Accident Analysis and Prevention 80 (2015) 57–66

Contents lists available at ScienceDirect

Accident Analysis and Prevention journal homepage: www.elsevier.com/locate/aap

Enforcement avoidance behavior near automated speed enforcement areas in Korean expressways Jisup Shim 1,a , Shin Hyoung Park 2,b , Sungbong Chung 3,c, Kitae Jang a, * a The Cho Chun Shik Graduate School for Green Transportation, Korea Advanced Institute of Science and Technology, 291 Daehak-ro, Yuseong-gu, Daejeon 305-701, South Korea b Department of Transportation Engineering, College of Engineering, Keimyung University, 1095 Dalgubeol-daero, Dalseo-gu, Daegu 704-701, South Korea c Graduate School of Railroads, Seoul National University of Science & Technology, 232 Gongneung-ro, Nowon-gu, Seoul 139-743, South Korea

A R T I C L E I N F O

A B S T R A C T

Article history: Received 23 October 2014 Received in revised form 20 March 2015 Accepted 27 March 2015 Available online xxx

Automated speed enforcement system (ASES) has been deployed as a safety countermeasure on Korean roadways to reduce speeding-related traffic crashes; information on ASES locations is mandated to be open to the public. However, because drivers are alerted about enforcement via on-board navigation systems and roadside signs, they can avoid enforcement by momentarily reducing their speeds near ASES locations. This enforcement avoidance behavior (EAB) can induce sudden changes in speed near the enforcement locations and thereby increase risk of crash occurrence. In light of this situation, the present study evaluates the effects of ASES on traffic behavior and safety. An analysis of traffic data shows that drivers indeed diminish their speeds near enforcement locations, and accelerate shortly after passing the locations. To investigate how this behavior affects safety performance, this study, by using Empirical Bayes analysis with comparison groups, compares crash occurrences along a certain section before and after the installation of ASES. The comparative analysis shows that overall crash occurrence dropped by 7.6% on average near the enforcement locations, although the reduction was not significant. However, an average 11% non-significant increase in crash occurrence is also observed in the upstream segment, where enforcement is announced to drivers and traffic starts to diminish speed. The findings suggest that the sudden changes in traffic speed induced by EAB substantially negate the benefits of ASES. Therefore, modification of the design of current ASES is required to mitigate EAB and further improve the effectiveness of ASES. ã 2015 Elsevier Ltd. All rights reserved.

Keywords: Automated speed enforcement Enforcement avoidance behavior Empirical Bayesian analysis Safety

1. Introduction Speeding is a behavior of driving at excessive speed and a major contributing factor to crash occurrence and its consequences, as speeding shortens time for drivers to react to events and increases the force of any impact. Crash data collected from different locations have shown that speeding increases the risk of crash occurrence (Maycock et al., 1998; Gambard et al., 1997) and that it has a direct influence on the severity of injuries (Nilsson, 1982;

* Corresponding author. Tel.: +82 42 350 1264; fax: +82 42 350 1250. E-mail addresses: [email protected] (J. Shim), [email protected] (S.H. Park), [email protected] (S. Chung), [email protected] (K. Jang). 1 Tel.: +82 42 350 1284; fax: +82 42 350 1250. 2 Tel.: +82 53 580 5285; fax: +82 53 580 5259. 3 Tel.: +82 10 8978 9682; fax: +82 2 975 6876. http://dx.doi.org/10.1016/j.aap.2015.03.037 0001-4575/ ã 2015 Elsevier Ltd. All rights reserved.

Elvik et al., 2004; Evans, 2004; Aarts and Schagen, 2006; Park et al., 2012). Since expressways are designed to facilitate faster travel, drivers on expressways are exposed to an environment in which speed is generally higher than that on other roadways. Statistics on traffic crashes on Korean expressways also indicate that speeding is one of the leading causes of traffic injuries and fatalities. In Korean expressways, a total of 8513 crashes occurred between 2007 and 2010, resulting in 197 fatalities and 1097 injuries. Among these crashes, 21% were caused by speeding, which accounts for 17.1% and 17.4% of fatalities and of injuries, respectively. To reduce injuries and fatalities caused by speeding, a simple but effective countermeasure has been enforcement of speed limits by police officers. Despite the effectiveness of police enforcement, broader deployment of manual enforcement measures has been impeded by limited resources and safety concerns for police officers. Thus, the automated speed enforcement system

58

J. Shim et al. / Accident Analysis and Prevention 80 (2015) 57–66

Fig. 1. Illustration of an ASES in Korea.

(ASES) was introduced on Korean expressways in 1997 as a more efficient speed enforcement measure. The ASES automatically measures vehicle speeds and photographs the plate numbers of vehicles that exceed the speed limit (See Fig. 1). Since this system is fully automated and cost-effective, ASESs are considered as a primary countermeasure for locations where speeding-related crashes frequently occur. As of 2013, ASESs have been deployed at more than 5200 locations. To further discourage drivers from speeding, information on enforcement locations is mandated by law to be open to the public so that drivers can be informed of the locations via navigation services, which are widely used in Korea. In response to this information, however, drivers tend to reduce their speeds momentarily only near enforcement locations, as they seek to avoid a penalty. Safety concerns with regard to this type of Enforcement Avoidance Behavior (EAB, often called the kangaroo effect) arise because sudden changes in speed may increase the risk of crash occurrence, especially upstream of the enforcement locations. In this study, using both cross-sectional and longitudinal approaches, we analyze a detailed set of traffic data to understand the attributes of this type of EAB and evaluate its impact on the safety performance of ASESs.

(i.e., drivers were without information on enforcement locations), whereas cameras were fully exposed for the other group (i.e., drivers had information on enforcement locations). The comparative analysis showed that drivers maintained their speeds below the speed limit on a longer section when the information on the enforcement location was not given. Other studies on ASES have postulated that, to avoid a penalty, drivers may momentarily reduce their speeds only at the enforcement locations and that this EAB can increase the risk of crash occurrences, specifically as drivers suddenly change their speeds (Decina et al., 2007; Elvik, 1997; ARRB Group, 2005; Christie et al., 2003; Thomas et al., 2008; Newstead and Cameron, 2003). De Pauw et al. (2014a) analyzed traffic speed data collected near speed enforcement cameras and showed that EAB occurred in the vicinity of speed cameras. Although identified as a potential risk factor, the impact of this behavior on traffic safety was not evaluated. In this study, we use vehicle trajectories and loop detector data to understand the attributes of EAB; we also examine crash data collected near ASESs to evaluate the safety performance of the ASES. 3. Enforcement avoidance behavior

2. Literature review 3.1. Descriptions of the study sites and data Extensive research has shown that ASESs have a positive effect on traffic safety because they reduce vehicle speeds near enforcement areas (ETSC, 1999; ARRB Group, 2005). Some studies have found that ASESs enhance traffic safety not only in the immediate vicinity of enforcement locations but also along the extended section associated with the system (Chen et al., 2000; Goldenbeld and van Schagen, 2005). However, these studies were conducted based on crash data collected from sites for which the location of the ASES was undisclosed. Therefore, as drivers did not know precisely where the enforcement locations were they tended to maintain their speeds below the speed limit throughout the section to avoid a penalty. This implies that the positive effect of the ASES is attenuated if drivers are aware of the enforcement locations, as in such cases they can avoid speeding tickets by reducing their speeds only at the ASES and not throughout the section. Evidence from a comparative study by Keall et al. (2001) confirms this implication. Keall et al. (2001) used hidden and visible cameras for ASESs to evaluate the effects of revealed location information of these systems. In the experiment, drivers were informed that speeding was enforced by ASESs somewhere along the roadways, but cameras were hidden for one group

This section analyzes detailed traffic data collected from GPSequipped taxis and inductive loop detectors to observe how ASESs affect the trend of vehicle movements. The study sites were selected based on data availability. Trajectory data were obtained from the driving records of 259 taxis in Daegu, South Korea during all days in May of 2013. These data include the vehicles' speeds and locations in terms of longitude and latitude collected every second. For site selection, we used a geographic information system (GIS) to spatially match the trajectory data with the ASES locations; we selected ASES sites at which the largest number of vehicles passed. Four sites were selected and are marked by white circles in Fig. 2. Although trajectory data provide detailed information on vehicle movements, those trajectories were from a small subset of all passing vehicles, and were collected from limited locations and at various times. To supplement this, we analyzed two years (from 2010 to 2011) of five-minute aggregated traffic data including traffic volume, occupancy (detector occupancy is a dimensionless measure of density, which is the percentage of time that vehicles are above the detector), and average speed, as measured by inductive loop detectors along the Gyeongbu Expressway. To conduct a before-and-after study, we selected

J. Shim et al. / Accident Analysis and Prevention 80 (2015) 57–66

ASES locations based on the following conditions: (i) ASESs were installed during the two years in which detector data were available; and (ii) multiple detectors were installed in the proximity. Three sites were selected and are labeled with white rectangles in Fig. 2. 3.2. Analysis of trajectory data The location information of ASESs and the trajectories were spatially matched to query the data points collected near each ASES (within a 1.5-km radius of the ASES). Trajectory data from 27 taxis passing four ASESs were queried. The speed profiles of the trajectories over the section near one of the selected ASESs are

59

plotted in Fig. 3(a). A visual inspection of the speed profiles in the figure indicates that vehicle speeds were lower near the ASES, as evident from the narrower bandwidths in the close upstream of the ASES. To examine overall patterns of speed variation over the section, speed measurements from all four ASES sites were combined. The speed measurements were then partitioned into 100-m segments, and average speed and standard deviation were computed for each segment. Fig. 3(b) presents the speed profile over a 3-km section that encompasses the ASES. This figure shows that drivers start to reduce speeds from about 700-m upstream of the ASES, after which they start to accelerate when they have passed the ASES. Also, bandwidth representing a range of
1 standard deviation

Fig. 2. Study sites for different data analysis.

60

J. Shim et al. / Accident Analysis and Prevention 80 (2015) 57–66

indicates that the speed variance across vehicles became smaller as they approached the ASES, returning to its original value after the ASES. The observed patterns (vehicles diminish their speeds shortly before the ASES and accelerate immediately after they pass the ASES) confirm that EAB indeed occurs near ASESs. The next section analyzes two years of loop detector data to show that this behavior is reproducible across sites and days. 3.3. Analysis of loop detector data In this section, we perform two types of comparative analysis – cross-sectional and before-and-after – using aggregated traffic data from inductive loop detectors. To this end, two years (years 2010 and 2011) of loop detector data from the Gyeongbu Expressway were obtained from the Korean Expressway Corporation (Korea Expressway Corporation, 2013). For the cross-sectional comparison, we selected three ASESs that were located close to multiple-loop detectors that were in good condition. Because two of the three ASESs were installed in the middle of 2010, beforeand-after comparisons were also feasible. Table 1 summarizes the location information of the ASESs and of the loop detectors at the

selected sites. It also summarizes the types of analyses conducted in this study. For the comparative analysis, box-and-whisker plots (also known as box plots) were constructed to characterize the traffic behavior near the ASESs. These plots are used to represent summary statistics of the distribution. For a more quantitative representation of differences in speed, we indicated the mean and standard deviation of used data at the top of each box plot. In this analysis, we used traffic data from the median lane during off-peak hours (from 9 p.m. to 7 a.m. on the following day) to exclude data that may have been measured during constrained traffic conditions (For more details on the site conditions, please see Jang et al., 2014). 3.3.1. Cross-sectional analysis For each ASES site, box-and-whisker plots were constructed using three loop detectors in the vicinity of the ASES – the closest loop detector to observe the distribution of speed at the ASES and one detector each to observe the distribution of speed upstream and downstream of the ASES (Fig. 4). The top and bottom of each box shows first and third quartiles, respectively; and each vertical bar displays the minimum and the maximum. Since only data from

Fig. 3. (a) Speed profiles near an ASES area. (b) Mean and variance of vehicle speeds across four ASESs.

J. Shim et al. / Accident Analysis and Prevention 80 (2015) 57–66

61

Table 1 Loop detector locations by analysis method. Site

ASES location (post kilometer: PK)

Distance from the ASES (kilometer)

Cross-sectional analysis

Before–after analysis

1

358.90

Yes

Yes

2

380.90

Yes

Yes

3

413.50

Upstream At camera Downstream Upstream At camera Downstream Upstream At camera Downstream

Yes

No (system installed in 2008)

357.20 (1.70 km) 358.70 (0.20 km) 359.70 (+0.80 km) 380.19 (0.71 km) 380.74 (0.16 km) 382.45 (+1.55 km) 412.40 (1.10 km) 413.40 (0.10 km) 414.04 (+0.54 km)

unconstrained, free-flow traffic conditions were plotted, data points are distributed in a high-speed range near 110 km/h, which is the speed limit along this expressway section. In the figures, boxes and whiskers of data from the upstream and downstream detectors are plotted higher than those from the detector near the ASES. Using the data, the mean and standard deviation of the speed at each detector were computed and are noted on the tops of the box plots. Compared with the mean and standard deviation of the speed near the ASES, the means and standard deviations at the upstream and downstream detectors are greater. This pattern shows that vehicles reduce their speeds with smaller variance prior to an ASES and recover to higher speed after passing the ASES. 3.3.2. Before-and-after analysis At two of the selected ASESs, before-and-after comparisons were conducted to observe how traffic behavior changed after the installation of the ASES. Given that the ASESs were installed in the middle of 2010, traffic data from all days in January of 2010 (before) and in January of 2011 (after) were used. Box-and-whisker plots using the speed data are presented for the before and after periods in Fig. 5. The box plots from the after-period are located lower than those from the before-period, signifying that vehicle speeds became slower after the installation of the ASES. These patterns

suggest that vehicles would travel at a higher speed without the ASES (i.e., ASES indeed has the effect of reducing vehicle speeds). Both cross-sectional and before-and-after comparisons confirm that the ASESs induce speed reductions when drivers are near them. These temporary speed reductions are due to EAB, and may increase the risk of traffic crashes because increases of speed variance along a roadway increase the likelihood of contact between vehicles (Liu and Popoff, 1997; Lee et al., 2002). Intuitively, it can also be postulated that speed disturbances can create situations in which vehicles suddenly decelerate upstream of the ASES; if the following vehicles do not react in time, contact between vehicles can occur. Therefore, a potential negative effect on traffic safety arises due to the installation of an ASES. This concern underscores the need to evaluate the spatial variation of safety performance of ASESs. 4. Effects of ASESs on traffic safety This section examines the effectiveness of an ASES as a safety countermeasure by comparing crash occurrences in nearby sections before and after an ASES were installed. The effectiveness of the system is measured in terms of the reduction in the number of crashes that occurred after implementing the ASES. The simplest

Fig. 4. Box-whisker plots in the proximity of the ASES at: (a) site 1–358.9 k; (b) site 2–380.9 k; (c) site 3–413.5 k.

62

J. Shim et al. / Accident Analysis and Prevention 80 (2015) 57–66

information about all crashes that occurred during the period ^ ib ) using the following before the installation of the ASES (m equation: ^ ib ¼ wb  Pbi þ ð1  wb Þ  xbi m

(1)

In this equation, xbi is the observed number of crashes at site i and Pbi is the predicted number of crashes during the period before the installation of the ASES, which can be estimated using a crash prediction model. To construct the crash prediction model, we have used crash data from 2007 to 2010 and annual average daily traffic data from 1128 roadway segments of Korean expressways. The descriptive statistics of the data are provided in Table 2. The number of crashes per year for jth segment, lj, was specified as in Eq. (2):

lj ¼ lj  expðb0 þ b1  AADTj þ eij Þ

Fig. 5. Box-whisker plots before and after ASES installation at: (a) site 1–358.9 k; (b) site 2–380.9 k.

way to measure the effectiveness is to look at the ratio of the number of crashes during the period after the installation to that before the installation. Although this approach is easy to implement and to understand, it often results in biased estimates because it does not reflect the regression-to-the-mean (RTM), the effect of changes in traffic volume, or general changes in the number of crashes from before to after the countermeasure is installed. To correct for these possible biases, the empirical Bayes (EB) method with a comparison group was adopted from Hauer (1997) in this study (Section 4.1). Five years (from 2007 to 2011) of crash data were collected from all Korean expressways; 28 study sites at which an ASES was installed in 2009 were selected, as marked by white rhombuses in Fig. 2. These sites were selected because they had no special geometric features: e.g., these were sites that were not near rest areas, bridges, tunnels, or curved sections. Also, all types of crashes that occurred at the study sites were included in the analysis. This was done for two reasons. First, since ASESs are installed where speed-related crashes are common, most of the crashes that occurred at the sites were related to speeding (even if speeding was not identified as a primary collision factor). Second, speeding reduces reaction times for events in front and thus causes various types of crashes, including rear-end collisions and sideswipes (Shin et al., 2009; Washington et al., 2007; Decina et al., 2007). Using the outcomes from the EB analysis for the 28 sites, we conducted a meta-analysis to estimate the overall effectiveness of the ASES (Section 4.2). 4.1. Empirical Bayes (EB) method with comparison group For site i, the EB method was used to compare the number of crashes that occurred in the period after the installation of the ASES (mai ) with the expected number of crashes that would have occurred without the ASES. The latter can be estimated based on

(2)

where lj is the observed number of crashes per year for the jth segment; lj is the length of the jth segment; AADTj is the annual average daily traffic for the jth segment; and ej is the error for the jth segment. Negative binomial (NB) regression was used to estimate the coefficients, b0 and b1, via maximum likelihood estimation procedure. The estimated coefficients are provided in Table 3. All the coefficients are statistically significant, as evident from the low p-values. The over-dispersion parameter, a, is 0.183, indicating that the data are over-dispersed. Using the estimated model, we were able to compute all the input values for Eq. (1). To calculate the predicted number of crashes for each study site, AADTs for five 1-km segments near the ASES were input. The over-dispersion parameter, a, is used to compute the weight factor (Hauer, 1997): wb ¼

1 1 þ a  Pbi

(3)

In this case, wb assigns a relative weight between the ^ ib predicted and the observed numbers of crashes to compute m b ^ ib , and vice versa). (i.e., as w increases, xbi has less of an effect on m This EB method is known to address issues that arise due to RTM and changes in traffic volume (a crash prediction model is a function of AADT, so that changes in traffic volume can be reflected). However, the effect of general changes in traffic crashes ^ ib should be from the before to after periods still exists. Thus, m adjusted by how much the number of crashes would have changed without the ASES. The adjusted amount is estimated as the ratio of the observed number of crashes in the comparison group in the period after the installation (Ca) to the observed number of crashes in the comparison group in the period before the installation (Cb).1 The effectiveness (ei) can be computed as follows: ^ ib  ðC a =C b ÞÞ ei ¼ mai =ðm

(4)

1 Before applying the adjustment, the suitability of the selection of the comparison group should be tested. The odds ratio between the treated group t1 and the comparison group is used to this end. Odds ratio = CT tt =T =C t1 , where T is the total number of crashes in all of the treated groups, C is the total number of crashes in all of the comparison groups, and the subscripts t and t  1 indicate the observation year. If the odds ratio is close to 1, the comparison group is suitable. In the present study, all of the Korean expressways were selected as a comparison group. The calculated odds ratio was found to be 0.95, confirming fair suitability as a comparison group (Hauer, 1997; De Pauw et al., 2014b).

J. Shim et al. / Accident Analysis and Prevention 80 (2015) 57–66

63

Table 2 Summary of data used for crash prediction model. Variables

Description

Mean

Standard deviation

Min

Max

Crashes Length AADT

Number of crashes that occurred in the segment for four years (2007–2010) Length of segment (km) Annual average daily traffic on the segment

16.3 8.4 33727.0

11.36 5.26 23217.30

0 0.6 3382

70 27.8 111356

Although the countermeasure may substantially reduce crash occurrences, it is unlikely that the site will have zero crashes after the implementation of the countermeasure (Elvik, 2011; De Pauw et al., 2013). Yet it is possible that mai will be zero due to short spatial interval and observation duration. In our study, therefore, it is assumed that true mean of crash occurrences at a site is not zero although observed crashes were zero at the site. Elvik (2011) explained that it is highly implausible that the true long-term mean number of crashes at any section will be zero even if the observed number of crashes is zero. Furthermore, zero crashes after countermeasure implementation means 100% effectiveness (i.e., crash reduction factor (CRF) is 100%), which is implausible according to the extensive list of CRF documented by FHWA (2007), which shows that there is no countermeasure that can result in 100% effectiveness. This zero count problem can be resolved by applying the EB method to the period after the installation, as in ^ ia was used instead of mai . Eq. (5); here, m

study relies on small crash samples due to short segment length and observation period (Elvik and Mysen, 1999). By applying metaanalysis, therefore, we were able to draw statistically more reliable outcomes even though the estimated effectiveness at each study site was inconclusive. In this study, we adapt the fixed-effect metaanalysis by Fleiss (1981) for our purposes. With this method, we produce an overall index of the effectiveness (ei in Eq. (6)) and confidence interval (see Eq. (7)) across the sites by assigning the inverse of the variance as a weight factor. $ n % ^ ¼ exp Si¼1 f i  lnðei Þ (6) E n Si¼1 f i

^ ia ¼ wa  Pai þ ð1  wa Þ  xai m

where the weight factor (fi) is estimated based on the variance of ei, s2i for each site.

(5)

^ ia is the estimated number of crashes at the treated where m location i during the period after the installation of the ASES, wa is the weight that is given to the predicted number of crashes at the treated location during the after-period, Pai is the predicted number of crashes during the after-period, which is estimated using a crash prediction model described in Table 4, and xai is the observed number of crashes that occurred at the treated location i during the after installation period. The predicted number of crashes, Pai , and the weight factor, wa, for the after-period were estimated using the data collected from the after-period but the estimation procedure is the same as for the before-period. Finally, it was possible to calculate the effectiveness for each ASES location (ei). However, the number of crashes varies across the sites because the study sites have different attributes (e.g., different AADT and other unobserved factors). Hence, we were unable to combine the crashes from all the sites and apply conventional before-and-after analysis because the estimated effectiveness would be more influenced by the outcomes at sites with larger numbers of crashes. To draw more general implications, therefore, we need to evaluate the overall effectiveness of ASESs across all sites. To do this, a meta-analysis was conducted. 4.2. Meta-analysis Meta-analysis is often used to reduce the statistical uncertainty of the estimated effectiveness at each site because the present

6 7 6 n 7 6S f  lnðe Þ 7 1 6 i¼1 i 7 i
qffiffiffiffiffiffiffiffiffiffiffiffiffi5 95% Confidence interval ¼ exp4 n n Si¼1 f i Si¼1 f i

fi ¼

1 s2i

(8)

s2i ¼

1 1 1 1 þ bþ aþ b ^i mai m C C

(9)

4.3. Results As revealed in Section 3, traffic speeds vary along sections near ASESs and, thus, the safety performance may also differ spatially along the sections. To examine the spatial distribution of changes in the safety performance near an ASES, we divide the 5-km section with the ASES in the middle into five 1-km segments. The segments are labeled 1 through 5 from the upstream segment. All five segments have almost uniform geometric conditions other than the ASES because sites in which special geometric features exist were excluded. The descriptive statistics of crash data are summarized in Table 5. Also, the index of effectiveness (ei) for each of the five ^ ia ; segments of the 28 study sites was estimated using Eq. (4) with m ^ was estimated and then, the overall index of effectiveness (E) ^ values for segments 1 and values are summarized in Table 6. The E 5 are 0.98 and 1.00, respectively, implying that these segments

Table 3 Estimated coefficients for crash prediction model (for the before-period). Crashes/year

Coefficient

Std. Err.

z (p > |z|)

b0 b1

0.242 1.46E-05 1 0.183

0.027 6.74E-07 (exposure) 0.011

8.83 (0.000) 21.66 (0.000)

ln(l)

a

(7)

95% Confidence interval Lower

Upper

0.189 0.000

0.296 0.000

0.163

0.205

64

J. Shim et al. / Accident Analysis and Prevention 80 (2015) 57–66

Table 4 Estimated coefficients for crash prediction model (for the after-period). Crashes/year

Coefficient

Std. Err.

z (p > |z|)

95% Confidence interval Lower

Upper

b0 b1

0.260 1.45e-05 1 0.189

0.039 9.47e-07 (exposure) 0.016

6.68 (0.000) 15.36 (0.000)

0.183 0.000

0.336 0.000

0.161

0.222

ln(l)

a

Table 5 Descriptive statistics for crashes in both the treatment and the comparison groups. Groups

Period

Segment

Mean

Standard deviation

Min

Max

Treated locations

Before

2.5 k to 1.5 k 1.5 k to 0.5 k 0.5 k to +0.5 k +0.5 k to +1.5 k +1.5 k to +2.5 k 2.5 k to 1.5 k 1.5 k to 0.5 k 0.5 k to +0.5 k +0.5 k to +1.5 k +1.5 k to +2.5 k Various expressways according to each treatment site Various expressways according to each treatment site

2.8 3.1 3.1 2.9 4.8 2.4 3.3 2.1 2.4 4.3 902.6 857.5

2.45 3.17 2.46 2.29 7.65 2.20 2.69 2.29 2.13 4.38 664.91 583.26

0 0 0 0 0 0 0 0 0 0 72 30

8 16 8 9 39 8 10 9 7 17 2071 1841

After

Comparison locations

Before After

Table 6 Estimated overall index of effectiveness. Segment

Segment 1 (2.5 k to 1.5 k)

Segment 2 (1.5 k to 0.5 k)

Segment 3 (0.5 k to +0.5 k)

Segment 4 (+0.5 k to +1.5 k)

Segment 5 (+1.5 k to +2.5 k)

Overall (1.5 k to +1.5 k)

^ [95% Confidence interval] E

0.98 [0.68; 1.39]

1.11 [0.81; 1.52]

0.80 [0.56; 1.16]

0.94 [0.67; 1.33]

1.00 [0.76; 1.30]

0.92 [0.76; 1.12]

were not affected by the ASES. Meanwhile, there were changes in the numbers of crashes in other segments – segments 2, 3, and 4 – after the installation of the ASES. In segment 2, where information on speed enforcement is provided to drivers via roadside signs or an on-board navigation ^ is 1.11, meaning that the number of crashes increased by system, E 11%. This strongly implies that the number of crashes increased after the installation of the ASES, as drivers started to reduce their speeds within this segment. In the segment including the ASES (segment 3), a 20% reduction in crash occurrence was observed after the installation of the ASES. This indicates that the ASES can reduce the number of crash occurrences near the enforcement location, as traffic speeds are reduced and stabilized in this ^ is 0.94, meaning that crash occurrence segment. In segment 4, E has diminished by 6% downstream of the ASES. This can be viewed as a ‘spillover effect’, in which the system not only improved safety at the ASES location but also in nearby sections (Chen et al., 2002). However, these changes were not significant. The outcomes show that the ASES has an effect on a 3-km section, from segments 2 through 4. The effects for different segments within this 3-km section have split outcomes – an increase in segment 2 and ^ for these segments reductions in segments 3 and 4. The overall E combined is 0.924, indicating that the overall impact of the ASES on traffic safety is a 7.6% reduction in the number of crash occurrences. This outcome can be explained by looking at the changes in traffic behavior because it is conjectured that perturbations in the traffic speed initiated in segment 2 become stabilized at a lower speed near the ASES in segment 3, after which vehicles tend to accelerate back to their desired speeds in segment 4.

5. Conclusions and discussion ASESs have been widely deployed as a safety countermeasure to discourage speeding behavior and thereby to reduce crash occurrences. Many previous studies have found that ASES is effective at enhancing overall traffic safety in nearby roadway sections, and that the safety effects are different in accordance with the section location and the distance from the ASES. Although a few studies have hypothetically explained that the difference in safety effects may be caused by drivers’ EAB (De Pauw et al., 2014b; De Pauw et al., 2014b), the existence of EAB and its causal sequence have not been fully unveiled. In this study, we observed drivers’ EAB near an ASES by analyzing real vehicle speed data collected from taxi trajectories and loop detectors. The analysis showed that drivers start to reduce their speeds at about 1000-m upstream of the ASES; they then recover their desired speed shortly after passing the ASES. Based on this observation, we divided our study sites into segments and conducted comparative analysis using the EB method for each segment before and after the ASES installation. The comparisons showed that the ASES had a positive effect on overall traffic safety in the sections under study, reducing total crashes by 7.6%. Although the ASES has a positive effect on overall traffic safety, the magnitude of the effect is small and not statistically significant. Considering that most previous studies have found that ASES reduced crashes by over 20% (Elvik, 1997; Hess and Polak, 2003; Mountain et al., 2004), the magnitude of the positive effect in our study was relatively small. This is because there was an 11% increase in crash occurrences in the 1500-m and 500-m segments upstream of the ASES, where drivers were alerted of enforcement

J. Shim et al. / Accident Analysis and Prevention 80 (2015) 57–66

by on-board navigation or by roadside signs and started to diminish their speeds. This negative effect is consistent with the recent finding in De Pauw et al. (2014b), which showed that the favorable safety effect of an ASES vanishes between 500 and 750-m upstream of the ASES, resulting in an 8% increase in overall crashes. To explain this increase, De Pauw et al. (2014b) conjectured that it may have been caused by drivers’ EAB; however, the authors were unable to validate the existence of an EAB. The present study provides empirical evidence that EAB indeed exists and creates perturbations in traffic speed; the study concludes that the increase in crash occurrences upstream of the ASES was due in large part to perturbations induced by EAB. It was possible to arrive at this conclusion based on the fact that variations in traffic speed cause crash occurrences (Yeo et al., 2013; Oh et al., 2006; Chung et al., 2010; Abdel-Aty et al., 2006; Golob et al., 2004; Liu and Popoff, 1997), though observations of traffic speed and safety were conducted separately in the present study. It has been reported that ASES is effective at enhancing safety by keeping vehicle speeds below the speed limit, as well as by reducing speed differences across vehicles. However, the findings in this study suggest that the benefits of the ASES may be nullified by informing drivers of enforcement locations, which induces EAB. Therefore, strategies mitigating EAB may be considered to further improve the safety benefit from installing ASES. One way is to provide information only on speed enforcement without detailed location information. By doing so, drivers’ compliance with the speed limit could be made greater throughout the roadway, as has been reported in a hidden versus visible camera experiment conducted by Keall et al. (2001). Another promising alternative is section ASES (also known as section control, point-to-point speed enforcement, trajectory control, tutor, and average speed enforcement) (ETSC, 2009). Such a system records a vehicle’s entrance and exit times for a section and computes the actual travel time. Then, the travel time is converted to the average travel speed over the section. If the average speed exceeds the speed limit, the corresponding vehicle is penalized. Since drivers must comply with the speed limit over the section if they are to avoid enforcement under this system, it is expected that EAB will be discouraged. In this perspective, some studies have evaluated the effectiveness of such a system and reported promising outcomes (Høye, 2014; De Pauw et al., 2014c; Montella et al., 2012). However, since this type of system has been implemented only in a limited number of locations, the effectiveness of the system has not yet been fully validated (Soole et al., 2013). Thus, studies are necessary to evaluate various speed control systems, including section ASES, to determine the most effective system for reducing the number of speeding crashes. Such an evaluation remains a topic for future research. Acknowledgement This research was supported by the National Research Foundation of Korea grant funded by the Korea government (MSIP) (NRF-2010-0028693). Appendix A. Supplementary data Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/j.aap.2015.03.037. References Aarts, L., Schagen, I.V., 2006. Driving speed and the risk of road crashes: a review. Accid. Anal. Prev. 38 (2), 215–224. Abdel-Aty, M., Dilmore, J., Hsia, L., 2006. Applying variable speed limits and the potential for crash migration. Trans. Res. Rec.: J. Trans. Res. Board 1953 (1), 21–30.

65

ARRB Group. Evaluation of the Fixed Digital Speed Camera Program in NSW, for Roads and Traffic Authority http://www.roadsafetyobservatory.com/Evidence/ Details/11123, NSW. RC2416. 2005. Chen, G., Wilson, J., Meckle, W., Cooper, P., 2000. Evaluation of photo radar program in British Columbia. Accid. Anal. Prev. 32, 517–526. Chen, G., Wilson, J., Meckle, P., 2002. Speed and safety effect of photo radar enforcement on a highway corridor in British Columbia. Accid. Anal. Prev. 34, 129–138. Christie, S.M., Lyons, R.A., Dunstan, F.D., Jones, S.J., 2003. Are mobile speed cameras effective? A controlled before and after study. Inj. Prev. 9, 302–306. Chung, K., Jang, K., Oum, S., Kim, Y., Song, K., 2010. Investigation of attributes of kinematic waves preceding traffic collisions. In 17th ITS World Congress . De Pauw, E., Daniels, S., Brijs, T., Hermans, E., Wets, G., 2013. The effect of combined speed and red light cameras on safety. Transportation Research Board 92nd Annual Meeting No. 13-1970. De Pauw, E., Daniels, S., Brijs, T., Hermans, E., Wets, G., 2014a. Behavioural effects of fixed speed cameras on motorways: overall improved speed compliance or kangaroo jumps? Accid. Anal. Prev. 73, 132–140. De Pauw, E., Daniels, S., Brijs, T., Hermans, E., Wets, G., 2014b. An evaluation of the traffic safety effect of fixed speed cameras. Saf. Sci. 62, 168–174. doi:http://dx. doi.org/10.1016/j.ssci.2013.07.028. De Pauw, E., Daniels, S., Brijs, T., Hermans, E., Wets, G., 2014c. Automated section speed control on motorways: an evaluation of the effect on driving speed. Accid. Anal. Prev. 73, 313–322. Decina, L.E., Thomas, L., Srinivasan, R., Staplin, L., 2007. Automated Enforcement: A Compendium of Worldwide Evaluations of Results. Office of Research and Technology. Behavioral Technology Research Division, National Highway Traffic Safety Administration, Washington, D.C. Elvik, R., 1997. Effects on accidents of automatic speed enforcement in Norway. Trans. Res. Rec. 1595, 14–19. Elvik, R., Christensen, P., Amundsen, A., 2004. Speed and Road Accidents: An Evaluation of the POWER Model. Report 740/2004. Institute of Transport Economics (TØI), Oslo. Elvik, R., 2011. Treatment of zero counts in before-and-after road safety evaluation studies: exploratory study of continuity corrections. Transportation Research Board 90th Annual Meeting, No. 11-0133 . Elvik, R., Mysen, A.B., 1999. Incomplete accident reporting: meta-analysis of studies made in 13 countries. Trans. Res. Rec.: J. Trans. Res. Board 1665 (1), 133–140. ETSC, 1999. Police Enforcement Strategies To Reduce Traffic Casualties in Europe. European Transport Safety Council (ETSC), Brussels. ETSC, 2009. Section control: towards a more efficient and better accepted enforcement of speed limits. ETSC Speed Fact Sheet, No. 5. Evans, L., 2004. Traffic Safety. Science Serving Society, Bloomfield Hills, MI, USA. Fleiss, J.L., 1981. Statistical Methods for Rates and Proportions. John Wiley, New York. FHWA, 2007. Desktop Reference for Crash Reduction Factors. U.S. Department of Transportation Federal Highway Administration. Report No. FHWA-SA-07-015. FHWA. Gambard, J.M., Fabre, P., Boulanger, P., 1997. An automatic speed recording system on French motorways: the speed observatory. The 4th World Congress on ITS, Berlin, Germany. Goldenbeld, C., van Schagen, I., 2005. The effects of speed enforcement with mobile radar on speed and accidents: an evaluation study on rural roads in the Dutch province of Friesland. Accid. Anal. Prev. 37, 1135–1144. Golob, T.F., Recker, W.W., Alvarez, V.M., 2004. Freeway safety as a function of traffic flow. Accid. Anal. Prev. 36 (6), 933–946. Hauer, E., 1997. Observational Before–After Studies in Road Safety. Elsevier Science, Oxford. Hess, S., Polak, J., 2003. An analysis of the effects of speed limit enforcement cameras on accident rates. Trans. Res. Rec. 1830, 25–33. Høye, A., 2014. Speed cameras, section control, and kangaroo jumps – a metaanalysis. Accid. Anal. Prev. 73, 200–208. Jang, K., Shim, J., Chung, S., Park, S.H., 2014. Effectiveness of managed lanes on South Korean expressways. Transportation Research Board 93rd Annual Meeting, No. 14-3623 . Keall, M.D., Povey, L.J., Frith, W.J., 2001. The relative effectiveness of a hidden versus a visible speed camera programme. Accid. Anal. Prev. 33, 277–284. Korea Expressway Corporation, 2013. Open Oasis Portal. http://www.ex.co.kr (accessed 06.05.13.). Lee, C., Saccomanno, F., Hellinga, B., 2002. Analysis of crash precursors on instrumented freeways. Trans. Res. Rec. 1784 (1), 1–8. Liu, G.X., Popoff, A., 1997. Provincial-wide travel speed and traffic safety study in Saskatchewan. Trans. Res. Rec.: J. Trans. Res. Board 1595, 8–13. Maycock, R., Brocklebank, P.J., Hall, R.D., 1998. Road Layout Design Standards and Driver Behavior Transport Research Laboratory Report, No. 332, Crowthorne, Berkshire. Montella, A., Persaud, B., D'Apuzzo, M., Imbriani, L.L., 2012. Safety evaluation of automated section speed enforcement system. Trans. Res. Rec. J. Trans. Res. Board 2281 (1), 16–25. Mountain, L., Hirst, W., Maher, M., 2004. Costing lives or saving lives? a detailed evaluation of the impact of speed cameras on safety. Traffic Eng. Control 45, 280–287. Newstead, S., Cameron, M., 2003. Evaluation of Crash Effects of the Queensland Speed Camera Program. Monash University Accident Research Centre Report No. 204. Monash University Accident Research Centre, Victoria, Australia.

66

J. Shim et al. / Accident Analysis and Prevention 80 (2015) 57–66

Nilsson, G., 1982. The effects of speed limits on traffic crashes in Sweden. Proceedings of the International Symposium on the Effects of Speed Limits on Traffic Crashes and Fuel Consumption, Organization for Economic Co-operation and Development, Ireland. Oh, C., Park, S., Ritchie, S.G., 2006. A method for identifying rear-end collision risks using inductive loop detectors. Accid. Anal. Prev. 38 (2), 295–301. Park, S., Jang, K., Park, S., Kim, D., Chon, K., 2012. Analysis of injury severity in traffic crashes: a case study of Korean expressways. Kor. Soc. Civil Eng. (KSCE) J. Civil Eng. 16 (7), 1280–1288. doi:http://dx.doi.org/10.1007/s12205-012-1527-3. Shin, K., Washington, S.P., Van Schalkwyk, I., 2009. Evaluation of the Scottsdale Loop 101 automated speed enforcement demonstration program. Accid. Anal. Prev. 41 (3), 393–403.

Soole, D.W., Watson, B.C., Fleiter, J.J., 2013. Effects of average speed enforcement on speed compliance and crashes: a review of the literature. Accid. Anal. Prev. 54, 46–56. doi:http://dx.doi.org/10.1016/j.aap.2013.01.018. Thomas, L.J., Srinivasan, R., Decina, L.E., Staplin, L., 2008. 2Safety effects of automated speed enforcement programs: critical review of international literature. Trans. Res. Rec. J. Trans. Res. Board 2078 (1), 117–126. Washington, S.P., Shin, K., Van Schalkwyk, I., 2007. Evaluation of the City of Scottsdale Loop 101 Photo Enforcement Demonstration Program: Final report. Arizona Department of Transportation. Yeo, H., Jang, K., Skabardonis, A., Kang, S., 2013. Impact of traffic states on freeway crash involvement rates. Accid. Anal. Prev. 50, 713–723.