trb 2013 annual meeting paper revised from original...

Downloaded from amonline.trb.orgDownloaded from amonline.trb.orgDownloaded from amonline.trb.orgDownloaded from amonline.trb.orgDownloaded from amonline.trb.orgDownloaded from amonline.trb.orgDownloaded from amonline.trb.orgDownloaded from amonline.trb.orgDownloaded from amonline.trb.orgDownloaded from amonline.trb.orgDownloaded from amonline.trb.orgDownloaded from amonline.trb.orgDownloaded from amonline.trb.orgDownloaded from amonline.trb.org

Crash Frequency Modeling for Highway Construction Zones 1

2

Ozgur Ozturk (Corresponding Author) 3 Graduate Research Assistant 4

Rutgers Intelligent Transportation Systems (RITS) Lab 5

Department of Civil and Environmental Engineering 6

Rutgers, The State University of New Jersey 7

623 Bowser Road, Piscataway, NJ 08854, USA 8

Tel: (732) 445-5496, Fax: (732) 445-0577 9

Email: [email protected] 10

11

Kaan Ozbay, Ph.D. 12 Professor & Director 13




96 Frelinghuysen Road, Piscataway, NJ 08854, USA 17

Tel: (732) 445-2792 Fax: (732) 445-0577 18


20

Hong Yang, Ph.D. 21 Post-Doctoral Associate 22




623 Bowser Road, Piscataway, NJ 08854, USA 26

Tel: (732) 445 0579 ex.127 Fax: (732) 445-0577 27


29

Bekir Bartin, Ph.D. 30 Research Associate 31




623 Bowser Road Piscataway, NJ 08854, USA 35

Tel: (732) 445-5496 Fax: (732) 445-0577 36

E-mail: [email protected] 37

38

39

40

41

42

Word count: 4683 Texts + 4 Figures + 6 Tables=7183 43

Abstract: 217 44

Submission Date: August 1, 2012 45

46

47

48

TRB Paper ID: 13-4555 49

Transportation Research Board’s 92nd Annual Meeting, Washington, D.C., 2013 50

TRB 2013 Annual Meeting Paper revised from original submittal.Downloaded from amonline.trb.org

Ozturk, Ozbay, Yang and Bartin 2

ABSTRACT 1 Throughout the country, work zone safety issues have received considerable attention in recent years due 2

to increasing work zone crashes along numerous highway renovation and reconstruction projects. In 3

general, previous studies were able to consider limited number of contributing factors mainly due to the 4

limitations in data availability. The main goal of this paper is to remedy this major problem related to data 5

availability and estimate improved models using data from multiple sources. Work zone project drawings, 6

crash database and straight line diagrams are used to create an integrated work zone safety database. 7

Work zone crash data is plotted in time and space to validate, locate and adjust work zone related 8

information. The negative binomial regression approach is used as the appropriate model to predict crash 9

frequency within these work zones. Traffic volume is adjusted for daytime and night time conditions in 10

terms of hourly distribution of daily traffic. The duration-based and period-based models are also 11

developed to address relationship between potential factors and to predict crash frequency on work zones 12

in terms of property damage only (PDO) and injury crashes. Compared with previous frequency models, 13

additional parameters such as number of lanes closed and speed reduction are used. These additional 14

factors identified as significant can help traffic engineers to further improve safety of work zone projects. 15



INTRODUCTION 1 Work zone safety has received much attention in the recent years due to numerous highway renovation 2

projects that resulted in many work zone crashes. There were 87,606 nationwide work zone crashes 3

occurred in 2010 and 37,476 of which were an injury crashes, resulting approximately one injury work 4

zone crash at every 14 minutes (1). 5

Similar statistics are observed at the state level. New Jersey (NJ) has been called the corridor 6

state, an apparent consequence of its perceived role and proximity to the strong markets in New York 7

City, Philadelphia and the Boston-Washington Northeast Corridor. It experiences the second highest 8

travel delay and total congestion cost in the nation (2). Moreover, its aging infrastructure requires regular 9

maintenance and replacement. About half of the state-maintained roads are in deficient condition (3). 10

Furthermore, NJ is the state with the highest capital and bridge disbursement and the maintenance 11

disbursement per mile (4). Accordingly, an increase is expected in number of work zone projects and 12

work zone crashes as well. In 2010, NJ crash statistics shows this increase distinctly for work zone 13

labeled crashes. Yearly average number of work zone crashes between 2004 and 2009 is 5,395; however 14

this number increased in 2010 by 26.7 percent to 6,837 crashes equal to 7.8 percent of the nationwide 15

total number of work zone crashes (5). Thus, there is a need to study the specific factors that contribute to 16

work zone crash frequencies in NJ. 17

The majority of the work zone crash studies focused on severity analysis and the related 18

contributing factors. There are only a few studies that directly addressed work zone crash frequency 19

modeling (6-11). Negative binomial regression method is generally used to model work zone crash 20

frequency with model parameters such as annual average daily traffic (AADT), duration, length, work 21

zone type and urban indicator to find contributory factors related to work zone crashes. Some of these 22

studies reported limitations in terms of data. By using different available data sources, an enhanced model 23

can be developed for work zone crash frequency by accounting more contributing factors. 24

The main objective of this study is to develop an improved work zone crash frequency model by 25

incorporating additional important parameters such as speed reduction, the number of lanes closed, road 26

systems and number of intersections. These parameters are not generally included in the work zone crash 27

frequency models proposed by previous studies (e.g. 7, 9). Seasonally and hourly adjusted directional 28

AADT values are also used in the model. NJ crash database (2004-2010), project drawings and NJ 29

straight line diagrams are combined to create an integrated database. Crash frequency is investigated for 30

injury and property damage only (PDO) crashes for day and night time conditions separately. In the 31

duration-based model, the number of total crashes is taken into account for the work zone duration, and in 32

the period-based model three monthly crash counts are used to set models. Poisson and negative binomial 33

regressions are employed to model crash count data. 34

This paper is organized as follows. Next section provides a comprehensive review of the 35

literature related to crash frequency modeling of the work zones. The following section describes the 36

integrated database used for modeling and presents the model specification. The last section presents 37

estimation results and summarizes major findings. 38

39

LITERATURE REVIEW 40 Previous studies provide important information for work zone impact on crash rates. Juergens (12) 41

reported an increase in crash rate between 7.0 to 21.4 percent for ten work zone locations. Graham (13) 42

stated an average increase of 7.5 percent for crash rates within 79 work zone sites. Rouphail et al. (14) 43

and Hall and Lorenz (15) concluded in their study that crash rates increased with the presence of work 44

zones more than 88 percent and 26 percent respectively. Garber and Woo (16) reported that the crash 45

rates at work zones on multilane highways increased on the average by 57 percent and the crashes at work 46

zones on two-lane urban highways increased about 168 percent on average. The research by Pigman and 47

Agent (17) also showed increasing crash rates for 14 out of 19 work zones when compared to the before 48

work zone conditions. Rouphail et al. (14) and Ha and Nemeth (18) found that work zone crashes are less 49

severe than pre-construction crashes. On the contrast, Pigman and Agent (17) and Garber and Zhao (19) 50

stated that work zone crashes are more severe than non-work zone crashes. Benekohal et al. (20) reported 51



the survey among truck drivers that 90 percent of them consider traveling through work zone is more 1

dangerous than non-work zone areas. Rouphail et al. (14) and Wang et al. (21) found that rear-end crashes 2

increased significantly during work zone periods. Khattak et al. (9) found that crash rates were higher 3

with the presence of work zones 23.5 percent for non-injury crashes and 17.5 percent for injury crashes. 4

Contrary to other studies described above, Jin et al. (22) observed lower crash rates during construction 5

periods in Utah. Dissanayake and Akepati (23) found that most of the work zone crashes at five states 6

(Iowa, Kansas, Missouri, Nebraska, and Wisconsin) occurred during the daytime and under clear weather 7

(no adverse) conditions. According to the descriptive studies about work zone crash rates, general 8

opinions are that work zones have an increasing effect on crash frequency when compared to pre-work 9

zone conditions. 10

Crash frequency models have been used for road safety analysis in many studies. Negative 11

binomial models (Poisson-Gamma) are generally used for modeling crash count data. In addition, there 12

are several methods to model crash frequencies such as Poisson model, zero inflated negative binomial 13

and Poisson, truncated regression, generalized additive model, Conway Maxwell Poisson model, negative 14

multinomial model etc. (24). There are only a few studies in the literature that are directly related to the 15

work zone crash frequency modeling. The following studies are focused on modeling work zone crash 16

occurrence by using different factors as model components. 17

As shown in Table 1, different methodologies and parameters were applied to develop work zone 18

crash frequency models. Pal and Sinha (6) investigated the safety effects of lane closure strategies at 19

interstate work zones in Indiana. Normal regression, negative binomial and Poisson regression models 20

were developed by using duration, traffic interaction and crash rate at normal conditions. They found that 21

normal regression predicts crash counts better when compared to negative binomial and Poisson 22

regression results. Venugopal and Tarko (7) proposed two models for crash frequency on approaches to 23

work zones and inside the work zones. They used project cost and work zone type as additional 24

parameters based on previous studies. Duration and length variables were statistically significant and the 25

dependence was almost linear. Elias and Herbsman (8) used Monte Carlo simulation to develop a work 26

zone crash frequency probability function. According to results, work zone crashes increase incrementally 27

when daily traffic volume (ADT), duration, and work zone length increase. Khattak et al. (9) developed 28

negative binomial model for crash modeling by using following parameters: ADT, length, duration, urban 29

indicator, injury indicator, work zone indicator. Separate models were constructed for both injury and 30

non-injury in the pre-work zone and during work zone periods. According to negative binomial regression 31

results, duration has significant effect on both injury and non-injury work zone crash frequencies. Qi et al. 32

(10) modeled rear-end work zone crashes by using truncated negative binomial regression. They found 33

that occurrence of rear-end crashes are more likely in work zones with flaggers, alternating one way 34

traffic, and higher AADT conditions. Srinivasan et al. (11) attempted to model the location of crashes 35

within work zones as a function of the length of work zone segments, traffic volume and weather 36

parameters. Multinomial logit model was used to construct crash probabilities per lane-mile for different 37

segments. The work zone components were investigated separately. Among various weather and traffic 38

volume scenarios, bad weather and high traffic volumes made the advance warning area relatively unsafe, 39

and the termination area was found to be the most unsafe segment during off-peak hours. 40

Work zone length, AADT, duration, project cost and work zone type are the most common 41

parameters used to set crash frequency models statistically. Based on previous studies, additional 42

parameters related to work zone safety structure such as number of lanes closed, speed reduction, road 43

system and number of intersections are identified as possible additional variable to model the relationship 44

between work zone and crash frequency. Several statistical methods are used to model work zone crash 45

frequency. The negative binomial regression is one of the most favored techniques adopted to model 46

crash frequency because of its performance on possible overdispersion in the data (24). 47



TABLE 1 Work Zone Crash Frequency Modeling Literature Summary 1

Work Zone Crash

Frequency Models

Pal & Sinha

(1996)(6)

Venugupal

& Tarko

(2000)(7)

Elias &

Herbsman

(2000)(8)

Khattak

et. al

(2002)(9)

Qi et. Al

(2005)

(10)

Srinivasan

et. Al

(2008)(11)

This

Study

Duration * * * * *

*

AADT * * * * * * *

Length * * * *

* *

No. of Operating Lanes

*

Work Zone Speed Limit

*

Cost of Project

*

Lane Closure

*

*

Speed Reduction

*

Urban Indicator

* *

Road System

*

Weather *

*

Crash Rate

*

Intersection

*

*

Ramp

*

*

Daytime-Nighttime

*

PDO-Injury

*

*

Control Device

*

Type of Work

*

*

Sample Size (Site) 34 116 - 36 - 1 60

Model NB, NLR,P NB Monte Carlo NB TNB MNL NB

NB:Negative Binomial Regression, P:Poisson, NLR:Normal Linear Regression, TNB:Truncated NB, MNL:Multinomial Logit Regression

2

DATA DESCRIPTION 3 Statistical software R is used to merge and analyze the crash data for this study. Crash counts are 4

clustered by three monthly periods (seasonal) for each work zone and separated into four categories as 5

following; daytime PDO, daytime injury, nighttime PDO and nighttime injury crashes. Fatal crashes are 6

assumed as injury crashes to avoid bias. Table 2 shows the number of crashes for each set in this study. 7

8

TABLE 2 Number of Work Zone Crashes for Each Category 9

Category PDO Injury Total

Daytime Crashes 2915 862 3777

Nighttime Crashes 1192 413 1605

Total Crashes 4107 1275 5382

10

AADT is a critical parameter for crash models since it represents the exposure of traffic. Hence, 11

accuracy of the AADT values is important for modeling. Khattak et al. (9) stated that directional AADT 12

should be used for modeling to determine crash distribution more precisely. To avoid over-representation 13

of AADT within the model, directional AADT from the NJ Straight Line Diagrams is used for given 14

milepost and time post of work zones. AADT values are adjusted by using New Jersey Department of 15

Transportation (NJDOT) seasonal adjustment factors since these values are estimated yearly (5). Instead 16

of using AADT directly, a traffic volume parameter is used for daytime and nighttime conditions 17

separately. It is generated by adjusting the seasonal-directional-AADT value based upon hourly 18



adjustment factors. Bourne et al. (25) reported an example of normalization issues for results that daytime 1

work zone crashes are overrepresented. Nighttime traffic is approximately estimated as a quarter of the 2

total daily traffic. In this way, a biased relationship between AADT and crash counts is avoided by using 3

reduced traffic volume for nighttime condition. Figure 1 shows hourly distribution of all work zone 4

crashes occurred in NJ between 2004-2010. 5

6 FIGURE 1 Hourly distribution of all work zone crashes in NJ, 2004-2010 (5). 7

8

Data from the NJ crash database for the period between 2004 and 2010, work zone project 9

drawings and NJ straight line diagrams are combined to create the integrated database for modeling work 10

zone crash frequencies. After separating each work zone by direction, and filtering incomplete data 11

mainly because of unclear project drawings, 60 work zones are used in frequency modeling. Work zone 12

spatial information is gathered from project drawings. Project drawings do not include time and date 13

information regarding work zones. Accordingly, spatio-temporal diagrams of work zone crashes for each 14

work zone are plotted to better determine location and duration of the work zones. Advance warning, 15

transition and termination areas are not included in project mileposts. Therefore project mileposts are 16

sometimes updated to capture all of work zone labeled crashes. A sample spatio-temporal diagram of the 17

work zone crashes along I-78 and related project drawing are shown in Figure 2 and Figure 3. 18

19

20 FIGURE 2 Temporal spatial plotting of construction zone crashes at I-78. 21

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23

Time of Day

Nu

mb

er

of W

ork

Zo

ne

Cra

sh

es

0500

1000

1500

2000

2500

3000

608524 483

368 333437

911

1747

2485

22192095

2381

267826582724 2762

26782690

2010

1447

12171314 1276

940



1 FIGURE 3 Sample project drawing (cover page) for the work zone project on I-78 (26). 2

3

Figure 4 shows a sample for the spatio-temporal distribution of the work zone crashes within 4

different work zone sections. Milepost information for work zone components is gathered from project 5

drawings. Ramp and intersection locations are also included. From the Figure 4, it can be seen that work 6

zone crashes are likely to be denser near ramps and intersections. 7

8

9 FIGURE 4 Spatial distribution of crashes within work zone on US 1 North (5). 10

11

Work zone length, milepost, number of operating lanes and lane closure information are obtained 12

from project drawings. NJDOT crash records provide seven different light conditions, however, for the 13

sake of simplicity; these are categorized into two levels as daytime and nighttime. Estimated project 14

duration is specified in terms of days. NJ Straight line diagrams provide posted speed limit for normal 15

conditions. Work zone speed limits are gathered from NJ crash records according to the distribution of 16

2007

2008

2009

2010

2011

Milepost

Date

31.0 31.4 31.8 32.2 32.6 33.0 33.4 33.8 34.2 34.6

Adv ance Warning

( 24 % )

Transition

( 0.5 % )

Activ ity Area

( 74 % )

Termination

( 1.5 % )AADT ( 33899 )

Non-WZ Crashes

WZ Crashes

Intersection

Ramp



posted speed limits within work zones. Work zone speed reduction and lane drop parameters are 1

generated by estimating difference between work zone and normal conditions. 2

Road types are categorized in two levels: Interstate and state highways. Number of lanes, number 3

of ramps and intersection for each work zones are obtained from NJ straight line diagrams. The duration-4

based model is developed by using 120 work zone crash count data for entire duration. Period-based PDO 5

and injury models are developed based on 950 work zone crash count data within 3 monthly seasonal 6

periods. Since work zones represent temporary conditions on the roadway and each work zone has its 7

own construction schedule, it is not possible to aggregate crashes over a longer time period. Thus, a three 8

month period is selected as the shortest appropriate duration. 9

10

MODEL SPECIFICATION 11 There are many statistical methods employed by previous related studies to model crash frequencies 12

including Poisson model, zero inflated negative binomial and Poisson, truncated regression, generalized 13

additive model, Conway Maxwell Poisson model, and negative multinomial model etc. (24). Crash counts 14

are non-negative integers that are affected by several factors. Elias and Herbsman (8) found best fitted 15

distributions for crash counts as negative binomial and Poisson. To model such crash count data, negative 16

binomial modeling (Poisson-Gamma) is favored since it allows over dispersion caused by other variables 17

(7, 9). Accordingly, crash frequency is modeled by using negative binomial method, where the dependent 18

variable is the total work zone crash counts for a given period. 19

Let Yi represents the number of crashes at work zone i for an exact length and duration, crash 20

occurrence for work zone i is independent and probability density can be Poisson distributed (9,27); 21

22

(1) 23

24

Where, is observed number of crashes and is the expected crash frequency for the work 25

zone . represents explanatory variables such as duration, length and AADT. Mean and variance values 26

of Yi are equal to mean and variance of . Eq. (2) describes this statement, where is the estimated 27

coefficient and is the value of explanatory variables. Overdispersion is included by error term , which 28

represents a random effect due to omitted explanatory variables and unmeasured heterogeneity. 29

30

(2) 31

32

In the negative binomial model, is assumed as gamma distributed with mean 1 and 33

variance (9, 28). Natural form of overdispersion is; 34

35

(3) 36

37

If we insert instead of , dispersion rate will be; 38

39

(4) 40

41

The negative binomial model differentiates from the Poisson model by the parameter , which is 42

necessary to determine overdispersion. If is significantly different than zero, the negative binomial 43

model is appropriate to use with such a count data, else Poisson model can be used. Safety performance 44

function (SPF) for predicting the number of work zone crashes in an interval of given length and duration 45

can thus be built as follows; 46

(5) 47

48



Where is the expected number of crashes within work zone; represents the work zone length 1

for the whole period of construction, is the traffic volume during the period of study Di represents 2

duration of the work zone. represents other explanatory variables and represents the coefficients for 3

model parameters. 4

Based on the crash information available from NJDOT crash database, project files and straight 5

line diagrams, following variables are selected for crash frequency modeling explained in Table 3. These 6

variables are length, light conditions, adjusted traffic volume, posted speed, speed reduction, the number 7

of operated lanes, the number of lane closures, road type, the number of ramps, and the number of 8

intersections within the work zone. Natural log values are used for duration, traffic and work zone length 9

parameters. Based on these parameters, the duration-based and period-based models are as follows: 10

11

Duration-based model; 12

(7) 13

14

Period-based model; 15

(8) 16

17

TABLE 3 Variables Considered in the Negative Binomial Model 18

Variable Symbol Type Description

Length length continuous length of the work zone (mile)

Light Conditions Night indicator daytime = 0, nighttime = 1

Traffic Volume traffic continuous adjusted ADT per lane (by direction,

seasonal factor, time factor)

WZ Speed wzspeed continuous reduced posted speed limit (mph)

Operated Lane operatedlane indicator number of operating lanes

Lane Drop lanedrop continuous number of closed lanes

Speed Reduction speedreduction continuous reduction in posted speed limit (mph)

Road System roadsystem indicator Interstate = 0, state = 1

Ramp Number ramp continuous number of ramps

Intersection intersection Continuous number of intersections

Duration duration continuous duration of the work zone (days)

19

MODEL ESTIMATION RESULTS 20 The estimated parameters shown in Table 4, Table 5 and Table 6 are used to determine the relationship 21

between independent variables and the frequency of work zone crashes. The results of the duration based 22

and period based models are presented at the 95 percent level of significance. 23

According to the duration-based work zone crash frequency modeling results shown in Table 4, 24

the project duration is found to be as the most significant parameter affecting the occurrence of crashes. 25

Similar results were found in previous studies for the duration of the work zones (7, 9). Work zone length 26

and traffic volume within the construction site are also found to be significant for the duration-based 27

model. For one percent increase in duration, traffic volume and length variables crash frequency increases 28

by 0.71 percent, 0.51 percent and 0.48 percent, respectively. Traffic parameter which represents the 29

volume per lane in conjunction with the number of operating lanes captures the effect of exposure. 30



TABLE 4 Estimated Parameters of Duration-Based Crash Frequency Model (N=120) 1

Variable Coefficient Std. Err. Z P>|z| Significant

ln(length) 0.477 0.133 3.580 0.000 ***

night -0.080 0.227 -0.350 0.725

ln(traffic) 0.512 0.158 3.230 0.001 ***

wzspeed -0.023 0.014 -1.600 0.110

operatedlane 0.642 0.141 4.540 0.000 ***

lanedrop 0.158 0.129 1.230 0.220

speedreduction 0.025 0.013 2.000 0.045 *

roadsystem 0.203 0.193 1.050 0.292

ramp 0.011 0.028 0.410 0.680

intersection 0.014 0.010 1.380 0.167

ln(duration) 0.710 0.084 8.500 0.000 ***

intercept -6.731 1.097 -6.140 0.000 ***

alpha 0.190 0.030

chibar2 = 614.89 Prob>=chibar2 = 0.000

Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

2

Work zone PDO crash counts for three month periods are modeled using the negative binomial 3

regression. The period-based PDO crash frequency model results are shown in Table 5. Results depict 4

that traffic volume has significant incremental effect on PDO crashes. One percent raise in traffic volume 5

causes 0.45 percent increase in total number of work zone PDO crashes. Additional operating lane causes 6

78.8 percent increase in total PDO crashes since traffic volume is estimated per lane. Similarly, lengths' 7

marginal effect of one percent increase causes 0.48 percent more PDO crashes. Different than normal 8

crash modeling parameters, lane drop and speed reduction are also found to have a significant effect on 9

the number of PDO crashes. One additional lane drop causes 28.8 percent increase in PDO crashes when 10

other parameters are kept constant. Since we used reduced volume for nighttime conditions, interpretation 11

is not biased due to the less traffic during nighttime. Considering volume adjustment, overall crash counts 12

at night are decreased by 26.9 percent when compared to daytime condition. If the work zone is on state 13

highways, 60.4 percent more PDO crashes are expected based on the modeling results. One more ramp 14

within the work zone section causes 4.3 percent more PDO crashes for the construction site. Proportional 15

increase in PDO crashes is found to be higher than injury crashes. 16

According to the period-based injury crash frequency modeling results at 95 percent level of 17

significance, length of the work zone is one of the most important variables for estimating the injury 18

crashes when compared to the PDO crashes (Table 6). One percent increase in project length causes 0.64 19

percent increase in injury crash counts. Traffic volume and number of operating lanes which represent the 20

exposure to traffic are both significant. Ullman et al. (29) stated that nighttime does not cause 21

significantly greater risk for crash severity than daytime for motorists at work zone condition. Similarly, 22

nighttime injury crash occurrence probability at work zone is found to be 31.8 percent less than daytime 23

work zone. Lane drop is another efficient factor to increase injury crash frequency at work zones. One 24

more lane drop causes 48.1 percent more injury crashes at the work zone site. Srinivasan et al. (30) 25

reported in their study that total crashes were increased by 66.9 percent for daytime and 60.1 percent for 26

night time compared to normal traffic conditions by temporary lane closure conditions at work zones. 27

28

29

30



TABLE 5 Estimated Parameters of Period-Based PDO Crash Frequency Model (N=950) 1

Variable Coefficient Std. Err. z P>|z| Significant

ln(length) 0.476 0.118 4.04 0.000 ***

night -0.314 0.152 -2.07 0.039 *

ln(traffic) 0.446 0.110 4.06 0.000 ***

wzspeed -0.011 0.009 -1.22 0.223

operatedlane 0.581 0.099 5.87 0.000 ***

lanedrop 0.253 0.084 3.00 0.003 **

speedreduction 0.036 0.010 3.46 0.001 ***

roadsystem 0.473 0.138 3.42 0.001 ***

ramp 0.042 0.020 2.15 0.032 *

intersection 0.013 0.007 1.84 0.066 .

intercept -4.648 0.876 -5.30 0.000 ***

alpha 0.501 0.040


Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

2

TABLE 6 Estimated Parameters of Period-Based Injury Crash Frequency Model (N=950) 3

Variable Coefficient Std. Err. z P>|z| Significant

ln(length) 0.642 0.153 4.20 0.000 ***

night -0.381 0.189 -2.02 0.043 *

ln(traffic) 0.325 0.137 2.36 0.018 *

wzspeed -0.016 0.011 -1.46 0.143

operatedlane 0.590 0.126 4.69 0.000 ***

lanedrop 0.393 0.102 3.84 0.000 ***

speedreduction 0.024 0.014 1.81 0.071 .

roadsystem 0.959 0.184 5.21 0.000 ***

ramp 0.023 0.024 0.96 0.337

intersection -0.001 0.009 -0.16 0.873

intercept -4.752 1.138 -4.18 0.000 ***

alpha 0.489 0.061


Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

4

CONCLUSIONS 5 In this study, the duration-based and the period-based safety performance functions are developed to 6

identify risk factors that affect crash frequency in work zones. Negative binomial regression is found to 7

be an appropriate approach to model count data because mean and variances differ significantly for the 8

data set used in this study. Crash database, work zone project drawings and NJ straight line diagrams are 9

used to estimate parameters for frequency models. The dataset used in safety performance function 10

models consists of 60 construction work zones on NJ interstate and state highways. Project drawings and 11



crash database provide satisfactory information for analyzing work zone crashes. By identifying the 1

effects of work zone length, daily traffic exposure, and other explanatory variables on the frequency 2

models, this study provides considerable information to transportation agencies. 3

According to the estimation results, “project duration”, “length of the work zone”, and “traffic 4

volume” are found to be the most important factors that increase work zone crash frequency. This result is 5

consistent with previous studies (6-11). By considering adjusted traffic volume based on hourly 6

distribution, nighttime PDO and injury crash occurrence probability is found to be 26.9 and 31.8 percent 7

lower than daytime condition, respectively. Additionally, lane closure is found to be an increasing factor 8

for number of PDO and injury work zone crashes. Speed variance between posted and work zone speed 9

limits are observed to increase PDO crashes. Number of ramps and intersections are found to be 10

important factors that increase crash frequency as well. 11

Based on findings from the modeling results, following recommendations could be useful to 12

increase safety within work zones: 13

“Project duration” could be minimized by including incentives and disincentives within 14

contracts , an approach adopted some states (31), 15

To avoid exposure resulting from “heavy volume”, traffic could be diverted to alternate 16

routes or work schedule could be switched to off-peak and weekend periods when 17

appropriate conditions exist (32), 18

Nighttime schedule for work zones can also be recommended to reduce injury and PDO 19

work zone crashes, 20

The variance between the posted and work zone speed limits should be optimized to 21

prevent increase of rear-end crash occurrence. A sharp reduction in speed limits may 22

cause more risk within work zones, and 23

Lane closure strategies should be revised to minimize the number of lane drops. If a lane 24

closure is necessary, countermeasures such as an intelligent lane merging system (33) can 25

be implemented to allow a smoother transition within the work zone to reduce crash 26

frequency. 27

28

ACKNOWLEDGEMENTS 29 The contents of this paper reflect views of the authors who are responsible for the facts and accuracy of 30

the data presented herein. This research is partially supported by NJDOT and also fellowship from 31

Turkish Government. The contents of the paper do not necessarily reflect the official views or policies of 32

the agencies. 33

34

REFERENCES 35 1. FHWA. Facts and Statistics Work Zone Injuries and Fatalities 36

www.ops.fhwa.dot.gov/wz/resources/facts_stats/injuries_fatalities.htm. Accessed May 15, 2012. 37

2. Schrank, D., T. Lomax, and B. Eisele. TTI’s 2011 Urban Mobility Report Powered by INRIX 38

Traffic Data. Texas Transportation Institute, 2011. 39

http://tti.tamu.edu/documents/mobility-report-2011.pdf Accessed May 18 2012. 40

3. FY 2013 - 2022 Statewide Capital Investment Strategy. New Jersey Department of 41

Transportation. http://www.state.nj.us/transportation/capital/cis/pdf/scis1322.pdf Accessed May 42

13 2012. 43

4. Hartgen, D. T., R.K. Karanam, M. G. Fields, T. A. Kerscher. 19th Annual Report on the 44

Performance of State Highway Systems (1984-2008). Report No. Policy Study 385. 45

Transportation Research Board of the National Academies, Washington, D.C., 2010. 46

5. Ozbay, K., H. Yang, O. Ozturk, M. Yildirimoglu, S. Demiroluk, and B. Bartin. Work Zone Safety 47

Analysis. Working report. New Jersey Department of Transportation, New Jersey. 48



6. Pal, R. and K.C. Sinha. Analysis of Crash Rates at Interstate Work Zones in Indiana. 1

Transportation Research Record: Journal of the Transportation Research Board, No. 1529, 2

Transportation Research Board of the National Academies, Washington, D.C., 1996, pp. 43–53. 3

7. Venugopal, S. and A. Tarko. Safety Models for Rural Freeway Work Zones", Transportation 4

Research Record: Journal of the Transportation Research Board, No. 1715, Transportation 5

Research Board of the National Academies, Washington, D.C., 2000, pp. 1-9. 6

8. Elias, A.M. and Z.J. Herbsman. Risk Analysis Techniques for Safety Evaluation of Highway 7

Work Zones. Transportation Research Record: Journal of the Transportation Research Board, 8

No. 1715, Transportation Research Board of the National Academies, Washington, D.C., 2000, 9

pp. 10–17. 10

9. Khattak, A. J. and F. M. Council. Effects of Work Zone Presence on Injury and Non-Injury 11

Crashes. Accident Analysis and Prevention, Vol. 34, 2002, pp. 19-29. 12

10. Qi, Y., R. Srinivasan, H. Teng, and R.F. Baker. Frequency of Work Zone Accidents on 13

Construction Projects. Report No: 55657-03-15. Region 2, University Transportation Research 14

Center, City College of New York, New York, 2005. 15

11. Srinivasan, S., G. Carrick, X. Zhu, K. Heaslip, and S. Washburn. Analysis of Crashes in Freeway 16

Work Zone Queues: A Case Study. Final Report Research No: 07-UF-R-S3. Transportation 17

Research Center, University of Florida, Florida, 2008. 18

12. Juergens, W.R. Construction Zone, Detour and Temporary Connection Accidents. Business and 19

Transportation Agency, California Division of Highways, 1972. 20

13. Graham, J., R. Paulsen, and J. Glennon. Accident and Speed Studies in Construction Zones. 21

Report No. FHWA-RD-77-80. FHWA, U.S. Department of Transportation, 1977. 22

14. Rouphail, N.M., Z.S. Yang, and J. Fazio. Comparative Study of Short- and Long-Term Urban 23

Freeway Work Zones. Transportation Research Record: Journal of the Transportation Research 24

Board, No. 1163, Transportation Research Board of the National Academies, Washington D.C., 25

1988, pp. 4-14. 26

15. Hall, J.W. and V.M. Lorenz. Characteristics of Construction-Zone Accidents. Transportation 27


Research Board of the National Academies, Washington D.C., 1989, pp. 20-27. 29

16. Garber, N. J. and T. H. Woo. Accident Characteristics at Construction and Maintenance Zones in 30

Urban Areas. Research Report VTRC 90 – R12, Virginia Transportation Research Council, 31

Charlottesville, Virginia, 1990. 32

17. Pigman, J.G. and K.R. Agent. Highway Accidents in Construction and Maintenance Work Zone. 33

Transportation Research Record: Journal of the Transportation Research Board, No. 1270. 34

Transportation Research Board of the National Academies, Washington D.C., 1990, pp. 12-21. 35

18. Ha, T. and Z. Nemeth. Detailed Study of Accident Experience in Construction and Maintenance 36

Zones. Transportation Research Record: Journal of the Transportation Research Board, No. 37

1509, Transportation Research Board of the National Academies, Washington, D.C., 1995, pp. 38

38-45. 39

19. Garber, N.J. and M. Zhao. Distribution and Characteristics of Crashes at Different Work Zone 40

Locations in Virginia. Transportation Research Record: Journal of the Transportation Research 41

Board, No. 1794, Transportation Research Board of the National Academies, Washington, D.C., 42

2002, pp. 19–25. 43

20. Benekohal, R., E. Shim, and P.T.V. Resende. Truck Drivers' Concerns in Work Zones: Travel 44

Characteristics and Accident Experiences. Transportation Research Record: Journal of the 45

Transportation Research Board, No.1509, Transportation Research Board of the National 46

Academies, Washington D.C., 1995, pp. 55-64. 47

21. Wang, J., W. E. Hughes, F.M. Council, and J.E. Paniati. Investigation of Highway Work Zone 48

Crashes: What We Know and What We Don’t Know. Transportation Research Record: Journal 49

of the Transportation Research Board, No.1529, Transportation Research Board of the National 50

Academies, Washington D.C., 1996, pp. 54-64. 51



22. Jin, T. G., M. Saito, and D. L. Eggett. Statistical Comparisons of the Crash Characteristics on 1

Highways between Construction Time and Non-Construction Time. Accident Analysis and 2

Prevention, Vol.40, 2008, pp. 2015-2023. 3

23. Dissanayake, S. and S.R. Akepati. Characteristics of Work Zone Crashes in the Swzdi Region: 4

Differences and Similarities. Proceedings of the 2009 Mid-Continent Transportation Research 5

Symposium, Ames, Iowa, 2009. 6

24. Lord, D. and F. Mannering. The statistical analysis of crash-frequency data: A Review and 7

Assessment of Methodological Alternatives. Transportation Research Part A, 2010, Vol. 44, pp. 8

291–305. 9

25. Bourne, J.S., T.A. Scriba, C. Eng, R.D. Lipps, , G.L. Ullman, , D.L. Markow, K.C. Markow, D. 10

Gomez, D.L. Holstein, B. Zimmerman, and R. Stargell Best Practices in Work Zone Assessment, 11

Data Collection, and Performance Evaluation. Scan Team Report (Scan 08-04), National 12

Cooperative Highway Research Program (NCHRP) Project 20-68A U.S. Domestic Scan, 2010. 13

26. State of New Jersey, Department of Transportation, Plans of Route I-78 Local and Express 14

Roadway Improvements, Contract No. 050003731, 2006. 15

27. Greene, W. Econometric Analysis, Third edition. Macmillan, New York, 1997. 16

28. Cameron, C.A. and P. K. Trivedi. Regression Analysis of Count Data. Econometric Society 17

Monograph No. 30. Cambridge University Press, Cambridge, 1998. 18

29. Ullman, G.L, M.D. Finley, J.E. Bryden, R. Srinivasan, and F.M. Council. Traffic Safety 19

Evaluation of Nighttime and Daytime Work Zones. NCHRP Report No. 627, Transportation 20

Research Board of the National Academies, Washington D.C., 2008. 21

30. Srinivasan, R., G.L. Ullman, M.D. Finley, and F.M. Council. Use of Empirical Bayesian Methods 22

to Estimate Temporal-Based Crash Modification Factors For Construction Zones. Transportation 23


Research Board of the National Academies, Washington, D.C., 2011, pp. 29-38 25

31. Zhu, Y., I. Ahmad, and L. Wang. "Estimating work zone road user cost for alternative contracting 26

methods in highway construction projects." Journal of Construction Engineering and 27

Management, Vol. 135 (7), 2009, pp. 601–608. 28

32. Ullman, G.L., V. Iragavarapu, and D. Sun. Work Zone Positive Protection Guidelines. Report No. 29

FHWA/TX-11/0-6163-1, Texas Transportation Institute, 2010. 30

33. Pei, Y. and L. Dai. Study on Intelligent Lane Merge Control System for Freeway Work Zones. 1. 31

Intelligent Transportation Systems Conference. IEEE, Seattle, WA, 2007, pp. 586-591. 32


trb 2013 annual meeting paper revised from original...

Documents