Accident Analysis and Prevention 60 (2013) 384 395

Contents lists available at ScienceDirect

Accident Analysis and Prevention

jo u r n al hom epa ge: www.elsev ier .co

Road safety performance indicators for the interu

George Y , MaAntonis C evea National Techb SWOV Instituc Technion, Israd LNEC Nationa

a r t i c l

Article history:Received 30 DReceived in reAccepted 17 N

Keywords:Road safetyPerformance indicatorsRoad networkPilot studies

catorour (ed netI for tntialne w

safety. This paper presents a theoretical concept for the determination of this SPI as well as a translationof this theory into a practical method. Also, the method is applied in a number of pilot countries namelythe Netherlands, Portugal, Greece and Israel. The results show that the SPI could be efciently calculatedin all countries, despite some differences in the data sources. In general, the calculated overall SPI scoreswere realistic and ranged from 81 to 94%, with the exception of Greece where the SPI was relatively lower(67%). However, the SPI should be considered as a rst attempt to determine the safety level of the road

1. Introdu

In orderroad safety numbers ofever, are ofas they depthe road tralties someTherefore, athe safety ctheir progre

Road safures (indicroad trafc (Hakkert etsafety cond

Corresponof TransportatAthens, Greece

E-mail add

0001-4575/$ http://dx.doi.onetwork. The proposed method has some limitations and could be further improved. The paper presentsdirections for further research to further develop the SPI.

2012 Elsevier Ltd. All rights reserved.

ction

to monitor the efciency of road safety measures andprogress in general, the most common indicators are the

accidents, fatalities and injuries. These numbers, how-ten not sufcient to illustrate the level of road safety,ict the worst case of unsafe operational conditions ofafc system. Moreover, counts of accidents and casu-times do not reveal the processes that produce them.dditional safety indicators are required for assessingonditions of a road trafc system and for monitoringss (ETSC, 2001; Wegman et al., 2008).ety performance indicators (SPIs) are dened as meas-ators) that reect the operational conditions of thesystem that inuence the systems safety performance

al., 2007). The SPIs purpose is to reect the currentitions of a road trafc system; to measure the inuence

ding author at: National Technical University of Athens, Departmention Planning and Engineering, 5 Heroon Polytechniou Str., GR-15773. Tel.: +30 2107721326; fax: +30 2107721454.ress: geyannis@central.ntua.gr (G. Yannis).

of various safety interventions and to enable comparisons betweendifferent road trafc systems (e.g. countries, regions, etc.). SPIsmay also provide information about underlying causes of accidents.They may concern particular groups of road users e.g. children, newdrivers or professional drivers, or compliance with important safetyrules e.g. seat belt use, or cover specic areas such as the urban roadnetwork or the trans-European network (EC, 2003).

SPIs have been increasingly researched during the last years,with particular emphasis on methodological and data issues(Wegman et al., 2008; Assum and Srensen, 2010). Moreover, thereis a rapid development of composite indicators e.g. indexes whichare a combination of individual indicators (De Leur and Sayed,2002; Hermans et al., 2008; Gitelman et al., 2010), since the mul-tidisciplinary character of road safety implies that policymakersshould take various inuential factors into account.

Within SafetyNet, a 6th FP European Integrated Project, safetyperformance indicators were developed for seven road safetyrelated areas, namely alcohol and drug-use, speeds, protective sys-tems, daytime running lights, vehicles (passive safety), traumamanagement and roads (Hakkert and Gitelman, 2007). An exhaus-tive literature review on SPIs (SafetyNet, 2005) was carried out inthe SafetyNet project, and demonstrated that there were no SPIsfor road networks in use. However, based on the Sustainable Safety

see front matter 2012 Elsevier Ltd. All rights reserved.rg/10.1016/j.aap.2012.11.012annisa,, Wendy Weijermarsb, Victoria Gitelmanc

hazirisa, Eleonora Papadimitrioua, Carlos Lima Aznical University of Athens, Greecete for Road Safety Research, The Netherlandsel Institute of Technology, Israell Laboratory for Civil Engineering, Portugal

e i n f o

ecember 2011vised form 9 August 2012ovember 2012

a b s t r a c t

Various road safety performance indiareas, mainly as regards driver behavisafety); however, no SPIs for the roaresearch is the development of an SPcomparisons. The developed SPI essetheoretically required one, dened as om/locate /aap

rban road network

rtijn Visb,dod

s (SPIs) have been proposed for different road safety research.g. seat belt use, alcohol, drugs, etc.) and vehicles (e.g. passivework and design have been developed. The objective of thishe road network, to be used as a benchmark for cross-regionly makes a comparison of the existing road network to thehich meets some minimum requirements with respect to road

G. Yannis et al. / Accident Analysis and Prevention 60 (2013) 384 395 385

approach (Koornstra et al., 1992; CROW, 1997), Dijkstra (2003) pro-posed a framework to assess network and design quality aspects ofa safe road infrastructure at the regional level.

The objective of the research presented in this paper was todevelop andifferent cocountry cosystem stroinfrastructuapply sampaccident prehazards, i.eenvironmenroads aim tlayout and rconsidered method prolate the roaRoad Asses2003, 2004)the road nethe SPI andof national region. Themethodolog

2. SPI deve

Wegmantial for a safprinciples athat roads ahierarchicaspeeds are in differentother (hombehaviour (predictabilforgiving wple).

From throad netwo

The road right placof a roadnetwork.

The road safe way.

Within tels. This redesign SPI. Aroad design

2.1. Road d

The Eurdesigned asAssessmentAccording tfor roads shvehicle safefor safety, wOne of the

infrastructure safety is the EuroRAP Road Protection Score (RPS).The RPS is a measure for the protection that is provided in rela-tion to three main accident types: run-off road, head-on impactsand severe impacts at intersections. EuroRAP designed a method to

te th star

or vae: spnd pnd inroRAay, theneieeds

RPSype ds. Type futch an, 2road

are ation orkP.

ad n

s secn 2.2pecied in

SPI c

d ne

roadent

ructua (20d ne

whicfetyyingfety (utch

meth Thes typeV ha

ent roadhe riactuamati

=

Ti iss, frories,

is thi; LTgory SPI for the road network, which may be applicable inuntries and may be used as a benchmark for cross-

mparisons. In general, the safety of a road transportngly depends on the layout and design of the roadre. Many ongoing practices in infrastructure researchling of casualty data for safety assessment. In addition,vention can be improved by early assessments of safety. by monitoring the physical appearance of the roadt and the operational conditions of trafc. Thus, SPIs for

o assess the safety hazards in relation to infrastructureoad design. Therefore, within SafetyNet, two SPIs werefor roads: a road network SPI and a road design SPI. Theposed by Dijkstra (2003) was used as a basis to formu-d network SPIs, where for road design SPI the Europeansment Programme (EuroRAP) experience (Lynam et al.,

was examined. The paper presents the rationale behindtwork SPI, the methodology developed for estimating

the results of its rst application through a numberpilot studies, each one focusing on a different country

SPI is still in the development phase and the proposedy is just a rst suggestion that could be improved.

lopment for road networks

and Aarts (2006) discuss ve principles that are essen-e performance of a road transport system. Four of thesere relevant for the road network. First, it is importantre monofunctional and are part of a road network that islly structured (functionality principle). Second, in casehigh, different types of road users and trafc driving

directions should be physically separated from eachogeneity principle). Third, road course and road usershould be predictable by a recognizable road designity principle). Fourth, the road environment should behen an accident occurs (physical forgivingness princi-

is perspective, it is important to assess the safety of ark at two levels:

network level; the right road should be located at thee from a functional point of view, i.e. the road category

should be appropriate given its function in the road

design level: individual roads should be designed in a

he SafetyNet project, SPIs were developed for both lev-sults in two SPIs: the road network SPI and the roadlthough this paper focuses on the road network SPI, the

SPI will be discussed briey for completeness purposes.

esign SPI

opean Road Assessment Programme (EuroRAP) was a complementary activity to the European New Car

Programme (EuroNCAP), developed in the 1990s.o EuroRAP (Lynam et al., 2003, 2004) a rating systemould help optimize the combined effect of road andty. EuroRAP was piloted to rate Europes various roadshile currently it is distributed throughout the world.

tools suggested by the EuroRAP for estimating road

calculato fourclassesistic ar(type ations athe Euthis whomoghigh sp

Thesame tfor roaprototto a DGitelmof the scoresmentaframewEuroRA

2.2. Ro

Thi(Sectiomore sdescribof the

3. Roa

TheassessminfrastDijkstring roathose road sain studand sain the Ditative1988).centreof FGSassessm

Theis on tpriate mathe

SPI(Ti)

whereegoriecategoSPI(Ti)egory Tof catee RPS for each road segment or route, expressed in ones, depending on a number of road characteristics. Thelues that are used for the scoring of each road character-eed limit, median treatment, hard obstacles or barrierslacement), road site areas (cut and embankment), junc-tersections (type and access). For more information onP RPS see Lynam et al. (2003, 2004) and iRAP (2009). Ine EuroRAP RPS assesses to some extent whether the

ty principle (separation of directions at medium and) and the physical forgivingness principle are met.

focuses on the road design and appears to relate to theof philosophy that is being aimed at in SafetyNet SPIsherefore, it was decided to adopt the EuroRAP RPS as aor the road design SPI. The road design SPI was appliedcase study (Duivenvoorden et al., 2007; Hakkert and007), which demonstrated the possibility of calculation

design SPI for the road network once the EuroRAP RPSvailable for each road section. However, further imple-of the EuroRAP RPS scores is required for a co-operative

to be established between the SafetyNet team and the

etwork SPI

tion rstly describes the concept of the developed SPI.1) which serves as a guideline for the development ofc SPI applications. Such an application is subsequently

Sections 2.2.2 and 2.2.3 which present the translationoncept into a practical method.

twork SPI concept

network SPI is based on a quantitative method for the of network and design quality aspects of a safe roadre at the regional level developed in a Dutch study by03). The purpose of this study was to compare the exist-twork to the theoretically required roads, dened ash meet some minimum requirements with respect to. Recently, this method was further improved and used

the relationship between network design, route choiceDijkstra, 2011). The classication method of urban areas

study originated from the rather descriptive and qual-od of the German guidelines for road categories (FGSV,e guidelines dened a classication of roads and urbans in a qualitative way. In the Dutch study the methods been adjusted to a more quantitative method for the

of the infrastructure. network SPI aims to measure whether the right roadght location. It is dened as the percentage of appro-l road category length, per road category. This can be

cally expressed as:

n(LTiti

|ti Ti)

nLTiti

(1)

the theoretically required road category i:{six cat-m AAA to C}; ti is the actual road category i:{sixfrom AAA to C}; n is the number of road connections;e value of the safety performance indicator for road cat-iiis the actual road length which should theoretically be

Ti and actually is of category ti.

386 G. Yannis et al. / Accident Analysis and Prevention 60 (2013) 384 395

The basic idea behind the developed SPI is that the traf-c demand determines the type of road required. The SPI thenmeasures to what extent the actual roads in a network are appro-priate, given the theoretically required roads.

To estimgories betwthe trafc obtained. Th

Dab = f (X1a ,where variac demandnumber of centres indeculture), wterrain, etc.

The actuto dene thfrom the trwith differmight existthe respecttrafc assig

Howevewhether twonly one couated. The depends ontrafc voluthe selectedrequired ro

Ti = f (Qi) It should

that the thetrafc volumcan be deteservice, the

Once thnection arecategory of for the calcu

4. Road ne

In ordertical metholimitations parisons, a ndeveloped rof the SPI p

Assumptioence genera

Assumptiobetween th

Assumptio

Assumptionecessary:

Concernquantify ththe rest of t

Denition area where

tion the c

tion num.

ntifyand

nd

D(N1

houldinto n be

N2n:

c(N1

umpe ro

s. Hoe lo

luateove vlowin

12

ch QR C2. N

actuwingumpfc dry, tos a prhe ction

mand estimrenttc. (Mump

roadvoluperfoty. Then trd safned

tion nt des per

known that, for a given road type (T), the above denednt density (rT) is a function of trafc volume

(Q ) (7)ate the above SPI, the theoretically required road cate-een centres need to be dened. To achieve this, initially,demand between pairs of centres a and b should beis will be derived by a formula of the following form:

X2a , . . . , Xia, X

1b , X

2b , . . . X

ib, X

2ab, . . . X

jab

) (2)

bles Xa, Xb and Xab are the determinants of the traf- between centres a and b, while i and j indicate thevariables. Variables Xa and Xb concern each of the twopendently (i.e. population, business, industry, tourism,

hile variables Xab concern both centres (i.e. distance,).al trafc volume between centres (which is requirede theoretically required road category) will be derivedafc demand. Due to the fact that several connections,ent route costs (distance, time, comfort, safety, etc.)

between pairs of centres, trafc volumes resulting fromive demand need to be derived from an appropriatenment model (Shef, 1985).r, the purpose of the road network SPI is to determineo centres are connected by an appropriate road, thusnnection between each pair of centres should be eval-criterion to select the most appropriate connection

policy (i.e. fastest route vs. safest route). Therefore, theme resulting from the previous trafc assignment for

connection will be used to determine the theoreticallyad category for this connection:

(3)

be mentioned that although the above formula impliesoretically required road category is only a function ofe, in practice, the way this relationship is established

rmined by additional characteristics like the level of road accident risk, the generalized cost, etc.e theoretical road categories required for each con-

determined, and the data concerning the length andthe actual connections are available, Eq. (1) can be usedlation of the SPI.

twork SPI development

to translate the above theoretical concept into a prac-d, which can be applied to several countries (within comparable data), and to enable cross-country com-umber of assumptions (which will lie at the basis of theoad network SPI) should be made. In the formulationroposed here, these are the following:

n 1. Two centres that are in each others area of inu-te trafc to and from each other.

n 2. The size of centres determines the trafc demandose centres.

n 3. All trafc between two centres uses the same road.

n 4. The trafc demand determines the type of roadmore trafc requires a higher-level road.

ing the rst and second assumption, there is a need toe relation between centre size and trafc demand. Forhis paper, the following denitions are used:

1. A centre (C) is dened as a geographically localized people travel to and from.

Deniwhich

Denias the (Fig. 1)

Quatres C1means

D12 = It s

taken tion caN1 andfunctio

D12 = Ass

the samcentrewith thbe evacan prthe fol

QR = DIn whiC1 andnot thefree o

Assing traindustused astrate tapplicathe defor thethe curcosts, e

Asshighertrafc safety of safethe givfor roabe de

Deniaccidecrashe

It isaccide

rT = rTFig. 1. Trafc demand D12 between centres C1 and C2.

2. The size (N) of a centre is dened as the extent toentre generates trafc.

3. Trafc demand (D) between two centres is denedber of motor vehicles travelling between the centres

ing the relationship between the size N1 and N2 of cen- C2, and the trafc demand D12 between these centresing the following function D:

, N2) (4)

be noted that the direction of the trafc ow is notaccount. Without using measurements, a simple func-derived by assuming that D12 linearly depends on bothwith the same coefcient (c). This yields the following

+ N2) (5)tion 3 states that all trafc between two centres usesad. In reality, often multiple routes exist between twowever, given the fact that only one connection (the pathwest cost) between each origin-destination pair willd (assuming that all trafc is assigned to this path), italid enough to serve as a starting point. This results ing function:

(6)

denotes the trafc volume on road R between centresote that Eqs. (4)(6) are meant to model trafc demand,al trafc expected. Whether or not the actual trafc is

is irrelevant here.tion 2 states that although there are many factors affect-emand between two centres (i.e. population, business,urism, culture) only the size of centres (population) isoxy in this study, to simplify the process and to demon-alculation of the SPI. More factors can be used in furthers of the method to provide more robust estimations of

between centres. In that case a more detailed methodation of the demand should be adopted, using not only

trafc ows but also the employment, trip rates, travelcNally, 2000).

tion 4 states that a larger trafc demand requires a type. Higher road types generally allow for higher

mes, thus this is mostly benecial for mobility. A roadrmance indicator, though, should determine the levelerefore, Assumption 4 is valid if a higher road type forafc demand (and resulting volume) is more benecialety. In this case, the level of unsafety of a road type canas follows:

4. The level of unsafety of a road type is dened by itsnsity (r), dened as the average yearly number of injury

km.

G. Yannis et al. / Accident Analysis and Prevention 60 (2013) 384 395 387

Q

rmax

r

T1

T2

Fig. 2. The redifferent road

Further otions could

Assumptiovolume.

Assumptiounsafety is

Assumptioc ow impincreases: r

Howevefor accidentThe functiomonotonic.

A maximbecause ottype lower dened as t

rmax denotM denote tTi denote t

Then, if rthere are trFig. 2 show

The funhowever, fomultiplicatirT(Q) = a(T)

Moreovehowever, thpurpose canecessary. Abe used forof differenteach countrson using thextent to w

Given a trafc volurequired roume range

Table 1Trafc volume ranges and associated minimally required road types.

volume range Minimally required road type

ethod

prevtionteps

ide wrmin

ide wrminrmin

typ

ally, ensitum wev

ious fnowes, rolt, trah onore, she prin prhe fo

ide wrmin

trafrminrminmax

give

obtarisoned. Ttegonto sictedQ1 Q2 Q3 Q4

T3

T4

lation between road accident density, r, and trafc volume, Q, fortypes, T.

n, to interpret Assumption 4, some underlying assump- be made:

n 4.1. A higher order road is safer at each given trafc

n 4.2. A predened maximum level of acceptablerequired.

n 4.3. For a given road type T, having a higher traf-lies decreases the level of safety (i.e. accident density

T(Q) is a positive-monotonic function).

r, it should be noted that although Assumption 4.3 holds density, it does not necessarily apply for accident risk.n of accident risk with trafc volume may not be positive

um level of acceptable unsafety should be predened,herwise there is no criterion to allow for any roadthan the highest-order road. The level of unsafety washe road type accident density, rT, so let:

e the maximum level of unsafety that is acceptable.he number of possible road types.he ith road type, where i = 1, . . ., M.

max is chosen high enough, Assumption 4.3 implies thatafc ows Qi (i = 1, . . ., M) such that rT(Qi) equals rmax.s this in a graphical form.ction rT(Q) can have different forms per road type,r the scope of this study it can be supposed that a

Trafc

0Q1Q1Q2Q2Q3Q3Q4

4.1. M

Theapplicasome s

1. Dec2. Dete3. Dec4. Dete5. Dete

road

Idedent dmaximers. Hoon varover, kvoluma resuresearcThereffrom t

So, were t

1. Dec2. Dete

ing 3. Dete4. Dete

the with

To compaproposroad casied iis restrve factor exists and a common risk function such as:r(Q).r, the general form of these risk functions is provided,e specic form in each country can be different. On thatlibration of the functional forms with national data isdditionally, different maximum levels of unsafety can

different networks taking into account the existence standards due to different landscape, policies, etc. iny. Nevertheless, none of the above affects the compari-e SPI, because it is dimensionless and reects only thehich each countrys national standards are achieved.maximum level of acceptable unsafety, rmax, ranges ofmes (in red cf. Fig. 2) are associated with minimallyad types. The following table shows which trafc vol-dictates which minimal road type (Table 1).

These rocombinatioare assumecentre sizeson empiricclasses in Tas an indica

The clas(2003). He dof inhabitanassumed toa centre. Asindustry, aithat is geneshould be gT1T2T3T4

for the determination of the road network SPI

ious section discussed the rationale behind the practical of the road network SPI. To be able to apply this theory,

have to be taken:

hen centres are in each others sphere of inuence.ation of the trafc volumes QR.hat is the maximum level of unsafety rmax.ation of characteristics of road types T1 to T4.ation of the accident density functions for the different

es and the corresponding trafc volumes.

the areas of inuence, the trafc volumes and the acci-y functions are known from empirical studies and thelevel of acceptable unsafety is dened by policy mak-er, in reality, trafc volumes between centres dependactors and are difcult to obtain in a simple way. More-ledge about exact quantitative relations between trafcad characteristics and accident density is limited. Asfc volumes and road types are unknown. Empirical

these functions was beyond the scope of this research.ome choices were made to be able to apply the theoryevious section.actice, the steps taken for the estimation of the indicatorllowing:

hen centres are in each others sphere of inuence.ation of the populations Pi, generating the correspond-

c volumes Qi.ation of the characteristics of road types T1 to T4.ation of theoretical road categories Ti corresponding to

imum level of unsafety rmax, for pairs of urban centresn population size.

in a road network SPI that allows for internationals, an internationally harmonized road categorization isable 2 shows the minimal requirements for differentries that are proposed within SafetyNet. Roads are clas-ix categories ranging from AAA to C. This classication

to rural roads and motorways.ad types are assigned to connections between differentns of centre sizes. Table 3 shows which road categoriesd to be necessary between different combinations of. It is herewith stressed again that this table is not basedal evidence, but on expert judgement. Therefore, theable 3 should not be considered as strict classes buttion.sication of the types of centres is adopted from Dijkstraistinguishes ve types of centres, based on the numberts. This implies that only the number of inhabitants is be relevant for the amount of trafc that is generated by mentioned before, in reality, also other factors, such asrports and tourist attractions, affect the amount of trafcrated. When applying the method, special considerationiven to these situations.

388 G. Yannis et al. / Accident Analysis and Prevention 60 (2013) 384 395

Table 2Minimal requirements for different road categories.

SafetyNet road categories Ti Rural areas (outside built-up areas)

AAA: AA: A: BB: B: C:Rural distributorroad 1

Rural distributorroad 2

Rural access road

Functional r Distributor road Access roadSeparation o

directions,e

Dual carriageway Single carriageway,preferable with laneseparation

Single carriageway

Lane congu 2 1, 2 2 1 2, 1 3, (1 4) 1 2, 1 1Obstacle-fre rier Medium Medium Small

Intersection Preferableroundabout

Preferableroundabout

Table 3Minimum requ

Urban centre Type 3 Type 4 Type 5

Type 1 (P1 > AA Indirectly IndirectlyType 2 (P2: 1 AA BB IndirectlyType 3 (P3: 3 BB BB BType 4 (P4: 1 B BType 5 (P5 < C

Table 4Overview of th

Start centre tres in

1 d 3 2 d 4 3 3 4

To deterin a given rareas, in lin1961; Grossurban or rurcircular seacentre of thlocation of tdescribed btype. The arof that speother centrcentres are centre type2 and 5, etcis that it is search areaadjusted (frprevents tahave a relat

As an exfrom the Poareas withinto the criterclassied asfor connect

The nex(existing) rthis step, thtions are cothe actual cMotorway A-level road 1 A-level road 2

oad category Through-road (road with a ow function)f opposing Dual carriageway Dual carriageway Single carriageway

preferable with lanseparation

ration 2 2 or more 2 1, 2 2 1 2, 1 3, (1 4)e zone Very wide or

safety barrierWide or safety barrier Wide or safety bar

s Grade-separated Preferablegrade-separated

Preferablegrade-separated

ired road types Ti connecting two centres of given size.

type (# inhabitants) Type 1 Type 2

200,000) AAA AAA 00,000200,000) AA 0,000100,000) 0,00030,000) 10,000)

e search for centres.

type Search for the centre of the same type (radius) Cen

The nearest 1 2 anThe nearest 2 3 anThe nearest 3 4 The nearest 4 5 The nearest 4 5 mine which centres should be connected to each otheregion or country, we used the so-called circular searche with hierarchical graph theory (Nystuen and Dacey,

and Yellen, 1998). According to this method, for eachal centre examined, a circular search area is drawn. Thisrch area is determined by the distance to the closeste same type: the centre of the circular search area is thehe urban centre assessed and the radius of the circle isy the shortest distance to the closest centre of the sameea within each circle can be seen as the area of inuencecic town or village. Within this area, connections toe types are considered. Table 4 shows which types ofsearched for. It is suggested that connections betweens of very different size, for instance 1 and 4, 1 and 5,., are not meaningful to examine. The reason for thishighly unlikely that such connections exist at all. The

for connections between type 3 and type 5 centres isom the centre type 3 to the nearest centre type 4). Thisking into account centre types 5 which probably do notion with the centre type 3.ample, Fig. 3 shows the application of the methodologyrtuguese pilot project. More specically, the circular

which connections are sought are presented accordingia of Table 4. The urban centre Albufeira, for example, is

type 3, having thus two different circular search areasions with type 4 and type 5 urban centres.t step is the identication and matching of the actuaload connections, in relation to the theoretical ones. Ine road categories of the theoretically required connec-mpared to the actual ones. In order to perform this step,onnections have to be identied and the roads that are

part of the the SafetyN

The idecriteria choon the Susprocedure iare lteredthe connectbetween tw

F search area Assessment of the connections between

1 and 1, 1 and 2, 1 and 32 and 2, 2 and 3, 2 and 43 and 3, 3 and 43 and 54 and 4, 4 and 5actual connection need to be categorized according toet road categories presented in Table 2.ntication process depends mainly on the specicsen to represent the route choice by drivers. Basedtainable Safety approach, the SafetyNet identications carried out in two steps. First, all existing connections

using a detour factor that limits the total length ofion. Initially, it was decided that the actual connectiono cities should be less than 1.6 times the Euclidean

ig. 3. Circular search areas in the Portuguese study area.

G. Yannis et al. / Accident Analysis and Prevention 60 (2013) 384 395 389

distance between the cities. This detour factor is similar to thedetour factor applied by Dijkstra (2003) and is evaluated in thepilot studies presented in the next sections. Secondly, it is alsopossible that more than one connections between two centres areindentied tions needsThis can beGIS applica

For eachcategory shegorizationbetween twferent roadpart inside context of tthe connectcategories ttheir length

The naparing the taggregatingtheoreticallthat meets the SPI for Eq. (1).

5. Road ne

The devof pilot stuPortugal. Thpresented i

5.1. Pilot st

For the sen as the pprovincial cties in the pmade up of100 and 12with speed2025 km ofits of 60 kmrural distrib30 km/h lim

The regiofor the calcinhabitantsarea and itpopulationsclosed intwhich wereis interestin

For Israeroad netwowork. The Central, Hamore denseand is scadistrict is cof the couwork is 17of urban arunits (mun

which is maintained by the Central Bureau of Statistics. Urban cen-tres for the pilot were sampled based on this list.

For the Portuguese pilot study, the national continental terri-tory bellow Tagus River was chosen as a study area. This area is

34,tion)s roan, anudy.ntry

occf the

e indry ocllectistrate Potaine

ata a

eneracedaisedle toilt uampls insl villaal ano eaco ovndedthermeasuentr

takensidcilit

n a fes werc thenetionafacendicaot core pat wed bnal ie lima (20ross rk.

eterm

he pinat

areainedme ainlyth lo). In from the previous step. In that case, one of the connec- to be selected as the actual connection to be evaluated.

done in several ways, e.g. using a route planner or ation, depending on the instruments and data available.

connection that is nally selected, the SafetyNet roadould be determined by translating the national road cat-

into the SafetyNet categorization. An actual connectiono centres can consist of several road sections with dif-

categories. Moreover, a connection may consist of athe built-up area and an interurban/rural part. In thehe present research, only the interurban/rural part ofion needs to be assessed. For each connection, the roadhat make up the rural part of the connection, as well ass, should be determined.l step is the calculation of the SPI. This is made by com-heoretically desired and the actual road category and by

the scores for each road category. In this way, for eachy required road category, the percentage of actual roadsthe requirements can be calculated. The calculation ofa given road category was provided in Section 2.2.1 by

twork SPI application in pilot studies

eloped methodology was evaluated through a numberdies carried out in the Netherlands, Greece, Israel andeir characteristics are summarized in Table 5 and are

n detail below.

udies characteristics

Netherlands, The Province of South Holland was cho-ilot area. The province has 3,455,097 inhabitants. Theapital is The Hague which is one of the 82 local authori-rovince. The province has a road network of 15,884 km

740 km of State roads (freeways with speed limits of0 km/h); 694 km of provincial roads (distributor roads

limits of typically 80 km/h and sometimes 100 km/h); Water Board roads (rural access roads with speed lim-/h) and 12,425 km of local authority roads (80 km/hutors, 50 km/h urban distributors and access roads withits).n of Peloponnese, situated in south Greece was selectedulation of the SPI in Greece. The region has 1,045,000

and an area of 21,493 km2. It covers a large geographical includes numerous cities/towns of various sizes and. Finally, it includes all types of roads in a relativelyerurban road network (7730 km in total, 6979 km of

examined) and it has a mountainous mainland, whichg to study.l, the whole country served as a pilot area and therk SPI was calculated for a sample of the road net-majority of population is concentrated in Tel-Aviv,ifa and Jerusalem districts. The Northern district isly populated, while the Southern district is the largestrcely populated. A signicant part of the Southernovered by deserts. In 2006, the average populationntry was 7,054,000 inhabitants. The total road net-,686 km, of which 7854 km are rural roads. The listeas considered was the whole list of administrativeicipalities/towns/villages/communities) of the country,

aroundestimavarioucatiopilot stthe couhumancoast ointensterritowas coadminand thwas ob

5.2. D

In gwere fwere rpossibthe buFor expalitieseveraPortugclose torder tsurrou

Furonly msome cnot bewas cotheir faover, icentreof trafwere drecreathe surthese ithe pilin a motres thupgrad

A that thDijkstrtres acnetwo

5.3. D

In tdetermsearchdetermcase sowas mbut witype 5000 km2 and has a total of 1,700,000 habitants (2006. This area has several types and sizes of urban areas,d types and also good conditions for urban areas identi-d thus satises the criteria for the implementation of a

However it is not representative of all urban centres at level. The pilot area is characterized by a very disperseupation except in the Setbal Peninsula and the South

Algarve. Both these regions are also characterized byustrial, commercial or touristic activities. Data related tocupation (geometry, population and number of houses)ed for a total of 438 Freguesias, the Portuguese smallestive unit level, from the National Institute of Statisticsrtuguese Geographic Institute. The road network datad from the InfoPortugal S.A. geo-referenced data base.

vailability

al no signicant obstacles concerning data availability within the pilot studies. However, a number of issues

and should be highlighted. Firstly, it was not always dene urban and rural centres by the boundaries ofp area due to lack of data availability on this level.e, in the Netherlands, data was available for munici-tead of villages/cities (one municipality may consist ofges that are separated by rural area). Moreover, in Israel,d the Netherlands, a number of urban centres are veryh other such that no rural area exists between them. Inercome this obstacle, each centre (village, city) that is

by rural area should be dened as one centre.ore, despite the fact that the size of the centre is the

re which denes the centre type, it was revealed thates with facilities of special interest (like ports) wouldn into account in case only the number of inhabitantsered. These places generate and attract trafc due toies and therefore need to be included as centres. More-w cases, using the population as a proxy for demand,e classied to a lower category considering the amountat was generated or attracted by them. In case centresd on the basis of other characteristics, like industry andl areas, indicators would be needed for these factors, e.g.

of industrial area or the number of employees. Data ontors were not available (or at least not easy to obtain) inuntries. In Greece and Portugal, the problem was solvedragmatic way; the list of centres was evaluated and cen-ere considered missing or misclassied were added ory experts with local knowledge.ssue with regard to the classication of the centres isits of the classes were chosen from the Dutch study of03). By modifying the limits, the distribution of cen-

types will change along with the theoretically required

ination of theoretical connections

ilot studies of Israel and Portugal the process for theion of the theoretical connections through the circulars was automated, whereas Greece and The Netherlands

the theoretically required connections manually. Intheoretical connections were found to be missing, this

due to missing centres (i.e. points of special interestw residing population, being erroneously classied asPortugal, not all relevant theoretical connections were

390 G. Yannis et al. / Accident Analysis and Prevention 60 (2013) 384 395

Table 5Characteristics of the study areas.

Netherlands Greece Israel Portugal

Pilot study region South Holland Peloponnese Whole country Territory bellow TagusArea (km2) 22,073 34,000Population 7,054,000 1,700,000Interurban r 17,686 28,000

identied b(caused by Moreover, connected tarea. Such added man

On the ctions were mainly obseThe search determine ttains, riverslack a direcin the Greeksidered witactual conn

Finally, stances, resbe assessedtres is very will result ipossible walar area andcloser type in areas witbe quite un

The theomethodolog

5.4. Identi

Pilot counections beplanner to sroute. Portution of the a GIS tool ahighest roa

It is impgories) is sesatised thwhile the bever, in somsafest. Therexpert withsomewhat the fastest r

To specidetour factoactual distaselected fordetour factobe assessedappropriatewhile the din Greece asea) often c

he Euclidean distance, given that the respective connectionsst. On that purpose, the detour factor should be used in aatic way. In principle, the detour factor should be 1.6 andtions exceeding this detour factor should be further ana-In case there is a (natural) barrier causing a detour factorigher, a detour factor of 2 can be accepted. Another possi-n dealing with natural barriers is to remove a theoreticallyd connection. This can be done in case the barrier makes

between these cities unlikely. Finally, in case the detour fac-oo high and the conclusion is that an actual connection is

missing, this should be noted as a aw in the road network.ase a route is selected as an actual connection, the roads thatute the actual connection should be classied on the basisSafetyNet road categorization (see Table 2). This appearedossible in all pilot countries, although the data sources andation used differed per country. For some pilots, knowledgelocal situation was necessary in order to determine the roadries.

actual connections in each pilot country are shown in Fig. 5,he actual road length per road category for these connectionsn in Table 6.

lculation of the SPI

nais str. Fotionined

oad ne

ry

rlands

al2818 21,493 3,455,097 1,045,000

oad network length (km) 15,884 7730

y the search areas, due to relatively small search areasshort distances between two centres of the same type).some small urban centres (type 5 centres) were noto any higher class centre since it was not in any searchconnections, which were considered missing, wereually in the pilot studies.ontrary, in some cases, theoretically required connec-determined that could not be found in reality. This wasrved when a natural barrier existed between centres.circles do not take natural barriers into account andhe distance on a straight line. In case of a barrier (moun-, lakes, etc.), cities that are geographically close mayt connection. This was the case in several connections

pilot. As discussed below, these connections were con-h special attention when identifying and matching theections.the use of circular search areas may, in rare circum-ult in excessive connections (that normally should not). For example, if the distance between two type 4 cen-long, the radius of the search area will be great and thusn meaningless type 4type 5 theoretical connections. Ay to overcome such a problem is to discard this circu-

use a new radius instead, dened by the distance to a5 centre. Such needs for exceptions could occur easierh mountainous layout as the distribution of centres caneven.retical connections resulting from the application of they in each pilot country are shown in Fig. 4.

cation and matching of actual connections

ntries used different methods to dene the actual con-tween centres. Greece and the Netherlands used a routeelect an actual connection and searched for the fastestgal built a routing model in a GIS tool for the identica-fastest route without hierarchy preference. Israel usednd searched for the shortest route, going through thed classes.ortant that a safe route (using appropriate road cate-lected for each connection. In general, the fastest routeis criterion in the Netherlands, Greece and Portugal,asic connection was found appropriate for Israel. How-e cases in the Netherlands, the fastest route was not theefore, the actual connections need to be evaluated by an

local knowledge and in case a safer route exists that islonger than the fastest, it should be selected instead ofoute.fy whether an actual connection should be assigned, ar was applied. This factor is dened as the rate of thence to the Euclidean one. The initial maximum value

the detour factor was 1.6. If for a specic connection ther had a value greater than 1.6 the connection would not

times tdo exipragmconneclyzed. to be hbility irequiretravel tor is tindeed

In cconstitof the to be pinformof the catego

Thewhile tis show

5.5. Ca

Thevalue, appliedconnecdeterm

Table 6Actual r

Count

Nethe

Greece

Israel

Portug. However, it was observed that this initial value was not for all connections in the pilot areas. More specically,etour factor was found appropriate for the Netherlands,nd Portugal large natural barriers (like mountains andause distances between centres to be higher than 1.6l step, which entails the actual calculation of the SPIaightforward if the preceding steps of the method arer each connection, the length (expressed in km) of the

that meets the requirement and the length that was not by using Eq. (1). By summing up the results for each

twork in the four pilot countries by road category.

Connectiontype

Length (km)

Avg Max Min St. dev.

AAA 14.8 52.9 1.1 11.7AA 8.8 20.6 1.5 4.4A 6.7 12.1 3.1 3.9BB 5.4 18.0 1.2 3.5B 5.6 24.7 0.6 4.1C 4.3 8.2 2.0 2.3

AAA 28.6 80.8 0.9 26.8AA 16.4 16.5 16.2 0.2A 21 66.4 0.4 19BB 0 0 0 0B 12.7 116.5 0.03 18.2C 1.5 16.4 0.01 3.33

AAA 10.6 34.7 1.4 6.8AA 6.5 33.9 0.7 5.5A 5.1 30.9 0.5 6.5BB 5.2 12.3 0.8 3.1B 5.7 19.0 1.1 3.7C 2.5 8.5 0.3 1.9

AAA 4.1 240.6 3.0 19.6

AA 0.8 25.4 0.1 2.3A 0.1 5.6 0.06 0.5BB 0.4 5.7 0.06 0.9B 10.6 61.3 0.03 12.5C 1.1 28.6 0.02 3.7

G. Yannis et al. / Accident Analysis and Prevention 60 (2013) 384 395 391

(theoreticalate road cabe noted thbetween tytres were oclass C roadand Greek pndings froAdditional iing each pil

The theonicantly bof Greece, one) while Netherlandulated. Thisand makes Fig. 4. Theoretically required connections in the fou

ly required) road category, the proportion of appropri-tegory was calculated for each road category. It mustat, according to Table 3, class C roads are required onlype 5 centres. However, connections between type 5 cen-nly assessed in the Portuguese pilot study therefore nos are stated as theoretically required in the Dutch, Israeliilot studies. This section presents the most importantm the calculation of the SPI for the four pilot countries.nformation as well as more analytical results concern-ot study can be found in Weijermars (2008).retically required road network was found to differ sig-etween the case study areas. In the case study areathe largest centre was of type 2 (and there was onlyseveral type 5 centres existed. In the pilot area in thes, two type 1 centres existed and the area is densely pop-

imposes different requirements on the road networkit difcult to compare the scores for different countries.

In general, scores of Gr

In the Neach road cbe observeply quite wthe requireneed to be B roads areroad are lescategory. Omeets or ex

To an exorder roadsas lower orof the roadcaused by r pilot countries.

the resulting scores seem quite realistic, although theeece seem low compared to the other pilot areas.etherlands, the road network SPI was calculated forategory and the results are presented in Fig. 6. It cand that for most road categories, the actual roads com-ell with the theoretically desired road category: 87% ofd AAA roads are in fact AAA roads, 87% of the roads thatAA are in reality AA or higher, and 91% of the required

in fact B or higher. However, the requirements for BBs often met, with about 22% of the roads being of a lowerverall, about 82% of the road network in South Hollandceeds the SafetyNet classication requirements.tent this score is inuenced by the presence of higher

fullling dual roles, as high order connection but alsoder connection. If one looks at individual classes, 22%s that should be BB are of a lower category. This isthe extensive use of single carriageway rural roads in

392 G. Yannis et al. / Accident Analysis and Prevention 60 (2013) 384 395

the Netherlcarriagewayis hardly likto comply roads have ures which and which aBased on thshould be gvarious roadifferent ty

In the Gout of whicategory thto 66.5% inoverall; howFig. 5. Actual connections in the four pilot c

ands which, according to SafetyNet, should be double. Considering the area already occupied by roads, itely that the majority of these roads could be rebuilt

with these requirements. Besides that, many of theseovertaking bans and other trafc management meas-conform to many of the Dutch regulation requirementsre currently not considered by the SafetyNet approach.e results of the pilot in South Holland considerationiven to reviewing the functional requirements for thed types and/or the road categories that are desired forpes of connections.reek pilot study 6.979 km of road were examined,

ch 4.344 were of appropriate or higher actual roadan the theoretical one, resulting in a total road SPI equal

this study area. This nding is not very satisfactoryever a more detailed consideration of the SPI reveals

an interestiof type AA connectionmeets BB though, thacarriagewaBB and B rotype wouldlevel conneIt is also inthese connea limited nu

Overall, the road canumber of satisfactoryountries.

ng picture. As shown in Fig. 7, theoretical connectionsare met only by 10% of the total length of the actuals. Respectively, around 52% of the total road lengthor higher standards. One should take into account,t no actual BB connections exist in the study area (dualy is very rare for lower level connections in Greece). Ifad categories were merged, the SPI for this connection

be extremely high in the study area. As regards lowerctions, the SPI is equal to 93% for type B connections.teresting to note that about 10% of the total length ofctions corresponds to AAA and A roads, indicating thatmber of roads is used for many connections.it is indicated that the total SPI score (aggregated overtegories) is the result of putting together an increasedlower level theoretical connections presenting a very

SPI, with a small number of higher level theoretical

G. Yannis et al. / Accident Analysis and Prevention 60 (2013) 384 395 393

Fig. 6. Percen

Fig. 7. Percen

Fig. 8. Percen

connectionnetwork inexplained bpopulation

The resuSPI is prese

It can becategories ccategory: 8of the requi81% of the r

The requof the road

ercentage of roads meeting the requirements in the Dutch pilot study. Fig. 9. Ptage of roads meeting the requirements in the Greek pilot study.

tage of roads meeting the requirements in the Israeli pilot study.

s presenting a poor SPI. A great unbalance of the road the Greek study area is thereby revealed possiblyy the nodes denition taking into account only theand not other criteria (e.g. touristic attractions).lts from the Israeli pilot study and the calculation of thented in Fig. 8.

observed that for most road categories, the actual roadomply fairly well with the theoretically desired road

5% of the required AAA roads are in fact AAA roads, 89%red BB roads are in fact BB or higher road categories, andequired B category roads are B or higher level roads.irements for AA roads are less often met: only 74.5%s that need to be AA are in reality AA or AAA roads,

implying tcategories (categories aBB-type cosimilar to th

Overall, road netwoments.

In Portutheoreticallis providedin the pilot possibility o

It can bereality AA oity BB or higB or higherpriate or hia total roadfrequent (3high SPI scoical connecclasses of rindicated bare not condoes not chto the needrelatively s

5.6. Additio

The pilother improMore speciularities ofimplement

5.7. Theore

As far aoretical conthe implemconnectionbetween twof the resping in conneither a smtage of roads meeting the requirements in the Portuguese pilot study.

hat 25.5% of the connections belong to lower roadhowever the differences between the AA and BB roadre not strict in Israeli conditions, meaning that actual

nnections sometimes have road design characteristicse AA-type roads).some 81% of the connections considered for the Israelirk meet or exceed the SafetyNet classication require-

gal, the comparison of the actual road network with they required and the calculation of the road network SPI

in Fig. 9. It should be pointed out that no urban centresarea were classied as type 1, eliminating therefore thef any AAA theoretical road connection.

observed that 75.5% of all roads that should be AA are inr higher, 86.5% of all roads that should be BB are in real-her and 95.8% of all roads that should be B are in reality

. In total, 14,975 km had an actual road category appro-gher than the theoretical category needed, resulting in

network SPI of 93.7%. Type B connections are the most81 out of 693 connections). These connections have are, justifying the high total network SPI. Type C theoret-tions are always associated with appropriate or higheroads, as this category is the minimum road standardy the SafetyNet classication. When Type C connectionssidered for the calculation, the total road network SPIange much (92.9%), suggesting that their contributioned network is small (total Type C connection length ismall).

nal observations

t studies revealed a number of issues that could fur-ve the methodology and its efciency on future pilots.

cally, according to the real conditions and the partic-

each study area, the methodology can be adapted bying small modications.

tical connections

s the circular areas for the determination of the the-nections are concerned, under certain circumstancesentation of the methodology could result in theoreticals that normally should not be assessed. If the distanceo centres of the same type is very large, the radius

ective circular search area would be very long result-ections that normally do not have sense. In that case,aller circular area could be assigned based on a different

394 G. Yannis et al. / Accident Analysis and Prevention 60 (2013) 384 395

criterion or the resulting theoretical connections should be logicallyassessed before any connection is assigned.

A detour factor of 1.6 was concluded not to be suitable for theGreek and Portugese pilot area. Detour factors ranging from 1.8 to3 were testin the Portutions a detosensible. Tha detour facshould be dnection is thhas an acce

5.8. Calcula

From thoverall scorpreferably boverall scorof B roads thview. Anothtype (type 1as was donethe interpreexact probl

When thsome road and BB roadprocedure dpoorer scorscores for B

With regcases relatisections conarises whetrequiremenroute to bethe primary

6. Discussi

The piloreveal somsection discto improve

Firstly, tattention. Itrural area athe data avaquite arbitralso other faof trafc thatrafc may the number

Secondlysome limitalar search aTherefore, wsituation junections. Acentres coumodel.

Thirdly, two centreselected as in a pragma

When assigning a route as the actual connection, it is importantthat this route is as safe as possible (using high road categories).Hakkert and Gitelman (2007) suggest both the fastest and safestroute (the route via the highest road categories) to be determined.

fest ias leicateconcries. antl

edgehat rth, n thradeom tty of ad s

esireres. Iuld b

resures old b

greamene sco

ervedry inal, duwhic

reovecombt men rT(d chad.

clus

s paprbanual rad Pht roixin

or coch o2000

roadmbe

tempdologclude re

wheting ry chs, the

necenedp thethermn thed. The detour factor of 1.8 seemed to be more realisticguse pilot, while in the Greek pilot, for some connec-ur factor of 2 (and in some cases even 3) appeared to beerefore, it can be concluded that each connection withtor higher than 1.6 should be analyzed separately and itecided on the basis of local knowledge whether a con-eoretically needed and whether the actual connection

ptable detour factor.

tion of the SPI

e observation of the results it can be claimed that ane estimated from the combination of all results shoulde avoided as it could provide misleading results. The

e was highest for Portugal, due to a very high percentageat have a high score. One overall score may give a biaseder possibility is to analyze the score for each connection

centretype 1 centre; type1 centretype2 centre, etc.) in the Israeli pilot. This complicates the calculation andtation of the results, but provides more insight into theems.e results are analyzed in more detail, there appear to becategories that have lower scores. In general, AA roadss are applied to a lesser extent, whereas the SafetyNetoes require this category to be examined, resulting ines for AA roads in Greece, Israel and Portugal and poorB roads in Greece and The Netherlands.ard to the scores of the individual connections, in somevely short sections of the connections (e.g. the roadnecting freeways to urban areas) failed. The questionher these sections should comply with the SafetyNetts or if one should allow a certain percentage of the

of a lower category and only base the assessment on part of the connection.

on

t studies that were discussed in the previous sectione limitations of the proposed road network SPI. Thisusses these limitations and makes some suggestionsthe method.he denition and classication of centres needs some

seems logical to dene a centre that is surrounded bys one centre. However, this is not always possible givenilable. Moreover, the limits of the classes were chosen

ary, there is no theoretical base for these limits. Besides,ctors than the number of inhabitants affect the amountt is generated by a centre. Production and attraction ofprovide more appropriate classication variables than

of inhabitants., also the determination of theoretical connections hastions. The connections that were found using the circu-reas do not always correspond to the actual situation.e advise to let an expert with knowledge of the local

dge and if necessary adjust the list of necessary con-lso, travel demand between different combinations ofld also be determined using a (macroscopic) trafc

in many cases, various routes were possible betweens but according to the methodology, one had to bethe actual connection. This step should be dealt withtic way, using the available tools.

The sathis wcomplerally categosignicknowlsomew

Fouroad. Ibe upglows frcapacieach rocally dSPI scoand co

Thethe scoIt shoudiffersrequirepare thbe obscategoPortugscore, gories.

Mosome currenfunctiofoundeon a ro

7. Con

Thi(interuindividRAP Rothe rigview. MindicatResearOppe,

Thein a nurst atmethowe conover, thGreececalculaarbitracentrethat isWe dedevelo

Furbetweenstead of the fastest route should be selected in casess than 5% slower than the fastest route. This slightlys the method, while from the pilot projects it was gen-luded that the fastest route also uses the higher roadTherefore, it is not expected that this criterion wouldy improve the results. We advise to ask someone with

of the local situation if there is a safer route that is justlonger and if so the safest route should be selected.in some cases, multiple connections use the sameis case the theoretically desired road category couldd when the estimated daily trafc volume (that fol-he combination of connections) exceeds the maximuma road category. On that purpose, the connections usingection should be determined. Upgrading the theoreti-d categories of certain roads will have an impact on then the pilot studies, this factor was not taken into accounte considered in a further application of the method.lting scores seem quite realistic in general, althoughf Greece seem low compared to the other pilot areas.e noted, that the theoretically desired road networktly between the case study areas. This puts differentts on the road network and makes it difcult to com-res for different countries. From the results it can also

that it may not be wise to combine the results per roadto one overall score. The highest score was recorded fore to a high weight of lower level roads in the total SPIh have a better score comparing to higher road cate-

r, the theoretically desired road categories betweeninations of centres may be somewhat strict in thethod. The determination of a typical accident densityQ) per road type, could allow for a more theoreticallyoice for the necessary road type given the trafc demand

ions

er presents two SPIs to assess the level of safety of the) road network. The road design SPI assesses whetheroads are designed in a safe way and is based on the Euro-rotection Score. The road network SPI assesses whetherad is located at the right place from a functional point ofg the road design with the road network performanceuld potentially allow for a more complete assessment.n this subject is currently under way (Wegman and).

network SPI is discussed in more detail and is appliedr of pilot countries. The SPI that is proposed is just at to determine the safety level of the road network. They could certainly be further improved. From the pilotse that it is possible to calculate the SPI scores. More-sulting scores seem quite realistic, with the exception ofre the SPI was relatively low. However, the method forthe SPI has some limitations. Most importantly, someoices had to be made concerning the classication of

amount of trafc between centres and the type of roadssary given the amount of trafc between two centres.

some directions for further research in order to further SPI.ore, we recommend to investigate the relation

e SPI scores and trafc safety (for example expressed in

G. Yannis et al. / Accident Analysis and Prevention 60 (2013) 384 395 395

the number of fatalities per km travelled) in order to obtain moreinsight into the validity of the road network SPI and the conse-quences of (changes in) SPI scores. Of course, a country could twell with the categorisation and still have a high risk because ofother structural factors such as unsafe driving behaviour. There-fore, the selection of the areas for such a study would have to becareful so that they have similar structural characteristics.

References

Assum, T., Srensen, M., 2010. Safety Performance Indicator for alcohol in roadaccidentsInternational comparison, validity and data quality. Accident Analy-sis and Prevention 42 (2), 595603.

CEC, 2003. European road safety action programme: Halving the number of roadaccident victims in the European Union by 2010: A shared responsibility.Commission of the European Communities (CEC), Communication from theCommission Com(2003) 311 nal.

CROW, 1997. Handboek Categorisering wegen op duurzaam veilige basis. Deel I(Voorlopige) Functionele en operationele eisen. Publicatie 116. C.R.O.W, Ede.(inDutch).

De Leur, P., Sayed, T., 2002. Development of a Road Safety Risk Index. TransportationResearch Record 1784, 3342.

Dijkstra, A., 2003. Kwaliteitsaspecten van duurzaam-veilige weginfrastructuur(Quality aspects of a sustainably safe road infrastructure). R-2003-10. Sticht-ing Wetenschappelijk Onderzoek Verkeersveiligheid SWOV, Leidschendam (inDutch).

Dijkstra, A., 2011. En route to safer roads: how road structure and road classicationcan affect road safety. PhD Thesis, University of Twente, Netherlands.

Duivenvoorden, K., Weijermars, W., Louwerse, R., Papadimitriou, E., Yannis, G., 2007.Detailed manual for SPIs on roads. In: Hakkert, A.S and V. Gitelman (Eds.), 2007.Road Safety Performance Indicators: Manual. Deliverable D3.8 of the EU FP6project SafetyNet.

European Transport Safety Council (ETSC), 2001. Transport safety performance indi-cators. Brussels.

FGSV, 1988. Richtlinien fr die Anlage von Straen RAS. Teil: Leitfaden fr diefunktionelle Gliederung des Straennetzes RAS-N. Forschungsgesellschaft frStraen- und Verkehrswesen FGSV, Kln (in German).

Gitelman, V., Doveh, E., Hakkert, S., 2010. Designing a composite indicator for roadsafety. Safety Science. Volume 48. Issue 9, 12121224.

Gross, J.L., Yellen, J., 1998. Graph Theory and Its Applications. CRC Press.Hakkert, A.S., Gitelman, V. and Vis, M.A. (Eds.), 2007. Road Safety Performance Indi-

cators: Theory. Deliverable D3.6 of the EU FP6 project SafetyNet.Hakkert, A.S and V. Gitelman (Eds.), 2007. Road Safety Performance Indicators: Man-

ual. Deliverable D3.8 of the EU FP6 project SafetyNet.Hermans, E., Van den Bossche, F., Wets, G., 2008. Combining road safety infor-

mation in a performance index. Accident Analysis and Prevention 40 (4),13371344.

International Road Assessment Programme (iRAP), 2009. Star Rating Roads forSafety: The iRAP Methodology. iRAP504.04. Site: www.irap.org

Koornstra, M.J., Mathijssen, M.P.M., Mulder, J.A.G., Roszbach, R. Amd Wegman,F.C.M., 1992. Towards a sustainable safe trafc system. Nationale verkeersvei-ligheidsverkenning voor de jaren 1990/2010. SWOV, Leidschendam (In Dutch).

Lynam, D, Sutch, T., Broughton, J, and Lawson, S.D. (2003) European Road Assess-ment Programme, Pilot Phase Technical Report, AA Foundation for Road SafetyResearch, Farnborough, UK.

Lynam, D., Hummel, T., Barker, J., Lawson, S.D., 2004. European road assessmentprogramme, EuroRAP! (2003) Technical report. AA Foundation for Road SafetyResearch, Farnborough, UK.

McNally, M. G., 2000 The Four Step Model. In: Handbook of Transport Modelling, ed.David A. Hensher and Kenneth J. Button, 35-52.

Nystuen, J.D., Dacey, M.F., 1961. A graph theory interpretation of nodal regions.Papers in Regional Science 7 (1), 2942.

SafetyNet, 2005. Deliverable D3.1. State of the art Report on Road Safety PerformanceIndicators, May 2005. SWOV Institute for Road Safety Research, Leidschendam,The Netherlands.

Shef, Y., 1985. Urban Transportation Networks: Equilibrium Analysis with Math-ematical Programming Methods. Prentice-Hall, Incorporated, Englewood Cliffs,NJ 07632, USA.

Wegman, F., Commandeur, J., Doveh, E., Eksler, V., Gitelman, V., Hakkert, S., Lynam,D., Oppe, S., 2008. SUNowerNext: Towards a Composite Road Safety Perfor-mance Index. SWOV Institute for Road Safety Research, Leidschendam, theNetherlands.

Weijermars, W.A.M. (Ed.), 2008. Safety Performance Indicators for Roads: Pilots inthe Netherlands, Greece, Israel and Portugal. Deliverable D3.10c of the EU FP6Project SafetyNet.

Wegman, F., Aarts, L., 2006. Advancing Sustainable Safety: National Road SafetyOutlook for 2005-2020. SWOV Institute for Road Safety Research, Leidschendam.

Wegman, F., Oppe, S., 2000. Benchmarking road safety performances of countries.Safety Science 48 (9), 12031211.

Road safety performance indicators for the interurban road network1 Introduction2 SPI development for road networks2.1 Road design SPI2.2 Road network SPI

3 Road network SPI concept4 Road network SPI development4.1 Method for the determination of the road network SPI

5 Road network SPI application in pilot studies5.1 Pilot studies characteristics5.2 Data availability5.3 Determination of theoretical connections5.4 Identification and matching of actual connections5.5 Calculation of the SPI5.6 Additional observations5.7 Theoretical connections5.8 Calculation of the SPI

6 Discussion7 ConclusionsReferences