Journal of Economic Perspectives—Volume 21, Number 3—Summer 2007—Pages 199 –222

Natural Disasters, Economic

Development, and Humanitarian Aid

David Strömberg

N atural disasters are one of the major problems facing humankind. Be-
tween 1980 and 2004, two million people were reported killed and five
billion people cumulatively affected by around 7,000 natural disasters,
according to the dataset maintained by the Centre for Research on the Epidemi-
ology of Disasters (CRED) at University of Louvain (Belgium). The economic costs
are considerable and rising. The direct economic damage from natural disasters
between 1980 –2004 is estimated at around $1 trillion.
Of course, the damage caused by natural disasters has been discussed for
centuries. For example, in 1755 an earthquake devastated Lisbon, which was then
Europe’s fourth-largest city. At the first quake, fissures five meters wide appeared in
the city center. The waves of the subsequent tsunami engulfed the harbor and
downtown. Fires raged for days in areas unaffected by the tsunami. An estimated
60,000 people were killed, out of a Lisbon population of 275,000. In a letter to
Voltaire dated August 18, 1756, Jean-Jacques Rousseau notes that while the earth-
quake was an act of nature, previous acts of men, like housing construction and
urban residence patterns, set the stage for the high death toll. Rousseau (as cited
by Dynes, 1999, from Masters and Kelly, 1990) wrote: “Without departing from your
subject of Lisbon, admit, for example, that nature did not construct twenty thou-
sand houses of six to seven stories there, and that if the inhabitants of this great city
had been more equally spread out and more lightly lodged, the damage would have
been much less and perhaps of no account.”
Following Rousseau’s line of thought, disaster risk analysts (for example,
Coburn, Spence, and Pomonis, 1994; Mileti, 1999) distinguish three factors con-
tributing to a disaster: the triggering natural hazard event (such as the earthquake

y David Strömberg is Associate Professor, Institute for International Economic Studies (IIES),
Stockholm University, Stockholm, Sweden. His e-mail address is 具[Link]@[Link]典.
200 Journal of Economic Perspectives

striking in the Atlantic Ocean outside Portugal); the population exposed to the event
(such as the 275,000 citizens of Lisbon); and the vulnerability of that population
(higher for the people in seven-story buildings). Distinguishing these factors will
prove useful in discussing how the effects of natural disasters evolve over time and
across countries.
The Lisbon earthquake was the first natural disaster in which the state ac-
cepted the responsibility for emergency response and reconstruction (Dynes,
1999). Lisbon suffered no epidemics of disease and was already being rebuilt within
less than a year. The new downtown of Lisbon was designed to resist subsequent
earthquakes. A possible explanation for the rapid and appropriate response is that
Portugal was relatively wealthy at the time and that important structural and
political changes were moving Portugal toward more modern economic and polit-
ical institutional forms (Dynes, 1999).
The general point is that societal factors affect the impact of disasters. Today,
disaster fatalities are, on average, higher in low-income countries and nondemo-
cratic countries. There could be many reasons for this; for example, perhaps
low-income countries are more exposed to natural hazards, or perhaps fatalities in
nondemocratic countries are higher only because these countries are also poor.
Below we will try to sort out how societal factors, such as economic and political
development, are related to the impact of disasters.
Rousseau mentions another important point in his letter, which is that
whether an event is elevated to social discussion and considered a disaster
depends on who is affected. He wrote: “You might have wished . . . that the
quake had occurred in the middle of a wilderness rather than in Lisbon. . . . But
we do not speak of them, because they do not cause any harm to the Gentlemen
of the cities, the only men of whom we take account.” Concerns continue today
that the decision about what to call a disaster and how much relief to provide
depends on who is suffering. For example, international relief may neglect
disasters in countries that are culturally and geographically distant from the
major donors and may favor high-profile emergencies at the expense of more
invisible suffering far from the media or the political spotlight.
This paper starts by describing the incidence of natural disasters, where
they strike, and their development over time. It then discusses how societal
factors act to protect people from or expose them to natural hazards. The final
section discusses the determinants and targets of international aid to disaster
victims.

Where Do Disasters Occur and Have They Increased Over Time?

This section discusses where natural disasters strike in geographic terms and
whether they have increased over time. I primarily use data from the Emergency
Events Database (EM-DAT), maintained by the Centre for Research on the Epide-
 David Strömberg 201

Table 1
Global Humanitarian Consequences of Natural Disasters, 1980 –2004

Number killed Number affected

Disaster type Number (thousands) (millions)

Drought 510 559 1,590

Wind storm 1,813 264 537
Wave/surge 28 229 2
Earthquake 621 215 78
Flood 2,102 171 2,490
Extreme temperature 237 68 11
Volcano 109 25 3
Slides 358 19 6
Wildfires 250 1 4
Total 6,028 1,551 4,710

miology of Disasters (CRED) at the University of Louvain (Belgium).1 Natural

disasters occur when extreme natural phenomena like earthquakes, floods, or
storms cause loss of lives, human suffering, or extensive damage to property. An
event qualifies as a disaster in the CRED database if at least one of the following
criteria is fulfilled: 10 or more people are reported killed; 100 or more people are
reported affected, injured, and/or homeless; the government declares a state of
emergency; or the government requests international assistance.
We limit the discussion to natural disasters caused by geophysical events, such
as volcanoes and earthquakes, or hydro-meteorological events, such as floods,
winds, and landslides. Between 1980 and 2004, 6,028 natural disasters of these types
occurred, causing around 1.5 million deaths and affecting 4.7 billion individuals, as
summarized in Table 1. Droughts have been the most deadly disaster, followed by
windstorms and wave surges (tsunamis), while floods have affected the largest
number of people. We do not discuss epidemics, insect infestations, and famines—
although these are also contained in the CRED data.
One general concern is that reporting (and misreporting) of disasters varies
systematically across time, levels of income, and political regimes. Reporting of
small disasters is probably more complete in later years and in more developed
countries. Closed undemocratic societies may also underreport natural disasters.
For example, China reported an average of less than one disaster per year over the
period 1960 –1979 (under Mao Zedong and Hua Guofeng) and never reported less
than six disasters per year after 1980 (with Deng Xiaoping taking over as de facto
leader). These systematic reporting differences make it hard to estimate how
economic development and political regime affect disaster causualties. Despite
these problems, the CRED database is the best one available, and we discuss it
keeping these problems in mind.

1
There exist two other global sources on natural disasters maintained by private insurance companies,
namely Sigma from Swiss Re and NatCat from Munich Re. However, these are not in the public domain.
202 Journal of Economic Perspectives

Figure 1
Number and Magnitude of Disasters, 1960 –2005

600
Total number killed

100,000
Number of disasters

400

Number killed
10,000

Number of disasters
1,000
200

Number killed per disaster 100

0 10
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005

Although the total number of disasters reported worldwide has been rapidly
increasing, the number of killed has not (except in Africa). The yearly numbers of
disasters are plotted in Figure 1, together with their total and average magnitudes.
Since 1960, the number of disasters has, on average, increased by around 5 percent
per year and the number of affected by 4 percent. However, the number of killed
has only increased by an insignificant 0.1 percent per year. In other words, the
average magnitude of the reported disasters has fallen.
What explains the growth in the number of natural disasters? More frequent
extreme weather events may explain part of the increase in droughts, floods, and
storms. According to the Intergovernmental Panel on Climate Change (2007),
there is likely to have been an increase in heavy precipitation events and, in some
regions, increases in tropical cyclone activity and extreme weather events causing
droughts in the period 1970 –2000.
More complete reporting may also be driving this increase. Although, the
number of earthquakes that hit populated areas has not increased since 1970
(Peduzzi, 2005), the numbers of reported earthquakes display the same 5 percent
yearly increase from 1960 –2004 as do disasters in general. The reason may be that
the propensity to report earthquakes has increased roughly at this rate. Over the
periods 1970 –79, 1980 –92, and 1993–2003, the EM-DAT reported, respectively, 11,
25, and 31 percent of the earthquakes that hit populated areas (Peduzzi, 2005).
Over the same three time periods, the average reported number of earthquakes per
year in EM-DAT was 10, 23, and 27.
Population growth may also turn more natural hazard events into disasters, by
exposing more people. To get a rough estimate of the importance of this factor, I
regressed the number of disasters in each country and each year from 1960 to 2004
on the country’s population, allowing a separate population coefficient for India
and China, and only including country and year fixed effects. The population
 Natural Disasters, Economic Development, and Humanitarian Aid 203

Table 2
Regional Humanitarian Consequences of Natural Disasters, 1980 –2004

Number of Killed Affected Killed per Affected per

disasters (thousands) (millions) 100,000 100,000

Africa 861 581 324 2.61 1,453

Asia 2,352 726 4,210 0.74 4,303
Americas 1,626 131 125 0.59 564
Europe 863 110 37 0.60 206
Oceania 324 4 19 0.46 2,363

Notes: The first three columns show the total number of disasters, number killed, number affected over the
period 1980 –2004. The last two columns divide those numbers by the population in 2006 in 100,000s.

coefficients are highly significant and imply that an increased population of ten
million is associated with an average increase in the number of disasters by 0.9
outside of China and India, and 0.3 in these two countries. Between 1960 and 2000,
population grew by 1.2 billion in India and China, and 1.8 billion in the remaining
countries in our sample. The associated increase in the number of disasters is
around 190, which is around half of the total actual increase shown in Figure 1.
Given that extreme weather events have become more frequent, world popu-
lation has doubled, and disaster reporting has become more complete, why is it that
the total death toll has not increased? To get an idea of what can explain the falling
per capita vulnerability, we now turn to the geographic distribution of natural
disasters.
Comparing continents, Asia has the most disasters, fatalities, and people
affected, as shown in Table 2. This finding is not surprising since Asia contains over
60 percent of the world population and one-third of the world land mass. In
relation to the population, the death rate is highest in Africa, and the share of the
population affected was highest in Asia.
To look more closely at the geographical distribution of disasters, it is useful to
study acts of nature and acts of men separately. Starting with acts of nature, global
data has been compiled by Dilley, Maxx, Chen, Deichmann, Lerner-Lam, and
Arnold (2005); for most types of events, the data gathered covers the period
1980 –2000. This data covers just the triggering natural phenomena: 1,600 storm
tracks; 4,000 cases of volcanic activity categorized as moderate or above; earth-
quakes of magnitude 4.5 or higher on the Richter scale; drought events when the
magnitude of monthly precipitation was less than or equal to 50 percent of its
long-term median value for three or more consecutive months; data on flood
events; and data on landslide risk based on soil moisture, precipitation, seismicity,
and temperature.
The global distribution of natural hazards is displayed in Figure 2 on a latitude–
longitude grid, whose boxes are 2.5 by 2.5 minutes (5 by 5 kilometers at the equator).
The graph shows areas in the top three risk deciles of exposure to earthquakes,
volcanoes, drought, floods, storms, or landslides. Floods, storms, and landslides
204 Journal of Economic Perspectives

Figure 2
Geographic Distribution of Natural Hazard Events, 1980 –2000

Top 3 deciles exposed to:

Drought, hydro-, and/or
geophysical hazards

Note: Computed by Dilley, Chen, Deichmann, Lerner-Lam, and Arnold (2005) for grid cells of 2.5 by
2.5 minutes (5 by 5 kilometers at the equator). The graph shows the grid cells that are in the top
three deciles of exposure to droughts or hydro- or geophysical hazards.

strongly affect eastern coastal regions of the major continents and the interior regions
of North and South America, Europe, and Asia. Earthquakes and volcanoes cluster
along fault boundaries with mountainous terrain. In the Americas, such a boundary
stretches from the western coast of the United States through Central America and the
western coast of South America. In Eurasia, one such boundary stretches from south-
ern Europe, through Turkey, Iran, and central Asia, all the way to southwest China and
Nepal. It then turns southeast through the waters outside Indonesia and then north-
east, passing New Zealand, the Philippines, Taiwan, and Japan. Drought is more widely
dispersed, affecting parts of Africa, the Middle East, India, and Southeast Asia, as well
as Brazil and the interior regions of North America.
Interestingly, no apparent relation exists between economic development and
exposure to natural hazards. High-income areas in Europe, North America, and
Japan are as highly exposed to natural hazards as are low-income areas in Africa and
Asia. This impression will be confirmed when we look at the numbers below.
The historical pattern of fatalities is very different. Figure 3 shows estimated
mortality risks, computed by allocating regional mortalities from 1980 to 2000 by
disaster type to smaller geographical units using data on population and frequency of
natural hazard events. Mortality risks are clearly lower in countries with developed
economies. The United States is noteworthy in that more than one-third of its popu-
lation lives in hazard-prone areas but only 1 percent of its land area ranks high in
mortality risk (Dilley et al., 2005). In contrast, many people have died from natural
disasters in central Africa, without being highly exposed to natural hazards. This
pattern suggests that high-income countries have been able to protect themselves
 David Strömberg 205

Figure 3
Geographic Distribution of Mortality Risk from Natural Hazards, Based on Events
1980 –2000

Note: Computed by Dilley, Chen, Deichmann, Lerner-Lam, and Arnold (2005) for grid cells of 2.5 by
2.5 minutes (5 by 5 kilometers at the equator). The graph shows the grid cells that are in the top
three and five deciles of the risk distribution.

against fatalities from natural disasters. The next section will investigate this pattern in
more detail.

What Characterizes Countries Vulnerable to Disasters?

High-income countries face a lower mortality risk from natural hazards, for
obvious reasons. High-income societies can better afford measures to limit the
effects of natural disasters. Buildings can be constructed of stronger and more
durable materials. Houses can have raised platforms to better withstand floods.
Agricultural areas can be irrigated to reduce losses during droughts. Warning
systems for certain natural disasters, such as hurricanes, can save lives (Sheets and
Williams, 2001). After a disaster strikes, measures such as bringing in medical care
and food as well as mass evacuations can limit the negative consequences of the
disaster.
Since the government typically handles many of these measures, disasters may
be less severe in countries with efficient and accountable governments. Govern-
ment efficacy is positively related to having firm civil liberties and a free press
(Isham, Kaufmann, and Pritchett, 1997). In a comparison across states within India,
Besley and Burgess (2002) found that calamity relief has been more responsive to
needs in states where more people read newspapers. For this reason, it is easy to
believe that disasters are less severe in those countries that are more democratic,
having a free press, as forcefully argued by Sen (1990).
206 Journal of Economic Perspectives

Table 3
Comparing High- and Low-Income Countries

Country income Number of Population Exposed pop. Killed in GDP per Democracy
category disasters (million) (million) disasters capita index

High-income 1,476 828 440 75,425 23,021 9.5

Low-income 1,533 869 496 907,810 1,345 3.2

Note: The first and the fourth columns contain the numbers of natural disasters and number killed,
respectively, over the period 1980 –2004. The other columns contain characteristics in 1996. “Exposed
pop.” is the population share in each country that live in areas in the top three deciles of risk exposure
to volcanic activity, earthquakes, storms, floods, landslides, or droughts—multiplied by the population
in the country and summed over the countries in the income group. The “Democracy index” is the
population-weighted average POLITY IV Democracy index.

Economic inequality may also increase vulnerability, because inequality in-

creases the share of the very poor and this group is particularly vulnerable (Chou,
Huang, Lee, Tsai, Chen, and Chang, 2004). Poor individuals typically have a worse
initial health status, live in lower-quality housing, more frequently live in areas
facing disaster risk, have lower savings and less insurance, and have poorer access
to alternative food supply routes. In addition, economic inequality may reflect a
society or a political system that is ill-equipped to make a collective effort to limit
the consequences of natural hazards (Kahn, 2005; Anbarci, Escaleras, and Register,
2005).
A small but growing body of work in economics has investigated these hypoth-
eses. We discuss the results and provide some empirical evidence using the EM-DAT
data merged with data on country characteristics. We will focus on disaster fatalities,
since there are fewer issues in measuring the number of people killed than the
number affected or economic losses.
Starting at the widest perspective, the one-third of the world’s population that
lives in low-income countries suffers almost two-thirds of all fatalities from disasters
(using the World Bank’s official analytical classification of high-, middle- and
low-income countries in 2006).2 In other words, disaster-related deaths per capita
are four times higher in this group of countries than elsewhere. It is instructive to
compare just the groups of high- and low-income countries. These are roughly
comparable in terms the number of disasters 1980 –2000, population size, and
populations exposed to high disaster risk. However, the high-income countries
suffered an order of magnitude fewer fatalities than did low-income countries, as
shown in Table 3.
How much of the 90 percent lower fatalities in high-income countries can be
explained by reduced vulnerability related to development? There could be many
reasons for the different findings for high- and low-income countries: perhaps different

2
World Bank list of economies (July 2006) available at 具[Link]
resources/[Link]典.
 Natural Disasters, Economic Development, and Humanitarian Aid 207

Figure 4
Fraction of Disasters by Magnitude for Low- and High-Income Countries, 1980 –2004

.15 Low-income
High-income
Smoothed fraction
.1 Normal distribution

.05

0
0 1 2 3 4 5 (Magnitude)
1 10 100 1,000 10,000 100,000 (Killed)

types of disasters strike in rich and poor countries; perhaps reporting is different; or
perhaps other factors drive both development and disaster-related deaths. It could also
be that the difference in fatalities appears by chance. The difference is completely
dominated by a few large disasters. Total fatalities would, in fact, have been higher in
high-income countries had the two largest disasters struck there. (The four most deadly
natural disasters in our data 1980 –2004 are: the 1984 droughts in Ethiopia and Sudan,
killing 300,000 and 150,000; the 2004 Indian Ocean Tsunami, killing 230,000; and the
1991 cyclone in Bangladesh, killing 140,000. All of these struck low-income countries,
except the 2004 tsunami that affected mainly middle-income countries.)
To assess the systematic differences, we use a linear regression to analyze the
determinants of the magnitude of the disaster measured in terms of the base-ten
logarithm of the number killed. Following Richardson (1960), a disaster that kills
one person has a magnitude of 0, one that kills ten has a magnitude of 1 in terms
of number killed, and one that kills a million has a magnitude of 6. Analyzing
magnitudes limits the influence of extreme observations. Figure 4 shows the
distributions of disaster magnitudes in the two income groups. For example,
disasters of magnitudes between 1 and 1.2 (numbers of killed between 10 and
101.2⫽16) were the most frequent in high-income countries, as 15 percent were of
this kind. In high-income countries, the whole disaster distribution is skewed
towards smaller magnitudes. This pattern indicates that the very large disasters are
much less likely in high-income countries.
Note that regressions on disaster magnitudes implicitly assume that the effects
on fatalities are proportional, independent of disaster size; for example, an increase
in income of 1 percent is associated with 0.4 percent fewer fatalities, whether the
disaster kills 100 or 10,000 people. This assumption is what enables us to draw
inferences on total fatalities, dominated by the large disasters, after observing
mainly small and medium-sized disasters.
We will analyze the 3,200 natural disasters (approximately) that occurred in
208 Journal of Economic Perspectives

1980 –2004 striking in countries and years for which we have background data.
Again, the dependent variable is the disaster magnitude. Our key independent
variables are various economic and political country-specific variables potentially
related to vulnerability to natural hazards. I will call these “vulnerability factors.” To
measure income, I use the income categories by the World Bank presented above
and real GDP per capita from the World Penn Tables.3 I use two variables as proxies
for government effectiveness and accountability. The first is the government effec-
tiveness index produced by the World Bank (Kaufmann, Kraay, and Mastruzzi,
2006). This broad, survey-based index aims at measuring the quality of public
services, infrastructure, and civil service. For each country, I use the 1996 –2004
average of this index. The next row is a yearly democracy index produced by the
POLITY IV (2004) project at the University of Maryland. The index is on a
zero-to-ten scale, and measures the competitiveness of political participation, the
openness and competitiveness of executive recruitment, and constraints on the
chief executive. I use this index divided by ten so that the range is between zero and
one. To measure inequality, I use the country’s average Gini coefficient 1980 –2004,
as reported by the UN-WIDER World Income Inequality Database.
To account for differences in the triggering natural hazard events and the
populations exposed, a set of controls is included in all regressions. This includes
dummy variables for type of disaster, the country population in levels and logs, and
the share of each country’s population that was exposed to high risk of natural
hazards in 1980 –2000. Of course, one would have liked to control more carefully
for the severity and the number of people affected by each particular disaster.
Peduzzi, Dao, and Herold (2005) have assembled this type of data, although it is
not yet publicly available. To control for other continent-specific and common
time-varying factors, continent and year dummy variables are included.
To simplify the exposition, I will sometimes refer to the estimates as “effects,”
although they may not measure causal effects. For example, societies that have high
per capita incomes or that experience economic growth are likely to be different
from poor, stagnant countries in a multitude of ways that are hard to measure and
are likely to affect disaster deaths. We cannot hope to control for all these factors.
It is important at least to correct the standard errors for the presence of these
unobserved factors, which is done by clustering by country.
How large is the total systematic difference in fatalities between high- and
low-income countries? In Table 4, column 1, I regress the disaster magnitudes on
only the indicators for middle-income and low-income countries and the controls.
The left-out category is high-income countries, so these coefficients can be inter-
preted relative to high-income countries. The estimates suggest around 70 percent
lower fatalities in high- than low-income countries from disasters of the same type
striking on the same continent in the same year, controlling for country population
and population shares exposed to high natural hazard risk. (The table shows that,

3
And I use the Laspeyres index, as reported in Heston, Summer, and Aten (2002).
 David Strömberg 209

Table 4
Characteristics of Countries with Severe Natural Disasters

Killed Killed Killed

(log10) (log10) (log10)

Middle-income country 0.29

(0.06)***
Low-income country 0.51
(0.08)***
Real GDP per capita (log10) ⫺0.21 ⫺0.43
(0.09)** (0.18)**
Government effectiveness ⫺0.13
(0.04)***
Democracy index 0.14 0.12
(0.06)** (0.06)*
Gini coefficient 0.30
(0.26)
Observations 3262 3137 4064
R-squared 0.25 0.26 0.31

Notes: Results from ordinary least squares regressions. The unit of observation is a natural disaster in the
CRED database. The dependent variable is log10(killed). All regressions include year-, continent-, and
disaster type fixed effects, and control for the log and level of population, and the share of the country’s
population living in areas exposed to high natural disaster risk. Column 3 also includes country fixed
effects. Robust standard errors, clustered by country, in parentheses.
* significant at 10 percent, ** significant at 5 percent, *** significant at 1 percent.

with the controls, the average disaster is about half a magnitude lower in high-
income countries than low-income countries, implying that high-income countries
have 30 percent of the fatalities of low-income countries, as 10-0.5 ⫽ 0.3.)
We now explore whether democracies are better at dealing with disasters. Sen
(1990) argues this, comparing the democratic India with China. Interestingly, the
Chinese government is more effective by the World Bank measures. Which of these
factors— democracy or effectiveness—is more important for disaster fatalities? The
specification in column 2 includes all the vulnerability factors discussed above, but
drops the middle- and low-income indicators and instead uses the (base-ten log of) real
per capita GDP to capture the economic status of countries. Income is still negatively
related to disaster deaths. Strikingly, countries with more effective government suffer
fewer fatalities while more democratic countries suffer more. This finding lends little
support to Sen’s argument. However, the benefits of democracy may appear as gov-
ernment effectiveness. Once government effectiveness is controlled for, the remaining
positive effect of democracy may just reflect more complete reporting of fatalities.
Consistent with this idea, the democracy index is insignificant when government
effectiveness is removed from the regression (result not shown).
I find little evidence that more unequal countries experience more disaster-
related deaths. The estimated Gini-coefficient is not significant. I also tried other
proxies for inequality, such as the income share of the poorest quintile (from
UN-WIDER), but these were not systematically related to disaster magnitudes
210 Journal of Economic Perspectives

either. In contrast, both Kahn (2005) and Anbarci, Escaleras, and Register (2005)
find that the Gini coefficient is positively correlated with disaster-related deaths,
controlling for income. Interestingly, Anbarci, Escalera, and Register also find that
the earthquake preparedness of countries, as judged by three international agen-
cies, is negatively correlated with the Gini coefficient (although this finding is just
a simple correlation). Thus, the evidence regarding the Gini-coefficient and disas-
ter effects is mixed.
The third column investigates whether changes in disaster magnitudes within
a country over time are related to changes in the potential vulnerability factors. By
including country-dummy variables, we effectively control for all differences be-
tween countries that are constant over time, such as geographical location and most
of the systematic natural hazard risk. Consequently, only time-varying country-level
variables, such as income and democracy, can now be included. To get more
statistical power, the sample period is expanded to 1960. The coefficient implies an
elasticity of disaster-related deaths to GDP per capita of – 0.43. Kahn (2005) finds
a similar elasticity for earthquake fatalities. To compare this to the earlier estimates,
an increase in per capita GDP from the average income among low-income
($1,300) countries to that of high- and low-income countries ($23,000) is associated
with a decrease in disaster fatalities of around 70 percent, as (23,000/1,300) – 0.43 ⫽
0.3. The democracy index is also positively correlated with disaster magnitude over
time, although not significantly so.
Recall that, despite population growth, total disaster-related deaths showed no
trend between 1960 and 2004. How much larger would disaster fatalities have been,
absent development? In an average country, GDP per capita increased 2.4 times
between 1960 and 2000. An estimated elasticity of – 0.4 implies that the current
number of disaster-related deaths would be around 30 percent higher, absent
economic development (2.4⫺0.4 ⫽ 0.7). However, this estimate assumes that all
countries had the same average development.
Taking into account growth and disaster death rates in individual countries,
disaster deaths are only 20 percent lower than they would have been absent
development. The reason is the weak development in Africa. Some African coun-
tries with many disaster-related deaths, notably Sudan and Mozambique, experi-
enced zero or negative economic growth from the 1960s to 2000. Summing the
effects across all African countries, the expected number of people killed in natural
disasters in Africa is the same evaluated at the levels of income in 1960 and in 2000.
Europe and Asia, on the other hand, would have had an estimated 30 percent more
disaster-related fatalities absent development. The high levels of growth in China,
Indonesia, and India account for much of the estimated reduction of fatalities in
Asia, somewhat balanced by the poor development in Bangladesh. The estimated
reduction is around 10 percent in the Americas.
I will now briefly discuss three remaining potentially important issues: misre-
porting; the effect of disasters on the economy; and whether the variance of disaster
magnitudes differs between high- and low-income countries. Starting with the last
issue, we may underestimate the effect of income by just looking at average disaster
 Natural Disasters, Economic Development, and Humanitarian Aid 211

magnitudes and not the full distribution. Total fatalities are dominated by extreme
disasters, and these are less likely if disaster magnitudes have less variance. High-
income countries do seem to have a lower variance, as shown in Figure 4. The
difference in estimated variance between high- and low-income countries implies
an additional 25 percent reduction of fatalities in high-income countries (under
the assumption that the errors are normally distributed). Note also that the
variance in disaster magnitudes is higher for some types of disasters, like droughts.
Another concern is that low-income countries may report fewer disasters
and, in particular, fewer small disasters. The average disaster magnitude in
low-income countries would then be larger in the data than in reality, making
reported disaster magnitude more negatively related to income. Hence, our
estimated – 0.4 elasticity of deaths per disaster with respect to income could be
an overestimate. However, the underreporting would also make the total num-
ber of reported disaster-related deaths in low-income countries smaller than in
reality. This would lead us to underestimate the effect of income on total
disaster deaths. In a pure cross-country regression, I estimated the elasticity of
total disaster-related deaths from 1980 to 2004 to average GDP per capita to be
– 0.5 and statistically significant.
We have discussed how economic development affects natural disasters, but
natural disasters also affect the economy. In theory, the impact of natural disasters
on GDP is unclear. The disaster may provide a positive contribution to measured
GDP as reconstruction efforts are part of GDP while the direct losses of human and
physical capital are not. On the other hand, the loss of productive capital may
reduce GDP. Growth in real incomes is not significantly different in years when
disasters strike than in an average year. This result was confirmed in our data after
running a regression of the percentage change in real GDP per capita on a dummy
variable indicating disaster years, including year and country fixed effects. For
other studies of this relationship, see Albala-Bertrand (1993), Chavériat (2000),
and Rasmussen (2004).
Looking forward, there is clearly scope to improve on the types of estimates
made in this section. Data at finer geographical and temporal levels already exist
for natural hazards as well as for demographic and economic variables. These could
be used to identify more convincingly the relationship between, for example,
income and disaster vulnerability. Developing joint statistical models of the trig-
gering natural phenomena (for example, weather and geological events) and
disaster effects would be a useful improvement to the “top three risk-deciles”– data
used in this paper. It may also prove a fruitful way to better understand the crucial
statistics of very large disasters.

Disaster Relief

Most victims of natural disasters live in low-income countries, with limited

resources to mitigate the effects of such disasters. Consequently, international relief
212 Journal of Economic Perspectives

from richer countries could play a key role. Given the large humanitarian stakes, it
is essential that relief be given where it can do most good. However, some suspect
that disaster relief favors high-profile emergencies at the expense of more invisible
suffering far from the media or the political spotlight. Other suspicions are that
disaster relief favors disaster victims in countries that have historical ties or lie close
to the large donors, or that it favors countries that are important from a foreign
policy or economic perspective. This section will discuss these motives and some
empirical evidence.

Who Gives Relief, Who Receives Relief, and How Much?

The two main sources of international disaster relief data are the Development
Assistance Committee (DAC) of the OECD and the Financial Tracking System
(FTS) of the UN Office for the Coordination of Humanitarian Affairs. The DAC
reports annual spending on emergency aid by donor–recipient country pair. This
data is reasonably comprehensive, but includes emergencies other than natural
disasters. The FTS is more specific; it records how much international relief was
provided to individual natural disasters. But the FTS data is far from comprehen-
sive, since reporting is voluntary. For this reason, we will mainly use the DAC data
and restrict the attention to years when natural disasters struck.
According to the Development Assistance Committee data, net disbursements
on emergency and distress relief over 1995–2004 were around $4.6 billion per year
(in constant 2004 dollars). A few donors supplied most of this aid; in fact, eleven
countries provided 90 percent. The United States has been the largest donor by far,
accounting for around a third of all relief. However, European countries as a group
account for 57 percent of the funds. Among them, the largest donors are the
Netherlands, the United Kingdom, Sweden, Norway, Germany, and France, each
contributing 6 –9 percent. Other large donor countries have been Canada, Japan,
and Australia. The bulk of the relief has been given to Africa (40 percent), Asia
(35 percent), and Europe (19 percent).
We will not analyze relief from private individuals through organizations such
as the Red Cross, OXFAM, UNICEF, and the Bill and Melinda Gates Foundation,
since this is not contained in the Development Assistance Committee data. Private
individuals contributed 10 percent of the total relief funds to natural disasters over
2000 –2004, based on records of the Financial Tracking Service of the United
Nations.

Why Do Countries Provide Disaster Relief?

In July 2001, heavy rains caused floods in Poland, killing approximately
30 people and affecting 15,000. This disaster was met by international relief efforts
totaling around $15 million, as reported by the FTS. In that same year, a flood in
Thailand killed 100 people and affected 450,000, while one in Angola killed 50 and
affected 40,000 people, and another in Brazil killed 50 and affected 2,200. Here are
the corresponding relief efforts: Thailand received $25,000; Angola, $75,000; and
Brazil, $230,000.
 David Strömberg 213

Can the reasons countries provide disaster relief help explain such divergent
responses? Private donations are probably based on humanitarian motivations for
the most part, and this motive is certainly the most stressed in official guidelines for
relief. The humanitarian motive is to provide relief where it would do most good,
in terms of saving lives and reducing human suffering. Arguably, this motivation
would drive relief to larger disasters in low-income countries, although such an
outcome is not completely obvious, because low-income countries may have poor
infrastructure and more prevalent corruption limiting the effectiveness of relief
(see the discussion in Collier and Dollar, 2002).
However, people must be aware of a disaster in order to help, and so disasters
in the media spotlight may get more relief. Donor government officials may also
react more swiftly to disasters reported in the media.
Donor countries may also provide relief with an eye to their own economic or
geostrategic political interest (for example, Alesina and Dollar, 2000, and the
references therein). Large disasters may destabilize governments. Aid to friendly
governments could help these stay in power; withholding aid from not-so-friendly
governments could destabilize them (Drury, Olson, and Van Belle, 2005). Disaster
relief may also be used to protect investments in foreign countries, driving relief
towards countries where the donors have large economic stakes.
People may be more moved by the suffering of those close by— geographically
or culturally—and media coverage also may focus more on these disasters. Geo-
strategic and economic interests are likely to be more important in countries
geographically close to donors or in countries where donors have historical or
cultural ties. We will investigate the role of geographical distance, common colonial
past, common language, bilateral trade flows, and similar voting patterns in the
United Nations in predicting aid flows. This approach is very much in line with
research on general aid flows, as in Alesina and Dollar (2000).

Disaster Severity and News

Given that the United States is the largest single donor, I will focus here on
relief provided by the USAID’s Office of Foreign Disaster Assistance (OFDA)
between 1968 and 2002. The OFDA is generally the first U.S. agency to be on the
scene of a disaster, and it also influences other relief efforts. For example, the
largest U.S. disaster relief program, under the Office of Food for Peace, requires an
OFDA disaster declaration to trigger some of its disaster assistance. Although the
OFDA only contributes around 20 percent of total U.S. disaster relief, each dollar
spent by the OFDA on a particular disaster was, on average, matched by four dollars
from other U.S. agencies in the fiscal year 2002.
The OFDA relief decisions respond strongly to fatalities. The OFDA provided
relief to 11 percent of the disasters with less than 100 fatalities, to 30 percent of the
disasters with fatalities ranging from 100 to 1000, and to 60 percent of the disasters
with more than 1,000 fatalities. These large differences partly arise from other
factors correlated with many deaths. Table 5 presents these results in a regression
framework. In the first column, the dependent variable is a dummy variable for
214 Journal of Economic Perspectives

Table 5
Disaster Relief, Need, and News

Whether U.S. OFDA Whether U.S. OFDA DAC donor country

provided relief ? provided relief ? provided relief ?

Killed (log10) 0.10 0.02 0.05

(0.01)*** (0.04) (0.01)***
Affected (log10) 0.06 0.04 0.014
(0.01)*** (0.01)*** (0.004)***
Real GDP per capita ⫺0.08 ⫺0.23 ⫺0.29
(log10)
(0.03)*** (0.07)*** (0.02)***
Disaster in the news 0.09 0.76 0.04
(0.02)*** (0.32)** (0.02)***
Type of regression OLS IV OLS
Observations 4755 4755 17837
R-squared 0.32 0.27

Notes: Results from ordinary least squares (OLS) regressions, except column 2, which shows an instru-
mental variable (IV) regression where news coverage has been instrumented by whether there was a
concurrent Olympic game. The dependent variable in the first two columns is a dummy variable
indicating whether the U.S. Office of Foreign Disaster Assistance (OFDA) provided relief to an
individual disaster. In the third column, the dependent variable indicates whether each Development
Assistance Committee (DAC) donor country provided emergency aid to a potential recipient country in
a year in which a disaster struck. All regressions include year fixed effects and control for the logarithm
of population. The first two columns also include fixed effects for the continent, the disaster type, and
the month of the disaster. The third column includes a donor fixed effect. Robust standard errors,
clustered by recipient country in parentheses.
* significant at 10 percent, ** significant at 5 percent, *** significant at 1 percent.

whether OFDA provided relief for a given disaster. The explanatory variables are
the number killed; the number affected; the real per capita GDP of the country;
and whether the disaster was in the news. The “news” variable measures whether a
disaster was covered by one of the major U.S. television networks: ABC, CBS, NBC,
or CNN. For example, Dan Rather covered the flood in Poland 2001 on the CBS
evening news, but no network covered the flood in Angola. The coefficients in the
regression suggest that an increase in fatalities by a factor of ten is associated with
an increase in the probability of receiving relief by 10 percent. Around 29 percent
of the disasters in the news received relief, as compared to 13 percent among those
that were not covered. Taking into account factors such as the number of killed and
affected, disasters covered by U.S. network news are 9 percent more likely to get
relief than other similar disasters.
However, this simple regression is clearly insufficient to show that the media
affects disaster relief. For example, it could be the case that both media coverage
of natural disasters and disaster relief efforts are focused on more severe disasters,
as measured in some way that we are unable to account for by the reported number
of killed and affected. News coverage could also be affected by the geographic or
cultural closeness between the recipient country and the primarily Western donors.
To investigate further the extent to which news coverage influences disaster
 Natural Disasters, Economic Development, and Humanitarian Aid 215

relief, Eisensee and Strömberg (2007) use the availability of other newsworthy
material as an instrument for whether or not the disaster was in the news. In other
words, they are asking whether a natural disaster is less likely to receive relief
because news about this disaster was crowded out by news about, for example, the
Olympics, or the O. J. Simpson trial. Thus, the first stage of the estimation process
uses, as the dependent variable, whether the natural disaster was covered by the
news media and, as the instrumental explanatory variable, measures of the amount
of competing news material. Then the coefficients from this regression are used to
calculate an estimate of media coverage of natural disasters, which is then plugged
into the original regression. The second column shows the results from this type of
estimation, using whether there was a concurrent set of Olympic Games as instru-
ment. The significant coefficient on news coverage indicates that news coverage of
natural disasters does have a causal effect on relief efforts. The estimated coefficient
is larger than the ordinary least squares estimate.
Moreover, the news effect biases relief towards disasters that are more news-
worthy, in the journalistic sense of the word. Earthquakes and volcanoes are more
often covered by the news than equally severe disasters of other types. For example,
Eisensee and Strömberg (2007) estimate that for every person that dies in a volcano
disaster, over 2,000 people must die in a drought to receive the same expected news
coverage. Similarly, disasters in Europe or the Americas are more covered than
similar disasters in Asia, Africa, and the Pacific. This coverage induces a relief bias
favoring the disasters that appear more in the news for reasons unrelated to
severity.
The third column of Table 5 analyzes relief from a much larger set of donors.
This column uses data from the Development Assistance Committee (DAC) cov-
ering the ten years 1995–2004, for 23 donor countries. This relief data is merged
with information on 143 potential recipient countries that experienced at least one
natural disaster during these ten years according to the CRED. The regression only
includes years when natural disasters struck, in order to reduce the problem that
this data also covers emergency relief given in complex emergencies like civil wars.
The dependent variable in this column indicates whether a DAC-donor provided
relief to a potential recipient country in a particular year (for example, the United
States provided relief to Thailand in 2001). Here, we control for the total number
of killed and affected in all natural disasters in that country in that year (for
example, the seven natural disasters in Thailand in 2001 killed a total of 222 people
and affected 470,000). This larger group of donors is more likely to give relief the
larger is the disaster and the poorer is the potential recipient country, and if the
disaster is in the news.

Donor–Recipient Relations
Data on bilateral emergency aid can be used to investigate the effects of
donor–recipient relations, as measured by factors like colonial ties and distance. In
effect, we can compare relief from different donors: for example, how much the
216 Journal of Economic Perspectives

United States and France gave to Thailand and Poland, respectively, in 2001. The
aid flows are converted into constant 2004 dollars.
Starting with the effect of colonial ties, the former colonial powers are
easily identified by the number of countries with which donors ever had colonial
ties.4 The donors with most colonial ties are the United Kingdom (ties to
35 percent of the recipients), France (19 percent), Spain (11 percent), Turkey
(7 percent), and Germany and Portugal (4 percent). The majority of the
recipient countries have colonial ties with just one donor, although some have
ties with more than one country. Notably, Palau and the Federal States of
Micronesia have ties to four donors. Colonial ties are completely lacking for 36
states, notably China, Iran, Thailand, and a set of states around the Caspian and
Black Seas. The spending patterns appear very different for the old colonial
powers. France, Spain, Portugal, and Turkey clearly give more to their former
colonies. For example, 40 percent of the former Spanish colonies receive
Spanish relief as compared to 10 percent of the other countries in an average
year. Other former colonial powers show no such preference, notably the larger
donor countries like Germany and the United Kingdom.
Simple comparisons like these are open to alternative interpretations. For
example, former colonies of some countries might be faring worse and need more
relief. To analyze this more carefully, we regress emergency relief on some possible
determinants. The unit of observation is donor country by potential recipient
country by year. Only years when recipient countries were struck by natural disas-
ters are included in the sample. We include a dummy variable for each recipient-
year pair, thus accounting for all factors constant for a recipient country in a given
year. The results are shown in Table 6. The first column has as the dependent
variable whether a donor provided relief, the second has (the natural logarithm of)
the amount of relief provided in those years when any relief was provided, and the
third has (the natural logarithm of the) share of each donor’s spending.5
Colonial history is clearly important. Having a common colonial history in-
creases the probability of getting relief by 8 percentage points (estimated coeffi-
cient 0.08 in the table). If the donor is located in Latin Europe (France, Spain,
Portugal, and Italy), the probability increases an additional 10 percentage points.
These numbers are very large, given that the average donor only gives relief to
20 percent of the recipient countries. At that level, an increase of 18 percentage
points roughly doubles the chance of getting relief. Colonial history is also of
importance for the amount of relief, when relief is provided. This is 46 percent
higher for countries with colonial ties, and three times higher for countries with
colonial ties to Latin Europe (with spending in logs, a coefficient of 0.38 on

4
From the dataset dist_cepii, provided by the Centre d’Etudes Prospectives et d’Informations
Internationales.
5
Here we assume that, conditional on positive amounts of aid being given, the conditional expectation
of the log amount is linear. The estimated coefficients do not change significantly if both equations are
estimated jointly using a Heckman selection model.
 David Strömberg 217

Table 6
Disaster Relief and Donor–Recipient Relations

Amount relief provided (loge), Share provided

Is relief provided? when relief is provided (loge)

Colony 0.08 0.38 0.66

(0.03)*** (0.18)** (0.16)***
Latin European colony 0.10 0.66 0.66
(0.05)** (0.27)** (0.24)***
Common language 0.02 0.38 0.22
(0.02) (0.15)** (0.09)**
Geographic distance ⫺0.11 ⫺0.79 ⫺1.13
(0.05)** (0.38)** (0.23)***
Trade value (log10) 0.08 0.24
(0.01)*** (0.10)**
UN friend ⫺0.26 1.20
(0.11)** (0.62)*
Observations 15819 4155 22910
R-squared 0.40 0.55 0.45

Notes: Results from ordinary least squares regressions. The dependent variables are the following: first
column, whether a donor country provided relief to a recipient country struck by a natural disaster that
year; second column, the logarithm of the amount provided, when relief was provided; third column,
the logarithm of the share of all relief provided by that donor ⫹ 0.00001. All regressions include donor
and recipient-year fixed effects. Robust standard errors, clustered by recipient country, in parentheses.
* significant at 10 percent, ** significant at 5 percent, *** significant at 1 percent.

colonial ties implies that relief from the donor is increased by a factor e0.38 ⫽ 1.46
and so is 46 percent higher for colonies). Including both the increased chance of
getting relief and the increased amount when relief is given, former colonies of
Latin European countries can expect expenditures almost four times as high from
their former colonial power than other aid recipients.
Donors also give more to countries with a common language. The dummy
variable “common language” indicates that at least 9 percent of the populations in
both countries speak a common language. For example, Canada has a common
language with 46 percent of the recipient countries, where French or English is
spoken by at least 9 percent of the population. The estimates in Table 6 imply that
having a common language does not significantly affect whether aid is given.
However, when relief is provided, the amount given is around 46 percent higher
(e0.38⫽1.46).
More distant countries are less likely to receive relief. The variable “geographic
distance” contains the distance between the capitals of the donor and the recipient
country. The unit is half the circumference of the earth, so a measurement of 1 means
that the country is at the other side of the earth (roughly 20,000 km). The estimated
coefficients imply that a country on the other side of the earth is 11 percent less likely
to receive relief than a country at distance zero. To place this estimate in perspective,
recall that the estimated probability of receiving relief increases by 5 percent when
fatalities increase by a factor of ten. Combined, the estimates imply that to have the
218 Journal of Economic Perspectives

same chance of receiving relief, a country at the other side of the earth must have
160 times as many fatalities as a country at zero distance, all else equal. A country on the
other side of the earth also gets 45 percent as much relief when relief is provided
(e–.79 ⫽ .45). The total effect is that a country on the other side of the earth gets
one-third of the funds of a country at distance zero.
To investigate the hypothesis that economic interest motives may drive disaster
relief, we include the variable “trade value,” which is the (base-ten logarithm of the)
bilateral trade flow between the donor and the recipient. (The data is provided by
UN Comtrade.) Disaster relief is clearly increasing in the importance of the trade
partner. The estimates imply that as trade flows increase by a factor of ten, the
chance of receiving relief increases by 8 percent and the amount of relief increases
by 27 percent (e0.24 ⫽ 1.27).
This evidence that economic donor interests affect relief is suggestive, but far
from conclusive. The problem is that trade and aid would be positively related even
if economic interests played no role for relief. The two are driven by similar factors,
such as geographic and cultural closeness. For example, the measures of distance,
common language, and colonial relationships that we have used to explain aid are
standard in tests of the “gravity model” of trade (for example, Glick and Rose,
2002). We cannot control for all factors driving both trade and aid, and part of the
positive relation between the two is due to this.
Finally, we investigate whether foreign policy motives drive relief to friendly
governments. This test requires a measure of government friendliness. A simple
proxy is whether the donor has a formal alliance with the country, which is tracked
in the Correlates of War Formal Interstate Alliance Data Set (Gibler and Sarkees,
2003, v3.03). Another proxy is the similarity between the donor’s and the recipi-
ent’s voting pattern in the United Nations (Gartzke, 2006).
We find little evidence that these measures of government friendliness are of
importance for disaster relief. Table 6 shows the effect of similar voting patterns in
the United Nations. A recipient with similar voting patterns to a donor is less likely
to receive relief from that donor, and the effect on the amount is positive, although
only significant at the 10 percent level. The net effect is not significant. Using the
number of alliances instead of voting patterns yields similar results. Other studies,
notably Alesina and Dollar (2000) and Drury, Olson, and Van Belle (2005) have
found aid to be positively related to these measures of government friendliness.
The difference in results might be a reflection of our sample period being post–
Cold War.

Which Countries Lose Out?

Which countries are likely to get less relief, given the same need? Aggregating
across donors, the estimated effect of distance, colonial ties, and common language on
total relief funds is shown for continents and a selection of countries in Table 7.
The first column contains the average distance in kilometers to the donors,
weighted by each donor’s share of relief expenditures. The second column shows
how much this distance is decreased because of colonial ties. Having a colonial tie
 Natural Disasters, Economic Development, and Humanitarian Aid 219

Table 7
Average Aid-Dollar Distance (km), and Implied Effect on Relief

Latin
Geographic European Common Generalized Implied effect
Continents distance Colony colony language distance on relief

Europe 3,728 ⫺386 0 ⫺191 3,295 1.86

America 8,767 ⫺548 ⫺299 ⫺713 4,576 1.60
Africa 7,180 ⫺582 ⫺283 ⫺680 5,865 1.54
Asia 7,236 ⫺328 ⫺86 ⫺244 7,381 1.37
Pacific 14,029 ⫺815 ⫺173 ⫺816 10,967 1.00

Countries

Algeria 4,296 ⫺677 ⫺681 ⫺628 2,309 2.57

Poland 3,873 ⫺700 0 0 3,173 2.52
Lebanon 5,798 ⫺677 ⫺681 ⫺2,299 2,141 2.48
Anguilla 5,717 ⫺955 0 ⫺1,869 2,893 2.26
Gambia 6,049 ⫺955 0 ⫺1,869 3,225 2.24
China 8,729 0 0 0 8,729 1.50
Brazil 8,734 ⫺12 ⫺12 ⫺4 8,706 1.47
Swaziland 10,758 ⫺955 0 0 9,803 1.41
East Timor 13,310 ⫺12 ⫺12 0 13,286 1.03
Tonga 14,245 ⫺955 0 0 13,290 1.00

Note: The first column shows the average distance a relief dollar would have to travel, if it was sent from
the donor country’s capital to the recipient country’s capital. The second column shows the reduction
in distance that would produce the same expected increase in relief as having colonial ties to the donor.
The following two columns show analogous numbers for “Latin European colony” and “common
language.” The generalized distance is the sum of geographic and cultural distance. The final column
shows the implied effect on relief. This column is computed by predicting the share of each donor’s
budget going to a recipient country (from column 3 of Table 6), multiplying this by the size of the
donor’s budget, and aggregating across donors.

has the same effect on expected aid flows as moving the country 12,000 kilometers
closer to the donor (based on estimates from Table 6, column 3). However, most
countries only have a colonial relationship to one country. For example, East Timor
and Brazil are former Portuguese colonies. Portugal is a minor donor, only con-
tributing 0.1 percent of the aid funds. The colonial tie only decreases the distance
to these Portuguese funds, and when weighted in this way, the average aid-dollar
distance only falls by 12 kilometers (as shown near the bottom of the second
column). In contrast, Tonga has a colonial relationship with the United Kingdom,
which provides 8 percent of the relief funds. This decreases its average distance to
the aid dollar by 955 kilometers. The columns for a Latin European colony and
common language are constructed in an analogous fashion.
Aggregating across donors, language is more important than colonial status,
since language connects the recipient to multiple donors. For example, Lebanon
has language ties to both English- and French-speaking donors, which move
Lebanon over 2,000 kilometers closer to the average aid dollar.
220 Journal of Economic Perspectives

The last column reports the estimated effect on relief. This column is com-
puted by predicting the share of each donor’s budget going to a recipient country
(from column 3 of Table 6), multiplying this by the size of the donor’s budget, and
aggregating across donors.
The differences are large. Compared to disasters in Europe, the estimates
imply that this general donor–recipient distance reduces funds to Asian disasters by
36 percent, to African disasters by 21 percent, and to American disasters by 16
percent.
The individual middle- and low-income countries “closest” to the large donors
are located in Eastern Europe (notably states from former Yugoslavia, the Baltic
states, Poland, and the Czech Republic) and states in the Middle East and northern
Africa (Lebanon, Jordan, Egypt, Algeria, and Morocco) and islands in the Carib-
bean (Anguilla, St. Kitts and Nevis, and Montserrat). For example, Lebanon lies
geographically close to the large European donors, has colonial ties to France, and
language ties to both French- and English-speaking donors. Note that generalized
distance reduces relief less to Poland than to Lebanon, even though the average
distance to the aid-dollar is longer. Relief decreases more than proportionally to
distance, which means that it is important to have a few large donors very close.
At the other end, we find islands in the South Pacific, such as Tonga,
Tuvalu, and the Solomon Islands, to be most distant from the important donors.
Another group at this end of the spectrum includes remote former Portuguese
colonies, notably East Timor. These countries generally lack language ties to the
major donors, and, as mentioned, their colonial ties are less valuable because
Portugal is a very minor donor. The estimated effect of donor distance makes
disasters in Poland receive 2.5 times more than similar disasters in East Timor.
Other former Portuguese colonies, such as Angola, Brazil, and Mozambique,
are also far from the average aid-dollar. A third group far from the large donors
are Asian countries without colonial ties, such as China, Nepal, Bhutan, and
Thailand.
What is the cost of these differences in relief in terms of human lives or
suffering? To my knowledge, there are no estimates of the net effect of moving
relief funds from one disaster to another. It is straightforward to study the kinds
of relief given: how much cash spent per capita, amount of blankets, food, and
clean water provided. However, it is harder to estimate their effects: to what
extent did relief save lives or improve health and nutrition? The underlying
problem is that provision of relief depends on these outcomes. For example,
comparing mortality rates between victims that have received relief and those
who have not will obviously not do, because relief was presumably given to those
victims still alive whose health was most endangered. From a technical point of
view, randomized field experiments would be desirable, but in the context of
disaster relief, such an approach is impractical. Seeking natural experiments
where relief provision differs for reasons unrelated to health status might be a
way forward.
 David Strömberg 221

Conclusions

Although the extreme natural phenomena causing natural disasters are not
more severe in developing countries, their consequences are. Significantly more
people are affected and killed in disasters striking low- or middle-income countries,
taking into account the type of disaster and the population at risk. Our estimates
imply that fatalities would have been around 20 percent higher today, absent
economic growth since 1960. This average hides strong regional differences, with
considerably larger fatality reductions in Europe and Asia, and no reductions in
Africa. Looking forward, the present strong growth rates in China and India hold
promise of further reductions in the toll from natural disasters. These two countries
hold more than one-third of the world population and three-quarters of all people
affected by disasters over the period 1980 to 2004.
International relief for natural disasters does increase with the severity of the
disaster, as measured by the number of killed and affected, and also rises when the
income of the affected country is lower. However, relief is also driven by factors
other than need. News coverage appears to drive disaster relief. Donors also give
more to countries that lie closer, and with which they share a common language
and colonial ties. These effects are sizeable. For example, to have the same chance
of receiving relief from a particular donor, a country at the other side of the earth
must have 160 times as many fatalities as a country at zero distance from that donor.
A country without colonial ties must have 50 times as many fatalities to have the
same chance of receiving relief as a former colony. For this reason, countries that
lie far away from the major donors are systematically disfavored when it comes to
disaster relief. Aggregated across donors, the estimated combined effects of dis-
tance, colonial ties, and common language imply that a disaster in Poland or
Algeria is expected to receive around 2.5 times as much relief as a similar disaster
in East Timor or Tonga.

References

Albala-Bertrand, J. 1993. “Political Economy Besley, Timothy, and Robin Burgess. 2002.
of Large Natural Disasters.” Oxford: Clarendon “The Political Economy of Government Respon-
Press. siveness: Theory and Evidence from India.”
Alesina, Alberto, and David Dollar. 2000.” Quarterly Journal of Economics, 117(4): 1415–51.
Who Gives Foreign Aid to Whom and Why?” Charvériat, Céline. 2000. “Natural Disasters in
Journal of Economic Growth, 5(1): 33– 63. Latin America and the Caribbean: An Overview
Anbarci, Nejat, Monica Escaleras, and of Risk.” Inter-American Development Bank
Charles A. Register. 2005. “Earthquake Fatali- Research Department Working Paper 434.
ties: The Interaction of Nature and Political Chou, Yiing-Jenq , Nicole Huang, Cheng-Hua
Economy.” Journal of Public Economics, 89(9 – Lee, Shu-Ling Tsai, Long-Shen Chen, and Hong-
10): 1907–33. Jen Chang. 2004. “Who Is at Risk of Death in an
222 Journal of Economic Perspectives

Earthquake?” American Journal of Epidemiology, Kahn, Matthew E. 2005. “The Death Toll from
160(7): 688 –95. Natural Disasters: The Role of Income, Geogra-
Coburn, A. W., R. J. S. Spence, and A. Pomo- phy, and Institutions.” Review of Economics and
nis. 1994. Vulnerability and Risk Assessment, 2nd Statistics, 87(2): 271– 84.
ed. New York: Disaster Management Training Kaufmann, Daniel, Aart Kraay, and Massimo
Programme and United Nations Development Mastruzzi. 2006. “Governance Matters V: Aggre-
Programme. gate and Individual Governance Indicators for
Collier, Paul, and David Dollar. 2002. “Aid 1996 –2005.” [Link]
Allocation and Poverty Reduction.” European INTWBIGOVANTCOR/Resources/1740479-
Economic Review, 46(8): 1475–1500.
1150402582357/2661829-1158008871017/gov_
Dilley, Maxx, Robert S. Chen, Uwe Deich-
matters_5_no_annex.pdf.
mann, Arthur L. Lerner-Lam, and Margaret
Masters, Roger D., and Christopher Kelly, eds.
Arnold. 2005. “Natural Disaster Hotspots: A
1990. The Collected Writings of Rousseau, vol. 3.
Global Risk Analysis.” Washington, DC: World
Hanover: The University Press of New England.
Bank.
Drury, A. Cooper, Richard Stuart Olson, and Mileti, Dennis S. 1999 Disasters by Design: A
Douglas A. Van Belle. 2005. “The Politics of Reassessment of Natural Hazards in the United
Humanitarian Aid: U.S. Foreign Disaster Assis- States.” Washington, DC: Joseph Henry Press.
tance, 1964 –1995.” The Journal of Politics, 67(2): Peduzzi, Pascal. 2005. “Is Climate Change In-
454 –73. creasing the Frequency of Hazardous Events?”
Dynes, Russell R. 1999. “The Dialogue Be- Environment and Poverty Times, February 22, p. 7.
tween Voltaire and Rousseau on the Lisbon Peduzzi, Pascal, Hy Dao, and Christian
Earthquake: The Emergence of a Social Science Herold. 2005. “Mapping Disastrous Natural Haz-
View.” University of Delaware Disaster Research ards Using Global Datasets.” Natural Hazards,
Center Preliminary Paper 293. 35(2): 265– 89.
Eisensee, Thomas, and David Strömberg. 2007. Rasumssen, Tobias. 2004. “Macroeconomic
“News Floods, News Droughts, and U.S. Disaster Implications of Natural Disasters in the Carib-
Relief.” Quarterly Journal of Economics, 122(2):693– bean.” International Monetary Fund Working
728. Paper WP/04/224.
Gartzke, Erik. 2006. “The Affinity of Nations Richardson, Lewis F. 1960. Statistics of Deadly
Index, 1946 –2002.” Mimeo, Columbia Univer- Quarrels. Edited by Quincy Wright and C. C.
sity. Data, codebook, etc. available at http:// Lienau. Pittsburgh: Boxwood Press.
[Link]/⬃eg589/[Link]. Sen, Amartya. 1990. “Public Action to Remedy
Glick, Reuven, and Andrew K. Rose. 2002.
Hunger.” Arturo Tanco Memorial Lecture, Au-
“Does a Currency Union Affect Trade? The
gust 2, London. [Link]
Time Series Evidence.” European Economic
sen/[Link].
Review, 46(6): 1125–51.
Sheets, Bob, and Jack Williams. 2001. “Hurri-
Heston, Alan, Robert Summers, and Bettina
Aten. 2002. Penn World Table Version 6.1, Cen- cane Watch: Forecasting the Deadliest Storms
ter for International Comparisons at the Univer- on Earth.” Vintage Books at Random House.
sity of Pennsylvania (CICUP), October. U.S. Agency for International Development
Intergovernmental Panel on Climate Change. (USAID). 2005. “Major Functional Series 200
2007. Climate Change 2007: The Physical Science Program Assistance: ADS 251—International
Basis: Summary for Policymakers. Disaster Assistance.” [Link]
Isham, Jonathan, Daniel Kaufmann, and Lant policy/ads/200/[Link].
H. Pritchett. 1997. “Civil Liberties, Democracy, Walter, Jonathan, ed. 2003. World Disasters
and the Performance of Government Projects.” Report 2003. Geneva: International Federation of
World Bank Economic Review, 11(2): 219 – 42. Red Cross and Red Crescent Societies.
This article has been cited by:

1. Lodewijk Smets, Stephen Knack, Nadia Molenaers. 2013. Political ideology, quality at entry and the
success of economic reform programs. The Review of International Organizations . [CrossRef]
2. Ashlesha Datar, Jenny Liu, Sebastian Linnemayr, Chad Stecher. 2013. The impact of natural disasters
on child health and investments in rural India. Social Science & Medicine 76, 83-91. [CrossRef]
3. Alejandro Quiroz Flores, Alastair Smith. 2012. Leader Survival and Natural Disasters. British Journal
of Political Science 1-23. [CrossRef]
4. C. Tubilewicz. 2012. The politics of compassion: examining a divided China's humanitarian assistance
to Haiti. International Relations of the Asia-Pacific 12:3, 449-481. [CrossRef]
5. Prema-chandra Athukorala. 2012. Indian Ocean Tsunami: Disaster, Generosity and Recovery. Asian
Economic Journal 26:3, 211-231. [CrossRef]
6. Joakim Kreutz. 2012. From Tremors to Talks: Do Natural Disasters Produce Ripe Moments for
Resolving Separatist Conflicts?. International Interactions 38:4, 482-502. [CrossRef]
7. Vikash Ramiah. 2012. Effects of the Boxing Day tsunami on the world capital markets. Review of
Quantitative Finance and Accounting . [CrossRef]
8. PATRICK WARD, GERALD SHIVELY. 2011. Vulnerability, Income Growth and Climate Change.
World Development . [CrossRef]
9. Ingmar Schumacher, Eric Strobl. 2011. Economic development and losses due to natural disasters:
The role of hazard exposure. Ecological Economics . [CrossRef]
10. Suresh De Mel, David McKenzie, Christopher Woodruff. 2011. Enterprise Recovery Following
Natural Disasters*. The Economic Journal no-no. [CrossRef]
11. Namsuk Kim. 2011. How much more exposed are the poor to natural disasters? Global and regional
measurement. Disasters no-no. [CrossRef]
12. Aziz Hebal, Boualem Remini. 2011. Choix du modèle fréquentiel le plus adéquat à l’estimation des
valeurs extrêmes de crues (cas du nord de L’Algérie). Canadian Journal of Civil Engineering 38:8,
881-892. [CrossRef]
13. Michael Dunford, Li Li. 2011. Earthquake reconstruction in Wenchuan: Assessing the state overall
plan and addressing the ‘forgotten phase’. Applied Geography 31:3, 998-1009. [CrossRef]
14. William Solecki, Robin Leichenko, Karen O’Brien. 2011. Climate change adaptation strategies
and disaster risk reduction in cities: connections, contentions, and synergies. Current Opinion in
Environmental Sustainability 3:3, 135-141. [CrossRef]
15. Chang Hoon Oh, Jennifer Oetzel. 2011. Multinationals' response to major disasters: how does
subsidiary investment vary in response to the type of disaster and the quality of country governance?.
Strategic Management Journal n/a. [CrossRef]
16. Monica Escaleras, Charles A. Register. 2010. Fiscal decentralization and natural hazard risks. Public
Choice . [CrossRef]
17. Günther Fink, Silvia Redaelli. 2010. Determinants of International Emergency Aid—Humanitarian
Need Only?. World Development . [CrossRef]
18. Eliza M. Lis, Christiane Nickel. 2010. The impact of extreme weather events on budget balances.
International Tax and Public Finance 17:4, 378-399. [CrossRef]
19. Martin Gassebner, Alexander Keck, Robert Teh. 2010. Shaken, Not Stirred: The Impact of Disasters
on International Trade. Review of International Economics 18:2, 351-368. [CrossRef]
20. Chang Hoon Oh, Rafael Reuveny. 2010. Climatic natural disasters, political risk, and international
trade. Global Environmental Change 20:2, 243-254. [CrossRef]
21. Amir AghaKouchak, Nasrin Nasrollahi. 2010. Semi-parametric and Parametric Inference of Extreme
Value Models for Rainfall Data. Water Resources Management 24:6, 1229-1249. [CrossRef]
22. C. Y. Olivola, N. Sagara. 2009. Distributions of observed death tolls govern sensitivity to human
fatalities. Proceedings of the National Academy of Sciences 106:52, 22151-22156. [CrossRef]
23. Philippe Régnier, Bruno Neri, Stefania Scuteri, Stefano Miniati. 2008. From emergency relief to
livelihood recovery: Lessons learned from post-tsunami experiences in Indonesia and India. Disaster
Prevention and Management 17:3, 410-430. [CrossRef]