Shi et al. 2011). Therefore, both prevailing traffic conditions and adverse weather conditions should be taken into account in analyzing truck safety. Although there has been significant research in the domain of truck crash analysis, we are not aware of any study that has attempted to comprehensively relate prevailing traffic conditions and adverse weather conditions to truck safety. This research opportunity motivated us to uncover valuable insights into truck traffic operations in terms of safety based on analyzing traffic and adverse weather conditions together with crash data. The availability of speed and traffic volumes obtained from traffic surveillance systems allowed us to prepare for potential
Choi et al. variables representing traffic conditions, which were matched with crash data. This study used freeway crashes occurring during a recent 3-year period, 2008–2010, with corresponding traffic surveillance data and weather data. Several statistical tests were conducted to obtain statistically significant evidence to associate truck traffic safety and prevailing traffic conditions. In addition, a binary logistic regression (BLR) technique was applied to model the relationship between truck crash severity and potential contributing factors obtained from the above data sources. A set of independent variables representing traffic conditions was established and analyzed together with weather and geometric conditions data. The severity of each truck crash was used as a dependent variable. Based on the findings from these analyses, this study suggests some recommendations for safer truck traffic operations on freeways.
Data Descriptions This study used 3 main data sources: the Korean freeway crash database, traffic surveillance systems data, and weather conditions data. Data from all crashes involving trucks on Korean freeways in a recent 3-year period, 2008–2010, were extracted and analyzed together with the corresponding traffic and weather conditions data. Crash Data The Korean freeway crash database contains useful information on crashes. In addition to crash-related information such as time, location, crash type, and severity, relevant geometric conditions such as horizontal and vertical alignment were included in crash record files and used to prepare for a set of independent variables. The truck crash data were obtained from the Korean Expressway Corporation (KEC) and consisted of 377 truckinvolved crashes. Truck crash cases that occurred in the study area were sequentially reduced based on a set of selection criteria for more systematic and reliable analyses. The selection criteria include crash location, crash cause, weather conditions, driver characteristics, and consistency of weather information. First, crashes that occurred on the freeway mainline were selected. Crashes that occurred on ramps, at toll gates, and at rest areas were excluded from the data set to be analyzed. Second, crashes resulting from vehicle malfunctioning were excluded. Because this study focused on the interactions among driver, geometry, traffic, and weather conditions, crashes resulting from driver-related causes were only included in the data set. Third, crash cases with incomplete driver information—for example, driver age, gender, driving experience, etc.—were excluded from the data set. Lastly, crash case, which weather information from crash database presented by KEC and Korean weather agency was inconsistent, was also excluded from the data set. As a result, a total of 972 truck crashes were reduced to 377. A total of 77 people were severely injured, which includes fatalities, and 300 were slightly injured. Several freeways that
Fatal Truck Crashes on Korean Freeways
75
Table 1. Characteristics of truck crash data by weather conditions Normal
Downloaded by [The Aga Khan University] at 08:29 09 October 2014
Freeway line
Rain
Snow
Fatal
Nonfatal
Fatal
Nonfatal
Fatal
Nonfatal
Total
Namhae Youngdong Jungbu naeryuk
8 20 32
46 62 93
7 3 1
37 33 16
0 4 2
2 4 7
100 126 151
Total
60
201 261
11
86
6
13
377
97
19
have significant truck volumes were selected. Because our main focus was to address the impact of traffic and weather conditions on truck safety, crashes that did not have matched traffic and weather data were excluded. A total of 377 truck crashes were used in this study for the analysis, excluding 68 crash cases for which corresponding traffic and weather data were not available. The KEC crash database describes crash severity using 4 different categories: A, B, C, and D. Crashes including 3 or more fatalities fall into category A. Crashes including 1–2 fatalities are defined as category B. Crashes involving injuries fall into category C. Category D includes other minor crashes including pedestrian–vehicle crashes that occurred at rest areas. This study used crashes from categories A, B, and C that included fatalities and injuries. Because fatal crashes fall into categories A and B, this study classified injury severity into 2 groups: fatal (A and B) and nonfatal (C). Table 1 presents a summary of the crash data used in this study. Traffic and Weather Data To develop safety insights into truck traffic operations, the impact of trucks on traffic streams needs to be evaluated. From a disaggregate point of view, the interaction between trucks and neighboring vehicles would be obviously different from that of homogeneous vehicles. From an aggregate point of view, the prevailing traffic conditions, such as average speed and traffic volume, would vary by the proportion of trucks. Because analysis of interactions among vehicles including trucks requires maneuvering data for each vehicle, which are not readily available in practice, analysis of prevailing traffic conditions is of much interest. The second source used was freeway traffic surveillance systems. The KEC collects and maintains traffic data for both point measurements and section measurements. Point measurements include speed, volume, and occupancy, which are collected from detector stations in the field. The detector spacing is approximately 1 km. Travel times are also available for specific freeway segments, where automatic vehicle identification systems are implemented based on dedicated short-range communication. This study used point measurements to derive variables representing traffic conditions because automatic vehicle identification systems have not been sufficiently implemented to provide traffic data corresponding to the truck crashes to be analyzed in this study. Traffic data aggregated in 15-min intervals collected from neighboring detector stations upstream and downstream from the crash location were used in this study. More specifically, these data included speed and volume data for 15-min before
crash occurrences. The spatiotemporal characteristics of prevailing traffic conditions need to be taken into consideration in developing independent variables. In addition to the average values representing the central tendency, the standard deviation of speed and volume over a period longer 15-min and the difference between upstream measurements and downstream measurements were explored to establish a set of candidate independent variables. Because speed and traffic count data were sometimes missing, a proper imputation method needed to be applied to prepare for the traffic data. When missing data occurred, adjacent upstream or downstream detector data for the same period was used to impute missing points. When adjacent detector data were also missing, we excluded the corresponding crash data from the analysis because the reliability of traffic data could not be assured. A set of candidate traffic parameters to represent the prevailing traffic conditions that can be potentially associated with truck crashes on freeways was prepared for the analysis and is presented in Table 2. The weather data used in this study were rainfall intensity and snowfall intensity. Although the crash database includes weather-related information, that information is mainly focused on roadway surface conditions, such as dry, wet, and icy. We used more detailed information to consider the severity of the conditions that would have an impact on a driver’s visibility. The intensity data were obtained from the Korean weather agency’s archives. The nearest weather station was matched with each crash in archiving the intensity data. This study defines adverse weather conditions as those involving rainfall and snowfall, and normal weather conditions include clear and cloudy conditions where drivers are not restricted by visibility and slippery road surfaces.
Characteristics of Traffic Conditions Before Truck Crash Based on Weather Conditions To identify the overall characteristics of traffic conditions before a truck crash occurred on freeways, several statistical analyses were conducted. The focus of the analysis was on evaluating traffic conditions based upon weather conditions. Analysis of variance (ANOVA) and Scheffe’s post hoc test were conducted to investigate whether the proposed traffic variables representing the prevailing traffic conditions would be differentiated by weather conditions such as normal, rain, up dn and snow. The null hypotheses for 4 variables, v15 , v15 , Vcv , and speeding, were rejected, which implies that the traffic conditions before a crash occurrence were significantly different based on the weather conditions. Appendix A (see online supplement) presents the ANOVA results for the variables that exhibited a significant difference. It can be said that if the variance between groups was greater than the variance within groups, the difference between the groups was statistically significant. Useful findings were identified by the post hoc test. A up prominent distinction for averages of v15 under different weather conditions was observed. Speed reduction due to snowfall was much greater than for other conditions.
76
Choi et al.
Table 2. Variable descriptions Variables
Variable names
Injury severity Crash location Temporal characteristics
severity bri/tun day night week age grade up grade dn curve up v15 dn v15 up v/c15
Driver Roadway characteristics
Traffic conditions
dn v/c15
Downloaded by [The Aga Khan University] at 08:29 09 October 2014
d v15 Vcv d v/c15 V/Ccv speeding
Description
ANOVA
BLR
0: nonfatal, 1: fatal 0: mainline, 1: bridge or tunnel 0: daytime, 1: nighttime 0: weekend, 1: weekdays Continuous/countable variable 0: no uphill, 1: uphill grade 0: no downhill, 1: downhill grade 0: tangent, 1: curve 15-Min average speed before crash occurrence at upstream detection station 15-Min average speed before crash occurrence at downstream speed 15-Min volume-to-capacity ratio before crash occurrence at upstream detection station 15-Min volume-to-capacity ratio before crash occurrence at downstream detection station 15-Min average speed difference between upstream and downstream detection stations before crash occurrence Coefficient of variation of speeds collected from upstream and downstream detection stations 15-Min volume-to-capacity ratio difference between upstream and downstream detection stations before crash occurrence Coefficient of variation of volume-to-capacity ratio calculated from upstream and downstream detection stations up up up Difference between v15 and speed limit (when v15 exceeds speed limit) 0: v15 does not exceed speed limit
— — — — — — — — Central tendency
DV IV IV IV IV IV IV IV IV — IV —
Spatiotemporal variation
IV IV — —
Driving pattern
IV
DV = dependent variable, IV = independent variable.
In addition, snowfall led to higher speed variation based on the analysis of Vcv , representing the magnitude of speed variations in prevailing traffic conditions before crash occurrences. Therefore, an effort to reduce speed variations during snowfall would be a viable approach to prevent crashes. Another interesting finding was observed for the variable speeding. The results indicated that drivers were more likely to exceed speed limits during rainfall compared to normal or snowfall conditions. Thus, speed enforcement under rainy conditions would be more effective at preventing crashes than enforcement under normal or snowfall conditions. Appendix B presents the post hoc results.
Analysis of Factors Contributing to Truck Crash Severity A logistic regression modeling approach was applied to the binary classification of the level of crash severity. The relationship between the dependent variable, which was a nonmetric variable in particular, and one or more independent variables was modeled. For the BLR model, multivariate procedure, the dependent variable took the value 1 or 0. The BLR model predicts the probability that the dependent variable takes the value of 1. In this case, 1 indicates that a truck-involved crash would belong to the group including fatal crashes. The form of the logistic regression model in this study is Prob (yi = 1) =
exp[ f (Xk , βk )] . 1 + exp[ f (Xk , βk )]
Prob (yi = 1) is the probability that the response variable yi includes a fatal truck crash; xk is a vector of explanatory variable
representing factors affecting the severity of the truck crash; βk is a vector of parameters to be estimated. Probability values can be any value between 0 (nonfatal crash) and 1 (fatal crash), but the predicted value cannot be outside of this range. To define a relationship bounded by 0 and 1, logistic regression uses an assumed relationship between the independent and dependent variables that resembles an Sshaped curve. The odds ratio represents the probability of an event divided by the probability of the event not happening. Because this is a ratio, its values range from zero to infinity (Hosmer and Lemeshow 2000). Factors with larger odds ratio have a larger impact on injury severity. First, both rainfall and snowfall intensities were applied as independent variables that represent adverse weather conditions. However, we were not able to uncover meaningful results that would be useful for traffic operations. A feasible reason would be the limitation associated with our data set. In other words, the range of rain and snowfall intensity data was not sufficiently wide to cover actual situations in practice. Because ANOVA results showed that traffic conditions are different by weather conditions, we divided our data set into 2 groups: one that involves crashes that occurred under normal weather conditions and one that involves crashes that occurred under adverse weather conditions. A set of candidate variables that would be potentially associated with truck safety were prepared for the analysis. These factors included driver characteristics and geometric design elements in addition to traffic and weather conditions. Table 2 presents the variables that were used in this study for conducting the crash severity analysis to derive useful insights into truck traffic operations on freeways for safety enhancement. Some variables for traffic conditions were excluded from the analysis based on the correlation analysis.
Fatal Truck Crashes on Korean Freeways
77
Table 3. BLR model of truck crash severity under normal weather conditions 95% Confidence interval for odds ratio
Downloaded by [The Aga Khan University] at 08:29 09 October 2014
Result
day night week age grade up grade dn curve bri/tun up v15 speeding d v15 Vcv up v/c15 constant Log likelihood Number of observations LR χ 2 (12) Pseudo R2
Coefficient
SE
t
0.916 0.275 0.006 −0.006 −0.275 −0.074 0.346 −0.009 0.128 0.042 −10.179 0.791 −1.257
0.375 0.449 0.015 0.409 0.406 0.331 0.648 0.016 0.065 0.029 5.582 1.114 1.821
2.440 0.610 0.360 −0.010 −0.680 −0.220 0.530 −0.560 1.970 1.470 −1.820 0.710 −0.690
P > |t| .015 .540 .716 .989 .499 .824 .594 .578 .048 .141 .068 .478 .490 −132.993 261 15.580 0.055
Odds ratio
Lower
Upper
2.498 1.317 1.006 0.994 0.760 0.929 1.413 0.991 1.137 1.043 3.8∗10−5 2.205 —
1.198 0.547 0.976 0.446 0.343 0.486 0.396 0.960 1.001 0.986 6.73∗10−10 0.248 —
5.208 3.173 1.036 2.216 1.684 1.776 5.034 1.023 1.291 1.103 2.143 19.577 —
Analysis of Crash Severity Under Normal Weather Conditions
Analysis of Crash Severity Under Adverse Weather Conditions
The Hosmer and Lemeshow goodness-of-fit test provides a χ 2 value of 15.58. The χ 2 value is less than the target χ 2 (significance level = .05, df = 12) value of 21.00, as presented in Table 3. In addition, the P-value was less than 0.05. Therefore, the Hosmer and Lemeshow test indicated that the model did not have a significant lack of fit. Two independent variables were identified as significant risk factors affecting crash severity at a 5 percent significance level. In addition to day night, speeding, which represents the overall tendency of drivers’ speed selection behavior before a crash occurs, was found to be significant. The odds ratio for day night was 2.5975, which implies that when a truckinvolved crash occurs at nighttime, the probability of a fatal crash would be 2.5 times greater than that of a daytime crash. Regarding speeding, it was found that when the prevailing speed exceeded a posted speed limit, a 1 km/h speed increment increased the possibility of a fatal crash by approximately 14 percent. When a 10 percent significance level was used, which is a more generous evaluation criterion, Vcv was found to be meaningful. Vcv indicates the speed variation, which is a well-known surrogate measure for safety before a crash occurs. Interestingly, the sign of the coefficient for Vcv was observed to be negative. Vcv decreased when the standard deviation for speed decreased while the average speed increased. Therefore, this finding also supports the idea that a higher prevailing speed would lead to a more severe crash. In addition, d v15 was identified as the most prominent factor affecting the crash severity (P = .141) among the insignificant variables. This result may be because the speed difference between upstream and downstream was closely associated with rear-end collisions. In particular, when a smaller vehicle collides with a large truck, the crash severity would be expected to increase remarkably. However, an extensive analysis with a larger data set is necessary to draw more generalized and objective conclusions on this matter.
The Hosmer and Lemeshow goodness-of-fit test yielded a χ 2 value of 21.01, which is a slightly higher than that calculated for the normal weather conditions. The χ 2 value was almost the same as the target χ 2 (significance level = .05, df = 12) value of 21.00. Table 4 presents the BLR modeling results for truck crash severity under adverse weather conditions. up Three independent variables including bri/tun, v15 , and Vcv were identified as significant risk factors affecting crash severity at a 5 percent significance level. When a 10 percent significance level was used, speeding and grade dn were found to be meaningful. The positive sign of the coefficient for bri/tun implies that a truck crash occurring on a bridge or in a tunnel was more up likely to be fatal. The result for v15 indicated that higher speed increased fatalities. It was also found that a 1 km/h increase in prevailing speed led to an approximately 9 percent increase in the possibility of a fatal crash. Regarding grade dn, it can be said that the truck crash severity increased when a crash occurred on a downhill grade. Interestingly, 2 variables, Vcv and speeding, were found to produce opposite results. Unlike the result for normal weather conditions, as the speed variation increased, the crash severity increased. This finding is possibly the reason why rainfall or snowfall would cause driver’s behavior for speed selection to vary. However, it should be noted that this impact was dependent on the intensity of rain and snow. In other words, both the average speed and speed variation decreased in the case of heavy snow or rain because the driver’s speed selection was significantly restricted. We do not believe that our data set contained a high enough intensity level to address this issue. Therefore, a further study to identify a cutoff point for the intensity level resulting in different impacts of weather conditions on speed variation would be of interest. Regarding speeding, when the prevailing speed was less than the posted speed limit, the crash severity decreased. A possible reason for
78
Choi et al.
Table 4. BLR model of truck crash severity under adverse weather conditions 95% Confidence interval for odds ratio
Downloaded by [The Aga Khan University] at 08:29 09 October 2014
Result
day night week age grade up grade dn curve bri/tun up v15 speeding d v15 Vcv up v/c15 constant Log likelihood Number of observations LR χ 2 (12) Pseudo R2
P > |t|
Coefficient
SE
t
0.374 1.003 −0.010 1.229 1.695 0.537 3.476 0.088 −0.114 −0.047 20.806 0.326 −12.071
0.721 0.962 0.031 1.057 0.995 0.639 1.580 0.042 0.065 0.040 8.280 2.100 4.504
0.520 1.040 −0.320 1.160 1.700 0.840 2.200 2.060 −1.740 −1.180 2.510 0.160 −2.680
this finding is that adverse weather conditions reduce capacity, resulting in a speed decrease such that most crashes occur under low-speed conditions. Another possible reason is that most drivers are more likely to conform to speed limits under adverse weather conditions. Investigation of Feasibility for Separate Modeling by Weather Conditions This study applied the likelihood ratio test used in previous studies (Chen and Chen 2011; Ulfarsson and Mannering 2004) to evaluate the feasibility of establishing separate models for normal and adverse weather conditions. The following statistics were used to conduct this test: β βN βA LR = −2 LLTotal − LLNormal , − LLAdverse β
where LLTotal is the log-likelihood for the integrated model βN is the log-likelihood for the model using all data; LLNormal βA is the logusing normal weather conditions; and LLAdverse likelihood for the model using adverse weather conditions. As a result, a χ 2 value of 26.77 was obtained, which is greater than the threshold point 21.00 with df = 12 and P < .05, indicating that the difference between normal and adverse weather conditions was statistically significant. From this result, the approach for separate modeling is reasonable.
.604 .297 .748 .245 .089 .400 .028 .039 .081 .240 .012 .877 .007 −37.829 116 21.010 0.217
Odds ratio
Lower
Upper
1.454 2.725 0.990 3.419 5.444 1.711 32.334 1.092 0.892 0.954 1.09∗109 1.385 —
0.354 0.414 0.933 0.431 0.774 0.489 1.462 1.004 0.785 0.882 97.238 0.023 —
5.967 17.950 1.051 27.139 38.302 5.986 715.077 1.186 1.014 1.032 1.21∗1016 84.908 —
In addition, the result of the goodness-of-fit test indicated that Pearson’s P-values were .345 and .127 for normal and adverse weather conditions, which are both greater than .05. These results indicate that the null hypothesis could not be rejected at a 95 percent confidence interval, which means that both models for normal and adverse weather conditions had acceptable fitness. Table 5 summarizes the results of the log-likelihood and goodness-of-fit tests.
Recommendations for Safer Truck Traffic Operations Speed-related variables representing prevailing traffic conditions before crash occurrences were found to be the most statistically significant factors affecting truck crash severity, compared to volume-related variable such as the volume-tocapacity ratio (v/c). It is inferred that speed management is an effective tool for safer truck traffic operations on freeways. The major findings can be further discussed to derive valuable insights into truck traffic operations based on different weather conditions such as normal and adverse. First, the finding that truck crash severity was greater at nighttime indicates that an emphasis on speeding enforcement should be made at nighttime under normal weather conditions. Another important result was that speeding was found to be statistically significant. In addition, the improvement of lighting conditions and proper provision of dedicated truck rest facilities
Table 5. Results of model comparison and goodness-of-fit test
Goodness-of-fit test
Model comparison
Number of observation covariate patterns Pearson’s χ 2 Prob > χ 2 Log likelihood (T) Comparison Prob > χ 2
Normal weather conditions
Adverse weather conditions
261 260 255.480 0.345
116 116 119.530 0.127 −184.206 26.77 26.77 > 21.00
Downloaded by [The Aga Khan University] at 08:29 09 October 2014
Fatal Truck Crashes on Korean Freeways would also be effective because crashes occurring at nighttime are associated with poor visibility and drowsy driving. Second, a variable speed limit (VSL) strategy is a promising tool for preventing truck crashes under adverse weather conditions. In particular, the VSL would be more effective for freeway segments including downhill grades, tunnels, and bridges. In particular, because Vcv was a dominant variable affecting severity, the major focus for VSL should be on minimizing the speed variation to enhance truck traffic safety. Moreover, advance warning information systems that inform drivers of upcoming hazards can be applied together with VSL strategies. In addition to speed management strategies, another viable approach is lane usage. Because truck lane restriction can make the traffic stream more homogeneous by separating trucks from other vehicles, the speed variation decreases. Therefore, it is also expected that larger safety benefits would be obtained if a truck lane restriction strategy was implemented on freeway segments that experience larger speed variations under adverse weather conditions, which also include downhill grades, tunnels, and bridges. However, the proposed recommendations are based on findings on risk factors identified at a 10 percent significance level. It should therefore be noted that more extensive analyses with larger data sets need to be conducted to draw more objective and generalized recommendations. Understanding characteristics associated with crash occurrence is crucial for developing effective countermeasures for preventing crashes and alleviating crash severity. Among various crash types, crashes involving trucks have a greater potential for fatalities. The purpose of this study was to analyze prevailing traffic conditions before crash occurrences using weather conditions to derive useful insights into truck traffic operations on freeways for safety enhancement. Truck crashes occurring on Korean freeways during a recent 3-year period, from 2008 to 2010, were analyzed. Traffic surveillance data and weather data were also used to identify prevailing traffic conditions and weather conditions that can potentially affect truck safety. Several statistical tests including ANOVA, post hoc, and t-tests were conducted to uncover statistically significant evidence to associate truck traffic safety and prevailing traffic conditions. In addition, a BLR modeling technique was applied to identify factors affecting truck crash severity. Some recommendations for safer truck traffic operations were presented based on the results obtained. It is expected that the outcomes of this study could be effectively utilized to support the development of various traffic operation strategies and policies for truck traffic safety. Although this study presents useful insights, further studies need to be conducted to obtain more reliable and valuable findings. First, more effective traffic variables representing prevailing traffic conditions should be derived. In addition to variables obtained from the point measurements used in this study, an analysis of section measurements, such as section speed, should be conducted because section measurements more effectively represent spatiotemporal variation in traffic conditions. Second, the use of road weather information systems data would allow for better opportunities to account for weather conditions in analyzing crash data, rather than using neighboring weather station data. To accomplish this data collection, road weather information systems must be widely
79 deployed. Third, studies that take crash types into consideration in analyzing crash severity with risk factors should be conducted because crash severity is likely to be associated with crash types, such as single-vehicle or vehicle-to-vehicle crashes. Lastly, applying more sophisticated modeling techniques to a larger crash data set is an important future research subject.
Acknowledgment This work was supported by the National Research Foundation of Korea grant funded by the Korea government (MEST) (NRF-2010-0029449).
References Abdel-Aty M, Abdelwahab H. Modeling rear-end collisions including the role of driver’s visibility and light truck vehicles using a nested logit structure. Accid Anal Prev. 2004;36:447–456. ¨ Cantor DE, Corsi TM, Grimm CM, Ozpolat K. A driver focused truck crash prediction model. Transp Res Part E: Logistics Transp Rev. 2010;46:683–692. Chang L, Mannering F. Analysis of injury severity and vehicle occupancy in truck- and non-truck-involved accidents. Accid Anal Prev. 1999;31:579–592. Chen F, Chen S. Injury severities of truck drivers in singleand multi-vehicle accidents on rural highways. Accid Anal Prev. 2011;43:1677–1688. Duncan CS, Khattak AJ, Council FM. Applying the ordered probit model to injury severity in truck–passenger car rear-end collisions. Transp Res Rec. 1998;1635:63–71. El-Tantawy S, Djavadian S, Roorda MJ, Abdulhai B. Safety evaluation of truck lane restriction strategies using microsimulation modeling. Transp Res Rec. 2009;2099:123–131. Evans L, Frick MC. Mass ratio and relative driver fatality risk in twovehicle crashes. Accid Anal Prev. 1993;25:213–224. Fontaine MD. Effect of truck lane restrictions on four-lane freeways in mountainous areas. Transp Res Rec. 2008;2078:135–142. Fontaine DM, Bhamidipati CS, Dougald LE. Safety impact of truck lane restrictions on multilane freeways. Transp Res Rec. 2009;2096:25–32. Gabler HC, Hollowell WT. The crash compatibility of cars and light trucks. Crash Prev Inj Contr. 2000;2:19–31. Gander PH, Marshall NS, James I, Quesne LL. Investigating driver fatigue in truck crashes: trial of a systematic methodology. Transp Res Part F: Traffic Psychol Behav. 2006;1:65–76. Golob T, Recker WW, Leonard LD. An analysis of the severity and incident duration of truck-involved freeway accidents. Accid Anal Prev. 1987;19:375–395. H¨akk¨anen H, Summala H. Fatal traffic accidents among trailer truck drivers and accident causes as viewed by other truck drivers. Accid Anal Prev. 2001;33:187–196. Hallmark SL, Hsu YY, Maze T, McDonald T, Fitzsimmons E. Investigating Factors Contributing to Large Truck Lane Departure Crashes Using the Federal Motor Carrier Safety Administration’s Large Truck Crash Causation Study (LTCCS) Database. Ames, IA: Iowa State University; 2009. Hickman JS, Hanowski RJ. An assessment of commercial motor vehicle driver distraction using naturalistic driving data. Traffic Inj Prev. 2012;13:612–619. Hosmer DW, Lemeshow S. Applied Logistic Regression. 2nd ed. Hoboken, NJ: Wiley; 2000. Keay K, Simmonds I. The association of rainfall and other weather variables with road traffic volume in Melbourne, Australia. Accid Anal Prev. 2005;37:109–124.
Downloaded by [The Aga Khan University] at 08:29 09 October 2014
80 Khattak AJ, Schneider RJ, Targa F. Risk factors in large truck rollovers and injury severity: analysis of single-vehicle collisions. Paper presented at: Transportation Research Board 82nd Annual Meeting; January 12–16, 2003; Washington, DC. Khorashadi A, Niemeier D, Shankar V, Mannering F. Differences in rural and urban driver-injury severities in accidents involving largetrucks: an exploratory analysis. Accid Anal Prev. 2005;37:910–921. Knapp KK, Smithson LD. Winter storm event volume impact analysis using multiple-source archived monitoring data. Transp Res Rec. 2000;1700:10–16. Korkut M, Ishak S, Wolshon B. Freeway truck lane restriction and differential speed limits: crash analysis and traffic characteristics. Transp Res Rec. 2010;2194:11–20. Lemp JD, Kockelmanb KM, Unnikrishnanc A. Analysis of large truck crash severity using heteroskedastic ordered probit models. Accid Anal Prev. 2011;43:370–380. National Institute for Occupational Safety and Health. Truck Driver Occupational Safety and Health. 2003 Conference Report and Selective Literature Review. Washington, DC: Centers for Disease Control and Prevention, Department of Health and Human Services; 2007.
Choi et al. Shi L, Cheng Y, Jin J, Ran B, Chen X. Effects of rainfall and environmental factors on traffic flow characteristics on urban freeway. Paper presented at: Transportation Research Board 90th Annual Meeting; January 23–27, 2011; Washington, DC. Traffic Accident Analysis System. The Korea Road Traffic Authority. Available at: http://taas.koroad.or.kr/Eng/indexMain.jsp/. Accessed September 17, 2013. Ulfarsson G, Mannering F. Differences in male and female injury severities in sport-utility vehicle, minivan, pickup and passenger car accidents. Accid Anal Prev. 2004;36:135–147. US Department of Transportation. Report to Congress on the Large Truck Crash Causation Study. Washington, DC: Federal Motor Carrier Safety Administration, US Department of Transportation; 2005. Young RK, Liesman J. Estimating the relationship between measured wind speed and overturning truck crashes using a binary logit model. Accid Anal Prev. 2007;39:574–580. Zhu X, Srinivasan S. A comprehensive analysis of factors influencing the injury severity of large-truck crashes. Accid Anal Prev. 2011a;43:49–57. Zhu X, Srinivasan S. Modeling occupant-level injury severity: an application to large-truck crashes. Accid Anal Prev. 2011b;43:1427–1437.
Traffic Injury Prevention Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/gcpi20
Risk Factors Related to Fatal Truck Crashes on Korean Freeways a
a
Saerona Choi , Cheol Oh & Mijeong Kim
a
a
Department of Transportation and Logistics Engineering , Hanyang University , Ansan-si , Korea Accepted author version posted online: 28 Feb 2013.Published online: 26 Nov 2013.
Click for updates To cite this article: Saerona Choi , Cheol Oh & Mijeong Kim (2014) Risk Factors Related to Fatal Truck Crashes on Korean Freeways, Traffic Injury Prevention, 15:1, 73-80, DOI: 10.1080/15389588.2013.778989 To link to this article: http://dx.doi.org/10.1080/15389588.2013.778989
PLEASE SCROLL DOWN FOR ARTICLE Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and should be independently verified with primary sources of information. Taylor and Francis shall not be liable for any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of the Content. This article may be used for research, teaching, and private study purposes. Any substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http:// www.tandfonline.com/page/terms-and-conditions
Traffic Injury Prevention (2014) 15, 73–80 C Taylor & Francis Group, LLC Copyright ISSN: 1538-9588 print / 1538-957X online DOI: 10.1080/15389588.2013.778989
Risk Factors Related to Fatal Truck Crashes on Korean Freeways SAERONA CHOI, CHEOL OH, and MIJEONG KIM Department of Transportation and Logistics Engineering, Hanyang University, Ansan-si, Korea
Downloaded by [The Aga Khan University] at 08:29 09 October 2014
Received 14 November 2012, Accepted 18 February 2013
Objectives: The major purpose of this study was to identify risk factors affecting truck crashes on freeways and propose recommendations for safer truck traffic operations. Methods: Truck crashes were analyzed to identify how truck traffic safety is related to prevailing traffic and weather conditions. Prevailing traffic conditions were characterized by central tendencies and the spatiotemporal variation of traffic parameters collected from freeway traffic surveillance systems. A total of 377 truck crashes occurring on Korean freeways in a recent 3-year period, 2008–2010, were analyzed together with corresponding prevailing traffic conditions and weather conditions. Several statistical tests were conducted to understand the characteristics of prevailing traffic conditions before crash occurrence based on different weather conditions. In addition, a binary logistic regression technique was applied to identify causal factors affecting truck crash severity under normal and adverse weather conditions. Results: Major findings from the analyses are discussed with truck operations strategies including speed enforcement, variable speed limit, and truck lane restriction from the safety enhancement point of view. Speed-related variables representing prevailing traffic conditions before crash occurrences were found to be the most statistically significant factors affecting truck crash severity, compared to volume-related variables such as the volume-to-capacity ratio (v/c). It is inferred that speed management is an effective tool for safer truck traffic operations on freeways. The major findings can be further discussed to derive valuable insights into truck traffic operations based on different weather conditions, such as normal and adverse. Conclusions: Some recommendations for safer truck traffic operations were presented based on the results obtained. The outcomes of this study could be effectively utilized to support the development of various traffic operations strategies and policies for truck traffic safety. Supplemental materials are available for this article. Go to the publisher’s online edition of Traffic Injury Prevention to view the supplemental file. Keywords: traffic safety, truck crashes, risk factors, truck traffic operations, injury severity, environmental conditions
Introduction Interest in truck traffic has increased largely due to greater safety concerns regarding truck-related crashes, in addition to the substantial economic impact of truck traffic on freight logistics. Recent Korean crash statistics (Traffic Accident Analysis System) revealed that the total number of fatalities caused by truck-involved crashes was 1266 in 2010, which makes up approximately 24 percent of total traffic fatalities. Safety issues regarding trucks can be described by several characteristics. First, the truck mass is generally greater than that of other vehicle types; thus, the resulting crash outcomes of truck crashes are more severe (Evans and Frick 1993). Sec-
Address correspondence to Saerona Choi, Department of Transportation and Logistics Engineering, Hanyang University, Sa 3-dong, Sangrok-gu, Gyeonggi-do, Ansan-si, Korea. E-mail: [email protected]
ond, when a truck is the leading vehicle in a car-following situation, the following smaller vehicle will struggle with ensuring a larger safety distance because the follower is not able to readily identify unexpected upcoming hazards. This situation is highly associated with an increased potential of rear-end collisions (Abdel-Aty and Abdelwahab 2004). Third, commercial truck drivers travel longer distances compared to the general public and tend to have increased nighttime driving times to avoid traffic congestion. Due to these circumstances, truck drivers are exposed to more hazardous situations caused by fatigued driving (Cantor et al. 2010). How to prevent truck-related crashes and reduce injury severity to minimize economic loss is an important consideration. One of the promising approaches to this issue is to devise effective truck traffic operations strategies that not only enhance safety but also alleviate congestion. The purpose of this study is to derive useful insights from analyses of truck crashes that can be further used for effective truck traffic operations for safety enhancement.
Downloaded by [The Aga Khan University] at 08:29 09 October 2014
74 To determine a research opportunity in accordance with the study purpose, a literature review was conducted from the traffic engineering point of view that a better understanding of the characteristics of truck crashes is necessary for safer traffic operations. The relevant literature can be broadly categorized into 2 areas. One area concerns truck traffic operations and the other concerns truck crash analysis. One of the countermeasures to alleviate safety concerns for truck traffic is to develop effective truck traffic operation strategies such as lane operations for trucks and differential speed limits by vehicle type. Significant efforts have been made to separate trucks from other vehicle types to provide more homogeneous and stable traffic streams (El-Tantawy et al. 2009; Fontaine 2008; Fontaine et al. 2009). Various lane operation scenarios have been evaluated based on field tests and simulation evaluations. Another approach is to apply differential speed limits to take the performance of truck maneuvering into account (Korkut et al. 2010). In addition, various existing studies have been conducted to analyze truck-related crashes. Their major purposes are to identify the dominant factors affecting crash severity (Chang and Mannering 1999; Duncan et al. 1998; Golob et al. 1987; Khattak et al. 2003; Khorashadi et al. 2005; Lemp et al. 2011; Zhu and Srinivasan 2011a). Among them, Golob et al. (1987) analyzed incident duration as a risk factor associated with truck-related crashes. Truck driver behavior, such as cell phone use and human factor characteristics, has also been explored to understand the characteristics of truck crashes (Cantor et al. 2010; Gander et al. 2006; H¨akk¨anen and Summala 2001; Hickman and Hanowski 2012; Zhu and Srinivasan 2011b). Truck crashes by lanes (Hallmark et al. 2009) and overturning crashes were also analyzed to capture more dominant factors (Young and Liesman 2007). Crash compatibility between different vehicle types was investigated by Gabler and Hollowell (2000). A focus of this study was to derive useful variables from prevailing traffic conditions that can be associated with truck safety. An important characteristic of truck traffic is that the flexibility of determining travel schedules for truck drivers is significantly limited, unlike for passenger vehicles, because the travel schedule is highly associated with commercial values. As a result, previous studies (National Institute for Occupational Safety and Health 2007; U.S. Department of Transportation 2005) have reported that truck drivers are often exposed to worse driving conditions compared to other drivers, such as adverse weather conditions. It is also known that adverse weather conditions tend to deteriorate traffic performance (Keay and Simmonds 2005; Knapp and Smithson 2000; Shi et al. 2011). Therefore, both prevailing traffic conditions and adverse weather conditions should be taken into account in analyzing truck safety. Although there has been significant research in the domain of truck crash analysis, we are not aware of any study that has attempted to comprehensively relate prevailing traffic conditions and adverse weather conditions to truck safety. This research opportunity motivated us to uncover valuable insights into truck traffic operations in terms of safety based on analyzing traffic and adverse weather conditions together with crash data. The availability of speed and traffic volumes obtained from traffic surveillance systems allowed us to prepare for potential
Choi et al. variables representing traffic conditions, which were matched with crash data. This study used freeway crashes occurring during a recent 3-year period, 2008–2010, with corresponding traffic surveillance data and weather data. Several statistical tests were conducted to obtain statistically significant evidence to associate truck traffic safety and prevailing traffic conditions. In addition, a binary logistic regression (BLR) technique was applied to model the relationship between truck crash severity and potential contributing factors obtained from the above data sources. A set of independent variables representing traffic conditions was established and analyzed together with weather and geometric conditions data. The severity of each truck crash was used as a dependent variable. Based on the findings from these analyses, this study suggests some recommendations for safer truck traffic operations on freeways.
Data Descriptions This study used 3 main data sources: the Korean freeway crash database, traffic surveillance systems data, and weather conditions data. Data from all crashes involving trucks on Korean freeways in a recent 3-year period, 2008–2010, were extracted and analyzed together with the corresponding traffic and weather conditions data. Crash Data The Korean freeway crash database contains useful information on crashes. In addition to crash-related information such as time, location, crash type, and severity, relevant geometric conditions such as horizontal and vertical alignment were included in crash record files and used to prepare for a set of independent variables. The truck crash data were obtained from the Korean Expressway Corporation (KEC) and consisted of 377 truckinvolved crashes. Truck crash cases that occurred in the study area were sequentially reduced based on a set of selection criteria for more systematic and reliable analyses. The selection criteria include crash location, crash cause, weather conditions, driver characteristics, and consistency of weather information. First, crashes that occurred on the freeway mainline were selected. Crashes that occurred on ramps, at toll gates, and at rest areas were excluded from the data set to be analyzed. Second, crashes resulting from vehicle malfunctioning were excluded. Because this study focused on the interactions among driver, geometry, traffic, and weather conditions, crashes resulting from driver-related causes were only included in the data set. Third, crash cases with incomplete driver information—for example, driver age, gender, driving experience, etc.—were excluded from the data set. Lastly, crash case, which weather information from crash database presented by KEC and Korean weather agency was inconsistent, was also excluded from the data set. As a result, a total of 972 truck crashes were reduced to 377. A total of 77 people were severely injured, which includes fatalities, and 300 were slightly injured. Several freeways that
Fatal Truck Crashes on Korean Freeways
75
Table 1. Characteristics of truck crash data by weather conditions Normal
Downloaded by [The Aga Khan University] at 08:29 09 October 2014
Freeway line
Rain
Snow
Fatal
Nonfatal
Fatal
Nonfatal
Fatal
Nonfatal
Total
Namhae Youngdong Jungbu naeryuk
8 20 32
46 62 93
7 3 1
37 33 16
0 4 2
2 4 7
100 126 151
Total
60
201 261
11
86
6
13
377
97
19
have significant truck volumes were selected. Because our main focus was to address the impact of traffic and weather conditions on truck safety, crashes that did not have matched traffic and weather data were excluded. A total of 377 truck crashes were used in this study for the analysis, excluding 68 crash cases for which corresponding traffic and weather data were not available. The KEC crash database describes crash severity using 4 different categories: A, B, C, and D. Crashes including 3 or more fatalities fall into category A. Crashes including 1–2 fatalities are defined as category B. Crashes involving injuries fall into category C. Category D includes other minor crashes including pedestrian–vehicle crashes that occurred at rest areas. This study used crashes from categories A, B, and C that included fatalities and injuries. Because fatal crashes fall into categories A and B, this study classified injury severity into 2 groups: fatal (A and B) and nonfatal (C). Table 1 presents a summary of the crash data used in this study. Traffic and Weather Data To develop safety insights into truck traffic operations, the impact of trucks on traffic streams needs to be evaluated. From a disaggregate point of view, the interaction between trucks and neighboring vehicles would be obviously different from that of homogeneous vehicles. From an aggregate point of view, the prevailing traffic conditions, such as average speed and traffic volume, would vary by the proportion of trucks. Because analysis of interactions among vehicles including trucks requires maneuvering data for each vehicle, which are not readily available in practice, analysis of prevailing traffic conditions is of much interest. The second source used was freeway traffic surveillance systems. The KEC collects and maintains traffic data for both point measurements and section measurements. Point measurements include speed, volume, and occupancy, which are collected from detector stations in the field. The detector spacing is approximately 1 km. Travel times are also available for specific freeway segments, where automatic vehicle identification systems are implemented based on dedicated short-range communication. This study used point measurements to derive variables representing traffic conditions because automatic vehicle identification systems have not been sufficiently implemented to provide traffic data corresponding to the truck crashes to be analyzed in this study. Traffic data aggregated in 15-min intervals collected from neighboring detector stations upstream and downstream from the crash location were used in this study. More specifically, these data included speed and volume data for 15-min before
crash occurrences. The spatiotemporal characteristics of prevailing traffic conditions need to be taken into consideration in developing independent variables. In addition to the average values representing the central tendency, the standard deviation of speed and volume over a period longer 15-min and the difference between upstream measurements and downstream measurements were explored to establish a set of candidate independent variables. Because speed and traffic count data were sometimes missing, a proper imputation method needed to be applied to prepare for the traffic data. When missing data occurred, adjacent upstream or downstream detector data for the same period was used to impute missing points. When adjacent detector data were also missing, we excluded the corresponding crash data from the analysis because the reliability of traffic data could not be assured. A set of candidate traffic parameters to represent the prevailing traffic conditions that can be potentially associated with truck crashes on freeways was prepared for the analysis and is presented in Table 2. The weather data used in this study were rainfall intensity and snowfall intensity. Although the crash database includes weather-related information, that information is mainly focused on roadway surface conditions, such as dry, wet, and icy. We used more detailed information to consider the severity of the conditions that would have an impact on a driver’s visibility. The intensity data were obtained from the Korean weather agency’s archives. The nearest weather station was matched with each crash in archiving the intensity data. This study defines adverse weather conditions as those involving rainfall and snowfall, and normal weather conditions include clear and cloudy conditions where drivers are not restricted by visibility and slippery road surfaces.
Characteristics of Traffic Conditions Before Truck Crash Based on Weather Conditions To identify the overall characteristics of traffic conditions before a truck crash occurred on freeways, several statistical analyses were conducted. The focus of the analysis was on evaluating traffic conditions based upon weather conditions. Analysis of variance (ANOVA) and Scheffe’s post hoc test were conducted to investigate whether the proposed traffic variables representing the prevailing traffic conditions would be differentiated by weather conditions such as normal, rain, up dn and snow. The null hypotheses for 4 variables, v15 , v15 , Vcv , and speeding, were rejected, which implies that the traffic conditions before a crash occurrence were significantly different based on the weather conditions. Appendix A (see online supplement) presents the ANOVA results for the variables that exhibited a significant difference. It can be said that if the variance between groups was greater than the variance within groups, the difference between the groups was statistically significant. Useful findings were identified by the post hoc test. A up prominent distinction for averages of v15 under different weather conditions was observed. Speed reduction due to snowfall was much greater than for other conditions.
76
Choi et al.
Table 2. Variable descriptions Variables
Variable names
Injury severity Crash location Temporal characteristics
severity bri/tun day night week age grade up grade dn curve up v15 dn v15 up v/c15
Driver Roadway characteristics
Traffic conditions
dn v/c15
Downloaded by [The Aga Khan University] at 08:29 09 October 2014
d v15 Vcv d v/c15 V/Ccv speeding
Description
ANOVA
BLR
0: nonfatal, 1: fatal 0: mainline, 1: bridge or tunnel 0: daytime, 1: nighttime 0: weekend, 1: weekdays Continuous/countable variable 0: no uphill, 1: uphill grade 0: no downhill, 1: downhill grade 0: tangent, 1: curve 15-Min average speed before crash occurrence at upstream detection station 15-Min average speed before crash occurrence at downstream speed 15-Min volume-to-capacity ratio before crash occurrence at upstream detection station 15-Min volume-to-capacity ratio before crash occurrence at downstream detection station 15-Min average speed difference between upstream and downstream detection stations before crash occurrence Coefficient of variation of speeds collected from upstream and downstream detection stations 15-Min volume-to-capacity ratio difference between upstream and downstream detection stations before crash occurrence Coefficient of variation of volume-to-capacity ratio calculated from upstream and downstream detection stations up up up Difference between v15 and speed limit (when v15 exceeds speed limit) 0: v15 does not exceed speed limit
— — — — — — — — Central tendency
DV IV IV IV IV IV IV IV IV — IV —
Spatiotemporal variation
IV IV — —
Driving pattern
IV
DV = dependent variable, IV = independent variable.
In addition, snowfall led to higher speed variation based on the analysis of Vcv , representing the magnitude of speed variations in prevailing traffic conditions before crash occurrences. Therefore, an effort to reduce speed variations during snowfall would be a viable approach to prevent crashes. Another interesting finding was observed for the variable speeding. The results indicated that drivers were more likely to exceed speed limits during rainfall compared to normal or snowfall conditions. Thus, speed enforcement under rainy conditions would be more effective at preventing crashes than enforcement under normal or snowfall conditions. Appendix B presents the post hoc results.
Analysis of Factors Contributing to Truck Crash Severity A logistic regression modeling approach was applied to the binary classification of the level of crash severity. The relationship between the dependent variable, which was a nonmetric variable in particular, and one or more independent variables was modeled. For the BLR model, multivariate procedure, the dependent variable took the value 1 or 0. The BLR model predicts the probability that the dependent variable takes the value of 1. In this case, 1 indicates that a truck-involved crash would belong to the group including fatal crashes. The form of the logistic regression model in this study is Prob (yi = 1) =
exp[ f (Xk , βk )] . 1 + exp[ f (Xk , βk )]
Prob (yi = 1) is the probability that the response variable yi includes a fatal truck crash; xk is a vector of explanatory variable
representing factors affecting the severity of the truck crash; βk is a vector of parameters to be estimated. Probability values can be any value between 0 (nonfatal crash) and 1 (fatal crash), but the predicted value cannot be outside of this range. To define a relationship bounded by 0 and 1, logistic regression uses an assumed relationship between the independent and dependent variables that resembles an Sshaped curve. The odds ratio represents the probability of an event divided by the probability of the event not happening. Because this is a ratio, its values range from zero to infinity (Hosmer and Lemeshow 2000). Factors with larger odds ratio have a larger impact on injury severity. First, both rainfall and snowfall intensities were applied as independent variables that represent adverse weather conditions. However, we were not able to uncover meaningful results that would be useful for traffic operations. A feasible reason would be the limitation associated with our data set. In other words, the range of rain and snowfall intensity data was not sufficiently wide to cover actual situations in practice. Because ANOVA results showed that traffic conditions are different by weather conditions, we divided our data set into 2 groups: one that involves crashes that occurred under normal weather conditions and one that involves crashes that occurred under adverse weather conditions. A set of candidate variables that would be potentially associated with truck safety were prepared for the analysis. These factors included driver characteristics and geometric design elements in addition to traffic and weather conditions. Table 2 presents the variables that were used in this study for conducting the crash severity analysis to derive useful insights into truck traffic operations on freeways for safety enhancement. Some variables for traffic conditions were excluded from the analysis based on the correlation analysis.
Fatal Truck Crashes on Korean Freeways
77
Table 3. BLR model of truck crash severity under normal weather conditions 95% Confidence interval for odds ratio
Downloaded by [The Aga Khan University] at 08:29 09 October 2014
Result
day night week age grade up grade dn curve bri/tun up v15 speeding d v15 Vcv up v/c15 constant Log likelihood Number of observations LR χ 2 (12) Pseudo R2
Coefficient
SE
t
0.916 0.275 0.006 −0.006 −0.275 −0.074 0.346 −0.009 0.128 0.042 −10.179 0.791 −1.257
0.375 0.449 0.015 0.409 0.406 0.331 0.648 0.016 0.065 0.029 5.582 1.114 1.821
2.440 0.610 0.360 −0.010 −0.680 −0.220 0.530 −0.560 1.970 1.470 −1.820 0.710 −0.690
P > |t| .015 .540 .716 .989 .499 .824 .594 .578 .048 .141 .068 .478 .490 −132.993 261 15.580 0.055
Odds ratio
Lower
Upper
2.498 1.317 1.006 0.994 0.760 0.929 1.413 0.991 1.137 1.043 3.8∗10−5 2.205 —
1.198 0.547 0.976 0.446 0.343 0.486 0.396 0.960 1.001 0.986 6.73∗10−10 0.248 —
5.208 3.173 1.036 2.216 1.684 1.776 5.034 1.023 1.291 1.103 2.143 19.577 —
Analysis of Crash Severity Under Normal Weather Conditions
Analysis of Crash Severity Under Adverse Weather Conditions
The Hosmer and Lemeshow goodness-of-fit test provides a χ 2 value of 15.58. The χ 2 value is less than the target χ 2 (significance level = .05, df = 12) value of 21.00, as presented in Table 3. In addition, the P-value was less than 0.05. Therefore, the Hosmer and Lemeshow test indicated that the model did not have a significant lack of fit. Two independent variables were identified as significant risk factors affecting crash severity at a 5 percent significance level. In addition to day night, speeding, which represents the overall tendency of drivers’ speed selection behavior before a crash occurs, was found to be significant. The odds ratio for day night was 2.5975, which implies that when a truckinvolved crash occurs at nighttime, the probability of a fatal crash would be 2.5 times greater than that of a daytime crash. Regarding speeding, it was found that when the prevailing speed exceeded a posted speed limit, a 1 km/h speed increment increased the possibility of a fatal crash by approximately 14 percent. When a 10 percent significance level was used, which is a more generous evaluation criterion, Vcv was found to be meaningful. Vcv indicates the speed variation, which is a well-known surrogate measure for safety before a crash occurs. Interestingly, the sign of the coefficient for Vcv was observed to be negative. Vcv decreased when the standard deviation for speed decreased while the average speed increased. Therefore, this finding also supports the idea that a higher prevailing speed would lead to a more severe crash. In addition, d v15 was identified as the most prominent factor affecting the crash severity (P = .141) among the insignificant variables. This result may be because the speed difference between upstream and downstream was closely associated with rear-end collisions. In particular, when a smaller vehicle collides with a large truck, the crash severity would be expected to increase remarkably. However, an extensive analysis with a larger data set is necessary to draw more generalized and objective conclusions on this matter.
The Hosmer and Lemeshow goodness-of-fit test yielded a χ 2 value of 21.01, which is a slightly higher than that calculated for the normal weather conditions. The χ 2 value was almost the same as the target χ 2 (significance level = .05, df = 12) value of 21.00. Table 4 presents the BLR modeling results for truck crash severity under adverse weather conditions. up Three independent variables including bri/tun, v15 , and Vcv were identified as significant risk factors affecting crash severity at a 5 percent significance level. When a 10 percent significance level was used, speeding and grade dn were found to be meaningful. The positive sign of the coefficient for bri/tun implies that a truck crash occurring on a bridge or in a tunnel was more up likely to be fatal. The result for v15 indicated that higher speed increased fatalities. It was also found that a 1 km/h increase in prevailing speed led to an approximately 9 percent increase in the possibility of a fatal crash. Regarding grade dn, it can be said that the truck crash severity increased when a crash occurred on a downhill grade. Interestingly, 2 variables, Vcv and speeding, were found to produce opposite results. Unlike the result for normal weather conditions, as the speed variation increased, the crash severity increased. This finding is possibly the reason why rainfall or snowfall would cause driver’s behavior for speed selection to vary. However, it should be noted that this impact was dependent on the intensity of rain and snow. In other words, both the average speed and speed variation decreased in the case of heavy snow or rain because the driver’s speed selection was significantly restricted. We do not believe that our data set contained a high enough intensity level to address this issue. Therefore, a further study to identify a cutoff point for the intensity level resulting in different impacts of weather conditions on speed variation would be of interest. Regarding speeding, when the prevailing speed was less than the posted speed limit, the crash severity decreased. A possible reason for
78
Choi et al.
Table 4. BLR model of truck crash severity under adverse weather conditions 95% Confidence interval for odds ratio
Downloaded by [The Aga Khan University] at 08:29 09 October 2014
Result
day night week age grade up grade dn curve bri/tun up v15 speeding d v15 Vcv up v/c15 constant Log likelihood Number of observations LR χ 2 (12) Pseudo R2
P > |t|
Coefficient
SE
t
0.374 1.003 −0.010 1.229 1.695 0.537 3.476 0.088 −0.114 −0.047 20.806 0.326 −12.071
0.721 0.962 0.031 1.057 0.995 0.639 1.580 0.042 0.065 0.040 8.280 2.100 4.504
0.520 1.040 −0.320 1.160 1.700 0.840 2.200 2.060 −1.740 −1.180 2.510 0.160 −2.680
this finding is that adverse weather conditions reduce capacity, resulting in a speed decrease such that most crashes occur under low-speed conditions. Another possible reason is that most drivers are more likely to conform to speed limits under adverse weather conditions. Investigation of Feasibility for Separate Modeling by Weather Conditions This study applied the likelihood ratio test used in previous studies (Chen and Chen 2011; Ulfarsson and Mannering 2004) to evaluate the feasibility of establishing separate models for normal and adverse weather conditions. The following statistics were used to conduct this test: β βN βA LR = −2 LLTotal − LLNormal , − LLAdverse β
where LLTotal is the log-likelihood for the integrated model βN is the log-likelihood for the model using all data; LLNormal βA is the logusing normal weather conditions; and LLAdverse likelihood for the model using adverse weather conditions. As a result, a χ 2 value of 26.77 was obtained, which is greater than the threshold point 21.00 with df = 12 and P < .05, indicating that the difference between normal and adverse weather conditions was statistically significant. From this result, the approach for separate modeling is reasonable.
.604 .297 .748 .245 .089 .400 .028 .039 .081 .240 .012 .877 .007 −37.829 116 21.010 0.217
Odds ratio
Lower
Upper
1.454 2.725 0.990 3.419 5.444 1.711 32.334 1.092 0.892 0.954 1.09∗109 1.385 —
0.354 0.414 0.933 0.431 0.774 0.489 1.462 1.004 0.785 0.882 97.238 0.023 —
5.967 17.950 1.051 27.139 38.302 5.986 715.077 1.186 1.014 1.032 1.21∗1016 84.908 —
In addition, the result of the goodness-of-fit test indicated that Pearson’s P-values were .345 and .127 for normal and adverse weather conditions, which are both greater than .05. These results indicate that the null hypothesis could not be rejected at a 95 percent confidence interval, which means that both models for normal and adverse weather conditions had acceptable fitness. Table 5 summarizes the results of the log-likelihood and goodness-of-fit tests.
Recommendations for Safer Truck Traffic Operations Speed-related variables representing prevailing traffic conditions before crash occurrences were found to be the most statistically significant factors affecting truck crash severity, compared to volume-related variable such as the volume-tocapacity ratio (v/c). It is inferred that speed management is an effective tool for safer truck traffic operations on freeways. The major findings can be further discussed to derive valuable insights into truck traffic operations based on different weather conditions such as normal and adverse. First, the finding that truck crash severity was greater at nighttime indicates that an emphasis on speeding enforcement should be made at nighttime under normal weather conditions. Another important result was that speeding was found to be statistically significant. In addition, the improvement of lighting conditions and proper provision of dedicated truck rest facilities
Table 5. Results of model comparison and goodness-of-fit test
Goodness-of-fit test
Model comparison
Number of observation covariate patterns Pearson’s χ 2 Prob > χ 2 Log likelihood (T) Comparison Prob > χ 2
Normal weather conditions
Adverse weather conditions
261 260 255.480 0.345
116 116 119.530 0.127 −184.206 26.77 26.77 > 21.00
Downloaded by [The Aga Khan University] at 08:29 09 October 2014
Fatal Truck Crashes on Korean Freeways would also be effective because crashes occurring at nighttime are associated with poor visibility and drowsy driving. Second, a variable speed limit (VSL) strategy is a promising tool for preventing truck crashes under adverse weather conditions. In particular, the VSL would be more effective for freeway segments including downhill grades, tunnels, and bridges. In particular, because Vcv was a dominant variable affecting severity, the major focus for VSL should be on minimizing the speed variation to enhance truck traffic safety. Moreover, advance warning information systems that inform drivers of upcoming hazards can be applied together with VSL strategies. In addition to speed management strategies, another viable approach is lane usage. Because truck lane restriction can make the traffic stream more homogeneous by separating trucks from other vehicles, the speed variation decreases. Therefore, it is also expected that larger safety benefits would be obtained if a truck lane restriction strategy was implemented on freeway segments that experience larger speed variations under adverse weather conditions, which also include downhill grades, tunnels, and bridges. However, the proposed recommendations are based on findings on risk factors identified at a 10 percent significance level. It should therefore be noted that more extensive analyses with larger data sets need to be conducted to draw more objective and generalized recommendations. Understanding characteristics associated with crash occurrence is crucial for developing effective countermeasures for preventing crashes and alleviating crash severity. Among various crash types, crashes involving trucks have a greater potential for fatalities. The purpose of this study was to analyze prevailing traffic conditions before crash occurrences using weather conditions to derive useful insights into truck traffic operations on freeways for safety enhancement. Truck crashes occurring on Korean freeways during a recent 3-year period, from 2008 to 2010, were analyzed. Traffic surveillance data and weather data were also used to identify prevailing traffic conditions and weather conditions that can potentially affect truck safety. Several statistical tests including ANOVA, post hoc, and t-tests were conducted to uncover statistically significant evidence to associate truck traffic safety and prevailing traffic conditions. In addition, a BLR modeling technique was applied to identify factors affecting truck crash severity. Some recommendations for safer truck traffic operations were presented based on the results obtained. It is expected that the outcomes of this study could be effectively utilized to support the development of various traffic operation strategies and policies for truck traffic safety. Although this study presents useful insights, further studies need to be conducted to obtain more reliable and valuable findings. First, more effective traffic variables representing prevailing traffic conditions should be derived. In addition to variables obtained from the point measurements used in this study, an analysis of section measurements, such as section speed, should be conducted because section measurements more effectively represent spatiotemporal variation in traffic conditions. Second, the use of road weather information systems data would allow for better opportunities to account for weather conditions in analyzing crash data, rather than using neighboring weather station data. To accomplish this data collection, road weather information systems must be widely
79 deployed. Third, studies that take crash types into consideration in analyzing crash severity with risk factors should be conducted because crash severity is likely to be associated with crash types, such as single-vehicle or vehicle-to-vehicle crashes. Lastly, applying more sophisticated modeling techniques to a larger crash data set is an important future research subject.
Acknowledgment This work was supported by the National Research Foundation of Korea grant funded by the Korea government (MEST) (NRF-2010-0029449).
References Abdel-Aty M, Abdelwahab H. Modeling rear-end collisions including the role of driver’s visibility and light truck vehicles using a nested logit structure. Accid Anal Prev. 2004;36:447–456. ¨ Cantor DE, Corsi TM, Grimm CM, Ozpolat K. A driver focused truck crash prediction model. Transp Res Part E: Logistics Transp Rev. 2010;46:683–692. Chang L, Mannering F. Analysis of injury severity and vehicle occupancy in truck- and non-truck-involved accidents. Accid Anal Prev. 1999;31:579–592. Chen F, Chen S. Injury severities of truck drivers in singleand multi-vehicle accidents on rural highways. Accid Anal Prev. 2011;43:1677–1688. Duncan CS, Khattak AJ, Council FM. Applying the ordered probit model to injury severity in truck–passenger car rear-end collisions. Transp Res Rec. 1998;1635:63–71. El-Tantawy S, Djavadian S, Roorda MJ, Abdulhai B. Safety evaluation of truck lane restriction strategies using microsimulation modeling. Transp Res Rec. 2009;2099:123–131. Evans L, Frick MC. Mass ratio and relative driver fatality risk in twovehicle crashes. Accid Anal Prev. 1993;25:213–224. Fontaine MD. Effect of truck lane restrictions on four-lane freeways in mountainous areas. Transp Res Rec. 2008;2078:135–142. Fontaine DM, Bhamidipati CS, Dougald LE. Safety impact of truck lane restrictions on multilane freeways. Transp Res Rec. 2009;2096:25–32. Gabler HC, Hollowell WT. The crash compatibility of cars and light trucks. Crash Prev Inj Contr. 2000;2:19–31. Gander PH, Marshall NS, James I, Quesne LL. Investigating driver fatigue in truck crashes: trial of a systematic methodology. Transp Res Part F: Traffic Psychol Behav. 2006;1:65–76. Golob T, Recker WW, Leonard LD. An analysis of the severity and incident duration of truck-involved freeway accidents. Accid Anal Prev. 1987;19:375–395. H¨akk¨anen H, Summala H. Fatal traffic accidents among trailer truck drivers and accident causes as viewed by other truck drivers. Accid Anal Prev. 2001;33:187–196. Hallmark SL, Hsu YY, Maze T, McDonald T, Fitzsimmons E. Investigating Factors Contributing to Large Truck Lane Departure Crashes Using the Federal Motor Carrier Safety Administration’s Large Truck Crash Causation Study (LTCCS) Database. Ames, IA: Iowa State University; 2009. Hickman JS, Hanowski RJ. An assessment of commercial motor vehicle driver distraction using naturalistic driving data. Traffic Inj Prev. 2012;13:612–619. Hosmer DW, Lemeshow S. Applied Logistic Regression. 2nd ed. Hoboken, NJ: Wiley; 2000. Keay K, Simmonds I. The association of rainfall and other weather variables with road traffic volume in Melbourne, Australia. Accid Anal Prev. 2005;37:109–124.
Downloaded by [The Aga Khan University] at 08:29 09 October 2014
80 Khattak AJ, Schneider RJ, Targa F. Risk factors in large truck rollovers and injury severity: analysis of single-vehicle collisions. Paper presented at: Transportation Research Board 82nd Annual Meeting; January 12–16, 2003; Washington, DC. Khorashadi A, Niemeier D, Shankar V, Mannering F. Differences in rural and urban driver-injury severities in accidents involving largetrucks: an exploratory analysis. Accid Anal Prev. 2005;37:910–921. Knapp KK, Smithson LD. Winter storm event volume impact analysis using multiple-source archived monitoring data. Transp Res Rec. 2000;1700:10–16. Korkut M, Ishak S, Wolshon B. Freeway truck lane restriction and differential speed limits: crash analysis and traffic characteristics. Transp Res Rec. 2010;2194:11–20. Lemp JD, Kockelmanb KM, Unnikrishnanc A. Analysis of large truck crash severity using heteroskedastic ordered probit models. Accid Anal Prev. 2011;43:370–380. National Institute for Occupational Safety and Health. Truck Driver Occupational Safety and Health. 2003 Conference Report and Selective Literature Review. Washington, DC: Centers for Disease Control and Prevention, Department of Health and Human Services; 2007.
Choi et al. Shi L, Cheng Y, Jin J, Ran B, Chen X. Effects of rainfall and environmental factors on traffic flow characteristics on urban freeway. Paper presented at: Transportation Research Board 90th Annual Meeting; January 23–27, 2011; Washington, DC. Traffic Accident Analysis System. The Korea Road Traffic Authority. Available at: http://taas.koroad.or.kr/Eng/indexMain.jsp/. Accessed September 17, 2013. Ulfarsson G, Mannering F. Differences in male and female injury severities in sport-utility vehicle, minivan, pickup and passenger car accidents. Accid Anal Prev. 2004;36:135–147. US Department of Transportation. Report to Congress on the Large Truck Crash Causation Study. Washington, DC: Federal Motor Carrier Safety Administration, US Department of Transportation; 2005. Young RK, Liesman J. Estimating the relationship between measured wind speed and overturning truck crashes using a binary logit model. Accid Anal Prev. 2007;39:574–580. Zhu X, Srinivasan S. A comprehensive analysis of factors influencing the injury severity of large-truck crashes. Accid Anal Prev. 2011a;43:49–57. Zhu X, Srinivasan S. Modeling occupant-level injury severity: an application to large-truck crashes. Accid Anal Prev. 2011b;43:1427–1437.
