t9? J3}b)
POLICY RESEARCH WORKING PAPER 3 03 5
Traffic Fatalities and Economic Growth
Elizabeth Kopits
Maureen Cropper
The World Bank
Development Research Group
Infrastructure and Environment
April 2003
U POLICY RESEARCH WORKING PAPER 3035
Abstract
Kopits and Cropper examine the impact of income fatalities peaks is approximately $8,600 in 1985
growth on the death rate due to traffic fatalities, as well internationial dollars. This is within the range of income
as on fatalities per motor vehicle and on the at which other externalities, such as air and water
motorization rate (vehicles/population) using panel data pollution, have been found to peak.
from 1963-99 for 88 countries. Specifically, they Projections of future traffic fatalities suggest that the
estimate fixed effects models for fatalities/population, global road death toll will grow by approximately 66
vehicles/population, and fatalities/vehicles and use these percent between 2000 and 2020. This number, however,
models to project traffic fatalities and the stock of motor reflects divergent rates of change in different parts of the
vehicles to 2020. world-a decline in fatalities in high-income countries of
The relationship between motor vehicle fatality rate approximately 28 percent versus an increase in fatalities
and per capita income at first increases with per capita of almost 92 percent in China and 147 percent in India.
income, reaches a peak, and then declines. This is The authors also predict that the fatality rate will rise to
because at low income levels the rate of increase in approximately 2 per 10,000 persons in developing
motor vehicles outpaces the decline in fatalities per countries by 2020, while it will fall to less than 1 per
motor vehicle. At higher income levels, the reverse 10,000 in high-income countries.
occurs. The income level at whlich per capita traffic
This paper-a product of Infrastructure and Environment, Development Research Group-is part of a larger effort in the
group to study the externalities associated with motorization. Copies of the paper are available free from the World Bank,
1818 H Street NW, Washington, DC 20433. Please contact Viktor Soukhanov, room MC2-205, telephone 202-473-5721,
fax 202-522-3230, email address vsoukhanov@worldbank.org. Policy Research Working Papers are also posted on the
Web at http://econ.worldbank.org. The authors may be contacted at kopits@rff.org or mcropper@worldbank.org. April
2003. (42 pages)
The Policy Research Working Paper Senes disseminates the findings of work in progress to encourage the exchange of ideas about
development issues An obhective of the series is to get the findings out quickly, even if the presentations are less than fully polished The
papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this
paper are entirely those of the authors. They do not necessarily represent the view of the World Bank, its Executive Directors, or the
countries they represent.
Produced by the Research Advisory Staff
TRAFFIC FATALITIES AND ECONOMIC GROWTH
Elizabeth Kopits
University of Maryland and Resourcesfor the Future
Maureen Cropper
World Bank
We would like to thank the World Bank Research Board for funding and the Bank's Road Safety
Thematic Group for comments and encouragement. We would especially like to thank Zmarak
Shalizi for his encouragement during early stages of the project. Paulus Guitink and Hua Wang
generously provided data. We would also like to thank the following people for their comments
and suggestions: Amy Aeron-Thomas, Roger Betancourt, Bill Evans, Goff Jacobs, Matthijs
Koornstra, Seth Sanders, Jeff Smith, V. Kerry Smith, and Jagadish Guria for his careful reading
of the paper. We appreciate the comments of seminar participants at Resources for the Future,
the University of Maryland, Camp Resources and the International Conference on Vehicle Safety
and Reliability in Keszthely, Hungary.
Traffic Fatalities and Economic Growth
Elizabeth Kopits and Maureen Cropper
I. Introduction
As countries develop death rates usually fall, especially for diseases that affect the
young and result in substantial life-years lost. Deaths due to traffic accidents are a
notable exception: the growth in motor vehicles that accompanies economic growth
usually brings an increase in road traffic accidents. Indeed, the World Health
Organization has predicted that traffic fatalities will be the sixth leading cause of death
worldwide and the second leading cause of disability-adjusted life-years lost in
developing countries by the year 2020 (Murray and Lopez 1996). Table 1 highlights the
increasing importance of the problem in several developing countries. For example,
between 1975 and 1998, road traffic deaths per capita increased by 44% in Malaysia and
by over 200% in Colombia and Botswana.
The situation in high-income countries is quite different. Over the same period,
traffic fatalities per person decreased by 60% in Canada and Hong Kong, and by amounts
ranging from 25% to 50% in most European countries. This reflects a downward trend in
both the fatality rate (deaths/population) and in fatalities per kilometer traveled that began
in most OECD countries in the early 1970's and has continued to the present.
Table 1. CIsaimge ann Tlrafn IFatifty Rate (DDeaths/10O,O0O IPersonos), 1975-1990
% Chnage 1% Chabnge
Country (675-'90) j Counntry (75 -!2
Canada -63.4% alaysia 44.3%
on Kong -61.7% diaA 79.3%
Finland -59.8% ri Lanka 84.5%
ustria -59.1% sotho 192.8%
Sweden -58.3% olombia 237.1%
[srael -49.7% China 243.0%
Belgtum -43.8% Botswana"t 383.8%
France -42.6%
Italy* -36.7%
New Zealand -33.2%
Taiwan -32.0%
United States -27.2%
apan -24.5%
*%change ('75-'97), *%/ochange ('76-'98), A%chage ('80-'98).
These patterns are not surprising. The traffic fatality rate (fatalities/population) is
the product of vehicles per person (V/P) and fatalities per vehicle (FN). How rapidly
fatality risk grows depends, by definition, on the rate of growth in motorization (V/P) and
the rate of change in fatalities per vehicle (F/V).' In most developing countries over the
past 25 years, vehicle ownership grew more rapidly than fatalities per vehicle fell. The
experience in industrialized countries, however, was the opposite; vehicles per person
grew more slowly than fatalities per vehicle fell. From these observations, two questions
emerge: Why did these patterns occur? and What trends can be expected in the future?
To answer these questions we examine how the death rate (F/P) associated with
traffic accidents and its components-V/P and F/V-change as countries grow. The,
topic is of interest for two reasons. For planning purposes it important to forecast the
growth in traffic fatalities. Equations relating F/P to per capita income can be used to
'The fatality rate may also be expressed as the product of fatalities per vehicle kilometers traveled (F/VKT)
and distance traveled per person (VKT/P). Lack of reliable time-series VKT data, especially. for
developing countries, prevents us from using this measure for our analysis.
2
predict traffic fatalities by region. These forecasts should alert policymakers to what is
likely to happen if measures are not enacted to reduce traffic accidents.
A second motive for our work comes from the literature on Environmental
Kuznets Curves (Grossman and Krueger 1995). This literature examines the relationship
between environmental externalities, such as air and water pollution, and economic
growth. A focus of this literature has been in identifying the income levels at which
externalities begin to decline. Road traffic fatalities are, indeed, an externality associated
with motorization, especially in developing countries where pedestrians comprise a large
share of casualties and motorists are often not insured. It is of interest to examine the
income level at which the traffic fatality rate (F/P) historically has begun to decline and
to compare this with the pattern observed for other externalities.
We investigate these issues by estimating equations for the motor vehicle fatality
rate (F/P), the rate of motorization (V/P) and fatalities per vehicle (F/V) using panel data
from 1963-99 for 88 countries. We estimate fixed effects models in which the natural
logarithm of F/P, V/P and F/V are expressed as (a) a quadratic function of ln(Y) and (b) a
spline function of ln(Y), where Y = real per capita GDP (measured in 1985 international
prices). Time trends during the period 1963-99 are modeled in four ways: (1) a common
linear time trend; (2) a common log-linear time trend; (3) regional, linear time trends; and
(4) regional, log-linear time trends. These models are used to project traffic fatalities and
the stock of motor vehicles to 2020.
Our main results are as follows. The per capita income at which the motor
vehicle fatality rate begins to decline is in the range of incomes at which other
externalities (specifically the common air pollutants) begin to decline-approximately
3
$6100 (1985 international dollars) when a common time trend is assumed for all
countries and $8600 (1985 international dollars) when separate time trends are used for
each geographic region. This turning point is driven by the rate of decline in F/V as
income rises since V/P, while increasing with income at a decreasing rate, never declines
with economic growth.
Projections of future traffic fatalities suggest that the global road death toll will
grow by approximately 66% over the next twenty years. This number, however, reflects
divergent rates of change in different parts of the world: a decline in fatalities in high-
income countries of approximately 28% versus an increase in fatalities of almost 92% in
China and 147% in India. We also predict that the fatality rate will rise to
approximately 2 per 10,000 persons in developing countries by 2020, while it will fall to
less than 1 per 10,000 in high-income countries.
The paper is organized at follows. Section II presents trends in fatality rates
(F/P), motorization rates (VIP), and fatalities per vehicle (Fly) for various countries.
Plots of each variable against per capita income motivate our econometric models.
Section III describes the econometric models estimated and Section IV presents our
projections of road traffic fatalities. Section V concludes.
II. How Fatality Risk, Motorization Rates and Fatalities/Vehicle Vary Across (and
Within) Countries
Death rates (F/P) due to motor vehicle crashes are the product of the motorization
rate (V/P) and fatalities per motor vehicle (FN). Before estimating statistical models
relating these ratios to per capita income it is useful to examine data showing how these
quantities vary with income both within and across countries.
4
It is widely recognized that the motorization rate rises with income (Ingram and
Liu (1999), Dargay and Gately (1999), Button et al. (1993)), implying that one should
find large differences in vehicles per capita across countries at different stages of
development and within countries as per capita incomes grow. Table 2 presents data on
motorization rates for various countries in 1999.2 Figure 1 plots the motorization rates
for these countries against per capita income, pooling data from all countries and years,
while Figure 2 shows how motorization rates have grown with income over time for a
sample of countries.
Table 2. Motorization Rates, 1999, 60 Countries
Vehicles* Vehicles*
Country /1,000 Persons Country /1,000 Persons
HD] Countries: HD2 Countries:
United States 779 Malaysia 451
Luxe ourg 685 Bulgaria 342
Japan 677 Thailand 280
Italy/ 658 Latvia 267
Iceland 629 Mauritius 195
Switzerland 622 Romania 169
Australia3 616 South Africa 144
Austria 612 Panama 112
Canada' 585 Turkey 100
Gen_nany_ 572 Indonesia 81
New Zealand 565 Sri Lanka' 74
Norway 559 Botswana 72
Cyprus 551 Swaziland' 69
Belgium 522 Colombia 67
Spain 499 Benin3 52
Finland 498 Morocco 51
Sweden' 496 Ecuador' 47
Czechoslovakia 440 Philippines' 42
2`The data in Table 2 and Figure I are displayed according to development status. Observations for
countries with a UN Human Development Index (HDI) less than 0.8 are denoted as "HD2" and countries
with an HDI value greater than 0.8 are labeled as "HD ".
5
United Kingdom' 434 Togo3 39
Netherlands 427 Mongolia 38
Denmark 424 Egyptt 335
portugal2 423 India 34
Bahrain 339 Nigeria3 29
Poland 323 Pakistan 23
Ireland3 312 ena3 14
Israel 301 Senegal 14
Korea, Rep. 296 anglad esh 3.1
Hungary 283 Ethiopia 1.5
Singapore 164
Costa Rica 162
Chile 138
Hong Kong 80
*Including passenger cars, buses, trucks, and motorized two-wheelers.
1- 1998 data,2- 1997 data,3- 1996 data.
Figure 1. Motorlzation Rate vs. lincoamie: AIR Counmtries samid Years
8,000 - Ic%5 i�0 0
0 O
o 0 c 0
E 6,000- 0 0
O' 4,000 - E oo a � HD2 Obsevtios
; 2,000 0 00000 0 0 0 0 e 0
0 0 0 00c~~0 0~
I I 0 I o I O srain
0 5,000 10,000 15,000 20,000 25,000
Per Capita GDP, 1985 int'l$
The cross-sectional variation in motorization rates in Table 2 is striking. Vehicles
per capita range from a high of 780 per 1,000 persons in the United States to fewer than
30 per 1,000 persons in countries such as Pakistan and Nigeria. High-income countries
6
tend to have more vehicles per capita than lower income countries but there are important
exceptions. Low motorization rates in Hong Kong, Singapore, Chile, and Costa Rica are
notable outliers. Figure 1, which plots data on V/P for all countries and years in the
dataset, suggests that, overall, motorization is strongly correlated with income. The
within-group variation in motorization varies from country to country, however, as
shown in Figure 2. Growth in vehicle ownership appears to have slowed down (but not
declined) in many high-income countries such as Norway, Australia, Hong Kong, and the
United States. In countries experiencing lower levels of per capita GDP such as Greece,
Malaysia, and Thailand, however, vehicle fleets have continued to expand rapidly with
income in recent decades.
7
Figure 2. Motorization Rates vs. I[ncone: Selected Conuiatries
0 Greece a Korea, Rep. of D United States + Hong Kong
O Norway o Ausahlia Taiwan, Province of China
8000
7000 -
6000 -
5000
d 4000- VP
. 3000 -
2000-
1000 -
0
3,000 6,000 9,000 12,000 15,000 18,000 21,000
Per Capita GDP, 1985 intl$
o Mauritius A Thailand O Turkey
+ Botswana O Malaysia o Sri Lanka
4500 -
4000 -
3500 -
E 3000 -
2500-
2000-
*~1500-
0 -
500 2,000 3,500 5,000 6,500 8,000
Per Capita GDP, 1985 inels
Fatalities per vehicle, by contrast, appear to decline steadily with income, at least
after some low level of income, and then reach a floor. Both Figure 3, which plots the
fatality rate (FlV) against income using data for all countries and years, and Figure 4,
which shows how F/V has changed with income over time for a sample of countries,
3
attest to this fact. In part, the sharp decline in F/V with income reflects the fact that, as
income rises, a higher percentage of travelers are vehicle passengers rather than
pedestrians, and thus, are less likely to die in the event of a crash.4 It also may reflect the
move to safer vehicles (e.g., from two-wheelers to four-wheelers), safer roads, and/or
changing attitudes toward risk as incomes grow.
Figure 3. Fatalities/Vehicle vs. Income: All Countries and Years
250 -
A
200 - 0 HD I Observations
A HD2 Observations
150 o
1100-
50 - .. :. 0 0
0 5,000 10,000 15,000 20,000 25,000
Per Capita GDP, 1985 int'l$
3The fatality figures used in Figures 3 and 4 have not been adjusted for underreporting of road deaths.
Thus, FN levels in developing countries may be underestimated.
4This point was first publicized by Smeed (1949), who demonstrated that F/V declines as V/P increases.
9
igre 4. lFatalities/Vehiclle v. Ilcome: Selected Coununtries
o Grce A United States [ Hong Kong
+ Norway 0 Australia o Taiwan, Provmce of China
40 -
30 -
,20 320-
10 1-A
0
2,000 8,000 14,000 20,000
Per Capita GDP, 1985 inf 1$
o Bangladesh A Egypt 0 Uganda + Ethiopia C Monjcoc
o India Turkey 0 Mauntrus A Mlaysia 0 Komea, Rep. of
250 -
200 -
~ 150
*~100
50
0
0 2,000 4,000 6,000 8,000 10,600
Per Capita GDP, 1985 int?l$
1 0
The foregoing data suggest that one would expect to see the motor vehicle fatality
rate (F/P) first increase and then decrease with income. Figure 5, which plots of F/P
versus income for all years and countries in our dataset, supports this inverted U-shaped
pattern. As incomes grow and vehicle fleets increase during initial stages of
development, traffic fatality rates tend to worsen. At higher income levels, however, as
growth in motorization slows and governments and individuals invest more in road
safety, the decline in F/V drives the fatality rate (F/P) down.
Figure 5. Traffic Fatality Rate vs. Income: All Countries and Years
4 o
o 0 o HDI Observations
&L 0 0 � � HD2 Observations
3 - 0it ADpO O4F 0
% 00,& * 00ccP:r4So
0~~~
a 00 �o
70 0 0
3 q
0 0 00
0 ~~~~~~~~~~~~~'~0 0
5,000 10,000 15,000 20,000 25,000
Per Capita GDP, 1985 int'l$
m. Statistical Models of Fatalities, Vehicle Ownership and Economic Growth
A. Models of Fatalities per Person
In fitting statistical models to the data on fatality risk we employ models of the
general form:
(1) ln(F/P)it = ai + G(t) + FTln(Yit)] + sit
where F/P = Fatalities/10,000 Persons, Y = Real Per Capita GDP (measured in 1985
international prices), as is a country-specific intercept, and G and F are functions. Two
specific forms of F are used, a quadratic specification (equation (2)) and a spline
(piecewise linear) specification (equation (3)):
(2) ln(F/P)it = ai + G(t) + b Inyit + c (lnYi02 + Fit
(3) ln(F/P)it = ai + G(t) + b lnYit + Ys [csDs(lnYit - lnYs)] + Sit
where D. is a dummy variable = 1 if Yit is in income category s+1, and Y. is the cutoff
income value between the s and s+I income group. Following Schmalensee et al. (2000),
we divide the observations into 10 income groups with an equal number of observations
in each spline segment.
Four different forms of G(t) are used: (1) a common linear time trend; (2) a
common log-linear time trend (In t); (3) regional, linear time trends; and (4) regional, log-
linear (In t) time trends. For purposes of defining time trends, we divide countries into
two groups: highly developed countries (HD 1)-i.e., countries that have a Human
Development Index in 1999 of 0.8 or greater-and all other countries (HD2).5 In
practice, this division of countries (shown in detail in the Appendix) corresponds closely
5 The United Nations Human Development Index measures per capita income, life expectancy and
educational achievement.
12
to highly-motorized countries versus other countries. All HD1 countries are treated as a
single region for the purposes of computing time trends. HD2 countries, in turn, are
classified according to region. Table 3 shows the number of countries in each geographic
region, for both HD1 and HD2 countries.
Table 3. Regional Distribution of Countries Used in Model Estimation
WB Region HD2 HD1
East Asia & Pacific 10 1
E. Europe & Central Asia 5 3
Latin America & Caribbean 5 2
Middle East & North Africa 8 1
South Asia 5 _
Sub-Saharan Africa 20
High-Income Countries 28
Total: 53 35
The inclusion of country-specific intercepts in equation (1) implies that the impact
of income on the fatality rate will reflect within- rather than between-country variation in
ln(F/P) and ln(Y). This is desirable for two reasons. For the purposes of predicting
future trends in F/P it is more desirable to rely on within-country experience rather than
on largely cross-sectional variation in income and fatality risk. Using only cross-country
variation to predict the future pattern of traffic fatalities is equivalent to saying that once
Indonesia reaches the income level of Greece, its road safety record will mirror that of
Greece. The second reason is that countries differ in their definition of what constitutes a
traffic death and in the percentage of deaths that are reported. (This topic is discussed
more fully in Section IV.) To the extent that the degree of under-reporting remains
13
constant over time but varies across countries it will not affect estimates of the impact of
economic growth on fatality risk.6
Equations (2) and (3) are estimated using panel data for 88 countries for the
period 1963-99. To be included in the dataset a country must have at least 10 years of
data on traffic fatalities. Table A. 1 in the Appendix lists the countries used to estimate
the models and the number of years of data that are available for each country. The data
on traffic fatalities used in this study come primarily from the International Road
Federation Yearbooks, which have been supplemented by and cross-checked against
various other sources. The sources of traffic fatality and all other data are described in
the Appendix.
B. Empirical Results
Table 4 summarizes the results of estimating the 8 models formed by combining
the quadratic and spline functions with 4 methods of treating time trends. (Complete
regression results are displayed in Table A.2. in the Appendix.) The table shows the
income level at which F/P first begins to decline as well as the time coefficients for each
model. To give a more complete picture of the model results, Figures 6 and 7 plot F/P as
a function of per capita income.7 Figure 6 shows the quadratic and spline models with a
common linear time trend while Figure 7 plots the quadratic and spline models with
region-specific log-linear time trends.
6 To illustrate, when equation (2) is estimated using country-specific intercepts, the coefficients b and c
reflect within-country variation in fatalities and income. Multiplying country i's fatality rate by a constant
to reflect under-reporting would not change the estimates of b and c.
7In both figures, F/P results are displayed using the country intercept for India with the time trend set equal
to 1999.
14
Table 4. Reg ession Results from Fatalities/Population Models
QuSadratic _S_ine
_ _Pot 1 2- 3 4 5 6 7 8
Turing Point $5,385 $4,825 $10,784 $5,738 $6,095 $6,095 $8,592 $8,592
(1985$int'l)
9o% C.I.: [$4977, [$4489, [$8644, [$5141,
95/o%C.I.: $5826] $5186] $13455] $6403]
Common time -0.0010 0.0346* 0.0013 0.0555*
trend, t: (0.0010) (0.0123) (0.0010) (0.0122)
Regional ts:
EAP 0.0054 0.2377* 0.0065* 0.2573*
(0.0029) (0.0417) (0.0029) (0.0416)
ECA . -0.0100* 0.0162 -.0126* -0.0305
(0.0032) (0.0399) (0.0032) (0.0403)
India 0.0194* 0.2789* 0.0225* 0.3189*
(0.0048) (0.0591) (0.0048) (0.05 79)
LAC 0.0145* 0.3100* 0.0120* 0.2639*
(0.0032) (0.0625) (0.0032) (0.0615)
MNA -0.0093* -0.0301 -.0067* 0.0106
(0.0023) (0.0301) (0.0025) (0.0324)
SA -0.0015 0.0361 0.0041 0.0956*
(0.0031) (0.0389) (0.0032) (0.0388)
SSA 0.0101* 0.1468* 0.0106* 0.1609*
(0.0013) (0.0191) (0.0013) (0.0186)
High Income -0.0191* -.0783* -.0154* -.0567*
(0.0016) (0.0177) (0.0017) (0.0176)
Significant at the 95% confidence level.
Standard Errors are given in parentheses.
8 Model Specifications:
1. Quadratic, common linear time trend
2. Quadratic, commnon log-linear time trend
3. Quadratic, regional linear time trends
4. Quadratic, regional, log-linear time trends
5. Spline, common linear time trend
6. Spline, common log-linear time trend
7. Spline, regional linear time trends
8. Spline, regional, log-linear time trends
Several results are worth emphasizing. The income levels at which the fatality
rate first declines are higher when region-specific time trends are included in the models
rather than a common-time trend. For example, F/P begins to decline at approximately
$8600 in the spline models with region-specific time trends but at approximately $6100
in the spline models with a common time trend. This reflects the fact that, in many
15
developing countries, fatality risk over the period 1963-99 grew faster than could be
explained by income growth alone. Because the region-specific time trends are jointly
significant, we believe that more emphasis should be placed on these models than on
models with a common time trend.
Whether the time trend enters the models linearly or in log-linear form, the
differences in trends across regions are generally similar. Over the estimation period
(1963-99) the fatality rate grew fastest in India and in Latin America (LAC) (holding
income constant), and almost as fast in Sub-Saharan Africa as in LAC. By contrast,
(holding income constant) F/P declined in high-income countries. Results for other
regions are statistically insignificant in at least some specifications.
Figure 6. Fatalities/Populadom Results, Commnioim Linear Tlime Tremd
1.8 l Quadratic,
1.6 commonT
1.4 Spline,
{ 1.2 1 g \' ' commonT
0 1.2
U0.8
~0.6
0.4
0.2
0
200 4,200 8,200 12,200 16,200 20,200 24,200 28,200
Per Capita GDP (1985 $inf1)
16
Figure 7. Fatalities/Population Results, Log-Linear Regional Time Trends
1.8-
1.8 -Quadratic,
1.6 ln(regionarr)
1.4 - Spine,
1.2 //\i(regionanl)
1.2
0.8
g 0.6
0.4
0.2
0
200 4,200 8,200 12,200 16,200 20,200 24,200 28,200
Per Capita GDP (1985 Sint'l)
It is also clear from examining Figures 6 and 7 that the relationship between per
capita income and the fatality rate is quite similar (holding the treatment of time constant)
whether one uses the quadratic or spline function. When a common time trend is
assumed F/P begins to decline at an income of $5,400 using the quadratic specification.
and at an income of $6,100 with the spline model. When region-specific, log-linear time
trends are included F/P begins to decline at an income of $5,700 in the quadratic model
and at an income of $8,600 in the-spline model. In both figures, the two models are
almost identical at low levels of income; however, the fatality rate peaks at a higher level
of income in the spline model and falls faster than in the quadratic model after it peaks.
In the Kuznets Curve literature it is standard practice to focus on the income level
at which the externality in question begins to decline. The usual interpretation is that, if a
17
country follows historical trends, the problem in question will eventually lessen once per
capita income reaches this turning point. Because the spline is a more flexible functional
form, we focus on the spline results in Table 4. The income levels at which fatalities per
person peak in the spline models, $6100 and $8600 (1985 international dollars) are within
the range of incomes at which Kuznets curves for common air and water pollutants peak
(Grossman and Krueger 1995). To better understand why this occurs, in the next sections
we examine models similar to those in Table 4 for the two components of fatalities per
person-vehicles per person and fatalities per vehicle.
C. Models of Vehicles per Person
Models for vehicles per person (V/P) are summarized in Table 5 and Figure 8.
Table 5 shows how motorization (V/P) varies with income in the both the quadratic and
spline models. Our discussion, however, focuses on the spline models. Figure 8 plots the
four spline models using the country intercept for India, with the time trend set equal to
1999.
Of the four spline models in Table 5, only two (Models 6 and 8) show vehicles
per person increasing with per capita income at a decreasing rate for all relevant values of
per capita income. This result occurs when time is entered in a log-linear fashion; when it
enters the motorization equation linearly, V/P peaks at a value of income observed in the
data, and the time trend associated with vehicle ownership is large and positive. The log-
linear time trend thus seems to yield more reasonable results than the linear. Of the two
models with log-linear time trends, only the model with regional log-linear time trends
18
yields non-negative income elasticities for all levels of income; hence we focus on this
model.
Figure 8. Vehicles/Population Results, Spline-Model
300
.h(regionarr)
__ , . I~~n(ca6umonT)
250 -conunonT
regionalT
200-
10
10
, 150
4w100-
50
$200 $4,200 $8,200 $12,200 $16,200 $20,200 S24,200 $28,200
Per Capita GDP (1985 $iit<D
Notes for Table 5:
Standard Errors are given in parentheses.
The constant term reflects the intercept terrn for India. Country fixed effects were included in all
regressions but are not displayed here. i
Asterisks indicate significance at the 95% confidence level.
Model Specifications:
1. Quadratic, common linear time trend
2. Quadratic, common log-linear time trend
3. Quadratic, regional linear time trends
4. Quadratic, regional, log-linear time trends
5. Spline, common linear time trend
6. Spline, common log-linear time trend
7. Spline, regional linear time trends
8. Spline, regional, log-linear time trends
'19
-'rable S. Regi iessonm IResults firon VehAcRes/PopunI om Models
Independent S__dratic Spline
Variables 1 2 3 4 5 6 7 8
5.52280 5.0464� 3.0785� 3.0826�
LnY (0.2120) (0.2301) (0.2976) (0.3042)
-0.28600 -0.2407� -0.1420� -0.1242�
(lnYit)2 (0.0129) (0.0138) (0.0182) (0.0183)
InY for: 0.8569� 0.95510 0.4247� 0.1899
$1 - $946 (0.1281) (0.1473) (0.1239) (0.1471)
$946 - $1,535 1.4314� 1.7748� 0.9087� 1.6752�
(0.1Q020) (0.1149) (0.1088) (0.1168)
$1,535 - $2,290 0.67480 1.0435� 0.7078� 1.0849�
(0.1 027) (0.1157) (0.1139) (0.1258)
$2,290 - $3,441 1.3786� 1.50830 0.88250 1.0385�
(0.1115) (0.1285) (0.1127) (0.1332)
1.2179� 1.57020 1.1359� 1.5656�
$3,441 - $4,682 (0.1347) (0.1524) (0.1250) (0.1426)
0.80520 1.14380 0.8378� 1.18700
$4,682 - $6,911 (0.1025) (0.1156) (0.0955) (0.1081)
$6,911 -$9,238 0.56910 0.59490 0.68830 0.82230
(0.1267) (0.1462) (0.1193) (0.1399)
$9,238-$11,263 -0.43590 -0.0216 -0.33670 0.1884
.9,238-$1 1,263 (0.1756) (0.1999) (0.1663) (0.1902)
$11,263-13,663 -0.71420 0.2125 -0.55040 0.4123�
(0.1750) (0.1941) (0.1693) (0.1835)
-0.61190 0.0978 .0.5092* 0.1897
>____________ $13,663 _______ (0.1477) (0.1650) (0.1396) (0.1540)
Tuming Point $15,587 $35,723 $51,179 $244117 $9,238 $9,238
(1 985$int'l)
95% C.I.: [$12995, ($26690, ($24,687, [$71,420,
9S%___ C.I.:___$18,696] $47,813] $106099] $834411]
Common time 0.0281 0 0.2358� 0.03220 0.263 10
trend: t (0.0010) (0.0132) (0.0010) (0.0130)
Regional t: 0.05390 0.55870 0.05820 0.59900
EAP (0.002 7) (0.0417) (0.0026) (0.0418)
ECA 0.05970 0.54000 0.0559� 0.5046�
(0.0031) (0.0389) (0.0030) (0.0400)
India 0.06570 0.66190 0.07470 0.73310
(0.0039) (0.0501) (0.003 7) (0.0483)
LAC 0.02560 0.21840 0.02340 0.20170
(0.0042) (0.0574) (0.0040) (0.0572)
0.01570 0.04490 0.02010 0.0358
MNA (0.0029) (0.0368) (0.0030) (0.0385)
SA 0.03590 0.25470 0.04060 0.24680
(0.0027) (0.0365) (0.0027) (0.0361)
SSA 0.02900 0.31190 0.03040 0.35290
(0.0014) (0.0202) (0.0013) (0.0196)
0.02150 0.16040 0.02920 0.19770
High Income _______ (0.0014) (0.0168) (0 0014) (0.0168)
F statistic on F8,1791 = F8,1791 = F8,1783 = F8,1783=
regional ts: 151.15 80.03 202.95 96.64
-20.6880 -19.6750 -11.3780 -12.8060 -2.28720 -3.10380 -0.0692 0.8421
Constant (0.8645) (0.9430) (1.1 795) (1.2143) (0.8586) (0.9835) (08261) (0.9753)
Adjusted R2: 0.9783 0.9742 0.9818 0.9775 0.9820 0.9767 0.9849 0.9799
No. of Countries: 75, No. of observations: 1876
20
In the preferred model, Model 8, the income elasticity of vehicle ownership
attains a maximum value of 1.67 in the second spline segment ($946 - $1,535 (1985
intemational dollars)) and decreases to a low of 0.18 in the highest income category.
Above the lowest income category, income elasticities in income categories 2 through 5
are significantly higher than income elasticities in income categories 7 to 10.8 These
results are broadly consistent with previous studies of motorization, which find that the
income elasticity of demand for motor vehicles declines with income (Ingram and Liu
(1998), Dargay and Gately (1999), Button et al. (1993)). Figure 8 suggests that the rate
of increase in motorization slows down considerably after reaching a per capita income
of $9400 (1985 international dollars), the level of income attained by Norway and the
United Kingdom in 1974.
D. Models for Fatalities per Vehicle
Table 6 and Figure 9 confirm that fatalities per vehicle decline sharply with
income. Focusing once again on the spline models with log-linear time trends, F/V
declines with income for per capita GDP in excess of $1,180 (1985 international dollars)
when either common or regional log-linear time trends are used. Figure 9, which plots
the four spline models for India (t = 1999) indicates exactly how fast F/V declines as
income grows. Fatalities per vehicle decline by a factor of 3 (e.g., from 360 to 120 per
100,000 vehicles for India) as per capita income grows from $1200 to $4400. After
reaching a per capita income of $15,200 (1985 international dollars), however, F/V
8Income elasticity estimates generated from a two-segment spline model are statistically different from
each other, decreasing from 1.32 (0.05)) to 0.719 (0.052) once per capita income exceeds $4,682 (1985
international dollars).
21
Tab1Re 6. Regressoim IResults froim ]FaUallles/Vehinees ModeRs
Independent Quadratic Spline
Variables 1 2 3 4 5 6 7 8
2.3796� 3.0043� 2.2926� 3.0550�
LnY (0.2509) (0.2915) (0.3829) (0.4240)
j0.1612� -0.2313� -0.1458� -0.23400
(lnYit)2 (0.0151) (0.0173) (0.023 1) (0.0252)
InY for: 0.6190� 0.30240 1.05170 0.6548�
$1-$1,179 (0.1167) (0.1372) (0.1268) (0.1551)
$1,179-$1,730 -0.2292 -0.9258* -0.50220 -1.52270
(0.1468) (0.1697) (0.1585) (0.1867)
$1 ,730-$2,698 0.2538 -0.2607 0.44620 -0.2433
(0.1334) (0.1552) (0.1426) (0.1 71 1)
$2,698-$3.813 0.5814� -1.09520 -0.0908 -0.83210
$2,698-$3,813 (0.1485) (0.1 751) (0.1519) (0.1834)
$3,8 1 3-$5,39 1 -0.3880� .0.93400 -0.1747 -0.87270
(0.1415) (0.1640) (0.1359) (0.1620)
$5,391-$7,532 -0.55500 -1.00290 .0.4913� 1.03570
(0.1447) (0.1686) (0.1391) (0.1668)
-0.6347� .0.94440 .0.3684� -0.8750�
(0.1745) (0.2075) (0.1697) (0.2083)
$9,614-$11,469 -0.28262- .. 19764 0.0859 ;1.14072
(0.242)) (0.2824) (0.2354) (0.2810)
$11,469-13,682 -0.4438 -1.8455� 0.0051 -1.79770
(0.2363) (0.2690) (0.2333) (0.2664)
>$13,682 -0.2782 -1.2485� 0.0143 -1.2226�
>$____________ ,______ ________ (0.1905) (0.2185) (0.1855) (0.2151)
Turing Point $1,603 $661 $2,593 $683 $2,698 $1,179 $2,698 $1,179
(1985$int'l)
95% C.Il: [$1,189, [$475, [$1,859, [$440,
. $2,161] $920] $3,616] $1,061]
Conmon time -0.03700 -0.23350 -0.03860 -0.22850
trend: t (0.0013) (0.0170) (0.0013) (0.0176)
Reaional t: 0.05120 -0.26610 -0.0561 0 -0.28260
EAP (0.0034) (0.0540) (0.0034) (0.0560)
ECA -0.0671 0 -0.44160 .0.06980 -0.46090
(0.0036) (0.0487) (0.0038) (0.0522)
India 0.0445* -0.355 1 �0.0536� .0.44160
Indla (0.0049) (0.0686) (0.0050) (0.0701)
LAC .0.01230 -0.0219 -0.0127� -0.0142
(0.0060) (0.1362) (0.0061) (0.138 7)
-0.0268� -0.0503 -0.0234� 0.0922
MNA (0.0035) (0.0490) (0.0039) (0.0543)
SA -0.0392� .0.19250 �0.03750 -0.1230�
(0.0033) (0.0462) (0.0035) (0.04 73)
SSA -0.02890 -0.25270 -0.02840 -0.23650
(0.0020) (0.0311) (0.0019) (0.0306)
High Income -0.04240 -0.22930 .0.04700 -0.24500
High Incorne (0.0017) (0.0215) (0.0018) (0.0223)
F statistic on Fg,1615 = Fs,161s= F8,1607= F8,1607=
regional ts: 132.45 29.78 140.67 31.56
-11.1050 .12.1660 .11.0920 -12.0550 -6.58290 -4.51750 -9.24310 -6.32740
Constant (1.0305) (1.2035) (1.5275) (1.7003) (0.7967) (0.9333) (0.8365) (1.0105)
Adjusted R2: 0.9532 0.9363 0.9569 0.9377 0.9540 0.9368 0.9588 0.9394
No. of Countries: 70, No. of observations: 1695
22
declines very slowly in absolute terms: from 25 per 100,000 vehicles at an income of
$20,000 to 15 per 100,000 vehicles at an income of $30,000.
Figure 9. Fatalities/Vehicle Results, Spline Models
400 - In(regionarl)
350 - _ - In(commnT)
comnmonT
>300 - regionalT
250 -
C1200-IN
150-
o2
50
50
$200 $4,200 $8,200 $12,200 $16,200 $20,200 $24,200 $28,200
Per Capita Income (1985 $infl)
Combining the results of model 8 for F/V and for V/P explains the results for
deaths per capita observed in Figure 7. The elasticity of V/P with respect to income
exceeds in absolute value the elasticity of F/V with respect to income for incomes up to
the $7,000-$9,000 interval, when the two elasticities are approximately equal-the
condition for (FNV)*(V/P) to peak.9 At higher incomes, the elasticity of fatalities per
vehicle with respect to income exceeds the elasticity of motorization with respect to
income.
9 This comparison is approximate since the width of the income intervals differs in Tables 5 and 6.
23
IV. Predictions of Future Traffic Fatalities and Motorization
One reason for estimating the preceding models is to predict what will happen to
traffic fatalities if historic trends continue. Future traffic fatalities can be predicted
directly from equation (1); i.e., by predicting future fatality rates (F/P) and multiplying by
estimates of future population, or by predicting vehicle ownership, V, from the V/P
equation and multiplying the vehicle stock by fatalities per vehicle. The second method
serves as a check on the first since more is known about vehicle ownership. In particular,
one can reject models that yield unbelievably high rates of vehicle ownership; e.g.,
ownership significantly in excess of one vehicle per person in the year 2020.
To project future vehicle ownership and traffic fatalities assumptions must be
made about income and population growth. The real per capita GDP series is projected
to 2020 using the World Bank's forecasts of regional growth rates (2000-2010) (Global
Economic Prospects 2002) with the assumption that the average annual 2001-2010
growth rates continue to 2020. (A list of the growth rates is provided in Appendix Table
A.3.) Population projections are taken from the U.S. Census International Data Base. In
total, the explanatory variables are available for 156 countries (representing 92% of total
world population in 2000), including 45 highly developed countries (HD 1) and Ill
developing countries (HD2). Table 7 shows the number of countries in each geographic
region for which predictions are made.
24
Table 7. Regional Distribution of Countries for Which Predictions are Made
WB Region HD2 HD1
East Asia & Pacific 14 1
E. Europe & Central Asia 5 4
Latin America & Caribbean 27 4
Middle East & North Africa 12 1
South Asia 7
Sub-Saharan Africa 46
High-Income Countries 35
Total: 111 45
To calculate the point estimates for the out-of-sample countries, assumptions must
be made regarding the country-specific intercept. The coefficient on the country dummy
variable for Chile is used to compute the predicted values for the 10 out-of-sample HDI
countries.'0 For the HD2 countries, the intercept is set equal to the mean of the country
intercepts for the corresponding region.
A. Projections of the World Vehicle Fleet to 2020
We begin by examining the implications of the models in Table 5 for future
growth in vehicle ownership. Figure 10 displays projections of the vehicle fleet
corresponding to all 8 models in Table 5. Not surprisingly, it is the form of the time
trend, rather than the choice between the spline and quadratic specifications, that makes
the greatest difference in the projections. Both the spline and quadratic models with
linear regional time trends yield unbelievably large estimates of the world motor vehicle
stock in 2020, as well as estimates of vehicle ownership per capita for certain groups of
countries that are well over 1. For this reason we focus on the spline models with log-
'� The choice of Chile is motivated by the fact that the most populous out-of-sample HD1 countries for
which predictions must be made are Argentina and Uruguay.
25
linear time trends. The model with region-specific log-linear time trends (Model 8)
generates forecasts of 1.47 billion vehicles in 2020, whereas the vehicle stock is predicted
to be over 1.37 billion vehicles when a common log-linear time trend is used (Model 6)."
IFigure 10. World Vehicle leet 1Projecdons Correspoimdalmg to Models iAmD Tlable 5
3,500 - l Model I
3,000 - X b ~~~~~~~-CModel 2
3,000 - Model 3
t < 2< ~~~~~~~~~~~~~~:Model 4
2,500 - Model 5
EB 2,000 - ~~~~~~~~~~~~Model 6
~32,000-
1,500 -
1,000 -..'"'''
500
1990 1995 2000 2005 2010 2015 2020
The predictions of these models agree fairly well with other estimates of vehicle
growth in the literature. Dargay and Gately (1999) project that the total vehicle fleet in
OECD countries will reach 705 million by 2015 (a 62% increase from 1992 values). The
sphne model (with the common time trend or the log-linear regional time trends) yields a
2015 estimate of 687 million vehicles for the same group of countries.
l The corresponding quadratic models give almost identical vehicle projections but we continue to focus
on the more flexible spline specifications.
26
Other studies have made projections of vehicle growth for the automobile fleet
only or for passenger cars and commercial vehicles. Since our motor vehicle counts
include all buses and two-wheelers, direct comparisons with these studies is difficult.
However, our estimates of the total vehicle fleet do exceed their automobile forecasts
under all specifications. Under Schafer's (1998) results, the global automobile fleet
would more than double from 470 million in 1990 (this includes light trucks for personal
travel in the U.S.) to 1.0-1.2 billion automobiles in 2020. This amnounts to a 113%-155%
increase in total automobiles. The spline model with log-linear regional time trends
generates a 140% increase in the total vehicle fleet during the same period (from 609
million to 1.47 billion total vehicles). Because it yields reasonable predictions of the
vehicle fleet, as well as reasonable income elasticity estimates, we focus on the spline
model with regional, log-linear time trends.
B. Projections of World Traffic Fatalities to 2020
Figure 11 shows predictions of road traffic fatalities for all countries to the year
2020, based on the spline model with log-linear regional time trends. Ninety-five percent
confidence levels for our predictions also appear on the graph.12 We emphasize that
these figures represent traffic fatalities unadjusted for under-reporting. To compare these
figures with fatality rates from other causes of death, it is necessary to adjust for the fact
that (a) the definition of what constitutes a traffic fatality differs across countries and (b)
the percentage of traffic fatalities reported by the police also varies across countries.
'2The model generates point estimates of the log of the fatality rate (ln(Fatalities/10,000 People)).
Therefore, the confidence intervals for the predicted values of In(Fatalities/10,000 People) are symmnetric,
but the forecast intervals for the total number of fatalities are not.
27
Figure 11. GDobalR oadT lraffi Fatalities, 1Before Adjusdtlg for Uideir-Reportng,
1990-2020
Estimated Number of Traffic Fatalities
C: � Lower Bound of 95% Forecast Interval
Upper Bound of 95% Forecast Interval
1,400,000 -
1,200,000 -
1,000,000-
800,000 -
600,000 - = = -
400,000- C -
200,000
1990 1995 2000 2005 2010 2015 2020
Our under-reporting adjustments follow the conservative factors used by Jacobs,
Aeron-Thomas and Astrop (2000).13 To update all point estimates to the 30-day traffic
fatality definition, a correction factor of 1. 15 was applied in the developing countries and
the standard ECMT correction factors were used for the high-income countries.14 Then
the estimates were adjusted to account for general under-reporting of traffic fatalities, by
25% in developing countries and by 2% in highly developed countries.15 With these
adjustments, global road deaths are projected to climb to over 1.2 million by 2020 (a 40%
adjustment over the base estimate of 864,000 presented in Figure 11). Although this
'3Jacobs et al. reviewed numerous underreporting studies and found evidence of underreporting rates
ranging from 0-26% in high motorized countries and as high as 351% in less motorized countries.
Fatalities in China, for example, were 42% higher in 1994 than reported in official statistics (Liren 1996).
14 High-Income countrnes with ECMT correction factors greater than 1 include: France: 1.057, Italy: 1.07,
Portugal: 1.3, Japan: 1.3 (ECMT 1998, 2000,2001).
15 The 25% under-reporting adjustment is applied to 111 HD2 countries and the 2% adjustment is used for
45 HD 1 countries. See Table 7 for regional breakdown of countries.
28
represents a 66% increase over the 2000 world estimate, the trend varies considerably
across different regions of the world. Table 8 and Figure 12 indicate that, between 2000
and 2020, fatalities are projected to increase by over 80% in developing countries, but
decrease by nearly 30% in high-income countries. Witiin the developing world, the
greatest percentage increases in traffic deaths between 2000 and 2020 will occur in South
Asia (144% increase), followed by East Asia and Sub-Saharan Africa (both showing an
80�% increase). It is also interesting to note that the number of traffic fatalities per
100,000 persons is predicted to diverge considerably by 2020. By 2020 the fatality rate is
predicted to be less than 8 in 100,000 in high-income countries but nearly 20 in 100,000
in low-income countries.
Table 8. Predicted Road Traffic Fatalities by Region (OOOs), Adjusted for Under-
Repor1ing, 19902020
Fatality Rate
(Deaths/100,000
Persons)
No. of % change
Region* Countries 1990 2000 2010 2020 'O-'20 2000 2020
EAP 15 112 188 278 337 79.8% 10.9 16.8
ECA 9 30 32 36 38 18.2% 19.0 21.2
LAC 31 90 122 154 180 48.1% 26.1 31.0
A 13 41 56 73 94 67.5% 19.2 22.3
A 7 87 135 212 330 143.9% 10.2 18.9
SA 46 59 80 109 144 79.8% 12.3 14.9
Subtotal. 121 419 613 862 1,124 83.3% 13.3 19.0
High Income
Countries: 35 123 110 95 80 -27.8% 11.8 7.8
World Total: 156 542 723 957 1,204 66.4% 13.0 17.4
*The results are displayed according to the World Bank regional classifications.
29
FiAgure 12. GilobaD Tlraefc Fatalities, Adjusted for Unideir-Reportnmg, 1990-2020
1,400,000 -Z High Income
1,200,000 - OSSA
1,000,000 - 0 MNA
800,000 - iECA
, l O ~~~~EAP
600,000
400,000 ,q,, ,
200,000
0 . -l ....... r-''''+_- -'
1990 1995 2000 2005 2010 2015 2020
The forecasts presented here are significantly lower than the 1990-2020 estimates
of road traffic fatalities presented by the World Health Organization in The Global
Burden of Disease (GBD) (Murray and Lopez 1996). WHO estimates that 1.39 million
people will die in road traffic accidents in 2000 and that 2.34 million will die in 2020.
The reason for the higher figures is that WHO starts from a higher base (999,000 deaths
in 1990). In part, the high GBD base estimate for 1990 may be due to severe data
limitations in developing regions. For example, 1990 estimates for the entire Sub-
Saharan Africa region were based on data from South Africa only (Cooper et al. 1997,
Jacobs et al. 2000). South Africa has by far the highest reported fatality rate (F/P) of
nearly 20 SSA countries for which we have 1990 data. Even after adjusting predicted
values for non-reporting and under-reporting of fatalities, our SSA estimate is 59,150
deaths for 1990 whereas the GBD baseline is 155,000 for the same year. Despite such
30
large differences between our base estimates and theirs, Murray and Lopez predict that
global traffic fatalities will grow at approximately the same rate as the present
projections. (Fatalities grow by 62% between 2000 and 2020 according to WHO and by
66% according to our estimates (see Table 8).)
We believe that Murray and Lopez (1996) have over-estimated road traffic
fatalities and stand behind the estimates presented here. One reason for this is that our
estimate of fatalities in 2000 (723,439) agrees with the TRL estimate of global road
deaths for 1999 (Jacobs et al. 2000), i.e., 745,769 fatalities worldwide (low under-
reporting adjustment case). The TRL 1999 estimate is based on published 1996 data
from 142 countries updated to 1999 levels and adjusted for non-reporting and under-
reporting of fatalities. Since this seems to be the most comprehensive, bottom-up
approach to estimating the global road death toll to date, we feel that it is the most
appropriate estimate against which to compare our projections. Our prediction of traffic
fatalities in 2020 (1.2 million deaths worldwide) also lies within the range suggested by
TRL, for 2020 (1 to 1.3 million deaths), although the latter is not based on a statistical
model.
V. Conclusions
The results presented above suggest that, if developing countries follow historic
trends, it will take many years for them to achieve the motor vehicle fatality rates of high-
income countries. Provided that present policies continue into the future, the traffic
fatality rate of India, for example, will not begin to decline until 2042.16 (The projected
16 This assumes the annual real per capita GDP growth rate of 3.87% and India's log-linear time trend
(from model 8) will continue into the future.
31
peak corresponds to approximately 24 fatalities per 100,000 persons prior to any
adjustment for underreporting but becomes 34 fatalities per 100,000 persons if we
maintain the underreporting adjustment factors chosen above.) This is primarily due to
the fact that India's per capita income (in 1985 international dollars) was only $2,900 in
2000, whereas F/P peaks at a per capita income of approximately $8,600. Similarly, in
Brazil F/P will not peak until 2032, and the model projects over 26 deaths per 100,000
persons as far out as 2050.
In other developing countries, the traffic fatality rate will begin to decline before
2020 but F/P rates will still exceed the levels experienced in High-Income countries
today (which average about 11 fatalities per 100,000 persons). Malaysia, for exaample, is
estimated to have over 20 fatalities per 100,000 persons (after adjusting for
underreporting) in 2020. If 5.1% growth continues beyond 2020, F/P will reach 11.1 by
2033 (using the same under-reporting adjustment as above); however, if the growth rate
decreases to 2.5% after 2020, F/P will reach 11.0 only in 2049.
The predictions in this paper, and the estimates of the income levels at which
traffic fatality rates begin to decline, assume the policies that were in place from 1963
through 1999 will continue in the future. Clearly, this may not be the case. In many
developing countries fatalities per vehicle could be reduced significantly through
interventions that are not reflected in our data. For example, drivers of two-wheelers
could be required to wear white helmets, traffic calming measures could be instituted in
towns, and measures could be taken to separate pedestrian traffic from vehicular traffic.
Whether such measures should be undertaken depends, of course, on their costs and their
32
effectiveness. The purpose of this paper has been to increase awareness of the nature and
growing magnitude of the problem.
33
IReferences
American Automobile Manufacturers Association. 1993. World Motor Vehicle Data
1993. Washington, D.C.: AAMA.
Bangladesh Bureau of Statistics. National Data Bank. http://www.bbsgov.org/ .
China Statistical Yearbook. (various years). State Statistical Bureau, People's Republic of
China.
Cooper, R.S., B. Osotimehim, J. Kaufman, and T. Forrester. 1998. Disease burden in sub-
Saharan Africa: what should we conclude in the absence of data? The Lancet 351:
208-210.
Cross National Time Series Database (CNTS). Access through Rutgers University
Libraries. http://www.scc.rutgers.edu/cnts/about.cfm.
Dargay, Joyce, and Dermot Gately. 1999. Income's effect on car and vehicle ownership,
worldwide: 1960-2015. Transportation Research A 33:101-138.
European Conference of Ministers of Transport. 2001. Statistical Report on Road
Accidents 1997-1998. Paris: OECD Publications Service.
European Conference of Ministers of Transport. 2000. Statistical Report on Road
Accidents 1995-1996. Paris: OECD Publications Service.
European Conference of Ministers of Transport. 1998. Statistical Report on Road
Accidents 1993-1994. Paris: OECD Publications Service.
Grossman, Gene and Alan Krueger. 1995. "Economic Growth and the Environment."
Quarterly Journal of Economics 110(2): 675-708
Inter-American Development Bank. 1998. Review of Traffic Safety Latin America and
Caribbean Region. Transport Research Laboratory and Ross Silcock.
International Road Federation. Various Years. I.R.F. World Road Statistics. Geneva,
Switzerland: International Road Federation.
International Road Traffic Accident Database (IRTAD). Paris, France: Federal Highway
Research Institute (BASt)/OECD Road Transport Research Programme.
Israel National Road Safety Authority. 2000. The National Road Safety Authority
Structure, Activities, & Responsibilities. State of Israel Ministry of Transport.
Ingram, Gregory K., and Zhi Liu. 1998. Vehicles, Roads, and Road Use: Alternative
Specifications. Working Paper 2036. World Bank.
Jacobs, G, A. Aeron-Thomas, and A. Astrop. 2000. Estimating Global Road Fatalities.
TRL Report 445. London, England: Transportation Research Laboratory.
34
Liren, D. 1996. Viewing China Road Traffic Safety and the Countermeasures in
Accordance with International Comparison. Beijing Research in Traffic Engineering,
Second Conference in Asian Road Safety, 28-31 October 1996.
Murray, C. and A. Lopez (eds.). 1996. The Global Burden of Disease. Cambridge, MA:
Harvard Press.
Road Safety in Latin America and the Caribbean: Analysis of the Road Safety Situation
in Nine Countries. Copenhagen, Denmark: Ministry of Transport Road Directorate,
1998.
Schmalensee, Richard, Thomas M. Stoker, and Ruth A. Judson. 2000. World Carbon
Dioxide Emissions: 1950-2050. Review of Economics and Statistics 80(1):15-27.
Schafer, Andreas. 1998. The Global Demand for Motorized Mobility. Transportation
Research A 32(6):455-477.
Smeed, R. J. 1949. "Some Statistical Aspects of Road Safety Research." Journal of the
Royal Statistical Society Series A. 112:1-23.
Statistical Economic and Social Research and Training Centre for Islamic Countries
(SESRTCIC). Basic Socio-Economic Indicators Database (BASEIND).
http://www.sesrtcic.orgldefaulteng.shtml.
United Nations Economic and Social Commission for Asia and the Pacific. 1997. Asia-
Pacific Road Accident Statistics and Road Safety Inventory. New York: United
Nations.
U.S. Census Bureau. International Data Base (IDB).
http://www.census.gov/ipc/www/idbsprd.html
World Bank. Global Development Network Growth Database Macro Time Series.
http://www.worldbank.ora/research/lrowth/GDNdata.htm .
World Bank. Global Economic Prospects and the Developing Countries 2002.
http://www.worldbank.org/nrospects/gep2002/toc.htm
World Resources Institute. 1998-1999. World Resources. Washington, D.C.
http://www.wri.org/wr-98-99/autos.htrn.
35
VI. Appendix
Data Sources
Data on the number of traffic fatalities and vehicle fleet17 composition were taken
from various editions of the International Road Federation's (IRF) World Road Statistics
(WRS) annual yearbooks, 1968-2000. Since each WRS edition contains data for the
previous five years, each series was compared across editions to check for accuracy and
to ensure that all revisions were properly recorded. Selected IRF data were also
compared to numerous regional and country-specific road safety studies. Supplementary
data was added from several sources where appropriate, including studies published by
the
o Inter-American Development Bank (1998)
o Danish Road Directorate (1998)
o Transportation Research Laboratory (Jacobs et al. 2000)
o United Nations Economic and Social Commission for Asia and the Pacific
(1997)
o Statistical Bureau of the People's Republic of China
o Ministry of Transport of Israel (2000)
o European Conference of Ministers of Transport (ECMT)
o Global Road Safety Partnership
o OECD Intemational Road Traffic Accident Database (IRTAD)
o Cross-National Time Series Database (CNTS)
o Statistical Economic and Social Research and Training Centre for Islamic
Countries (SESRTCIC)
o Bangladesh Bureau of Statistics
o American Automobile Manufacturers Association
7Vehicle counts include all passenger cars, buses, trucks, and motorized two-wheelers.
36
Population figures came from the U.S. Census Bureau's International Data Base
and income data were taken from the World Bank Global Development Network Growth
Database Macro Time Series. The income variable most appropriate for this analysis is
the Real Per Capita GDP, chain method (1985 international prices) (RGDPCH), since it
accounts for differences in purchasing power across countries and allows for comparisons
over time.'8
i8This series was created from the Penn World Tables 5.6 RGDPCH variable for 1960-1992 and the 1992-
1999 data was estimated using the 1985 GDP per capita and GDP per capita growth rates from the Global
Development Finance and World Development Indicators.
37
Table A.1. Classification of Countries for Which Fatalities Are Projected
The number of observations for each country used to estimate the model is given in parentheses after the
country name.
Re-Rion _u_t_ Regon Countr|
South Asia (SA) Latin America &Caribbean (LAC)
HD2:Bangladesh (18) HDI:Argentina
Bhutan Chile (34)
India (27) Costa Rica (14)
Maldives Uruguay
Nepal (14, HD2:Antigua & Barbuda
Pakistan (33) Belize
Sri Lanka (34) Bolivia
Brazil (16)
East Asia & Pacific (EAP) Colombia (27)
HDl:Korea, Rep. (29) Dominica
HD2:China (22, Dom Republic
Fiji (15, Ecuador (13)
Indonesia (24, El Salvador
Kiribati Grenada
Lao PDR Guatemala
Malaysia (37, Guyana
Mongolia (14, Haiti
Papua New Guinea (12, Honduras
Philippines (18, Jamaica
Samoa (13, Mexico
Solomon Islands Nicaragua
Thailand (28 Panama (16J
Tonga (13) Paraguay
Vanuatu Peru (1)
Puerto Rico
Middle East & N. Africa (MNA) St Kitts & Nevis
HDI:Bahrain (12) SL Lucia
HD2:Algeria St. Vincent & Grenadines
Djibouti Suriname
Egypt (I 1 Trinidad & Tobago
Iran Venezuela
Iaq (11,
Jordan (32) Europe & Central Asia (ECA)
Morocco (34, HDl:Czech Republic (28)
Oman Hungary (33)
Saudi Arabia (20, Poland (19)
Syria (22, Slovak Republic
Tunisia (26, HD2:Bulgaria (18)
Yemen (12, Georgia
Latvia (11)
Romania (13)
Turkey (34)
Yugoslavia (20)
38
(Table A.l. continued)
Sub-Saharan Africa (SSA) High-Income OECD
HD2:Angola HDI:Australia (31,
Benin (23, Austria ( 37)
Botswana (31, Belgium (36,
Burkina Faso Canada (30,
Burundi Denmark (37,
Cameroon (17, Finland (37,
Cape Verde France (37,
Central African Republic Germany (30,
Chad Greece (36,
Comoros Iceland (35,
Congo, Der. Rep. Ireland (31,
Congo, Rep. Italy (35,
Cote d'Ivoire (17, Japan (37
Equatorial Guinea Luxembourg (34,
Ethiopia (30) Netherlands (37,
Gabon New Zealand (37,
Gambia, The Norway (37)
Ghana Portugal (37,
Guinea Spain (32,
Guinea-Bissau Sweden (35)
Kenya (27) Switzerland (36)
Lesotho (21) United Kingdom (35,
Liberia United States (36,
Madagascar
Malawi (28, Other High-Income
Mali HDI:Bahamas
Mauritania Barbados
Mauritius (26) Bermuda
Mozambique (12) Cyprus (30,
Narnibia Hong Kong (36,
Niger (24) Israel (32,
Nigeria (15, Kuwait
Rwanda Malta
Sao Tome &Principe Qatar
Senegal (19) Singapore (16,
Seychelles Taiwan (25,
Sierra Leone (17, U.A.E.
Somalia
South Africa (35)
Sudan
Swaziland (13)
Tanzania
Togo (16)
Uganda (19)
Zambia (13)
Zimbabwe .1 (IS,__
Total: 156 Countries, 2,178 Country-year Observations
39
Notes for T1aDbe A.2:
Standard Errors are given in parentheses.
The constant term reflects the intercept term for India. Country fixed effects were
included in all regressions but are not displayed here.
Model Specifications:
1. Quadratic, common linear time trend
2. Quadratic, common log-linear time trend
3. Quadratic, regional linear time trends
4. Quadratic, regional, log-linear time trends
5. Spline, common linear time trend
6. Spline, common log-linear time trend
7. Spline, regional linear time trends
8. Spline, regional, log-linear time trends
40
Table A.2. Regression Results from Fatalities/Population Models
Independent dratic Spline
Variables 1 2 3 4 5 6 7 8
7.7500* 7.8146* 5.1714* 6.1926*
InY (0.2243) (0.2223) (0.3058) (0.2886)
-0.4510* -0.4607* -0.2785* -0.3578*
(1nYi2 fo(0.0137) (0.0134) (0.0191) (0.0/ 77)
InY for: 1.6837* 1.6067* 1.4445* 1.2526*
$1- $938 (0.1432) (0.1434) (0.1400) (0.1439)
$938- $1,395 1.1798* 1.1445* 1.1193* 1.0604*
(0.1266) (0.1258) (0.1322) (0.1283)
$1,395- $2,043 0.5064* 0.4277* 0.5119* 0.3256*
(0.1219) (0.1209) (0.1250) (0.1278)
$2,043- $3045 0.9199* 0.8177* 0.9757* 0.7651*
(0.1304) (0.1302) (0.1299) (0.1345)
$3,045- $4,065 0.6995* 0.6221 * 0.9602* 0.7007*
(0.1526) (0.1520) (0.1492) (0.1495)
$4,065- $6,095 0.3225* 0.2758* 0.6021 * 0.3903*
(0.1146) (0.1130) (0.1119) (0.1107)
$6,095- $8,592 -0.0477 -0.1119 0.2579* 0.0746
(0.1318) (0.1315) (0.1288) (0.1299)
-0.7910* -0.9382* -0.2074 -0.5422*
$8,592-$10,894 (0.1645) (0.1658) (0.1633) (0.1668)
$10,894-13,234 -1.5720* -1.6733* -0.6682* -1.3381*
(0.1952) (0.1904) (0.1999) (0.1886)
>$13,234 -1.1509* -1.2061* -0.5216* -0.9964*
$__3,234 (0.163 7) (0.1603) (0.163 7) (0.1571)
Tuming Point $5,385 $4,825 $10,784 $5,738 $6,095 $6,095 $8,592 $8,592
(I 985$intl)
95% C.I.: [$4977, [$4489, [$8644, [$5141,
$58261 $5186] $13455] $6403]
Commnon time -0.00 10 0.0346* 0.0013 0.0555*
trend: t (0.0010) (0.0123) (0.0010) (0.0122)
Regional t: 0.0054* 0.2377* 0.0065* 0.2573*
EAP (0.0029) (0.04 17) (0.0029) (0.0416)
ECA -0.01 00' 0.0162 -0.0126* -0.0305
(0.0032) (0.0399) (0.0032) (0.0403)
India 0.0194* 0.2789* 0.0225* 0.3189*
(0.0048) (0.0591) (0.0048) (0.05 79)
LAC 0.0145* 0.3100* 0.0120* 0.2639*
(0.0032) (0.0625) (0.0032) (0.0615)
MNA -0.0093* -0.0301 -0.0067* 0.0106
(0.0023) (0.0301) (0.0025) (0.0324)
SA -0.0015 0.0361 0.0041 0.0956*
(0.003 1) (0.0389) (0.0032) (0.0388)
SSA 0.0101* 0.1468* 0.0106* 0.1609*
(0.0013) (0.0191) (0.0013) (0.0186)
High Income -0.0191* -0.0783* -0.0154* -0.0567*
High Income (0.0016) (0.0177) (0.001 7) (0.01 76)
F statistic on F8,2102 = F8,2102= F8,2os4= F8,2094=
regional ts: 35.60 19.72 29.50 20.99
-32.948* -33.046* -23.796* -27.427* -12.550* -12.149* -11.359* -10.462*
constant (0.9128) (0.9092) (1.1952) (1.1416) (0.9606) (0.9596) (0.9360) (0.9599)
Adjusted R2: 0.8455 0.8460 0.8634 0.8557 0.8554 0.8567 0.8695 0.8656
No. of Countries: 88, No. of observations: 2200
41
TsabD A.3. IFoirecasts of Re2a Peir Capift GIDIP Amunuall Girowth Rates (%)9 2000-2020
elioim/Counatry I 2000 I 2001 2002=2020
South Asia 3.0 2.8 3.8
India 3.3 2.8
East Asia and Pacific 6.4 3.6 5.1
China 7.0 6.4
Korea, Rep. 7.9 1.7
Indonesia 3.5 2.1
Europe and Central Asia 6.1 1.9 3.3
Russian Federation 8.6 5.0
Turkey 5.5 -8.7
Poland 4.1 1.4
Latin America and the Caribbean 2.2 -0.7 2.1
Brazil 3.0 0.2
Mexico 5.2 -1.3
Argentina -1.7 -3.2
MidldRe East anmd North Arica 1.9 1.5 1.4
Saudi Arabia 0.7 -1.5
Iran 3.5 2.5
Egypt 3.5 2.7
Sub-Saharan Africa 0.5 0.3 1.3
South Africa 1.4 1.0
Nigeria 0.4 0.3
High-income Economies
Industrial
G-7 2.7 0.3 2.1
US 3.2 0.3 2.0
Japan 1.5 -0.9 2.0
G-4 Europe 3 1.3 2.3
GerTnany 3.1 0.7 2.1
Euro Area 3.3 1.4 2.5
Non-G7 Industrial 3.5 1.6 2.7
Other High-income 4.7 -0.7 2.8
Asian NEEs (KOR, HKG, SGP) 6.4 -0.7 3.4
Source: World Bank Global Economic Prospects 2002.
42
Policy Research Working Paper Series
Contact
Title Author Date for paper
WPS3011 Renegotiation of Concession J Luis Guasch April 2003 J Troncoso
Contracts in Latin America Jean-Jacques Laffont 37826
St6phane Straub
WPS3012 Just-in-Case lnventories A Cross- J Luis Guasch April 2003 J Troncoso
Country Analysis Joseph Kogan 37826
WPS3013 Land Sales and Rental Markets in Klaus Deininger April 2003 M Fernandez
Transition Evidence from Rural Songqing Jin 33766
Vietnam
WPS3014 Evaluation of Financial Liberalization: Xavier Gine April 2003 K Labrie
A General Equilibrium Model with Robert M Townsend 31001
Constrained Occupation Choice
WPS3015 Off and Running9 Technology, Trade, Carolina S6nchez-PAramo April 2003 H Sladovich
and the Rising Demand for Skilled Norbert Schady 37698
Workers in Latin America
WPS3016 Partisan Politics and Stuti Khemani April 2003 H Sladovich
Intergovernmental Transfers in India 37698
WPS3017 Why is Unemployment so High Jan Rutkowski April 2003 J Rutkowski
in Bulgaria? 84569
WPS3018 Integrating the Least Developed Paul Brenton Apnl 2003 L Tabada
Countries into the World Trading 36896
System The Current Impact of EU
Preferences under Everything but Arms
WPS3019 Foreign Aid, Conditionality, and Ghost Thilak Ranaweera Apnl 2003 P Rathan Raj
of the Financing Gap: A Forgotten 33705
Aspect of the Aid Debate
WPS3020 Consumption, Health, Gender, Anne Case April 2003 D Ballantyne
and Poverty Angus Deaton 87198
WPS3021 Institutional Arrangements for Public Elizabeth Currie April 2003 M Rosenquist
Debt Management Jean-Jacques Dethier 82602
Eriko Togo
WPS3022 Sending Farmers Back to School Gershon Feder April 2003 P Kokila
The Impact of Farmer Field Schools Rinku Murgai 33716
in Indonesia Jaime B Quizon
WPS3023 Productivity Growth and Product Douglas M Addison April 2003 D Addison
Variety Gains from Imitation and 31188
Education
WPS3024 R&D and Development Daniel Lederman April 2003 P Soto
William F Maloney 37892
WPS3025 Trade Structure and Growth Daniel Lederman April 2003 P Soto
William F Maloney 37892
WPS3026 Strategic Approaches to Science Robert Watson April 2003 R. Watson
and Technology in Development Michael Crawford 36965
Sara Farley
Policy Research Working Paper Series
Contact
Title Author Date for paper
WPS3027 Financial Intermediation and Growth: Genevieve Boyreau- April 2003 P Sintim-Aboagye
Chinese Style Debray 38526
WPS3028 Does a Country Need a Promotion Jacques Morisset April 2003 M. Feghali
Agency to Attract Foreign Direct 36177
Investment? A Small Analytical Model
Applied to 58 Countries
WPS3029 Who Benefits and How Much9 How Alessandro Nicita Apnl 2003 P. Flewitt
Gender Affects Welfare Impacts of a Susan Razzaz 32724
Booming Textile Industry
WPS3030 The Impact of Bank Regulations, Asli Demirgu,-Kunt April 2003 A. Yaptenco
Concentration, and Institutions on Luc Laeven 31823
Bank Margins Ross Levine
WPS3031 Imports, Entry, and Competition Law Hiau Looi Kee April 2003 P Flewitt
as Market Disciplines Bernard Hoekman 32724
WPS3032 Information Diffusion in International Alejandro lzquierdo April 2003 M. Feghali
Markets Jacques Morisset 36177
Marcelo Olarreaga
WPS3033 The Role of Occupational Pension Dimitri Vittas Apnl 2003 P. Infante
Funds in Mauritius 37642
WPS3034 The Insurance Industry in Dimitri Vittas April 2003 P Infante
Mauritius 37642