Effects of the Open Road Tolling on Safety Performance of ...

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

Effects of the Open Road Tolling on Safety Performance of Freeway Mainline 1

Toll Plazas 2 3 4 Hong Yang (Corresponding Author) 5 Ph. D. Candidate, 6 Rutgers Intelligent Transportation Systems (RITS) Laboratory, 7 Department of Civil and Environmental Engineering, 8 Rutgers, The State University of New Jersey, 9 623 Bowser Rd. Piscataway, NJ 08854 USA, 10 Tel: (732) 445-0576 x119 11 Fax: (732) 445-0577 12 E-mail: [email protected] 13 14 Kaan Ozbay, Ph. D. 15 Professor & Director, 16 Rutgers Intelligent Transportation Systems (RITS) Laboratory, 17 Department of Civil and Environmental Engineering, 18 Rutgers, The State University of New Jersey, 19 623 Bowser Rd. Piscataway, NJ 08854 USA, 20 Tel: (732) 445-2792 21 Fax: (732) 445-0577 22 E-mail: [email protected] 23 24 Bekir Bartin, Ph. D. 25 Research Associate, 26 Rutgers Intelligent Transportation Systems (RITS) Laboratory, 27 Department of Civil and Environmental Engineering, 28 Rutgers, The State University of New Jersey, 29 623 Bowser Rd. Piscataway, NJ 08854 USA, 30 Tel: (732) 445-3162 31 Fax: (732) 445-0577 32 E-mail: [email protected] 33 34 35 36 Abstract: 263 37 Word count: 4712 Text + 7 Tables + 4 Figures = 7462 38 Submission Date: August 1, 2011 39 Resubmission Date: November 15, 2011 40

41 42

43 Paper submitted for Presentation and Publication in the 44

Transportation Research Record, Journal of Transportation Research Board after being presented 45 Transportation Research Board’s 91st Annual Meeting, Washington, D.C., 2012 46

TRB 2012 Annual Meeting Paper revised from original submittal.Downloaded from amonline.trb.org

Yang, Ozbay and Bartin 2

ABSTRACT 1 Advances of Intelligent Transportation Systems (ITS) technologies promoted the implementation 2

of open road tolling (ORT) on tolled freeways worldwide. This new tolling solution converts 3 existing barrier tollbooths to express lanes capable of collecting tolls at high-speeds. ORT has 4

demonstrated numerous benefits in reducing traffic congestion and air pollution. However, 5 effects of ORT on safety are still not clear, as most of ORT systems have only been operated for 6

a relatively short period of time. Therefore, this study aims to evaluate the safety impacts of ORT 7 by studying locations where such tolling solution was recently deployed on the Garden State 8

Parkway in New Jersey. Multiple-year crash data at the toll plazas before-after the 9 implementation of the ORT systems were used for analysis. Full Bayes methodology is 10

employed to estimate crash frequency models as a function of traffic and toll plaza 11 configurations. These models were used to estimate the crash frequency assuming that the ORT 12

systems were not installed. Then, these estimations were compared with the observed number of 13 crashes occurred after the deployment of the ORT systems. Individual comparisons show that 14

crash reductions are observed at most of the toll plazas. The overall comparison shows that 15 crashes at locations where ORT systems were deployed are decreased by about 24 percent after 16

deployment of these systems. It can thus be concluded that the use of ORT is a beneficial 17 solution towards improving toll road safety. From an implementation point of view, the analyses 18

results indicate that special attention should be paid to operational elements such as signage, 19 diversion and merge designs of the ORT systems. 20



INTRODUCTION 1 Open road tolling (ORT) is a new generation of tolling solution that will eventually lead to a 2 conversion of conventional toll plazas to barrier-free electronic toll collections in the future. By 3 design, ORT consisting of high-speed (express) electronic toll collection (ETC) lanes allows 4 vehicles to electronically and automatically pay tolls without slowing down from highway 5 speeds. Typically, there are two types of implementation of ORT (1). The first type is the all-6 electronic ORT which completely replaces barrier tollbooths by express ETC lanes (2, 3). 7 Without the presence of tollbooths, this type of ORT design enables automatic debiting via in-8 vehicle transponders (e.g., E-ZPass tags) or other automatic vehicle identification (AVI) 9 technologies. The second is the interim type of ORT implementation, which is being deployed by 10 many toll authorities. It installs express ETC lanes by retrofitting existing tollbooths to permit 11 high-speed non-stop toll collection for ETC users and other registered users only (1, 4). Cash or 12 coin users are still diverted to use remaining barrier booths off the express ETC lanes. 13

In the United States, many highway authorities in states like New Jersey, Florida and 14 Illinois have implemented the ORT concept in recent years (e.g., 5-7). Demonstration projects 15 have shown that the implementation of ORT is an effective means of relieving congestion. For 16 example, Klodzinski et al. (1) evaluated the addition of ORT to a mainline toll plaza in Florida 17 and found that installation of express ETC lanes reduced delays by 49.8 percent for cash users 18 and by 55.3 percent for automatic coin machine (ACM) users. According to Levinson and 19 Odlyzko (8), express ETC lanes of ORT can increase throughput, from 350 to 400 vehicles per 20 hour per lane with manual collection up to 2200 vehicles per hour per lane. The use of ORT has 21 also been shown to significantly reduce emissions. For instance, Lin and Yu (9) quantified 22 various ORT deployment scenarios on Illinois toll highways and suggested that the near roadside 23 carbon monoxide concentration levels and diesel particulate matter emissions can be reduced by 24 up to 37 percent and 58 percent, respectively. 25

While ORT can sharply reduce transaction-related delays and pollution at toll plazas, 26 safety impacts of retrofitting existing toll plazas and installing express ETC lanes, however, is 27 still not so clear. ORT systems can be deemed safer since high-speed tolls avoid the safety 28 deficiency of barrier toll plazas as they eliminate many stop-and-go traffic, dangerous 29 interactions and distractions (10). On the other hand, diverging and merging of vehicles that use 30 express ETC lanes at higher speeds might increase traffic conflicts (e.g., 11, 12). Cash and coin 31 users must exit to use the barrier tollbooths and then merge with high-speed users on express 32 lanes. These maneuvers may raise more safety issues. Unlike those easily measurable benefits 33 such as capacity improvement and reduction of costs in toll collection (e.g., 13-16), it is difficult 34 to evaluate safety performance of the new tolling solution shortly after its implementation 35 because of the random and rare occurrence of motor vehicle crashes. Quantifying safety 36 performance of ORT requires long-term crash data collected at toll plazas with and without the 37 deployment of ORT. 38

This study aims to evaluate the impact of implementing ORT on crash rates at toll plazas 39 by using extensive data collected at multiple mainline toll plazas with and without the 40 deployment of express ETC lanes on Garden State Parkway in New Jersey. The crash data 41 available in this study cover the period from January 2001 to December 2009. Therefore, safety 42 performance of toll plazas with and without express ETC lanes can be analyzed and compared 43 using these multiple-year crash data (17). 44



OPEN ROAD TOLLING IN NEW JERSEY 1 Gardens State Parkway (GSP) is a 172.4-mile limited-access toll parkway with 359 exits and 2 entrances. Over 380 million vehicles travel the GSP which stretches the length of New Jersey 3 (NJ) from the New York (NY) state line at Montvale to Cape May at the southern tip of the state. 4 Tolls are collected at 50 locations, including 11 mainline toll plazas and 39 on entrance and exit 5 ramps (18). It is among America’s busiest highways, serving users from NJ and NY’s most 6 marketable communities (19, 20). 7

The GSP operator (GSP was operated by the NJ Highway Authority (NJHA) until 2003 8 and later by the NJ Turnpike Authority (NJTA)) always focused on using new toll collection 9 technologies to improve tolling efficiency and reduce congestion. After the first toll collected 10 manually in 1954, automatic coin machines (ACM) were introduced in early 1950s and had 11 spread to most toll plazas and ramps on the Parkway by 1959 (21). When a toll increased beyond 12 the quarter, tokens were introduced in the early 1980s. They continued to be available until 13 January 2002 and ceased to be accepted in payment in January 2009 (21). Regular (low-speed) 14 ETC system was implemented in 1999. The entire ETC system was completed in August 2000 15 (20). The ETC system has been widely adopted by travelers with an ETC penetration rate beyond 16 70 percent on GSP (22). 17

In 2001, the state government issued an order to promote a 10-year congestion relief plan 18 for GSP (13). Under the plan, elimination of mainline barriers in one direction and use of express 19 ETC lanes of ORT in the other were recommended. By 2010, all mainline barriers except Toms 20 River were converted to one-way tolling (express ETC lanes were added to both directions at the 21 Toms River toll plaza). Between 2004 and 2006, the open road tolling program was implemented. 22 Express ETC lanes of ORT have been installed at a number of toll plazas listed in TABLE 1. 23

TABLE 1 Open Road Tolling Operation at the Mainline Toll Plazas on GSP 24 Toll Plaza Milepost ORT Operation Date No. of Express ETC Lanes

Cape May NB 19.4 May 2006 2 Toms River NB 84.7 May 2005 2 Toms River SB 84.7 May 2005 2 Asbury Park NB 104.0 May 2005 3

Raritan SB 125.4 May 2005 5 Pascack Valley NB 166.1 January 2004 2 Pascack Valley SB 166.1 January 2004 2

25 Installation of express ETC lanes at the Pascack Valley toll plaza in January 2004 marks 26

a major milestone of ORT deployment on GSP. FIGURE 1 shows the progression of toll 27 collection at the toll plaza. The conventional toll plaza with seven tollbooths in one direction was 28 retrofitted to an interim version of ORT system, which allows ETC drivers to drive at 55mph 29 through the two high-speed lanes in each direction. Followed this demonstration project, this 30 type of ORT system has been successfully deployed at several other sites listed in TABLE 1. 31 FIGURE 2 shows an example of the layout of the converted plaza. Signs are installed upstream 32 of the toll plaza to guide the selection of tollbooths. Upon the operation of such ORT system, 33 electronic readers mounted on the gantry automatically charge the ETC vehicles. Meanwhile, 34 overhead cameras capture the license plates of vehicles without transponders (e.g., E-ZPass tag). 35 Cash and coin users are diverted to use barrier tollbooths. The new ORT system has been widely 36 accepted as over 90 percent E-ZPass vehicles use the high-speed lanes on GSP. It is estimated 37 that express ETC lane can process about 800 more vehicles per hour than traditional ETC lanes 38 (10). Compared to the observed benefits such as capacity improvement and reduction of costs in 39



toll collection (e.g., 10, 13 & 14), little information about the safety impact of the deployed ORT 1 system is available. 2

3

4

5

6 FIGURE 1 Toll collection evolution at Pascack Valley toll plaza (Source: 10, 14) 7 8

9 FIGURE 2 Separation between barrier tollbooths and express E-ZPass lanes at Cape May 10

Northbound toll plaza (Source: Google Map) 11

(a) Toll Plaza Configuration in 1960s

(b) Toll Plaza with Traditional ETC Lanes

(c) Toll Plaza with Express ETC Lanes



SAFETY PERFORMANCE MODELING 1 Data Description 2 To investigate crash characteristics at various toll plaza locations, data from a number of 3 resources were collected. The toll plaza configurations and monthly average daily traffic (MADT) 4 data for GSP toll plazas between January 2001 and December 2009 were obtained from the New 5 Jersey Turnpike Authority (NJTA). The corresponding crash records for toll areas are obtained 6 from the raw crash database of the New Jersey Department of Transportation (NJDOT). The 7 records provide detailed information on each crash such as the time, location, collision type and 8 severity etc. In order to capture the impact of toll plaza, it is necessary to limit the toll plaza 9 analysis to certain section. Generally, crash data are analyzed from sections of 0.5 miles or 10 longer (e.g., 23, 24). Considering the signposting distances and physical lengths of the toll plazas 11 on GSP, toll plaza major impact area is assumed to be one mile which covers 0.5 miles before 12 and after the tollbooths, respectively. Crashes occurred within the impact area of each toll plaza 13 were extracted from the raw crash records. To avoid the influence of construction, data of one-14 year before the operation of the ORT systems were excluded from analysis. TABLE 2 presents 15 the mean and standard deviation (SD) of MADT and the corresponding crash counts at the toll 16 plazas where express ETC lanes were implemented. TABLE 3 summarizes similar information 17 for other barrier toll plazas which do not have open road tolling. 18 19

TABLE 2 Statistical Summary of MADT and Monthly Crash Counts (Mean ±SD) 20 Index Dataset 1: without ORT Dataset 2: with ORT

Toll Plaza Time Period MADT (×1000) Crash Time Period MADT (×1000) Crash

16 Cape May NB 01/01-04/05 15.66 ± 6.20 0.52 ± 0.75 05/06-12/09 16.01 ± 6.26 0.86 ± 0.90 17 Toms River NB 01/01-04/04 45.70 ± 6.75 3.23 ± 1.73 05/05-12/09 48.80 ± 6.47 2.11 ± 2.19 18 Toms River SB 01/01-04/04 43.56 ± 6.27 3.88 ± 2.89 05/05-12/09 45.52 ± 6.05 2.34 ± 1.95 19 Asbury Park NB 01/01-04/04 76.50 ± 9.72 5.40 ± 2.99 05/05-12/09 78.16 ± 10.06 7.18 ± 4.09 20 Raritan SB 01/01-04/04 114.90 ± 12.59 8.23 ± 3.53 05/05-12/09 111.87 ± 12.23 6.05 ± 3.28 21 Pascack Valley NB 01/01-12/02 41.35 ± 3.01 0.88 ± 0.99 01/04-12/09 42.68 ± 2.94 1.10 ± 1.02 22 Pascack Valley SB 01/01-12/02 41.00 ± 2.70 2.42 ± 1.61 01/04-12/09 42.20 ± 2.84 1.36 ± 1.18

21 TABLE 3 MADT and Monthly Crash Counts for Other Barrier Toll Plazas (Mean ±SD) 22

Index Toll Plaza Milepost Available Time Period MADT (×1000) Crash 1 Great Egg SB 28.8 01/01-12/09 20.66 ± 6.68 1.17 ± 1.36

2 New Gretna NB 53.5 01/01-12/09 22.71 ± 3.70 0.98 ± 0.98

3 Barnegat SB 68.9 01/01-12/09 34.60 ± 5.15 2.87 ± 2.14

4 Union NB 142.7 01/01-12/09 99.62 ± 6.32 12.18 ± 4.75

5 Essex SB 150.7 01/01-12/09 76.87 ± 4.79 7.72 ± 3.46

6 Bergen NB 160.5 01/01-12/09 70.02 ± 4.35 5.45 ± 2.43

7 Cape May SB 19.4 01/01-12/05 16.13 ± 6.44 0.57 ± 0.79

8 Great Egg NB 28.8 01/01-12/05 20.34 ± 6.74 0.72 ± 1.03

9 New Gretna SB 53.5 01/01-12/05 22.57 ± 3.74 1.37 ± 0.99

10 Barnegat NB 68.9 01/01-02/07 35.29 ± 5.31 2.09 ± 1.85

11 Asbury Park SB 104 01/01-08/04 76.95 ± 11.34 4.86 ± 2.43

12 Raritan NB 125.4 01/01-08/04 118.72 ± 13.33 11.32 ± 4.44

13 Union SB 142.7 01/01-02/05 102.80 ± 6.91 15.52 ± 5.24

14 Essex NB 150.7 01/01-06/05 82.29 ± 5.11 6.13 ± 2.61

15 Bergen SB 160.5 01/01-11/05 68.16 ± 4.20 4.27 ± 2.38



As seen from TABLE 2, the highest number of monthly crash counts is observed at the 1 Raritan SB toll plaza when the ORT was not deployed. Its traffic volume is the highest, as well. 2 After ORT was deployed, the mean crash frequency decreased by about 26.5 percent (from 8.23 3 per month to 6.05 per month). Crash reductions are also observed at both directions of Toms 4 River toll plazas and Pascack Valley southbound toll plaza despite of their increasing MADTs. 5 The remaining three sites seem to have higher crash risk after ORT systems were deployed. 6

However, the preliminary analysis in TABLE 2 cannot clarify the true effectiveness of 7 deploying express ETC lanes at the toll plazas on GSP because the change of crash frequencies 8 seems to be affected by the change in MADT. More importantly, this naïve before-after 9 comparison is subject to problems such as regression-to-the-mean, crash migration, maturation, 10 and external causal factors (e.g., 25-27). Thus, more rigorous evaluation methodologies are 11 needed to address the safety effects of the ORT systems at those toll plazas. Next section 12 presents the modeling and estimation methods used in this study. 13 14 Crash Modeling Using Poisson Mixture Models 15 To estimate the safety impacts of the ORT systems, the crash reduction rate (CRR) at the thi toll 16 plaza is calculated as follows: 17

1 1

Observed Crashes with ORT1 1Expected Crashes without ORT

i iN N

it itt t

CRR Y ��

� � � �� (1)

where itY is the observed crash frequency at the thi toll plaza in the tht time interval (month) after 18

the express ETC lanes were installed. it� is the corresponding expected number of crashes at 19

the thi toll plaza in the tht month if the express ETC lanes have not been installed. iN is the total 20

number of months after the express ETC lanes were deployed at the thi toll plaza. 21 To calculate CRR, the key step is to obtain the expected number of crashes it� through 22

some analytical models. The Poisson regression model is a natural first choice for modeling the 23 longitudinal crash data because it can capture the more desirable statistical characteristics of 24 traffic crashes that are rare, random, nonnegative, discrete, and typically sporadic (28). However, 25 past research has indicated that the crash counts are likely to be over-dispersed (e.g., 29, 30). To 26 deal with the problems of over-dispersion, accounting for site-specific attributes and other 27 complex variations, etc, many extensions of simple Poisson models have been suggested. Two of 28 the frequently used alternatives are Poisson-Gamma (PG) mixture model and Poisson-Lognormal 29 (PL) mixture model. These models have the following model structures: 30

Assuming ity denotes the number of independently observed crashes at a given site i in 31

the tht time interval. Assuming ity follows the Poisson distribution with the mean of it� , we have 32

the basic Poisson regression model: 33 ( ) ( 1, 2,..., ; 1, 2,..., )it ity Poisson i n t T� � � (2)

Let it� describe the systematic variation of the expected number of crashes it� . it� is 34

specified as a function of a set of explanatory variables. The commonly adopted function in crash 35 analysis is the log-linear model: 36

log( ) Tit iX� �� (3)

where iX is a vector of the explanatory variables such as geometric attributes, environment 37

conditions, etc. � is a k dimensional vector of unknown coefficients and ~ (0, )Normal �� ; 38

�� is the k k� dimensional variance-covariance vector and the non-diagonal elements are set to 39



zero 2( 0, )ij

i j� � � and diagonal elements are set to large positive values 1 2( . . 1000, 1, 2,..., ).

iii e i k� � � 2

To address issues of over-dispersion for unobserved heterogeneity, the noisy 3 measurement it� is introduced into it� : 4

log( ) log( )it it it� � �� (4)exp( )it it it� � �� (5)

The term exp( )it� in equation (5) denotes a multiplicative random effect (31). For 5

Poisson-Gamma model, it usually assumes that exp( )it� is independent and follows Gamma 6

distribution for all sites i and t . Similarly, Poisson-Lognormal model assumes that the noisy 7 structure exp( )it� is Lognormal distributed. 8

1 2exp( ) ( , )it Gamma� � � (PG model) (6)2 2 2exp( ) ~ (0, ) | (0, )it itLognormal or Normal� � �� (PL model) (7)

where 1� and 2� are the shape parameter and reciprocal of a scale parameter of Gamma 9

distribution, respectively; the Gamma distribution has 1 2/mean � �� and 21 2/variance � �� . By 10

specifying 1 2� � �� , the Poisson–Gamma model becomes Negative-Binomial (NB) model; � is 11

the inverse dispersion parameter and 0� � . 2� denotes extra Poisson variance among observations. 12

Under the Poisson-Gamma (or NB) model, the expected number of crashes and its variance are: 13 ( ) exp( )T

i it it iE Y X� � �� (8)2( ) ( ) [ ( )] /i i iVar Y E Y E Y �� (9)

Similarly, the expected number of crashes and its variance under the Poisson-Lognormal 14 model are follows: 15

2( ) (exp( )) exp(0.5 )i it it it itE Y E �� (10)2 2( ) ( ) [ ( )] [exp( ) 1]i i iVar Y E Y E Y �� (11)

The variance shown in equations (9) and (11) is always larger than the mean. Thus, these 16 models are more appropriate for modeling the over-dispersed crash data. 17

Moreover, hierarchical Bayes version of the PG model and PL model can be obtained by 18 imposing extra hyper-priors for the parameters to consider more stages of randomness. TABLE 4 19 summarizes the two model structures used in this study. 20

21 TABLE 4 Hierarchical Bayes Models for Crash Frequency at the Toll Plazas on GSP 22

Poisson-Gamma Poisson-Lognormal exp( )it it it� � �� exp( )it it it� � ��

exp( ) ( , )it Gamma� � � 2exp( ) ~ (0, )it Lognormal �� 1 2( , )Gamma� � � 2

1 2( , )Gamma� � �� 1 ~ ( )Exponential a� 1 ~ ( )Exponential a�

2 ( , )Gamma b c� 2 ( , )Gamma b c� 23 Data listed in TABLE 3 are combined with the dataset 1 in TABLE 2 to develop the link 24

function of it� and estimate the parameters. Specifically, the structure of the link function is 25 shown in equation (12). The explanatory variables are summarized in TABLE 5. 26

27



0 1 2 3

4 5 6 7

log( ) log( ) log( ) log( )it MADT AL DLACM CASH ETC S

� � � � ��

� � � �

� � � �

(12)

1 TABLE 5 Model Covariates and Their Corresponding Statistical Summary 2

Covariates Symbol Description Mean ± SD Min Max Number of observed crashes Y Crashes per month 4.67 ± 4.86 0 30 Monthly average daily traffic MADT Thousand vehicles per day 54.80 ±32.75 8.86 143.79 Length of approach zone AL Mile 0.21 ± 0.06 0.12 0.40 Length of departure zone DL Mile 0.21 ± 0.06 0.12 0.4 Number of ACM tollbooths ACM Tollbooths primarily use ACM 3.60 ± 1.41 2 7 Number of cash tollbooths CASH Tollbooths primarily use cash 1.71 ± 0.75 1 3 Number of regular ETC lanes ETC Tollbooths primarily use low-speed ETC 3.57 ± 2.08 1 11 Season S Winter(Month=12,1,2), Spring(Month=3,4,5)

Summer(Month=6,7,8), Fall (Month=9,10,11) 2.48 ± 1.12 1 4

3 Finally, the model parameters shown in TABLE 4 and the parameters of equation (12) 4

are estimated in a Full Bayes context via Markov Chain Monte Carlo (MCMC) sampling in 5 WinBUGS (32). The methodology has been widely used in road safety studies (e.g., 31, 33 & 6 34). Non-informative prior distributions are imposed on hyper-parameters in TABLE 4 to reflect 7 the lack of prior information. Convergence of each model parameter is identified by monitoring 8 MCMC trace plots and the Brooks-Gelman-Rubin (BGR) convergence statistic set in WinBUGS. 9 Deviance Information Criterion (DIC) is used to compare different models (35, 36). To assess 10 the accuracy of the posterior estimates for each parameter, the ratio of the Monte Carlo error 11 relative to the standard deviation of each parameter is suggested to be less than 0.05 (37). 12

13 RESULTS & DISCUSSION 14 The posterior estimates of model parameters were obtained from the Full Bayes analysis based 15 on 10,000 posterior samples via WinBUGS. The estimation results are summarized in TABLE 6. 16 The convergence of each model parameter was verified as the BGR convergence statistics were 17 below 1.2. The ratio of the Monte Carlo error relative to the standard deviation of each parameter 18 is about 0.01 to 0.08. The two models were compared by computing the DIC as presented by 19 Spiegelhalter et al. (35). The computed DIC statistics were 5674 and 5678 under the Poisson-20 Gamma model, Poisson-Lognormal model, respectively. As a rule of thumb, the smaller the DIC 21 the better the model fit, and a difference larger than 10 suggests in favor of the model with 22 smaller DIC. However, if the difference in DIC is less than 5, and the models make very 23 different inferences, then it might be misleading to report the model with the lowest DIC (35, 36). 24 Using the DIC values in TABLE 6, we see that the DIC difference in this study is less than 5. 25 Thus, the results of both models are reported. 26



TABLE 6 Model Parameter Estimates and 95-Percent Bayesian Credible Intervals (CI) 1 Parameter Description Poisson-Gamma Poisson-Lognormal

0� Constant -7.531 (-7.937, -7.050) -7.621 (-8.106, -7.119)

1� Monthly average daily traffic 1.599 (1.501, 1.706) 1.617 (1.551, 1.693)

2� Length of approach zone -0.795(-0.933, -0.672) -0.798 (-0.956, -0.644)

3� Length of departure zone -0.564 (-0.674, -0.380) -0.569 (-0.719, -0.425)

4� Number of ACM tollbooths 0.035 (0.007, 0.060) 0.031 (0.006, 0.057)

5� Number of cash tollbooths 0.068(0.000, 0.123) 0.059 (0.006, 0.110)

6� Number of regular ETC lanes 0.038 (0.016, 0.058) 0.037 (0.015, 0.062)

7� Season 0.013 (-0.015, 0.042) 0.011 (-0.019, 0.038) � Gamma distribution parameter 11.490 (8.961, 14.820) ---

2� Variance of normal distribution --- 0.087 (0.067, 0.111)

1� Hyper-parameter 1 1.846 (0.234, 5.123) 1.857 (0.226, 5.106)

2� Hyper-parameter 2 0.230 (0.020, 0.670) 0.229 (0.018, 0.658) DIC Deviance Information Criterion 5674 5678

2 � From TABLE 6 we note that all parameter estimates except 7� are significantly different 3

from zero as their 95-percent Bayesian credible intervals do not cover zero under both 4 models. The estimates of � for Poisson-Gamma model and 2

� for Poisson-Lognormal are 5

significant, demonstrating the presence of over-dispersion in the crash data. 6 � As shown in TABLE 6, crash occurrence rate is expected to increase with daily traffic. 7

This result is consistent with the general finding in traffic safety studies as more vehicles 8 are likely to cause more complex interactions among vehicles and therefore more crashes 9 (31). 10

� The length of the approach zone and the departure zone at a toll plaza can significantly 11 influence the crash occurrence rate. Their corresponding coefficients 2� and 3� have 12

negative signs indicating that the increase in length of these zones reduces crash 13 likelihood. As expected, when the length of these sections increases, the total amount of 14 conflicts between vehicles may be reduced because drivers have more time to make 15 critical decisions (e.g., lane-changing and merging) at the toll plazas. 16

� The positive coefficients of 4� and 5� indicate that the number of manual payment 17

tollbooths is positively associated with the number of crashes at toll plazas. It can be 18 attributed to the fact that more cash and coin tollbooths result in more stop-to-go traffic. 19

� Similar to the findings reported in a previous study (38), the positive coefficient of 20

6� suggests that the implementation of regular (low-speed) ETC system is also likely to 21 increase the likelihood of the crash occurrence. An explanation for this is that drivers are 22 more likely to be confused to locate appropriate lane for payment. 23

� The variable 7� which turned out to be statistically not significant suggests that there are 24 no distinct seasonal differences in terms of toll plaza safety at these locations. 25 26 TABLE 7 presents the evaluation results in terms of crash reduction rate (CRR) for 27

implementing ORT at the mainline toll plazas on Garden State Parkway. The estimated results 28 under both models show a reduction in crash occurrence rate at most of the toll plazas after 29 installing express ETC systems. The highest crash reduction rate is observed at Raritan SB toll 30 plaza, where the estimated CRR is more than 40 percent. This toll plaza is the busiest one among 31 other toll plazas. Before the deployment of express ETC lanes it used to have 20 tollbooths and 32



served more than 100,000 vehicles per day. Five express ETC lanes were added to allow vehicles 1 with transponders (e.g., E-ZPass tags) to travel at 55 mph. This is the largest ORT system on the 2 Garden State Parkway. The decrease among Toms River and Pascack Valley toll plazas ranges 3 from 17 percent to 36 percent. The composite results are also provided in TABLE 7. Overall, 4 motor vehicle crashes are decreased by about 24 percent at the toll plazas with a standard 5 deviation of 2.4 percent. The overall results support the findings of a previous study (1) that ORT 6 can cause a significant reduction in the number of crashes. 7

8 TABLE 7 Estimates of Crash Reduction Rates for Implementing ORT at Toll Plazas 9

Poisson-Gamma Poisson-Lognormal Toll Plaza Observed Crashes

with ORT Estimated Crashes without ORT CRR (95% CI) Estimated Crashes

without ORT CRR (95% CI)

Cape May NB 38 22 -0.759 (-1.049, -0.499) 21 -0.781 (-1.053, -0.537) Toms River NB 118 184 0.358 (0.283, 0.427) 186 0.363 (0.288, 0.436) Toms River SB 131 200 0.343 (0.274, 0.406) 200 0.342 (0.272, 0.407) Asbury Park NB 402 392 -0.031 (-0.199, 0.121) 386 -0.045 (-0.194, 0.091)

Raritan SB 339 571 0.404 (0.322, 0.474) 568 0.401 (0.319, 0.475) Pascack Valley NB 79 114 0.303 (0.222, 0.385) 115 0.306 (0.204, 0.393) Pascack Valley SB 98 119 0.174 (0.080, 0.275) 120 0.178 (0.056, 0.282)

Overall 1205 1602 0.247 (0.198, 0.294) 1596 0.244 (0.196, 0.290) 10

Despite the overall reduction, it should be noted that the number of crashes is observed to 11 have increased after the implementation of the ORT systems at Cape May NB and Asbury Park 12 NB toll plazas. The estimated results under both models suggest that the number of crashes at 13 Asbury Park NB toll Plaza increased about 3-4 percent after three express ETC lanes were added. 14 Whereas this small number of increase does not suggest a clear impact of the ORT system as its 15 95-percent credible interval covers zero. Cape May NB toll plaza has the lowest number of crash 16 counts among all toll plazas. However, it shows more than 70 percent increase of crash 17 frequency after the ORT system was installed at this toll plaza. 18

A more careful review of the reported crashes at these two toll plaza was conducted. 19 FIGURE 3 shows the spatial distribution of the reported crashes before and after the ORT system 20 was deployed. As seen from FIGURE 3a and FIGURE 3c, more than 40 percent of crashes 21 occurred at the approach section of the tollbooths during 2001 to 2004 when the ORT system 22 was not installed yet. Most of these crashes are rear end, side swipe and fixed object collisions. 23 In contrast, FIGURE 3b and FIGURE 3d illustrate the observed crashes during 2006 to 2009 24 after the ORT system was in operation. Interestingly, only about 10 to 20 percent of crashes 25 occurred at the immediate vicinity of the tollbooths at both toll plazas. Notably, more crashes 26 occurred at the sections before the ORT signs (FIGURE 3b) and the merge section (FIGURE 3d). 27 The major crash types are found to be rear end, side swipe and fixed object collisions at these 28 sections. As a comparison, we also reviewed the reported crashes at other toll plazas that have 29 significant crash reductions after the ORT system was implemented. FIGURE 4 shows an 30 example of the crash distributions of the Raritan SB toll plaza. The crash distributions at this toll 31 plaza did not vary significantly before and after the implementation of the ORT system. No 32 obvious changes in crash frequencies were found around diverge and merge sections. 33

Vehicles have to make a decision whether to use the express lanes or to divert to barrier 34 tollbooths when approaching the ORT signs. After leaving the barrier tollbooths, the slow 35 moving vehicles have to merge with the high-speed vehicles coming from the express ETC lanes, 36 which might lead to higher number of traffic conflicts. FIGURE 3 and FIGURE 4 suggest that 37



when retrofitting an existing toll plaza to add ORT system, attention should be paid to factors 1 such as signage, diversion and merge designs. 2

3

Milepost

Cra

sh F

requ

ency

(%

)0

1020

3040

5060

19 19.1 19.3 19.5 19.7 19.9

Tollbooths (Milepost: 19.4)

(a) Observ ed Crash Frequency without ORT (2001-2004)(Cape May NB)

MilepostC

rash

Fre

quen

cy (

%)

010

2030

4050

60

19 19.1 19.3 19.5 19.7 19.9


(b) Observ ed Crash Frequency with ORT (2006-2009)(Cape May NB)

ORT Signs Start to merge

Milepost

Cra

sh F

requ

ency

(%

)0

1020

3040

5060

103.6 103.8 104 104.2 104.4

Tollbooths (Milepost: 104)

(c) Observ ed Crash Frequency without ORT (2001-2004)(Asbury Park NB)

Milepost

Cra

sh F

requ

ency

(%

)0

1020

3040

5060

103.6 103.8 104 104.2 104.4

Tollbooths (Milepost: 104)

(d) Observ ed Crash Frequency with ORT (2006-2009)(Asbury Park NB)

ORT Signs Start to merge

4 FIGURE 3 Crash distributions at Cape May NB & Asbury Park NB toll plazas 5

6

Milepost

Cra

sh F

requency (

%

01

020

30

40

50

60

125 125.2 125.4 125.6 125.8


(a) Observ ed Crash Frequency without ORT (2001-2004)(Raritan SB)

Milepost

Cra

sh F

requency (

%)

01

020

30

40

50

60

125 125.2 125.4 125.6 125.8


(b) Observ ed Crash Frequency with ORT (2006-2009)(Raritan SB)

ORT SignsStart to merge

7 FIGURE 4 Crash distributions at Raritan SB toll plaza before and after deploying ORT 8

9 10 CONCLUSIONS 11



This study aims to evaluate the impact of the open road tolling (ORT) on safety performance of 1 the mainline toll plazas. Garden State Parkway (GSP) in New Jersey was used as the case study. 2

Seven toll plazas where the ORT systems were deployed were used in the analysis. Crash data, 3

related traffic data and toll plaza configurations between 2001 and 2009 were used to support the 4 analysis. Crash modeling based on Full Bayes methodology was conducted to determine the 5

crash reduction rate (CRR) of the ORT system. 6 The safety performance functions developed via Full Bayes methodology show that the 7

number of crashes increase with daily traffic volume and the number of barrier tollbooths. The 8 results also suggest that the number of regular (low-speed) ETC toll lanes might also lead to 9

increased likelihood of crash occurrence just as cash and ACM toll lanes do. An explanation for 10 this is that drivers are more likely to be confused when selecting appropriate toll lane. On the 11

other hand, increasing the length of toll plaza arrival and departure zones is likely to reduce the 12 number of crashes. As expected, when the length of these zones increases, the total amount of 13

conflicts between vehicles may reduce because drivers have more time to make critical decisions 14 (e.g., lane-change and merge) along approach and departure sections. 15

The safety evaluations show that the overall crash frequency at the toll plazas on GSP 16 was reduced by a crash reduction rate (CRR) of 24 percent after the deployment of ORT system 17

at toll plazas. The hypothesis that the ORT with express ETC lanes reduces the crash rates at toll 18 plazas on GSP and positively impacts the safety is supported by the analysis results. Individually, 19

five of the toll plazas show a decrease in crash occurrence rate, with a reduction rate ranging 20 from 17 to 40 percent. Raritan SB toll plaza had the highest reduction of 40 percent. In contrast, 21 crash model estimated that the other two toll plazas Cape May NB and Asbury Park NB have 22

more number of crashes after the deployment of ORT. A detailed review of the crash occurred 23 along these two toll plazas was conducted. The results suggest that less crashes occurred at the 24

immediate vicinity of the tollbooths at both toll plazas after the ORT systems were installed, 25 whereas high number of crashes are likely to occur at the diversion and merging sections. As 26

vehicles without transponders (e.g., E-ZPass tags) have to leave the mainline to use barrier 27 tollbooths and then merge back, lane changes and speed differential among vehicles increase at 28

these sections. This increase may be the reason of the higher likelihood of rear end and side 29 swipe collisions observed in the crash data. However, more research is needed for better 30

understanding this assumption. 31 The analyses results indicate that the implementation of the ORT system improved safety 32

on most of the toll plazas on GSP. In addition, when deploying ORT systems, special attention 33 should be paid to some critical components such as signage guidance, diversion and merge 34

designs. 35 36

ACKNOWLEDGEMENTS 37 The contents of this paper reflect the views of the authors who are responsible for the facts and 38 the accuracy of the data presented herein. The contents of the paper do not necessarily reflect the 39 official views or policies of the agencies. 40

REFERENCES 41



1. Klodzinski, J., Gordin, E., and Al-Deek, H. M. (2007) Evaluation of Impacts from 1 Deployment of an Open Road Tolling Concept for a Mainline Toll Plaza. Transportation 2 Research Record: Journal of the Transportation Research Board, No. 2012, pp. 72-83. 3

2. Stewart, S. (1999). Electronic Tolling and Traffic Management (ETTM): The 407 ETR-4 the Experience, the Opportunity and a Concept. IEE Seminar on Electronic Tolling and 5 Congestion Charging, London, UK, 25 March. 6

3. Mendoza, E., Gold, M., Carter, P., and Parmar, J. (1999). The Sale of Highway 407 7 Express Toll Route: A Case Study. The Journal of Structured Finance, Vol. 5, No. 3, pp. 8 5-14. 9

4. Klodzinski, J., and Al-Deek, H. M. (2004). Evaluation of Toll Plaza Performance after 10 Addition of Express Toll Lanes at Mainline Toll Plaza. Transportation Research Record: 11 Journal of the Transportation Research Board, No. 1867, pp. 107-115. 12

5. Muriello, M. F., and Jiji, D. (2004). Value Pricing Toll Program at Port Authority of New 13 York and New Jersey- Revenue for Transportation Investment and Incentives for Traffic 14 Management. Transportation Research Record: Journal of the Transportation Research 15 Board, No. 1864, pp. 9-15. 16

6. Suarez, E., and Hoeflich, K. (2005). The End of the Tollbooth? Public Roads, Volume 68, 17 Issue 6, pp.64-70. 18

7. Kovacs, P., and Abou-Sabh, S. (2008). The Implementation of Open Road Tolling on the 19 Illinois Tollway. Tollways, December 2008-Asset Management, pp. 92-101. 20

8. Levinson, D., and Odlyzko, A. (2008). Too Expensive To Meter: The Influence of 21 Transaction Costs in Transportation and Communication. Philosophical Transactions of 22 The Royal Society, A, Volume 366, No. 1872, pp. 2033-2046. 23

9. Lin, J., and Yu, D. (2008). Traffic-Related Air Quality Assessment for Open Road 24 Tolling Highway Facility. Journal of Environmental Management, Volume 88, Issue 4, 25 pp. 962-969. 26

10. Siegel, J., Jon Orcutt, J., and Slevin, K. (2004). The Open Road - The Region’s Coming 27 Toll Collection Revolution. Report, Tri-State Transportation Campaign. 28

11. Benda, J., Hochmuth, J., Kowshik, R., Medgyesy, L. (2009). Open Road Tolling Signing 29 Studies. In TRB 88th Annual Meeting Compendium of Papers CD-ROM. Transportation 30 Research Board, Washington D.C. 31

12. Tri-State Transportation Campaign (2000). Non-Stop Parkway Bill Absent in NJ Senate. 32 Mobilizing the Region, Issue 258 (http://www.tstc.org/bulletin/20000225/mtr25805.htm). 33

13. Weinstein, J. (2001). Garden State Parkway Congestion Relief Plan - A Proposed 34 Framework. Department of Transportation, New Jersey. 35

14. Raczynski, R., and Finn, J. (2005). Garden State Parkway Plaza Improvement Program. 36 Presented at IBTTA Facilities Management Workshop, May 14-18, Toronto, Ontario. 37

15. Ciszewski, S., Hubicki, D., and Tang, W. (2005). Garden State Parkway Raritan South 38 Express Lanes. Presented at IBTTA 73rd Annual Meeting & Exhibition, September 17-39 21, Cleveland, Ohio. 40

16. Ciszewski, S. (2004). Express Lanes and One Way Tolls. Presented at IBTTA Fall 41 Technology Workshop, November 13-16, Madrid, Spain. 42

17. NJDOT (2011). http://www.state.nj.us/transportation/refdata/accident/. Accessed on 43 March 16, 2011. 44

18. NJTA (2011a). http://www.state.nj.us/turnpike/who-we-are.html. Accessed on May 12, 45 2011. 46



19. Travelers’ Market (2011). Garden State Parkway. Available online at: 1 http://www.travelersmarketing.com/NewJersey.htm. Accessed on July 10, 2011. 2

20. Currie, J., and Walker, R. (2011). Traffic Congestion and Infant Health: Evidence from 3 E-ZPass. American Economic Journal: Applied Economics, Vol. 3, No. 1, pp. 65-90. 4

21. Toll Road News (2008). Garden State Parkway Tokens Won't Be Accepted from Jan 1- 5 Cash Going Too? http://www.tollroadsnews.com/node/3818. Accessed on 05/15/11. 6

22. NJTA (2011b). Financial Summary for the Five Months Ended May 31, 2011. 7 http://www.state.nj.us/turnpike/documents/MonthlyFinancialMay2011.pdf. Accessed on 8 May 19, 2011. 9

23. Resende, P. T. V. and Benekohal, R. F. (1997). Effects of Roadway Section Length on 10 Accident Modeling. Proceedings of the Conference-Traffic Congestion and Traffic Safety 11 in the 21st Century: Challenges, Innovations, and Opportunities, Chicago, IL, June 8-11, 12 pp. 403-409. 13

24. Abdel-Aty, M., Devarasetty, P. C., Pande, A. (2009). Safety Evaluation of Multilane 14 Arterials in Florida. Accident Analysis & Prevention, Volume 41, Issue 4, pp. 777-788. 15

25. Hauer, E. (1997). Observational Before-After Studies in Road Safety: Estimating the 16 Effect of Highway and Traffic Engineering Measures on Road Safety. Pergamon Press, 17 Elsevier Science, Ltd., Oxford, United Kingdom. 18

26. Elvik, R. (2002). The Importance of Confounding in Observational Before-and-After 19 Studies of Road Safety Measures. Accident Analysis & Prevention, Volume 34, Issue 5, 20 pp. 631-635. 21

27. Shen, J., and Gan, A. (2003). Development of Crash Reduction Factors: Methods, 22 Problems, and Research Needs. Transportation Research Record: Journal of the 23 Transportation Research Board, No. 1840, pp. 50-56. 24

28. Miaou, S. and Lum, H. (1993). Modeling Vehicle Accidents and Highway Geometric 25 Design Relationships. Accident Analysis & Prevention, Volume 25, Issue 6, pp. 689-709. 26

29. Hauer, E. (2001). Overdispersion in Modelling Accidents on Road Sections and in 27 Empirical Bayes Estimation Accident Analysis & Prevention, Volume 33, Issue 6, pp. 28 799-808. 29

30. Lord, D., Washington, S.P., and Ivan, J.N. (2005). Poisson, Poisson-Gamma and Zero 30 Inflated Regression Models of Motor Vehicle Crashes: Balancing Statistical Fit and 31 Theory. Accident Analysis & Prevention, Volume 37, Issue 1, pp. 35-46. 32

31. El-Basyouny, K., and Sayed, T. (2009). Accident Prediction Models with Random 33 Corridor Parameters. Accident Analysis & Prevention, Volume 41, Issue 5, pp. 1118-34 1123. 35

32. Lunn, D., Tomas, A., Best, N., and Spiegelhalter, D. (2000). WinBUGS – A Bayesian 36 Modelling Framework: Concepts, Structure, and Extensibility. Statistics and Computing, 37 Vol. 10, No. 4, pp. 325-337. 38

33. Lan, B., Persaud, B., Lyon, C., and Bhim, R. (2009). Validation of a Full Bayes 39 Methodology for Observational Before-After Road Safety Studies and Application to 40 Evaluation of Rural Signal Conversions. Accident Analysis & Prevention, Volume 41, 41 Issue 3, pp. 574-580. 42

34. Yanmaz-Tuzel, O., and Ozbay, K. (2010). A Comparative Full Bayesian Before-and-43 After Analysis and Application to Urban Road Safety Countermeasures in New Jersey. 44 Accident Analysis & Prevention, Volume 42, Issue 6, pp. 2099-2107. 45



35. Spiegelhalter, D., Best, N., Carlin, B. P., Van der Linde, A. (2002). Bayesian Measures of 1 Model Complexity and Fit. Journal of the Royal Statistical Society. Series B (Statistical 2 Methodology), Vol. 64, No. 4, pp. 583-639. 3

36. Gelman, A., Meng, X. L., and Stern, H. (1996). Posterior Predictive Assessment of 4 Model Fitness via Realized Discrepancies (with discussion). Statistica Sinica, Vol. 6, pp. 5 733-807. 6

37. Burnham, Kenneth P. and Anderson, David R. Model Selection and Multimodel 7 Inference: A Practical Information - Theoretic Approach, 2nd ed.. Springer: New York, 8 2002. 9

38. Mohamed, A., Abdel-Aty, M., and Klodzinski, J. (2001). Safety Considerations in 10 Designing Electronic Toll Plazas: Case Study. ITE Journal, pp. 20-24. 11


Effects of the Open Road Tolling on Safety Performance of ...

Documents

Transcript of Effects of the Open Road Tolling on Safety Performance of ...