S

ET

a

ARRA

KBCST

1

(f2grtapcatithbwco

s

0h

Accident Analysis and Prevention 62 (2014) 426– 431

Contents lists available at ScienceDirect

Accident Analysis and Prevention

jo u r n al hom epa ge: www.elsev ier .com/ locate /aap

afety effects of reducing the speed limit from 90 km/h to 70 km/h

llen De Pauw ∗, Stijn Daniels, Melissa Thierie, Tom Brijsransportation Research Institute, Hasselt University, Wetenschapspark 5, BE-3590 Diepenbeek, Belgium

r t i c l e i n f o

rticle history:eceived 14 December 2012eceived in revised form 30 April 2013ccepted 2 May 2013

eywords:efore- and after evaluationrashespeed limit reductionraffic safety

a b s t r a c t

Speed is one of the main risk factors in traffic safety, as it increases both the chances and the severity of acrash. In order to achieve improved traffic safety by influencing the speed of travel, road authorities maydecide to lower the legally imposed speed limits. In 2001 the Flemish government decided to lower speedlimits from 90 km/h to 70 km/h on a considerable number of highways. The present study examines theeffectiveness of this measure using a comparison group before- and after study to account for generaltrend effects in road safety. Sixty-one road sections with a total length of 116 km were included. The speedlimits for those locations were restricted in 2001 and 2002. The comparison group consisted of 19 roadsections with a total length of 53 km and an unchanged speed limit of 90 km/h throughout the researchperiod. Taking trend into account, the analyses showed a 5% decrease [0.88; 1.03] in the crash rates

after the speed limit restriction. A greater effect was identified in the case of crashes involving seriousinjuries and fatalities, which showed a decrease of 33% [0.57; 0.79]. Separate analyses between crashesat intersections and at road sections showed a higher effectiveness at road sections. It can be concludedfrom this study that speed limit restrictions do have a favorable effect on traffic safety, especially onsevere crashes. Future research should examine the cause for the difference in the effect between roadsections and intersections that was identified, taking vehicle speeds into account.

. Introduction

Speed is defined as an important risk factor in traffic safetyElvik and Vaa, 2004). Although crashes are caused by differentactors and it is difficult to examine the role of speed (Nilsson,004), higher speeds are proven to increase the likelihood ofetting involved in a crash. Different causes can contribute to thiselationship. One of these causes is that drivers have less time toake in information and react, which leads to lower chances ofvoiding a crash. At the same time the car covers an extensivelyrolonged distance before it stops. Not only is there an increasedhance of getting involved in a crash, but the severity of the crashlso increases with speed, as the degree of kinetic energy at theime of the collision is higher (OECD, 2006). In order to favorablynfluence traffic safety, the Flemish government decided to lowerhe speed limits from 90 km/h to 70 km/h on a large number ofighways, based on four main criteria, at least one of which had toe met. Those criteria were: road sections without cycle paths or

ith cycle lanes close to the roadway; road sections with obstacles

lose to the roadways with a high risk of collision; road sectionsutside urban areas but with a high building density, and a high

∗ Corresponding author. Tel.: +32 11269111.E-mail addresses: ellen.depauw@uhasselt.be (E. De Pauw),

tijn.daniels@uhasselt.be (S. Daniels), tom.brijs@uhasselt.be (T. Brijs).

001-4575/$ – see front matter © 2013 Elsevier Ltd. All rights reserved.ttp://dx.doi.org/10.1016/j.aap.2013.05.003

© 2013 Elsevier Ltd. All rights reserved.

number of vulnerable road users; and road sections on whichseveral severe crashes occurred in the past (Juvyns, pers. comm.).The speed limit was often only restricted at specific sections ofroads, for example at sections between two intersections or atsections between two parts of an urban environment. The speedlimit reduction was introduced for the majority of the locationsin 2001–2002. No enforcement and educational efforts werecombined with this change; only the traffic signs were adapted.

2. Background

Previous studies that examined the traffic safety effects of speedlimit restriction, commonly showed a favorable effect on trafficsafety. A review of Elvik (2009), who analyzed 115 studies with526 estimates, generally found a decrease in crash numbers whenthe speed limit was reduced. The studies that did find unfavorableeffects on traffic safety, were for the most part small studies fromwhich the results are rather unreliable. In the fixed-effect statis-tical weight these counted for only 5.7%. The effect on crashes isoften expressed through a power function to which the differencein speed has to be raised (Elvik et al., 2004; Nilsson, 2004). Elvik(2009) revised this Power Model and made a distinction between

rural roads and freeways on the one hand and urban and residen-tial roads on the other hand. For the category of the freeways/ruralroads, which are also the type of roads that are included in thepresent study, a power estimate of 4.1 was found for fatal crashes.

sis and Prevention 62 (2014) 426– 431 427

Foctts(lftnttif

ttfiIsoacm

3

t2ipoairtttrr1osatlT1ltTar

uHt1o1yeu

Table 1Main characteristics of the treated and comparison locations (Agency of Roads andTraffic, Ministry of Mobility and Public Works).

Number of locations (%)

Treated group Comparison group

Road categoryPrimary 3 (5%) 1 (5%)Secondary 9 (15%) 5 (26%)Local 49 (80%) 13 (68%)Urban areaInside 17 (28%) 2 (10%)Outside 44 (72%) 17 (90%)Number of lanes2 × 1 56 (92%) 16 (84%)2 × 2 4 (7%) 0 (0%)

selected for the treated group, with a proportion of 48% at inter-sections. On average, the comparison group comprised 21 severecrashes, with 37% occurring at intersections. An initial view is given

E. De Pauw et al. / Accident Analy

or serious injury crashes he found a power of 2.60. The analysisf all injury crashes, without a distinction to the severity of therash, resulted in a power of 1.6. When these powers are appliedo the change in speed from 90 km/h to 70 km/h this would leado a decrease of 64% in the number of fatal crashes, 48% in theevere injury crashes and 33% in all injury crashes. In addition, Elvik2013) analyzed this relationship, according to the initial speedimit, through the application of two models: (1) an exponentialunction; (2) the Power Model. A slightly higher support was giveno the exponential function, which showed an increase of 1.58 in theumber of fatal crashes if speed increased with 1 km/h from an ini-ial speed of 85 km/h. The number of injury crashes was estimatedo increase with 1.21. Starting from an initial speed of 75 km/h anncrease of 0.79 fatal crashes with an increase by 1 km/h was found,or the injury crashes this was 0.86.

A restriction in the obeyed speed limit will not necessarily leado a proportional effect on driving speeds. McCarthy (1998) showedhat a lot of factors can mediate the effect of speed limits on traf-c safety; in particular, the driver’s chosen speed is important.

n turn, this choice is influenced by different elements, such asocio-economic factors, personal risk perception and the extentf enforcement. In addition, road conditions and the vehicle haven effect. When a speed limit is not in accordance with the roadonditions, this limit will not be maintained or it will barely beaintained.

. Data

The treated group included all road sections that had a reduc-ion in the speed limit from 90 km/h to 70 km/h during 2001 and002 located in the province of Limburg, one of the five provinces

n Flanders, Belgium. Road sections on which other measures wereerformed during the research period that could have had an effectn travel speeds or traffic safety were excluded. Therefore, localuthorities were asked to report whether, in addition to lower-ng the speed limit, other measures were implemented during theesearch period. Examples of possible other treatments are changeso traffic regulations such as the right-of-way rules and changes inhe infrastructure, such as narrowing or broadening roads. Loca-ions that only had some small changes in infrastructure, such asepair and maintenance works, were not excluded. Eventually, 61oad sections were included with a total length of 116 km, located in6 different municipalities in the province of Limburg. The lengthf the sections ranged from 0.1 to 6.04 km. For most of the roadections a speed limit restriction was applied in 2002, 13 had andaptation in 2001. The comparison group consisted of 19 sec-ions, with a total length of 53 km. Furthermore, the comparisonocations were all located in the province of Limburg. As shown inable 1, most of the road sections (80%) are situated at local roads,5% are secondary roads that connect, collect and distribute at the

ocal and intercity level, 5% are primary roads which have the func-ion of connection, collection and distribution at the Flemish level.he majority of the road sections are situated outside the urbanrea (72%), and have 2 × 1 lanes (92%). Picture 1 shows examples ofoads that were adapted.

At the time of the study, crash data for Belgium was available upntil 2009 (Federal Public Service Economy, Statistics Department).owever, geo-coded crash data was required, in order to select

he crashes at the treated locations. This data was available from996 until 2007 (Ministry of Mobility and Public Works, Agencyf Roads and Traffic). Subsequently, the before period starts from

996 to 2000/2001, the after period from 2002/2003 to 2007. Theear during which the speed limit was adapted, 2001 or 2002, wasxcluded from the research period. Two groups of crash data weresed, based on the severity of crashes. The first group included all

3 × 1 1 (2%) 3 (16%)

injury crashes, which are all crashes in which at least one personwas slightly injured. Crashes with property damage only were notincluded, as data on these crashes is not gathered systematically. Inthe case of the second group, a subgroup of the injury crashes wasselected, and only crashes with seriously injured persons (definedas any person who got involved in a traffic crash and needed a hos-pitalization for more than 24 h) and fatally injured persons (anyperson who, as a result of a traffic crash, died on the spot or within30 days of the crash) were included. The spatial analysis programArcGIS version 9.3 was used to select the crashes. A buffer of 10 mwas applied to make sure all crashes at the selected locations wereincluded. Furthermore, a distinction was made according to thelocation of the crash, with a distinction between road section andintersection. On average, 322 injury crashes per annum occurred atthe treated locations. Fifty-five percent took place at intersections,45% at road sections. The comparison group consisted of 64 injurycrashes per year, with an occurrence of 44% at intersections. In thecase of severe crashes an average of 74 crashes per annum were

Picture 1. Examples of roads at which the speed limit was restricted from 90 km/hto 70 km/h.

Source: Google Street View.

428 E. De Pauw et al. / Accident Analysis and Prevention 62 (2014) 426– 431

Table 2Odds ratios (OR) and standard deviations (s) for all injury crashes and severe crashes for all years before the implementation of the measure.

Total OR (s) Intersections OR (s) Road sections OR (s)

Injury crashes1996–1997 1.39 (0.20) 0.95 (0.28) 2.05 (0.29)1997–1998 0.92 (0.20) 1.00 (0.27) 0.84 (0.29)1998–1999 0.93 (0.18) 0.94 (0.25) 0.90 (0.26)1999–2000 0.95 (0.18) 1.25 (0.26) 0.76 (0.24)

Average 1.05 (0.19) 1.03 (0.27) 1.14 (0.27)

Fatal and serious injury crashes1996–1997 1.62 (0.32) 1.07 (0.51) 2.23 (0.42)1997–1998 0.96 (0.34) 0.75 (0.49) 1.23 (0.49)1998–1999 0.70 (0.32) 0.85 (0.44) 0.56 (0.47)1999–2000 0.81 (0.29) 1.15 (0.

Average 1.02 (0.32) 0.96 (0.

D1

iacit

s

4

4

c(aaseacuocccm

iagram 1. Mean crash numbers per km in the treated and comparison group from996 to 2007, both for injury crashes and more severe crashes.

n Diagram 1, which shows the mean crash rates per km, both forll injury crashes and the more severe crashes in the treated andomparison group. A decrease can be observed for all groups andn the case of severe crashes in particular, this is stronger in thereated group, when compared to the comparison group.

No data was available in relation to traffic volumes or travelpeeds at the treated and comparison locations.

. Study design

.1. Before- and after study

To evaluate the effectiveness of a traffic safety measure, the mostommonly used study design is a before- and after (B&A) studyElvik, 2002; Shinar, 2007), which compares the number of crashesfter the implementation of a measure with the number of crashest the same location before the implementation. Within those B&Atudies, different methods can be used, which mainly differ in thextent they control for confounding variables. A confounding vari-ble is defined as any exogenous variable affecting the number ofrashes or injuries whose effects, if not estimated, can be mixedp with the measure under evaluation (Elvik, 2002). A first groupf B&A studies are the simple B&A methods, which compare therash rates after the measure with before, without control for any

onfounding variable. A second group are the B&A methods thatontrol for trend effects. As before to after the measure different ele-ents are changed that could have had an autonomous effect on the

length of roads in research group with a certain clength of roads in comparison group with a certain c

45) 0.64 (0.39)

47) 1.16 (0.45)

occurrence of crashes, it is important to take those effects intoaccount. These elements are for example traffic safety campaigns,stronger enforcement, adaptations of infrastructure, changes intraffic volume, etc. A third group are the Bayesian methods, of whichone is the empirical Bayes (EB) method. Preference is given to thisEB method as it controls for the most important confounding fac-tors (Elvik, 2002; Hauer, 1997), such as common trend, effects ofchance and the regression to the mean (RTM) phenomenon.

4.2. Comparability of the comparison group

The comparison group has an important role in B&A studies, asit is applied to control for the trend and often also for RTM. Thelocations in the comparison group have to be similar to the treatedgroup on a couple of characteristics, that is for example geometricdesign, traffic volumes and vehicle fleet (Persaud & Lyon, 2007). Thecomparability of the comparison group can be examined throughthe calculation of the odds-ratio (OR) for the crash rates during theyears before the measure.

OR = Rt/Rt−1

Ct/Ct−1(1)

Rt = number of crashes in the treated group in year t; Rt−1 = numberof crashes in the treated group in year t − 1; Ct = number of crashesin the comparison group in year t; Ct−1 = number of crashes in thecomparison group in year t − 1.

When the OR is near to 1, the comparison group is compara-ble to the treated locations. Maximum standard deviation shouldnot be higher than 0.20 (Hauer, 1997). The ORs are calculated forthe total comparison group for the years before the speed limitrestriction. The results of these calculations are shown in Table 2.The odds ratios for the injury crashes show that the comparisongroup is comparable to the treated group. A subdivision betweencrashes that occurred at intersections and at road sections, shows aslightly better comparability for crashes that occurred at intersec-tions compared to road sections. The ORs for fatal and serious injurycrashes are less comparable, with an OR more different from 1, andstandard deviations that exceed 0.20. This can be explained by thelow number of crashes, where small fluctuations result in higherrelative changes. However, on average, the odds ratios are near toone, and the use of the total comparison group can be consideredas comparable with the treated group.

The qualitative characteristics, as shown in Table 1 can alsobe compared. Therefore, a comparison is made for the classifica-tion and urbanization of roads. The following equation is used to

examine this:

haracteristic/total length of roads in research groupharacteristic/total length of roads in comparison group

(2)

E. De Pauw et al. / Accident Analysis and Prevention 62 (2014) 426– 431 429

Table 3Results of the B&A study with correction for trend effects.

Total Intersections Road sections

# eff < 1 # eff > 1 # eff < 1 # eff > 1 # eff < 1 # eff > 1

Analysis per locationInjury crashes 38 (62%) 23 (38%) 26 (43%) 35 (57%) 43 (70%) 18 (30%)Severe crashes 41 (67%) 20 (33%) 30 (49%) 31 (51%) 41 (67%) 13 (21%)

Total Intersections Road sectionsEff [95% CI] Eff [95% CI] Eff [95% CI]

filttrstRipRr

ftllamarstbtsittssicbpcataaet

5

5

p(f

Meta-analysisInjury crashes 0.95 [0.88; 1.03]

Severe crashes 0.67 [0.57; 0.79]

Five equations were calculated: three for the functional classi-cation of roads (local, secondary and primary), and two for the

evel of urbanization (inside or outside built-up areas). In ordero analyze whether differences are significant, the Fisher’s exactest is calculated. The comparison group is more or less compa-able with the treated group for local roads (1.12; p = 0.6707);econdary roads (0.83; p = 0.2619) are slightly underrepresented inhe treated group, in common with primary roads (0.74; p = 0.4262).oads outside urban areas are comparable (0.92; p = 0.1708), roads

nside urban areas are overrepresented in the treated group (1.50; = 0.3579). However, none of these differences were significant.egarding the results of those analyses and the calculated oddsatios, we consider the comparison group to be acceptable.

However, in addition to these analyses, the absolute crash ratesrom the treated and comparison group need some specific atten-ion. As can be seen from Diagram 1, the crash rates at the treatedocations are systematically higher than those at the comparisonocations, both for the period before and after the measure. This isn important remark, which has consequences for the evaluationethod. Preference goes to the EB method, as this controls for RTM

nd trend effects. However, because of this discrepancy in crashates between the treated and comparison group, it was not pos-ible to use the EB method. This method controls for RTM throughhe estimation of the number of crashes that would have occurredefore, instead of using the recorded number before. This estima-ion is based on statistical analyses, using a comparison group or onome available crash prediction model. In this study the compar-son group encompasses much lower crash rates compared to thereated group. Diagram 1 shows considerable differences betweenhe mean numbers of crashes in the treated group and the compari-on group throughout the full period 1996–2000. These differenceseem to be rather structural, as the speed limit reduction was onlyntroduced starting from 2001. Consequently, an EB estimation ofrash rates in the period before implementation of the measure,ased on this comparison group, would result in a biased (in theresent case: an unreasonably low) estimated number of crashesompared to the recorded crash rates. It was not possible to selectnother comparison group, as there were no other locations withinhe province of Limburg with an unchanged speed limit of 90 km/hnd no information could be obtained from locations elsewhere. As

result, no attempt could be made to correct any RTM-bias and thevaluation was continued as a B&A study with comparison groupo account for trend effects.

. Method

.1. Evaluation per location

The effectiveness of lowering the speed limit is first calculateder location, and can be expressed in an index of effectiveness (Eff)Hauer, 1997), which shows the relative change in the crash ratesrom before to after. When the index is lower than 1, this shows

1.11 [1.00; 1.23] 0.89 [0.80; 0.99]0.94 [0.73; 1.20] 0.64 [0.52; 0.73]

that the crashes decreased and the measure had a favorable effecton traffic safety. An index higher than 1 indicates a higher crash rateafter the implementation of the measure, compared to before. Theequation has to be adapted for trend effects. Therefore, it is assumedthat the treated locations followed the same trend as the compari-son group. This trend is reflected by the evolution of the crash ratesfrom before to after in the comparison group. Consequently, theeffect estimate can be expressed as:

Effl = Lafter

Lbefore + (Cafter/Cbefore)= Lafter/Lbefore

Cafter/Cbefore(3)

Lafter = number of crashes on location L after the measure;Lbefore = number of crashes on location L before the measure;Cafter = number of crashes in the comparison group after the mea-sure; Cbefore = number of crashes in the comparison group beforethe measure.

The reliability of this estimate is assessed by the 95% confidenceinterval (CI):

95% CI, lower limit = exp[ln(eff) − 1.96 × s]

95% CI, upper limit = exp[ln(eff) + 1.96 × s](4)

The standard deviation (s) is the root of the variance (s2):

s2 = 1Lafter

+ 1Lbefore

+ 1Cafter

+ 1Cbefore

(5)

5.2. Effectiveness across different locations

In addition to an individual analysis per location, a meta-analysis– which calculates the overall effect of all locations with an adaptedspeed limit – can be carried out (Fleiss, 1981). To count the overalleffect, every location is given a value that is inversely proportionalto the variance:

W = 1s2

(6)

Suppose that the measure was implemented at m different loca-tions, the weighted mean index of effectiveness of the measure overall locations (EFF) is:

EFF = exp

[∑ml=1wl × ln(effl)∑m

l=1wl

](7)

The estimation of a 95% confidence interval is:

95% CI EFF = exp

⎡⎣

∑ml=1wl × ln(effl)∑m

l=1wl

± 1.96 × 1√∑ml=1wl

⎤⎦ (8)

6. Results

The results of the analyses are shown in Table 3. When each loca-tion is analyzed separately, a decrease in injury crashes is found at

4 sis an

67tdrootlh

adoatfafdscchs

7

mlmttuchofiwfttAoweroce

wttcllHsoilmp

30 E. De Pauw et al. / Accident Analy

2% of the locations after lowering the speed limit from 90 km/h to0 km/h. Furthermore, a separate analysis is carried out for crasheshat occurred at intersections and at road sections. There was aecrease in crash rates at intersections at 43% of the locations. Atoad sections a decrease is found at 70% of the locations. In the casef the fatal and serious injury crashes, a decrease is found at 67%f the locations. A distinction between road sections and intersec-ions showed a decrease in severe crashes at 49% and 67% of theocations, respectively. At the road sections 7 locations (12%) alsoad an index equal to 1.

The meta-analysis for the total number of injury crashes showed decrease in crash rates of 5% after lowering the speed limit. Thisecrease was only significant at the 25% level. For crashes thatccurred at intersections, an increase of 11% is found, significantt the 10% level. On the contrary, analysis of crashes at road sec-ions resulted in a significant decrease of 11%. A meta-analysisor the more severe crashes showed a significant decrease of 33%t all treated locations. This strong decrease was mainly foundor crashes that occurred at road sections, for which a significantecrease of 36% was found. The severe crashes at intersectionshowed a decrease of 6%, which was non-significant. These resultslearly show a more favorable effect was found for the severerashes, compared to the total number of injury crashes. Also aigher effectiveness is found for the occurrence of crashes at roadections compared to intersections.

. Discussion

Ideally, a traffic safety measure should be evaluated using the EBethod (Elvik, 2002; Hauer, 1997). Since the crash rates are much

ower in the comparison group, compared to the treated group, thisethod was not applicable here, as it would lead to biased estima-

ions. One possible explanation for this discrepancy between thereated and comparison group arises from the criteria that weresed to select the road sections for reducing the speed limits. Oneriterion was roads that are localized outside urban areas, but stillave a high building density. Locations in the comparison group,n the other hand, are mainly situated at rural roads, where traf-c volumes may be lower compared to the treated locations. Itas not possible to select another comparison group and there-

ore, it was not feasible to use the EB method to its full extent ando control for RTM. However, roads that had a speed limit restric-ion were not mainly selected on the basis of a high crash count.

range of different criteria were used: the absence of cycle pathsr paths close to the roadway, presence of obstacles close to road-ays or a high number of vulnerable road users. From this it can be

xpected that the occurrence of RTM is possible, though probablyather limited in size as the registered number of crashes was onlyne of several criteria for inclusion. Moreover the before periodounted 5–6 years, which probably considerably reduced the RTMffect.

Traffic volume data was not taken into account, since this dataas not available. This volume data could have given an explana-

ion of the low crash rates in the comparison group compared tohe treated group. Next to this it would have been interesting toompare the traffic volumes after the implementation of the speedimit restriction with before, in order to analyze whether the speedimit restriction had an effect on the route choice of the drivers.owever, we can expect this effect was limited, due to the road

ystem structure in Flanders. This structure does only give limitedpportunity for drivers to choose alternative roads, as these mainly

nclude local roads with a speed limit of 70 or 50 km/h. The speedimit restriction from 90 km/h to 70 km/h, which was mainly imple-

ented at short road sections at the upper category of roads, willrobably have had a limited effect on the rerouting choices of the

d Prevention 62 (2014) 426– 431

driver. Supposing that traffic volumes would have been available,a model could be fitted, through which a better estimation of thecrash numbers during the before period would be possible, andsubsequently RTM could be controlled. Furthermore changes in thetraffic volumes as a consequence of the measure could be analyzed.Since these traffic volumes were not available, it is assumed theRTM effect is small and no changes in traffic volumes occurred dueto the installation of speed cameras. However, these can be consid-ered as reasonable assumptions. Therefore, the present results atthe meta-level are regarded as the best possible estimation of theresults that would be found, if exposure data were available. Nev-ertheless, it was possible to take general trend effects into accountusing a comparison group.

The analyses clearly showed a higher effectiveness for moresevere crashes with serious injuries and fatalities compared toall injury crashes. This can be ascribed to the fact that speed isdirectly related to injury severity in a crash. This is different thanthe probability of being involved in a crash, which is more com-plex, as the occurrence of crashes can seldom be attributed to asingle factor (Transportation research board, 1998). The analysesalso showed a stronger effectiveness at road sections compared tointersections, both for injury crashes, for which even a contradic-tory was found, and the more severe crashes. This is more difficult toexplain. Crashes that occur at intersections may be less influencedby speed compared to road stretches, and causation might rather berelated to maneuvers, for example turning left. This explains whyno decrease was found, but this does not explain why an increase isfound. A possible cause for this increase in the number of crashes,is the increase in the variance of travel speeds.

Lowering the speed limit will not automatically lead to achange in travel speeds by all drivers. Factors such as habits, non-acceptance of the new measure or inattentiveness might explainwhy the actual speed adaptation is lower than the required speedadaptation (McCarthy, 1998). Furthermore, the speed limit changewas only communicated by adapting traffic signs. Parker (1997)stated that changing posted speed limits alone, without additionalenforcement, educational programs or other engineering meas-ures, only has a minor effect on driver behavior. Furthermorethe infrastructure of the road was not adapted, which makes itless appealing for drivers to adapt their behavior, whereas oth-ers will strictly follow the rules. This can lead to an increasein the variance in travel speeds, which is an important riskfactor for the occurrence of crashes (Aarts and Van Schagen,2006).

In turn, changes in speed behavior will not necessarily resultin an equivalent effect on traffic safety. As formulated by Nilsson(2004), Elvik et al. (2004) and Elvik (2009, 2013), this relationshipcan be expressed by power estimations of the difference in speeds.In a back-of-the-envelope calculation this theory can be comparedwith the results from the present study, and the power estima-tions can be applied to observed travel speeds at Flemish roads.For this calculation the power estimations are used that resultedfrom the study by Elvik (2009), from which we applied the powerestimations for rural roads and freeways, as these are also thetype of roads that are included in the present study. In 2007 themean speed at 70 km/h roads was 75 km/h, at roads with a limit of90 km/h this was 82.5 km/h. The V85 was respectively 95.1 km/h and85.6 km/h (Riguelle, 2009). Using the power estimations on thesespeeds resulted in an estimated decrease in crash rates between14% and 35%, as shown in Table 4.

The results from our analyses are less favorable with respectto all injury crashes. In the case of severe crashes the results aremore in line with these theoretically expected results. However,

this reasoning lacks validity due to its non-experimental setting.At the least, we would recommend a more detailed analysis of thespeed behavior on the roads in question.

E. De Pauw et al. / Accident Analysis an

Table 4Estimation of effect in number of crashes using power estimations by Elvik (2009)for mean and V85 speeds at 90 and 70 km/h roads in Flanders in 2007.

Mean speeds82.5 → 75 km/h

V85 speeds95.1 → 85.6 km/h

Fatal injury crashes −32% −35%

8

sf

-

-

R

A

Serious injury crashes −22% −24%Injury crashes −14% −15%

. Conclusions

The data for the study sample suggests that the restriction of thepeed limit from 90 km/h to 70 km/h at highways in Flanders has aavorable effect on crashes, mainly the most severe ones.

With respect to injury crashes, a more favorable effect was foundfor crashes that occurred at road sections, for which a decreasewas found, whereas an increase was identified for crashes atintersections. Changes in variances of travel speeds may be acausal factor. However, future research should examine this ina more in-depth manner.

This study was unable to obtain data on travel speeds. Futureresearch is required to examine the relationship between thespeed limit and the travel speeds of the driver, and the effect ontraffic volumes.

eferences

arts, L., Van Schagen, I., 2006. Driving speed and the risk of road crashes: a review.Accident Analysis & Prevention 38, 215–224.

d Prevention 62 (2014) 426– 431 431

Elvik, R., 2002. The importance of confounding in observational before-and-after studies of road safety measures. Accident Analysis & Prevention 34,631–635.

Elvik, R., 2009. The Power Model of the Relationship Between Speed and Road Safety.Update and New Analyses (No. TOI Report 1034/2009). Institute of TransportEconomics, Norwegian Centre for Transport Research, Oslo.

Elvik, R., 2013. A re-parameterisation of the Power Model of the relationshipbetween the speed of traffic and the number of accidents and accident victims.Accident Analysis & Prevention 50, 854–860.

Elvik, R., Vaa, T., 2004. The Handbook of Road Safety Measures. Elsevier, Oxford.Elvik, R., Christensen, P., Amundsen, A., 2004. Speed and Road Accidents. An Eval-

uation of the Power Model (No. 740/2004). Institute of Transport Economics,Oslo.

Fleiss, J.L., 1981. Statistical Methods for Rates and Proportions. John Wiley, NewYork.

Hauer, E., 1997. Observational Before–after Studies in Road Safety: Estimating theEffect of Highway and Traffic Engineering Measures on Road Safety. Pergamon,Kidlington.

McCarthy, P., 1998. Effect of speed limits on speed distributions and highway safety:a survey of literature in managing speed. In: Review of Current Practice for Set-ting and Enforcing Speed Limits (No. Special Report 254). Perdue University,Department of Economics, Indiana.

Nilsson, G., 2004. Traffic Speed Dimensions and the Power Model to Describe theEffect of Speed on Safety. Lund Institute of Technology, Lund.

Organization for Economic Co-operation and Development (OECD), 2006. SpeedManagement (No. 55921). Organization for Economic Co-operation and Devel-opment, Paris.

Parker, M.R., 1997. Effect of Raising and Lowering Speed Limits on Selected RoadwaySections (No. FHWA-RD-92-084). U.S. Department of Transportation, FederalHighway Administration, McLean, Virginia.

Persaud, B., Lyon, C., 2007. Empirical Bayes before-after safety studies: lessonslearned from two decades of experience and future directions. Accident Analysis& Prevention 39, 546–555.

Riguelle, F., 2009. Nationale gedragsmeting snelheid 2003–2007 (No.

D/2009/0779/68). BIVV, Observatorium voor de verkeersveiligheid, Brussel.

Shinar, D., 2007. Traffic Safety and Human Behavior. Elsevier, Oxford.Transportation Research Board, National Research Council, 2007. Managing speed.

In: Review of Current Practice for Setting and Enforcing Speed Limits (No. SpecialReport 254). National Academy Press, Washington, DC.