Gasoline prices and road fatalities: international evidence.
Burke, Paul J. ; Nishitateno, Shuhei
I. INTRODUCTION
While a negative impact of gasoline prices on road fatalities has
been documented for the United States, there is limited international
evidence on whether road death rates are higher in countries with lower
gasoline prices. In this study, we employ data for a panel of 144
countries during 1991-2010 to present international estimates of the
gasoline price elasticity of road fatalities. To address the potential
endogeneity of gasoline prices, we use each country's underground
oil reserves and the international crude oil price as instruments for
that country's gasoline price. We find that the mean long-run
gasoline price elasticity of road deaths is in the order of -0.3 to -0.6
and that around 35,000 lives could be saved on roads each year by
phasing out global fuel subsidies. We also use our results to estimate
the number of deaths that could be avoided on U.S. roads by an increase
in fuel taxes.
Road safety is a leading public health issue. Road crashes are the
cause of 1.3 million deaths every year; the ninth-leading cause of death
globally and the number-one cause of death for people between 15 and 29
years of age (data for 2011; World Health Organization 2013a). Road
death rates are particularly high in middle- and low-income countries,
which each year see an average of 20 and 18 deaths per 100,000
population, respectively. There are around 9 road deaths per 100,000
population each year in high-income countries. Up to 50 million people
worldwide also suffer nonfatal injuries each year, bringing large human
and financial costs. The global road death toll is expected to increase
to around 2.4 million per year by 2030 in a business-as-usual scenario,
making road crashes the fifth-leading cause of death (World Health
Organization 2013b). Finding ways to reduce global road deaths is an
increasingly important policy imperative.
A negative relationship between the gasoline pump price (in U.S.
cents) and annual road deaths per 100,000 population for a large cross
section of countries in 2010 is presented in Figure 1. Countries with
low gasoline prices, such as Venezuela and Iran, have among the highest
road death rates, whereas road fatalities tend to be less frequent in
high-price countries. Figure 1 also demonstrates substantial variation
in road fatality rates among countries with similar gasoline prices. A
number of additional variables, including per capita incomes, road-use
laws, and road infrastructure, will be considered in explaining this
variation.
[FIGURE 1 OMITTED]
The large cross-country variation in retail prices for gasoline--a
tradable commodity--exists primarily because of differences in tax and
subsidy policies. Venezuela had the lowest average price of gasoline in
2010: just 2 U.S. cents per liter. Venezuela's gasoline price is
substantially below the international price for crude oil (51 cents per
liter in 2010; GIZ 2012), and so involves a large subsidy for consumers.
In contrast, some governments impose high taxes on gasoline that result
in high retail pump prices. In Turkey, for instance, the average
gasoline pump price was 252 U.S. cents per liter in 2010, including 139
cents of taxes (International Energy Agency 2013a).
There are several ways in which higher gasoline prices may reduce
road deaths. It is likely that the principal channel is a reduction in
the distance traveled in motor vehicles as people respond to the
incentive to substitute away from a more expensive commodity. Reduced
driving decreases the exposure of both vehicle occupants and others to
road crashes. Reductions in distance traveled may be a result of people
transitioning to less transport-intensive activities, alternative
transport options, and closer workplaces. Some of these responses take
time, so the long-run gasoline price elasticity of road deaths is likely
to exceed the short-run elasticity.
In addition to reducing distance traveled, higher gasoline prices
might also lead to a reduction in road deaths per kilometer driven. One
reason is that, to conserve fuel, drivers might reduce high-speed
driving and also their rates of acceleration and braking. (1) Another is
that high-risk drivers, including the young, the old, and those taking
leisure-related trips, are particularly sensitive to gasoline prices
(Cullotta 2008; Grabowski and Morrisey 2004; Morrisey and Grabowski
2011). Higher gasoline prices also result in substitution from heavier
to lighter, more fuel-efficient, private vehicles (e.g., light trucks to
automobiles), and lighter vehicles are associated with a lower overall
number of road deaths per kilometer traveled (Gayer 2004; White 2004).
Substitution to bus travel may also reduce overall road safety risks.
There are ways in which higher gasoline prices might actually lead
to more rather than fewer road deaths. By reducing congestion, higher
gasoline prices can allow remaining drivers to travel at faster speeds
(Burger and Kaffine 2009), which increases the risk of fatal crashes.
Substitution to particularly risky types of fuel-efficient vehicles,
such as motorcycles, may also cause additional road deaths when gasoline
prices rise (Hyatt et al. 2009; Wilson, Stimpson, and Hilsenrath 2009).
(2)
Existing evidence for the United States indicates that higher
gasoline prices reduce road fatalities and/or crashes (Chi, McClure, and
Brown 2012; Chi et al. 2010, 2011, 2013a, 2013b; Grabowski and Morrisey
2004, 2006; Haughton and Sarkar 1996; Leigh and Wilkinson 1991; Montour
2011; Sivak 2009). As for gasoline demand itself, the response of road
deaths to gasoline prices in the United States is inelastic. Grabowski
and Morrisey (2004), for instance, use data for 48 U.S. states for the
period 1983-2000 and find a gasoline price elasticity of road fatalities
of -0.3 when responses over a 2-year period are considered.
Most international studies on the determinants of road fatalities
(e.g., Anwaar et al. 2012; Noland 2005; Page 2001) concentrate on other
issues, although Litman (2012) presents a scatterplot for 16
Organization for Economic Co-operation and Development (OECD) countries
that shows a negative association between average gasoline prices and
traffic fatality rates. As far as we are aware, there has been no prior
international estimate of the gasoline price elasticity of road deaths.
II. APPROACH
We estimate the following specification:
(1) ln [D.sub.c,t] = [alpha] + [[beta].sub.1] ln [G.sub.c,t] +
[[beta].sub.2] ln [Y.sub.c,t], + [[beta].sub.3] ln [P.sub.c,t] + [gamma]
[X.sub.c,t] + [[delta].sub.c] + [[omega].sub.t] + [[epsilon].sub.c,t]
where D is road deaths in country c in year r, G is the gasoline
price in year-2010 U.S. cents, Y is gross domestic product (GDP) in real
purchasing power parity-adjusted U.S. dollars, P is population, and X is
a vector of additional controls included in later estimations.
[[delta].sub.c] and [[omega].sub.t] are country and year fixed effects,
and e is an error term. We also present specifications that control for
country-specific time trends.
Our primary interest is in identifying the long-run gasoline price
elasticity of road deaths. (3) To this end, we initially estimate
Equation (1) for a cross section of countries in the year 2010, as
so-called between variation has a natural long-run interpretation. We
then proceed to panel estimates using the pooled ordinary least squares
(OLS), between, and fixed-effects estimators. The between estimator uses
average data for each country and provides estimates of long-run effects
(Baltagi 2008; Baltagi and Griffin 1983, 1984; Pesaran and Smith 1995;
Pirotte 1999, 2003; Stem 2010). Fixed-effects estimations control for
time-invariant factors such as the extent of mountainous terrain, but
when a static fixed-effects equation is estimated the coefficients
represent shorter-run effects. To explore the dynamics of the response
to higher gasoline prices and for a further estimate of the long-run
gasoline price elasticity of road deaths, we then table results for
distributed lag specifications (with country fixed effects). For checks
on the importance of functional form, we also estimate negative binomial
models (with and without country fixed effects) and models using per
capita measures of road deaths and GDP. (4)
An issue of concern is that the gasoline price term in Equation (1)
may be correlated with the error. The level of demand for road transport
might have a material effect on a country's average gasoline pump
price, for instance, while also affecting road deaths (Grabowski and
Morrisey 2004, 2006; Morrisey and Grabowski 2011). Alternatively,
governments may impose higher gasoline taxes in countries with low
demand for road use, as argued by Hammar, Lofgren, and Sterner (2004).
It is also possible that the set of factors affecting gasoline
tax/subsidy policies might include road safety concerns. Perhaps even
more importantly, there might be omitted policy variables that are
associated with gasoline prices: interventionist governments might tax
gasoline and have strict road rules, for example. These concerns mean
that we cannot be sure that single-equation estimation of Equation (1)
will produce unbiased and consistent estimates of the effect of gasoline
prices on road deaths.
To address the potential endogeneity of gasoline prices, we present
estimates using the two supply-side instruments for the gasoline price
employed in our recent study of the gasoline price elasticity of demand
(Burke and Nishitateno 2013). The first is a country's per capita
underground oil reserves, as oil-rich countries such as Venezuela are
more likely to subsidize gasoline and oil-poor countries such as the
Republic of Korea are more likely to tax it. The second is a measure of
the annual average international crude oil price, as higher crude oil
prices flow through to higher gasoline pump prices in most countries. It
is likely that our instruments affect road deaths via gasoline prices
rather than other channels.
The advantage of using two instruments is that doing so allows
verification of the effect of gasoline prices on road deaths using
independent sources of variation in gasoline prices. The exclusion
restrictions are that oil reserves and the global oil price are not
correlated with unobserved determinants of road deaths (across countries
and over time, respectively). In addition to using the instruments
separately, we also present estimates using both instruments together.
Existing studies on the effect of gasoline prices on road deaths in the
United States do not use instrumental variable (IV) approaches. (5)
Our estimates are for a panel of 144 countries for 1991, 1993,
1995, 1998, 2000, 2002, 2004, 2006, 2008, and 2010: ten years for which
average gasoline price data are available from the November surveys of
GIZ (2012). Our data on road fatalities are primarily from the
International Road Federation (2012) and include all reported deaths
that occur within 30 days of a road crash. Alternative estimates of road
deaths in 2010 from the WHO (2013b), as used in Figure 1, provide
similar results. We focus on fatalities because international data on
nonfatal road crashes are less reliable (Luoma and Sivak 2007;
Sauerzapf, Jones, and Haynes 2010; WHO 2013b). Nevertheless, the
accuracy of the data on road deaths is likely to vary. If the reporting
of road deaths improves in a way that is correlated with economic
development, our GDP variable will control for some data quality
differences. Because of missing data, on average each country is
included in our sample for 5.8 of the 10 years. The countries in our
sample represented 94% of the world's population in 2010. A full
list of data sources is in the Appendix.
III. RESULTS
A. Main Specifications
Results for single-equation specifications, controlling for GDP and
population, are in Table 1. Column 1 is for a year-2010 cross section of
countries and indicates that a 1% higher gasoline price on average
reduces road fatalities by 0.4%. Cross-sectional estimates utilize only
between variation and so this is a first estimate of the long-run effect
of gasoline prices on road deaths. Columns 2 and 3 present results using
the pooled OLS (with year dummies) and between estimators. The point
estimates of the gasoline price elasticity of road deaths are slightly
smaller (-0.3), but remain distinguishable from zero at the 1%
significance level.
Column 4 of Table 1 controls for both year and country fixed
effects, which removes most of the variation in gasoline prices in our
sample. (6) This makes it difficult for within-country gasoline price
movements to affect road deaths. Static models relying on only within
variation are also likely to provide short-run effects because of the
underspecification of dynamics (Baltagi 2008). Likely as a result of
these factors, the fixed-effects estimate in Column 4 is insignificant.
We obtain a significant coefficient in the fixed-effects specification
in Column 5 that includes a linear time trend for each country in the
sample.
Table 2 shows our IV results. Our cross-sectional and panel
estimates instrumenting with per capita oil reserves (Columns 1-2)
indicate that higher gasoline prices significantly reduce road deaths,
with the cross-sectional estimate providing an elasticity of -0.3. We
are prevented from controlling for country fixed effects when
instrumenting with oil reserves per capita as there is almost no useful
time-series variation in per capita oil reserves. Columns 3 and 4
instrument with the log real international crude oil price.
Country-specific linear time trends are included instead of year
dummies, as year dummies would be perfectly collinear with our
instrument. (7) The gasoline price elasticities in these estimates are
-0.5 and -0.4. It is reassuring that we obtain similar results using
different instruments. Column 5 uses both instruments and obtains a
gasoline price elasticity of road deaths of -0.5.
Column 6 of Table 2 uses both instruments and controls for the full
set of variables that will be used in Table 4. The results suggest a
stronger negative effect of gasoline prices on road deaths, with an
elasticity of -0.9. Given the smaller sample size, weaker first-stage
identification, and the possibility that some of the controls could
themselves be endogenous, we do not include the Column 6 estimate in our
"headline" results. The robustness of our IV estimates to the
addition of controls such as road infrastructure variables does,
however, reduce the concern that the result is driven by a violation of
the IV exclusion restriction.
The tests of Stock and Yogo (2005) indicate that our instruments
provide adequate identification strength. Specifically, the null
hypothesis of 15% maximal IV size is rejected in each of the IV
specifications. The first-stage coefficients, as expected, indicate that
oil reserves and the oil price are negatively and positively correlated
with the gasoline price, respectively. (8) Overidentification tests in
Columns 5-6 do not reject the null hypothesis that the instruments are
valid. The IV results instrumenting with oil reserves per capita
(Columns 1-2) are similar to the single-equation estimates, meaning that
endogeneity tests fail to reject the null that the log gasoline price is
exogenous. (9) In contrast, endogeneity tests in Columns 3-6 suggest
that there is due cause to treat gasoline prices as endogenous.
The estimated coefficients for the control variables in Tables 1
and 2 are of interest. As expected, countries with larger populations
typically have more road deaths, which is merely a scale effect. The
panel results indicate that countries with larger economies also on
average have more road fatalities, presumably because more people can
afford private road vehicle travel. The income elasticities are smaller
than the income elasticities of gasoline consumption of around +1.0
obtained by Burke and Nishitateno (2013), likely because richer
countries dedicate more resources to improving road safety. A nonlinear
relationship between GDP per capita and road deaths will be considered
in coming specifications.
Table 3 shows distributed-lag estimates with country and year fixed
effects. Because these rely solely on within variation, we commence the
gasoline price terms from year t - 1 to allow time for responses to
November prices. As a result, our estimation sample here extends to 2009
rather than 2010. Lags are included for every second year, given the
biennial nature of GIZ's gasoline price data, and the sample
reduces with each additional lag. The long-run gasoline price elasticity
is the sum of the coefficients for each gasoline price term.
The results in Table 3 provide an estimate of the long-run gasoline
price elasticity of -0.6 when lags to year t - 9 are considered
(significant at the 10% level). Similar, and statistically stronger,
long-run multipliers are obtained in pooled OLS and between estimates of
distributed lag models. We also find generally similar estimates in
specifications with country-specific time trends (see base of Table 3).
We cannot rule out that even larger elasticities may be obtained from
distributed lag models once longer time-series are available.
Based on our single-equation and IV estimates using between
variation in Tables 1 and 2 and our estimates using distributed lags in
Table 3, we conclude that the average long-run gasoline price elasticity
of road deaths is likely in the order of -0.3 to -0.6. This is an
inelastic response, meaning that higher gasoline prices do reduce road
deaths, but in a less-than-proportionate manner. In earlier work (Burke
and Nishitateno 2013) we obtained similar estimates of the long-run
gasoline price elasticity of demand, suggesting that the effect of
gasoline prices on road deaths is primarily related to the relationship
between gasoline prices and the propensity for road travel. We do not
have sufficient data for our international sample to decompose the
effects of gasoline prices on road deaths into specific channels such as
distance traveled or travel speeds, although such research would be of
interest when data permit. (10)
B. Robustness
In some countries, a large share of the vehicle fleet runs on
diesel rather than gasoline. Table 4 presents estimates using the
average of the gasoline and diesel prices and controlling for additional
variables: land area, the length of each country's road network,
the share of roads that is paved, the vehicle and motorcycle stocks,
measures of the importance of rail and air transport, the share of the
population aged 15-24 (who are typically overrepresented in road
crashes), the urban population share, alcohol consumption, blood alcohol
limits for drivers, the maximum speed in urban areas, measures of the
rule of law and control of corruption, economic growth, and infant
mortality. Controlling for road infrastructure variables helps address
the concern that our main result operates via the additional road
funding that is possible when gasoline taxes are high. The log infant
mortality rate is included as a proxy of overall health conditions in
each country (noting that few infant deaths are caused by road crashes).
We show between estimates for static models given our desires to
estimate long-run effects and maximize sample size.
The results in Table 4 provide fuel price elasticities of road
deaths of -0.3 to -0.45 (significant at the 1% level) and suggest that
fuel prices are one of the most statistically robust cross-country
determinants of road death rates. Countries with better control of
corruption, higher dependence on air travel, lower alcohol consumption,
and stricter speed laws have fewer road deaths. Interestingly, countries
with better "rule of law" ratings have more road deaths
(holding all other variables, including corruption, constant), perhaps
because reporting of road deaths is more complete. Once the full set of
other variables has been controlled for, we find no evidence that the
number of motor vehicles in each country is a strong predictor of road
deaths. (11)
The effect of gasoline prices on road deaths may operate via some
of the control variables in Table 4, including GDP. We obtain similar
estimates for the gasoline price term if the controls are lagged,
however. Pooled OLS estimates also provide similar results (although
with slightly smaller point estimates of the gasoline price elasticity
of road deaths). Because we control for log real GDP, our coefficient
estimates for (3! are identical if our gasoline price measure is scaled
by real GDP. While our list of controls in Table 4 is long, there are
many other factors that influence road safety. Current data constraints
for our large international sample mean that we leave the task of
including additional control variables to future research, perhaps for a
smaller set of countries. (12)
Table 5 presents results using per capita measures of road deaths
and GDP, and controlling for population density instead of population.
The table also shows estimates for subsamples of OECD and non-OECD
countries and an estimate controlling for regional dummies. The results
are similar to those in Table 1, confirming that it makes little
difference if variables are in total or per capita terms. (13) Column 2
controls for the square of log GDP per capita to account for the road
death Kuznets curve (see, for instance, Law, Noland, and Evans 2011).
The results suggest that the road death rate typically increases until a
mid-range GDP per capita and subsequently falls (holding other factors
constant). We continue to observe a negative and statistically
significant gasoline price elasticity of road deaths.
Column 3 of Table 5 includes the squared log gasoline price to test
whether the gasoline price elasticity of road deaths varies at different
gasoline price levels. Road deaths appear to be more responsive to
changes in the gasoline price when the price is already high. The
estimated gasoline price elasticity of road deaths at the
25th-percentile gasoline price is -0.5, increasing to -0.8 at the 75th
percentile. (14) Column 4 includes an interaction between the log
gasoline price and log GDP per capita. The estimate provides no evidence
that the gasoline price elasticity of road deaths varies by development
level. We also obtain statistically significant estimates of the effect
of gasoline prices on road deaths for subsets of OECD and non-OECD
countries (Columns 5 and 6). Data on road fatalities are likely to be
more reliable for OECD countries, and so the gasoline price elasticity
of road deaths of -0.5 for the OECD subsample increases our confidence
in the main results. Column 7 controls for regional dummy variables,
which allows some time-invariant regional-specific characteristics such
as driving culture to be considered. The results are similar.
Table 6 presents cross-sectional, pooled, and fixed-effect negative
binomial estimates. Negative binomial models are suited to a
count-dependent variable and are preferred over Poisson models because
our road death data exhibit overdispersion (variance exceeds the mean).
As in Table 5, we use road deaths weighted by population (now in
unlogged form). The coefficients for the gasoline price, which can again
be interpreted as elasticities, are significantly different from zero,
and range from -0.4 (using between variation in the cross-sectional
estimate) to -0.2 (using static within variation, and so likely
representing a shorter-run effect). An unreported fixed-effect negative
binomial model with additional lags provides a long-run gasoline price
elasticity of road deaths of -0.6 (significant at the 5% level). In an
additional check--available on request--we also estimated an IV negative
binomial model, obtaining similar results to our linear IV estimates
(but for which weak instrument test and other information is not
available). In short, results using negative binominal models fall
within our reported range.
A lingering concern may be that there are additional time-varying
policies affecting road deaths that are not possible to control for and
may be correlated with gasoline prices. While our specifications have
relatively high [R.sup.2] values, there are clearly other variables
(such as road safety advertising campaigns) that are likely to affect
road deaths. It is important to note, however, that omitted variables
could only be causing a serious identification problem across our full
suite of estimates if they are correlated with gasoline prices (in our
single-equation estimates) and both of our instruments (in our various
IV estimates). This is unlikely. The world oil price is unlikely to be
affected by or have any short-term influence on road safety policies,
for instance. Our IV strategy, together with our use of numerous
controls (alcohol consumption; country fixed effects; country-by-country
time trends, regional dummies; etc.), makes us confident that our
results represent consistent estimates of the causal effect of gasoline
prices on road deaths.
It is important to explicitly note that many other factors,
including those that we have not been able to represent in our
estimations, also have important influences on road death rates. There
are many, sad, stories behind individual road crashes. There is also
substantial evidence that specific interventions such as helmet laws can
reduce road death rates (e.g., Passmore et al. 2010). The results in
this article do not challenge this evidence. Instead, the results
provide macro-level guidance on one economic variable--the price of
gasoline--that has a macro-level effect on road deaths. This variable is
amenable to policy.
IV. ESTIMATING THE NUMBER OF AVOIDED DEATHS FROM FUEL PRICE REFORM
Some countries, particularly the oil-rich, provide large price
subsidies to consumers of gasoline. Table 7 presents estimates of the
number of road deaths that could be avoided if countries with gasoline
prices lower than those in the United States (76 cents per liter in
2010) increased their average gasoline price to the U.S. level. The
estimates are based on a conservative long-run gasoline price elasticity
of road deaths of -0.4 (e.g., Column 1 of Table 1). Like GIZ (2012), we
consider the gasoline price in the United States--the lowest of all OECD
countries--as the divider between countries that subsidize gasoline
consumption and the rest. While the United States does apply state and
federal taxes on gasoline, these could be considered to be the minimum
required to adequately cover road infrastructure and externality costs
(GIZ 2012). There are alternative approaches that could be used to
measure the size of fuel subsidies (e.g., Davis 2014).
The results in Table 7 suggest that around 35,000 lives per annum
could be saved in 23 countries by removing the subsidies that were in
place in 2010 (15); 35,000 lives is 3% of the global annual road death
toll. The countries in which fuel subsidy reform offers the largest
potential reductions in road deaths are Iran (10,600 avoided deaths per
year) and Venezuela (>5,000 avoided deaths per year). The removal of
fuel subsidies would also result in large reductions in road deaths in
Indonesia (4,500), Nigeria (4,200), Saudi Arabia (2,700), Egypt (1,800),
and Algeria (1,700). (16)
How many road deaths could be avoided if the United States itself
had higher taxes on gasoline? A simulation using our results implies
that around 10,000 lives per year could be saved if U.S. gasoline taxes
were increased to bring the U.S. gasoline price to the UK level (192
cents per liter in 2010). This would reduce U.S. road fatalities by more
than a quarter. A reduction of this magnitude is not historically
infeasible: the number of annual road deaths in the United States fell
by 9,600 (17%) as gasoline prices spiked during the years 1973-1975, for
instance (Leigh and Geraghty 2008). Annual road deaths in the United
States also reduced by 9,800 between 2006 and 2010 as gasoline prices
increased and the economy entered recession (and due to other factors;
Sivak and Schoettle 2010).
How many more road deaths would occur if countries that currently
have high gasoline taxes move down to U.S.-level gasoline prices? The
case of the United Kingdom is illustrative. Our estimates indicate that
the United Kingdom would have around 1,800 additional road deaths per
year if it had U.S.-level gasoline prices, a 95% increase over current
levels. The United Kingdom's road death toll has been falling over
recent years; our estimates indicate that moving to U.S.-level prices
would return the country to a circa-1995 road death toll.
We ask one final question: How large a role has increases in real
gasoline prices played in the reductions in road deaths that have been
achieved in most developed countries? The answer is that higher gasoline
prices have had a material effect in reducing road deaths, but are
typically not the majority of the story. This results from the relative
inelasticity of road deaths to gasoline prices. An example will help.
During 2002-2010, France's annual road death toll fell from around
7,700 to around 4,000. During this period, real gasoline pump prices in
France increased by 57%. Our estimates suggest that the contribution of
this price increase to the reduction in France's road death toll is
about 800 annual road deaths or around 20%. Other factors explain the
majority of the reduction in road deaths in France in recent years.
Similar is true of most other developed countries. (17)
V. CONCLUSION
This study has utilized the substantial variation in international
gasoline pump prices to examine the effect of gasoline prices on the
number of people dying in road crashes. Our results indicate that higher
gasoline prices significantly reduce road deaths, with our point
estimates of the mean long-run gasoline price elasticity of road deaths
lying between -0.3 and -0.6. The effect is an inelastic one, as also
obtained in studies of the United States (e.g., Grabowski and Morrisey
2004).
The international community is mobilizing a number of strategies to
improve road safety during the United Nations' Decade of Action for
Road Safety 2011-2020. The Global Plan for the Decade of Action (WHO
2010) is silent on the potential role of fuel pricing. Reductions in
fuel subsidies and increases in fuel taxes could, however, make a large
contribution to the plan's objectives. Countries providing the
largest fuel subsidies are particularly compelling candidates for
reform. Globally, around 35,000 road deaths could be avoided each year
by the removal of the fuel subsidies that were in place in 2010.
Finally, there are likely to be large changes in road transport
over coming decades. Moving toward non-oil powered vehicles may involve
a reduction in the marginal cost of driving. If so, our results suggest
that this could feed into higher road death rates. At the same time,
however, vehicle safety will continue improving. The economic and other
factors affecting road death rates will remain a stimulating field of
research.
ABBREVIATIONS
BAC: Blood Alcohol Concentration
GDP: Gross Domestic Product
IEA: International Energy Agency
IRF: International Road Federation
IV : Instrumental Variable
OECD: Organization for Economic Co-operation and Development
OLS: Ordinary Least Squares
WHO: World Health Organization
doi: 10.1111/ecin. 12171
Online Early publication November 9, 2014
APPENDIX: VARIABLE DESCRIPTIONS
Road deaths: Number of reported deaths that occur within 30 days of
a road crash. Includes all deaths (e.g., of vehicle occupants,
motorcyclists, cyclists, and pedestrians; International Road Federation
2012). Data for nine countries were supplemented with figures from the
Organization for Economic Co-operation and Development (2013a) and the
United Nations Economic Commission for Europe (2013).
Gasoline price: Average retail gasoline pump price in year-2010
U.S. cents per liter. Prices were collected by GIZ (2012) in
mid-November surveys. Data are for unleaded octane 95 gasoline.
Indonesia's price is for subsidized gasoline. The U.S. GDP deflator
from the World Bank (2013a) was used to deflate prices.
GDP: Expenditure-side real GDP at chained purchasing power
parities, in 2005 US$ (Feenstra, Inklaar, and Timmer 2013).
Population: Total population, in people (Feenstra, Inklaar, and
Timmer 2013).
Oil reserves per capita: Proved underground reserves of crude oil,
thousand tons per capita (U.S. Energy Information Administration 2011).
One year's lag or lead used for a small number of missing
observations.
Real world oil price: Average cost of total crude imports of the
members of the IEA in year-2010 U.S. dollars per barrel (IEA 2013a). The
U.S. GDP deflator from the World Bank (2013a) was used to deflate
prices.
Average gasoline and diesel price: Simple average of the gasoline
and diesel retail pump prices in year-2010 U.S. cents per liter. Prices
were collected by GIZ (2012) in mid-November surveys. Gasoline prices
are for unleaded octane 95 gasoline. Indonesia's gasoline price is
for subsidized gasoline. The U.S. GDP deflator from the World Bank
(2013a) was used to deflate prices.
Land area: A country's total land area, excluding inland water
bodies, national claims to continental shelf, and exclusive economic
zones, in square kilometers (World Bank 2013a).
Road distance: Length of the total road network, in kilometers
(International Road Federation 2012). Data linearly interpolated.
Paved road share (%): Percentage of road length that is surfaced
with crushed stone, hydrocarbon binder, bituminized agents, concrete, or
cobblestones (International Road Federation 2012). Data linearly
interpolated.
Motor vehicle stock (4+ wheels): Number of motor vehicles with four
or more wheels. Includes cars, buses, lorries, and vans (International
Road Federation 2012). Several apparent errors were removed. Data
linearly interpolated.
Motorcycle stock: Two- or three-wheeled road motor vehicles
(International Road Federation 2012). Several apparent errors were
removed. Data linearly interpolated.
Rail share of energy used in transport (%): Percentage share of
rail sector's energy use in total energy used in the road, rail,
and domestic aviation sectors (IEA 2013b).
Air passengers: Domestic and international passengers of air
carriers registered in the country (World Bank 2013a).
Population aged 15-24 (%): Percentage of population aged 15-24.
Five-yearly United Nations (2010) data on the 15- to 24-year-old
population were linearly interpolated. Data on total population from
Feenstra, Inklaar, and Timmer (2013).
Urban population (%): People living in urban areas as defined by
national statistical offices as a share of the total population (World
Bank 2013a).
Alcohol consumption per adult: Annual alcohol consumption (in
liters of pure alcohol) per adult (age 15+) (WHO 2013c).
Blood alcohol limit for drivers in 2011: Legal blood alcohol
concentration (BAC) for general drivers in 2011 (or nearby year),
expressed as a percentage. This variable is missing for countries with
no limit (which only slightly reduces the sample). This variable is not
time varying (WHO 2013d).
Maximum speed in urban areas in 2011: Maximum speed limit for cars
on residential roads in 2011, in kilometers per hour. This variable is
not time varying (WHO 2013d).
Rule of law score: A measure of perceptions of the extent to which
agents have confidence in and abide by the rules of society, and in
particular the quality of contract enforcement, property rights, the
police, and the courts, as well as the likelihood of crime and violence.
Approximate possible range is -2.5 (worst) to 2.5 (best) (World Bank
2013b).
Control of corruption score: A measure of perceptions of the extent
to which public power is exercised for private gain, including both
petty and grand forms of corruption, as well as capture of the state by
elites and private interests. Approximate possible range is -2.5 (worst
control of corruption) to 2.5 (best) (World Bank 2013b).
Economic growth rate (%): Annual percentage change in
expenditure-side real GDP at chained purchasing power parities, in 2005
US$ (Feenstra, Inklaar, and Timmer 2013).
Infant mortality rate: Number of infants dying before reaching 1
year of age, per 1,000 live births (World Bank 2013a).
Population density: Population per squared kilometer of land area
(World Bank 2013a).
REFERENCES
Anwaar, A., P. Anastasopoulos, G. P. Ong, S. Labi, and M. B. Islam.
"Factors Affecting Highway Safety, Health Care Services, and
Motorization--An Exploratory Empirical Analysis Using Aggregate
Data." Journal of Transportation Safety & Security, 4(2), 2012,
94-115.
Baltagi, B. H. Econometric Analysis of Panel Data. 4th ed.
Chichester, UK: John Wiley & Sons, 2008.
Baltagi, B. H., and J. M. Griffin. "Gasoline Demand in the
OECD: An Application of Pooling and Testing Procedures." European
Economic Review, 22(2), 1983, 117-37.
--. "Short and Long Run Effects in Pooled Models."
International Economic Review, 25(3), 1984, 631-45.
Burger, N. E., and D. T. Kaffine. "Gas Prices, Traffic, and
Freeway Speeds in Los Angeles." Review of Economics and Statistics,
91(3), 2009, 652-57.
Burke, P. J., and S. Nishitateno. "Gasoline Prices, Gasoline
Consumption, and New-Vehicle Fuel Economy: Evidence for a Large Sample
of Countries." Energy Economics, 36, 2013, 363-70.
Chi, G., A. G. Cosby, M. A. Quddus, P. A. Gilbert, and D. Levinson.
"Gasoline Prices and Traffic Safety in Mississippi." Journal
of Safety Research, 41(6), 2010, 493-500.
Chi, G., X. Zhou, T. E. McClure, P. A. Gilbert, A. G. Cosby, L.
Zhang, A. A. Robertson, and D. Levinson. "Gasoline Prices and Their
Relationship to Drunk-Driving Crashes." Accident Analysis and
Prevention, 43(1), 2011, 194-203.
Chi, G., T. E. McClure, and D. B. Brown. "Gasoline Prices and
Traffic Crashes in Alabama, 1999-2009." Traffic Injury Prevention,
13(5), 2012, 476-84.
Chi, G., J. R. Porter, A. G. Cosby, and D. Levinson. "The
Impact of Gasoline Price Changes on Traffic Safety: A Time Geography
Explanation." Journal of Transport Geography, 28, 2013a, 1-11.
Chi, G., M. A. Quddus, A. Huang, and D. Levinson. "Gasoline
Price Effects on Traffic Safety in Urban and Rural Areas: Evidence from
Minnesota, 1998-2007." Safety Science, 59(1), 2013b, 154-62.
Congressional Budget Office. Effects of Gasoline Prices on Driving
Behavior and Vehicle Markets. Washington, DC: Congressional Budget
Office, 2008.
Cullotta, K. A. "As Gas Prices Rise, Teenagers' Cruising
Declines." New York Times, June 29, 2008.
Davis, L. W. "The Economic Cost of Global Fuel
Subsidies." American Economics Review: Papers and Proceedings,
104(5), 2014, 581-5.
Feenstra, R. C., R. Inklaar, and M. P. Timmer. "The Next
Generation of the Penn World Table." Version 8.0,2013. Accessed
October 13, 2014. http://www.ggdc.net/pwt.
Gayer, T. "The Fatality Risks of Sport-Utility Vehicles, Vans,
and Pickups Relative to Cars." Journal of Risk and Uncertainty,
28(2), 2004, 103-33.
GIZ. International Fuel Prices 2010/2011. 7th ed. Bonn and
Eschborn, Germany: GIZ, 2012.
Grabowski, D. C., and M. A. Morrisey. "Gasoline Prices and
Motor Vehicle Fatalities." Journal of Policy Analysis and
Management, 23(3), 2004, 575-93.
--. "Do Higher Gasoline Taxes Save Lives?" Economics
Letters, 90(1), 2006, 51-55.
Hammar, H., A. Lofgren, and T. Sterner. "Political Economy
Obstacles to Fuel Taxation." Energy Journal, 25(3), 2004, 1-17.
Haughton, J., and S. Sarkar. "Gasoline Tax as a Corrective
Tax: Estimates for the United States, 1970-1991." Energy Journal,
17(2), 1996, 103-26.
Hyatt, E., R. Griffin, L. W. Rue III, and G. McGwin Jr. "The
Association between Price of Regular-Grade Gasoline and Injury and
Mortality Rates among Occupants Involved in Motorcycle- and
Automobile-Related Motor Vehicle Collisions." Accident Analysis and
Prevention, 41(5), 2009, 1075-79.
International Energy Agency. Energy Prices and Taxes. Paris:
International Energy Agency, 2013.
--. World Energy Statistics and Balances. Paris: International
Energy Agency, 2013b.
International Road Federation. World Road Statistics 2012. Geneva:
International Road Federation, 2012. Data also obtained from back
issues.
Law, T. H., R. B. Noland, and A. W. Evans. "The Sources of the
Kuznets Relationship between Road Fatalities and Economic Growth."
Journal of Transport Geography, 19(2), 2011, 355-65.
Leigh, J. R, and E. M. Geraghty. "High Gasoline Prices and
Mortality from Motor Vehicle Crashes and Air Pollution." Journal of
Occupational and Environmental Medicine, 50(3), 2008, 249-54.
Leigh, J. P., and J. T. Wilkinson. "The Effect of Gasoline
Taxes on Highway Fatalities." Journal of Policy Analysis and
Management, 10(3), 1991, 474-81.
Litman, T. "Pricing for Traffic Safety: How Efficient
Transport Pricing Can Reduce Roadway Crash Risks." Transportation
Research Record, 2318, 2012, 16-22.
Luoma, J., and M. Sivak. "Characteristics and Availability of
Fatal Road-Crash Databases in 20 Countries Worldwide." Journal of
Safety Research, 38(3), 2007, 323-27.
Montour, C. P. "Death and Taxes: The Effect of Gasoline Prices
on Traffic Fatalities." St. Olaf College Economics
Department's Omicron Delta Epsilon Journal of Economic Research,
1(3), 2011, 37-62.
Morrisey, M. A., and D. C. Grabowski. "Gas Prices, Beer Taxes
and GDL Programmes: Effects on Auto Fatalities among Young Adults in the
US." Applied Economics, 43(25), 2011, 3645-54.
Nishitateno, S., and P. J. Burke. "The Motorcycle Kuznets
Curve." Journal of Transport Geography, 36, 2014, 116-23.
Noland, R. B. "Fuel Economy and Traffic Fatalities:
Multivariate Analysis of International Data." Energy Policy,
33(17), 2005, 2183-90.
Organization for Economic Co-operation and Development.
"OECD.Stat Extracts." 2013a. Accessed October 13, 2014.
http://stats.oecd.org/.
--. OECD Factbook 2013: Economic, Environmental and Social
Statistics. Paris: Organization for Economic Co-operation and
Development, 2013b.
Page, Y. "A Statistical Model to Compare Road Mortality in
OECD Countries." Accident Analysis and Prevention, 33(3), 2001,
371-85.
Passmore, J., N. T. H. Tu, M. A. Luong, N. D. Chinh, and N. P. Nam.
"Impact of Mandatory Motorcycle Helmet Wearing Legislation on Head
Injuries in Viet Nam: Results of a Preliminary Analysis." Traffic
Injury Prevention, 11(2), 2010, 202-6.
Pesaran, M. H., and R. Smith. "Estimating Long-Run
Relationships from Dynamic Heterogeneous Panels." Journal of
Econometrics, 68(1), 1995, 79-113.
Pirotte, A. "Convergence of the Static Estimation toward the
Long Run Effects of Dynamic Panel Data Models." Economics Letters,
63(2), 1999, 151-58.
--. "Convergence of the Static Estimation toward the Long Run
Effects of Dynamic Panel Data Models: A Labour Demand
Illustration." Applied Economics Letters, 10(13), 2003, 843-47.
Sauerzapf, V., A. P. Jones, and R. Haynes. "The Problems in
Determining International Road Mortality." Accident Analysis and
Prevention, 42(2), 2010, 492-99.
Sivak, M. "Mechanisms Involved in the Recent Large Reductions
in US Road Fatalities." Injury Prevention, 15(3), 2009, 205-6.
Sivak, M., and B. Schoettle. "Toward Understanding the Recent
Large Reductions in U.S. Road Fatalities." Traffic Injury
Prevention, 11(6), 2010, 561-66.
Stem, D. 1. "Between Estimates of the Emissions-Income
Elasticity." Ecological Economics, 69(11), 2010, 2173-82.
Stock, J. H., and M. Yogo "Testing for Weak Instruments in
Linear IV Regression," in Identification and Inference for
Econometric Models: Essays in Honor of Thomas Rothenberg, edited by D.
W. K. Andrews and J. H. Stock. Cambridge, UK: Cambridge University
Press, 2005, 80-108.
U.S. Department of Energy. "Driving More Efficiently."
2013. Accessed October 13, 2014. http://www. fueleconomy.gov.
U.S. Energy Information Administration. "International Energy
Statistics." 2011. Accessed October 13, 2014.
http://www.eia.gov/cfapps/ipdbproject/iedindex3.cfm.
United Nations. "World Population Prospects: The 2010
Revision." 2010. Accessed October 13, 2014. http://esa.
un.org/wpp/.
United Nations Economic Commission for Europe. "Statistical
Database." 2013. Accessed October 13, 2014.
http://w3.unece.org/pxweb/.
White, M. J. "The 'Arms Race' on American Roads: The
Effect of Sport Utility Vehicles and Pickup Tracks on Traffic
Safety." Journal of Law and Economics, 47(2), 2004, 333-55.
Wilson, F. A., J. P. Stimpson, and P. E. Hilsenrath. "Gasoline
Prices and Their Relationship to Rising Motorcycle Fatalities,
1990-2007." American Journal of Public Health, 99(10), 2009,
1753-58.
World Bank. "World Development Indicators." 2013a.
Accessed October 13, 2014. http://data.worldbank.org/.
--. "Worldwide Governance Indicators." 2013b. Accessed
October 13, 2014. http://www.govindicators.org.
World Health Organization. Global Plan for the Decade of Action for
Road Safety 2011-2020. Geneva: World Health Organization, 2010.
--. "Top 10 Causes of Death." 2013a. Accessed October 13,
2014. http://www.who.int/mediacentre/factsheets/ fs310/en/index.html.
--. Global Status Report on Road Safety 2013: Supporting a Decade
of Action. Geneva: World Health Organization, 2013b.
--. "Global Information System on Alcohol and Health."
2013c. Accessed October 13,2014. http://apps.
who.int/gho/data/view.main?showonly=GISAH.
--. "Global Health Observatory Data Repository." 2013d.
Accessed October 13, 2014. http://apps.who.int/gho/data.
(1.) Speeds over 80 kmph are generally associated with lower fuel
efficiency (U.S. Department of Energy 2013). The Congressional Budget
Office (2008) reports that higher gasoline prices indeed cause drivers
on uncongested Californian roads to slightly reduce their speeds.
(2.) Recent evidence on the safety risks associated with motorcycle
travel is provided by Nishitateno and Burke (2014).
(3.) The long-run response is more important from a policy
viewpoint than the short-run response.
(4.) Negative binomial models are often used in studies of road
deaths (see, for instance, the papers of Chi et al.). As will be
documented, negative binomial models provide long-run gasoline price
elasticities of road deaths that fall within our reported range. Our
focus is primarily on linear models (in log-log form) because these are
better suited to an instrumental variable context.
(5.) In their study of Los Angeles freeway speeds, Burger and
Kaffine (2009) also use the world oil price to instrument the local
gasoline price. Grabowski and Morrisey (2006) do not use an IV approach
in their study of road deaths but hope that state gasoline taxes provide
an exogenous source of variation in gasoline prices.
(6.) A regression of log gasoline price on country and year dummies
has an [R.sup.2] of .89.
(7.) Results are similar without these country-specific time
trends.
(8.) The first stage in Column 5 of Table 2 indicates that the
partial effect of a 1% increase in the real world oil price is on
average a 0.3% increase in the domestic gasoline price, holding other
factors constant. Having an additional ton of in-ground oil reserves per
capita on average reduces the local gasoline price by 0.07%.
(9.) Burger and Kaffine's (2009) results on the effects of
gasoline prices on rush-hour vehicle speed and Burke and
Nishitateno's (2013) estimates of the gasoline price elasticity of
demand are also similar across their OLS and IV specifications.
(10.) The International Road Federation (2012) and OECD (2013a)
provide some international data on vehicle or passenger kilometers
traveled, but these are unavailable for the majority of our sample and
are of questionable quality. The OECD (2013b) notes that there is no
common international method for calculating passenger distance traveled
in road vehicles. Studies of the United States provide somewhat
conflicting results on how gasoline prices affect road deaths: Grabowski
and Morrisey (2004) find that the effect of higher gasoline prices on
road deaths operates via a reduction in vehicle distance traveled,
whereas Haughton and Sarkar (1996), Grabowski and Morrisey (2006), Chi
et al. (2010,2013b), and Montour (2011) report that there is also a
reduction in road deaths per vehicle-kilometer traveled.
(11.) In regressions with a smaller set of controls, a positive and
significant effect of motor vehicle numbers on road deaths is obtained.
Estimates of remain similar.
(12.) We obtain similar gasoline price elasticities of road deaths
in specifications that also control for seatbelt usage rates, rural road
speed limits, or expert assessments of the effectiveness of helmet law
enforcement (all measured in 2011; WHO 2013b, 2013d). Because data
limitations further reduce our sample, we omit these controls from Table
4.
(13.) Because the dependent variable is just a rescaling by
population, the effect of including our additional control variables in
this regression is the same as in Table 4.
(14.) In unreported specifications, we find that the gasoline price
elasticity of road deaths is similar when changes in gasoline prices are
either small or large.
(15.) These are ceteris paribus estimates for the year 2010.
Population and GDP growth will concurrently place upward pressure on
road deaths in most of these countries. There is also the chance that
subsidy removal could lower the global oil price and therefore increase
road deaths in other countries, but the magnitudes involved in such a
process are difficult to model.
(16.) Several of the countries listed in Table 7 (e.g., Indonesia,
Iran, and Nigeria) have reduced gasoline subsidies since 2010. Future
researchers might explore the effects of these recent subsidy reductions
on road safety.
(17.) Our estimates indicate that only 16% of the previously
mentioned reduction in U.S. road deaths over the period 2006-2010 was
due to rising gasoline prices.
PAUL J. BURKE and SHUHEI NISHITATENO *
* We are grateful for comments from Joseph Doyle, Ryan Edwards,
Yusaku Horiuchi, Brantley Liddle, Anthony Ockwell, Hideo Yunoue, two
referees, and participants at the Japanese Economic Association Fall
Meeting 2014 and seminars at the Australian National University, Beijing
Institute of Technology, Kwansei Gakuin University, Monash University,
and the University of Tasmania.
Burke: Arndt-Corden Department of Economics, Australian National
University, Canberra, ACT 2601, Australia. Phone +61 2 6125 6566, Fax
+61 2 6125 3700, E-mail
[email protected]
Nishitateno: School of Policy Studies, Kwansei Gakuin University,
Hyogo 669-1337, Japan. Phone +81 79 565 7957, Fax +81 79 565 7957,
E-mail
[email protected]
TABLE 1
Results for Single-Equation Specifications
Dependent Variable: Ln Road Deaths
Fixed Effects
2010 Pooled Between
Specification (1) (2) (3) (4) (5)
Ln gasoline price -.40 *** -.30 *** -.35 *** .01 -.10 *
-0.11 (.07) (.08) (.10) (.06)
Ln GDP .00 .17 *** .23 *** 34 ** 34 ***
(.07) (.04) (.04) (.15) (.10)
Ln population .98 *** .79 *** .72 *** 1.09 ** .12
(.08) (.05) (.05) (.46) (.37)
Year fixed effects No Yes No Yes No
Country-specific No No No No Yes
time trends
[R.sup.2] .87 .86 .88 .16 .61
Observations 101 837 837 837 837
Countries 101 144 144 144 144
Notes: Standard errors are robust and clustered at the country
level (except for the between estimate). The [R.sup.2]s reflect
the power of the explanatory variables and year dummies (but not
the country fixed effects). Coefficients on constants not reported.
***, **, and * indicate statistical significance at 1%, 5%, and
10%, respectively.
TABLE 2
Instrumental Variable Results
Dependent Variable: Ln Road Deaths
Oil Reserves per Ln Real-World
Capita ('000 tons) Oil Price
Fixed
Instruments 2010 Pooled Pooled Effects
Specification (1) (2) (3) (4)
Ln gasoline price -.31 *** -.24 *** -.51 *** -.39 **
(.07) (.06) (.12) (.16)
Ln GDP .00 .16 *** .14 ** .46 ***
(.07) (.04) (.06) (.12)
Ln population .98 *** .79 *** .92 *** .02
(.07) (.04) (.08) (.39)
Year fixed effects No Yes No No
Country-specific time No No Yes Yes
trends
[R.sup.2] .87 .86 .95 .58
First stage
Coefficient on oil .-49 *** .-33 *** -- --
reserves per capita
Coefficient on Ln real- -- -- 29 *** 24 ***
world oil price
Partial [R.sup.2] on .23 .13 .15 .13
instruments
F statistic on 12.31 12.71 108.75 59.05
instruments
Robust endogeneity test .45 .50 .04 .06
p value
Sargan overidentification -- -- -- --
test p value
Observations 101 837 837 830
Countries 101 144 144 137
Dependent Variable: Ln Road Deaths
Both Both
Pooled, with Full
Set of Controls
Instruments Pooled from Table 4
Specification (5) (6)
Ln gasoline price -.46 *** -.91 ***
(.08) (.20)
Ln GDP .14 ** .00
(.06) (.09)
Ln population 92 *** 1.00 ***
(.08) (.10)
Year fixed effects No No
Country-specific time Yes Yes
trends
[R.sup.2] .95 .99
First stage
Coefficient on oil -.69 *** .-79 ***
reserves per capita
Coefficient on Ln real- 28 *** 13 ***
world oil price
Partial [R.sup.2] on .26 .19
instruments
F statistic on 59.79 14.41
instruments
Robust endogeneity test .00 .00
p value
Sargan overidentification .35 .36
test p value
Observations 837 408
Countries 144 91
Notes: Standard errors are robust and clustered at the country
level. The [R.sup.2]s reflect the power of the explanatory
variables (except country fixed effects). Coefficients on constants
and the additional controls in Column 6 are not reported. The
instrumented variable is the log gasoline price. The null of weak
instruments is rejected if the F statistic on the instruments
exceeds the Stock-Yogo critical value. The Stock-Yogo 5% critical
value for 10% (15%) maximal IV size is 16.38 (8.96) with one
instrument and 19.93 (11.59) with two instruments. The
overidentification test is for specifications with robust but
unclustered standard errors. Column 4 drops seven singletons.
***, **, and * indicate statistical significance at 1%, 5%, and
10%, respectively.
TABLE 3
Distributed Lag Results
Dependent Variable: Ln Road Deaths
(1) (2) (3)
Ln gasoline [price.sub.t-1] -.04 .02 .13
(.08) (.07) (.08)
Ln gasoline [price.sub.t-3] -.12 -.23 *
(.12) (.13)
Ln gasoline [price.sub.t-5] -.19 ***
(.07)
Ln gasoline [price.sub.t-7]
Ln gasoline [price.sub.t-9]
Ln [GDP.sub.t] .28 * .31 * 37 ***
(.16) (.17) (.11)
Ln [population.sub.t] 1.19 ** 1.11 * .42
(.48) (.61) (.64)
Country fixed effects Yes Yes Yes
Year fixed effects Yes Yes Yes
Long-run gasoline price elasticity -.04 -.10 -.29 **
Same elasticity: pooled OLS estimate -.31 *** -.35 *** -.39 ***
Same elasticity: between estimate -.27 *** -.29 *** -.40 ***
Same elasticity: estimate with -.06 -.12 -.35 **
country-specific time trends (as
well as country fixed effects)
[R.sup.2] .15 .14 .14
Observations 762 569 442
Years 1992-2009 1994-2009 1996-2009
Countries 149 145 140
Dependent Variable: Ln Road Deaths
(4) (5)
Ln gasoline [price.sub.t-1] .05 -.06
(.09) (.17)
Ln gasoline [price.sub.t-3] -.04 .17
(.12) (.19)
Ln gasoline [price.sub.t-5] -.17 -.34 *
(.11) (.19)
Ln gasoline [price.sub.t-7] -.20 ** -.34 **
(.09) (.17)
Ln gasoline [price.sub.t-9] -.06
(.14)
Ln [GDP.sub.t] .53 ** .71 *
(.26) (.40)
Ln [population.sub.t] .19 .26
(.85) (.94)
Country fixed effects Yes Yes
Year fixed effects Yes Yes
Long-run gasoline price elasticity -.36 * -.63 *
Same elasticity: pooled OLS estimate -.47 *** -.57 ***
Same elasticity: between estimate -.37 *** -.59 ***
Same elasticity: estimate with -.58 ** -.96 **
country-specific time trends (as
well as country fixed effects)
[R.sup.2] .12 .16
Observations 336 237
Years 2005-2009 2007-2009
Countries 133 129
Notes: Standard errors are robust and clustered at the country
level. The [R.sup.2]s reflect the power of the explanatory
variables and year dummies (but not the country fixed effects).
Coefficients on constants not reported.
***, **, and * indicate statistical significance at 1%, 5%, and
10%, respectively.
TABLE 4
Results with Additional Controls
Dependent Variable: Ln Road Deaths. Estimator: Between
(1) (2) (3)
Ln average gasoline and diesel price -.31 *** -.41 ***
(08) (.13)
Ln gasoline price -.45 ***
(.13)
Ln GDP 23 *** .23 .21
(.04) (.15) (.15)
Ln population 22 *** .67 *** 71 ***
(.05) (.13) (.13)
Ln land area .05 .05
(.05) (.05)
Ln road distance .04 .04
(.10) (.10)
Paved road share (%) .00 .00
(.00) (.00)
Ln motor vehicle stock (4+ wheels) .11 .09
(.12) (.12)
Ln motorcycle stock -.02 -.02
(.04) (.04)
Rail share of energy used in .01 .01
transport (%) (.01) (.01)
Ln air passengers -.11 * -.11 *
(.07) (.06)
Population aged 15-24 (%) .03 .04
(.03) (.03)
Urban population (%) -.00 -.00
(.00) (.00)
Ln alcohol consumption per adult .11 ** .11 **
(.05) (.05)
Blood alcohol limit for drivers in 1.11 1.22
(2.25) (2.24)
Maximum speed in urban areas in 2011 .01 *** .01 **
(.00) (.00)
Rule of law score 40 ** .40 **
(.20) (.20)
Control of corruption score -.36 ** -.37 **
(.17) (.17)
Economic growth rate (%) -.02 ** -.02 **
(.01) (.01)
Ln infant mortality rate .02 -.02
(.14) (.14)
[R.sup.2] .88 .95 .95
Observations 832 408 408
Countries 144 91 91
Notes: Coefficients on constants not reported. Log alcohol
consumption per adult is lagged 1 year to increase sample size.
***, **, and * indicate statistical significance at 1%, 5%, and
10%, respectively.
TABLE 5
Estimates for Road Deaths per 100,000 Population
Dependent Variable: Ln Road Deaths per 100,000 Population. Estimator:
Between
Sample Full (1) Full (2) Full (3) Full (4)
Ln gasoline price -.30 *** -.24 *** 1.26 *** -.80
(.08) (.08) (.45) (.69)
Ln GDP per capita .26 *** 2.88 *** .29 *** .02
(.04) (.49) (.04) (.34)
Ln population density -.09 *** -.07 ** -.09 *** -.09 ***
(.03) (.03) (.03) (.03)
[(Ln GDP per capita). -.15 ***
sup.2] (.03)
[(Ln gasoline price). -.22 ***
sup.2] (.06)
Ln gasoline price * Ln GDP .05
per capita (.08)
GDP per capita at turning -- 12,828 -- --
point ($)
Estimated gasoline price elasticity for xth-percentile gasoline price
25th -- -- -.52 *** --
50th -- -- -.70 *** --
75th -- -- -.83 *** --
[R.sup.2] .29 .41 .35 .30
Observations 837 837 837 837
Countries 144 144 144 144
Dependent Variable: Ln Road Deaths per 100,000 Population. Estimator:
Between
Full, with
Non-OECD Regional
Sample OECD (5) (6) Dummies (7)
Ln gasoline price -.47 * -.17 * -.28 ***
(.24) (.10) (.09)
Ln GDP per capita -.09 .39 *** .20 ***
(.15) (.06) (.06)
Ln population density .06 -.15 *** -.09 **
(.05) (.04) (.04)
[(Ln GDP per capita).
sup.2]
[(Ln gasoline price).
sup.2]
Ln gasoline price * Ln GDP
per capita
GDP per capita at turning -- -- --
point ($)
Estimated gasoline price elasticity for xth-percentile gasoline price
25th -- -- --
50th -- -- --
75th -- -- --
[R.sup.2] .16 .39 .33
Observations 270 567 837
Countries 34 110 144
Notes: Coefficients on constants not reported. The OECD subsample
includes all 34 current member countries. The regional dummies are
based on the seven World Bank (2013a) regions. Year dummies are not
included because the between estimator is being employed.
***, **, and * indicate statistical significance at 1%, 5%, and
10%, respectively.
TABLE 6
Negative Binomial Models
Dependent Variable: Road Deaths per 100,000 Population
Pooled, Pooled,
with Full with
2010 Pooled Set of Country
Sample (1) (2) Controls Fixed
from Table Effects
4 (3) (4)
Ln gasoline price -.42 *** -.25 *** -.29 *** -.17 ***
(.09) (.07) (.11) (.06)
Ln GDP per capita 2.79 *** 2.43 *** 2.21 *** 2.55 **
(.67) (.45) (.61) (.99)
Ln population density -.08 ** -.08 *** -.13 .19
(.04) (.03) (.10) (.21)
[(Ln GDP per capita). -.16 *** -.13 *** -.17 *** -.12 **
sup.2] (.04) (.03) (.04) (.05)
Year fixed effects No Yes Yes Yes
GDP per capita at 7,206 11,289 9,629 30,372
turning point ($)
Observations 101 837 408 830
Countries 101 144 91 137
Notes: Standard errors are robust and clustered at the country
level. Column 4 results have been obtained by including country
dummies in a negative binomial estimation. Coefficients on
constants not reported.
***, **, and * indicate statistical significance at 1%, 5%, and
10%, respectively.
TABLE 7
Gasoline-Subsidizing Countries: Estimated Road Deaths Avoided if
Gasoline Price Were Equal to the Level in the United States (76
Cents per Liter, 2010)
(1) (2) (3) (4) (5)
Estimate:
Avoided
Road Road
Gasoline Deaths per Deaths if
Pump 100,000 Road Gasoline
Price Population Deaths Price Were
(U.S. (WHO (WHO 76 U.S.
Country Cents) 2013b) 2013b) Cents
Venezuela 2 37 10,791 >5,000
Iran 10 34 25,224 10,600
Saudi Arabia 16 25 6,800 2,700
Libya 17 n.a. n.a. 584
Qatar 19 14 247 158
Bahrain 21 11 132 103
Turkmenistan 22 n.a. n.a. 382
Kuwait 23 17 452 202
Oman 31 30 845 144
Algeria 32 n.a. n.a. 1,700
Yemen 35 24 5,698 1,000
Brunei Darussalam 39 7 27 15
Nigeria 44 34 53,339 4,200
United Arab Emirates 47 13 956 210
Egypt 48 13 10,729 1,800
Indonesia 51 18 42.434 4,500
Ecuador 53 27 3,911 259
Malaysia 59 25 7,085 345
Sudan 62 25 10,935 331
Angola 65 23 4,407 141
Bolivia 70 19 1,910 38
Kazakhstan 71 22 3,514 51
Azerbaijan 75 13 1,202 6
Sum for 23 countries ~35,000
Notes: Countries are ordered by Column 2 value. Column 5 estimates
use a gasoline price elasticity of road deaths of -0.4 and are
rounded to the nearest hundred if >1,000. Estimates are capped at
">5,000" for Venezuela, given the imprecision associated with
estimates using such large price changes. Regression estimates use
IRF reported road death data rather than the WHO death estimates.
The WHO data are their estimates and are shown because they provide
superior country coverage in 2010 (only).
