Professional Documents
Culture Documents
Abstract: Many studies suggest that more crashes occur due to mixed traffic flow at unsignalized inter-
sections. However, very little is known about the injury severity of these crashes. The objective of this
study is therefore to investigate how contributory factors affect crash injury severity at unsignalized in-
tersections. The dataset used for this analysis derived from police crash reports from Dec. 2006 to Apr.
2009 in Heilongjiang Province, China. An ordered probit model was developed to predict the probabili-
ty that the injury severity of a crash will be one of four levels: no injury, slight injury, severe injury ,
and fatal injury. The injury severity of a crash was evaluated in terms of the most severe injury sus-
tained by any person involved in the crash. Results from the present study showed that different factors
had varying effects on crash injury severity. Factors found to result in the increased probability of seri-
ous injuries include adverse weather, sideswiping with pedestrians on poor surface, the interaction of
rear-ends and the third-class highway, winter night without illumination, and the interaction between
traffic signs or markings and the third-class highway. Although there are some limitations in the current
study, this study provides more insights into crash injury severity at unsignalized intersections.
Key words: traffic crash; injury severity; ordered probit model; unsignalized intersection
?l994-2014 China Academic Journal Electronic Publishing House. All rights reserved. http://www.cnki.net
Journal of Traffic and Transportation Engineering( English Edition) 273
vestigating the crash frequencies (Poch and Manne- crash is more reliably and accurately recorded than the
ring 1996; Persaud et al. 2002; Retting et al. 2003; injury severities sustained by all occupants in the traffic
Schepers et al. 2011). However, very few researches police crash reports (Shankar et al. 1996; Chang and
have explored the crash injury severity at this kind of Mannering 1999). Therefore, this study takes each
intersection, especially in China. crash as the research subject unit. The crash injury se-
So far, to our knowledge, there is almost only one verity is defined as the highest injury severity of the
study conducted for examining crash injury severity at crash, which is classified into the following four cate-
unsignalized intersections (Haleem and Abdel-Aty gories (levels): no injury, slight injury, severe inju-
2010). The ordered probit, binary probit, and nested ry, and fatal injury according to the records from Chi-
logit models were employed to analyze crash injury nese police crash reports.
severity at three-legged and four-legged unsignalized Many statistical approaches have been employed to
intersections in six counties in the state of Florida examine the injury severity of the crash, such as
from 2003 to 2006. They concluded that several traf- ordered logit/probit model (Eluru et al. 2008; Hal-
fic and roadway covariates (such as traffic volume, eem and Abdel-Aty 2010) , binary Iogit/probit model
geometric factors, shoulder width, left tum move- (Haleem and Abdel-Aty 2010; Peek-Asa et al.
ment, and number of right and left tum lanes) had 2010), heteroskedastic ordered probit model (Lemp
different influences on crash injury severity. They al- et al. 2011), multinomial and nested logit models
so suggested that the probabilities of fatal injury for (Haleem and Abdel-Aty 2010 j Eluru 2013) , bivari-
at-fault drivers aged 15-19 and 20-24 were much low- ate ordered probit model (Yamamoto and Shankar
er than other age groups. 2004), bivariate generalized ordered probit model
Analysis of crash injury severity has been conduc- (Chiou et al. 2013), and mixed logit model (Kim et
ted based on different research subject units, which al. 2013). A more comprehensive review of models
include: taking each crash as the subject unit, the available for examining crash injury severity can be
crash injury severity is defined as the most severe in- found in the literature (Mannering and Bhat 2014).
jury in the crash (Haleem and Abdel-Aty 2010; Zhu The ordered probit model is not only much easier to
and Srinivasan 2011 a); taking each vehicle involved interpret than other models, but also accounts for the
in the crash as the subject unit, the injury severity of ordinal and categorical nature of the dependent varia-
each vehicle is also defined as the most severe injury ble (i. e. , crash injury severity). Furthermore, this
in such vehicle (Chang and Mannering 1999; Chang model also has considerable appeal as a result of being
and Wang 2006) ; taking each driver involved in the parsimonious in the number of independent variables
crash as the subject unit, focusing on the driver's in- (Zhu and Srinivasan 2011a). In view of these
jury severity (Kockelman and Kweon 2002; Abdel- vantages, the ordered probit model is applied for this
Aty 2003; Xie et al. 2012); taking the specific occu- analysis.
pant as the subject unit, such as the front and rear seat Thus, the main goal of this study is to further in-
occupant (Shimamura et al. 2005; Newgard 2008) , vestigate the effects on crash injury severity of envi-
studying on the injury severity of the specific occu- ronmental, roadway, and crash characteristics factors
pant; taking each occupant involved in the crash as (including factors not considered in the previous
the subject unit, focusing on the injury severity of study) at unsignalized intersections in Heilongjiang
each occupant (Yamamoto arid Shankar 2004; Province, China. The remainder of this study is or-
Christoforou et al. 2010; Eluru et al. 2010; Zhu and ganized as follows. Data sources and variables de-
Srinivasan 2011b) . scription are presented in Section 2. The ordered
It is more useful to describe the overall injury severi- probit model is briefly introduced in Section 3. The
ty of a crash by taking each occupant involved in the model results are presented in Section 4 and discussed
crash as the research subject unit (Zhu and Srinivasan in Section 5. Finally, Section 6 gives conclusions and
2011 b). However, in fact, the most severe injury of a presents the area of future study.
?l994-2014 China Academic Journal Electronic Publishing House. All rights reserved. http://www.cnki.net
274 Yulong Pei et al.
x" x" =1 if the crash took place under the visibility 50-100 m, 0 for otherwise 0.170 0.376
X7 = 1 if the crash took place under the visibility 100-200 m , 0 for otherwise 0.169 0.375
x" = 1 if the crash was an angular collision. 0 for otherwise 0.493 0.500
X10 =1 if the crash was a rear-end collision. 0 for otherwise 0.085 0.278
XII =1 if the involved vehicle sideswiped with one pedestrian, 0 for otherwise 0.067 0.251
X I2 X 12 = 1 if the intersection was controlled by traffic signs or markings. 0 for otherwise 0.386 0.487
X I4 =1 if the crash occurred on the second-class highway, 0 for otherwise O.123 0.328
XIS =1 if the crash occurred on the third-class highway. 0 for otherwise 0.070 0.254
X, 6 X 16 =1 if the crash happened on the urban road, 0 for otherwise 0.531 0.499
XI' =1 if the crash took place at night without illumination, 0 for otherwise 0.127 0.333
?l994-2014 China Academic Journal Electronic Publishing House. All rights reserved. http://www.cnki.net
Journal of Traffic and Transportation Engineering( English Edition) 275
Pj(j)[givenx jk = I] -PjU) [given X ik -0] (4) creases by 15.75%). Crashes in the urban settings
P j (j) [given X ik = 0] were less injurious (no injury increases by 90. 92% ,
where X jk is the kth indicator variable for the ith crash. slight injury increases by 15.87%, severe injury de-
Hence, there is a direct pseudo-elasticity for each creases by 15.79%, and fatal injury decreases by
observation (crash). The average direct pseudo-elas- 34.98%) than rural area. No significant differences
ticity is estimated and reported to represent the whole about the likelihoods of injury severity levels were
dataset. For example, -0. 91 average direct pseudo- found during daylight, whereas the interaction between
elasticity of a particular probability with respect to an daylight and adverse weather was found to significantly
indicator variable suggests that the probability decrea- reduce the odds of fatal injury and severe injury in cra-
ses by 91 % on average due to the variable switching shes. Although the factor of night without illumination
from 0 to 1. did not significantly influence the probability of each
injury severity level, there was an interaction effect of
4 Results
winter and night without illumination. This leaded to a
The estimation results for the ordered probit model higher probability of increased injury severity as com-
are displayed in Tab. 2. Significance was assessed pared to other seasons or light conditions.
when ex is O.05. The Chi square test result of likeli- Angular collisions were less injurious than other
?l994-2014 China Academic Journal Electronic Publishing House. All rights reserved. http://www.cnki.net
276 Yulong Pei et al.
crash types. Rear-end collisions occurring at unsignal- collisions and sideswiping. The joint effect of side-
ized intersections of the third-class highways were swiping with pedestrians and poor surface was associ-
more likely to produce fatal injuries (fatal injury in- ated with an increase of the likelihood of severe injury
creases by 240.29%) compared to head-on, angular and fatal injury.
-0.109 0.088
0.288 0.088
Note: S. E. denotes standard error; n. s. indicates no significant found; Of. stands for degree of freedom.
?1994-2014 China Academic Journal Electronic Publishing House. All rights reserved. http://www.cnki.net
Journal of Traffic and Transportation Engineering( English Edition) 277
?1994-2014 China Academic Journal Electronic Publishing House. All rights reserved. http://www.cnki.net
278 Yulong Pei et al.
?1994-2014 China Academic Journal Electronic Publishing House. All rights reserved. http://www.cnki.net
Journal of Traffic and Transportation Engineering( English Edition) 279
Chang, L. Y. , Wang, H. W. ,2006. Analysis of traffic injury severi- Pei, Y. , Ma, Y. , 2005. The relationship between condition and traf-
ty: an application of non-parametric classification tree techniques. fic accidents of urban road intersections in the cold region. Journal
Accident Analysis & Prevention, 38 (5): 1019-1027. of Harbin Institute of Technology, 37 (1): 130-133.
Chiou, Y. C. , Hwang, C. C. , Chang, C. C. , et aI. ,2013. Model- Persaud, B. , Lord, D. , Palmisano, J. , 2002. Calibration and trans-
ing two-vehicle crash severity by a bivariate generalized ordered ferability of accident prediction models for urban intersections.
probit approach. Accident Analysis & Prevention, 51: 175-184. Transportation Research Record, 1784: 57-64.
Christoforou, Z. , Cohen, S. , Karlaftis, M. G.• 2010. Vehicle occu- Poch, M. , Mannering, F. , 1996. Negative binomial analysis of inter-
pant injury severity on highways: an empirical investigation. Acci- section-accident frequency. Journal of Transportation Engineering,
dent Analysis & Prevention, 42(6): 1606-1620. 122(2): 105-113.
Eluru, N. , 2013. Evaluating alternate discrete choice frameworks for Quddus, M. A. , Noland, R. B. , Chin, H. C. , 2002. An analysis
modeling ordinal discrete variables. Accident Analysis & Preven- of motorcycle injury and vehicle damage severity using ordered
tion, 55: 1-11. probit models. Journal of Safety Research, 33 (4) : 445-462.
Eluru, N. , Bhat, C. R., Hensher, D. A., 2008. A mixed general- Retting, R. A. , Weinstein, H. B. , Solomon, M. G. , 2003. Analy-
ized ordered response model for examining pedestrian and bicyclist sis of motor-vehicle crashes at stop signs in four U. S. cities. Jour-
injury severity level in traffic crashes. Accident Analysis & Preven- nal of Safety Research, 34 ( 5 ) : 485-489.
tion, 40(3): 1033-1054. Schepers, J. P. , Kroeze, P. A. , Sweers, W. , et al. , 2011. Road
Eluru, N. , Paleti, R. , Pendyala, R. M. , et al. ,2010. Model- factors and bicycle-motor vehicle crashes at unsignalized priority
ing injury severity of multiple occupants of vehicles copula- intersections. Accident Analysis & Prevention, 43 (3) : 853-861.
based multivariate approach. Transportation Research Record, Shankar, V. , Mannering, F. , Barfield, W.• 1996. Statistical analysis
2165: 1-11. of accident severity on rural freeways. Accident Analysis & Pre-
Haleem, K. , Abdel-Aty, M. , 2010. Examining traffic crash injury se- vention, 28 (3) : 391-401.
verity at unsignalized intersections. Journal of Safety Research, 41 Shimamura, M., Yamazaki, M., Fujita, G., 2005. Methodtoevalu-
(4) : 347-357. ate the effect of safety belt use by rear seat passengers on the injury
Kim, J. K. , Ulfarsson, G. F. , Kim, S. , et al. ,2013. Driver-injury severity of front seat occupants. Accident Analysis & Prevention,
severity in single-vehicle crashes in California: a mixed logit anal- 37( 1): 5-17.
ysis of heterogeneity due to age and gender. Accident Analysis & Wang. X. , Kockelman, K. M. , 2005. Use of heteroscedastic ordered
Prevention, 50: 1073-1081. logit model to study severity of occupant injury distinguishing
Kockelman, K. M.• Kweon, Y. J., 2002. Driver injury severity: an effects of vehicle weight and type. Transportation Research Re-
application of ordered probit models. Accident Analysis & Preven- cord, 1908: 195-204.
tion, 34(3): 313-321. Wu, Z. , Yang, J. , Huang, L. , 2013. Study on the collision avoid-
Lemp, J. D. , Kockelman, K. M. , Unnikrishnan, A. ,2011. Analy- ance strategy at unsignalized intersection based on PreScan simula-
sis of large truck crash severity using heteroskedastic ordered probit tion. Procedia-Social and Behavioral Sciences, 96: 1315-1321.
models. Accident Analysis & Prevention, 43 ( 1) : 370-380. Xie, Y. , Zhao, K. , Huynh, N. , 2012. Analysis of driver injury se-
Mannering, F. L. , Bhat, C. R. ,2014. Analytic methods in accident verity in rural single-vehicle crashes. Accident Analysis & Preven-
research: methodological frontier and future directions. Analytic tion, 47: 36-44.
Methods in Accident Research, I: 1-22. Yamamoto, T. , Shankar, V. N., 2004. Bivariate ordered-response
Mouyid, I. , Hernandez, S. , 2012. Multi-vehicle collisions involving probit model of driver's and passenger's injury severities in colli-
large trucks on highways: an exploratory discrete outcome analy- sions with fixed objects. Accident Analysis & Prevention, 36 ( 5 ) :
sis. 53rd Annual Transportation Research Forum, Tampa. 869-876.
Newgard, C. D.• 2008. Defining the "older" crash victim: the rela- Zhu, X. , Srinivasan, S. , 2011a. A comprehensive analysis of factors
tionship between age and serious injury in motor vehicle crashes. influencing the injury severity of large-truck crashes. Accident
Accident Analysis & Prevention, 40 ( 4 ) : 1498-1505. Analysis & Prevention, 43 ( 1 ) : 49-57.
Peek-Asa , C. , Britton, C. , Young. T. , et aI. ,2010. Teenage driver Zhu, X. , Srinivasan, S. , 2011b. Modeling occupant-level injury se-
crash incidence and factors influencing crash injury by rurality. verity: an application to large-truck crashes. Accident Analysis &
Journal of Safety Research, 41 ( 6) : 487-492. Prevention, 43(4): 1427-1437.
?1994-2014 China Academic Journal Electronic Publishing House. All rights reserved. http://www.cnki.net