Road safety development in Europe: A decade of changes (2001–2010)

Accident Analysis and Prevention 60 (2013) 85–94

Contents lists available at ScienceDirect

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

Road safety development in Europe: A decade of changes (2001–2010) Yongjun Shen ∗ , Elke Hermans, Qiong Bao, Tom Brijs, Geert Wets Transportation Research Institute (IMOB) – Hasselt University, Wetenschapspark 5 bus 6, 3590 Diepenbeek, Belgium

a r t i c l e

i n f o

Article history: Received 19 March 2013 Received in revised form 8 August 2013 Accepted 14 August 2013 Keywords: Road safety development Fatality risk European Union Data envelopment analysis Malmquist productivity index

a b s t r a c t To evaluate the road safety development of a country over time, the percentage change in the number of road fatalities is traditionally the main indicator. However, simply considering the reduction in the road fatalities may not correctly reflect the real improvement in road safety because the transport circumstances of a country underlying the road fatalities also change every year. In this study, we present a new way for measuring the road safety performance change over time, which is to use the technique of data envelopment analysis (DEA) and the Malmquist productivity index. In doing so, we can not only focus on the evolution of road safety final outcomes within a given period, but also take the changes of different measures of exposure in the same period into account. In the application, the DEA-based Malmquist productivity index (DEA-MI) is used to measure the extent to which the EU countries have improved their road safety performance over the period 2001–2010. More objective and insightful results are obtained compared to the ones based on the traditional indicator. The results show considerable road safety progress in most of the Member States during these ten years, and the fatality risk rather than the fatality number on Europe’s roads has actually been reduced by approximately half. However, the situation differed considerably from country to country. The decomposition of the DEA-MI into ‘efficiency change’ and ‘technical change’ further reveals that the bulk of the improvement during the last decade was attained through the adoption of productivity-enhancing new technologies throughout the road transport sector in Europe, rather than through the relatively underperforming countries catching up with those best-performing ones. © 2013 Elsevier Ltd. All rights reserved.

1. Introduction During the past decades, rapid growth of traffic volume, especially motorized road mobility, has resulted in continuously increasing safety problems. Europe, recognized as one of the safest road traffic regions in the world, also suffers from the road crash problems. Due to the high level of car ownership, road transport has emerged as the dominant segment in Europe’s transport sector accounting for roughly 84% of all passenger transport and 46% of freight transport (European Commission, 2012). However, it is also responsible for the majority of negative impacts on safety, which accounts for over 100 times more deaths than all other transportation modes (rail, air, maritime, etc.) together (Forum of European Road Safety Research Institutes & European Conference of Transport Research Institutes, 2009). The estimated annual direct and indirect cost of road traffic injuries in the European Union (EU) exceeds D 130 billion, or over 1% of the EU Gross Domestic Product

∗ Corresponding author. Tel.: +32 11 26 91 43; fax: +32 11 26 91 99. E-mail addresses: [email protected] (Y. Shen), [email protected] (E. Hermans), [email protected] (Q. Bao), [email protected] (T. Brijs), [email protected] (G. Wets). 0001-4575/$ – see front matter © 2013 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.aap.2013.08.013

(GDP) in 2010 (European Commission, 2010a), which is becoming increasingly socially unacceptable and difficult to justify to citizens. Under these circumstances, a large number of road safety strategies and programmes have been launched and put into effect, especially over the last decade. In particular, at the early 21st century, the EU has set itself an ambitious target of halving the yearly number of road fatalities between 2001 and 2010 (European Commission, 2001). Although the initial target was not completely met by the end of 2010 (European Transport Safety Council, 2011), significant progress has been achieved across the EU, and the action has therefore been considered as a strong catalyst of efforts made by Member States to improve their road safety. However, the development in different Member States was unbalanced. Their contribution to this progress was therefore dissimilar. Consequently, benchmarking across countries in terms of their road safety development over the past decade is desirable, which enables policy makers to monitor the effectiveness of the implemented programme in each Member State, to update the strategy and target in the light of the gathered experience, and to further motivate all the Member States to continue their efforts in the future. To compare the development of road safety between countries, the percentage change in the number of road fatalities is traditionally the main indicator, with a higher reduction indicating a better

86

Y. Shen et al. / Accident Analysis and Prevention 60 (2013) 85–94

Fig. 1. Fatality change of 27 EU countries between 2001 and 2010. Data source: European Commission (2012).

rank (European Transport Safety Council, 2011). For instance, Fig. 1 shows the fatality change of each of the 27 EU countries during the last decade and the average change within the EU (i.e., EU27 in the figure). As can be seen, since 2001, road fatalities have been cut by 43% in the EU as a whole. Amongst others, eight countries including Latvia, Estonia, Lithuania, Spain, Sweden, Luxembourg, France and Slovenia, reached the EU 2010 target. Ireland, Germany, United Kingdom, the Netherlands, Belgium and Portugal achieved reductions above the EU average, while the other countries progressed to a lesser extent. In general, there was no country in which the number of road fatalities recorded in 2010 exceeded that of 2001. Such an illustration is intuitive and the results are easy to obtain because the number of road fatalities in two years is the only information needed for the calculation. However, simply considering the reduction in the road fatalities may not correctly reflect the real improvement in road safety because the transport circumstances of a country underlying the road fatalities also change every year. For instance, consider a country that recorded 100 road fatalities in one year with a participation of 10 billion passenger-kilometres (pkm) in traffic, and 90 road fatalities with 9 billion pkm in the second year. Although the number of fatalities is reduced by 10% between these two years, there is actually no improvement in road safety performance because the degree of participation in traffic also decreases by 10% in this country and its fatality rate has thereby not changed during these two years. Consequently, to capture the dynamic road safety development in a country, we should not only focus on its evolution of road fatalities within a given period, but also take the changes of exposure in the same period into account. In other words, the concept of risk, which is defined as the ratio of road safety outcomes and some measure of exposure, should be applied in the context of benchmarking. In this respect, the population size, the number of registered vehicles, and the distance travelled are the three most frequently used measures of exposure to risk (International Traffic Safety Data and Analysis Group, 2012). Nevertheless, there has been considerable debate in the past about which one is the most appropriate indicator of exposure, because they describe risk from different points of view and are not consistent in most cases. In other words, countries may have different evaluation results or ranking positions using different exposure information, which to a great extent baffles decision makers in assessing their relative performance. Therefore, the main objective of this study is, by simultaneously considering the evolution of road fatalities and the change in different measures of exposure, to comprehensively explore the extent to which the EU countries have improved their road safety performance over the period of 2001–2010. In doing so, data envelopment analysis (DEA) (Charnes et al., 1978), which applies a mathematical optimization technique to measure the relative efficiency of a set of decision making units (DMUs) on the basis of multiple inputs and multiple outputs, and the Malmquist

productivity index (Malmquist, 1953), which evaluates productivity change of DMUs over time, are employed to undertake the assessment. The results are further compared with the ones from only considering the percentage change of road fatalities. The remaining of this paper is organized as follows. In Section 2, we briefly introduce the technique of DEA and the construction of the Malmquist productivity index for road safety evaluation. In Section 3, we demonstrate the application of this DEA-based Malmquist productivity index for measuring the road safety development of EU countries over the last decade, and the results are subsequently provided. The paper ends with discussion and concluding remarks in Section 4. 2. Methodology Data envelopment analysis originally developed by Charnes et al. (1978) is a powerful benchmarking tool for measuring the relative efficiency of a homogeneous set of DMUs by constructing a so-called efficient production frontier. In microeconomic production theory, such a frontier indicates the maximum quantity of outputs that can be obtained from a given combination of inputs (Seiford and Thrall, 1990; Coelli et al., 2005). Without any assumptions about the functional form of the frontier which is often complex or even unknown in the real world situation, DEA aims at finding the ‘best’ DMU(s) which is viewed as the most efficient one under the given circumstances, and is used to construct the frontier. The others that either make less outputs with the same inputs or make the same outputs with more inputs are inefficient, and the degree of their inefficiency can be measured based on the distance from the frontier. Since its first introduction in 1978, DEA has been quickly recognized as a powerful analytical research tool for modelling operational processes in terms of performance evaluation, benchmarking, and decision making (Emrouznejad et al., 2008; Cook and Seiford, 2009). In road safety context, the effectiveness of applying the DEA model has also been demonstrated by several recent studies, such as Hermans et al. (2009), and Shen et al. (2011a, 2012). As opposed to the original application field of economics, in which the definition of the best practices relies on the assumption that inputs have to be minimized and outputs have to be maximized, to use DEA for road safety evaluation, the output, i.e., the number of road fatalities, is expected to be as low as possible with respect to the level of exposure. Therefore, the DEA frontier DMUs, or the road safety best-performing countries in this study are those with minimum output levels given certain input levels. For instance, consider the number of fatalities per pkm travelled of two countries P(x0 , y0 ) and Q(x1 , y1 ) in Fig. 2. According to the DEA-based road safety principle that for a given amount of pkm travelled (x), countries having a lower number of fatalities (y) are the efficient (or low-risk) ones, we can thus identify

Y. Shen et al. / Accident Analysis and Prevention 60 (2013) 85–94

y F P(x0, y0) Q(x1, y1) B

O

x

A

Fig. 2. Graphic representation of the efficiency frontier based on the DEA-based road safety model.

that country Q is efficient. Thereby the efficiency frontier (F) is the ray extending from the origin through Q, and the area above this frontier constitutes the production possibilities set, i.e., the set of feasible activities, in which country P is located. Hence, P is inefficient (or has a high risk relative to Q), and its efficiency score can be computed as: AB/AP, which is also defined as the distance function of P, denoted as Do (xo , yo ). Mathematically, consider a set of n DMUs, or the EU countries in this study, each having m different inputs and s different outputs. The efficiency score of a particular DMUo can be obtained by solving the following adjusted output-oriented DEA model (Shen et al., 2012): Do (xo , yo ) = min o s.t.

n 

xij j ≥xio ,

i = 1, · · ·, m

j=1 n



(1) yrj j ≤ o yro ,

r = 1, · · ·, s

j=1

j ≥0,

j = 1, · · ·, n

This linear program is computed separately for each DMU, and the subscript, o, refers to the DMU whose efficiency is to be evaluated. xij and yrj are the ith output and rth input, respectively of the jth DMU. j is an n × 1 nonnegative vector of the weight given to the jth DMU’s inputs and outputs in constructing for DMUo a hypothetical composite unit (HCU) that outperforms it. In other words, those DMUs that contribute to the construction of the HCU, will obtain a non-zero value of , and make up the reference set for DMUo to learn from (El-Mahgary and Lahdelma, 1995). Moreover,  (0 <  ≤ 1) is the uniform proportional reduction in DMUo ’s outputs. Its minimum amount is known as the DEA efficiency score for DMUo , which also equals to its distance function, i.e., Do (xo , yo ). Generally, solving this linear programming problem enables us to find the lowest possible value of , for which there exists a HCU that owns at least as much of each input as DMUo , meanwhile leading to no more than  times each of the outputs of that DMU. Hence, if the value of  equals to one, it means no reduction is needed for this DMU, so it is efficient and its input–output combination lies on the efficient production frontier, such as is the case for country Q in Fig. 2. In the case that  < 1, the DMU is inefficient, and it lies inside the frontier, such as country P whose efficiency score equals to AB/AP < 1. Therefore, without any change of the exposure level, the number of fatalities in country P should be proportionally

87

reduced by 1-AB/AP = BP/AP to become efficient, and thus point B could be treated as the HCU, or the benchmark point of country P. Furthermore, to measure the performance change of DMUs over time, Färe et al. (1992) combined the ideas on the measurement of efficiency and the measurement of productivity to construct a Malmquist productivity index (originally introduced by Malmquist (1953) as a quantity for analyzing the consumption of inputs) directly from input and output data using DEA. Different from other statistical techniques (e.g., time series analysis (Page, 2001; Dupont and Martensen, 2007)), the DEA-based Malmquist productivity index, hereafter referred to as DEA-MI, is a non-parametric method imposing no assumptions on the specific statistical distribution of the error terms, which in most cases is a priori unknown. By using longitudinal data, the DEA-MI relies on firstly constructing an efficient production frontier over the whole sample realized by DEA, and then computing the distance of individual observations to this frontier. In this way, the data are believed to be able to ‘speak for themselves’ and the specification error is minimized. Moreover, in contrast to conventional production functions or other index approaches, the DEA-MI owns an advantage of being able to be further decomposed into two components, one measuring the change in efficiency (EFFCH) and the other measuring the change in the frontier technology (TECHCH). The first term, i.e., EFFCH, indicates the magnitude of the efficiency change from the period t to t + 1, which also reflects the capability of an inefficient DMU in catching up with those efficient ones. The second one, i.e., TECHCH, measures the shift in the efficient production frontier between two time periods. It therefore captures the genuine progress or regress of those efficient DMUs. During the last decade, several relevant studies on road traffic evaluation over time have been carried out by using the DEA-MI approach, such as Odeck (2000, 2006), and Shen et al. (2010, 2011b). Different from the traditional concept of efficiency which is commonly defined as the ratio between the consumption of societal and economical resources (e.g., safety expenditures) and the road traffic achievement (e.g., crash reduction), in this study, the relationship between the number of road fatalities and different measures of exposure is the main focus. Therefore, the efficiency score of a country calculated here can be interpreted as its relative road safety risk, with a higher efficiency score indicating a lower risk. Thus, from the output-oriented view of road safety development assessed in this study, an improvement in efficiency (i.e., EFFCH) means a reduction of the relative fatality risk in each country, which could occur when there are decreases in the quantities of output (i.e., road fatalities) based on a given set of inputs (i.e., measures of exposure). In other words, the number of road fatalities could be reduced without a change in exposure. For instance, maintaining the same population size, more and better road user education and driver training is a useful way to reduce road crash risk and consequent injuries and fatalities (European Commission, 2010b). Moreover, without changing the number of registered vehicles and/or the distance travelled, encouraging citizens to use public transport instead of private cars is also widely recognized as an effective solution for reducing road crash risk in a country (Organization for Economic Co-operation and Development, 2008; Organization for Economic Co-operation and Development/International Transport Forum, 2008). In contrast to a change in efficiency, technical change (i.e., TECHCH) occurs through the adoption of new road safety technologies that reduce the minimum quantities of road fatalities given a certain level of exposure. In this case, the number of road fatalities would be reduced even if the exposure is enlarging. In this respect, the introduction of safer vehicles, betterment of road infrastructure, and improvement in medical treatment of people involved in crashes are all practical examples of improving road safety via technical enhancement (European Commission, 2010b).

88

Y. Shen et al. / Accident Analysis and Prevention 60 (2013) 85–94

y

Ft P(x0t, y0t) Q(x1t, y1t)

H

P’(x0

B

t+1,

y0t+1)

F t+1

Q’(x1t+1, y1t+1) D

G

O

A

C

x

Fig. 3. Graphic representation for EFFCH and TECHCH computation.

TECHCHP =

Towards a safer use of the road, both efficiency enhancements and technical improvements are required. The DEA-MI calculated here will allow us to measure the combined effect of EFFCH and TECHCH of each country within the given period, and it will also capture the separate impact of each effect. Mathematically, the DEA-MI is computed as the product of EFFCH and TECHCH. Therefore, to obtain the total factor productivity change of a DMU over time, we need to firstly derive its EFFCH and TECHCH, respectively. In doing so, consider the same example as presented in Fig. 2, but with two time periods t and t + 1 for the two countries P(x0 , y0 ) and Q(x1 , y1 ) now, which is illustrated in Fig. 3. By identifying the efficient country in each time period, which is Q (x1t , y1t ) and Q  (x1t+1 , y1t+1 ), respectively, we derive the efficiency frontiers Ft and Ft+1 as in Fig. 3, and the efficiency score of country P in each time period can be measured as AB/AP and CD/CP’, respectively. Thus, the magnitude of the efficiency change of country P from period t to t + 1 can be computed as the ratio of these two efficiency scores, which can be further expressed in the corresponding distance function forms as follows: Dt+1 (xt+1 , yt+1 ) EFFCH = o t o t t o Do (xo , yo )

t xt+1 , and yt+1 denote the ith input and rth outwhere xijt , yrj ij rj put of the jth DMU at a given point in time t and t + 1, respectively. For the effect of the efficiency change, which also reflects the capability of an inefficient DMU in catching up with those efficient ones, EFFCH > 1 indicates progress in the relative efficiency of the DMUo from period t to t + 1, while EFFCH = 1 and EFFCH < 1 means respectively no change and regress in efficiency. To fully evaluate the total factor productivity change, we should also take into account the technical change, which measures the shift in the technology frontier between two time periods. In Fig. 3, we notice that the production possibilities set expands from period t to t + 1, as a greater number of input–output combinations become feasible when the frontier moves from Ft to Ft+1 , and the HCU of country P also moves from B to G. Thus, the TECHCH at P(xot , yot ) is evaluated by: AB/AG, which is equivalent to:

(2)

TECHCHP  =



t xijt j ≥xio ,



TECHCH =

(3) r = 1, · · ·, s

s.t.

t+1 xijt+1 j ≥xio ,

i = 1, · · ·, m (4)

t+1 t+1 yrj j ≤ o yro ,

j=1

j = 1, · · ·, n

i = 1, . . ., m

j=1 n



(8) t+1 t yrj j ≤ yro ,

r = 1, . . ., s

j=1

Dot (xot+1 , yot+1 )

j = 1, . . ., n



= min 

n

t+1 xijt j ≥xio ,

i = 1, · · ·, m (9)

t yrj j

j ≥0,

j=1

j ≥0,

t xijt+1 j ≥xio ,

≤

t+1 yro ,

r = 1, · · ·, s

j=1

n

n 

n 



j = 1, · · ·, n

Dot+1 (xot+1 , yot+1 ) = min o



(7)

Dot+1 (xot , yot ) = min 

j=1 n

and

s.t.

Dot+1 (xot , yot ) Dot+1 (xot+1 , yot+1 )

1/2

where the two mixed-period measures, i.e., Dot+1 (xot , yot ) and Dot (xot+1 , yot+1 ), can be derived from the following modification of model (1):

s.t.

j=1

j ≥0,

(6)

and

i = 1, · · ·, m

t t yrj j ≤ o yro ,

Dot (xot+1 , yot+1 )

Dot (xot , yot )

j ≥0,

j=1 n



CH/CP  Dot (xot+1 , yot+1 ) = t+1 CH/CP  Do (xot+1 , yot+1 )

where the numerator Dot (xot+1 , yot+1 ) represents the relative efficiency of P  (xot+1 , yot+1 ) relative to the frontier at time t, i.e., Ft . The overall TECHCH is therefore defined as the geometric mean of the above two TECHCHs:

Dot (xot , yot ) = min o n

(5)

where the denominator Dot+1 (xot , yot ) denotes the relative efficiency of P(xot , yot ) with respect to the frontier at time t + 1, i.e., Ft+1 . Similarly, the TECHCH at P  (xot+1 , yot+1 ) can be expressed by:

where the two distance functions can be computed by means of model (1), and they are represented as below:

s.t.

Dot (xot , yot ) AB/AP = t+1 AB/AP Do (xot , yot )

r = 1, · · ·, s

j = 1, · · ·, n

For the change in the frontier technology, values greater than one indicate an improvement in this aspect, while values equal to and less than one imply status quo and deterioration, respectively. By now, the DEA-MI, which measures the total factor productivity change of a particular DMUo from period t to period t + 1, can be computed as the product of EFFCH and TECHCH:

Y. Shen et al. / Accident Analysis and Prevention 60 (2013) 85–94

89

Fig. 4. The evolution in road fatalities and MI of the 26 EU countries and its decomposition into efficiency change and technical change in 2001–2010.



MIo =

 =

Dot+1 (xot+1 , yot+1 ) Dot (xot , yot ) Dot (xot+1 , yot+1 ) Dot (xot , yot ) Dot+1 (xot , yot ) Dot+1 (xot+1 , yot+1 ) Dot (xot+1 , yot+1 ) Dot+1 (xot+1 , yot+1 ) Dot (xot , yot ) Dot+1 (xot , yot )

1/2

1/2

(10)

MIo > 1 indicates progress in the total factor productivity of the DMUo from period t to t + 1, while MIo = 1 and MIo < 1 means respectively status quo and decay in productivity. In the following section, the DEA-MI is applied to evaluate the road safety development in Europe over time. Meanwhile, the two effects on efficiency enhancements and technical improvements are measured separately for cross-country comparisons. 3. Application and results To assess the dynamic road safety progress in Europe during the last decade, this study considers simultaneously the evolution in the number of road fatalities and the changes in three common measures of exposure to risk, i.e., the number of inhabitants, passenger-kilometres travelled and passenger cars,1 as output and input, respectively. Data from 2001 to 2010 are collected for the 26 EU countries (the EU-27 except Malta2 ), which are Belgium (BE), Bulgaria (BG), Czech Republic (CZ), Denmark (DK), Germany

1 The total number of motor vehicles and the number of motor vehicle kilometres travelled are more preferred to be used as a true representation of the level of a country’s motorization, as soon as the data on these measures are available. 2 The initial examination revealed the very distinct nature of the data for Malta (see also Shen et al., 2012) and consequently, it was decided to eliminate this outlier and to consider only 26 EU countries.

(DE), Estonia (EE), Ireland (IE), Greece (EL), Spain (ES), France (FR), Italy (IT), Cyprus (CY), Latvia (LV), Lithuania (LT), Luxembourg (LU), Hungary (HU), the Netherlands (NL), Austria (AT), Poland (PL), Portugal (PT), Romania (RO), Slovenia (SI), Slovakia (SK), Finland (FI), Sweden (SE), and United Kingdom (UK) (European Commission, 2012). The data for 2001 and 2010 are shown in Table 1. 3.1. The overall results The DEA-MI is now adopted to measure the extent to which the 26 EU countries have improved their level of road safety during the period under study. The overall results are shown in Fig. 4, together with the evolution of road fatalities during the same time period. The upper portion of the figure indicates the cumulative MI of the 26 EU countries and its decomposition (i.e., EFFCH and TECHCH) from 2001 to 2010 by sequential multiplication of the improvements in each year with 2001 as the index year (equal to one). From the trend of MI, we can see that the EU as a whole achieved considerable improvement in road safety performance during the last decade (i.e., 95%). Although a slight decrease existed in 2007, the total ‘productivity’ went steadily up during this period, and the trend was much steeper in the last three years (improved from 45% in 2008 to 95% in 2010). Further comparing the MI result with the evolution of road fatalities during the same time period shown in the lower part of Fig. 4, we can see that they are mirror images of each other, but are not identical. First, the number of road fatalities was slightly decreased between 2006 and 2007 (from 43,104 to 42,540). However, from the fatality risk point of view, the situation was actually deteriorating in this period. Moreover, considering the total reduction of the number of road fatalities, around 43% has been cut during

90

Y. Shen et al. / Accident Analysis and Prevention 60 (2013) 85–94

Table 1 Input and output data of the 26 EU countries for 2001 and 2010. Country

BE BG CZ DK DE EE IE EL ES FR IT CY LV LT LU HU NL AT PL PT RO SI SK FI SE UK

Input

Output

Population (million)

Passenger-kilometres (10 billion)

Passenger cars (million)

Fatalities

2001

2010

2001

2010

2001

2010

2001

2010

10.29 8.02 10.24 5.36 82.35 1.36 3.87 10.95 40.72 59.48 56.98 0.70 2.36 3.48 0.44 10.19 16.05 8.04 38.25 10.29 22.13 1.99 5.38 5.19 8.90 59.11

10.90 7.53 10.52 5.55 81.78 1.34 4.47 11.31 46.07 62.96 60.48 0.80 2.24 3.29 0.51 10.00 16.62 8.39 38.18 10.64 21.44 2.05 5.43 5.36 9.38 62.23

10.69 2.79 6.35 4.96 85.26 0.68 3.98 6.80 30.80 71.57 72.05 0.40 1.20 2.60 0.58 4.62 14.16 6.71 15.77 7.32 5.25 2.08 2.41 5.70 9.28 65.38

10.91 4.69 6.36 5.10 88.70 1.01 4.60 9.96 34.16 72.73 70.02 0.59 1.65 2.99 0.65 5.26 14.12 7.30 29.79 8.37 7.55 2.56 2.69 6.47 9.92 65.38

4.71 2.04 3.48 1.86 39.22 0.44 1.37 3.31 17.80 30.07 32.91 0.27 0.57 1.15 0.28 2.42 6.62 4.14 10.25 3.52 2.83 0.87 1.28 2.15 4.01 25.44

5.23 2.55 4.47 2.14 42.02 0.55 1.92 5.17 22.07 31.55 36.56 0.46 0.77 1.69 0.33 3.00 7.58 4.40 16.87 4.47 4.28 1.06 1.63 2.83 4.32 29.24

1486 1011 1333 431 6977 199 412 1880 5517 8162 7096 98 558 706 70 1239 993 958 5534 1670 2450 278 614 433 583 3598

812 776 802 255 3648 78 212 1258 2479 3992 4090 60 218 300 32 740 537 552 3908 937 2377 138 371 272 266 1905

Source: European Commission (2012).

the last decade. If we use the concept that 50% reduction of road fatalities means the improvement in road safety by 100%, we can calculate that during the last decade, the EU has achieved a road safety improvement of 75% (= 1/(1 − 43%) − 1). However, if we also take the changes of exposure into account, the improvement over these ten years was actually 20% more. In addition, by decomposing the MI score into EFFCH and TECHCH, we can see that the road safety improvement in Europe during the last decade was mostly dominated by the technical component, which means that the main source of this growth came about through the adoption of productivity-enhancing new technologies throughout the road transport sector in Europe, which was partially offset through a slight decrease in efficiency. Taking the EFFCH and TECHCH of the last three years as an example, in 2008 and 2009, both road safety efficiency and technology of these 26 EU countries were improved, which enabled the total factor productivity, i.e., the overall road safety performance, to increase dramatically.3 However, when it comes to 2010, great improvement has been made in the best-performing countries, thereby resulting in a remarkable shift in the technology frontier. On the other hand, the underperforming countries showed difficulties in keeping pace with their benchmarks.

3.2. Cross-country comparisons Although remarkable improvement in terms of road safety performance has been achieved in Europe during the last decade, the situation could differ considerably from one country to another. Therefore, apart from analyzing the road safety development of the EU countries by considering them as a whole, investigation of the

3 The effects of the 2008 financial and economic crisis on road traffic perhaps partly explain this favourable development.

progress in each of these countries is also important. The crosscountry comparisons are thus provided in the following sections. 3.2.1. Efficiency change: the catch-up effects To compare the road safety progress in these 26 EU countries during the past decade, we first look at the changes in their relative efficiency, which actually reflect the capability of a underperforming country in catching up with those best-performing ones in terms of road safety risk. Tables 2 and 3 present the DEA efficiency scores and the corresponding efficiency changes of these countries over the period 2001–2010. It can be seen from Table 2 that the SUN countries, i.e., Sweden, United Kingdom, and the Netherlands, were the best-performing countries in terms of road safety because they obtained the efficiency score of one alternatively during these ten years. In other words, they had the lowest road fatality risk and also determined the risk levels of other countries since they were the ones that shifted the frontier in this period. The remaining countries, however, had an efficiency score less than one in each time period, and both improvement and decline occurred during these ten years. Within these countries, there were still more than half whose overall efficiency change (2010 compared to 2001) was less than one (see the last column of Table 3), which implies their weak capability in catching up with those best-performing countries. On the contrary, comparison of development up to 2010 shows that Spain, Estonia, Lithuania and Luxembourg achieved the best improvement in terms of efficiency (all over 15%). In other words, the road safety progress in these countries during the last decade was at least 15% faster than that of the SUN countries. 3.2.2. Technical change: the frontier-shift effects Having analyzed the efficiency changes for all these countries, we now take into account their changes in the frontier technology so as to fully evaluate the overall road safety improvement of each country. The results are shown in Table 4.

Y. Shen et al. / Accident Analysis and Prevention 60 (2013) 85–94

91

Table 2 Efficiency scores of the 26 EU countries over the period 2001–2010. Country

BE BG CZ DK DE EE IE EL ES FR IT CY LV LT LU HU NL AT PL PT RO SI SK FI SE UK

Efficiency score 2001

2002

2003

2004

2005

2006

2007

2008

2009

2010

0.448 0.483 0.467 0.757 0.795 0.417 0.571 0.355 0.456 0.521 0.656 0.436 0.257 0.300 0.559 0.501 0.984 0.611 0.421 0.375 0.550 0.445 0.533 0.729 0.972 1

0.499 0.495 0.431 0.701 0.792 0.368 0.631 0.406 0.472 0.545 0.657 0.456 0.253 0.300 0.627 0.429 0.988 0.585 0.396 0.378 0.546 0.452 0.532 0.756 0.985 1

0.516 0.481 0.417 0.737 0.786 0.487 0.700 0.406 0.459 0.660 0.675 0.440 0.258 0.288 0.714 0.451 0.932 0.563 0.400 0.400 0.576 0.487 0.493 0.854 1 1

0.485 0.407 0.365 0.725 0.794 0.392 0.564 0.327 0.466 0.652 0.645 0.315 0.221 0.226 0.686 0.385 1 0.538 0.330 0.401 0.438 0.398 0.441 0.804 1 0.999

0.479 0.372 0.366 0.756 0.803 0.364 0.501 0.308 0.477 0.612 0.628 0.360 0.239 0.211 0.687 0.363 1 0.573 0.322 0.389 0.378 0.394 0.409 0.739 1 0.915

0.484 0.330 0.431 0.801 0.825 0.294 0.569 0.300 0.511 0.700 0.629 0.432 0.251 0.238 0.740 0.345 1 0.584 0.325 0.488 0.373 0.403 0.392 0.853 1 0.948

0.480 0.330 0.366 0.597 0.822 0.296 0.633 0.301 0.541 0.708 0.671 0.427 0.235 0.236 0.678 0.353 1 0.593 0.296 0.471 0.333 0.371 0.350 0.750 0.951 1

0.491 0.296 0.399 0.563 0.838 0.418 0.683 0.297 0.642 0.668 0.688 0.472 0.295 0.298 0.844 0.415 1 0.570 0.289 0.494 0.289 0.453 0.358 0.719 0.993 1

0.439 0.318 0.440 0.689 0.803 0.517 0.705 0.293 0.653 0.587 0.687 0.511 0.336 0.365 0.553 0.461 0.970 0.548 0.315 0.479 0.290 0.518 0.539 0.792 0.982 1

0.397 0.275 0.372 0.617 0.710 0.487 0.599 0.255 0.548 0.489 0.551 0.474 0.291 0.348 0.644 0.383 0.878 0.491 0.277 0.322 0.256 0.498 0.415 0.640 1 0.946

We can see that although fluctuations occurred in every country within these ten years, the overall technology performance of these 26 EU countries were all doubled (see the last column of Table 4), which means that due to the adoption of productivity-enhancing new technologies (such as the introduction of safer vehicles and the betterment of road infrastructure) throughout the road transport sector in each country over the last decade, the same exposure on the roads in 2010 would lead up to only half or even lower number of fatalities relative to the situation in 2001. Among others, Lithuania, Cyprus, Italy, Austria and Germany were the

technological innovators, which recorded an improvement of around 130% compared to 2001.

3.2.3. Total factor productivity change: the overall road safety progress Considering both efficiency change and technical change together, the total factor productivity change, or the overall road safety progress in each of these 26 EU countries during the last decade can now be deduced, which is illustrated in Fig. 5.

Table 3 Efficiency changes of the 26 EU countries from 2001 to 2010. Country

BE BG CZ DK DE EE IE EL ES FR IT CY LV LT LU HU NL AT PL PT RO SI SK FI SE UK

EFFCH 02/01

03/02

04/03

05/04

06/05

07/06

08/07

09/08

10/09

10/01

1.114 1.026 0.921 0.926 0.996 0.881 1.105 1.145 1.035 1.046 1.001 1.047 0.983 1.001 1.121 0.857 1.004 0.957 0.941 1.008 0.993 1.017 0.998 1.037 1.013 1.000

1.033 0.972 0.967 1.051 0.993 1.325 1.109 0.999 0.973 1.212 1.029 0.965 1.022 0.958 1.138 1.051 0.944 0.962 1.010 1.057 1.055 1.077 0.925 1.129 1.015 1.000

0.94 0.847 0.876 0.984 1.011 0.804 0.806 0.806 1.014 0.988 0.955 0.717 0.858 0.784 0.961 0.854 1.073 0.955 0.825 1.002 0.761 0.818 0.895 0.942 1.000 0.999

0.988 0.912 1.002 1.042 1.011 0.928 0.888 0.941 1.024 0.938 0.973 1.143 1.081 0.935 1.002 0.942 1.000 1.065 0.976 0.970 0.862 0.991 0.927 0.918 1.000 0.916

1.012 0.887 1.179 1.060 1.027 0.808 1.137 0.976 1.072 1.143 1.002 1.197 1.050 1.130 1.077 0.952 1.000 1.019 1.008 1.255 0.985 1.021 0.960 1.155 1.000 1.037

0.990 1.000 0.849 0.746 0.997 1.007 1.112 1.000 1.059 1.012 1.066 0.990 0.936 0.991 0.916 1.023 1.000 1.016 0.910 0.966 0.894 0.922 0.893 0.879 0.951 1.055

1.023 0.898 1.089 0.942 1.019 1.411 1.079 0.990 1.186 0.943 1.025 1.106 1.256 1.262 1.245 1.174 1.000 0.96 0.977 1.048 0.869 1.221 1.022 0.959 1.044 1.000

0.895 1.076 1.103 1.224 0.959 1.237 1.033 0.985 1.017 0.880 0.999 1.082 1.137 1.225 0.655 1.111 0.970 0.963 1.093 0.969 1.004 1.142 1.507 1.102 0.989 1.000

0.904 0.865 0.845 0.895 0.883 0.942 0.849 0.870 0.840 0.832 0.802 0.927 0.868 0.952 1.165 0.831 0.904 0.896 0.878 0.673 0.881 0.962 0.770 0.808 1.018 0.946

0.886 0.571 0.794 0.814 0.893 1.165 1.049 0.718 1.203 0.938 0.840 1.089 1.135 1.159 1.151 0.765 0.892 0.803 0.658 0.858 0.465 1.121 0.779 0.878 1.028 0.947

92

Y. Shen et al. / Accident Analysis and Prevention 60 (2013) 85–94

Table 4 Technical changes of the 26 EU countries from 2001 to 2010. Country

BE BG CZ DK DE EE IE EL ES FR IT CY LV LT LU HU NL AT PL PT RO SI SK FI SE UK

TECHCH 02/01

03/02

04/03

05/04

06/05

07/06

08/07

09/08

10/09

10/01

1.033 1.008 1.008 1.008 1.033 1.008 1.008 1.008 1.029 1.033 1.033 1.008 1.008 1.008 1.033 1.008 1.008 1.033 1.008 1.008 1.008 1.030 1.008 1.015 1.033 1.024

1.051 1.022 1.022 1.022 1.051 1.022 1.022 1.022 1.034 1.049 1.051 1.022 1.022 1.022 1.051 1.022 1.022 1.051 1.022 1.022 1.022 1.040 1.022 0.982 1.049 0.988

1.122 1.196 1.196 1.194 1.127 1.196 1.131 1.196 1.143 1.109 1.127 1.190 1.196 1.196 1.127 1.196 1.196 1.127 1.196 1.196 1.196 1.110 1.196 1.097 1.110 1.099

1.091 1.075 1.075 1.069 1.087 1.075 1.088 1.075 1.087 1.097 1.087 1.086 1.075 1.079 1.087 1.075 1.077 1.087 1.075 1.075 1.075 1.095 1.075 1.095 1.097 1.095

1.017 1.029 1.029 1.023 1.035 1.029 0.994 1.029 1.044 0.990 1.044 1.044 1.029 0.993 1.041 1.029 1.035 1.044 1.029 1.029 1.029 0.985 1.029 0.985 0.992 0.985

1.025 1.032 1.032 1.030 1.037 1.032 1.016 1.032 1.051 1.017 1.051 1.051 1.032 1.027 1.040 1.032 1.041 1.051 1.032 1.032 1.032 1.025 1.032 1.027 1.015 1.027

1.115 1.051 1.051 1.066 1.089 1.051 1.146 1.051 1.070 1.144 1.070 1.070 1.051 1.148 1.077 1.051 1.059 1.070 1.051 1.051 1.051 1.145 1.051 1.145 1.137 1.150

1.133 1.089 1.089 1.099 1.132 1.089 1.135 1.089 1.129 1.136 1.129 1.129 1.089 1.135 1.131 1.089 1.094 1.129 1.089 1.089 1.089 1.135 1.089 1.135 1.131 1.137

1.304 1.333 1.333 1.333 1.304 1.333 1.317 1.333 1.304 1.298 1.304 1.304 1.333 1.304 1.304 1.333 1.333 1.304 1.333 1.333 1.333 1.281 1.333 1.292 1.326 1.294

2.282 2.146 2.146 2.164 2.295 2.146 2.193 2.146 2.285 2.239 2.299 2.301 2.146 2.304 2.287 2.146 2.208 2.299 2.146 2.146 2.146 2.188 2.146 2.034 2.265 2.084

Fig. 5. Overall road safety progress in the 26 EU countries from 2001 to 2010.

Except for Romania, which had an overall MI score less than one (0.998) indicating slight deterioration in its road safety development, all other countries have improved their road safety performance during this period. Further comparing the result with the one in Fig. 1, which is based only on the fatality change between 2001 and 2010, a relatively high correlation coefficient (0.856) can be derived, and the country ranking, especially of those worstperforming ones, is quite consistent. However, some differences can also be found. First, apart from the eight countries who have reached the EU 2010 target of halving the road fatalities over the last decade, four more countries have also doubled their performance if the fatality risk is considered. They are Cyprus, Ireland, Germany and Belgium, in which Cyprus has only reduced its road fatalities by less than 40%. However, due to its rapid growth in the degree of participation in traffic during the past decade (14.6% increase in population size (ranking third among the 26 EU countries), 47.5% more passenger-kilometres travelled compared to 2001 (ranking fourth), and 68.6% more registered passenger cars (ranking first), see also Fig. 6), the fatality risk in Cyprus has actually been reduced by more than half. Moreover, although Latvia achieved the highest reduction in road fatalities (i.e., 61%), its overall road safety performance change was inferior to countries like Spain based on the

DEA-MI, which could be mainly attributable to the fact that its population size also declined during the last decade (see Fig. 6). In other words, the great progress in the number of road fatalities in Latvia was partially offset by the reduction of its exposure during the same time period. On the contrary, due to the prominent reduction in road fatalities and steady growth in the degree of participation in traffic, Spain actually achieved the greatest road safety progress among all these 26 EU countries.

4. Discussion and concluding remarks In this study, we investigated the road safety development in Europe over the last decade (2001–2010). Different from only considering the percentage change in road fatalities, or the simple ratio analysis (i.e., using a single risk indicator), this paper presented a new way for assessing the road safety performance change of countries over time, which was to use the technique of data envelopment analysis and the Malmquist productivity index. In doing so, we could not only focus on the evolution of road safety final outcomes within a given period, but also take the change of different measures of exposure in the same period into account.

Y. Shen et al. / Accident Analysis and Prevention 60 (2013) 85–94

93

Fig. 6. Changes in the number of road fatalities and three measures of exposure between 2001 and 2010 for Cyprus, Latvia, and Spain.

From the methodological point of view, due to the built-in capability of allowing direct peer comparisons on the basis of multiple inputs and multiple outputs related to DMUs, DEA is superior to the simple ratio analysis (e.g., the number of fatalities per distance travelled) which provides only partial measures of the multiple input–output relations and thus often leads up to misclassifications and incorrect judgments (Lewin et al., 1982). Moreover, the inputs and outputs used in the model can be expressed in different units of measurement. In other words, the preliminary normalization (e.g., standardization) of raw data is not required, which is particularly convenient from a practical point of view and eliminates the sensitivity of the results with respect to the specific normalization scheme that is used (Organization for Economic Cooperation and Development, 2008). However, as a non-parametric method, DEA does not allow for measurement error or random shocks, thereby inferential statistics and some traditional mechanisms such as hypothesis testing are not directly applicable (Cooper et al., 2004). With respect to the Malmquist productivity index, it has been used in a variety of studies during the last decades for different sectors such as agriculture, airlines, banking, electric utilities and so on (Tone, 2004), and has proved to be a proper tool for measuring the productivity change of DMUs over time, especially due to its unique capability of decomposing the index score into ‘efficiency change’ and ‘technical change’. The first term indicates the magnitude of the change in the efficiency score of each DMU between two time periods, while the latter one measures the shift in the efficient production frontier during the same time periods. In this study, by applying the DEA-based Malmquist productivity index, these concepts have been translated to the road safety field. Specifically, the efficiency score of a country is interpreted as the inverse of its road safety risk relative to the best performers (i.e., a higher efficiency score indicates a lower risk). Thus, the efficiency change calculated here measures the change in the relative fatality risk in each country, which thereby indicates the capability of an underperforming country in catching up with those best-performing ones. The technical change, on the other hand, captures the genuine progress or regress of those best-performers, i.e., the change in the minimum number of road fatalities given a certain level of exposure, due to the adoption of productivity-enhancing new technologies. In the application, by considering simultaneously the three common measures of exposure, i.e., the number of inhabitants, passenger-kilometres travelled and passenger cars on the one hand, and the number of road fatalities on the other hand, the DEA-based Malmquist productivity index was used as a valuable benchmarking tool to measure the extent to which the EU countries have improved their road safety performance over the period

2001–2010. The result indicated first that although the ambitious target set in 2001 to halve the number of road fatalities by 2010 was not completely met, there was significant road safety progress in the EU over these ten years – an overall improvement of 95%. In other words, the fatality risk rather than the fatality number on Europe’s roads was actually reduced by approximately half. Moreover, the decomposition of the DEA-MI into efficiency change and technical change further revealed that the bulk of this progress was attained through an overall improvement in the technological environment of the road transport sector in Europe, rather than through the relatively underperforming countries catching up with those best-performing ones. Further looking at each country individually, we found that the development in these 26 EU countries was different considerably. In particular, the SUN countries (Sweden, United Kingdom, and the Netherlands) have proven themselves as best road safety performers since they obtained the efficiency score of one alternatively in each of these ten years, while most of the others were still half on their way to become best-performers. Nevertheless, their outstanding road safety performance on the other hand also restricted to a certain extent their further improvement. As can be seen, these three countries were not the ones who achieved the greatest road safety progress over the last decade. Two of them (i.e., United Kingdom, and the Netherlands) even didn’t reduce their fatality risk by half. In this respect, apart from the eight countries (i.e., Latvia, Estonia, Lithuania, Spain, Sweden, Luxembourg, France and Slovenia) who reached the EU 2010 target, four more countries also doubled their road safety performance if the fatality risk was taken into account. They were Cyprus, Ireland, Germany and Belgium. In other words, although the number of road fatalities in these four countries has not been halved during the past decade, due to the rapid growth in the degree of participation in traffic, their fatality risk has actually been reduced by more than half. Furthermore, among all these 26 EU countries, although the highest reduction in road fatalities was attained by Latvia, Spain actually gained the best improvement in terms of efficiency (or road safety risk), while Lithuania was the technological innovator. Both of them thereby achieved the greatest road safety progress during the last decade. With regard to the remaining countries, some of them were still getting stuck in the rut or even deteriorating in terms of their road safety performance, such as Romania (even though its fatality number actually went downwards during this period). All these verify the fact that simply considering the reduction of road fatalities may not correctly reflect the real improvement in road safety performance because the transport circumstances of a country which can impact on the fatalities also change every year. The approach used in this study thus makes the comparisons between countries more

94

Y. Shen et al. / Accident Analysis and Prevention 60 (2013) 85–94

justly. However, it should be noted here that since only the information on some common measures of exposure was considered in this study, the correlations between road fatalities and some safety measures, policies, and culture within the complex environment of cross-country comparisons were not addressed in this study. In 2010, the European Commission proposed in its ‘Policy Orientations on Road Safety 2011–2020 (European Commission, 2010a) to continue with the target of halving the total number of road fatalities in the EU by 2020, which is apparently more challenging than the previous one yet gives a clear signal of Europe’s commitment towards road safety. Achieving such an ambitious target requires great efforts from all the Member States to better understand their own relative road safety situation, and moreover, to learn from those best-performing ones as a basis for developing their own road safety policy. Consequently, by collecting the crash and exposure data at regular intervals, systematic country comparisons over time should be conducted throughout the whole target period so as to evaluate the results of policy interventions and to monitor the progress in road safety performance of each country. The approach applied in this study is therefore particularly valuable for this purpose, and it can be easily extended when other road safety outcomes, such as the number of crashes and the range of injury severities, as well as other measures of exposure or even other factors such as policy, economic, and cultural indicators, are taken into account. Furthermore, by using the concept of variable return to scale (Banker et al., 1984), the disparities between countries can also be captured, which can be realized by further decomposing the efficiency change calculated in this study (which is based on the constant return to scale) into scale efficiency change and pure technical efficiency change. For more information, we refer to Tone (2004). In the future, more research attention will be paid to the sensitivity and stability analysis of the methodology (see e.g., Jenkins and Anderson, 2003; Tortosa-Ausina et al., 2008). In addition, explorations on the reasons behind the progress or decline in each country and the prediction of its future development are also worthwhile (see e.g., Lassarre and Thomas, 2005; Yannis et al., 2011). In doing so, however, detailed information (to be determined, e.g., by means of interviewing national experts) is required. References Banker, R.D., Charnes, A., Cooper, W.W., 1984. Some models for estimating technical and scale inefficiencies in data envelopment analysis. Manage. Sci. 30, 1078–1092. Charnes, A., Cooper, W.W., Rhodes, E., 1978. Measuring the efficiency of decision making units. Eur. J. Oper. Res. 2, 429–444. Coelli, T.J., Rao, D.S.P., O’Donnell, C.J., Battese, G.E., 2005. An Introduction to Efficiency and Productivity Analysis, 2nd ed. Springer. Cook, W.D., Seiford, L.M., 2009. Data envelopment analysis (DEA) – thirty years on. Eur. J. Oper. Res. 192, 1–17. Cooper, W.W., Li, S., Seiford, L.M., Zhu, J., 2004. Sensitivity analysis in DEA. In: Cooper, W.W., Seiford, L.M., Zhu, J. (Eds.), Handbook on Data Envelopment Analysis. Kluwer Academic Publishers, Boston, pp. 75–98. Dupont, E., Martensen, H. (Eds.), 2007. Multilevel Modelling and Time Series Analysis in Traffic Research – Methodology. Deliverable D7.4 of the EU FP6 project SafetyNet. El-Mahgary, S., Lahdelma, R., 1995. Data envelopment analysis: visualizing the results. Eur. J. Oper. Res. 85, 700–710.

Emrouznejad, A., Parker, B.R., Tavares, G., 2008. Evaluation of research in efficiency and productivity: a survey and analysis of the first 30 years of scholarly literature in DEA. J. Socio-Econ. Plan. Sci. 42 (3), 151–157. European Commission (EC), 2001. White Paper: European Transport Policy for 2010: Time to Decide, Commission of the European Communities, Brussels. European Commission (EC), 2010a. Towards a European Road Safety Area: Policy Orientations on Road Safety 2011–2020. COM (2010) 389 final, Commission of the European Communities, Brussels. European Commission (EC), 2010b. Best practices in road safety: Handbook for measures at the country level. Commission of the European Communities, Brussels. European Commission (EC), 2012. EU Energy and Transport in Figures 2012. Commission of the European Communities, Brussels. European Transport Safety Council (ETSC), 2011. 2010 Road Safety Target Outcome: 100,000 Fewer Deaths Since 2001, 5th Road Safety PIN Report. ETSC, Brussels. Färe, R., Grosskopf, S., Lindgren, B., Roos, P., 1992. Productivity change in Swedish pharmacies 1980–1989: a nonparametric Malmquist approach. J. Product. Anal. 3, 85–102. Forum of European Road Safety Research Institutes & European Conference of Transport Research Institutes (FERSI/ECTRI), 2009. The Sustainable Safety Approach to Road Transport and Mobility. FERSI/ECTRI http://www.ectri. org/Publications/Strategic-Documents.htm Hermans, E., Brijs, T., Wets, G., Vanhoof, K., 2009. Benchmarking road safety: lessons to learn from a data envelopment analysis. Acc. Anal. Prevent. 41 (1), 174–182. International Traffic Safety Data and Analysis Group, 2012. IRTAD Road Safety Annual Report 2011. Organization for Economic Co-operation and Development/International Transport Forum, Paris, www.irtad.net. Jenkins, L., Anderson, M., 2003. A multivariate statistical approach to reducing the number of variables in data envelopment analysis. Eur. J. Oper. Res. 147, 51–61. Lassarre, S., Thomas, I., 2005. Exploring road mortality ratios in Europe: national versus regional realities. J. R. Stat. Soc. Ser. A 168 (1), 127–144. Lewin, A.Y., Morey, C.R., Cook, T.J., 1982. Evaluating the administrative efficiency of courts. Omega Int. J. Manage. Sci. 10 (4), 401–411. Malmquist, S., 1953. Index numbers and indifference surfaces. Trab. Estat. 4, 209–242. Odeck, J., 2000. Assessing the relative efficiency and productivity growth of vehicle inspection services: an application of DEA and Malmquist indices. Eur. J. Oper. Res. 126, 501–514. Odeck, J., 2006. Identifying traffic safety best practice: An application of DEA and Malmquist indices. Omega 34 (1), 28–40. Organization for Economic Co-operation and Development (OECD), 2008. Handbook on Constructing Composite Indicators: Methodology and User Guide. OECD, Paris. Organization for Economic Co-operation and Development/International Transport Forum (OECD/ITF), 2008. Towards Zero: Ambitious Road Safety Targets and the Safe System Approach. Joint Transport Research Centre of the OECD and the ITF, Paris. Page, Y., 2001. A statistical model to compare road mortality in OECD countries. Acc. Anal. Prev. 33 (3), 371–385. Seiford, L.M., Thrall, R.M., 1990. Recent developments in DEA: the mathematical programming approach to frontier analysis. J. Econ. 46, 7–38. Shen, Y., Hermans, E., Ruan, D., Vanhoof, K., Brijs, T., Wets, G., 2010. A DEA-based Malmquist productivity index approach in assessing road safety performance. In: Proceedings of the 9th International FLINS Conference, Chengdu, World Scientific Proceedings Series on Computer Engineering and Information Science, Vol. 4, pp. 923–928. Shen, Y., Hermans, E., Ruan, D., Wets, G., Brijs, T., Vanhoof, K., 2011a. A generalized multiple layer data envelopment analysis model for hierarchical structure assessment: a case study in road safety performance evaluation. Exp. Syst. Appl. 38 (12), 15262–15272. Shen, Y., Ruan, D., Hermans, E., Brijs, T., Wets, G., Vanhoof, K., 2011b. Sustainable road transport in the European Union: changes in undesirable impacts. Transport. Res. Record: J. Transport. Res. Board 2242, 37–44. Shen, Y., Hermans, E., Brijs, T., Wets, G., Vanhoof, K., 2012. Road safety risk evaluation and target setting using data envelopment analysis and its extensions. Acc. Anal. Prevent. 48, 430–441. Tone, K., 2004. Malmquist Productivity Index: Efficiency Change Over Time. In: Cooper, W.W., Seiford, L.M., Zhu, J. (Eds.), Handbook on Data Envelopment Analysis. Kluwer Academic Publishers, Boston, pp. 203–228. Tortosa-Ausina, E., Grifell-Tatje, E., Armero, C., Conesa, D., 2008. Sensitivity analysis of efficiency and Malmquist productivity indices: an application to Spanish savings banks. Eur. J. Oper. Res. 184, 1062–1084. Yannis, G., Antoniou, C., Papadimitriou, E., Katsohis, D., 2011. When may road fatalities start to decrease? J. Saf. Res. 42 (1), 17–25.