Accident Analysis and Prevention 40 (2008) 12001210

Dimensions of road safety problems and their measurement

Rune Elvik
Institute of Transport Economics, Gaustadalleen 21, NO-0349 Oslo, Norway
Received 1 January 2007; received in revised form 25 December 2007; accepted 29 January 2008

Abstract
This paper identifies nine characteristics of road safety problems that are all in principle amenable to numerical measurement. The nine
characteristics identified are:
1.
2.
3.
4.
5.
6.
7.
8.
9.

Magnitude
Severity
Externality
Inequity
Complexity
Spatial dispersion
Temporal stability
Perceived urgency
Amenability to treatment

The purpose of identifying these dimensions and of trying to measure them is to provide a basis for selecting problems for treatment by means
of safety programmes. Selecting problems for treatment usually cannot be done on the basis of a single dimension, as it is the mix of characteristics
that determine the prospects for successfully treating a problem. It is proposed that amenability to treatment is a function of complexity, perceived
urgency and the availability of cost-effective treatments. Speed and speeding is used as an example of a road safety problem to illustrate how the
various dimensions can be measured.
2008 Elsevier Ltd. All rights reserved.
Keywords: Road safety problem; Dimension; Numerical measurement; Selection for treatment

1. Introduction
Road safety has been greatly improved in most highly
motorised countries during the past 3035 years. Yet, no country
is satisfied with its road safety record, and a number of wellknown road safety problems persist. Speeding, for example, was
regarded as a major problem 35 years ago and remains so today.
The high accident rates of young drivers also seem to defy most
attempts at reducing them. Why are some road safety problems
more difficult to solve than others? Which are the characteristics of road safety problems that influence their amenability to
treatment? This paper will try to shed light on these questions

Tel.: +47 22 573800; fax: +47 22 60 92 00.

E-mail address: re@toi.no.

0001-4575/$ see front matter 2008 Elsevier Ltd. All rights reserved.
doi:10.1016/j.aap.2008.01.004

by identifying a number of dimensions of road safety problems

that may influence the prospects of finding effective solutions to
them.
One of the objectives of the paper is to show that road safety
problems are multidimensional and may therefore be viewed
from different perspectives, emphasising different dimensions.
Depending on which dimensions are focused, different measures
may be seen as important for solving the problems. Another
objective of the paper is to show how the various dimensions
that characterise road safety problems can be measured numerically. Numerical measurement is useful, because then different
problems can be compared in terms of more than one dimension.
This may facilitate the selection of problems for treatment, as
some dimensions may be less amenable to treatment than others. The paper does not focus on statistical analysis and does not
seek to develop new statistical methods, but illustrates how the

R. Elvik / Accident Analysis and Prevention 40 (2008) 12001210

various dimensions of road safety problems can be expressed

numerically. To show how a given road safety problem can be
analysed in terms of the various dimensions speed and speeding
is used as an example.
2. Dimensions of road safety problems and their
measurement
It is obvious that there does not exist any right way of defining a road safety problem. A reasonable definition is that a road
safety problem is any factor that contributes to the occurrence of
accidents or the severity of injuries. The term road safety problem is broadly synonymous with the term risk factor, interpreted
in a wide sense that includes traffic volume as a risk factor.
According to this definition, a road safety problem may exist
even if it is not recognised. Before the recent surge in research
concerning driver fatigue (Sagberg and Bjrnskau, 2004), this
was not recognised as a major problem. Moreover, even if
research has shown that a certain factor, such as speeding, contributes importantly to accidents, road users may not see it as
a problem and may not want action to be taken against it. One
should therefore distinguish between the statistical analysis of
road safety problems and the perception of such problems. This
distinction is particularly relevant as far as the prospects of solving road safety problems are concerned: Unless a problem is seen
as a problem, it is not likely to be solved. This paper focuses
on the statistical analysis of problems, not their perception. The
following nine dimensions of road safety problems are proposed.
1.
2.
3.
4.
5.
6.
7.
8.
9.

Magnitude
Severity
Externality
Inequity
Complexity
Spatial dispersion
Temporal stability
Perceived urgency
Amenability to treatment

All these dimensions are to some extent amenable to numerical measurement and comparative analyses. Short definitions of
each dimension and proposals for their measurement are given
in Table 1. Key points are elaborated below.

1201

risk can be estimated by this formula:

Population attributable risk =

PE(RR 1)
(PE(RR 1)) + 1

(1)

PE is the proportion of exposure in the presence of the risk factor,

RR is the relative risk associated with it. Sources of data for
estimating exposure and relative risk could be travel behaviour
surveys, traffic counts, questionnaires and accident statistics. As
an example, suppose that 10% of travel is exposed to a risk factor
A that involves a relative risk of accident involvement of 3, using
a reference value of 1 for the 90% of travel not exposed to the
risk factor. The population risk attributable to this risk factor is:
Risk attributable to factor A =

0.1(3 1)
= 0.167
(0.1(3 1)) + 1

This means that if the excessive risk is eliminated, the amount

of travel remaining unchanged, the number of accidents could
be reduced by 16.7%.
Estimating population attributable risk is, in principle,
straightforward when the risk factor is categorical. When the risk
factor is continuous, like traffic volume, the complexity of the
traffic environment, speed, blood alcohol concentration, headway to vehicles ahead, etc., the estimation of attributable risk
must rely on a functional relationship between the risk factor
and road safety. Consider, as an example, speed. Suppose one
wants to estimate the risk attributable to speeding. To do this,
one should first develop a speed distribution for two cases: (1)
actual driving speeds and (2) driving speeds consistent with perfect compliance with speed limits, i.e. no speeding. Fig. 1 shows
distributions of speed for these two cases for national roads in
Norway with a speed limit of 80 km/h.
The curve to the right is the current speed distribution; the
curve to the left is the speed distribution representing no speeding. It is seen that the elimination of speeding results in a
reduction of the mean speed of travel from 78.5 to 74.3 km/h.
The risk attributable to speeding can be estimated by applying
the Power model of the relationship between speed and road
safety (Elvik, 2005; Elvik et al., 2004). This model describes
the relationship between speed and road safety in terms of a set
of power functions. Thus, for fatalities on roads with a speed

2.1. Magnitude
The magnitude of a road safety problem is the size of the contribution it makes to accidents. The contribution a certain risk
factor makes to accidents is indicated by the risk attributable to it.
In epidemiology (Rothman and Greenland, 1998), several measures of attributable risk have been developed. For the purpose
of comparing the magnitude of different road safety problems,
population attributable risk is perhaps the best indicator. The
term population should not be taken literally; it refers to the
contribution a factor makes to the total number of cases (i.e. all
accidents) rather than a subset of cases. Population attributable

Fig. 1. Determining the risk attributable to speeding by way of shifting the speed
distribution.

1202

R. Elvik / Accident Analysis and Prevention 40 (2008) 12001210

Table 1
Definitions of dimensions of road safety problems and numerical indicators of the problems
Dimension of problem

Definition

Numerical indicator

1. Magnitude

The size of the contribution a problem makes to the total

number of accidents or killed or injured road users
The gradient of the attributable risk associated with a
problem with respect to levels of injury severity

Population attributable risk represented by a problem (for

numerical example, see text)
Comparison of the attributable risk associated with a
problem across levels of injury severity; comparing shares
of injuries by levels of severity
The net contribution to the overall risk of a road user group
attributable to risks imposed by other groups
Difference in number of injuries between: (a) current
distribution of risk and (b) a distribution of risk in
proportion to shares of exposure
Complexity is indicated by: (a) many factors each making a
small contribution to overall risk; (b) correlations among
factors preventing their unique contributions from being
identified; (c) interactions among risk factors with respect
to their effect on accident rate (d) over-determination, i.e.
the sum of known factors more than explain overall risk
Distance between accident concentrations that have been
identified statistically
A time-series of the population attributable risk represented
by a problem
Results of opinion polls regarding the support for stronger
action to solve specific road safety problems
A function of complexity, perceived urgency and
knowledge of effective treatments

2. Severity

3. Externality
4. Inequity

5. Complexity

6. Spatial dispersion
7. Temporal stability
8. Perceived urgency
9. Amenability to treatment

The fact that travel performed by one group of road users

imposes an additional risk on other groups of road users
The size of the contribution to risk made by a lack of
proportionality between the benefits of transport and the
risk run
The extent to which the specific contributions of individual
risk factors to the overall risk represented by a problem can
be identified

The degree to which an accident problem is concentrated

geographically
Changes over time with respect to the magnitude of a road
safety problem
The strength of the support in the population for stronger
action or regulations designed to solve a problem
The prospect of implementing effective safety treatments,
i.e. treatments that will reduce a problem (in particular its
magnitude)

limit of 80 km/h:

Fatality risk attributable to speeding = 1

74.3
78.5

4.5

= 1 0.781 = 0.219
If speeding was eliminated, the number of fatalities could be
reduced by about 22%. This is the risk attributable to speeding. The 95% confidence interval for the exponent is (4.1, 4.9).
To estimate the contribution of speeding to serious and slight
injuries, the exponent should be 3.0 (2.2, 3.8) and 1.5 (1.0, 2.0),
respectively, leading to estimates of 0.152 for serious injuries
and 0.079 for slight injuries.
2.2. Severity
It is widely agreed that fatalities are the most serious impact
of road safety problems, serious injuries the second most serious impact and pure property damage the least serious impact of
road safety problems. Hence, a road safety problem is severe if
it makes a greater contribution to fatalities and serious injuries
than the contribution it makes to slight injuries or property
damage.
There are many ways to assess the severity of a road safety
problem. An approach which employs the notion of attributable
risk is to compare estimates of attributable risk across levels
of injury severity. Thus, the following estimates of the risks
attributable to speeding have been developed for Norway for the
years 2004 and 2005, based on data about speed and injuries for
those years:

Fatalities attributable to speeding:

Serious injuries attributable to speeding:
Slight injuries attributable to speeding:

0.239
0.173
0.093

If the risks attributable to speeding did not have a severity

gradient, they would be 0.093 at all levels of injury severity.
For fatalities, therefore, 61% of the risk attributable to speeding
[(0.2390.093)/0.239] can be assigned to the contribution of the
severity gradient, 39% to the contribution of speeding to the
occurrence of injuries, irrespective of severity.
2.3. Externality
An externality can be defined as an activity performed by one
actor that has impacts on the welfare of another actor and where
the actor producing the impact does not consider it in decisions
about the activity (Baumol and Oates, 1988, 17). In the context
of road traffic, an externality in this sense exists when travel
performed by one group of road users imposes an additional
risk on another group of road users, i.e. a risk that would not
exist if the travel giving rise to it did not take place. To identify
the external risk produced by a certain group of road users, it
is instructive to form a table of injured road users according to
the combination of parties involved in accidents. Table 2 shows
these combinations for Norway. In case an accident involved
more than two parties, it has been assigned to heaviest party
involved.
Truck-trailer combinations (a truck pulling one or more trailers or semi-trailers) were classified as the heaviest type of
vehicle. Next follow (single) trucks, buses, vans and cars. The
first row of the table shows the number of persons who were
injured as occupants of truck-trailers. The first column of the

R. Elvik / Accident Analysis and Prevention 40 (2008) 12001210

1203

Table 2
Injured road users in police reported accidents in Norway 19982005 by combination of groups involved
Injured as occupant of

Counterpart in accident
Truck-trailer

Truck

Bus

Van

Car

Large MC

Truck-trailer
Truck
Bus
Van
Car
Large MC
Small MC
Moped
Cycle
Pedestrian
Other

73
80
120
115
1789
41
10
23
42
54
17

32
102
103
214
2736
84
14
68
144
220
54

10
37
63
80
1210
46
9
47
105
318
14

5
40
43
271
2815
128
47
150
286
409
20

96
197
290
1,038
31,355
1,926
474
2,350
4,388
5,635
158

0
3
1
3
203
107
4
17
42
61
5

Total

2364

3771

1939

4214

47,907

446

table shows the number of persons who were injured in accidents in which the counterpart was a truck-trailer. Seventy-three
truck-trailer occupants were injured in accidents in which the
other party involved was also a truck-trailer. Ninety-six truck
occupants were injured in accidents in which the truck-trailer
crashed with a car. This can be interpreted as a risk imposed
by cars on truck-trailer occupants. The first column of Table 2
shows that, for example, 1789 car occupants were injured in accidents in which the other party was a truck-trailer. In total, 2364
road users were injured in accidents in which truck-trailers were
involved, of which 2291 were not occupants of the truck-trailer.
Hence, the external risk imposed by truck-trailers accounts for
2291 injured road users, whereas the risk other road users impose
on occupants of truck-trailers (first row of Table 2) accounts
for only 149 injured occupants. The involvement of a trucktrailer in an accident therefore represents a large net external
risk: the number of injuries to other groups of road users in accidents involving truck-trailers is much greater than the number
of injuries to truck-trailer occupants. The ratio of the number
of injuries to other groups of road users (2291) to the number of injuries to truck-trailer occupants (149) may serve as a
numerical index of the external risk created by truck-trailers
(2291/149 = 15.38, i.e. 15 times as many injuries to others as the
injuries others impose on truck-trailer occupants).
By definition, the sum of the risks imposed by one group
of road users upon other groups and the risks inflicted on a
certain group of road users by all other groups is zero. That is,
risks exported (external risks) equal risks imported (inflicted
by other groups). Risks that only involve one group of road
users, such as injuries to truck-trailer occupants in accidents that
involve only truck-trailers (either collisions or single-vehicle
accidents) can be regarded as internal. Table 3 shows the net
externality contributed by various groups of road users at various levels of accident injury severity. Table 3 was derived from
a set of tables similar to Table 2. It is seen that truck-trailers and
other heavy vehicles (trucks, buses, vans) represent a significant
external risk and that there is a distinct severity gradient to the
external risk.
Perhaps somewhat surprisingly, the number of external
injuries is higher at lower speed limits than at higher speed lim-

Small MC

Moped

Cycle

Pedestrian

1
0
0
0
31
5
14
26
12
13
0

0
2
7
4
59
26
21
139
82
147
2

2
2
3
3
59
25
6
36
254
167
4

0
1
8
8
78
25
5
51
58
37
2

102

489

561

273

Other

None

Total

3
22
36
35
543
43
10
46
83
188
31

533
481
627
954
19,859
2,216
340
1,036
644
185
414

755
967
1,301
2,725
60,737
4,672
954
3,989
6,140
7,434
721

1040

27,289

90,395

its. This tendency can probably be explained by the fact that

there are more unprotected road users, such as pedestrians and
cyclists, on urban roads that have a low speed limit than on
rural roads having higher speed limits. The severity of external
injuries increases, however, as speed limit increases. This is evident from the fact that the severity gradient for external injuries
is steeper at high speed limits than at low speed limits.
2.4. Inequity
Risk may vary along many dimensions: different groups of
road users face different levels of risk; risk may vary between
regions of a country or between the rich and the poor. Considerations of equity with respect to the distribution of risk are relevant
for all these dimensions. While there is no standard definition
of fairness with respect to the distribution of risk, this paper
will approach the concept by relying on John Rawls theory of
justice as fairness (Rawls, 1971, 2001). He suggests that the
distribution of primary goods in a society is just if it is egalitarian. Unequal distributions of primary goods, he argues, can only
be justified if the departure from equality benefits everybody in
society, including those who get the smallest share of goods that
are unequally distributed. He refers to this principle of justice as
the difference principle.
Application of the difference principle to transport risk
requires a definition of what it means to be advantaged or disadvantaged. The advantage provided by a transport system is
the opportunity to travel (or transport goods). As far as personal
travel is concerned, the most advantaged group is therefore the
group that performs the largest amount of travel. This group
can be regarded as the most advantaged by making the greatest use of a transport system, which serves several groups of
road users. The least advantaged group is the one that makes
the least use of the system, i.e. performs the smallest number of
person-kilometres of travel.
Differences in risk are arranged to the benefit of the least
advantaged if that group has the lowest level of risk, and the
most advantaged group has the highest level of risk. To determine of this is the case today, estimates of fatality and injury
risk for various groups of road users in Norway were com-

1204

R. Elvik / Accident Analysis and Prevention 40 (2008) 12001210

Table 3
Net external injuries imposed by various groups of road users in Norway (19982005) by speed limit and injury severity
Ratio of injuries imposed on other groups of road users (external) to
injuries imposed by other groups of road users (imposed) by speed limit,
injury severity and group of road user
Speed limit of 3050 km/h

Truck-trailer
Truck
Bus
Van
Car
Large MC
Small MC
Moped
Bicycle
Pedestrian
Other
Total

Speed limit of 6070 km/h

Speed limit of 80100 km/h

Slight

Serious

Fatal

Slight

Serious

Fatal

Slight

Serious

Fatal

29.57
13.44
3.14
3.18
3.05
0.12
0.14
0.11
0.05
0.04
4.38

49.00
28.50
8.33
7.09
7.03
0.15
0.09
0.05
0.04
0.01
10.00

19.00
32.00
23.00
8.00
5.29
0.05
0.00
0.11
0.11
0.03
7.50

16.00
9.89
1.53
2.48
0.77
0.20
0.11
0.06
0.05
0.08
3.71

20.71
19.00
5.67
2.63
1.44
0.03
0.00
0.11
0.07
0.01
8.50

32.33
13.00
35.00
2.86
0.76
0.04
0.17
0.00
0.00
0.00
1.25

9.66
6.14
2.16
1.64
0.45
0.23
0.19
0.17
0.14
0.17
3.67

27.46
6.81
3.74
2.00
0.63
0.05
0.29
0.13
0.08
0.01
3.50

63.80
23.40
12.71
3.74
0.28
0.02
0.20
0.00
0.00
0.00
4.00

1.00

1.00

1.00

1.00

1.00

1.00

1.00

1.00

1.00

pared. The amount of travel (the advantage) was measured in

terms of person-kilometres of travel, estimated on the basis of
the national travel behaviour survey (Denstadli and Hjorthol,
2002; Denstadli et al., 2005). Risks were estimated on the basis
of official accident statistics for the period 19982005. Table 4
shows risks of fatality or injury for different groups of road users
in Norway. The groups have been arranged according to their
shares of travel. As can be seen from Table 4, current rates of
fatality or injury do not satisfy the difference principle, in the
sense that the least advantaged groups those representing the
smallest shares of travel enjoy the lowest risks. Hypothetical
levels of risk that would be minimally consistent with the difference principle (although not strictly proportional to the amount
of travel) have been entered in Table 4.
The disparity in risk between motor vehicle occupants and
pedestrians, in particular, is unfair since pedestrian risk is almost
entirely imposed by motor vehicles. Although it is not realistic
to reduce pedestrian risk to the same level as that of, for example,
bus occupants, it is nevertheless of interest to try to find out how
much of pedestrian risk can be attributed to differences between
different types of motor vehicles in the probability of hitting a
pedestrian, the mass of vehicles striking them, and speed. Table 5
explores these effects.
Motor vehicles have been categorised as heavy (truck-trailer,
truck, bus), light (van, car) or two-wheelers (including bicycles). Impact speed is not known, but speed limit can be used
as a proxy for it. Cases in which the speed limit was unknown
have been omitted. Light vehicles have the lowest rate of striking
pedestrians (2.38/100 million vehicle km). If the rate for heavy
vehicles and two-wheelers was reduced to the same level as
that of light vehicles, the number of pedestrian injuries would
be reduced from 3750 to 3610. The effects of mass on the
severity of injuries were assessed by comparing between different types of vehicles the percentage of pedestrians who were
killed or seriously injured at the lowest speed limit (3050 km/h).
Thus, for example, 2.2% of pedestrians struck by two-wheelers
at a speed limit of 3050 km/h were killed. The correspond-

ing percentage for heavy vehicles was 12.5%. The fatality risk
attributable to the greater mass of heavy vehicles is therefore
(12.52.2)/12.5 = 0.824. The effects of speed were determined
analogously by comparing the proportion of fatalities or
serious injuries between speed limits, keeping vehicle mass
constant.
If injury outcomes in all pedestrian crashes were identical to
those in which pedestrians are struck by two-wheelers at a speed
limit of 3050 km/h, the number of killed pedestrians could be
reduced from 196 to 93. Eliminating the contribution of large
mass would account for a reduction of 41 fatalities, eliminating
unsafe speeds would account for a reduction of 62 fatalities. A
further reduction from 93 to 88 fatalities would be possible if
all vehicles had the same rate of striking pedestrians as light
vehicles (cars and vans). While pedestrians would still have a
higher fatality rate than car occupants, the difference would be
greatly reduced, all else equal.
2.5. Complexity
The complexity of a road safety problem depends on whether
it is attributable to a single risk factor or to a multiplicity of risk
factors that each makes a small contribution and interact in ways
that are poorly understood and difficult to measure.
The more fragmented the contributions of a set of risk factors
is, the more complex is a problem, since one then has to know the
effects of many risk factors to fully account for the problem, and
since interventions targeted at a specific risk factor are likely to
make only a small contribution towards reducing risk. To show
how to assess complexity, an example based on an analysis of
the relationship between speed and road safety (Taylor et al.,
2002) will be used. The final version of the model fitted was of
the form:
Accidents = 3.152 107 AADT0.7282 L1.039 V 2.431 Gi
exp(0.1213DS+0.2865DX)

0.14
0.06
0.03
0.03
0.03
0.03
0.03
0.03
0.03
0.03
0.03
1.30
0.57
0.20
0.15
0.12
0.12
0.10
0.10
0.08
0.07
0.05
3.61
1.43
0.66
0.60
0.50
0.45
0.40
0.40
0.30
0.25
0.20
Car
Van
Bus
Pedestrian
Truck
Truck-trailer
Large MC
Bicycle
Moped
Other
Small MC

76.4
8.3
7.0
2.3
1.5
1.2
1.0
1.0
0.6
0.5
0.1

3.61
1.43
0.66
25.61
4.29
6.76
44.32
19.23
15.57
12.23
47.97

1.30
0.57
0.20
8.79
1.39
1.15
17.52
11.18
13.22
4.62
19.04

0.14
0.06
0.03
0.51
0.11
0.10
0.66
1.03
1.07
0.22
1.15

Serious injury rates

consistent with
difference principle
Fatality rates
consistent with
difference principle
Slight injury rate per
million km of travel
Serious injuries per
100 million km of
travel
Fatalities per billion
km of travel

Current fatality and injury

rates (19982005)
Share of travel (%)

Table 4
Observed fatality and injury rates in Norway and hypothetical rates consistent with the difference principle proposed by John Rawls

Fatality and injury rates consistent

with the difference principle of
justice proposed by John Rawls

Slight injury rates

consistent with
difference principle

R. Elvik / Accident Analysis and Prevention 40 (2008) 12001210

1205

Here, AADT is annual average daily traffic, L is the length of

a road section in kilometres, V is the mean speed of traffic in
miles/h, G is a dummy for class of road (four classes were used),
exp is the exponential function, DS is the number of sharp curves
per kilometre of road, and DX is the number of four leg junctions
per kilometre of road. By applying this model equation, it is possible to compare the size of the risks attributable to the various
factors by varying them one at a time, keeping the other factors
in the model equation constant. Table 6 shows an example of
such a use the model.
The effects of increasing each factor by 10% are very different. It is seen that speed has by far the largest effect. In fact,
by reducing mean speed from 44.2 to 34.8 miles/h (a reduction of 21%), roads in category 1 would become as safe as
roads in category 2. AADT would have to be reduced by 55%
to obtain the same effect. Since speed has a dominant influence on the number of accidents, it represents a case of low
complexity.

2.6. Spatial dispersion

Some road safety problems have a distinct geographic
dimension, some do not. It is typically problems related to
the quality of infrastructure roads, traffic control devices,
etc. that have a geographic dimension. The spatial dispersion or concentration of road safety problems on a road
system can be determined by conducting a network screening
(Harwood et al., 2002). The objective of network screening is to locate as accurately as possible those places or
parts of the road system that have the highest expected number of accidents or the highest incidence of fatal or severe
injuries.
Fig. 2 shows the expected number of injury accidents per
kilometre of road, estimated by means of the empirical Bayes
method (combining estimates based on an accident prediction
model and the recorded number of accidents for each kilometre),
for a period of eight years for a major national road in Norway.
The mean expected number of accidents per kilometre was 4.68.
It is seen that three sections of road, all marked as grey in Fig. 2,
had a higher expected number of accidents. The profiles-andpeaks routine of SafetyAnalyst was applied to the three road
sections, using window lengths of 3, 4 or 6 km (Harwood et
al., 2002). Two of the sections were identified as hazardous,
employing a critical value of 0.10 for the coefficient of variation
for the EB-estimates (the coefficient of variation is the standard
error divided by the mean). These sections of road are, however,
not located next to each other.
The two sections identified as hazardous are located at opposite ends of the road. These sections make up 16% the length
of the road, but had nearly 32% of the recorded number of
accidents. Unfortunately, it was not possible to assess the relationship between speed and the expected number of accidents.
Speed data were only available for km 13, 21, 23, 33, 37 and
51. These data are too sparse to estimate a speed profile for the
entire road.

1206

Table 5
Changes in the number of pedestrian injuries if all pedestrian were struck by vehicles of low mass, at low speed limits and only by vehicles that have the lowest rate of striking pedestrians per km of travel
Pedestrian injuries by speed limit

Heavy vehicle

Fatal
Serious
Slight
Total
N

Light vehicle

Fatal
Serious
Slight
Total
N

Two-wheeler

Fatal
Serious
Slight
Total
N

6070 km/h

80100 km/h

Total

Current number

0.125
0.130
0.745

0.105
0.289
0.605

0.345
0.276
0.379

0.145
0.166
0.689

41
47
195

1.000
216

1.000
38

1.000
29

1.000
283

283

0.029
0.130
0.841

0.083
0.234
0.683

0.180
0.265
0.555

0.046 152
0.153 506
0.801 2647

1.000
2646

1.000
448

1.000
211

1.000 3305
3305

0.022
0.124
0.854

0.000
0.385
0.615

0.000
0.250
0.750

0.019
0.154
0.827

3
25
134

1.000
137

1.000
13

1.000
12

1.000
162

162

Mass
22
1
23

Speed
6
10
16

Net effects of mass

and speed

Effects of changes
in accident rate

Final expected
number

13
36
234

5
14
90

8
22
144

283

19
17
36

56
75
131

77
414
2814

174

0
0
0

3305

0
0
0

0
5
5

3
20
139
162

77
414
2814
3305

1
4
16

2
16
113
131

R. Elvik / Accident Analysis and Prevention 40 (2008) 12001210

3050 km/h

Changes in the number of pedestrian injuries if (a) all pedestrian were

struck by the smallest vehicles (mass); (b) at the lowest speed limit and (c)
only by vehicles that have the lowest rate of striking pedestrians

R. Elvik / Accident Analysis and Prevention 40 (2008) 12001210

1207

Table 6
Effects of various variables on the number of accidents
Explanatory variables
Constant term
AADT
Length (km)
Speed (miles/h)
Category 1
Category 2
Category 3
Category 4
Sharp curves per km
Four leg junctions per km

Mean value

5990
3
44.2
1
1
1
0.50
0.14

Coefficient
3.15E-07
0.7282
1.039
2.431
1
0.558
0.391
0.285
0.1213
0.2865

Percentage change in accidents by adding

10% to value of explanatory variable
7.2
10.4
26.0

Change needed no make

road as safe as category 2
55%
21%

0.6
0.4

Derived from Taylor et al. (2002).

2.7. Temporal stability

The stability of a road safety problem over time can be examined in many ways. Tests have been developed to determine if
there is a trend in a series of accident counts, or if there has been
a sudden jump in the series (Hauer, 1996a,b). At the national
level, a simple indicator of stability is changes over time with
respect to the population attributable risk represented by a road
safety problem.
Employing speed data presented by Sakshaug (1986), and
accident data from the official accident record for Norway,
the risks attributable to speeding were estimated for the years
19801984 and 20042005. The population attributable risk was
estimated to 0.257 for fatalities in 19801984 and 0.239 for
20042005. The corresponding attributable risks were, respectively, 0.184 and 0.173 for serious injuries and 0.098 and 0.093
for slight injuries. Speeding does therefore not seem to have
become a greater problem over time in Norway.
2.8. Perceived urgency
One of the factors contributing to a road safety problem
may be a perception that it is not a problem. A problem

which is not seen as a problem may be less amenable to

treatment than a problem that everybody sees as a problem,
and for which stronger action to reduce the problem is supported.
The perceived urgency of road safety problems can be studied
in many ways. One useful indicator is the level of support for
stronger policy interventions. Table 7 shows the most recent
findings of a Norwegian survey regarding this (Nordbakke and
Fyhri, 2005).
Huge majorities of the Norwegian public favour a law
requiring cycle helmets to be worn, pedestrians to wear a
reflective device in the dark, imprisonment to be used more
often to punish drinking and driving and a speed limit of
30 km/h in residential areas. When it comes to policy interventions that deal more generally with speed or speeding,
opinions are more divided. A majority are opposed to reducing speed limits on most roads in towns to 30 km/h. There
is also little support for requiring cars to have a device that
will make speeding unpleasant (for example by means of an
active accelerator pedal giving resistance when the driver tries
to exceed the speed limit, or by means of a warning signal that gives an unpleasant noise when the speed limit is
exceeded).

Fig. 2. Expected number of injury accidents per kilometre of road for a national road in Norway. High-risk sections marked in grey.

1208

R. Elvik / Accident Analysis and Prevention 40 (2008) 12001210

Table 7
Attitudes to road safety policy interventions in Norway
Policy intervention

Mandatory helmet wearing law for cyclists

Mandatory use of pedestrian reflective device in darkness
Imprisonment for drink-driving
Speed limit of 30 km/h in residential areas
Speed limit of 30 km/h in towns in general (currently 50 km/h)
Reducing speed limits in general to improve safety
Higher tickets for speeding
Device in cars making speeding unpleasant (ISA-systems)

Percentage distribution of opinions (N = 2000)

Support

Oppose

No opinion

Total

79
90
78
73
40
46
41
32

20
10
20
25
60
52
54
66

1
0
2
2
0
2
5
2

100
100
100
100
100
100
100
100

Source: Nordbakke and Fyhri (2005).

2.9. Amenability to treatment

A road safety problem is not always easy to solve or reduce,
even if it makes a major contribution to accidents or injuries.
Amenability to treatment can be defined as the prospects of
implementing measures that will reduce the size of a road safety
problem, or, at best, eliminate the problem.
One can try to assess the amenability of various problems to
treatment by combining information on the size of these problems with information on the level of support for stronger policy
interventions. Fig. 3 shows such a combination of information
for four road safety problems in Norway. The four problems
are speeding, drinking and driving, pedestrian accidents in the
dark, and cyclist accidents. For each of these problems, the percentage of the public who support stronger policy interventions,
taken from Table 7, is shown on the abscissa. The higher the percentage, the easier one would think it would be to introduce the
road safety measures that can reduce the problem. Also shown
in Fig. 3 is the fatality risk attributable to each problem. This is
an indication of the importance of the problem.
As can be seen from Fig. 3, measures designed to reduce
drinking and driving, pedestrian accidents in the dark and cyclist
accidents enjoy wide support. Measures to curb speeding, which
is the most important of the problems shown in Fig. 3, are less
supported. This suggests that speeding may be less amenable to
treatment than the other three problems listed in Fig. 3.

Fig. 3. Relationship between support for stronger policy interventions and risk
attributable to various road safety problems in Norway.

Amenability to treatment depends not just on public support

for certain treatments, but also on the effectiveness of these treatments and on their cost-effectiveness. An index of amenability to
treatment could perhaps be formed by treating three conditions
as necessary (but not necessarily sufficient) for the introduction of a treatment: public support (which may be defined as
support by a majority), known effectiveness and favourable
cost-effectiveness. Thus, for speeding, ISA-systems may reasonably be expected to be both effective and cost-effective
(Elvik, 2007), but lack public support. Hence, the index takes the
value 2/3 (two conditions out three fulfilled). Imprisonment for
drink-driving is widely supported, but neither effective nor costeffective (Ross and Klette, 1995). It gets an amenability score of
1/3.
3. Discussion
Road safety problems come in all shapes and sizes. To help
identify the most important road safety problems, a standardised
description of them is useful. To help develop such a standardised
description, this paper has identified some dimensions of road
safety problems and briefly indicated how these dimensions can
be measured.
Selecting targets for safety interventions requires a systematic
analysis of road safety problems. The importance of the dimensions of road safety problems discussed in this paper depends in
part on policy objectives. If the prevention of fatal and serious
injuries is regarded as more important than the prevention of
slight injuries, severity becomes an important dimension, possibly also externality. If seeking the largest overall reduction of
traffic injury is an overriding target, the magnitude of a problem
is its most important dimension.
The nine dimensions of road safety problems that have been
identified in this paper are not exhaustive. A tenth dimension
is the ease of observing a problem. A problem that cannot be
observed, or only observed with great difficulty, may be overlooked or treated as unimportant, even if in fact it could be of
great importance. Inattention, distractions and fatigue are probably all quite important road safety problems, in the sense that
these factors make a major contribution to accidents. Observing
them is, on the other hand, quite difficult, although unobtrusive
monitoring devices are now available that can, in principle, help
collect data on the incidence of inattention and fatigue.

R. Elvik / Accident Analysis and Prevention 40 (2008) 12001210

There are a number of issues in the analysis of road safety

problems that have not been dealt with in this paper. This
includes, first of all, the potential of bias caused by incomplete and inaccurate accident reporting. Bicycle accidents, in
particular, are very incompletely reported in official accident
statistics (Elvik and Mysen, 1999). This problem will be underestimated in any analysis based on official accident statistics.
Another problem of official road accident statistics is that it
often employs a very crude scale for injury severity. Any analysis of the dimension of severity will become equally crude and
uninformative. Long-term impacts of traffic injury are nowhere
recorded in official statistics and can, at present, not be meaningfully addressed in analyses trying to quantify the dimensions of
road safety problems. The analysis of the severity dimension is
also hampered by the frequency-severity indeterminacy created
by incomplete accident reporting (Hauer, 2006). The indeterminacy arises because it is never possible to determine whether an
apparent overrepresentation of serious injuries is real or due to
the fact that serious injuries are more likely to be reported than
slight injuries.
A second issue concerns choice of level of analysis. Road
safety problems can be analysed at the international level (Peden
et al., 2004), the national level, the regional level or the local
level. The framework described in this paper is perhaps best
suited to an analysis at the national level of government, designed
to help in developing a national road safety programme. At the
local level of government, it may not be feasible to assess all
dimensions of road safety problems, or some of the dimensions
may be regarded as irrelevant at this level.
A third issue not discussed exhaustively in this paper is the
treatment of correlations between different road safety problems.
Roads in urban areas often have a higher injury accident rate than
roads in rural areas. In part, this is the result of a larger number of unprotected road users in urban areas than in rural areas.
Accident rate is higher in darkness than in daylight, but so is the
incidence of drinking and driving. Correlations between problems imply that estimates of population attributable risk could
be biased. In the limit, if two problems are perfectly correlated,
treating them as two distinct problems will double count their
contribution to accidents.
A fourth issue, about which very little is known, concerns
interactions between risk factors. This has to do with how the
effects of several risk factors combine. Do the effects of several
risk factors combine multiplicatively, additively, sub-additively,
exponentially, or not at all according to any known mathematical
function? This aspect is related to the dimension of complexity
a dimension that calls for considerably more research before
it can be adequately analysed.
Despite these problems, it is instructive to identify dimensions of road safety problems to help assess in a systematic
way the prospects of solving these problems. The externality
problem, for example, is not easily solved, as it is closely connected to incompatibilities among road users in terms of kinetic
energy and protective devices. So, one may ask, what is the point
of assessing the externality dimension, if nothing can be done
about it? First, it is not clear that nothing at all can be done about
it. It is a difficult target for intervention, but not impossible to

1209

influence at all. Second, by assessing externality, one may guard

against setting policy objectives that are completely unrealistic,
such as reducing the injury risk of a motorcycle rider to the same
level as that of a bus occupant.
4. Conclusions
The following conclusions can be drawn on the basis of the
exploratory analyses presented in this paper:
1. Several dimensions of road safety problems can be recognised. By better understanding the multidimensional nature
of such problems, it is possible to examine more systematically the implications of treatments aiming to solve particular
problems.
2. All dimensions discussed in this paper can to some extent be
measured numerically. This permits profiles of problems
to be developed enabling a more systematic comparison of
problems.
3. There is a need for further developing numerical measures
of the dimensions of road safety problems. This applies
in particular to the dimensions of severity, complexity and
amenability to treatment.
References
Baumol, W.J., Oates, W.E., 1988. The Theory of Environmental Policy, second
ed. Cambridge University Press, New York.
Denstadli, J.M., Engebretsen, ., Hjorthol, R., Vagane, L., 2005. Den
nasjonale reisevaneunderskelsen 2005 nkkelrapport. Rapport 844.
Transportkonomisk institutt, Oslo, RVU 2005.
Denstadli, J.M., Hjorthol, R., 2002. Den nasjonale reisevaneunderskelsen.
Nkkelrapport. Rapport 588. Transportkonomisk institutt, Oslo, RVU
2001.
Elvik, R., 2005. Speed and road safety. Synthesis of evidence from evaluation
studies. Transport. Res. Rec. 1908, 5969.
Elvik, R., 2007. Prospects for improving road safety in Norway. Report 897.
Institute of Transport Economics, Oslo.
Elvik, R., Christensen P., Amundsen, A., 2004. Speed and road accidents. Evaluation of the Power Model. Report 740. Institute of Transport Economics,
Oslo.
Elvik, R., Mysen, A.B., 1999. Incomplete accident reporting. Meta-analysis of
studies reported in 13 countries. Transport. Res. Rec. 1665, 133140.
Harwood, D.W., Torbic, D.J., Bauer, K.M., Persaud, B., Lyon, C., Hauer, E.,
2002. SafetyAnalyst: Software tools for safety management of specific
highway sites. White paper for module 1 Network Screening. US Department of Transportation, Federal Highway Administration, Washington
DC.
Hauer, E., 1996a. Statistical test of difference between expected accident frequencies. Transport. Res. Rec. 1542, 2429.
Hauer, E., 1996b. Detection of safety deterioration in a series of accident counts.
Transport. Res. Rec. 1542, 3843.
Hauer, E., 2006. The frequency-severity indeterminacy. Accid. Anal. Prev. 38,
7883.
Nordbakke, S., Fyhri, A., 2005. Trafikantenes kunnskaper om og holdninger til
trafikksikkerhet 2004. Rapport 3/2005, Veg- og trafikkavdelingen. Statens
vegvesen, Vegdirektoratet, Oslo.
Peden, M., et al., 2004. World Report on Road Traffic Injury Prevention. World
Health Organization, Geneva.
Rawls, J., 1971. A Theory of Justice. Harvard University Press, Boston.
Rawls, J., 2001. Justice as Fairness A Restatement. Cambridge University
Press, New York.

1210

R. Elvik / Accident Analysis and Prevention 40 (2008) 12001210

Ross, H.L., Klette, H., 1995. Abandonment of mandatory jail for impaired drivers
in Norway and Sweden. Accid. Anal. Prev. 27, 151157.
Rothman, K., Greenland, S., 1998. Modern Epidemiology, second ed. Lippincott
Williams and Wilkins, Philadelphia, PA.
Sagberg, F., Bjrnskau, T., 2004. Sovning bak rattet: Medvirkende faktorer,
omfang og konsekvenser. Rapport 728. Transportkonomisk institutt, Oslo.

Sakshaug, K. 1986. Fartsgrenseunderskelsen-85. Detaljerte resultater fra fartsdelen og ulykkesdelen. Notat 535/86 og 536/86. SINTEF, Avdeling for
samferdselsteknikk, Trondheim.
Taylor, M.C., Baruya, A., Kennedy, J.V., 2002. The relationship between speed
and accidents on rural single-carriageway roads. TRL Report TRL511.
Transport Research Laboratory, Crowthorne.