- Email: [email protected]
Safety effects of dynamic speed limits on motorways
Accident Analysis and Prevention xxx (xxxx) xxx–xxx
Contents lists available at ScienceDirect
Accident Analysis and Prevention journal homepage: www.elsevier.com/locate/aap
Safety effects of dynamic speed limits on motorways ⁎
Ellen De Pauwa, Stijn Danielsb,c, , Laurent Franckxd,1, Inge Mayeresd,e,2 a
Rondpunt, Uitbreidingstraat 518, 2600 Berchem, Belgium Belgian Road Safety Institute, Haachtsesteenweg 1405, 1130 Brussels, Belgium Hasselt University, Transportation Research Institute, Wetenschapspark 5, 3590 Diepenbeek, Belgium d VITO—Flemish Institute for Technological Research, Boeretang 200, 2400 Mol, Belgium e Katholieke Universiteit Leuven, Department of Economics, Naamsestraat 69, 3000 Leuven, Belgium b c
A R T I C L E I N F O
A B S T R A C T
Keywords: Safety Crash Motorway Dynamic speed limit Empirical bayes Cost-benefit
Dynamic speed limits (DSL) are limits that change according to real-time traffic, road or weather conditions. In DSL-schemes road users are typically informed of speed limit changes by electronic signs that are housed within gantries situated above lanes. Dynamic speed limit systems are increasingly applied worldwide, usually on motorways. One of the objectives of dynamic speed limits is to improve traffic safety through reductions in speed variations within and across lanes and between upstream and downstream flows. This paper shows the results of an empirical evaluation of the effects on traffic safety of a dynamic speed limit system on motorways in Flanders, Belgium. The evaluation was done by means of a before-after analysis of crashes, completed with a cost-benefit analysis. The results show that the number of injury crashes decreased significantly (−18%) after the introduction of the system. A separate analysis for serious and fatal injury crashes revealed a non-significant decrease of 6%. A distinction according to crash type showed an almost significant decrease of 20% in the number of rear-end crashes whereas the number of single-vehicle crashes decreased by 15% (ns). However, no effect was found for side crashes. In addition to the analysis of the effects, a cost-benefit analysis was applied. The costs of the implementation of these systems were compared with the benefits of crash prevention. The cost-benefit analyses of the crash effects showed a benefits-to-costs ratio of approximately 0.7, which means that the costs tend to exceed the benefits. Taking into account the important margins of uncertainty with respect to both costs and benefits, we have also explored how the net benefits are affected by some key assumptions. The general conclusion is that there is no convincing evidence that the costs of the system currently outweigh the expected benefits in terms of crash prevention.
1. Introduction
by generic legislation. In some countries the speed limit is reduced in case of rain, or speed limits nearby school zones are reduced at school start or end times. The focus of the present study is on DSLs, which are applied as a consequence of the real time situation. Through these systems, speed limits can be adapted remotely, either automatically by an algorithm or manually by an operator. This makes it possible to show different speed limits at different times of the day and different days of the week (Van Nes et al., 2010). DSLs are introduced to harmonize traffic flows which is assumed to improve both throughput and traffic safety. The traffic safety improvement is targeted through reductions in speed variations within and across lanes and between upstream and downstream flows
On the majority of the roads, fixed speed limits represent the appropriate speed for average conditions. However, in order to take account of the real time traffic, road and weather conditions, dynamic speed limits (DSLs) can be applied (European Commission, 2010). DSL systems are activated at a given time, as a consequence of traffic volume or other environmental conditions (Islam et al., 2013; OECD, 2006). Variable speed limits are often used as a synonym for DSLs. However, according to the OECD (2006) the term ‘variable speed limits’ refers to systems that are activated through general criteria (e.g. time of the day, season and certain weather conditions), which are usually set
⁎
Corresponding author at: Belgian Road Safety Institute, Haachtsesteenweg 1405, 1130 Brussels, Belgium. E-mail addresses: [email protected] (E. De Pauw), [email protected] (S. Daniels), [email protected] (L. Franckx), [email protected] (I. Mayeres). Present address: Belgian Federal Planning Bureau, Kunstlaan 47-49, 1000 Brussel, Belgium. 2 Present address: Transport & Mobility Leuven, Diestsesteenweg 57, 3010 Leuven, Belgium. 1
http://dx.doi.org/10.1016/j.aap.2017.06.013 Received 6 October 2016; Received in revised form 16 June 2017; Accepted 17 June 2017 0001-4575/ © 2017 Elsevier Ltd. All rights reserved.
Please cite this article as: Pauw, E.D., Accident Analysis and Prevention (2017), http://dx.doi.org/10.1016/j.aap.2017.06.013
Accident Analysis and Prevention xxx (xxxx) xxx–xxx
E. De Pauw et al.
This paper addresses the effects of DSL systems on traffic safety. As will be shown in Section 2, quite a number of studies analysed the effects of DSLs through simulation models and driving simulator studies. However, we didn’t find peer reviewed empirical studies that analysed the impact on traffic safety. Furthermore, to our knowledge, no literature about the cost-effectiveness of DSL systems is available. The present study therefore analyses the traffic safety effect of DSL systems in Flanders, Belgium, based on an empirical analysis of observed crash data. The effects are analysed through an empirical Bayes before-andafter study, which compares the crashes after the implementation of the measure with the number of crashes before, and controls for confounding variables. The empirical analysis is complemented by a costbenefit analysis of the applied DSL system. Section 2 provides an overview of previous studies that analysed the effects of DSL systems. Studies that analysed the effects on the traffic flow as well as studies that analysed the traffic safety effects are included in this section. In Section 3 the data, the methodology and the results of the evaluation of the effects on traffic safety are described; Section 4 describes the cost-benefit analysis. The results are discussed in Section 5 and conclusions are listed in Section 6.
DSLs to delay the onset of congestion and to help produce more rapid recovery from congestion, provided that demand volumes are not too far above the zone capacity. When demand volumes are high, they found no benefit over static speed limits. The simulation also showed the importance of appropriate DSL sign location and effective algorithm design. Habtemichael and de Picado Santos (2013) analysed the operational benefits of DSLs under different traffic conditions. They studied the combination of different compliance rates and congestion levels and found that the operational benefits depended on these two factors. The system had the highest operational benefits during lightly congested traffic conditions, little benefit during uncongested conditions, and no benefit during heavily congested conditions. Also impact of DSLs on traffic safety was assessed by means of simulation models in a number of papers. We briefly discuss them. Lee et al. (2004) used a real time crash prediction model integrated with a microscopic traffic simulation model. They found that temporarily reducing speed limits during risky traffic conditions can reduce the crash potential. The greatest reduction occurred at the location with a high traffic turbulence. Abdel-Aty et al. (2006a,b) studied how traffic safety could be increased at a motorway in Orlando. They found DSLs can be used to improve safety, through the implementation of lower speed limits upstream and higher speed limits downstream of the location where crash likelihood is observed in real time. This improvement was present in the case of medium-to-high-speed regimes but not in lowspeed situations. They furthermore analysed the potential for crash migration and found that the crash potential relocates to a location downstream of the detector of interest. Overall the safety of the freeway was improved (Abdel-Aty et al., 2006a,b). In a later study, Abdel-Aty et al. (2008) continued on this research and found that DSLs can be used to reduce crash risk and prevent crash occurrence in free-flow conditions and conditions approaching congestion. Habtemichael and de Picado Santos (2013) found however somewhat different results. In their study the most favourable traffic safety effects were found during highly congested traffic conditions, followed by lightly congested conditions and the least during uncongested situations. Furthermore, they found that the effects are highly dependent on the level of driver compliance. Lee et al. (2006) studied the safety benefits of DSLs and used simulated traffic conditions on a freeway in Toronto. They found that real time DSLs can reduce the overall crash potential by 5%–17%. Also Islam et al. (2013) analysed this impact. They proposed a model predictive DSL control strategy. The safety impact was quantified through a real time crash prediction model for an urban freeway corridor in Alberta. The results indicated that the DSL can improve safety by 50%.
2. Literature review
2.2. Field studies
A number of peer reviewed studies analysed the effects of DSL systems. These effects have mainly been stated in terms of traffic safety and traffic operations. In a recent paper, Lu and Shladover (2014) reviewed studies on DSLs and classified these studies as three types: simulations for algorithm development and evaluations, DSL implementation and field testing, and a combination of DSLs with ramp metering. We will briefly discuss the literature that has attempted to assess the effects of DSLs, by means of simulations as well as through field studies. Combinations of DSLs with ramp metering are outside the scope of this paper and not further treated.
The effect of DSL systems was also analysed through empirical studies. Lu and Shladover (2014) reviewed evidence from field studies that assessed effects on traffic flow and traffic safety in the UK, Germany, the Netherlands, France and the United States. The effects reported on traffic operations were mixed. Some authors reported a reduction of travel times (Chang et al., 2011; Hoogendoorn et al., 2013). Some studies reported an improvement in throughput (Chang et al., 2011; Kwon et al., 2007), but others not (DeGaspari et al., 2013). Papageorgiou et al. (2008) reported that effects on traffic flow are highly dependent on the saturation level of traffic with improved traffic flow at overcritical occupancies (dense traffic), but deteriorated traffic flow efficiency (lower average speeds) at undercritical traffic conditions. Multiple studies reported favourable effects on road safety for the implemented VSL strategies, but interestingly, none of the field studies that assessed effects on traffic safety has been published in peer reviewed journals and their methodological rigour is hard to assess. The road safety literature (Hauer, 1997; Elvik, 2002) has extensively reported that accident studies that don’t take into account confounding
(Lee et al., 2004; Islam et al., 2013). Throughput improvement is expected due to the speed harmonization benefits (Fudala and Fontaine, 2010). DSLs are sometimes also used in order to reduce vehicle emissions and road noise (Papageorgiou et al., 2008). DSL systems are also applied on motorways in the Flanders region in Belgium. The variable speed limit is denoted by electronic signs that are housed within gantries situated above motorway lanes. Speed limits imposed by DSL systems are compulsory and have – according to the road authority – three main objectives (Vlaams Verkeerscentrum, 2015): (1) Increase safety: upstream from an incident (e.g. traffic jams, crashes, road works) speed limits can be temporarily lowered in order to reduce the mean speed, lead the traffic smoothly to the incident and avoid the occurrence of crashes; (2) Indicate obstructions: DSLs can lead away traffic from a blocked lane; (3) Improve traffic flow through homogenization of speed: at moments with a high traffic flow, speed limits will be reduced, which will lead to a more homogeneous traffic flow and to less manoeuvres. Furthermore, headways are assumed to become smaller, which means that the available space is used more efficiently and the probability of traffic jams is subsequently lower.
2.1. Simulation models Many authors developed simulation models to assess the effects of DSLs. Islam et al. (2013) studied the effects on mobility. In the best case scenario, DSL control with a 5-min speed limit update frequency and a 10-km/h maximum speed difference between two successive time steps, they reported a 33% reduction of total travel time. Fudala and Fontaine (2010) did this for work zones specifically. They found potential of 2
Accident Analysis and Prevention xxx (xxxx) xxx–xxx
E. De Pauw et al.
factors such as trends and regression-to-the-mean tend to bias – usually to overstate – effects. Based on the reviewed evidence, Lu and Shladover (2014) concluded that DSLs can significantly improve freeway traffic safety if the compliance rate is high enough whereas the impact on traffic throughput is still controversial. The latter is explained by inherent difficulties in the observation of traffic flow status, by the large variations in driver behaviour and by the relative immaturity of DSL systems that often use algorithms that are unlikely to improve traffic flow. Hoogendoorn et al. (2013) evaluated effects of DSLs on emissions and noise. They reported a slight deterioration for both aspects.
5) the before period consisted of crashes from 1999 up to 2002 and the after period included crashes from 2006 until 2011. For the segment where a DSL system was installed in 2009 (segment 4) the before period included 2006–2008 and the after period 2010–2011. The before and after periods in the comparison group were each time the same as for the research group. Through this selection the most recent available years of crash data were used. The before period amounted to 3.8 years on average; the after period amounted to 5.2 years. The crashes were subdivided in two groups according to their severity: (1) injury crashes (all crashes with at least one injured person); (2) severe injury crashes, i.e. crashes with either severely injured persons (every person who needed more than 24 h of hospitalization as a result of a crash) or fatally injured persons (every person who died within 30 days after the crash as a consequence of the crash). Table 2 gives an overview of the average number of crashes at the treated locations per year for the full research period. These numbers clearly show that the rear-end crash is the crash type that occurs most frequently.
3. Evaluation of the traffic safety effect 3.1. Data In 2003 the first DSL systems were installed at Belgian motorways. Motorways are defined here as roads for motorized vehicles only with a median barrier and no at-grade junctions (Elvik et al., 2009). The general maximum speed limit on Belgian motorways is 120 km/h. Access to motorways is forbidden for pedestrians, cyclists, moped riders and all vehicles that cannot drive faster than 70 km/h. The speeds that are displayed by the dynamic signs are based on the data gathered by loop detectors and by automatic incident detection cameras. Data on speed and occupancy of the lane are used to set the speed limit. When the loops and cameras detect a high occupancy together with a low speed the speed limit is reduced. At the locations upstream from this incident the speed is gradually reduced in order to prevent sudden braking. Speed is reduced from 120 km/h to 110 km/h, 90 km/h, 70 km/h and 50 km/h as lowest speed limit. Weather conditions are not taken into account in the algorithm. Corthout et al. (2010) evaluated the DSL system around Antwerp and concluded that the performance of the system is good, with a reasonably high detection rate and very low false alarm rate. For the present study, police-reported crash data were available up until 2011. In order to have at least one year of crash data available in the after period, all DSL systems that were installed and operational up until 2010 were included in the study. In total five road segments with DSLs were assessed, covering a total distance of 59.54 km. These five segments are all located at access roads of the ring road of Antwerp. Table 1 gives an overview of the characteristics of the DSL locations, further called the ‘treated locations’. Fig. 1 gives a graphical overview of the locations. Each dot represents a gantry with DSLs. The comparison group, that was selected in order to control for the general trend effects, included all crashes that occurred on motorways in the Flanders region, at least 10 km away from locations with DSLs. All crashes from 1999 up to 2011 were used. Data were available from the national crash statistics (Statistics Belgium). The crashes from the year 2003 were excluded from the study, since there were generic problems with the crash registration in that year. Also the crashes from the years 2004–2005 were excluded since major road works were at that moment carried out on the ring road of Antwerp, nearby the five treated segments and thus possibly influencing traffic volumes or road user behaviour on the treated segments. Therefore, for the segments where DSL systems were installed in 2003–2004 (segments 1, 2, 3 and
3.2. Method The traffic safety effect is studied through an empirical Bayes (EB) before-and-after study, that compares the crash numbers after the implementation of the measure with the before situation. The EB approach increases the precision of estimation and corrects for the regression-tothe-mean (RTM) bias (Hauer et al., 2002). In a first step, the effect per segment is calculated (see Table 1, for more information on the location of the five segments). The analysis per segment (further referred to as ‘location’) can be expressed through an odds ratio, which results in an estimation of the index of effectiveness (θl): Ll
θl =
1 2 3 4 5
Length in km
Year of installation
No. of lanes
E17 Ghent-Antwerp E19 Breda-Antwerp E19 Brussels-Antwerp E313 Geel-Ranst E34/E313 Ranst-Antwerp
7.3 4.4 6.0 31.8 10.0
2004 2004 2003 2009 2003
3 2 3 2 2
N
M
(1)
E[κ|K]l = the expected number of crashes on the treated location l during the before period, controlled for RTM Ll = the observed number of crashes on the treated location l during the after period M = the observed number of crashes in the comparison group during the before period N = the observed number of crashes in the comparison group during the after period In order to increase the precision of the estimates, the empirical Bayes method makes joint use of two information sources on the safety of a location: the crash record of that location and the crash frequency expected at similar locations. A weighted average combines these two sources (Hauer et al., 2002): E[κ|K]l = w * E[κ] * T + (1 − w) * Kl
(2)
with w = the weight (between 0 and 1) that is given to the crashes at similar entities E[κ] = average number of crashes in the before period at similar entities T = number of years during the before period 1 − w = the weight given to the crashes at the treated location L Kl = observed number of crashes in the before period at the treated location L. To calculate the average number of crashes at similar entities (E[κ]) in Eq. (2), a Safety Performance Function (SPF) was used that earlier has been developed for crash occurrence of injury crashes on Flemish motorways. This study developed an SPF per motorway segment on the basis of several variables: (1) traffic volume, (2) length of road segment, (3) type of road segment and (4) number of lanes. The type of road segment included two main categories: road segments at entries/exits
Table 1 Characteristics of the treated segments. Road segment
E[κ K]l
3
Accident Analysis and Prevention xxx (xxxx) xxx–xxx
E. De Pauw et al.
Fig. 1. Geographical location of the treated segments. Table 2 Average number of crashes/year that occurred on the treated locations during the before and the after period.
Table 3 Results of the SPF. Injury crashes
Before After
Injury crashes
Severe crashes
Rear-end injury crashes
Single-vehicle injury crashes
Side injury crashes
108.7 61.3
22.9 15.4
51.1 31.3
31.6 17.7
13.7 8.3
and interchanges and road segments between two entries/exits or interchanges. The model has a negative binomial probability distribution with a log link function and is calculated using the SPSS GENLIN procedure.
−16.792 (SE:0.623)
Length of segment (β) Traffic volume (γ) Over dispersion Likelihood ratio test statistic (χ2)
0.939 (SE:0.029)*** 1.011 (SE:0.049)*** 0.313 (SE:0.031) 1013.5***
−18.493 (SE: 1.030)*** 0.951 (SE: 0.047)*** 1.035 (SE: 0.080)*** 0.325 (SE:0.070) 500.6***
*** significant at 1% level.
period to the annual average number of crashes that occurred in 2008–2010. To calculate the weight (w) in Eq. (2), next equation can be used:
2
E[κ] = eαL βV γe∑i= 1 δixi
α
Severe crashes ***
(3)
1
w=
with E[κ] = expected annual number of crashes α, β, γ, δ = model parameters L = length of road segment (in m) V = Traffic volume (in vehicles/24 h) x1 = Segment type (0 = at entries/exits and interchanges, 1 = between entries/exits and interchanges) x2 = Number of lanes {1,2,3} Traffic volume data were gathered through double inductive loops in the pavement. All segments with built-in inductive loops at all Flemish motorways were included in the SPF. In total 292 segments were included in the SPF for 2008, 381 segments for 2009 and 544 segments for 2010. The road segments included in the SPF have an average length of 2448 m (SD 2726 m) and an average daily traffic volume of 36047 (SD 20045) vehicles. Table 3 displays the final results of the SPFs, split up for all injury crashes and the severe crashes. The segment type and the number of lanes were not significant and thus not included in the final model. The model was based on crash data from 2008 to 2010. However, the before period in the present study was 1999–2002 for the locations at which DSL systems were installed in 2003 or 2004 and was 2006–2008 for the location at which DSLs were installed in 2009. Therefore the estimated number of crashes was multiplied with an adjustment factor to match the time frame of the observed data with the time frame of the SPF. This adjustment factor was expressed as the proportion of the annual average number of crashes during the before
1+
E[κ] *T k
(4)
with k the inverse value of the overdispersion parameter of the model, which is estimated per unit of length (Elvik, 2008).
3.2.1. Overall effect estimation The evaluation of each location separately has only limited significance. Therefore a fixed effects meta-analysis was carried out, which resulted in one overall effect estimate and shows more statistically reliable outcomes (Fleiss, 1981). Every location within the meta-analysis receives a weight, which is the inverted value of the variance. The variance is, based on the variables in Eq. (1), calculated as next:
sl ² =
1 1 1 1 + + + E[κ|K]l Ll M N
(5)
The weight can be calculated as follows:
wl =
1 s2l
(6)
Subsequently locations at which many crashes occurred are given higher weights. Supposing that the measure is executed at n different places, the weighted mean index of effectiveness of the measure over all places θ is as follows: 4
Accident Analysis and Prevention xxx (xxxx) xxx–xxx
E. De Pauw et al.
Table 4 Results of meta-analyses of the crash effects.
Table 5 Victim related costs according to Korzhenevych et al. (2014). Effect [95%CI]
Injury crashes Rear-end crashes Side crashes Single-vehicle crashes Severe crashes
0.82 0.80 1.00 0.85 0.94
[0.70; 0.96]* [0.64;1.01] [0.64; 1.56] [0.64;1.13] [0.68; 1.29]
n
(7)
The estimation of a 95% CI is as follows: n
⎡∑ wl *ln(θl ) 95%CI = exp ⎢ l=1 n ± 1.96* ⎢ wl ∑ l=1 ⎢ ⎣
1 n
∑l=1
⎤ ⎥ wl ⎥ ⎥ ⎦
Estimate in kEUROs
Fatality Severe injury Slight injury
2178.0 330.4 21.3
slightly injured persons was 78.50, the average annual number of severely injured persons was 22.50, and the number of fatalities was 2.50. In order to estimate the counterfactual, we will first apply the central values of the estimates of the index of effectiveness (see Table 4). For “severe injuries” and “fatalities”, we will apply the estimate for “severe crashes” (0.94), while for the number of slight injuries, we apply the estimate for “injury crashes” (0.82). This corresponds to 95.73 slightly injured persons, 23.94 severely injured persons and 2.66 fatalities in the counterfactual scenario. Combining the unit cost figures from Table 5 with the estimated number of fatalities and injuries avoided, we obtain the estimates of the annual economic value of crash prevention due to DSL, as described in Table 6.
* significant at 5% level.
wl *ln(θl ) ⎤ ⎡∑ θ = exp ⎢ l=1 n ⎥ ⎢ ∑l=1 wl ⎥⎦ ⎣
Cost category
(8)
A similar approach was used in De Pauw et al. (2014). 3.3. Results
4.2. The social costs of dynamic speed limits
The results of the meta-analyses are provided in Table 4. The metaanalysis of the injury crashes shows a significant decrease of 18% as a result of the implementation of DSL systems. The injury crashes were subdivided according to the type of crash and the three main crash types were analysed: (1) rear-end crashes; (2) side crashes; (3) singlevehicle crashes. As can be seen from Table 4, no effect on the number of side crashes is found. A decrease is found for the number of rear-end crashes (−20%) which is significant at the 10% level and almost significant at the 5% level as the upper limit of the 95% confidence interval is close to one. The number of single-vehicle crashes decreased by 15%, which is however not significant. An analysis of the fatal and serious injury crashes shows no significant effects.
In order to estimate the costs of DSL, we need to consider both investment costs and operational costs. In the absence of detailed cost accounts of the DSL, we have used indepth interviews with two officials from the Department of Mobility and Public Works as main source of information. One should realize that these figures are expert judgments of the interviewees. However, we esteem them to be a valid indication of the order of magnitude of the costs. Unless stated otherwise, all the figures below are based directly on the information we obtained during the interviews. The cost of installing the equipment is highly location specific. However, 269 kEUR was reported to us as being representative for the material investment costs per km of highway covered (including maintenance costs in the first two years of operation). To this material investment cost, we have added an estimated salary cost of the supervising personnel (47 kEUR per km). As taxes are a transfer, all these sums are net of VAT, income taxes and wage taxes. With an average distance between gantries of 750 m, this results in 237 kEUR as a representative investment cost per gantry. As a total distance of 59.54 km is covered by the studied DSL (see Section 3.1), this implies an initial investment cost of 18,815 kEUR. There are no official estimates of the economic lifetime of the installations. The oldest dynamic displays and DSL systems have been in use since 1999 and 2003, respectively. On the one hand, there is no indication that they will need to be replaced in the foreseeable future, and no major overhaul of the system is planned. On the other hand, taking into account current developments in the domain of automated and connected mobility, it seems reasonable to assume that, within a decade or so, the type of technology will be largely obsolete. We therefore assume a lifetime of 25 years. However, as we shall argue below, the principal results are robust for other assumptions. According to the Department of Mobility and Public Works, total annual maintenance costs (both preventive and curative) correspond to
4. Cost-benefit analysis As already discussed above, DSL systems are introduced for several reasons. We will stick here to the general approach of the paper, and limit ourselves to their benefits in terms of crash reduction. 4.1. The social benefits of the reduced crash risk In this section we calculate the social benefits of the reduced crash risk that was estimated in Table 4. First, we present the monetary values of the social costs of crashes that are proposed in the literature. Next, we apply these values to the impacts of DSL in Flanders. We use the figures from the ‘Update of the Handbook on External Costs of Transport’ (Korzhenevych et al., 2014) for the unit costs of crashes with injuries and fatalities. The advantage of this approach is that these figures have been obtained using an internationally recognized methodology. The following unit values are reported for Belgium (p 23), using the market prices (at Purchasing Power Parity3) of 2010. In order to translate these unit costs in total economic benefits of crash prevention, we also need estimates of the number of victims and/ or crashes, both the actual number and the number that would have prevailed in the absence of DSL systems (the counterfactual). In 2010–2011, on the treated locations, the average number of
Table 6 Total value of prevented injuries in caused by DSL in Flanders, Belgium.
3 In contrast to market exchange rates, Purchasing Power Parities reflect the differences in purchasing power between two currencies. They are thus the relevant metric to convert costs and benefits expressed in one currency in another currency.
5
Cost category
Estimate in kEUROs
Fatalities Severe injuries Slight injuries Total
348 475 367 1189
Accident Analysis and Prevention xxx (xxxx) xxx–xxx
E. De Pauw et al.
Flemish motorways (Magis, 2014) indicated that only one fourth of the drivers comply with the posted speed limit of 90 km/h and half of them keep the speed limit of 110 km/h. In comparison to moments with a fixed speed limit of 120 km/h, the average speed is only 6 km/h lower at moments with a dynamic speed limit of 90 km/h and is even 0.4 km/ h higher at moments with a speed limit of 110 km/h. Compared to moments with a speed limit of 120 km/h, the number of drivers that exceed the speed limit is 18 times higher during moments with a speed limit of 90 km/h and 5 times higher when the speed limit is 110 km/h. For the number of drivers that exceed the speed limit by more than 10%, this is respectively 63 and 8.5 times higher. The reduction of the speed limits through electronic signs does not only serve the purpose to reduce the driving speed, but also to warn drivers of the presence of an incident downstream. Therefore, it would be interesting to assess which effect DSLs have on the attention of the drivers and their behaviour when they approach an incident.
3% of the investment costs, thus to 564.45 kEUR. The system is exploited and maintained by the section Traffic Enforcement Systems of the Department Electromechanics and Telematics of the Road and Traffic Agency. The Department of Mobility and Public Works reckons that about six full time equivalents are needed for the exploitation of the entire system, which counts approximately 300 gantries. If we take the very conservative assumption that personnel costs are proportional to the number of gantries, the total net salary cost linked to the exploitation of the gantries under study is 23.57 kEUR (using standard salary costs from the Flemish public service for the personnel categories involved), and amounts thus to 4.2% of the maintenance costs. 4.3. Results In order to compare the costs and benefits of the system over its complete life cycle, we calculate its Net Present Value, defined as:
NPV = (B0 − C0) +
B1 − C1 B − Cn +⋯ n 1+r (1 + r )n
5.2. Cost-benefit analysis (2)
The cost-benefit analyses of the crash effects showed that, using unit values from the international literature for the valuation of crash prevention, the costs exceed the benefits, with a benefits-to-costs ratio of approximately 0.7. One can reasonably ask whether this small benefits-to-cost ratio is robust to changes in the discount rates and/or the expected economic lifetime of the installation. As argued in Section 4.2, the assumption of a lifetime of 25 years seems reasonable. However, even if the DSL would remain in operation for 50 years without major overhaul, it can easily be verified that the benefits-to-costs ratio would not exceed 0.87. With higher discount rates, the benefits-to-costs ratio would become even smaller. With a discount rate of 1%, the benefits-to-cost ratio for a 25years lifespan would be 0.87 as well. This implies that, for any reasonable assumptions regarding the lifespan of the system or the discount rates, the initial investment costs outweigh the expected traffic safety benefits. This benefits-to-cost ratio can be compared with values for other traffic safety measures, for instance such as discussed in SafetyNet (2009). Some conclusions from the review in SafetyNet (2009) were:
Where n is the expected lifetime of the system, Bi and Ci are the costs and the benefits in period i, respectively, and r is the discount rate. Following the official guidelines of the Flemish Government for conducting social cost-benefit analysis, we have used a discount rate of 4 percent. The net present value of investment costs, maintenance and exploitation costs is then 26.59 million EUR. The net present value of prevented crashes (using the values proposed by Korzhenevych et al. (2014)) is 19.32 million EUR – this corresponds to a benefits-to-costs ratio of approximately 0.7. 5. Discussion The present study focused on the effects of DSLs on the occurrence of crashes. In addition, the costs of the prevented crashes were compared with the costs of the implementation of these systems. 5.1. Evaluation of the traffic safety effect The study shows that the introduction of DSLs had a favourable effect on traffic safety. The number of injury crashes decreased by 18% and is significant (95% CI [−30%; −4%]). This effect is mainly attributable to a decrease in the number of rear-end crashes, for which an almost significant decrease of 20% (95% CI[−36%;+1%]) is found. A decrease is also found for the single-vehicle crashes (−15%), although not significant (95% CI [−36%;+13%]). An evaluation of the fatal and serious injury crashes shows a nonsignificant decrease of 6%. The lack of significance might be related to the sparsity of the data and therefore this result can hardly be interpreted. It would be interesting to execute a similar evaluation again later on, when more years of crash data are available in the after period and a larger number of Flemish motorways are equipped with the DSL systems. To the best of our knowledge, no previous empirical safety evaluations of DSL systems have been reported in the scientific literature. For now, the results of the current study will be the best available effect estimate for such system. This effect will have to be confirmed by additional research, preferably in different countries and settings. At this moment, it seems reasonable to assume that the effects of DSL systems are likely to be dependent on some implementation characteristics such as the applied algorithms and their input variables. In future research also the effects on speed behaviour should be analysed. The eventual traffic safety effects are likely to be dependent on the level of driver compliance (Habtemichael and de Picado Santos, 2013). It could be analysed to what extent drivers obey the DSLs and whether speed enforcement leads to higher speed compliance and furthermore to higher effects on the traffic safety level. A first analysis on
• For some traffic measures (such as infrastructural measures to re-
• •
duce speed), the benefit-to-cost ratios vary widely from country to country. Moreover, within a country, the ratios are very sensitive to the type of road or area to which the measures are applied. The ratio can vary from as much as 17 for German residential areas (Höhnscheid et al., 2006) to negative values in Sweden and Norway (Elvik, 2007; Elvik and Amundsen, 2000). For other measures (such as daytime running lights), estimated benefit-cost ratios range from 2 to 5, but these results have been found to be very sensitive to technical assumptions (such as the monetary valuation of safety) made by the researchers (see Knight et al., 2006). There are three fields where the findings do appear to be consistent across studies, though. Increased speed enforcement, intelligent speed adaptation and random breath testing are very cost-effective in all the studies that have been reviewed.
SafetyNet (2009) also gives an overview of 39 cost-effective road safety measures in Norway. The highest benefit-cost ratios are obtained for enhanced neck injury protection (20.25), seat belt reminders (16.21), lowering speed limit on hazardous roads (14.29) and alcolock for drivers convicted of drink-driving (8.75). Consequently, when compared with the traffic safety measures observed in SafetyNet (2009), the revealed benefits-to-cost ratio of the DSL systems appears to be rather weak. However, DSL systems might bring about more favourable effects than what was included in the cost-benefit analysis. Next to crash 6
Accident Analysis and Prevention xxx (xxxx) xxx–xxx
E. De Pauw et al.
effects, this measure could also have effects on traffic flows, congestion and travel times, and furthermore also on vehicle emissions and road noise. Nevertheless, no conclusive effects on any of these outcomes were found in previous implementations and experiments as described in Section 2. In addition, only the direct costs of crashes were taken into account in order to calculate the benefits. It can be argued that the prevention of crashes also leads to favourable effects on congestion levels, on emissions and fuel consumption. It was not possible to take these indirect effects into account. Furthermore, only injury crashes were analysed in this study. However the DSL systems are likely also to have had a favourable effect on property-damage-only crashes. Since information on the crashes is not gathered systematically, it was not possible to analyse the effects on this crash type, which also may lead to an underestimation of the benefits. Finally, it can be argued that due to economies of scale or further technological developments, costs for future installations might decrease, which could decrease installation costs and therefore shift the B/C ratio.
References Abdel-Aty, M., Dilmore, J., Dhindsa, A., 2006a. Evaluation of variable speed limits for real-time freeway safety improvement. Accid. Anal. Prev. 38 (2), 335–345. Abdel-Aty, M., Dilmore, J., Hsia, L., 2006b. Applying variable speed limits and the potential for crash migration. Transp. Res. Rec.: J. Transp. Res. Board 1953, 21–30. Abdel-Aty, M., Cunningham, R.J., Gayah, V.V., Hsia, L., 2008. Dynamic variable speed limit strategies for real-time crash risk reduction on freeways. Transp. Res. Rec.: J. Transp. Res. Board 2078, 108–116. Chang, G.-L., Park, S.Y., Paracha, J., 2011. Intelligent transportation system field demonstration: integration of variable speed limit control and travel time estimation for a recurrently congested highway. Transp. Res. Rec.: J. Transp. Res. Board 2243, 55–66. Corthout, R., Tampère, C.M.J., Deknudt, P., 2010. Assessment of variable speed limits from the drivers’ perspective. Intelligent Transportation Systems (ITSC), 2010 13th International IEEE Conference 499–506. De Pauw, E., Daniels, S., Brijs, T., Hermans, E., Wets, G., 2014. Safety effects of an extensive black spot treatment programme in Flanders-Belgium. Accid. Anal. Prev. 66, 72–79. DeGaspari, M., Jin, P.J., Wall, W.J., Walton, C.M., 2013. The effect of active traffic management on travel time reliability: a case study of I-5 in Seattle, washington. In: Presented at 92nd Annual Meeting of the Transportation Research Board. Washington, D.C. Elvik, R., Amundsen, A.H., 2000. Improving Road Safety in Sweden. Report 490. Institute of Transport Economics, Oslo quoted in SafetyNet (2009). Elvik, R., 2009. Cost-benefit analysis of safety measures for vulnerable and inexperienced road users. Work Package 5 of EU-project PROMISING. Report 435. Institute of Transport Economics, Oslo quoted in SafetyNet (2009). Elvik, R., 2002. The importance of confounding in observational before-and-after studies of road safety measures. Accid. Anal. Prev. 34, 631–635. Elvik, R., 2007. Prospects for improving road safety in Norway. A road safety impact assessment. Draft Report. Institute of Transport Economics, Oslo, Norway. Elvik, 2008. The predictive validity of empirical Bayes estimates of road safety. Accid. Anal. Prev. 40, 1964–1969. European Commission, 2010. Dynamic Speed Limits. Retrieved Nov 25, 2015, from. http://ec.europa.eu/index_en.htm. Fleiss, J.L., 1981. Statistical Methods for Rates and Proportions, Second Edition. John Wiley, New York. Flemish Traffic Control Centre [WWW Document], 2016. URL http://www.verkeerscentrum.be/verkeersinfo/verkeerscentrum/vc_middelen_rss (accessed 4.25.17). Fudala, N.J., Fontaine, M.D., 2010. Interaction between system design and operations of variable speed limit systems in work zones. Transp. Res. Rec.: J. Transp. Res. Board 2169, 1–10. Höhnscheid, K.J., et al., 2006. ROSEBUD Thematic Network. Examples of Assessed Road Safety Measures. A Short Handbook. Bundesanstalt für Strassenwesen, Bergisch Gladbach, Germany. Habtemichael, F.G., de Picado Santos, L., 2013. Safety and operational benefits of variable speed limits under different traffic conditions and driver compliance levels. Transp. Res. Rec.: J. Transp. Res. Board 2386, 7–15. Hauer, E., Harwood, D.W., Council, F.M., Griffith, M.S., 2002. Estimating safety by the empirical Bayes method: a tutorial. Transp. Res. Rec. 1784, 126–131. Hauer, E., 1997. Observational Before-After Studies in Road Safety: Estimating the Effect of Highway and Traffic Engineering Measures on Road Safety. Pergamon Press, Oxford. Hoogendoorn, S.P., Daamen, W., Hoogendoorn, R.G., Goemans, J.W., 2013. Assessment of dynamic speed limits on Freeway A20 near Rotterdam, Netherlands. Transp. Res. Rec.: J. Transp. Res. Board 2380, 61–71. Islam, M.T., Hadiuzzaman, M., Fang, J., Qiu, T.Z., El-Basyouny, K., 2013. Assessing mobility and safety impacts of a variable speed limit control strategy. Transp. Res. Rec.: J. Transp. Res. Board 2364, 1–11. Knight, I., Sexton, B., Bartlett, R., Barlow, T., Latham, S., McCrae, I., 2006. Daytime Running Lights (DRL): a review of the reports from the European Commission. TRL Report PPR 170. Transport Research Laboratory, Crowthorne. Korzhenevych, A., Dehnen, N., Bröcker, J., Holtkamp, M., Meier, H., Gibson, G., Varma, A., Cox, V., 2014. Update of the Handbook on External Costs of Transport, Final Report for the European Commission – DG MOVE. . http://ec.europa.eu/transport/ themes/sustainable/studies/doc/2014-handbook-external-costs-transport.pdf. Kwon, E., Brannan, D., Shouman, K., Isackson, C., Arseneau, B., 2007. Development and field evaluation of variable advisory speed limit system for work zones. Transp. Res. Rec.: J. Transp. Res. Board 2015, 12–18. Lee, C., Hellinga, B., Saccomanno, F., 2004. Assessing safety benefits of variable speed limits. Transp. Res. Rec.: J. Transp. Res. Board 1897, 183–190. Lee, C., Hellinga, B., Saccomanno, F., 2006. Evaluation of variable speed limits to improve traffic safety. Transp. Res. Part C 14 (3), 213–228. Lu, X.-Y., Shladover, S., 2014. Review of variable speed limits and advisories. Transp. Res. Rec. J. Transp. Res. Board 2423, 15–23. http://dx.doi.org/10.3141/2423-03. Magis, M., 2014. De effectiviteit van dynamische rijstrooksignalisatie in Vlaanderen. Hasselt University. OECD, 2006. Speed Management. OECD, Paris No. 55921. Papageorgiou, M., Kosmatopoulos, E., Papamichail, I., 2008. Effects of variable speed limits on motorway traffic flow. Transp. Res. Rec.: J. Transp. Res. Board 2047, 37–48. SafetyNet, 2009. Cost-Benefit Analysis. Retrieved 15 June 2017. https://ec.europa.eu/ transport/road_safety. Van Nes, N., Brandenburg, S., Twisk, D., 2010. Improving homogeneity by dynamic speed limit systems. Accid. Anal. Prev. 42, 944–952.
5.3. Developments Roadside DSL systems might have to compete in the near future with in-vehicle systems. Some providers of navigation and mapping products are already providing “Jam Ahead Warning” features on a commercial basis. A drawback of these systems compared to DSL is that they are currently only in use in a largely unknown proportion of vehicles, whereas the information of DSL is publicly and freely available. As a result, not all road users are warned of oncoming jams in case only in-vehicle systems would exist. Moreover, current in-vehicle systems have only informative value and don’t impose – in contrast to DSL’s – any obligation on the road user. For instance, they do not set enforceable speed restrictions. In the foreseeable future, the same function could be taken over by cooperative vehicle systems, i.e. techniques that allow vehicles to communicate with each other and with the roadway infrastructure. 6. Conclusions It can be concluded that the investigated DSLs have had a favourable effect on injury crashes, with a significant decrease of 18%. Mainly the number of rear-end crashes (−20%) decreased, albeit just nearly significantly. The number of single-vehicle crashes showed a tendency to decrease, but this effect was not significant. No effect was found on the number of side crashes. A cost-benefit analysis shows no convincing evidence that the expected safety benefits currently exceed the costs of the system with an estimated benefit to cost ratio of 0.7. However, if DSL systems could arguably improve traffic throughput, reduce travel times and reduce noise and mitigate emissions, this ratio might shift. This ratio might also shift if costs for future installations would decrease, e.g. due to technological developments or market evolutions. As it remains somewhat unclear whether DSL’s can also serve other purposes, apart from improving traffic safety, further research in this area is recommended. Further research should also assess whether different implementation schemes of DSL-systems (e.g. different underlying algorithms or different feed-back messages to drivers) lead to different effects. Acknowledgements This research was carried out within the framework of the Policy Research Centre on Traffic Safety and was partly supported by a grant from the Research Foundation Flanders. The content of this paper is the sole responsibility of the authors. 7
Contents lists available at ScienceDirect
Accident Analysis and Prevention journal homepage: www.elsevier.com/locate/aap
Safety effects of dynamic speed limits on motorways ⁎
Ellen De Pauwa, Stijn Danielsb,c, , Laurent Franckxd,1, Inge Mayeresd,e,2 a
Rondpunt, Uitbreidingstraat 518, 2600 Berchem, Belgium Belgian Road Safety Institute, Haachtsesteenweg 1405, 1130 Brussels, Belgium Hasselt University, Transportation Research Institute, Wetenschapspark 5, 3590 Diepenbeek, Belgium d VITO—Flemish Institute for Technological Research, Boeretang 200, 2400 Mol, Belgium e Katholieke Universiteit Leuven, Department of Economics, Naamsestraat 69, 3000 Leuven, Belgium b c
A R T I C L E I N F O
A B S T R A C T
Keywords: Safety Crash Motorway Dynamic speed limit Empirical bayes Cost-benefit
Dynamic speed limits (DSL) are limits that change according to real-time traffic, road or weather conditions. In DSL-schemes road users are typically informed of speed limit changes by electronic signs that are housed within gantries situated above lanes. Dynamic speed limit systems are increasingly applied worldwide, usually on motorways. One of the objectives of dynamic speed limits is to improve traffic safety through reductions in speed variations within and across lanes and between upstream and downstream flows. This paper shows the results of an empirical evaluation of the effects on traffic safety of a dynamic speed limit system on motorways in Flanders, Belgium. The evaluation was done by means of a before-after analysis of crashes, completed with a cost-benefit analysis. The results show that the number of injury crashes decreased significantly (−18%) after the introduction of the system. A separate analysis for serious and fatal injury crashes revealed a non-significant decrease of 6%. A distinction according to crash type showed an almost significant decrease of 20% in the number of rear-end crashes whereas the number of single-vehicle crashes decreased by 15% (ns). However, no effect was found for side crashes. In addition to the analysis of the effects, a cost-benefit analysis was applied. The costs of the implementation of these systems were compared with the benefits of crash prevention. The cost-benefit analyses of the crash effects showed a benefits-to-costs ratio of approximately 0.7, which means that the costs tend to exceed the benefits. Taking into account the important margins of uncertainty with respect to both costs and benefits, we have also explored how the net benefits are affected by some key assumptions. The general conclusion is that there is no convincing evidence that the costs of the system currently outweigh the expected benefits in terms of crash prevention.
1. Introduction
by generic legislation. In some countries the speed limit is reduced in case of rain, or speed limits nearby school zones are reduced at school start or end times. The focus of the present study is on DSLs, which are applied as a consequence of the real time situation. Through these systems, speed limits can be adapted remotely, either automatically by an algorithm or manually by an operator. This makes it possible to show different speed limits at different times of the day and different days of the week (Van Nes et al., 2010). DSLs are introduced to harmonize traffic flows which is assumed to improve both throughput and traffic safety. The traffic safety improvement is targeted through reductions in speed variations within and across lanes and between upstream and downstream flows
On the majority of the roads, fixed speed limits represent the appropriate speed for average conditions. However, in order to take account of the real time traffic, road and weather conditions, dynamic speed limits (DSLs) can be applied (European Commission, 2010). DSL systems are activated at a given time, as a consequence of traffic volume or other environmental conditions (Islam et al., 2013; OECD, 2006). Variable speed limits are often used as a synonym for DSLs. However, according to the OECD (2006) the term ‘variable speed limits’ refers to systems that are activated through general criteria (e.g. time of the day, season and certain weather conditions), which are usually set
⁎
Corresponding author at: Belgian Road Safety Institute, Haachtsesteenweg 1405, 1130 Brussels, Belgium. E-mail addresses: [email protected] (E. De Pauw), [email protected] (S. Daniels), [email protected] (L. Franckx), [email protected] (I. Mayeres). Present address: Belgian Federal Planning Bureau, Kunstlaan 47-49, 1000 Brussel, Belgium. 2 Present address: Transport & Mobility Leuven, Diestsesteenweg 57, 3010 Leuven, Belgium. 1
http://dx.doi.org/10.1016/j.aap.2017.06.013 Received 6 October 2016; Received in revised form 16 June 2017; Accepted 17 June 2017 0001-4575/ © 2017 Elsevier Ltd. All rights reserved.
Please cite this article as: Pauw, E.D., Accident Analysis and Prevention (2017), http://dx.doi.org/10.1016/j.aap.2017.06.013
Accident Analysis and Prevention xxx (xxxx) xxx–xxx
E. De Pauw et al.
This paper addresses the effects of DSL systems on traffic safety. As will be shown in Section 2, quite a number of studies analysed the effects of DSLs through simulation models and driving simulator studies. However, we didn’t find peer reviewed empirical studies that analysed the impact on traffic safety. Furthermore, to our knowledge, no literature about the cost-effectiveness of DSL systems is available. The present study therefore analyses the traffic safety effect of DSL systems in Flanders, Belgium, based on an empirical analysis of observed crash data. The effects are analysed through an empirical Bayes before-andafter study, which compares the crashes after the implementation of the measure with the number of crashes before, and controls for confounding variables. The empirical analysis is complemented by a costbenefit analysis of the applied DSL system. Section 2 provides an overview of previous studies that analysed the effects of DSL systems. Studies that analysed the effects on the traffic flow as well as studies that analysed the traffic safety effects are included in this section. In Section 3 the data, the methodology and the results of the evaluation of the effects on traffic safety are described; Section 4 describes the cost-benefit analysis. The results are discussed in Section 5 and conclusions are listed in Section 6.
DSLs to delay the onset of congestion and to help produce more rapid recovery from congestion, provided that demand volumes are not too far above the zone capacity. When demand volumes are high, they found no benefit over static speed limits. The simulation also showed the importance of appropriate DSL sign location and effective algorithm design. Habtemichael and de Picado Santos (2013) analysed the operational benefits of DSLs under different traffic conditions. They studied the combination of different compliance rates and congestion levels and found that the operational benefits depended on these two factors. The system had the highest operational benefits during lightly congested traffic conditions, little benefit during uncongested conditions, and no benefit during heavily congested conditions. Also impact of DSLs on traffic safety was assessed by means of simulation models in a number of papers. We briefly discuss them. Lee et al. (2004) used a real time crash prediction model integrated with a microscopic traffic simulation model. They found that temporarily reducing speed limits during risky traffic conditions can reduce the crash potential. The greatest reduction occurred at the location with a high traffic turbulence. Abdel-Aty et al. (2006a,b) studied how traffic safety could be increased at a motorway in Orlando. They found DSLs can be used to improve safety, through the implementation of lower speed limits upstream and higher speed limits downstream of the location where crash likelihood is observed in real time. This improvement was present in the case of medium-to-high-speed regimes but not in lowspeed situations. They furthermore analysed the potential for crash migration and found that the crash potential relocates to a location downstream of the detector of interest. Overall the safety of the freeway was improved (Abdel-Aty et al., 2006a,b). In a later study, Abdel-Aty et al. (2008) continued on this research and found that DSLs can be used to reduce crash risk and prevent crash occurrence in free-flow conditions and conditions approaching congestion. Habtemichael and de Picado Santos (2013) found however somewhat different results. In their study the most favourable traffic safety effects were found during highly congested traffic conditions, followed by lightly congested conditions and the least during uncongested situations. Furthermore, they found that the effects are highly dependent on the level of driver compliance. Lee et al. (2006) studied the safety benefits of DSLs and used simulated traffic conditions on a freeway in Toronto. They found that real time DSLs can reduce the overall crash potential by 5%–17%. Also Islam et al. (2013) analysed this impact. They proposed a model predictive DSL control strategy. The safety impact was quantified through a real time crash prediction model for an urban freeway corridor in Alberta. The results indicated that the DSL can improve safety by 50%.
2. Literature review
2.2. Field studies
A number of peer reviewed studies analysed the effects of DSL systems. These effects have mainly been stated in terms of traffic safety and traffic operations. In a recent paper, Lu and Shladover (2014) reviewed studies on DSLs and classified these studies as three types: simulations for algorithm development and evaluations, DSL implementation and field testing, and a combination of DSLs with ramp metering. We will briefly discuss the literature that has attempted to assess the effects of DSLs, by means of simulations as well as through field studies. Combinations of DSLs with ramp metering are outside the scope of this paper and not further treated.
The effect of DSL systems was also analysed through empirical studies. Lu and Shladover (2014) reviewed evidence from field studies that assessed effects on traffic flow and traffic safety in the UK, Germany, the Netherlands, France and the United States. The effects reported on traffic operations were mixed. Some authors reported a reduction of travel times (Chang et al., 2011; Hoogendoorn et al., 2013). Some studies reported an improvement in throughput (Chang et al., 2011; Kwon et al., 2007), but others not (DeGaspari et al., 2013). Papageorgiou et al. (2008) reported that effects on traffic flow are highly dependent on the saturation level of traffic with improved traffic flow at overcritical occupancies (dense traffic), but deteriorated traffic flow efficiency (lower average speeds) at undercritical traffic conditions. Multiple studies reported favourable effects on road safety for the implemented VSL strategies, but interestingly, none of the field studies that assessed effects on traffic safety has been published in peer reviewed journals and their methodological rigour is hard to assess. The road safety literature (Hauer, 1997; Elvik, 2002) has extensively reported that accident studies that don’t take into account confounding
(Lee et al., 2004; Islam et al., 2013). Throughput improvement is expected due to the speed harmonization benefits (Fudala and Fontaine, 2010). DSLs are sometimes also used in order to reduce vehicle emissions and road noise (Papageorgiou et al., 2008). DSL systems are also applied on motorways in the Flanders region in Belgium. The variable speed limit is denoted by electronic signs that are housed within gantries situated above motorway lanes. Speed limits imposed by DSL systems are compulsory and have – according to the road authority – three main objectives (Vlaams Verkeerscentrum, 2015): (1) Increase safety: upstream from an incident (e.g. traffic jams, crashes, road works) speed limits can be temporarily lowered in order to reduce the mean speed, lead the traffic smoothly to the incident and avoid the occurrence of crashes; (2) Indicate obstructions: DSLs can lead away traffic from a blocked lane; (3) Improve traffic flow through homogenization of speed: at moments with a high traffic flow, speed limits will be reduced, which will lead to a more homogeneous traffic flow and to less manoeuvres. Furthermore, headways are assumed to become smaller, which means that the available space is used more efficiently and the probability of traffic jams is subsequently lower.
2.1. Simulation models Many authors developed simulation models to assess the effects of DSLs. Islam et al. (2013) studied the effects on mobility. In the best case scenario, DSL control with a 5-min speed limit update frequency and a 10-km/h maximum speed difference between two successive time steps, they reported a 33% reduction of total travel time. Fudala and Fontaine (2010) did this for work zones specifically. They found potential of 2
Accident Analysis and Prevention xxx (xxxx) xxx–xxx
E. De Pauw et al.
factors such as trends and regression-to-the-mean tend to bias – usually to overstate – effects. Based on the reviewed evidence, Lu and Shladover (2014) concluded that DSLs can significantly improve freeway traffic safety if the compliance rate is high enough whereas the impact on traffic throughput is still controversial. The latter is explained by inherent difficulties in the observation of traffic flow status, by the large variations in driver behaviour and by the relative immaturity of DSL systems that often use algorithms that are unlikely to improve traffic flow. Hoogendoorn et al. (2013) evaluated effects of DSLs on emissions and noise. They reported a slight deterioration for both aspects.
5) the before period consisted of crashes from 1999 up to 2002 and the after period included crashes from 2006 until 2011. For the segment where a DSL system was installed in 2009 (segment 4) the before period included 2006–2008 and the after period 2010–2011. The before and after periods in the comparison group were each time the same as for the research group. Through this selection the most recent available years of crash data were used. The before period amounted to 3.8 years on average; the after period amounted to 5.2 years. The crashes were subdivided in two groups according to their severity: (1) injury crashes (all crashes with at least one injured person); (2) severe injury crashes, i.e. crashes with either severely injured persons (every person who needed more than 24 h of hospitalization as a result of a crash) or fatally injured persons (every person who died within 30 days after the crash as a consequence of the crash). Table 2 gives an overview of the average number of crashes at the treated locations per year for the full research period. These numbers clearly show that the rear-end crash is the crash type that occurs most frequently.
3. Evaluation of the traffic safety effect 3.1. Data In 2003 the first DSL systems were installed at Belgian motorways. Motorways are defined here as roads for motorized vehicles only with a median barrier and no at-grade junctions (Elvik et al., 2009). The general maximum speed limit on Belgian motorways is 120 km/h. Access to motorways is forbidden for pedestrians, cyclists, moped riders and all vehicles that cannot drive faster than 70 km/h. The speeds that are displayed by the dynamic signs are based on the data gathered by loop detectors and by automatic incident detection cameras. Data on speed and occupancy of the lane are used to set the speed limit. When the loops and cameras detect a high occupancy together with a low speed the speed limit is reduced. At the locations upstream from this incident the speed is gradually reduced in order to prevent sudden braking. Speed is reduced from 120 km/h to 110 km/h, 90 km/h, 70 km/h and 50 km/h as lowest speed limit. Weather conditions are not taken into account in the algorithm. Corthout et al. (2010) evaluated the DSL system around Antwerp and concluded that the performance of the system is good, with a reasonably high detection rate and very low false alarm rate. For the present study, police-reported crash data were available up until 2011. In order to have at least one year of crash data available in the after period, all DSL systems that were installed and operational up until 2010 were included in the study. In total five road segments with DSLs were assessed, covering a total distance of 59.54 km. These five segments are all located at access roads of the ring road of Antwerp. Table 1 gives an overview of the characteristics of the DSL locations, further called the ‘treated locations’. Fig. 1 gives a graphical overview of the locations. Each dot represents a gantry with DSLs. The comparison group, that was selected in order to control for the general trend effects, included all crashes that occurred on motorways in the Flanders region, at least 10 km away from locations with DSLs. All crashes from 1999 up to 2011 were used. Data were available from the national crash statistics (Statistics Belgium). The crashes from the year 2003 were excluded from the study, since there were generic problems with the crash registration in that year. Also the crashes from the years 2004–2005 were excluded since major road works were at that moment carried out on the ring road of Antwerp, nearby the five treated segments and thus possibly influencing traffic volumes or road user behaviour on the treated segments. Therefore, for the segments where DSL systems were installed in 2003–2004 (segments 1, 2, 3 and
3.2. Method The traffic safety effect is studied through an empirical Bayes (EB) before-and-after study, that compares the crash numbers after the implementation of the measure with the before situation. The EB approach increases the precision of estimation and corrects for the regression-tothe-mean (RTM) bias (Hauer et al., 2002). In a first step, the effect per segment is calculated (see Table 1, for more information on the location of the five segments). The analysis per segment (further referred to as ‘location’) can be expressed through an odds ratio, which results in an estimation of the index of effectiveness (θl): Ll
θl =
1 2 3 4 5
Length in km
Year of installation
No. of lanes
E17 Ghent-Antwerp E19 Breda-Antwerp E19 Brussels-Antwerp E313 Geel-Ranst E34/E313 Ranst-Antwerp
7.3 4.4 6.0 31.8 10.0
2004 2004 2003 2009 2003
3 2 3 2 2
N
M
(1)
E[κ|K]l = the expected number of crashes on the treated location l during the before period, controlled for RTM Ll = the observed number of crashes on the treated location l during the after period M = the observed number of crashes in the comparison group during the before period N = the observed number of crashes in the comparison group during the after period In order to increase the precision of the estimates, the empirical Bayes method makes joint use of two information sources on the safety of a location: the crash record of that location and the crash frequency expected at similar locations. A weighted average combines these two sources (Hauer et al., 2002): E[κ|K]l = w * E[κ] * T + (1 − w) * Kl
(2)
with w = the weight (between 0 and 1) that is given to the crashes at similar entities E[κ] = average number of crashes in the before period at similar entities T = number of years during the before period 1 − w = the weight given to the crashes at the treated location L Kl = observed number of crashes in the before period at the treated location L. To calculate the average number of crashes at similar entities (E[κ]) in Eq. (2), a Safety Performance Function (SPF) was used that earlier has been developed for crash occurrence of injury crashes on Flemish motorways. This study developed an SPF per motorway segment on the basis of several variables: (1) traffic volume, (2) length of road segment, (3) type of road segment and (4) number of lanes. The type of road segment included two main categories: road segments at entries/exits
Table 1 Characteristics of the treated segments. Road segment
E[κ K]l
3
Accident Analysis and Prevention xxx (xxxx) xxx–xxx
E. De Pauw et al.
Fig. 1. Geographical location of the treated segments. Table 2 Average number of crashes/year that occurred on the treated locations during the before and the after period.
Table 3 Results of the SPF. Injury crashes
Before After
Injury crashes
Severe crashes
Rear-end injury crashes
Single-vehicle injury crashes
Side injury crashes
108.7 61.3
22.9 15.4
51.1 31.3
31.6 17.7
13.7 8.3
and interchanges and road segments between two entries/exits or interchanges. The model has a negative binomial probability distribution with a log link function and is calculated using the SPSS GENLIN procedure.
−16.792 (SE:0.623)
Length of segment (β) Traffic volume (γ) Over dispersion Likelihood ratio test statistic (χ2)
0.939 (SE:0.029)*** 1.011 (SE:0.049)*** 0.313 (SE:0.031) 1013.5***
−18.493 (SE: 1.030)*** 0.951 (SE: 0.047)*** 1.035 (SE: 0.080)*** 0.325 (SE:0.070) 500.6***
*** significant at 1% level.
period to the annual average number of crashes that occurred in 2008–2010. To calculate the weight (w) in Eq. (2), next equation can be used:
2
E[κ] = eαL βV γe∑i= 1 δixi
α
Severe crashes ***
(3)
1
w=
with E[κ] = expected annual number of crashes α, β, γ, δ = model parameters L = length of road segment (in m) V = Traffic volume (in vehicles/24 h) x1 = Segment type (0 = at entries/exits and interchanges, 1 = between entries/exits and interchanges) x2 = Number of lanes {1,2,3} Traffic volume data were gathered through double inductive loops in the pavement. All segments with built-in inductive loops at all Flemish motorways were included in the SPF. In total 292 segments were included in the SPF for 2008, 381 segments for 2009 and 544 segments for 2010. The road segments included in the SPF have an average length of 2448 m (SD 2726 m) and an average daily traffic volume of 36047 (SD 20045) vehicles. Table 3 displays the final results of the SPFs, split up for all injury crashes and the severe crashes. The segment type and the number of lanes were not significant and thus not included in the final model. The model was based on crash data from 2008 to 2010. However, the before period in the present study was 1999–2002 for the locations at which DSL systems were installed in 2003 or 2004 and was 2006–2008 for the location at which DSLs were installed in 2009. Therefore the estimated number of crashes was multiplied with an adjustment factor to match the time frame of the observed data with the time frame of the SPF. This adjustment factor was expressed as the proportion of the annual average number of crashes during the before
1+
E[κ] *T k
(4)
with k the inverse value of the overdispersion parameter of the model, which is estimated per unit of length (Elvik, 2008).
3.2.1. Overall effect estimation The evaluation of each location separately has only limited significance. Therefore a fixed effects meta-analysis was carried out, which resulted in one overall effect estimate and shows more statistically reliable outcomes (Fleiss, 1981). Every location within the meta-analysis receives a weight, which is the inverted value of the variance. The variance is, based on the variables in Eq. (1), calculated as next:
sl ² =
1 1 1 1 + + + E[κ|K]l Ll M N
(5)
The weight can be calculated as follows:
wl =
1 s2l
(6)
Subsequently locations at which many crashes occurred are given higher weights. Supposing that the measure is executed at n different places, the weighted mean index of effectiveness of the measure over all places θ is as follows: 4
Accident Analysis and Prevention xxx (xxxx) xxx–xxx
E. De Pauw et al.
Table 4 Results of meta-analyses of the crash effects.
Table 5 Victim related costs according to Korzhenevych et al. (2014). Effect [95%CI]
Injury crashes Rear-end crashes Side crashes Single-vehicle crashes Severe crashes
0.82 0.80 1.00 0.85 0.94
[0.70; 0.96]* [0.64;1.01] [0.64; 1.56] [0.64;1.13] [0.68; 1.29]
n
(7)
The estimation of a 95% CI is as follows: n
⎡∑ wl *ln(θl ) 95%CI = exp ⎢ l=1 n ± 1.96* ⎢ wl ∑ l=1 ⎢ ⎣
1 n
∑l=1
⎤ ⎥ wl ⎥ ⎥ ⎦
Estimate in kEUROs
Fatality Severe injury Slight injury
2178.0 330.4 21.3
slightly injured persons was 78.50, the average annual number of severely injured persons was 22.50, and the number of fatalities was 2.50. In order to estimate the counterfactual, we will first apply the central values of the estimates of the index of effectiveness (see Table 4). For “severe injuries” and “fatalities”, we will apply the estimate for “severe crashes” (0.94), while for the number of slight injuries, we apply the estimate for “injury crashes” (0.82). This corresponds to 95.73 slightly injured persons, 23.94 severely injured persons and 2.66 fatalities in the counterfactual scenario. Combining the unit cost figures from Table 5 with the estimated number of fatalities and injuries avoided, we obtain the estimates of the annual economic value of crash prevention due to DSL, as described in Table 6.
* significant at 5% level.
wl *ln(θl ) ⎤ ⎡∑ θ = exp ⎢ l=1 n ⎥ ⎢ ∑l=1 wl ⎥⎦ ⎣
Cost category
(8)
A similar approach was used in De Pauw et al. (2014). 3.3. Results
4.2. The social costs of dynamic speed limits
The results of the meta-analyses are provided in Table 4. The metaanalysis of the injury crashes shows a significant decrease of 18% as a result of the implementation of DSL systems. The injury crashes were subdivided according to the type of crash and the three main crash types were analysed: (1) rear-end crashes; (2) side crashes; (3) singlevehicle crashes. As can be seen from Table 4, no effect on the number of side crashes is found. A decrease is found for the number of rear-end crashes (−20%) which is significant at the 10% level and almost significant at the 5% level as the upper limit of the 95% confidence interval is close to one. The number of single-vehicle crashes decreased by 15%, which is however not significant. An analysis of the fatal and serious injury crashes shows no significant effects.
In order to estimate the costs of DSL, we need to consider both investment costs and operational costs. In the absence of detailed cost accounts of the DSL, we have used indepth interviews with two officials from the Department of Mobility and Public Works as main source of information. One should realize that these figures are expert judgments of the interviewees. However, we esteem them to be a valid indication of the order of magnitude of the costs. Unless stated otherwise, all the figures below are based directly on the information we obtained during the interviews. The cost of installing the equipment is highly location specific. However, 269 kEUR was reported to us as being representative for the material investment costs per km of highway covered (including maintenance costs in the first two years of operation). To this material investment cost, we have added an estimated salary cost of the supervising personnel (47 kEUR per km). As taxes are a transfer, all these sums are net of VAT, income taxes and wage taxes. With an average distance between gantries of 750 m, this results in 237 kEUR as a representative investment cost per gantry. As a total distance of 59.54 km is covered by the studied DSL (see Section 3.1), this implies an initial investment cost of 18,815 kEUR. There are no official estimates of the economic lifetime of the installations. The oldest dynamic displays and DSL systems have been in use since 1999 and 2003, respectively. On the one hand, there is no indication that they will need to be replaced in the foreseeable future, and no major overhaul of the system is planned. On the other hand, taking into account current developments in the domain of automated and connected mobility, it seems reasonable to assume that, within a decade or so, the type of technology will be largely obsolete. We therefore assume a lifetime of 25 years. However, as we shall argue below, the principal results are robust for other assumptions. According to the Department of Mobility and Public Works, total annual maintenance costs (both preventive and curative) correspond to
4. Cost-benefit analysis As already discussed above, DSL systems are introduced for several reasons. We will stick here to the general approach of the paper, and limit ourselves to their benefits in terms of crash reduction. 4.1. The social benefits of the reduced crash risk In this section we calculate the social benefits of the reduced crash risk that was estimated in Table 4. First, we present the monetary values of the social costs of crashes that are proposed in the literature. Next, we apply these values to the impacts of DSL in Flanders. We use the figures from the ‘Update of the Handbook on External Costs of Transport’ (Korzhenevych et al., 2014) for the unit costs of crashes with injuries and fatalities. The advantage of this approach is that these figures have been obtained using an internationally recognized methodology. The following unit values are reported for Belgium (p 23), using the market prices (at Purchasing Power Parity3) of 2010. In order to translate these unit costs in total economic benefits of crash prevention, we also need estimates of the number of victims and/ or crashes, both the actual number and the number that would have prevailed in the absence of DSL systems (the counterfactual). In 2010–2011, on the treated locations, the average number of
Table 6 Total value of prevented injuries in caused by DSL in Flanders, Belgium.
3 In contrast to market exchange rates, Purchasing Power Parities reflect the differences in purchasing power between two currencies. They are thus the relevant metric to convert costs and benefits expressed in one currency in another currency.
5
Cost category
Estimate in kEUROs
Fatalities Severe injuries Slight injuries Total
348 475 367 1189
Accident Analysis and Prevention xxx (xxxx) xxx–xxx
E. De Pauw et al.
Flemish motorways (Magis, 2014) indicated that only one fourth of the drivers comply with the posted speed limit of 90 km/h and half of them keep the speed limit of 110 km/h. In comparison to moments with a fixed speed limit of 120 km/h, the average speed is only 6 km/h lower at moments with a dynamic speed limit of 90 km/h and is even 0.4 km/ h higher at moments with a speed limit of 110 km/h. Compared to moments with a speed limit of 120 km/h, the number of drivers that exceed the speed limit is 18 times higher during moments with a speed limit of 90 km/h and 5 times higher when the speed limit is 110 km/h. For the number of drivers that exceed the speed limit by more than 10%, this is respectively 63 and 8.5 times higher. The reduction of the speed limits through electronic signs does not only serve the purpose to reduce the driving speed, but also to warn drivers of the presence of an incident downstream. Therefore, it would be interesting to assess which effect DSLs have on the attention of the drivers and their behaviour when they approach an incident.
3% of the investment costs, thus to 564.45 kEUR. The system is exploited and maintained by the section Traffic Enforcement Systems of the Department Electromechanics and Telematics of the Road and Traffic Agency. The Department of Mobility and Public Works reckons that about six full time equivalents are needed for the exploitation of the entire system, which counts approximately 300 gantries. If we take the very conservative assumption that personnel costs are proportional to the number of gantries, the total net salary cost linked to the exploitation of the gantries under study is 23.57 kEUR (using standard salary costs from the Flemish public service for the personnel categories involved), and amounts thus to 4.2% of the maintenance costs. 4.3. Results In order to compare the costs and benefits of the system over its complete life cycle, we calculate its Net Present Value, defined as:
NPV = (B0 − C0) +
B1 − C1 B − Cn +⋯ n 1+r (1 + r )n
5.2. Cost-benefit analysis (2)
The cost-benefit analyses of the crash effects showed that, using unit values from the international literature for the valuation of crash prevention, the costs exceed the benefits, with a benefits-to-costs ratio of approximately 0.7. One can reasonably ask whether this small benefits-to-cost ratio is robust to changes in the discount rates and/or the expected economic lifetime of the installation. As argued in Section 4.2, the assumption of a lifetime of 25 years seems reasonable. However, even if the DSL would remain in operation for 50 years without major overhaul, it can easily be verified that the benefits-to-costs ratio would not exceed 0.87. With higher discount rates, the benefits-to-costs ratio would become even smaller. With a discount rate of 1%, the benefits-to-cost ratio for a 25years lifespan would be 0.87 as well. This implies that, for any reasonable assumptions regarding the lifespan of the system or the discount rates, the initial investment costs outweigh the expected traffic safety benefits. This benefits-to-cost ratio can be compared with values for other traffic safety measures, for instance such as discussed in SafetyNet (2009). Some conclusions from the review in SafetyNet (2009) were:
Where n is the expected lifetime of the system, Bi and Ci are the costs and the benefits in period i, respectively, and r is the discount rate. Following the official guidelines of the Flemish Government for conducting social cost-benefit analysis, we have used a discount rate of 4 percent. The net present value of investment costs, maintenance and exploitation costs is then 26.59 million EUR. The net present value of prevented crashes (using the values proposed by Korzhenevych et al. (2014)) is 19.32 million EUR – this corresponds to a benefits-to-costs ratio of approximately 0.7. 5. Discussion The present study focused on the effects of DSLs on the occurrence of crashes. In addition, the costs of the prevented crashes were compared with the costs of the implementation of these systems. 5.1. Evaluation of the traffic safety effect The study shows that the introduction of DSLs had a favourable effect on traffic safety. The number of injury crashes decreased by 18% and is significant (95% CI [−30%; −4%]). This effect is mainly attributable to a decrease in the number of rear-end crashes, for which an almost significant decrease of 20% (95% CI[−36%;+1%]) is found. A decrease is also found for the single-vehicle crashes (−15%), although not significant (95% CI [−36%;+13%]). An evaluation of the fatal and serious injury crashes shows a nonsignificant decrease of 6%. The lack of significance might be related to the sparsity of the data and therefore this result can hardly be interpreted. It would be interesting to execute a similar evaluation again later on, when more years of crash data are available in the after period and a larger number of Flemish motorways are equipped with the DSL systems. To the best of our knowledge, no previous empirical safety evaluations of DSL systems have been reported in the scientific literature. For now, the results of the current study will be the best available effect estimate for such system. This effect will have to be confirmed by additional research, preferably in different countries and settings. At this moment, it seems reasonable to assume that the effects of DSL systems are likely to be dependent on some implementation characteristics such as the applied algorithms and their input variables. In future research also the effects on speed behaviour should be analysed. The eventual traffic safety effects are likely to be dependent on the level of driver compliance (Habtemichael and de Picado Santos, 2013). It could be analysed to what extent drivers obey the DSLs and whether speed enforcement leads to higher speed compliance and furthermore to higher effects on the traffic safety level. A first analysis on
• For some traffic measures (such as infrastructural measures to re-
• •
duce speed), the benefit-to-cost ratios vary widely from country to country. Moreover, within a country, the ratios are very sensitive to the type of road or area to which the measures are applied. The ratio can vary from as much as 17 for German residential areas (Höhnscheid et al., 2006) to negative values in Sweden and Norway (Elvik, 2007; Elvik and Amundsen, 2000). For other measures (such as daytime running lights), estimated benefit-cost ratios range from 2 to 5, but these results have been found to be very sensitive to technical assumptions (such as the monetary valuation of safety) made by the researchers (see Knight et al., 2006). There are three fields where the findings do appear to be consistent across studies, though. Increased speed enforcement, intelligent speed adaptation and random breath testing are very cost-effective in all the studies that have been reviewed.
SafetyNet (2009) also gives an overview of 39 cost-effective road safety measures in Norway. The highest benefit-cost ratios are obtained for enhanced neck injury protection (20.25), seat belt reminders (16.21), lowering speed limit on hazardous roads (14.29) and alcolock for drivers convicted of drink-driving (8.75). Consequently, when compared with the traffic safety measures observed in SafetyNet (2009), the revealed benefits-to-cost ratio of the DSL systems appears to be rather weak. However, DSL systems might bring about more favourable effects than what was included in the cost-benefit analysis. Next to crash 6
Accident Analysis and Prevention xxx (xxxx) xxx–xxx
E. De Pauw et al.
effects, this measure could also have effects on traffic flows, congestion and travel times, and furthermore also on vehicle emissions and road noise. Nevertheless, no conclusive effects on any of these outcomes were found in previous implementations and experiments as described in Section 2. In addition, only the direct costs of crashes were taken into account in order to calculate the benefits. It can be argued that the prevention of crashes also leads to favourable effects on congestion levels, on emissions and fuel consumption. It was not possible to take these indirect effects into account. Furthermore, only injury crashes were analysed in this study. However the DSL systems are likely also to have had a favourable effect on property-damage-only crashes. Since information on the crashes is not gathered systematically, it was not possible to analyse the effects on this crash type, which also may lead to an underestimation of the benefits. Finally, it can be argued that due to economies of scale or further technological developments, costs for future installations might decrease, which could decrease installation costs and therefore shift the B/C ratio.
References Abdel-Aty, M., Dilmore, J., Dhindsa, A., 2006a. Evaluation of variable speed limits for real-time freeway safety improvement. Accid. Anal. Prev. 38 (2), 335–345. Abdel-Aty, M., Dilmore, J., Hsia, L., 2006b. Applying variable speed limits and the potential for crash migration. Transp. Res. Rec.: J. Transp. Res. Board 1953, 21–30. Abdel-Aty, M., Cunningham, R.J., Gayah, V.V., Hsia, L., 2008. Dynamic variable speed limit strategies for real-time crash risk reduction on freeways. Transp. Res. Rec.: J. Transp. Res. Board 2078, 108–116. Chang, G.-L., Park, S.Y., Paracha, J., 2011. Intelligent transportation system field demonstration: integration of variable speed limit control and travel time estimation for a recurrently congested highway. Transp. Res. Rec.: J. Transp. Res. Board 2243, 55–66. Corthout, R., Tampère, C.M.J., Deknudt, P., 2010. Assessment of variable speed limits from the drivers’ perspective. Intelligent Transportation Systems (ITSC), 2010 13th International IEEE Conference 499–506. De Pauw, E., Daniels, S., Brijs, T., Hermans, E., Wets, G., 2014. Safety effects of an extensive black spot treatment programme in Flanders-Belgium. Accid. Anal. Prev. 66, 72–79. DeGaspari, M., Jin, P.J., Wall, W.J., Walton, C.M., 2013. The effect of active traffic management on travel time reliability: a case study of I-5 in Seattle, washington. In: Presented at 92nd Annual Meeting of the Transportation Research Board. Washington, D.C. Elvik, R., Amundsen, A.H., 2000. Improving Road Safety in Sweden. Report 490. Institute of Transport Economics, Oslo quoted in SafetyNet (2009). Elvik, R., 2009. Cost-benefit analysis of safety measures for vulnerable and inexperienced road users. Work Package 5 of EU-project PROMISING. Report 435. Institute of Transport Economics, Oslo quoted in SafetyNet (2009). Elvik, R., 2002. The importance of confounding in observational before-and-after studies of road safety measures. Accid. Anal. Prev. 34, 631–635. Elvik, R., 2007. Prospects for improving road safety in Norway. A road safety impact assessment. Draft Report. Institute of Transport Economics, Oslo, Norway. Elvik, 2008. The predictive validity of empirical Bayes estimates of road safety. Accid. Anal. Prev. 40, 1964–1969. European Commission, 2010. Dynamic Speed Limits. Retrieved Nov 25, 2015, from. http://ec.europa.eu/index_en.htm. Fleiss, J.L., 1981. Statistical Methods for Rates and Proportions, Second Edition. John Wiley, New York. Flemish Traffic Control Centre [WWW Document], 2016. URL http://www.verkeerscentrum.be/verkeersinfo/verkeerscentrum/vc_middelen_rss (accessed 4.25.17). Fudala, N.J., Fontaine, M.D., 2010. Interaction between system design and operations of variable speed limit systems in work zones. Transp. Res. Rec.: J. Transp. Res. Board 2169, 1–10. Höhnscheid, K.J., et al., 2006. ROSEBUD Thematic Network. Examples of Assessed Road Safety Measures. A Short Handbook. Bundesanstalt für Strassenwesen, Bergisch Gladbach, Germany. Habtemichael, F.G., de Picado Santos, L., 2013. Safety and operational benefits of variable speed limits under different traffic conditions and driver compliance levels. Transp. Res. Rec.: J. Transp. Res. Board 2386, 7–15. Hauer, E., Harwood, D.W., Council, F.M., Griffith, M.S., 2002. Estimating safety by the empirical Bayes method: a tutorial. Transp. Res. Rec. 1784, 126–131. Hauer, E., 1997. Observational Before-After Studies in Road Safety: Estimating the Effect of Highway and Traffic Engineering Measures on Road Safety. Pergamon Press, Oxford. Hoogendoorn, S.P., Daamen, W., Hoogendoorn, R.G., Goemans, J.W., 2013. Assessment of dynamic speed limits on Freeway A20 near Rotterdam, Netherlands. Transp. Res. Rec.: J. Transp. Res. Board 2380, 61–71. Islam, M.T., Hadiuzzaman, M., Fang, J., Qiu, T.Z., El-Basyouny, K., 2013. Assessing mobility and safety impacts of a variable speed limit control strategy. Transp. Res. Rec.: J. Transp. Res. Board 2364, 1–11. Knight, I., Sexton, B., Bartlett, R., Barlow, T., Latham, S., McCrae, I., 2006. Daytime Running Lights (DRL): a review of the reports from the European Commission. TRL Report PPR 170. Transport Research Laboratory, Crowthorne. Korzhenevych, A., Dehnen, N., Bröcker, J., Holtkamp, M., Meier, H., Gibson, G., Varma, A., Cox, V., 2014. Update of the Handbook on External Costs of Transport, Final Report for the European Commission – DG MOVE. . http://ec.europa.eu/transport/ themes/sustainable/studies/doc/2014-handbook-external-costs-transport.pdf. Kwon, E., Brannan, D., Shouman, K., Isackson, C., Arseneau, B., 2007. Development and field evaluation of variable advisory speed limit system for work zones. Transp. Res. Rec.: J. Transp. Res. Board 2015, 12–18. Lee, C., Hellinga, B., Saccomanno, F., 2004. Assessing safety benefits of variable speed limits. Transp. Res. Rec.: J. Transp. Res. Board 1897, 183–190. Lee, C., Hellinga, B., Saccomanno, F., 2006. Evaluation of variable speed limits to improve traffic safety. Transp. Res. Part C 14 (3), 213–228. Lu, X.-Y., Shladover, S., 2014. Review of variable speed limits and advisories. Transp. Res. Rec. J. Transp. Res. Board 2423, 15–23. http://dx.doi.org/10.3141/2423-03. Magis, M., 2014. De effectiviteit van dynamische rijstrooksignalisatie in Vlaanderen. Hasselt University. OECD, 2006. Speed Management. OECD, Paris No. 55921. Papageorgiou, M., Kosmatopoulos, E., Papamichail, I., 2008. Effects of variable speed limits on motorway traffic flow. Transp. Res. Rec.: J. Transp. Res. Board 2047, 37–48. SafetyNet, 2009. Cost-Benefit Analysis. Retrieved 15 June 2017. https://ec.europa.eu/ transport/road_safety. Van Nes, N., Brandenburg, S., Twisk, D., 2010. Improving homogeneity by dynamic speed limit systems. Accid. Anal. Prev. 42, 944–952.
5.3. Developments Roadside DSL systems might have to compete in the near future with in-vehicle systems. Some providers of navigation and mapping products are already providing “Jam Ahead Warning” features on a commercial basis. A drawback of these systems compared to DSL is that they are currently only in use in a largely unknown proportion of vehicles, whereas the information of DSL is publicly and freely available. As a result, not all road users are warned of oncoming jams in case only in-vehicle systems would exist. Moreover, current in-vehicle systems have only informative value and don’t impose – in contrast to DSL’s – any obligation on the road user. For instance, they do not set enforceable speed restrictions. In the foreseeable future, the same function could be taken over by cooperative vehicle systems, i.e. techniques that allow vehicles to communicate with each other and with the roadway infrastructure. 6. Conclusions It can be concluded that the investigated DSLs have had a favourable effect on injury crashes, with a significant decrease of 18%. Mainly the number of rear-end crashes (−20%) decreased, albeit just nearly significantly. The number of single-vehicle crashes showed a tendency to decrease, but this effect was not significant. No effect was found on the number of side crashes. A cost-benefit analysis shows no convincing evidence that the expected safety benefits currently exceed the costs of the system with an estimated benefit to cost ratio of 0.7. However, if DSL systems could arguably improve traffic throughput, reduce travel times and reduce noise and mitigate emissions, this ratio might shift. This ratio might also shift if costs for future installations would decrease, e.g. due to technological developments or market evolutions. As it remains somewhat unclear whether DSL’s can also serve other purposes, apart from improving traffic safety, further research in this area is recommended. Further research should also assess whether different implementation schemes of DSL-systems (e.g. different underlying algorithms or different feed-back messages to drivers) lead to different effects. Acknowledgements This research was carried out within the framework of the Policy Research Centre on Traffic Safety and was partly supported by a grant from the Research Foundation Flanders. The content of this paper is the sole responsibility of the authors. 7