- Email: [email protected]
Analysis of driver casualty risk for different work zone types
Accident Analysis and Prevention 43 (2011) 1811–1817
Contents lists available at ScienceDirect
Accident Analysis and Prevention journal homepage: www.elsevier.com/locate/aap
Analysis of driver casualty risk for different work zone types Jinxian Weng, Qiang Meng ∗ Department of Civil & Environmental Engineering, National University of Singapore, Singapore 117576, Singapore
a r t i c l e
i n f o
Article history: Received 8 December 2010 Received in revised form 15 March 2011 Accepted 11 April 2011 Keywords: Work zone Driver Logistic Regression Casualty Risk
a b s t r a c t Using driver casualty data from the Fatality Analysis Report System, this study examines driver casualty risk and investigates the risk contributing factors in the construction, maintenance and utility work zones. The multiple t-tests results show that the driver casualty risk is statistically different depending on the work zone type. Moreover, construction work zones have the largest driver casualty risk, followed by maintenance and utility work zones. Three separate logistic regression models are developed to predict driver casualty risk for the three work zone types because of their unique features. Finally, the effects of risk factors on driver casualty risk for each work zone type are examined and compared. For all three work zone types, five significant risk factors including road alignment, truck involvement, most harmful event, vehicle age and notification time are associated with increased driver casualty risk while traffic control devices and restraint use are associated with reduced driver casualty risk. However, one finding is that three risk factors (light condition, gender and day of week) exhibit opposing effects on the driver casualty risk in different types of work zones. This may largely be due to different work zone features and driver behavior in different types of work zones. © 2011 Elsevier Ltd. All rights reserved.
1. Introduction The work zone safety problem has become a high-priority issue for traffic engineer professionals. The presence of a work zone increases traffic conflicts and can cause severe traffic accidents. As shown by many researchers (Bedard et al., 2002; Ullman et al., 2006; Meng et al., 2010), the occurrence likelihood of severe crashes in work zones is higher than that in non-work zones and severe crashes are likely to lead to driver casualties. Hereafter, a driver casualty in this study is referred to as a driver being injured or killed in a vehicle accident that occurred in a work zone. Work zones can be categorized into three types: (1) construction; (2) maintenance; and (3) utility. Fig. 1 gives an example of the three types of work zones. Construction work zones are longterm work zones, where work activities last more than three days (MUTCD, 2003). In construction work zones, a complete traffic control plan should be implemented to improve work zone safety. For example, channelizing devices must be used to warn and alert drivers traveling across the construction work zones. Given that in this context construction is defined as more than three days duration, then construction activities will always extend into the night. Therefore, retroreflective and illuminated devices should be deployed for this work zone type (McAvoy et al., 2007). Also, the speed limit must be posted in construction work zones in order to
∗ Corresponding author. Tel.: +65 65165494; fax: +65 67791635. E-mail address: [email protected] (Q. Meng). 0001-4575/$ – see front matter © 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.aap.2011.04.016
improve the workers’ safety (Brewer et al., 2005). Compared with construction work zones, maintenance work zones usually have short work durations, where work occupies more than one hour but less than three days (MUTCD, 2003). Since the work duration is shorter, simplified traffic control procedures may be implemented. In comparison with construction work zones, maintenance work zones usually undertake lower intensity of work activities. In other words, there are often fewer workers in maintenance work zones than that in construction work zones. Another difference between the two types of work zones is that very few maintenance work zones are provided with law enforcement support (Pigman et al., 2006). Utility work zones significantly differ from the typical construction and maintenance work zones. Utility work is often of a very short duration (less than one hour) and involves very small crew sizes (MUTCD, 2003). In utility work zones, workers and equipment may move along a road at very slow speeds. As workers often need to work on or near the roadways, utility work zones pose unique challenges to driver’s health and safety (Datta et al., 2008). These facts underscore an urgent need to separately assess the driver casualty risk and the effects of its risk factors in each work zone type. 1.1. Literature review A number of studies have been conducted on work zone safety analysis. Bedard et al. (2002) employed a multivariate logistic regression model and found that those drivers who are female,
1812
J. Weng, Q. Meng / Accident Analysis and Prevention 43 (2011) 1811–1817
Fig. 1. (a) construction work zone, (b) maintenance work zone and (c) utility work zone.
drinking and traveling at higher speed have a higher likelihood of being involved in a fatal accident. Hill (2003) investigated the effectiveness of traffic control devices (e.g., flagman, a stop/go signal) on the occurrence likelihood of fatal work zone crashes. Srinivasan et al. (2007) modeled the location of work zone crashes as a multinomial logit function of the lengths of different work zone segments, traffic volume, weather, and other exogenous factors. Harb et al. (2008) explored work zone freeway crash characteristics by using multiple and conditional logistic regression techniques. According to the model results, roadway geometry, weather condition, age, gender, lighting condition, residence code, and driving under the influence of alcohol and/or drugs are the most significant risk factors associated with the work zone crashes. Based on Kansas highway work zone crash data for the years 2003–2004, Li and Bai (2008) developed the crash-severity-index models to measure the work zone crash risk. See et al. (2009) proposed a binary logistic regression model to examine the crash severity for two-lane closure strategies including the Iowa weave strategy and the conventional right-lane closure strategy. However, these existing work zone studies use the aggregated traffic accident data across all work zone types, which may give rise to biased or inaccurate results. For example, Kumar et al. (2008) found that there are no significant differences between weekdays and weekends in casualty risk. Li and Bai (2008) also pointed out that neither the day of week nor the gender has significant effects on the crash severity. In addition, the existing studies implicitly assume that the marginal effect of a risk factor on the casualty risk remains the same for all work zone types. In reality, the marginal effects of each risk factor vary with the work zone type because of their unique configurations. Theoretically, slightly different policies and guidelines should be developed for each type to reduce the casualty risk. Hence, there is a critical need to separately analyze the casualty risk for each type of work zone. 1.2. Objectives and contributions The objective of this study is to analyze the driver casualty risk and examine the effects of its risk contributing factors for the construction, maintenance and utility work zone types, respectively. To achieve this objective, the binary logistic regression technique is first employed to develop a driver casualty risk model for each work zone type. The marginal effect of each risk factor is finally examined and compared across the different types of work zones. The contributions of this study are twofold. First, by considering the differences in the features of the three work zone types, this study makes an initial attempt to separately analyze the driver casualty risk in each work zone type. Second, this study could
provide adequate supports that different policies and guidelines should be put forward for the mitigation of casualty risk in different types of work zones. 2. Logistic regression technique Many non-parametric models, such as classification and regression tree models (Kuhnert et al., 2000; Chang and Wang, 2006) and artificial neural network models (Mussone et al., 1999; AbdelAty and Abdelwahab, 2004), have been employed for traffic safety analysis. Although the non-parametric models can provide a high level of prediction accuracy, they have a weak interpretation of classification results. Non-parametric analysis is also less useful in examining the marginal effects of risk factors, which can provide valuable information for traffic engineers to establish the priorities for risk mitigation. Logistic regression technique is a suitable statistical method for predicting the correlations between a set of independent variables and a discrete target variable. In addition, logistic regression, whose outcomes are of a discrete or categorical nature, can predict the probability of the event of interest and estimate the marginal effect of each explanatory variable. For these reasons, a number of studies have already utilized this technique in traffic safety analysis (Hill, 2003; Li and Bai, 2008; See et al., 2009). For instance, Dissanayake and Lu (2002) developed a set of sequential binary logistic regression models to analyze the contributing factors and predict the crash severities of single-vehicle fixed-object work zone crashes involving young drivers. Since this study concentrates on the driver casualty risk in a work zone crash, the outcome of a driver involved in a work zone accident (the target variable) is assumed to be dichotomous: (i) injury (including death) and (ii) non-injury. A binary logistic regression model is thus employed to estimate the likelihood of each outcome. The target variable, y, can only take two values: y = 1 for the injury and y = 0 for the non-injury. The binary logistic regression model can be formulated as follows: ln
(x , . . . , x ) n 1 1 − (x1 , . . . , xn )
= ˛0 +
n
˛i xi
(1)
i=1
The probability that a driver is killed or injured can be calculated by
P(y = 1|x1 , . . . , xn ) = (x1 , . . . , xn ) =
e
n
˛0 +
1+e
˛x i=1 i i
n
˛0 +
˛x i=1 i i
(2)
J. Weng, Q. Meng / Accident Analysis and Prevention 43 (2011) 1811–1817
1813
Table 1 Variable descriptions. Utility work zone
Observation
Value
Weather
Clear Rain, Snow and Fog Other (blowing Sand)
1 2 3
91.6%a 7.8% 0.6%
83.2% 16.7% 0.1%
90.8% 9.2% 0.0%
Day of week
Weekend Weekday
1 2
24.2% 75.8%
20.0% 80.0%
13.7% 86.3%
Light condition
Daylight Dark with illumination Dark without illumination Dawn, dusk
1 2 3 4
64.1% 13.3% 20.0% 2.6%
68.5% 8.2% 20.8% 2.5%
87.0% 4.6% 8.4% 0.0%
Speed limit
<45 mph 45–60 mph >60 mph
1 2 3
11.3% 62.4% 26.3%
12.0% 50.4% 37.6%
15.3% 77.1% 7.6%
Traffic control devices
None Yes
0 1
33.9% 66.1%
37.8% 62.2%
38.9% 61.1%
Road surface condition
Dry Wet Other (snowy/sand)
1 2 3
87.7% 11.0% 1.3%
78.3% 16.1% 5.6%
85.5% 12.2% 2.3%
Road alignment
Straight Curve Actual number of lanes
1 2 –
88.3% 11.7% –
76.7% 23.3% –
80.9% 19.1% –
Age
Young (age < 30) Middle-age (30 ≤ age <60) Old (age ≥ 60)
1 2 3
26.3% 57.7% 16.0%
26.1% 56.2% 17.7%
23.7% 52.7% 23.6%
Gender
Male Female
1 2
75.1% 24.9%
77.5% 22.5%
77.1% 22.9%
Airbag availability
Not available Available
0 1
57.7% 42.3%
49.2% 50.8%
39.7% 60.3%
Restraint use
Non-used Used
0 1
21.9% 78.1%
25.1% 74.9%
29.8% 70.2%
Truck involvement
Not truck involved Truck involved
0 1
45.5% 54.5%
39.1% 60.9%
20.6% 79.4%
Vehicle age
0–5 year 5–10 year 10–15 year >15 year
1 2 3 4
4.7% 34.7% 33.5% 27.1%
4.6% 33.9% 31.3% 30.2%
1.5% 24.4% 50.4% 23.7%
Notification time Most harmful event
Actual notification time Pedestrian/vehicle ahead Overturn Other (fixed object, pole)
– 1 2 3
– 81.8% 7.3% 10.9%
– 76.5% 6.8% 16.7%
– 63.3% 9.2% 27.5%
Number of lanes
Number of datasets a
Construction work zone
Maintenance work zone
Variable
10,666
797
131
Percentage of drivers based upon the selected variables.
where xi , i = 1, . . ., n are the explanatory variables such as gender, weather and age; (x1 , . . ., xn ) is a conditional probability of the form P(y = 1|x1 , . . ., xn ), and ˛i is the coefficient of explanatory variable, which can directly determine the odds ratio (OR) of the explanatory variable (Harb et al., 2008). The OR describes the relative amount by which the odds of the outcome increase (i.e., OR greater than 1.0) or decrease (i.e., OR less than 1.0) when the value of the explanatory variable increases by 1.0 units. These parameters can be estimated using the maximum likelihood estimation technique. 3. Data 3.1. Database We first analyze the national database – Fatality Analysis Reporting System (FARS) managed by the National Highway Traffic Safety Administration (NHTSA). The work zone driver casualty data in FARS were obtained from the work zone crashes on public
roads within the 51 U.S. states between 2001 and 2006. The impact of traffic flow on driver casualty risk was not taken into account because the FARS database does not provide traffic flow information. According to this database, the number of driver casualties is not found to monotonically increase year by year though the annual traffic volume may increase from an empirical perspective. Therefore, there is inadequate evidence that the traffic flow is related to the driver casualty risk. The lack of consideration of traffic flow will not make the results less reliable. The FARS database provides the following information on each driver casualty for each work zone type: (i) environmental characteristics; (ii) road characteristics; (iii) driver characteristics; and (iv) crash information. Environmental characteristics include detailed information regarding the weather (clear, Rain/Fog/Snow, other), the day of week (weekend, weekday), the light condition (daylight; dark with illumination, dark without illumination, dawn/dusk), the speed limit (<45 mph, 45–60 mph, >60 mph) and the use of traffic control devices (none, yes). The road characteristics consist of the road surface condition (dry, wet, other), the road
1814
J. Weng, Q. Meng / Accident Analysis and Prevention 43 (2011) 1811–1817
Table 2 Multiple t-test results of driver casualty risk at different work zone types. Work zone type
Construction Maintenance Utility a
Sample size
10,666 797 131
Mean
t-test statistic (p-value)
0.722 0.691 0.672
Maintenance
Utility
2.47 (0.014)a
3.97 (<0.001) 1.18 (0.237)
p-value in parentheses.
alignment (straight, curve) and the number of available lanes. Driver characteristics include the driver’s age and gender. Crash information comprises the airbag availability (not available, available), the restraint use, truck involvement, vehicle age (0–5 year, 5–10 year, 10–15 year, >15 year), notification time and the most harmful event (pedestrian or vehicle ahead, overturn, other). 3.2. Data processing Table 1 presents 16 variables described above. Each variable is represented by a binary or an ordinal number. It should be noted that these 16 variables are selected based on the percentage of missing data (“Unknown”). The variables with more than 7.9% missing data (e.g., travel speed) are excluded. A preliminarily check of the data quality for the selected variables is performed before applying them for the analysis. For records with an “unknown” value for any of the sixteen selected variables, a value is generated based on the remaining records in the sample. For example, in approximately 24% of the records pertaining to construction work zones, the day of week has the value “Weekend”. Therefore, there is a 24% probability of replacing an unknown value for the day of week for a record from the construction work zone with the value “Weekend” and a 76% probability of replacing it with the “Weekday”. In the FARS database, injury severity is given as no injury, possible injury, non-incapacitating evident injury, incapacitating injury, fatality, injury-severity unknown and died prior to crash. Since this study focuses on the driver casualty risk in work zone crashes, the records where drivers died prior to the crash are excluded from the analysis. Therefore, a total of 11,594 records of drivers involved in work zone vehicle accidents between 2001 and 2006 have been collected. Of these records, 10,666 drivers are involved in construction work zone accidents; 797 drivers suffer different severity levels of
injuries in maintenance work zone accidents and 131 drivers are involved in utility work zone accidents. Due to the relatively small number of driver fatalities in utility work zones, this study assumes that there are only two outcomes of a driver involved in an accident: non-injury and injury. Thus, the possible injury, non-incapacitating evident injury, incapacitating injury, fatality and injury-severity unknown are aggregated into the injury classification. 4. Results and discussions 4.1. Descriptive analysis results Table 1 provides detailed summary on statistics of the selected variables for the three work zone types. Approximately 75% of the drivers involved in construction work zone accidents are male. Similar statistics are found for the maintenance and utility work zone accidents. The distribution of the use of traffic control devices is also similar across the three work zone types as are the data related to restraint use, weather, road surface condition, road alignment and age. However, other variables (e.g., light condition, airbag availability, most harmful event) vary substantially among the three work zone types. For example, a minority of the drivers involved in construction work zone accidents (42.3%) have airbags in their vehicles, whereas 60.3% of those involved in utility work zone accidents used airbags. In maintenance work zones, the proportion of drivers using the airbag approximates to that without using the airbag function. A multiple t-test is employed to investigate whether there is a statistical difference of the mean driver casualty risk between any two work zone types. Table 2 gives the multiple t-test results. The mean driver casualty risk in the construction work zone is 0.722, which is statistically larger than both 0.691 in the maintenance
Table 3 Statistic results of the logistic regression model for the construction work zone. Variable
Coefficient (ˇ)
Standard error
Wald Chi-square
p-Value
Intercept Weather Day of week Light condition Speed limit Traffic control devices Road surface condition Road alignment Number of lanes Age Gender Airbag availability Restraint use Truck involvement Vehicle age Notification time Most harmful event
−1.0189 0.0583 −0.0272 −0.1260 0.0842 −0.1565 0.1303 0.2865 0.0582 0.1561 0.6585 −0.4839 −1.9078 0.5888 0.1274 0.0044 0.2546
0.278 0.105 0.013 0.027 0.042 0.052 0.088 0.084 0.023 0.038 0.061 0.050 0.087 0.052 0.027 0.001 0.041
13.4 0.31 4.18 21.4 4.10 9.23 2.21 11.6 6.66 17.0 118 94.3 481 129 22.1 11.9 38.6
<0.01 0.58 0.04 <0.01 0.04 <0.01 0.14 <0.01 0.01 <0.01 <0.01 <0.01 <0.01 <0.01 <0.01 <0.01 <0.01
AIC SC -2log-likelihood
Intercept only 12,613.9 12,621.2 12,611.9
Intercept and covariates 11,108.2 11,231.9 11,074.2
J. Weng, Q. Meng / Accident Analysis and Prevention 43 (2011) 1811–1817
1815
Table 4 Statistic results of the logistic regression model for the maintenance work zone. Variable
Coefficient (ˇ)
Standard error
Wald Chi-square
p-Value
Intercept Weather Day of week Light condition Speed limit Traffic control devices Road surface condition Road alignment Number of lanes Age Gender Airbag availability Restraint use Truck involvement Vehicle age Notification time Most harmful event
−0.9012 −0.1037 0.0903 0.1563 −0.0168 −0.2305 0.1658 0.6112 0.0492 0.2177 −0.2738 −0.6668 −1.8642 0.3244 0.0888 0.0128 0.6149
0.463 0.257 0.043 0.082 0.114 0.135 0.184 0.164 0.061 0.096 0.149 0.131 0.198 0.153 0.050 0.004 0.098
3.79 0.16 4.47 3.62 0.02 2.92 0.82 13.9 0.66 5.15 3.38 25.9 89.1 4.51 3.15 11.1 39.3
0.05 0.69 0.04 0.05 0.88 0.08 0.37 <0.01 0.42 0.02 0.06 <0.01 <0.01 0.03 0.07 0.02 <0.01
AIC SC -2log-likelihood
Intercept only 987.1 991.8 985.1
work zone and 0.672 in the utility work zone. However, there is no statistical difference between the mean risks in the maintenance and utility work zones (at the significance level of 0.05). This may be due to the small sample sizes for these two work zone types. 4.2. Logistic regression model results for three different work zone types The logit procedure in the Statistical Analysis Software (SAS, 2008) is performed to calibrate a separate logistic regression model for each work zone type. Table 3 gives the results for the construction work zone. It can be seen that the driver casualty risk in a construction work zone is significantly (at a significance level of 0.10) increased with the following variables: speed limit, road alignment, number of lanes, age, gender, truck involvement, vehicle age, notification time and most harmful event. The coefficients associated with the day of week, light condition, traffic control devices, airbag availability and restraint use are negative and statistically significant at a significance level of 0.10, indicating that these five variables are significantly associ-
Intercept and covariates 874.0 953.6 840.0
ated with the decreased driver casualty risk in the construction work zone. In a maintenance work zone, similarly, the variables that can significantly increase driver casualty risk include the road alignment, age, truck involvement, vehicle age, notification time and most harmful event, as shown in Table 4. The traffic control devices, restraint use and airbag availability are shown to reduce the casualty risk. The results for the utility work zone are shown in Table 5. Road surface condition, road alignment, number of lanes, age, vehicle age and notification time all increase the casualty risk. Among the 16 variables, only four variables including speed limit, age and airbag availability do not significantly influence the driver casualty risk in utility work zones. 4.3. Marginal effects of risk factors for the three work zone types The marginal effects of the risk factors on driver casualty risk across different work zone types are shown in Table 6. For clarity, the results are discussed according to the categories mentioned
Table 5 Statistic results of the logistic regression model for the utility work zone. Variable
Coefficient (ˇ)
Standard error
Wald Chi-square
p-Value
Intercept Weather Day of week Light condition Speed limit Traffic control devices Road surface condition Road alignment Number of lanes Age Gender Airbag availability Restraint use Truck involvement Vehicle age Notification time Most harmful event
−7.2366 −3.5173 0.1853 −0.7325 0.2535 −3.0772 1.7234 1.0353 1.0106 0.0741 0.5977 −0.4300 −1.7939 2.3032 0.2143 0.0543 0.2904
2.477 0.554 0.084 0.176 0.335 0.488 0.657 0.409 0.263 0.158 0.267 0.293 0.312 0.542 0.112 0.010 0.131
8.54 40.3 4.91 17.4 0.57 39.7 6.88 6.42 14.8 0.22 5.00 2.16 33.1 18.0 3.67 29.8 4.93
<0.01 <0.01 0.03 <0.01 0.45 <0.01 0.01 0.011 <0.01 0.64 0.03 0.14 <0.01 <0.01 0.05 <0.01 0.03
AIC SC -2log-likelihood
Intercept only 167.8 170.7 165.8
Intercept and covariates 150.9 199.8 116.9
1816
J. Weng, Q. Meng / Accident Analysis and Prevention 43 (2011) 1811–1817
Table 6 Relative effects of risk factors on driver casualty risk at different work zone types. Variable
Weather Day of week Light condition Speed limit Traffic control devices Road surface condition Road alignment Number of lanes Age Gender Airbag availability Restraint use Truck involvement Vehicle age Notification time Most harmful event a b
Construction work zone
Maintenance work zone
Utility work zone
ORa
Lower CI
Upper CI
Significantb
OR
Lower CI
Upper CI
Significant
OR
Lower CI
Upper CI
Significant
1.06 0.97 0.88 1.09 0.86 1.14 1.33 1.06 1.17 1.93 0.62 0.15 1.80 1.14 1.004 1.29
0.86 0.95 0.84 1.00 0.77 0.96 1.13 1.01 1.09 1.72 0.56 0.13 1.63 1.08 1.002 1.19
1.30 1.00 0.93 1.18 0.95 1.35 1.57 1.11 1.26 2.18 0.68 0.18 2.00 1.20 1.007 1.40
N Y Y Y Y N Y Y Y Y Y Y Y Y Y Y
0.90 1.09 1.17 0.98 0.79 1.18 1.84 1.05 1.24 0.76 0.51 0.16 1.38 1.09 1.01 1.85
0.44 1.01 1.00 0.79 0.61 0.82 1.34 0.93 1.03 0.57 0.40 0.11 1.03 1.00 1.005 1.53
1.83 1.19 1.37 1.23 1.00 1.69 2.54 1.18 1.50 1.00 0.66 0.23 1.87 1.26 1.021 2.24
N Y Y N Y N Y N Y Y Y Y Y Y Y Y
0.03 1.20 0.48 1.29 0.05 5.60 2.82 2.75 1.08 1.82 0.65 0.17 10.0 1.24 1.05 1.34
0.01 1.02 0.34 0.67 0.02 2.10 1.26 1.64 0.79 1.08 0.37 0.10 3.46 1.00 1.035 1.04
0.09 1.42 0.68 2.48 0.12 14.9 6.27 4.60 1.47 3.07 1.16 0.31 28.9 1.69 1.077 1.73
Y Y Y N Y Y Y Y N Y N Y Y Y Y Y
Odds ratio (OR) is calculated by changing one unit of one variable at a time while controlling for the other variables, where OR = exp(). N, not significant at p ≤ 0.10 for the risk factor; Y, significant at p ≤ 0.10 for the risk factor.
earlier. Here, the marginal effect for a given risk factor can be represented by its odds ratio (OR) defined as the relative amount by which the odds of the injury increase or decrease when the risk factor’s value increases by 1.0 units. 4.3.1. Environmental characteristics The OR associated with the day of week is 0.97 for the construction work zones, 1.09 for the maintenance work zones and 1.20 for the utility work zones. This result implies that a driver incurs a higher casualty risk in the maintenance and utility work zones during the weekday than at the weekend. However, she/he has a 3% lower casualty risk on a weekday than at the weekend in construction work zones. The fact that the higher traffic may slow down drivers during weekdays could account in part for the 3% lower risk. This finding could also explain why Kumar et al. (2008) found no significant differences of the casualty risk between weekdays and weekends if the crash data they used were aggregated across different work zone types. In construction work zones, drivers are less likely to be injured in the dark with illumination condition (OR = 0.88) than in the daylight condition. However, in the maintenance work zones, drivers are 17% more likely to be injured under the dark with illumination condition (OR = 1.17). The opposing effects may be because retroreflective signs are always used for the night construction work so that drivers will be more vigilant driver (McAvoy et al., 2007). However, retroreflective signs are less used in the night maintenance work zone (MUTCD, 2003). As shown by many researchers (Lu et al., 2006; Li and Bai, 2008), the speed limit in the construction work zones (OR = 1.09) is highly correlated with driver casualty risk. A high speed limit increases the driver casualty risk in the construction work zone accidents, while it is not statistically significant in the maintenance and utility work zones. One possible reason may be that the speed limit works collectively with other factors (e.g., work intensity). For example, the maintenance work zone undertaking the very low intensity of work is often posed no speed limit while the high work intensity usually poses low speed limit. Therefore, further investigation of the relationship between the work intensity and the speed limit should be carried out for maintenance work zones. As expected, the presence of traffic control devices significantly reduces driver casualty risk for all the three work zone types though their marginal effects are not the same. More specifically, drivers have a 14% lower casualty risk when traffic control devices are used in the construction work zones, and a 21% lower casualty risk when traffic control devices are used in the maintenance work zones.
4.3.2. Road characteristics The road surface condition is found to be associated with increased driver casualty risk in utility work zones while it is not statistically associated with increased driver casualty risk in the other two types of work zones. Although the road alignment is a contributing factor, its marginal effect is not the same across the three work zone types. On a curved road, drivers have a 33% higher casualty risk in construction work zones, an 84% higher risk in maintenance work zones and a 182% higher risk in utility work zones. Similarly, the number of lanes is also associated with increased casualty risk in the construction and utility work zones. The increase of casualty risk associated with increase in the number of lanes may possibly be explained by the increase in the number of lane change conflicts, which may result in severe vehicle crashes (Kononov et al., 2008). However, it has no substantial impact on the casualty risk in maintenance work zones. 4.3.3. Driver characteristics Middle-aged drivers are 1.17 times more likely than young drivers to be injured or killed in construction work zones and this figure is 1.24 times in maintenance work zones. Although the OR associated with the age in utility work zones is larger than 1.0, it is not statistically associated with increased driver casualty risk at the significance level of 0.10. In addition, female drivers have a 93% higher casualty risk than male drivers in the construction work zones (CI = 1.72–2.18), and an 82% higher risk than male drivers in the utility work zones (CI = 1.08–3.07). The increased casualty risk associated with the female driver in construction and utility work zones results from two factors: (1) increased risk-taking in younger female drivers and (2) increased exposure in older female drivers (Kostyniuk et al., 1996). 4.3.4. Crash characteristics Consistent with previous studies, airbag availability is associated with a lower driver casualty risk in both construction and maintenance work zones (Evans, 1991). Compared with the drivers in the vehicles without airbags, the drivers in the vehicles equipped with airbags are less prone (=38%) to suffering casualties in the construction work zones, and a 49% less likely in the maintenance work zones. Accidents involving trucks are found to dramatically increase the driver casualty risk, especially in utility work zones. More specifically, drivers caught up in utility work zone accidents involving a truck have a 10.0 times greater casualty risk. However, the marginal effect of truck involvement in the construction work
J. Weng, Q. Meng / Accident Analysis and Prevention 43 (2011) 1811–1817
zones (OR = 1.80) is less than that in utility work zones but larger than that in maintenance work zones (OR = 1.38). Consistent with our expectations, the vehicle age also affects driver casualty risk. In construction work zones, a driver in a vehicle that is 5–10 years old has a 14% higher casualty risk than the driver in a newer vehicle (e.g., ≤5 years). Also, the longer notification time is associated with the higher driver casualty risk in all three work zone types (OR = 1.004 for the construction, OR = 1.010 for the maintenance and OR = 1.004 for the utility). Consistent with previous studies (Levine et al., 1999; Adadzi et al., 2008), our results also indicate that the type of collision (most harmful event) is associated with a decreased driver casualty risk in all the three work zone types. 5. Conclusions Taking into account the differences in work zone features, this study has separately analyzed the driver casualty risk and investigated the risk factors for the three work zone types: construction, maintenance and utility. Driver casualty data from 2001 to 2006 for each work zone type are taken from the FARS. The univariate analysis implies that the distributions of some of the risk factors vary with the work zone type. The multiple t-tests further show that the construction work zones have the largest driver casualty risk, followed by the maintenance and utility work zones. Based on these results, three separate logistic regression models are developed to predict driver casualty risk for the three work zone types. Finally, the marginal effects of each factor on the driver casualty risk in each work zone type are examined and compared. Seven factors (road alignment, truck involvement, most harmful event, traffic control devices, restraint use and vehicle age) are found to significantly impact driver casualty risk for all the three work zone types. Probably because of the small sample size for the utility work zones, age and airbag availability are not found to significantly influence the driver casualty risk in the utility work zones. The speed limit is not found to be a risk factor in maintenance and utility work zones probably as a result of the fact that imposed speed limits in these two work zone types may interact with work intensity. Hence, further investigation of the relationship between work intensity and speed limit should be carried out in the future. One finding from this study is that three factors (light condition, gender, and the day of week) exhibit heterogeneous effects on driver casualty risk in different types of work zones. Drivers in construction work zones tend to be more likely to suffer injuries or fatalities at the weekend while drivers in the maintenance and utility work zones are less likely to be injured or killed in the weekend accidents. These results provide supports for the argument that slightly different policies and guidelines should be developed for different types of work zones in order to reduce the casualty risks. For example, construction activities are encouraged to be carried out during the weekdays while the maintenance and utility works should be implemented at the weekends. Additional measures should be proposed to forbid risky driving behavior in construction and utility work zones.
1817
References Abdel-Aty, M., Abdelwahab, H., 2004. Predicting injury severity levels in traffic crashes: a modeling comparison. Journal of Transportation Engineering 130 (2), 204–210. Adadzi, K.D., Classen, S., Hall, A., Duncan, R.P., Garvan, C.W., 2008. Predictors of injury among younger and older adults in fatal motor crashes. Accident Analysis and Prevention 40, 1804–1810. Bedard, M., Guyatt, G., Stones, M., Hirdes, J., 2002. The independent contribution of driver, crash and vehicle characteristics to driver fatalities. Accident Analysis and Prevention 34 (6), 717–727. Brewer, M.A., Pesti, G., Schneider IV, W.H., 2005. Identification and testing of measures to improve work zone speed limit compliance. Report No. FHWA/TX06/0-4707-1, Texas Transportation Institute, Texas. Chang, L., Wang, H., 2006. Analysis of traffic injury severity: an application of nonparametric classification tree techniques. Accident Analysis and Prevention 34 (5), 1019–1027. Datta, T., Savolainen, Grillo, L., Schattler, K., 2008. Utility work zone traffic control guidelines. Technique Project # DTFH61-06-G-00006. Dissanayake, S., Lu, J., 2002. Analysis of severity of young driver crashes, sequential binary logistic regression modeling. Transportation Research Record 1784, 108–114. Evans, L., 1991. Airbag effectiveness in preventing fatalities predicted according to type of crash, driver age, and blood alcohol concentration. Accident Analysis and Prevention 23, 531–541. Harb, R., Radwan, E., Yan, X., Pande, A., Abdel-Aty, M., 2008. Freeway work-zone crash analysis and risk identification using multiple and conditional logistic regression. Journal of Transportation Engineering 134 (5), 203–214. Hill, R.W., 2003. Statistical analysis of fatal traffic accident data. Texas Tech University, Lubbock, Texas. Kononov, J., Bailey, B.A., Allery, B.K., 2008. Exploratory analysis of relationship between number of lanes and safety on urban freeways. In: Proceedings of the TRB 87th Annual Meeting, CDR-OM, National Research Council , Washington, DC. Kostyniuk, L., Molnar, L., Eby, D., 1996. Are women taking more risks while driving? In: Proceedings of the Women’s Travel Issues Second National Conference , Baltimore. Kumar, A., Lalwani, S., Agrawal, D., Rautji, R., Dogra, T.D., 2008. Fatal road traffic accidents and their relationship with head injuries: an epidemiological survey of five years. Indian Journal of Neurotrauma 5, 63–77. Kuhnert, P.M., Do, K., McClure, R., 2000. Combining non-parametric models with logistic regression: an application to motor vehicle injury data. Computational Statistics and Data Analysis 34 (3), 371–386. Levine, E., Bedard, M., Molloy, D.W., Basilevsky, A., 1999. Determinants of driver fatality risk in front impact fixed object collisions. Mature Medicine Canada 2, 239–242. Li, Y., Bai, Y., 2008. Development of crash-severity-index models for the measurement of work zone risk levels. Accident Analysis and Prevention 40 (5), 1724–1731. McAvoy, D.S., Schattler, K.L., Datta, T.K., 2007. Driving simulator validations for nighttime construction work zone devices. Transportation Research Record 2015, TRB, 55–63. Meng, Q., Weng, J., Qu, X., 2010. A probabilistic quantitative risk assessment model for the long-term work zone crashes. Accident Analysis and Prevention 42 (6), 1866–1877. Mussone, Li., Ferrari, A., Oneta, M., 1999. An analysis of urban collisions using an artificial intelligence model. Accident Analysis and Prevention 31 (6), 705–718. Pigman, J.G., Agent, K.R., Green, E.R., 2006. Evaluation of work zone safety operations and issues. Research Report KTC-06-08/SPR287-05-1F, University of Kentucky, (UK). SAS, 2008. SAS/STAT 9.2 User’s Guide. Cary, NC: SAS Publishing. See, C.F., Schrock, S.D., McClure, K., 2009. Crash analysis of work zone lane closures with left-hand merge and downstream lane shift. In: Proceedings of the TRB 88th Annual Meeting, CDR-OM , Washington, DC. Srinivasan, S., Carrick, G., Heaslip, K., Washburn, S., 2007. Analysis of crashes in freeway work-zone queues: a case study. Research Project #: 07-UF-R-S3. Ullman, G.L., Ullman, B.R., Finley, M.D., 2006. Analysis of crashes at active night work zones in Texas. In: Proceedings of the TRB 85th Annual Meeting, CDR-OM, National Research Council , Washington DC. U.S. Department of Transportation, 2003. Manual on Uniform Traffic Control Devices for Streets and Highways (MUTCD), Federal Highway Administration.
Contents lists available at ScienceDirect
Accident Analysis and Prevention journal homepage: www.elsevier.com/locate/aap
Analysis of driver casualty risk for different work zone types Jinxian Weng, Qiang Meng ∗ Department of Civil & Environmental Engineering, National University of Singapore, Singapore 117576, Singapore
a r t i c l e
i n f o
Article history: Received 8 December 2010 Received in revised form 15 March 2011 Accepted 11 April 2011 Keywords: Work zone Driver Logistic Regression Casualty Risk
a b s t r a c t Using driver casualty data from the Fatality Analysis Report System, this study examines driver casualty risk and investigates the risk contributing factors in the construction, maintenance and utility work zones. The multiple t-tests results show that the driver casualty risk is statistically different depending on the work zone type. Moreover, construction work zones have the largest driver casualty risk, followed by maintenance and utility work zones. Three separate logistic regression models are developed to predict driver casualty risk for the three work zone types because of their unique features. Finally, the effects of risk factors on driver casualty risk for each work zone type are examined and compared. For all three work zone types, five significant risk factors including road alignment, truck involvement, most harmful event, vehicle age and notification time are associated with increased driver casualty risk while traffic control devices and restraint use are associated with reduced driver casualty risk. However, one finding is that three risk factors (light condition, gender and day of week) exhibit opposing effects on the driver casualty risk in different types of work zones. This may largely be due to different work zone features and driver behavior in different types of work zones. © 2011 Elsevier Ltd. All rights reserved.
1. Introduction The work zone safety problem has become a high-priority issue for traffic engineer professionals. The presence of a work zone increases traffic conflicts and can cause severe traffic accidents. As shown by many researchers (Bedard et al., 2002; Ullman et al., 2006; Meng et al., 2010), the occurrence likelihood of severe crashes in work zones is higher than that in non-work zones and severe crashes are likely to lead to driver casualties. Hereafter, a driver casualty in this study is referred to as a driver being injured or killed in a vehicle accident that occurred in a work zone. Work zones can be categorized into three types: (1) construction; (2) maintenance; and (3) utility. Fig. 1 gives an example of the three types of work zones. Construction work zones are longterm work zones, where work activities last more than three days (MUTCD, 2003). In construction work zones, a complete traffic control plan should be implemented to improve work zone safety. For example, channelizing devices must be used to warn and alert drivers traveling across the construction work zones. Given that in this context construction is defined as more than three days duration, then construction activities will always extend into the night. Therefore, retroreflective and illuminated devices should be deployed for this work zone type (McAvoy et al., 2007). Also, the speed limit must be posted in construction work zones in order to
∗ Corresponding author. Tel.: +65 65165494; fax: +65 67791635. E-mail address: [email protected] (Q. Meng). 0001-4575/$ – see front matter © 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.aap.2011.04.016
improve the workers’ safety (Brewer et al., 2005). Compared with construction work zones, maintenance work zones usually have short work durations, where work occupies more than one hour but less than three days (MUTCD, 2003). Since the work duration is shorter, simplified traffic control procedures may be implemented. In comparison with construction work zones, maintenance work zones usually undertake lower intensity of work activities. In other words, there are often fewer workers in maintenance work zones than that in construction work zones. Another difference between the two types of work zones is that very few maintenance work zones are provided with law enforcement support (Pigman et al., 2006). Utility work zones significantly differ from the typical construction and maintenance work zones. Utility work is often of a very short duration (less than one hour) and involves very small crew sizes (MUTCD, 2003). In utility work zones, workers and equipment may move along a road at very slow speeds. As workers often need to work on or near the roadways, utility work zones pose unique challenges to driver’s health and safety (Datta et al., 2008). These facts underscore an urgent need to separately assess the driver casualty risk and the effects of its risk factors in each work zone type. 1.1. Literature review A number of studies have been conducted on work zone safety analysis. Bedard et al. (2002) employed a multivariate logistic regression model and found that those drivers who are female,
1812
J. Weng, Q. Meng / Accident Analysis and Prevention 43 (2011) 1811–1817
Fig. 1. (a) construction work zone, (b) maintenance work zone and (c) utility work zone.
drinking and traveling at higher speed have a higher likelihood of being involved in a fatal accident. Hill (2003) investigated the effectiveness of traffic control devices (e.g., flagman, a stop/go signal) on the occurrence likelihood of fatal work zone crashes. Srinivasan et al. (2007) modeled the location of work zone crashes as a multinomial logit function of the lengths of different work zone segments, traffic volume, weather, and other exogenous factors. Harb et al. (2008) explored work zone freeway crash characteristics by using multiple and conditional logistic regression techniques. According to the model results, roadway geometry, weather condition, age, gender, lighting condition, residence code, and driving under the influence of alcohol and/or drugs are the most significant risk factors associated with the work zone crashes. Based on Kansas highway work zone crash data for the years 2003–2004, Li and Bai (2008) developed the crash-severity-index models to measure the work zone crash risk. See et al. (2009) proposed a binary logistic regression model to examine the crash severity for two-lane closure strategies including the Iowa weave strategy and the conventional right-lane closure strategy. However, these existing work zone studies use the aggregated traffic accident data across all work zone types, which may give rise to biased or inaccurate results. For example, Kumar et al. (2008) found that there are no significant differences between weekdays and weekends in casualty risk. Li and Bai (2008) also pointed out that neither the day of week nor the gender has significant effects on the crash severity. In addition, the existing studies implicitly assume that the marginal effect of a risk factor on the casualty risk remains the same for all work zone types. In reality, the marginal effects of each risk factor vary with the work zone type because of their unique configurations. Theoretically, slightly different policies and guidelines should be developed for each type to reduce the casualty risk. Hence, there is a critical need to separately analyze the casualty risk for each type of work zone. 1.2. Objectives and contributions The objective of this study is to analyze the driver casualty risk and examine the effects of its risk contributing factors for the construction, maintenance and utility work zone types, respectively. To achieve this objective, the binary logistic regression technique is first employed to develop a driver casualty risk model for each work zone type. The marginal effect of each risk factor is finally examined and compared across the different types of work zones. The contributions of this study are twofold. First, by considering the differences in the features of the three work zone types, this study makes an initial attempt to separately analyze the driver casualty risk in each work zone type. Second, this study could
provide adequate supports that different policies and guidelines should be put forward for the mitigation of casualty risk in different types of work zones. 2. Logistic regression technique Many non-parametric models, such as classification and regression tree models (Kuhnert et al., 2000; Chang and Wang, 2006) and artificial neural network models (Mussone et al., 1999; AbdelAty and Abdelwahab, 2004), have been employed for traffic safety analysis. Although the non-parametric models can provide a high level of prediction accuracy, they have a weak interpretation of classification results. Non-parametric analysis is also less useful in examining the marginal effects of risk factors, which can provide valuable information for traffic engineers to establish the priorities for risk mitigation. Logistic regression technique is a suitable statistical method for predicting the correlations between a set of independent variables and a discrete target variable. In addition, logistic regression, whose outcomes are of a discrete or categorical nature, can predict the probability of the event of interest and estimate the marginal effect of each explanatory variable. For these reasons, a number of studies have already utilized this technique in traffic safety analysis (Hill, 2003; Li and Bai, 2008; See et al., 2009). For instance, Dissanayake and Lu (2002) developed a set of sequential binary logistic regression models to analyze the contributing factors and predict the crash severities of single-vehicle fixed-object work zone crashes involving young drivers. Since this study concentrates on the driver casualty risk in a work zone crash, the outcome of a driver involved in a work zone accident (the target variable) is assumed to be dichotomous: (i) injury (including death) and (ii) non-injury. A binary logistic regression model is thus employed to estimate the likelihood of each outcome. The target variable, y, can only take two values: y = 1 for the injury and y = 0 for the non-injury. The binary logistic regression model can be formulated as follows: ln
(x , . . . , x ) n 1 1 − (x1 , . . . , xn )
= ˛0 +
n
˛i xi
(1)
i=1
The probability that a driver is killed or injured can be calculated by
P(y = 1|x1 , . . . , xn ) = (x1 , . . . , xn ) =
e
n
˛0 +
1+e
˛x i=1 i i
n
˛0 +
˛x i=1 i i
(2)
J. Weng, Q. Meng / Accident Analysis and Prevention 43 (2011) 1811–1817
1813
Table 1 Variable descriptions. Utility work zone
Observation
Value
Weather
Clear Rain, Snow and Fog Other (blowing Sand)
1 2 3
91.6%a 7.8% 0.6%
83.2% 16.7% 0.1%
90.8% 9.2% 0.0%
Day of week
Weekend Weekday
1 2
24.2% 75.8%
20.0% 80.0%
13.7% 86.3%
Light condition
Daylight Dark with illumination Dark without illumination Dawn, dusk
1 2 3 4
64.1% 13.3% 20.0% 2.6%
68.5% 8.2% 20.8% 2.5%
87.0% 4.6% 8.4% 0.0%
Speed limit
<45 mph 45–60 mph >60 mph
1 2 3
11.3% 62.4% 26.3%
12.0% 50.4% 37.6%
15.3% 77.1% 7.6%
Traffic control devices
None Yes
0 1
33.9% 66.1%
37.8% 62.2%
38.9% 61.1%
Road surface condition
Dry Wet Other (snowy/sand)
1 2 3
87.7% 11.0% 1.3%
78.3% 16.1% 5.6%
85.5% 12.2% 2.3%
Road alignment
Straight Curve Actual number of lanes
1 2 –
88.3% 11.7% –
76.7% 23.3% –
80.9% 19.1% –
Age
Young (age < 30) Middle-age (30 ≤ age <60) Old (age ≥ 60)
1 2 3
26.3% 57.7% 16.0%
26.1% 56.2% 17.7%
23.7% 52.7% 23.6%
Gender
Male Female
1 2
75.1% 24.9%
77.5% 22.5%
77.1% 22.9%
Airbag availability
Not available Available
0 1
57.7% 42.3%
49.2% 50.8%
39.7% 60.3%
Restraint use
Non-used Used
0 1
21.9% 78.1%
25.1% 74.9%
29.8% 70.2%
Truck involvement
Not truck involved Truck involved
0 1
45.5% 54.5%
39.1% 60.9%
20.6% 79.4%
Vehicle age
0–5 year 5–10 year 10–15 year >15 year
1 2 3 4
4.7% 34.7% 33.5% 27.1%
4.6% 33.9% 31.3% 30.2%
1.5% 24.4% 50.4% 23.7%
Notification time Most harmful event
Actual notification time Pedestrian/vehicle ahead Overturn Other (fixed object, pole)
– 1 2 3
– 81.8% 7.3% 10.9%
– 76.5% 6.8% 16.7%
– 63.3% 9.2% 27.5%
Number of lanes
Number of datasets a
Construction work zone
Maintenance work zone
Variable
10,666
797
131
Percentage of drivers based upon the selected variables.
where xi , i = 1, . . ., n are the explanatory variables such as gender, weather and age; (x1 , . . ., xn ) is a conditional probability of the form P(y = 1|x1 , . . ., xn ), and ˛i is the coefficient of explanatory variable, which can directly determine the odds ratio (OR) of the explanatory variable (Harb et al., 2008). The OR describes the relative amount by which the odds of the outcome increase (i.e., OR greater than 1.0) or decrease (i.e., OR less than 1.0) when the value of the explanatory variable increases by 1.0 units. These parameters can be estimated using the maximum likelihood estimation technique. 3. Data 3.1. Database We first analyze the national database – Fatality Analysis Reporting System (FARS) managed by the National Highway Traffic Safety Administration (NHTSA). The work zone driver casualty data in FARS were obtained from the work zone crashes on public
roads within the 51 U.S. states between 2001 and 2006. The impact of traffic flow on driver casualty risk was not taken into account because the FARS database does not provide traffic flow information. According to this database, the number of driver casualties is not found to monotonically increase year by year though the annual traffic volume may increase from an empirical perspective. Therefore, there is inadequate evidence that the traffic flow is related to the driver casualty risk. The lack of consideration of traffic flow will not make the results less reliable. The FARS database provides the following information on each driver casualty for each work zone type: (i) environmental characteristics; (ii) road characteristics; (iii) driver characteristics; and (iv) crash information. Environmental characteristics include detailed information regarding the weather (clear, Rain/Fog/Snow, other), the day of week (weekend, weekday), the light condition (daylight; dark with illumination, dark without illumination, dawn/dusk), the speed limit (<45 mph, 45–60 mph, >60 mph) and the use of traffic control devices (none, yes). The road characteristics consist of the road surface condition (dry, wet, other), the road
1814
J. Weng, Q. Meng / Accident Analysis and Prevention 43 (2011) 1811–1817
Table 2 Multiple t-test results of driver casualty risk at different work zone types. Work zone type
Construction Maintenance Utility a
Sample size
10,666 797 131
Mean
t-test statistic (p-value)
0.722 0.691 0.672
Maintenance
Utility
2.47 (0.014)a
3.97 (<0.001) 1.18 (0.237)
p-value in parentheses.
alignment (straight, curve) and the number of available lanes. Driver characteristics include the driver’s age and gender. Crash information comprises the airbag availability (not available, available), the restraint use, truck involvement, vehicle age (0–5 year, 5–10 year, 10–15 year, >15 year), notification time and the most harmful event (pedestrian or vehicle ahead, overturn, other). 3.2. Data processing Table 1 presents 16 variables described above. Each variable is represented by a binary or an ordinal number. It should be noted that these 16 variables are selected based on the percentage of missing data (“Unknown”). The variables with more than 7.9% missing data (e.g., travel speed) are excluded. A preliminarily check of the data quality for the selected variables is performed before applying them for the analysis. For records with an “unknown” value for any of the sixteen selected variables, a value is generated based on the remaining records in the sample. For example, in approximately 24% of the records pertaining to construction work zones, the day of week has the value “Weekend”. Therefore, there is a 24% probability of replacing an unknown value for the day of week for a record from the construction work zone with the value “Weekend” and a 76% probability of replacing it with the “Weekday”. In the FARS database, injury severity is given as no injury, possible injury, non-incapacitating evident injury, incapacitating injury, fatality, injury-severity unknown and died prior to crash. Since this study focuses on the driver casualty risk in work zone crashes, the records where drivers died prior to the crash are excluded from the analysis. Therefore, a total of 11,594 records of drivers involved in work zone vehicle accidents between 2001 and 2006 have been collected. Of these records, 10,666 drivers are involved in construction work zone accidents; 797 drivers suffer different severity levels of
injuries in maintenance work zone accidents and 131 drivers are involved in utility work zone accidents. Due to the relatively small number of driver fatalities in utility work zones, this study assumes that there are only two outcomes of a driver involved in an accident: non-injury and injury. Thus, the possible injury, non-incapacitating evident injury, incapacitating injury, fatality and injury-severity unknown are aggregated into the injury classification. 4. Results and discussions 4.1. Descriptive analysis results Table 1 provides detailed summary on statistics of the selected variables for the three work zone types. Approximately 75% of the drivers involved in construction work zone accidents are male. Similar statistics are found for the maintenance and utility work zone accidents. The distribution of the use of traffic control devices is also similar across the three work zone types as are the data related to restraint use, weather, road surface condition, road alignment and age. However, other variables (e.g., light condition, airbag availability, most harmful event) vary substantially among the three work zone types. For example, a minority of the drivers involved in construction work zone accidents (42.3%) have airbags in their vehicles, whereas 60.3% of those involved in utility work zone accidents used airbags. In maintenance work zones, the proportion of drivers using the airbag approximates to that without using the airbag function. A multiple t-test is employed to investigate whether there is a statistical difference of the mean driver casualty risk between any two work zone types. Table 2 gives the multiple t-test results. The mean driver casualty risk in the construction work zone is 0.722, which is statistically larger than both 0.691 in the maintenance
Table 3 Statistic results of the logistic regression model for the construction work zone. Variable
Coefficient (ˇ)
Standard error
Wald Chi-square
p-Value
Intercept Weather Day of week Light condition Speed limit Traffic control devices Road surface condition Road alignment Number of lanes Age Gender Airbag availability Restraint use Truck involvement Vehicle age Notification time Most harmful event
−1.0189 0.0583 −0.0272 −0.1260 0.0842 −0.1565 0.1303 0.2865 0.0582 0.1561 0.6585 −0.4839 −1.9078 0.5888 0.1274 0.0044 0.2546
0.278 0.105 0.013 0.027 0.042 0.052 0.088 0.084 0.023 0.038 0.061 0.050 0.087 0.052 0.027 0.001 0.041
13.4 0.31 4.18 21.4 4.10 9.23 2.21 11.6 6.66 17.0 118 94.3 481 129 22.1 11.9 38.6
<0.01 0.58 0.04 <0.01 0.04 <0.01 0.14 <0.01 0.01 <0.01 <0.01 <0.01 <0.01 <0.01 <0.01 <0.01 <0.01
AIC SC -2log-likelihood
Intercept only 12,613.9 12,621.2 12,611.9
Intercept and covariates 11,108.2 11,231.9 11,074.2
J. Weng, Q. Meng / Accident Analysis and Prevention 43 (2011) 1811–1817
1815
Table 4 Statistic results of the logistic regression model for the maintenance work zone. Variable
Coefficient (ˇ)
Standard error
Wald Chi-square
p-Value
Intercept Weather Day of week Light condition Speed limit Traffic control devices Road surface condition Road alignment Number of lanes Age Gender Airbag availability Restraint use Truck involvement Vehicle age Notification time Most harmful event
−0.9012 −0.1037 0.0903 0.1563 −0.0168 −0.2305 0.1658 0.6112 0.0492 0.2177 −0.2738 −0.6668 −1.8642 0.3244 0.0888 0.0128 0.6149
0.463 0.257 0.043 0.082 0.114 0.135 0.184 0.164 0.061 0.096 0.149 0.131 0.198 0.153 0.050 0.004 0.098
3.79 0.16 4.47 3.62 0.02 2.92 0.82 13.9 0.66 5.15 3.38 25.9 89.1 4.51 3.15 11.1 39.3
0.05 0.69 0.04 0.05 0.88 0.08 0.37 <0.01 0.42 0.02 0.06 <0.01 <0.01 0.03 0.07 0.02 <0.01
AIC SC -2log-likelihood
Intercept only 987.1 991.8 985.1
work zone and 0.672 in the utility work zone. However, there is no statistical difference between the mean risks in the maintenance and utility work zones (at the significance level of 0.05). This may be due to the small sample sizes for these two work zone types. 4.2. Logistic regression model results for three different work zone types The logit procedure in the Statistical Analysis Software (SAS, 2008) is performed to calibrate a separate logistic regression model for each work zone type. Table 3 gives the results for the construction work zone. It can be seen that the driver casualty risk in a construction work zone is significantly (at a significance level of 0.10) increased with the following variables: speed limit, road alignment, number of lanes, age, gender, truck involvement, vehicle age, notification time and most harmful event. The coefficients associated with the day of week, light condition, traffic control devices, airbag availability and restraint use are negative and statistically significant at a significance level of 0.10, indicating that these five variables are significantly associ-
Intercept and covariates 874.0 953.6 840.0
ated with the decreased driver casualty risk in the construction work zone. In a maintenance work zone, similarly, the variables that can significantly increase driver casualty risk include the road alignment, age, truck involvement, vehicle age, notification time and most harmful event, as shown in Table 4. The traffic control devices, restraint use and airbag availability are shown to reduce the casualty risk. The results for the utility work zone are shown in Table 5. Road surface condition, road alignment, number of lanes, age, vehicle age and notification time all increase the casualty risk. Among the 16 variables, only four variables including speed limit, age and airbag availability do not significantly influence the driver casualty risk in utility work zones. 4.3. Marginal effects of risk factors for the three work zone types The marginal effects of the risk factors on driver casualty risk across different work zone types are shown in Table 6. For clarity, the results are discussed according to the categories mentioned
Table 5 Statistic results of the logistic regression model for the utility work zone. Variable
Coefficient (ˇ)
Standard error
Wald Chi-square
p-Value
Intercept Weather Day of week Light condition Speed limit Traffic control devices Road surface condition Road alignment Number of lanes Age Gender Airbag availability Restraint use Truck involvement Vehicle age Notification time Most harmful event
−7.2366 −3.5173 0.1853 −0.7325 0.2535 −3.0772 1.7234 1.0353 1.0106 0.0741 0.5977 −0.4300 −1.7939 2.3032 0.2143 0.0543 0.2904
2.477 0.554 0.084 0.176 0.335 0.488 0.657 0.409 0.263 0.158 0.267 0.293 0.312 0.542 0.112 0.010 0.131
8.54 40.3 4.91 17.4 0.57 39.7 6.88 6.42 14.8 0.22 5.00 2.16 33.1 18.0 3.67 29.8 4.93
<0.01 <0.01 0.03 <0.01 0.45 <0.01 0.01 0.011 <0.01 0.64 0.03 0.14 <0.01 <0.01 0.05 <0.01 0.03
AIC SC -2log-likelihood
Intercept only 167.8 170.7 165.8
Intercept and covariates 150.9 199.8 116.9
1816
J. Weng, Q. Meng / Accident Analysis and Prevention 43 (2011) 1811–1817
Table 6 Relative effects of risk factors on driver casualty risk at different work zone types. Variable
Weather Day of week Light condition Speed limit Traffic control devices Road surface condition Road alignment Number of lanes Age Gender Airbag availability Restraint use Truck involvement Vehicle age Notification time Most harmful event a b
Construction work zone
Maintenance work zone
Utility work zone
ORa
Lower CI
Upper CI
Significantb
OR
Lower CI
Upper CI
Significant
OR
Lower CI
Upper CI
Significant
1.06 0.97 0.88 1.09 0.86 1.14 1.33 1.06 1.17 1.93 0.62 0.15 1.80 1.14 1.004 1.29
0.86 0.95 0.84 1.00 0.77 0.96 1.13 1.01 1.09 1.72 0.56 0.13 1.63 1.08 1.002 1.19
1.30 1.00 0.93 1.18 0.95 1.35 1.57 1.11 1.26 2.18 0.68 0.18 2.00 1.20 1.007 1.40
N Y Y Y Y N Y Y Y Y Y Y Y Y Y Y
0.90 1.09 1.17 0.98 0.79 1.18 1.84 1.05 1.24 0.76 0.51 0.16 1.38 1.09 1.01 1.85
0.44 1.01 1.00 0.79 0.61 0.82 1.34 0.93 1.03 0.57 0.40 0.11 1.03 1.00 1.005 1.53
1.83 1.19 1.37 1.23 1.00 1.69 2.54 1.18 1.50 1.00 0.66 0.23 1.87 1.26 1.021 2.24
N Y Y N Y N Y N Y Y Y Y Y Y Y Y
0.03 1.20 0.48 1.29 0.05 5.60 2.82 2.75 1.08 1.82 0.65 0.17 10.0 1.24 1.05 1.34
0.01 1.02 0.34 0.67 0.02 2.10 1.26 1.64 0.79 1.08 0.37 0.10 3.46 1.00 1.035 1.04
0.09 1.42 0.68 2.48 0.12 14.9 6.27 4.60 1.47 3.07 1.16 0.31 28.9 1.69 1.077 1.73
Y Y Y N Y Y Y Y N Y N Y Y Y Y Y
Odds ratio (OR) is calculated by changing one unit of one variable at a time while controlling for the other variables, where OR = exp(). N, not significant at p ≤ 0.10 for the risk factor; Y, significant at p ≤ 0.10 for the risk factor.
earlier. Here, the marginal effect for a given risk factor can be represented by its odds ratio (OR) defined as the relative amount by which the odds of the injury increase or decrease when the risk factor’s value increases by 1.0 units. 4.3.1. Environmental characteristics The OR associated with the day of week is 0.97 for the construction work zones, 1.09 for the maintenance work zones and 1.20 for the utility work zones. This result implies that a driver incurs a higher casualty risk in the maintenance and utility work zones during the weekday than at the weekend. However, she/he has a 3% lower casualty risk on a weekday than at the weekend in construction work zones. The fact that the higher traffic may slow down drivers during weekdays could account in part for the 3% lower risk. This finding could also explain why Kumar et al. (2008) found no significant differences of the casualty risk between weekdays and weekends if the crash data they used were aggregated across different work zone types. In construction work zones, drivers are less likely to be injured in the dark with illumination condition (OR = 0.88) than in the daylight condition. However, in the maintenance work zones, drivers are 17% more likely to be injured under the dark with illumination condition (OR = 1.17). The opposing effects may be because retroreflective signs are always used for the night construction work so that drivers will be more vigilant driver (McAvoy et al., 2007). However, retroreflective signs are less used in the night maintenance work zone (MUTCD, 2003). As shown by many researchers (Lu et al., 2006; Li and Bai, 2008), the speed limit in the construction work zones (OR = 1.09) is highly correlated with driver casualty risk. A high speed limit increases the driver casualty risk in the construction work zone accidents, while it is not statistically significant in the maintenance and utility work zones. One possible reason may be that the speed limit works collectively with other factors (e.g., work intensity). For example, the maintenance work zone undertaking the very low intensity of work is often posed no speed limit while the high work intensity usually poses low speed limit. Therefore, further investigation of the relationship between the work intensity and the speed limit should be carried out for maintenance work zones. As expected, the presence of traffic control devices significantly reduces driver casualty risk for all the three work zone types though their marginal effects are not the same. More specifically, drivers have a 14% lower casualty risk when traffic control devices are used in the construction work zones, and a 21% lower casualty risk when traffic control devices are used in the maintenance work zones.
4.3.2. Road characteristics The road surface condition is found to be associated with increased driver casualty risk in utility work zones while it is not statistically associated with increased driver casualty risk in the other two types of work zones. Although the road alignment is a contributing factor, its marginal effect is not the same across the three work zone types. On a curved road, drivers have a 33% higher casualty risk in construction work zones, an 84% higher risk in maintenance work zones and a 182% higher risk in utility work zones. Similarly, the number of lanes is also associated with increased casualty risk in the construction and utility work zones. The increase of casualty risk associated with increase in the number of lanes may possibly be explained by the increase in the number of lane change conflicts, which may result in severe vehicle crashes (Kononov et al., 2008). However, it has no substantial impact on the casualty risk in maintenance work zones. 4.3.3. Driver characteristics Middle-aged drivers are 1.17 times more likely than young drivers to be injured or killed in construction work zones and this figure is 1.24 times in maintenance work zones. Although the OR associated with the age in utility work zones is larger than 1.0, it is not statistically associated with increased driver casualty risk at the significance level of 0.10. In addition, female drivers have a 93% higher casualty risk than male drivers in the construction work zones (CI = 1.72–2.18), and an 82% higher risk than male drivers in the utility work zones (CI = 1.08–3.07). The increased casualty risk associated with the female driver in construction and utility work zones results from two factors: (1) increased risk-taking in younger female drivers and (2) increased exposure in older female drivers (Kostyniuk et al., 1996). 4.3.4. Crash characteristics Consistent with previous studies, airbag availability is associated with a lower driver casualty risk in both construction and maintenance work zones (Evans, 1991). Compared with the drivers in the vehicles without airbags, the drivers in the vehicles equipped with airbags are less prone (=38%) to suffering casualties in the construction work zones, and a 49% less likely in the maintenance work zones. Accidents involving trucks are found to dramatically increase the driver casualty risk, especially in utility work zones. More specifically, drivers caught up in utility work zone accidents involving a truck have a 10.0 times greater casualty risk. However, the marginal effect of truck involvement in the construction work
J. Weng, Q. Meng / Accident Analysis and Prevention 43 (2011) 1811–1817
zones (OR = 1.80) is less than that in utility work zones but larger than that in maintenance work zones (OR = 1.38). Consistent with our expectations, the vehicle age also affects driver casualty risk. In construction work zones, a driver in a vehicle that is 5–10 years old has a 14% higher casualty risk than the driver in a newer vehicle (e.g., ≤5 years). Also, the longer notification time is associated with the higher driver casualty risk in all three work zone types (OR = 1.004 for the construction, OR = 1.010 for the maintenance and OR = 1.004 for the utility). Consistent with previous studies (Levine et al., 1999; Adadzi et al., 2008), our results also indicate that the type of collision (most harmful event) is associated with a decreased driver casualty risk in all the three work zone types. 5. Conclusions Taking into account the differences in work zone features, this study has separately analyzed the driver casualty risk and investigated the risk factors for the three work zone types: construction, maintenance and utility. Driver casualty data from 2001 to 2006 for each work zone type are taken from the FARS. The univariate analysis implies that the distributions of some of the risk factors vary with the work zone type. The multiple t-tests further show that the construction work zones have the largest driver casualty risk, followed by the maintenance and utility work zones. Based on these results, three separate logistic regression models are developed to predict driver casualty risk for the three work zone types. Finally, the marginal effects of each factor on the driver casualty risk in each work zone type are examined and compared. Seven factors (road alignment, truck involvement, most harmful event, traffic control devices, restraint use and vehicle age) are found to significantly impact driver casualty risk for all the three work zone types. Probably because of the small sample size for the utility work zones, age and airbag availability are not found to significantly influence the driver casualty risk in the utility work zones. The speed limit is not found to be a risk factor in maintenance and utility work zones probably as a result of the fact that imposed speed limits in these two work zone types may interact with work intensity. Hence, further investigation of the relationship between work intensity and speed limit should be carried out in the future. One finding from this study is that three factors (light condition, gender, and the day of week) exhibit heterogeneous effects on driver casualty risk in different types of work zones. Drivers in construction work zones tend to be more likely to suffer injuries or fatalities at the weekend while drivers in the maintenance and utility work zones are less likely to be injured or killed in the weekend accidents. These results provide supports for the argument that slightly different policies and guidelines should be developed for different types of work zones in order to reduce the casualty risks. For example, construction activities are encouraged to be carried out during the weekdays while the maintenance and utility works should be implemented at the weekends. Additional measures should be proposed to forbid risky driving behavior in construction and utility work zones.
1817
References Abdel-Aty, M., Abdelwahab, H., 2004. Predicting injury severity levels in traffic crashes: a modeling comparison. Journal of Transportation Engineering 130 (2), 204–210. Adadzi, K.D., Classen, S., Hall, A., Duncan, R.P., Garvan, C.W., 2008. Predictors of injury among younger and older adults in fatal motor crashes. Accident Analysis and Prevention 40, 1804–1810. Bedard, M., Guyatt, G., Stones, M., Hirdes, J., 2002. The independent contribution of driver, crash and vehicle characteristics to driver fatalities. Accident Analysis and Prevention 34 (6), 717–727. Brewer, M.A., Pesti, G., Schneider IV, W.H., 2005. Identification and testing of measures to improve work zone speed limit compliance. Report No. FHWA/TX06/0-4707-1, Texas Transportation Institute, Texas. Chang, L., Wang, H., 2006. Analysis of traffic injury severity: an application of nonparametric classification tree techniques. Accident Analysis and Prevention 34 (5), 1019–1027. Datta, T., Savolainen, Grillo, L., Schattler, K., 2008. Utility work zone traffic control guidelines. Technique Project # DTFH61-06-G-00006. Dissanayake, S., Lu, J., 2002. Analysis of severity of young driver crashes, sequential binary logistic regression modeling. Transportation Research Record 1784, 108–114. Evans, L., 1991. Airbag effectiveness in preventing fatalities predicted according to type of crash, driver age, and blood alcohol concentration. Accident Analysis and Prevention 23, 531–541. Harb, R., Radwan, E., Yan, X., Pande, A., Abdel-Aty, M., 2008. Freeway work-zone crash analysis and risk identification using multiple and conditional logistic regression. Journal of Transportation Engineering 134 (5), 203–214. Hill, R.W., 2003. Statistical analysis of fatal traffic accident data. Texas Tech University, Lubbock, Texas. Kononov, J., Bailey, B.A., Allery, B.K., 2008. Exploratory analysis of relationship between number of lanes and safety on urban freeways. In: Proceedings of the TRB 87th Annual Meeting, CDR-OM, National Research Council , Washington, DC. Kostyniuk, L., Molnar, L., Eby, D., 1996. Are women taking more risks while driving? In: Proceedings of the Women’s Travel Issues Second National Conference , Baltimore. Kumar, A., Lalwani, S., Agrawal, D., Rautji, R., Dogra, T.D., 2008. Fatal road traffic accidents and their relationship with head injuries: an epidemiological survey of five years. Indian Journal of Neurotrauma 5, 63–77. Kuhnert, P.M., Do, K., McClure, R., 2000. Combining non-parametric models with logistic regression: an application to motor vehicle injury data. Computational Statistics and Data Analysis 34 (3), 371–386. Levine, E., Bedard, M., Molloy, D.W., Basilevsky, A., 1999. Determinants of driver fatality risk in front impact fixed object collisions. Mature Medicine Canada 2, 239–242. Li, Y., Bai, Y., 2008. Development of crash-severity-index models for the measurement of work zone risk levels. Accident Analysis and Prevention 40 (5), 1724–1731. McAvoy, D.S., Schattler, K.L., Datta, T.K., 2007. Driving simulator validations for nighttime construction work zone devices. Transportation Research Record 2015, TRB, 55–63. Meng, Q., Weng, J., Qu, X., 2010. A probabilistic quantitative risk assessment model for the long-term work zone crashes. Accident Analysis and Prevention 42 (6), 1866–1877. Mussone, Li., Ferrari, A., Oneta, M., 1999. An analysis of urban collisions using an artificial intelligence model. Accident Analysis and Prevention 31 (6), 705–718. Pigman, J.G., Agent, K.R., Green, E.R., 2006. Evaluation of work zone safety operations and issues. Research Report KTC-06-08/SPR287-05-1F, University of Kentucky, (UK). SAS, 2008. SAS/STAT 9.2 User’s Guide. Cary, NC: SAS Publishing. See, C.F., Schrock, S.D., McClure, K., 2009. Crash analysis of work zone lane closures with left-hand merge and downstream lane shift. In: Proceedings of the TRB 88th Annual Meeting, CDR-OM , Washington, DC. Srinivasan, S., Carrick, G., Heaslip, K., Washburn, S., 2007. Analysis of crashes in freeway work-zone queues: a case study. Research Project #: 07-UF-R-S3. Ullman, G.L., Ullman, B.R., Finley, M.D., 2006. Analysis of crashes at active night work zones in Texas. In: Proceedings of the TRB 85th Annual Meeting, CDR-OM, National Research Council , Washington DC. U.S. Department of Transportation, 2003. Manual on Uniform Traffic Control Devices for Streets and Highways (MUTCD), Federal Highway Administration.