Content uploaded by Monica Escaleras
Author content
All content in this area was uploaded by Monica Escaleras
Content may be subject to copyright.
Public Sector Corruption and Natural Disasters:
A Potentially Deadly Interaction
Monica Escaleras Nejat Anbarci Charles A. Register*
August 30, 2006
Abstract: A number of recent studies have, separately, addressed the effects of public sector
corruption and natural disasters. In this paper, we intersect these lines of research to assess
whether corruption in the public sector plays a role in the havoc wrought by large scale natural
disasters, using major earthquakes as the example. We first develop a brief theoretical model of
the relation between these two variables and then empirically test the proposition by analyzing
344 major quakes occurring in 42 countries during the 1975 through 2003 period. We use a
Negative Binomial estimation strategy that takes into account the endogenous nature of
corruption and controls for a number of other factors such as earthquake frequency, magnitude,
distance from population centers, and a country’s level of development which have been shown
to influence a quake’s destructiveness. The results provide strong evidence that public sector
corruption is both positively and significantly related to the death toll a given earthquake takes on
a population.
Keywords: Earthquake fatalities, corruption, institutional variables.
JEL Classification(s): D31, H41, P16
____________
* Monica Escaleras: (Corresponding Author) Florida Atlantic University, Department of Economics, 777 Glades
Road, Boca Raton, FL 33431; E-mail: mescaler@fau.edu; Phone: (561) 297-1312.
Nejat Anbarci: Florida International University, Department of Economics, University Park, DM 316, Miami, FL
33199.
Charles A. Register: Florida Atlantic University, Department of Economics, 777 Glades Road, Boca Raton, FL 33431.
**We would like to thank Gokhan Karahan for his very useful comments on the intersection of earthquakes and
corruption in Turkey. Also, we thank Eric Chiang, Cem Karayalcin, Lonnie Stevans, and Mehmet Ulubasoglu for their
assistance in developing this project.
1
Public Sector Corruption and Natural Disasters:
A Potentially Deadly Interaction
Earthquakes don’t kill people: collapsing buildings do. While earthquakes may not be
preventable, it is possible to prevent the disasters they cause. In the past 15 years, there
have been more than 400 recorded earthquakes in 75 countries rendering almost 9
million people homeless, injuring 584,000 and causing 156,00 deaths. Many of these
deaths were the result of buildings that folded in on themselves because concrete was
diluted, steel bars were excised, or otherwise substandard building practices were
employed. It is difficult to evaluate the extent to which corruption might have played a
role. However, the accompanying examples…illustrate that the marriage of corrupt
contractors and corrupt building inspectors and other public officials resulted in ignored
building codes, lax enforcement and the absence of on-site inspection, which is deadly
when it occurs in earthquake-prone areas.
1
1. Introduction
In this study we bring together two veins of research that have, separately, generated a good
deal of recent interest in the literature on political economy; public sector corruption and natural
disasters. Prior studies have identified a number of deleterious effects of public sector corruption
on the economic, political, and social performance of countries.
2
The initial focus of this research
dealt with the summary issue of economic growth. Mauro (1995), relying on data for as many as
60 countries and various indices of corruption, found corruption to lower investment and, thus, to
limit economic growth. This result has since been confirmed by Mo (2001), Pellegrini and
Gerlagh (2004), and Meon and Sekkat (2005). Also relying on international data, Alesina and
1
Quote taken from James Lewis (2005), who is an architect, consultant, and writer on environmental hazards. He is
also a visiting fellow in development studies at the University of Bath.
2
This should not be taken to imply that corruption can have no positive effects. Discussions of such potentially
positive effects go back as far as Leff (1964), Leys (1965), and Huntington (1968) and can more recently be found in
Bardhan (1997), Beck and Maher (1986), and Lien (1986). A good summary analysis of the debate surrounding
corruption’s positive and negative effects is offered by Meon and Sekkat (2005).
2
Weder (2002) report that countries with high levels of perceived public sector corruption receive
less foreign aid than their less corrupt counterparts. Similarly, Habib and Zurawicki (2002) find
that countries perceived to be relatively corrupt are less attractive to foreign direct investment.
From a more local market perspective, Bliss and Di Tella (1997) report that corrupt government
regulators often assist in the erection of barriers to entry to provide monopoly profits for favored
producers which then become a source for income transfers from consumers to the corrupt
officials in the forms of bribes and other kickbacks from the empowered, protected producers.
Research has also been directed toward the impact of public sector corruption or
disproportionate public sector influence on political decision making, primarily within the U.S.
An early example is that of Anderson and Tollison (1991), who show that while federal spending
during the great depression was related to a specific locality’s economic need, strong evidence
also exists pointing to relief efforts flowing to parts of the country with unusual congressional
clout. In a unique analysis of voter turnout, Karahan, Coats, and Shughart (2006) stress a
demand-side approach to voting in which candidates for office expend greater effort for votes the
greater is the potential payoff of winning the election. Analyzing data on county supervisor
elections in Mississippi during the late 1980’s, the authors show that turnout was significantly
greater in those counties where supervisor corruption had been recently uncovered in a FBI sting
operation. That is, the greater potential return from winning an election in a corrupt county lead
those seeking office to more actively pursue votes. Also focusing on the U.S., Depken and
Lafountain (2006) find that state bond ratings increase in states with relatively higher levels of
corruption. Pointing to the pervasive nature of public sector corruption, Anbarci, Escaleras, and
Register (2006) report that traffic deaths are significantly higher in countries with relatively high
levels of public sector corruption. Finally and most closely related to the current paper, Garrett
and Sobel (2003) offer evidence that suggests that both a president’s declaration of a disaster area
and the flow of Federal Emergency Management (FEMA) money to such areas are highly
correlated with the political importance of a particular state to the sitting president and to a state’s
3
congressional representation on FEMA oversight committees.
As noted above (fn. 2), this brief survey should not be taken to imply that corruption in the
public sector is not, in specific cases, an efficient means of moving economic actors through what
would otherwise be painfully slow bureaucratic mazes. When this is the case, corruption may
well prove efficient. While, on the whole, the evidence to date seems to clearly point to a net
negative overall effect of public sector corruption (Meon and Sekkat, 2005), we leave this debate
to others.
While not as broad, a growing body of research also addresses the economic and social
consequences of natural disasters. This area of study has taken on added gravity with the great
Indian Ocean tsunami of December 26, 2004 which devastated large segments of Indonesia and
Sri Lanka and, according to the U.N., lead to the deaths of 186,983 individuals. More recently,
the experience of the U.S. Gulf Coast, especially in and around the city of New Orleans,
following Hurricane Katrina in 2005 brought home the potentially dire consequences that natural
disasters can have both in human and economic terms. As was true for the literature on public
sector corruption, the earliest economic research in this area focused on the effects of natural
disasters on economic growth. In this regard, Skidmore and Toya (2002) conclude, based on a
variety of natural disasters striking 89 countries, that long run growth is actually enhanced by
disasters, primarily through the accumulation of human capital and its substitution for physical
capital. Similarly, Ramcharan (2005) shows that consumption expenditures increase following
significant natural disasters; much of which is financed through inflows of monies from other
countries.
Regardless of impact of natural disasters on purely economic factors such as growth,
however, there is no debate as to the human costs that these events can bring about. In this
regard, we intersect these lines of research by considering the relation between public sector
corruption and the death toll of natural disasters. Our contention is that public sector corruption
can and often does lead to the construction of inferior public infrastructure, private buildings, and
4
housing units which ultimately fail in the face of a major disaster leading to a greater death toll
than would occur in the absence of corruption.
To address this question, we empirically analyze 344 major earthquakes occurring across
the globe during the period of 1975-2003. Our choice of earthquakes as the focus of the study is
driven by a number of related factors. Most importantly, quakes strike a very well-defined,
localized area allowing for rather precise estimates of the event’s destruction. Equally important,
quakes have very well-accepted scientific parameters associated with them that allow for an
accurate comparison across events. In addition, we choose earthquakes for the analysis since
there are, relative to many other natural disasters, a relatively large number of events during a
given time period. Finally, we focus on earthquakes because they allow for a strong test of the
hypothesis that public sector corruption plays a role in the devastation wrought by a natural
disaster in that most of the deaths from quakes result from the failure of houses, apartment
buildings, and other structures. Whether these structures result from the letting of public
contracts for large-scale infrastructure projects such as dams, levees, highways, and bridges,
maintenance of these structures, inspection and oversight of the construction of commercial
buildings and housing units, or the expansion of ‘un-zoned’ informal sector housing, bribes and
other forms of public sector corruption in any of these areas can lead to seismically-suspect
structures thus potentially adding to the destructiveness of a given quake. Previewing our results,
relying on two of the most commonly used measures of corruption, we find a consistently
positive and statistically significant relation between an economy’s perceived level of corruption
and the death toll taken when it suffers from a major earthquake.
In the following section, we discuss the potential interaction between public sector corruption
and the severity of natural disasters drawing on several recent earthquakes as an anecdotal
backdrop. Following this, we offer a brief theoretical sketch of the relation between corruption
and deaths due to earthquakes. In the next sections, we present our data and empirical analysis.
Finally, a conclusion and discussion is offered.
5
2. The Relation between Public Sector Corruption and Earthquake Deaths
Earthquakes and humankind share a long and deadly history. Quakes result from the
pressure build-up caused by the scrubbing of two or more tectonic plates. Since the earth’s
surface is made-up of these ever-moving plates, earthquakes occur both frequently and in nearly
every country on the globe. The National Geophysical Data Center (NGDC) notes the first
recorded quake as occurring in what is now Jordan in 2150 BC.
3
NGDC has catalogued 6,524
significant quakes since that time with 540 happening during the 1989-2003 period alone. While
many of these were relatively minor, leading to limited loss of property or life, some proved
catastrophic. Taken together, the quakes occurring between 1989 and 2003, lead to an estimated
166,245 casualties, more than $117 trillion in property losses, and countless millions of people
being dislocated.
4
As examples of the devastation that quakes can bring, consider the following four events:
1) the August 17, 1999 quake striking the Marmara Region of Turkey which measured 7.6 on the
Richter Scale and resulted in 17,118 deaths; 2) the January 26, 2001 Bhuj, India quake with a 7.7
Richter Scale measurement that lead to 20,005 deaths; 3) the 6.8 Richter Scale quake striking
near Algiers, Algeria on May 21, 2003 which lead to 2,266 deaths; and, finally 4) the 6.6 Richter
Scale quake devastating the area around the city of Bam, Iran on December 26, 2003, causing
41,000 fatalities. These quakes have a number of obvious commonalities. Each occurred in a
country where quakes are common, each released tremendous energy as noted by the relatively
high Richter values, each resulted in tremendous loss of life not to mention property, and each
took place in a country with moderate to good seismically-sensitive building codes in place.
5
A
3
The NGDC Significant Earthquake Database is available at www.ngdc.noaa.gov.
4
Data taken from the NGDC Significant Earthquake Database.
5
That Turkey, India, Algeria, and Iran suffer through quakes with great regularity as well as the relative strength of
these quakes is noted by considering the data available from the NGDC Significant Earthquake Database while the
6
common characteristic that might not be so obvious, however, is that in each case public sector
corruption took what was already a deadly act of nature and turned it into an even more deadly
conspiracy between man and nature.
6
Following the Marmara quake in Turkey, the country’s Interior Minister, commenting on
the typically shoddy building practices in the affected area stated, “The contractors who built
those buildings and those who issued permits committed murder. The builders and bureaucrats
were involved in organized crime (Bohlen, 1999).” Similarly, a report from the Turkish
Architects and Engineers Association notes that more than half of all buildings in Turkey fail to
comply with construction requirements (Bohlen, 1999). Part of the problem, no doubt, was the
rapidly growing population of the area that placed a premium on quick construction of houses,
apartments, and other structures. Unfortunately, quick construction is not necessarily consistent
with seismically-sensitive construction. Many of those who died did so as their homes and
workplaces collapsed upon them since contractors used concrete diluted with too much sand,
failed to follow accepted procedures for insuring that air pockets or voids in concrete poured
within a building’s steel substructure would be eliminated, used fewer and lower quality steel
supports in the construction process, added stories above what had been designed, and other
similar substandard building practices (Lewis, 2005). So commonplace was the collapse of
relatively new apartment buildings, these structures came to be known locally as “bribe
buildings”. And most importantly, the problem in Turkey is not one of lacking understanding
about seismically-sensitive engineering or appropriate construction standards, but rather one of
unscrupulous and often inappropriately trained contractors who ignore the country’s building
codes with the consent of government inspectors (Lewis, 2005; Bohlen, 1999; Kinzer, 1999). As
a sign of their complicity in substandard construction, recently forty municipal officials from
existence of seismically-sensitive building codes in each country is confirmed in the International Association for
Earthquake Engineering’s (1996) Regulations for Seismic Design: A World List-1996.
6
A good survey of intersection of construction and corruption can be found in Stansbury (2005).
7
three towns in Turkey were taken into custody for allowing unlicensed buildings to be erected in
exchange for brides from contractors (Turkish Daily News, 2006).
Unfortunately, the situation is eerily reminiscent in the case of the Bhuj earthquake.
While the Indian central government has adopted seismically-sensitive building standards, few of
its local areas fully enforce them (BBC News, 2001). At the time of the Bhuj quake, contractors
needed only to have their building plans approved—there was no actual onsite inspection of the
construction process. In their review of the Bhuj quake, the World Seismic Safety Initiative
(2001, p. 33) concluded that the lack of onsite inspections “…led to poor structural design and in
some cases the use of poor building material due to corruption, black marketeering, and simple
economic considerations on the part of the client/builder.” An editorial in the Times of India
added, “As often happens in India, corruption underlies our avoidable disasters (BBC News,
2001).” The problem of corruption in construction in India was so well-known and so
widespread that, according to the Cooperative Institute for Research in Environmental Sciences
(CIRES, 2001), within one week of the Bhuj quake, 37 builders, architects, and engineers of
failed buildings were charged with culpable homicide and criminal conspiracy.
The Algerian quake of 2003 offers a similar pattern of events. Following a major quake
in 1980 in Algeria, the country further strengthened its building codes. Yet the improved codes
proved to be little more than window dressing as they were frequently bent or completely
ignored. The president of the Council of Algerian Architects put it this way, “It is not a question
of laws—we have extraordinary laws—the problem is their enforcement” while the head of the
National Union of Construction Engineers adds that Algeria’s entire construction industry is
riddled with corruption often entailing architects and engineers being paid for their signatures not
for their professional expertise in construction (Smith, 2003).
Again, the situation is little different in the Iranian quake in Bam (Pejhan, 2003): “It is
not as if Iran does not have civil engineers whose expertise is in building structures that can
survive earthquakes. It is not as if these experts have not developed codes (similar to those in
8
existence in California and Japan) for building earthquake resistant structures… The experts, the
codes and the laws have all existed in Iran for decades. Yet with every major quake, structures
new and old, public and private are destroyed with great loss of life…. The heavens are not
responsible for the enormity of these tragedies. Greed, corruption and lack of knowledge are
obvious culprits: greed and corruption in the case of public buildings… where contractors seek to
increase their profits… and government officials are only too ready to oblige as long as the bribe
is sizable enough; lack of knowledge (and poverty) in the case of residents who rebuild their
houses using the same primitive materials and techniques.” Perhaps most to the point, we note
that the quake that struck roughly halfway between San Francisco and Los Angeles in Paso
Robles, California in 2003, with an identical release of energy as the Bam quake (each having a
6.6 Richter Scale reading), resulted in only two fatalities compared with the death toll in Bam of
41,000.
In these cases, the anecdotal evidence clearly shows that the substandard construction
practices encouraged by public sector corruption can significantly exacerbate the deadly effects of
a major earthquake. To determine the extent to which these cases are typical, we consider the
relation between corruption and the deaths arising from 344 significant quakes occurring between
1975 and 2003 in the multivariate analysis below. To direct that analysis, in the following section
we develop a simple theoretical model of the relation.
3. Theoretical Framework
The types of corruption noted above are, unfortunately, all too common in construction.
In fact, recent surveys indicate that corruption in construction is more common than in any other
sector of the typical economy.
7
There are, no doubt, many causes for this but the very nature of
the construction process is likely one of the most important. Whether considering the
7
See the 2002 Bribe Payers Index in Transparency International’s (2003) Global Corruption Report and Control Risks
Group’s (2002) Facing Up to Corruption.
9
construction of a modest home or a major public infrastructure project, all building involves a
physical process of layering materials, with each layer potentially providing “cover” for
substandard construction practices encouraged by public sector corruption as the following
simple example details. Moving vertically, the first step in any building project involves laying a
concrete foundation that covers not just the ground but also the structural steel needed to
strengthen and secure the concrete. Next come internal supports for the building’s walls, floors,
and roof. Depending on the size of the project, these supports may be made of wood or steel and
in the later case, may again be covered or filled with concrete. Following this, the building’s
floors, walls, and roofing are added. Finally, the entirety of most the final surfaces are covered
with paint, carpeting, tile, stucco or something similar.
What does this discussion have to do with public sector corruption? Since construction
involves many steps, each of which essentially physically covers the preceding step, unless
extremely vigilant inspectors are nearly ever-present the possibility of a substandard outcome is
relatively great. And while this is true with any construction process, as the project grows in size
and complexity, the potential problem grows and most likely exponentially so. Much the same is
true for unique, major infrastructure projects which provide little in the way of comparative
examples for builders, inspectors, or those to whom the project is delivered. Add to this the
pressure that penalties for late delivery and inclement weather create, and you find a situation in
which the temptation to cut corners in order to finish a project on time and at budget can be quite
severe. And as the anecdotal examples presented in the previous section and the surveys noted
above suggest, contractors in many countries have apparently little trouble in finding willing
accomplices to their substandard production practices in the form of unscrupulous government
officials involved with zoning and inspection of construction projects. Of course, in many
circumstances, the gamble pays-off in the sense that, while substandard, a building may not have
the misfortune of being exposed to the extreme stress of the type brought on by a severe natural
hazard in its effective lifetime. When such a hazard strikes, however, the results can and too
10
often do prove catastrophic. In this section, we present a simplified theoretical model which
brings together public sector corruption and the construction process. In the model’s
development, we focus on the production of new housing units though this just serves to ease the
discussion of the model. Given the preceding discussion, extensions to other types of
construction should be intuitively obvious.
8
In our setup, we consider a construction market which is perfectly competitive and
assume that all commercial developers have access to the same technology. With that common
technology, each developer is assumed to be able to construct one normalized unit of housing
which complies with existing earthquake codes at a normalized cost of one.
9
The demand, D(p),
for new housing by households is downward sloping, where p denotes the price. Perfect
competition among developers will lead to p = 1 and a profit of zero for developers who construct
earthquake-safe houses. Let n > 0 denote the equilibrium number of new houses. That is, when
demand equals supply at p = 1, the equilibrium quantity is n. Thus, since each developer can
produce one unit of housing, there will be n developers in the market. For simplicity, we assume
that, compared to buying a new house, the cost of owner-construction of a house is prohibitive.
10
By assuming that a house lasts k years, the total number of houses at a given point in time is kn,
where k > 0. Finally, we assume that a potential buyer in the market cannot distinguish between
earthquake-safe and earthquake-unsafe houses in the absence of an earthquake.
11
8
A very thorough treatment of public sector corruption in the construction of structures can be found in Stansbury
(2005).
9
As discussed in the empirical section, there are 42 countries in our sample. Each has adopted the International
Conference of Building Officials (1987) Uniform Building Code: Structural Engineering Design Provisions which
provides specifications for earthquake resistant design and seismic zone information for countries worldwide.
10
While it is true that for some households, owner-construction is not too costly (compared to houses provided by the
market). However, in many developing countries, most people cannot afford even a slightly higher cost of a safe
owner-constructed house. On the contrary, many owner-constructed houses in these countries are squatter houses
which are typically much less safe than the earthquake-unsafe houses constructed by developers; in addition these
squatter houses are built on highly earthquake-unsafe grounds as well.
11
Uniform building codes, in their most positive light, serve to protect building owners and occupants from their own
ignorance. Put more directly, all construction processes involve highly technical, complex engineering processes that
few other than trained professionals can reasonably be expected to understand. Of course, seismically-sensitive
11
Suppose there are m > 0 inspectors to enforce seismically-sensitive building codes at
each of several distinct stages of the construction process and that the inspectors are randomly
assigned to developers.
12
If an inspector is not corrupt, he will reject any bribe proposal made by
a developer and will inspect the construction process properly. This will result in a house that is
earthquake-safe. On the contrary, if an inspector is corrupt, he will accept a developer’s proposal
and the two parties will bargain over the surplus (the outcome of which will be determined by the
Nash bargaining solution). Having accepted the bribe, the inspector will not oversee the
construction process adequately allowing the developer to produce an earthquake-unsafe house at
a reduced cost of γ < 1. Trivially, the Nash bargaining solution outcome (or the outcome of any
bargaining solution which satisfies Symmetry and Pareto Optimality properties) is such that the
two parties each receive (1-γ)/2.
In the absence of any intervention by society, clearly bribery is the dominant strategy for
a developer and an inspector. Societies with good institutions can prevent corruption by
enforcing their laws (e.g., by taking the developer and the inspector to court after the unsafe
houses suffer damages as a result of an earthquake and by imposing the appropriate
punishments). Suppose the cost of having and maintaining such institutions to the government is
C > 0; this cost will also entail an adequate salary for state employees (including the inspectors).
Part of the punishment of the parties involved in bribery will usually take the form of a jail term,
with a cost of J > 0 to each party jailed. In addition, the guilty parties are required to reimburse
building codes and government enforcement are not the only way of protecting consumers in such a case. One could
imagine a private market of structural experts providing this informational service. This would likely prove inefficient,
however, for at least three reasons. First, when a structure is for sale it is typically considered by many potential buyers
and if the buyer is responsible for hiring the private inspector, needlessly redundant inspections will occur. Second,
when the seller is responsible for providing the inspection, a potential for collusion between the seller and the inspector
exists. Finally, there is the possibility of a negative spillover that might occur when unsafe structures collapse onto
adjacent structures, even if those structures were themselves earthquake-safe.
12
Whether m is greater or less than n does not matter here. If m > n, we can assume that one inspector will be
assigned to each developers and that m-n inspectors will not be assigned to any developer. If m < n, we can assume
that some inspectors will be assigned to more than one developer. Finally, if m = n, it can be assigned that exactly one
inspector will be assigned to each developer.
12
society for the damages they impose on others; let R > 0 denote this amount. If R < (1-γ)/2, then
the government would be providing an incentive for bribery, which would not be in line with the
concept of good institutions. Thus, we assume R > (1-γ)/2. Governments that do not incur the
cost of having and maintaining good institutions will be without such institutions and will stay on
the sidelines when society suffers from the consequences of corruption. In addition, they will pay
very low salaries to their state employees, which will be normalized to zero here.
13
Thus, the normal form game between the government and a typical inspector will be as
follows:
Note that if R < C, then the unique Nash equilibrium is such that the government doesn’t
maintain good institutions and inspectors accept bribes. Suppose that R > C is possible. In that
case, there is no pure strategy Nash equilibrium. But there is a mixed strategy Nash equilibrium
such that the government maintains good institutions with probability p and doesn’t with
probability 1-p, and a typical inspector rejects bribes with probability q and accepts bribes with
probability q. The level of p can be interpreted to measure the quality of institutions and the level
13
When asked how the state employees would afford living with virtually no salary increases in the face of more than
40% annual inflation year after year in the mid-1980s, the then prime minister of Turkey, Turgut Ozal, said: “my state
employees are canny” (implying that would find ways to supplement their incomes).
Government
Maintain good
Institutions
Don’t Maintain Good
Institutions
Inspector
Reject Bribes
s, -C
0, 0
Accept Bribes
s+(1-γ)/2-J-R, R-C
(1-γ)/2, 0
13
of 1-q can be interpreted to measure the prevalence of corruption.
Likewise, one can consider a similar version of the above game between the government
and a typical developer.
Proposition 0: In the mixed strategy equilibrium of the above games when R > C, the probability
that a typical developer will propose bribe and a typical inspector will accept the bribe will be
C/R.
The frequency of a major earthquake is f
r
∈ (0,1), which is typically a very small number on an
annual basis. The probability that a house is close to a fault line is f
o
∈ (0,1), which too is a very
small number.
14
An earthquake-safe house will prevent any casualties in any earthquake. An
unsafe house will be heavily damaged in a major earthquake if it is close to a fault line. Finally,
14
Clearly, the probabilities used in our setup can only be approximately known in reality. But because of simplicity
(and as is customary in any stochastic setup), we assume that they are exactly known by the agents. Furthermore,
especially f
r
and f
o
are not dichotomous in reality. But considering realistic continuous distributions of them would
make things unnecessarily complicated here. Note that a realistic distribution of f
r
should in effect combine two things,
namely frequency and magnitude of earthquakes, since f
r
is the frequency of a major earthquake. On the other hand, a
realistic distribution of f
o
would stand for the focal distances of the houses from the epicenter of the earthquake.
Government
Maintain good
Institutions
Don’t Maintain
Good Institutions
Developer
Don’t Bribe
0, -C
0, 0
Bribe
(1-γ)/2-J-R, R-C
(1-γ)/2, 0
14
as all reconnaissance reports on earthquakes indicate, fatalities will be assumed to increase in the
number of heavily damaged houses. The proof of the next result is straightforward and thus will
be omitted.
Proposition 1: Fatalities will increase in the number of houses, in the extent of corruption, in the
lack of good institutions, in the frequency of major earthquakes, and in the probability of houses
being close to a fault line.
Predictions of our Theoretical Setup: Proposition 1 provides the empirical predictions of our
simple theoretical setup. It states that earthquake deaths will increase (1) in the number of
houses—which is a proxy for the population in the earthquake-affected area, (2) in the extent of
corruption—since it will directly affect the number of earthquake-unsafe houses, (3) in the lack of
good institutions, (4) in the frequency of major earthquakes, that is, in essence, in the frequency
of earthquakes as well as in the magnitude of the earthquakes, and (5) in the probability of a
house being close to a fault line, that is, in essence, in the focal distance from the epicenter of the
earthquake.
3. Data and Empirical Analysis
The anecdotal summaries discussed above, as well as the theoretical sketch, suggest that
substandard construction due to corrupt conspiracies between contractors and government
inspectors may well make what would already be difficult naturally occurring events truly
disastrous. In this section, we attempt to test this proposition empirically. The source for the
earthquake-related data is the NGDC’s Significant Earthquake Database which, as discussed
above, contains information on more than 6,500 destructive quakes occurring worldwide since
2150 BC.
15
As the title suggests, this is a catalog of “significant earthquakes”. To be included, a
15
A complete description and discussion of the NGDC Significant Earthquake Database can be found at
15
quake must meet one of the following criteria: cause approximately $1 million or more in
property damages, have a Richter value of 7.5 or greater, or cause 10 or more deaths. We take
from this source all quakes measuring 6+ on the Richter-scale
16
, occurring world-wide, between
1975 and 2003, for which complete data is available. Our focus is on the number of lives lost—
FATALITIES—a focus that allows us to avoid complications associated with estimating the costs
of lost or damaged physical structures across countries and time.
17
This leaves us to analyze the
fatalities resulting from 344 major earthquakes, arising from 42 countries. Of perhaps most
importance, we should note that each complies with the International Conference of Building
Officials Uniform Building Code (see fn.9). The 42 countries come from around the globe: 9
being from Africa, 10 from Asia, 7 from Europe, and the remaining 16 from the Americas—as
detailed in Appendix 1, which also includes the average level of corruption, as is defined below,
for each country in the sample between 1975 and 2003.
Our primary measure of the level of public sector corruption within an economy is taken
from the International Country Risk Guide (ICRG), published by Political Risk Services Group,
as assembled by the IRIS Center at the University of Maryland (COR-ICRG). This source reports
complete data on more than 100 countries between 1982 and 2004. While broad in its time
coverage, this measure, which is most commonly used in empirical analysis, offers no annual
measure of corruption for the first seven years of our sample period. For the sake of sample size,
however, we take advantage of the remarkable degree of consistency over time in the corruption
www.ngdc.noaa.gov.
16
We chose this cut-off as even the most poorly constructed structures rarely fail in quakes of less magnitude.
17
While we would ideally like to be able to distinguish the cause of deaths (deaths due to the immediate collapse of a
structure, deaths due to secondary collapses, stress-related deaths, deaths due to quake-related illnesses days or weeks
later….) no such data exists. NGDC, as with all who catalog earthquake data, simply rely on the official death counts
provided by local government officials. To the extent that reporting practices differ, error in the dependent variable
will be imparted. Unless this is systematically related with public sector corruption, however, the likely outcome will
be to reduce the likelihood of finding a significant relation between deaths and corruption, providing for a strong test of
the proposition. Further, if one were to suppose that relatively corrupt governments, due to their complicity in shoddy
building practices, wished to misreport deaths due to earthquakes, the likely direction of the bias would be downward,
again providing for a strong test of the proposition.
16
index for a given country that this source reports and assign the 1982 corruption index for those
quakes occurring between 1975 and 1981 while assigning the actual year-value for those
occurring between 1982 and 2003.
18
The specific variable taken from this source evaluates the
overall level of corruption in government. COR-ICRG ranges from one to seven (zero to six on
the original scale, which, due to the regression model’s natural logs, is re-scaled) with greater
values indicating less corruption in the forms of high-level government officials being likely to
require special payments and whether such bribes are expected throughout all levels of
government (see Knack and Keefer (1995) for a thorough treatment of this measure). This index
is particularly attractive in our application since special payments or bribes required by
government officials can be expected to translate into substandard and thus cheaper construction
regardless of whether the corruption occurs at the very beginning of the process—that is, in
allowing building to take place on seismically-suspect land—all the way through to the project’s
final inspection. Further, this is true whether the project under consideration is that for a modest
home or a major infrastructure project such as a highway or dam. The only difference being, and
referring back to our discussion of the general process of construction, that the latter offers many
more individual steps at which the substandard work brought on by the corruption can be
“covered” by the next step in the overall construction process.
Since our measure of corruption is key to the analysis, we would be remiss if we did not
expressly note that, as with all measures of public sector corruption, that offered by ICRG is
“subjective” in that it is based on surveys. While this is true, several related facts support the use
of each of the commonly used surveys. Perhaps most importantly, as pointed out by Alesina and
Weder (2002), there is very high correlation between each of the available measures. Different
groups, using different methodologies produce measures of public sector corruption which tend to
be very strongly related with one another and, further, tend to be very highly correlated over time.
18
It should be noted that the results are not qualitatively different if the period 1975-1981 is omitted.
17
From a more practical perspective, it should also be noted that, while perhaps not perfect
measures of corruption, empirical analysis has shown that these subjective evaluations of public
sector corruption are powerful in explaining social and economic phenomena as varied as
investment decisions and growth (Mauro, 1995) and traffic fatalities (Anbarci, Escaleras, and
Register, 2006).
To place the current analysis in the context of prior research on earthquakes, we take
advantage of models reported by Anbarci, Escaleras, and Register (2005), Schulze, Brookshire,
Hageman, and Tschirhart (1989), and Dunbar, Bilham, and Laituri (2002) in selecting other
independent variables that help determine the destructiveness of an earthquake, either from a
practical or scientific perspective. From the former perspective, we take into account the
FREQUENCY with which a country suffers from a major quake by calculating the number of 6+
quakes occurring within a given country in the prior 100 years. The expectation here is that there
may be a “learning by doing” reduction in deaths for a quake of a given size in a country that is
regularly hit by such events relative to a country that is only rarely stricken. To this, we add the
population and population density of the province(s) affected by a given quake (POPULATION
and DENSITY, respectively).
19
In each case, a positive relation with deaths is expected. It is also
reasonable to expect that a country’s level of development will influence the death toll from a
given quake, given that, while nearly all countries have “moderate-to-good” building and safety
codes (Lewis, 2005, p. 24), greater levels of development should allow for higher-level building
codes, better zoning and land use, better health care infrastructures, and higher levels of self-
insurance from quake-risk on the part of individuals. Our primary measure of a country’s level of
development is its GDP per capita (GDPPC), measured in constant (1995) U.S. dollars, as
reported in the World Bank’s World Development Indicators. Finally to capture any region-
specific factors, including those related to development not captured by GDPPC, we include
19
Population and land area are taken from several sources: 1) The World Gazetteer (www.world-gazetter.com), 2)
Geo-Hive (www.xist.org
) and 3) Population Statistics offered by Jan Lahmeger (www.library.unn.nl).
18
dummy variables for AFRICA, ASIA, and EUROPE, relative to the omitted AMERICAS.
From a more scientific perspective, information on the dynamics and location of the
quakes is taken from the NGDC’s Significant Earthquake Database. From this, we take two key
variables that determine a quake’s destructiveness; the quake’s power or MAGNITUDE,
measured by the common Richter scale and the quake’s proximity to the affected geographic
region, known as the focal distance (DISTANCE) (this draws heavily on the ESRI ArcView
Geographic Information source).
20
We would, of course, also like to be able to take into account
such factors as the subsurface nature of the earth as a quake of a given magnitude can be expected
to lead to greater destruction if the subsurface is relatively dense compared to an area of sandy
soil and whether local fault lines tend to lead to lateral or the much more disastrous vertical
scrubbing of tectonic plates, but such data for all areas is not, at present available. The
expectations for these variables should be obvious: the death toll of a given quake should be
greater for quakes that are more powerful and which occur closer to highly populated areas.
Table 1 provides summary statistics for each of these variables, which are more
thoroughly defined, along with their sources, in Appendix 2. Given the complete coverage of
these variables in the tables, we only expressly note the descriptives for the two key variables,
deaths and corruption. The mean value for the number of quake fatalities, FATALITIES, is 849,
with a rather broad range of 0-50,000. The measure of corruption taken from the International
County Risk Guide, COR-ICRG, has a mean of 3.8 and takes on values throughout its entire
potential range of one to seven.
To gain some initial insight into the relations between a quake’s magnitude, its level of
destruction and corruption, we analyze in a univariate sense the two quartile tails of the sample
20
The latitude and longitude of the affected region were collected from the Getty Thesaurus of Geographic Names,
available online at www.getty.edu
. We thank Brian Anyzeski for assisting us in calculating the distance between the
affected region and the quake’s epicenter.
19
based on the corruption index, COR-ICRG, as reported in Table 2.
21
To be considered a Low-
Corruption observation, the index had to have a value of 4.975 or above while High-Corruption
applies to those observations with COR-ICRG values of 3 or less. Consistent with the proposition
that public sector corruption increases the death toll wrought by an earthquake of a given size,
there appears to be a rather clear pattern between fatalities and corruption. Specifically, while
both the High-Corruption and Low-Corruption countries experience quakes that are not
significantly different in MAGNITUDE or DISTANCE from the affected region, the resulting
differences in FATALITIES are both statistically significant, at least at the 0.1 level, and
practically much greater in the High-Corruption countries—1,530 on average, versus 351.
22
To more rigorously test this apparent relation between corruption and fatalities, we
estimate the following basic model:
23
)1(
10
98765
43210
ii
iiiii
iiiii
EUROPE
ASIAAFRICAGDPPCPOPULATIONDENSITY
MAGNITUDEDISTANCEFREQUENCYICRGCORFATALITIES
εα
ααααα
ααααα
+
+++++
++++−+=
where i references a specific country and each variable is as defined above.
Estimation of Equation (1) involves two complications that must be addressed before
reliable estimates can be made. The first deals with the nature of the independent variable,
FATALITIES. Given that deaths are non-negative count data, the appropriate class of estimators is
21
The sub-groups differ in size due to clumping of the data. The results presented here are not affected if different cut-
off points for the corruption variable are used. We chose these cut-offs because they gave us sub-groups of the most
equal size possible
.
22
The difference in means t-tests noted takes into account whether the sub-groups have equal variances.
23
A reasonable argument could be made that rational homeowners would be less likely to invest in their properties in
areas where the risk of expropriation is relatively high, thus arguing for the inclusion of such a variable throughout our
modeling. When a variable for risk of expropriation is included in the models reported, the variable is insignificant in
all but one case. Further, while reducing the significance of the coefficient on corruption in the deaths equations, that
key variable remains significant, at least at the .1 level, in each model. Clearly, risk of expropriation and corruption are
correlated and we suspect this to be the cause of the poor performance of the expropriation variable and the modest
weakening of the corruption variable. No other significant changes are made when expropriation risk is included.
Given that our focus is on corruption and that nothing of value is added by including the risk of expropriation variable,
we have opted to report the models without this variable.
20
the Poisson. The Poisson is best suited for counts with low variance as the model assumes the
conditional mean and variance of the dependent variable to be equal. In this case, given the
relative over-dispersion of our dependent variable, we employ a Negative Binomial specification
which relaxes the Poisson’s assumptions by introducing a parameter (identified as α in the tables
below) that explicitly accounts for unobserved heterogeneity between observations in the sample.
Further, as is customary in the Negative Binomial specification, all variables, except dummies,
are entered in natural logs. Finally, in all of the Negative Binomial regressions reported below,
the significance of the parameter that accounts for the unobserved heterogeneity is reported,
allowing for a test of the appropriateness of the Negative Binomial specification.
The second issue for estimation of Equation (1) is the fact that it is highly unlikely that
the key variable, COR-ICRG, is exogenous since public sector corruption is commonly known to
be highly correlated with a number of other, omitted institutional factors.
24
To take this into
account, prior to the estimation of Equation (1), preliminary regressions of the form in Equation
(1P) are estimated with COR-ICRG as the dependent variable:
iiiii
iiiii
iiiii
POPULATIONDENSITYMAGNITUDEDISTANCE
FREQUENCYEUROPEASIAAFRICAENGLISH
PROTESTANTTINTCONFLICDEMOCRACYGDPPCICRGCOR
εββββ
βββββ
βββββ
++++
+++++
+++++=−
13121110
98765
43210
)P1(
where COR-ICRG represents the extent of corruption in the public sector for country i, GDPPC is
a measure of development, DEMOCRACY reflects the degree of openness or democracy that
exists within a country, INTCONFLICT indicates the degree of political stability, PROTESTANT
indicates the share of the population that is protestant, ENGLISH is a dummy variable indicating
whether English is the legal origin of a country, and the remaining variables: FREQUENCY,
POPULATION, DENSITY, AFRICA, ASIA EUROPE, DISTANCE, and MAGNITUDE are defined
24
A Hausman test confirmed the endogenous nature of our corruption measures. For a discussion of omitted variables
as an endogeneity problem, see Wooldridge (2002, pp. 50-51).
21
as in Equation (1).
Equation (1P) includes all of the exogenous variables from Equation (1) as well as
several institutional variables closely correlated with corruption, but relatively uncorrelated with
deaths due to earthquakes, as instruments. While the literature suggests a great many such
possible instruments (see, for example, Treisman (2001)), we take advantage of a recent paper by
Serra (2006) which undertakes a modified Leamer Extreme Bounds Analysis (see Leamer (1985)
and Levine and Renelt (1992)) to test the robustness of 28 commonly used corruption
determinants, across 60 countries. From this analysis Serra (2006) concludes that per capita
income, the extent of democratic institutions, political stability, the percent of the population that
is Protestant, and being of English colonial heritage are the most significant determinants of
public sector corruption, each exerting negative influence on corruption. We mimic these results
by using GDPPC, DEMOCRACY, INTCONFLICT, PROTESTANT, and ENGLISH as instruments
in our preliminary regressions. DEMOCRACY is provided by Polity IV. This scale, which runs
from zero to ten, ranks countries annually as to the general openness of their public institutions
with lower values indicating lesser degrees of openness.
25
INTCONFLICT is taken from the
International Country Risk Guide, published by Political Risk Services Group, as assembled by
the IRIS Center at the University of Maryland. This scale runs from one to twelve with higher
values indicating less political violence in the country and its impact on governance.
PROTESTANT and ENGLISH are each taken from La Porta, Lopez-de-Silanes, Shleifer, and
Vishny (1999) and reflect the percentage of a country’s population belonging to the Protestant
religion as of 1980 and whether the country’s legal origin is English, respectively. Based on the
work of Treisman (2001), Serra (2006) and others, we expect each of these instruments to be
negatively correlated with corruption (positively correlated with COR-ICRG, that is).
26
25
Available at: http://www.bsos.umd.edu/cidcm/polity/index.html.
26
Given the nature of the empirical modeling, we have little choice but to accept that the models include some degree
of colinearity. At the same time, we view this as primarily an empirical matter. That is, the primary shortcoming due
22
From this regression, which is reported in column 1 of Appendix 3, a predicted value of
the corruption measure is created (COR-ICRGP) and then subsequently added to Equation (1) in
lieu of the actual value of COR-ICRG to control for the endogeneity of that variable. The
variables taken from Equation (1) that appear in Equation (1P) do so in the same form (logged or
linear) as they appear in Equation (1). Since the estimation of the preliminary regression simply
provides a means to correct for the endogeneity of our corruption index in Equation (1), our
discussion of the results is brief. First, note that since COR-ICRG is bounded by the values of
one and seven and the sample includes each of these extreme values, the preliminary regression
for this variable uses a two-bound Tobit methodology. The regression offers solid goodness of fit
measures as evidenced by a strong LR Chi-Square FM value indicating that the included
independent variables as a group are highly significant in determining corruption as well as a
solid Pseudo R-Square value. Most importantly, the model offers individual coefficient estimates
that are, based on prior work, of the expected signs and are, with the exception of PROTESTANT
statistically significant. Specifically, GDPPC, DEMOCRACY, INTCONFLICT, PROTESTANT,
and ENGLISH are each positively related to the corruption index—that is, each is found to reduce
the level of public sector corruption (recall that higher levels of both COR-ICRG reflect less
corruption). Conversely, EUROPE and ASIA are negatively related to the corruption index
suggesting significantly higher average levels of corruption in those regions, relative to the
Americas, even after income and the other independent variables are held constant. Incidental to
the model, we find that countries that experience relatively severe earthquakes, as given by
MAGNITUDE, tend to be relatively corrupt. We have no explanation for this result. All other
variables in the model are insignificant.
27
to colinearity is the inflation of standard errors and the potential for deflated t-statistics suggesting a coefficient is
insignificant when in reality a significant relation exists. Since our key variable, corruption, is uniformly significant,
we are relatively unconcerned with this problem though it lessens the confidence we have in all estimated insignificant
coefficients.
27
While our choice of instruments is driven by Serra’s (2006) analysis, it should be noted that other common
determinants of corruption, such as ethno-linguistic fractionalization, were added to the predictive equation. These
variables tended to be insignificant and lead to no significant differences in the estimation of Equation (1).
23
With the predicted value of corruption now in hand, we turn to the estimation of Equation
(1). This model is presented in Table 3. Prior to discussing individual variable results, it should
be noted that the model shows a very good fit. First, a test of the significance of the parameter
entered to control for unobserved heterogeneity between observations gives a likelihood ratio
Chi-Square (LR Chi-Square α) showing that, well beyond the .001 level, the likelihood of this
data arising from a Poisson process to be virtually nil, arguing strongly in favor of the Negative
Binomial specification. Further, the full model likelihood ratio Chi-Square test (LR Chi-Square
FM) is highly significant, again beyond the 0.001 level, indicating that the independent variables
taken as a group are quite significant in explaining the death toll of a given earthquake. Finally,
the model offers a quite reasonable maximum likelihood R-Square (ML R-Square) value.
With respect to specific variables, consider first the outcome for our variable of primary
interest, COR-ICRGP which is both negative and significant, well beyond the .01 level,
suggesting that, other relevant factors constant, an earthquake of a given size can be expected to
lead to significantly more deaths in a country that has a relatively high level of public sector
corruption than in a less corrupt country. While it is not possible to determine the exact linkage
of this relationship, its strength, even when corrected for omitted variables, clearly indicates a
very strong positive relationship between corruption in the public sector and deaths when a major
quake strikes. As discussed above, since most deaths from quakes result from collapsing
structures, the precise avenues this relationship might take include all those discussed above such
as substandard construction and poor oversight/management of public infrastructure projects and
substandard construction of housing and other structures resulting from bribes and other forms of
corruption in the letting of contracts and building practice inspections. An exact determination of
the avenue that corruption takes in a specific case would have to rely on a case study approach,
something that is well beyond the current analysis and, further, something that would require
substantial input from the forensic engineering field. At a minimum, however, this result is
completely in keeping with the case studies discussed in this study’s Section 2.
24
While we cannot identify the precise path from corruption to deaths, we can put its effect
into practical terms by evaluating the marginal impact of corruption, assuming all other variables
are held constant. The marginal effect for the predicted value of COR-ICRG, is roughly 50
percent, for a one standard deviation change in that index. For example, the mean value COR-
ICRG is 3.78 on a scale of one to seven, with a standard deviation of 1.59, while the mean
number of deaths in the sample is 849. Thus, if a major quake struck a relatively corrupt country
with a COR-ICRG value of 2.19 (one standard deviation below the mean of COR-ICRG), the
model would predict that rather than suffering 849 deaths, that toll would rise to roughly 1,275.
Conversely, the same quake striking in a less corrupt country with a COR-ICRG value of 5.37
would suffer only 425 deaths. We take this to be a remarkably strong indictment of public sector
corruption as it applies to the death toll arising from a major natural disaster. Some might
conclude that the estimated effect, even if believing in the underlying premise of the negative
effect of corruption, is too large to be accepted at face value. In this regard, however, it should be
pointed out that the estimated marginal effect is based on a one standard deviation change in the
corruption index. In the case presented, that moves the index from, 3.78 to 5.37 or to 2.19.
While these may not seem to be large moves, in absolute terms, it must be recalled that the range
of COR-ICRG is only one to seven, thus, in this case, a one standard deviation increase in the
index represents a very significant fall in the level of public sector corruption within a country
and vice-versa. To give an intuitive feel for these differences, in 2001 El Salvador had a COR-
ICRG value of 3.75, very near the sample mean while China’s COR-ICRG value in 1998 was 2.
Similarly, in 1999, the U.S. had a COR-ICRG value of 5.9. Thus, while the marginal effects are
truly large, they are based on differences in corruption that are rather pronounced as well.
Turning to the remaining variables, we find both expected outcomes and generally high
levels of significance. Consider first the quake-related variables, FREQUENCY, DISTANCE, and
MAGNITUDE. In each case the variables are highly significant with FREQUENCY and
DISTANCE being strongly associated with falling deaths while MAGNITUDE enters in an equally
25
strong positive fashion. The FREQUENCY result most like indicates that countries which tend to
suffer regularly from quakes, learn from the experiences and are thus better prepared for the
quakes. Not surprisingly, the DISTANCE outcome merely points out that, other things constant,
the further a quake’s epicenter is from major population centers, the fewer deaths result. Finally,
the outcome for MAGNITUDE—quakes of greater severity leading to more deaths—requires
little discussion.
A bit less well-behaved are the population-related variables. While DENSITY yields the
expected positive relation with deaths, the actual population of the affected area (POPULATION)
is both positive and significant. Apparently, the absolute number of potential victims is relatively
more important than the concentration of the population when it comes to determining the death
toll from an earthquake. Our income variable, GDPPC, performs well, with both the expected
negative sign and a high level of statistical significance suggesting that the general process of
development brings with it, on average, some degree of protection from the affects of natural
disasters. Finally, turning to the regional variables, the only variable that is significant is the
EUROPE dummy whose positive coefficient may result from the rather old stock of often multi-
story housing and other structures in that region, relative to the Americas.
28
4. Additional Empirical Considerations
While the results presented above follow standard empirical procedures, are generally
consistent with prior research on earthquakes and other natural disasters, and lead to a conclusion
about the affects of corruption that, especially in the light of the case studies presented above,
seem quite plausible, as with all empirical work, obvious potential shortcomings should be
addressed. We see two such potential shortcomings here and address each: 1) The sensitivity of
28
These results, other than for the corruption variable, are quite similar to those reported in Anbarci, Escaleras, and
Register (2005) and Kahn (2005).
26
our results to the particular index of corruption employed and, 2) The possibility of the results
being driven by outliers.
Even though others have shown the various readily available measures of public sector
corruption to be both highly consistent both with one another and over time, to evaluate the
sensitivity of our results to the ICRG corruption index employed, we replicate the analysis
presented above using an alternative corruption index taken from Transparency International (TI)
which reports a “poll of polls” summary of the level of perceived corruption within an economy,
for the period 1998 through 2003 (COR-TI).
29
The TI index is compiled by researchers at
Gottingen University from up to 12 individual surveys. Given that our full sample includes
quakes dating back to 1975, we assign the 1998 value of COR-TI to all quakes occurring between
1975 and 1997. This reduces the variability in the measure of corruption but, as such, should
provide a relatively strong test of the result we find with COR-ICRG, and is unavoidable since
limiting the sample to the post-1997 period during which we have COR-TI values would cause us
to lose more than two-thirds of the sample, effectively negating the use of COR-TI as a sensitivity
check on the COR-ICRG results. Further, while not as broad in country coverage as our primary
measure, COR-TI does allow us, when used as just described, to evaluate 305 of the 344 quakes
in the primary analysis. This index ranges from one to ten with higher values indicating less
corruption in the form of the perceptions of business people as to the degree of public corruption
resulting from officials abusing their offices for private gain. For the 305-unit sample, COR-TI
has a mean of 4.17.
With the exception of the alternative measure of corruption, the variables employed are
identical here, though the estimation procedure for the preliminary regression differs. To obtain
predicted values of COR-TI (COR-TIP) for use in the modified Equation (1), we estimate a
29
As pointed out by Alesina and Weder (2002), there is a high degree of correlation across the variously available
measures of corruption. In our case, the two indices have a simple correlation of 0.7. Further, while the TI measure is
available for 1995-1997, it has been shown to have been somewhat inconsistent prior to 1998, thus we only use the data
from that source beginning with 1998.
27
modified Equation (1P) with COR-TI as the independent variable. As with COR-ICRG, this
corruption index is bounded but in this case no observations take on the extreme values thus
Ordinary Least Squares is employed rather than the two-bound Tobit procedure. The results for
the estimation of the modified Equation (1P) are presented in column 2 of Appendix 3 and offer
strong goodness of fit measures both with respect to the F-test of the joint significance of the
independent variables (F-statistic FM) being significant well beyond the 0.001 level and the
model’s solid R-square value. Most importantly, note the generally good performance of the
Serra (2006) variables included to uniquely determine corruption, both on their own and in terms
of their similarity with the results when using COR-ICRG (column 1). In each case, GDPPC,
having an ENGLISH origin, INTCONFLICT, and PROTESTANT are each found to exert positive
pressure on the corruption index, that is, to reduce corruption, though INTCONFLICT does so
insignificantly.
The only major difference between the results for the instruments between this model and
that when the primary measure of corruption is used has to do with the DEMOCRACY variable
which, unexpectedly, is both negative and significant when COR-TI is the dependent variable.
The reason(s) for this are not at all clear. Three possibilities are that: 1) Since both measures of
corruption are based on surveys (as is the democracy variable for that matter), there may be
something inherent in the way the surveys are constructed that leads to this result, 2) Since we are
forced to use the 1998 value of COR-TI for the entire 1975-1997 period, this result might be
driven by the loss of variability in COR-TI and, 3) There may be something unique about the 39
observations lost when using COR-TI that is at work here. Whether either of these explanations
is correct or the result comes about due to something completely unrelated to these, the good
performance of the other instruments suggests that this outcome should not lead to any problems
for estimating the primary model. Finally, two additional variables are significant in this
regression, EUROPE and POPULATION. The finding that, on average, observations from
EUROPE are associated with higher levels of corruption is the same as was found for the COR-
28
ICRG model. Unlike that model where POPULATION was insignificant, here we do find that
countries with larger populations tend to have higher levels of corruption. Taken together then, it
seems that whether one uses COR-ICRG or the alternative COR-TI, while not identical, the results
do not change substantively enough to raise concerns as to which measure of corruption should
be used in empirical analysis.
Having estimated the modified Equation (1P) as just discussed, predicted values for
COR-TI (COR-TIP) are calculated and substituted into Equation (1) for the primary corruption
measure. The results for this Negative Binomial regression are presented in column 1 of Table 4.
Since this model differs with that of Table 3 only by the use of COR-TIP in lieu of COR-ICRGP
and given the overall similarity of results, our discussion here is brief. Note again that the model
offers very solid goodness of fit measures, with the maximum likelihood R-square values (ML R-
square) actually being identical, to the second decimal place. Most worth noting, however, is the
result for the predicted corruption index which is again negative and highly significant indicating
that, controlling both for that variable’s endogeneity and other factors which play a role in
determining the death toll of an earthquake, public sector corruption clearly takes what is already
a dire circumstance, a major earthquake, and makes it significantly worse. It is also worth noting
that as for the primary model, each of the quake-related variables is significant and of the same
sign. The only differences here are with respect to POPULATION, GDPPC, and EUROPE which
while remaining positive, are no longer significant. Again, whether these differing outcomes
result from the lost observations when COR-TI is used, are due to differences in the survey
methods used to create the two corruption indices, due to our having to assign the 1998 value of
COR-TI to all prior quakes, or to some other factor is not known and beyond our scope. Using
COR-TIP, the final difference with the primary measure concerns the dummy for AFRICA which
is now negative and significant. A plausible explanation for this outcome might be that,
traditionally, the structures built in that region, especially for rural housing, have not been multi-
story, densely packed structures that are capable of great destruction during a major earthquake.
29
Taken together, the results for each of the estimations using the alternate measure of
corruption from Transparency International are found to be highly consistent with those using the
primary measure of corruption, thus suggesting that the results presented in Table 3 do not appear
to be sensitive to the selection of the corruption index.
Finally, while the results presented in Table 3 and column 1 of Table 4 are based on an
estimation procedure, the Negative Binomial, which is expressly designed to take into account the
effect of outliers, it seems reasonable to question whether the outcomes presented are at least
somewhat influenced by outliers in that while the maximum number of deaths in the sample is
50,000, the mean is only about 850 suggesting that the sample is indeed rather skewed. To take
this into account, Equations (1P) and (1) are re-estimated for the 310 observations on COR-ICRG
resulting in less than 500 deaths and the 274 observations when COR-TI is used, with the same
deaths limit. With this roughly 10 percent reduction in the sample, the mean number of deaths
falls from 850 to just 31 in the 310 unit sub-sample. The significance of this dramatic fall in the
mean number of deaths can be seen by considering the ratio of the standard deviation of deaths to
that variable’s mean in the two cases. Specifically, in the full, 344 observation sample the
standard deviation to mean ratio is 5.29 which drops to just 2.29 when the 34 observations with
death tolls greater than 499 are omitted.
To be complete, we re-estimated both the models using COR-ICRG and COR-TI with the
deaths restriction. The predictive equation results for COR-ICRG are found in column 3 of
Appendix 3 while the results for FATALITIES using the predicted value of COR-ICRG are found
in column 2 of Table 4. Similarly, the results using COR-TI are presented in column 4 of
Appendix 3 and column 3 of Table 4. Taken together, both the predictive equations and the
Negative Binomial models of FATALITIES provide strong evidence of one conclusion: Public
sector corruption increases the death toll that major earthquakes cause and this result holds even
when the samples are limited to quakes that are comparatively non-lethal. In other words, there is
no evidence that the results presented above are in anyway driven by the existence of outliers.
30
This said, there are a few interesting outcomes from these regressions worth noting, over-
and-above the models’ good fits and the consistently positive relation between corruption and
deaths. First, for both the COR-ICRG and COR-TI models limited by the number of deaths, the
instruments used in the preliminary regressions of Appendix 3 hold-up well, offering the same
signs and generally the same levels of significance as when the full models are used. This is
especially true for the GDPPC and ENGLISH variables. Second, across-the-board, we find the
EUROPE dummy in these regressions to be significantly associated with higher levels of
corruption clearly suggesting a relationship that should not be overlooked. Turning to the
FATALITIES regressions reported in Table 4, two additional outcomes seem worth noting. The
first is that, once the sample is limited to less than 500 deaths, prior experience with major
earthquakes, as given by the FREQUENCY variable, becomes insignificant regardless of the
corruption measure used. This likely suggests nothing more than common sense: To the extent
that the FREQUENCY variable captures “learning by doing”, there is much more learning when
horrific quakes with enormous death tolls strike than when comparatively less deadly ones do so.
Finally, when the sample is restricted in this fashion, GDPPC also becomes insignificant (as it
was in the COR-TI model). Again, a plausible interpretation might simply be that the positive
effects of development on a country’s ability to deal with a natural disaster prove more effective
in particularly severe events relative to those that are less severe.
In short, while the results of the primary and auxiliary models are, as would be expected,
not perfectly consistent, the essential conclusion of them is: There appears to be a strong,
statistically significant, positive relationship between public sector corruption and the toll in
terms of deaths that major earthquakes cause.
5. Conclusions
Our primary contention in this study is that public sector corruption, especially as it
31
applies to the building trades, can be expected to lead to the construction of houses, apartment
buildings, and other structures in a seismically-insensitive manner. Should no quake hit in the
lifetime of the project, one would have to conclude that the conspiracy between builder and
public official did little harm. To the contrary, as both the vignettes described in Section 2 and
our empirical analysis show, when a quake strikes these substandard structures, the conspiracy
extends from builder and public official to a much more lethal one between nature and man. We
show this by analyzing 344 large earthquakes, from 42 countries, occurring between 1975 and
2003, using two of the most commonly employed measures of public sector corruption and
controlling for the obvious endogeneity of all indices of corruption. Across various
specifications, we find a country’s level of public sector corruption to be positively and
significantly correlated with the fatalities caused by a large quake, regardless of which corruption
index is used, the control variables included, or the estimation strategy employed.
While major earthquakes are the example used to show this, similar results may well
exist for other large-scale natural disasters that might afflict a country such as hurricanes,
fire/wind storms, mud slides, and the like. At a minimum, it is fair to say that the results
presented here are clearly consistent with the anecdotal evidence from the U.S. Gulf Coast’s
recent experience with Hurricane Katrina which struck, just to the east of New Orleans, Louisiana
on August 29, 2005 with winds up to 125 miles per hour accompanied by heavy rains.
30
No
doubt, the images of this much beloved U.S. city flooded, with thousands of its inhabitants
clinging to rooftops of buildings disparately awaiting rescue, remain firmly etched in most minds.
The relevant question for the present analysis is whether corruption of the type discussed here
played a role in the city’s plight? In this regard, Carrns (2005) notes that the much of the
flooding was due to ill-designed and poorly maintained levees and, further, that the levee control
boards were classic examples of wasteful contracts and patronage hiring which directly
contributed to substandard levee construction and maintenance. And perhaps most telling, long-
30
For a thorough discussion of Katrina, see Shughart (2006).
32
time U.S. Representative from the area, Billy Tauzin, put his state’s tradition of corruption this
way, “Half of Louisiana is under water and the other half is under indictment,” (Gibbs et al.,
2005, p. 45).
Clearly, even with the most honest contractors and inspectors employed, the most
benevolent of disaster planning and management staffs in place, the strongest possible political
commitment to protect the public from nature’s occasional wrath, and with the very best of
private foresight and planning, natural disasters will continue to take their toll. To the extent that
one accepts the foregoing analysis, however, it is equally clear that this toll can be significantly
reduced by eliminating public sector corruption wherever possible.
33
References
Alesina, A. and B. Weder (2002). “Do Corrupt Governments Receive less Foreign Aid?”
American Economic Review, 92, 1126-1137.
Anbarci, N., M. Escaleras, and C. Register (2005). “Earthquake Fatalities: The Interaction of
Nature and Political Economy,” Journal of Public Economics, 89, 1907-1933.
Anbarci, N., M. Escaleras, and C. Register (2006). “Traffic Fatalities and Public Sector
Corruption,” Kyklos, 59, 327-344.
Anderson, G. and R. Tollison (1991). “Congressional Influence and Patterns of New Deal
Spending, 1933-1939,” Journal of Law and Economics, 34, 161-175.
Bardhan, P. (1997). “Corruption and Development: A Review of Issues,” Journal of Economic
Literature, 35, 1320-1346.
Beck, P. and M. Maher (1986). “A Comparison of Bribery and Bidding in Thin Markets,”
Economics Letters, 20, 1-5.
Bliss, C. and R. Di Tella (1997). “Does Competition Kill Corruption?” Journal of Political
Economy, 105, 1001-1023.
BBC News (2001). “Press Blames Corruption for Quake Losses,” (January 29).
http://www.news.bbc.co.uk/2/hi/world/monitoring/media_reports/1142765.stm
.
Bohlen, C. (1999). “Turkish Earthquake Survivors Blame Corruption,” New York Times (August
20). http://www.wakingbear.com/turkey1.htm
.
Bribe Payers Index (2002). In Transparency International Global Corruption Report 2003. (Pluto
Press: London)
.
Carnes, A. (2005). “Long before the Flood: New Orleans System was Prime for Leaks,” Wall
Street Journal, November 25, 2005.
Control Risks Group (2002). Facing Up to Corruption. (Control Risks Group: London).
Cooperative Institute for Research in Environmental Sciences (2001). “26 January 2001 Bhuj
Earthquake, Gujarat, India.” http://www.cires.colorado.edu
Depken, C. and C.L. Lafountain (2006). “Fiscal Consequences of Public Corruption: Empirical
Evidence from State Bond Ratings,” Public Choice, 126, 75-85.
Dunbar, P., R. Bilham, and M. Laituri (2002). “Earthquake Loss Estimation for India Based on
Macroeconomic Indicators,” in T. Beer and A. Ismail-Zadeh, eds., Risk Science and
Sustainability: Science for Reduction of Risk and Sustainable Development of Society
(Kluwer Academic Publishers: Dordrecht, Holland).
Garrett, T. and R. Sobel (2003). “The Political Economy of FEMA Disaster Payments,”
Economic Inquiry, 41, 496-509.
34
Gibbs, N., et. al. (2005), “An American Tragedy,” Time, 166, 44-49.
Habib, M. and L. Zurawicki (2002). “Corruption and Foreign Direct Investment,” Journal of
International Business Studies, 33, 291-307.
Huntington, S. (1968). Political Order in Changing Societies. (Yale University Press: New
Haven, CT).
International Association for Earthquake Engineering (1996). Regulations for Seismic Design: A
World List-1996. (IAEE: Tokyo).
International Conference of Building Officials (1987). Uniform Building Code: Structural
Engineering Design Provisions. (ICBO: Whittier, CA).
Karahan, G., R. Coats, and W. Shughart II (2006). “Corrupt Political Jurisdictions and Voter
Participation,” Public Choice, 126, 87-106.
Khan, M. (2005). “The Death Toll From Natural Disasters: The Role of Income, Geography, and
Institutions,” Review of Economics and Statistics, 87, 271-284.
Kinzer, S. (1999). “The Turkish Quake’s Secret Accomplice: Corruption,” New York Times
(August 17). http://www.library.cornell.edu/colldev/mideast/tquak.htm
.
Knack, S. and P. Keefer (1995). “Institutions and the Economic Performance: Cross-Country
Tests Using Alternative Institutional Measures,” Economics and Politics, 7, 207-227.
La Porta, R., et. al., (1999). “The Quality of Government,” The Journal of Law, Economics and
Organization, 15, 222-279.
Leamer, E. E. (1985). “Sensitivity Analysis Would Help,” American Economic Review, 75, 31-
43.
Lewis, J. (2005). “Earthquake Destruction: Corruption on the Fault Line,” in Transparency
International, Global Corruption Report 2005, 23-30.
Leff, N. (1964). “Economic Development through Bureaucratic Corruption,” American
Behavioral Scientist, 8, 8-14.
Levine, R. and D. Renelt (1992). “A Sensitivity Analysis of Cross-Country Growth Regressions,”
American Economic Review, 82, 942-963.
Leys, C. (1965). “What is the Problem about Corruption?” Journal of Modern African Studies, 3,
215-230. Reprint in A.J. Heidenheimer, M. Johnston, and V.T. Le Vine (Eds.), Political
Corruption: A Handbook, 51-66, 1989. (Oxford University Press: Transaction Books).
Lien, D. (1986). “A Note on Competitive Bribery Games,” Economics Letters, 22, 337-341.
Mauro, P. (1995). “Corruption and Growth,” Quarterly Journal of Economics, 110, 681-712.
Meon, P. and K. Sekkat (2005). “Does Corruption Grease or Sand the Wheels of Growth?”
35
Public Choice, 122, 69-97.
Mo, P. (2001). “Corruption and Economic Growth,” Journal of Comparative Economics, 29, 66-
79.
Pejhan, S. (2003). “Ready for Future Bam’s?” The Iranian (January 3).
www.iranian.com/Opinion/2004/January/Again/index.html.
Pellegrini, L. and R. Gerlagh. (2004). “Corruption’s Effect on Growth and its Transmission
Channels,” Kyklos, 57, 429-456.
Ramcharan, R. (2005). “How Big are the Benefits of Economic Diversification? Evidence from
Earthquakes,” IMF Working Paper.
Schulze, W., D. Brookshire, R. Hageman, and J. Tschirhart (1987). “Benefits and Costs of
Earthquake Resistant Buildings,” Southern Economic Journal, 53, 934-951.
Serra, D. (2006). “Empirical Determinants of Corruption: A Sensitivity Analysis,” Public Choice,
126, 225-256.
Shughart II, W. (2006). “Katrinanomics: The Politics and Economics of Disaster Relief,” Public
Choice, forthcoming.
Skidmore, M. and H. Toya (2002). “Do Natural Disasters Promote Long-Run Growth?”
Economic Inquiry, 40, 664-687.
Smith, C. (2003). “Quake Demolishes Confidence in Algerian Rulers,” New York Times (May
30). http://www.algeria-watch.org/en/articles/2003/demolish_confidence.htm.
Stansbury, N. (2005). “Exposing the Foundations of Corruption in Construction,” in
Transparency International, Global Corruption Report 2005, 36-50.
Treisman, D. (2000). “The Causes of Corruption: A Cross-National Study,” Journal of Public
Economics, 76, 399-457.
Turkish Daily News (2006). “Ministry Takes over Bodrum’s Building Rights,” (August 8).
www.turkishdailynews.com.tr/article.php?enewsid=50947.
Wooldridge, J. (2002). Econometrics of Cross Section and Panel Data. (Cambridge, MA: MIT
Press).
World Seismic Safety Initiative (2001). Interdisciplinary Observations on the January 2001
Bhuj, Gujarat Earthquake. www.rms.com/Publications/Bhuj_EQ_Report.pdf.
36
Table 1. Descriptive Statistics
Variable Observations Mean Standard
Deviation
Minimum Maximum
FATATLITIES
344 849.03 4,438.77 0 50,000
COR-ICRG
344 3.78 1.41 1 7
COR-TI
299 4.20 1.94 1.7 9.4
FREQUENCY
344 0.51 0.24 0.01 0.80
DISTANCE
344 125.09 184.69 5.94 2,316.5
MAGNITUDE
344 6.80 0.59 6 8.5
DENSITY
344 269.41 1,095.73 0.20 13,148.15
POPULATION
344 9,330.63 20,212.79 1.98 166,052.8
GDPPC
344 6,652.37 10,423.57 101.58 44,774.71
AFRICA
344 0.03 0.17 0.00 1.00
AMERICA
344 0.35 0.48 0.00 1.00
ASIA
344 0.43 0.50 0.00 1.00
EUROPE
344 0.19 0.39 0.00 1.00
37
Table 2. Relations between Corruption, Fatalities, Magnitude, and Distance
Variable Low-
Corruption
Mean
(n=90)
Low-
Corruption
Std. Dev.
High-
Corruption
Mean
(n=86)
High-
Corruption
Std. Dev.
Difference
t-test
FATALITIES
351.42
2,607.76
1,530.36
5.840.81
1,178.96*
(1.79)
MAGNITUDE
6.73
0.60
6.81
0.63
0.08
(0.89)
DISTANCE
152.23
299.66
103.78
90.90
48.45
(1.45)
Note: t-statistics for differences in means are in parentheses; * denotes significance beyond the 0.10 level.
38
Table 3. Determinants of Earthquake Fatalities
Variable
FATALITIES
Negative Binomial Estimates
COR-ICRGP
-1.876**
(0.891)
FREQUENCY
-0.671**
(0.213)
DISTANCE
-1.042**
(0.192)
MAGNITUDE
20.428**
(2.517)
DENSITY
0.131
(0.132)
POPULATION
0.315**
(0.145)
GDPPC
-0.340**
(0.174)
AFRICA
0.375
(0.819)
ASIA
0.514
(0.401)
EUROPE
1.960**
(0.511)
Intercept
-27.063**
(5.368)
Number of Observations
344
LR Chi-Square α
84,000**
LR Chi-Square FM
141.76**
ML R-Square
0.34
Notes: * denotes significance beyond the 0.10 level while ** denotes
significance beyond the 0.05 level.
39
Table 4. Determinants of Earthquake Fatalities (Alternate Models)
Variable
(1)
FATALITIES
Negative Binomial
Estimates
(2)
FATALITIES
<500 Deaths
Negative Binomial
Estimates
(3)
FATALATIES
<500 Deaths
Negative Binomial
Estimates
COR-ICRGP
-1.804**
(0.855)
COR-TIP
-6.008**
(2.295)
-3.674**
(1.421)
FREQUENCY
-0.734**
(0.251)
-0.148
(0.194)
-0.204
(0.207)
DISTANCE
-1.143**
(0.208)
-0.480**
(0.178)
-0.521**
(0.184)
MAGNITUDE
22.540**
(2.718)
4.918**
(1.920)
5.456**
(2.104)
DENSITY
0.145
(0.136)
0.038
(0.108)
-0.018
(0.115)
POPULATION
0.184
(0.148)
0.256**
(0.110)
0.177
(0.115)
GDPPC
0.424
(0.462)
-0.138
(0.171)
0.320
(0.330)
AFRICA
-3.618**
(1.324)
0.709
(0.746)
-2.682**
(1.089)
ASIA
0.535
(0.340)
-0.495
(0.314)
-0.306
(0.308)
EUROPE
0.559
(0.814)
0.301
(0.476)
-0.295
(0.561)
Intercept
-30.303**
(5.902)
-2.486
(3.709)
-3.970
(3.909)
Number of
Observations
305
310
279
LR Chi-Square α
70,000** 2,300** 1,800**
LR Chi-Square FM
124.89** 36.62** 41.56**
Notes: * denotes significance beyond the 0.10 level while ** denotes significance beyond the 0.05 level.
40
Appendix 1: Sample Countries, Number of Earthquakes, Fatalities, and
Corruption
Country Number of
Earthquakes
Average Number of
Fatalities
Average Corruption
(COR-ICRG)
Algeria 2 3,633 2.3
Australia 2 0 6.0
Bolivia 1 105 4.0
Chile 15 15 4.2
China 24 1,131 4.0
Colombia 10 88 4.0
Congo 2 5 3.5
Costa Rica 5 11 6.0
Dominican Republic 2 4 3.0
Ecuador 6 171 4.0
Egypt 1 12 4.8
El Salvador 3 401 3.5
Ethiopia 1 2 4.0
Greece 12 10 4.4
Guatemala 5 4,486 3.4
Guinea 1 443 4.0
Honduras 1 0 3.0
India 8 4,027 3.3
Indonesia 39 248 2.2
Iran 21 5,932 3.3
Italy 3 1,567 4.5
Japan 21 283 5.6
Malawi 1 9 5.0
Mexico 20 501 3.1
New Zealand 6 0 6.8
Nicaragua 2 90 5.7
Pakistan 9 50 3.1
Panama 2 14 3.0
Papau New Guinea 13 174 3.4
Peru 15 27 3.7
Philippines 21 393 2.4
Portugal 2 33 4.5
Romania 2 8 3.8
Russia 7 286 3.1
Sudan 2 16 3.0
Taiwan 9 290 3.9
Tanzania 1 0 2.0
Turkey 16 1,740 3.1
Uganda 1 7 4.0
United States 30 5 6.2
Venezuela 2 42 4.0
Yemen 2 1,433 4.0
41
Appendix 2. Description of Data and Sources
Variables Description Source
FATATLITIES
Number of casualties due to an
earthquake.
NGDC Significant Earthquake
Database
COR-ICRG
Corruption index from ICRG, annual
surveys from 1982-1985: 6(lowest
corruption), 0(highest corruption).
International Country Risk Guide
(2005)
COR-TI
Corruption index from Transparency
International, survey 1998, 10 (lowest
corruption), 1 (highest corruption).
Transparency International
The data set is available online at
www.transparency.org
.
FREQUENCY
Ratio of the number of 6+ Richter scale
earthquakes occurring within a country
during the prior 100 years, to 100.
NGDC Significant Earthquake
Database
DISTANCE
Distance between the epicenter and the
affected geographic region measured by
the square root of the sum of the squares
of the depth and surface-distance.
Latitude, longitude and depth of
epicenter: NGDC Significant
Earthquake Database.
Latitude and longitude of
affected region: Getty Thesaurus
of Geographic Names on Line
(www.getty.edu
).
MAGNITUDE
Magnitude of an earthquake determined
from the logarithm of the amplitude of
waves recorded by seismographs;
measured by the Richter-Scale.
NGDC Significant Earthquake
Database
DENSITY
Population of the province(s) affected per
square kilometers.
The World Gazetteer
(www.world-gazetteer.com
),
GeoHive (www.xist.org
).
POPULATION
Population of the province(s) affected,
expressed in “thousands”.
The World Gazetteer
(www.world-gazetteer.com
),
GeoHive (www.xist.org
), and
Population Statistics
(www.library.unn.nl
).
GDPPC
Real GDP per capita, expressed in
constant (1995) U.S. dollars.
World Bank World Development
Indicators 2002
Instruments
DEMOCRACY
Index scaled 0-10 with higher values
indicating more thoroughgoing
democratic institutions.
Polity IV database
INTCONFLICT
Index scaled 0-12 with high values
indicating very low risk of political
violence in the country.
International Country Risk Guide
(2005)
PROTESTANT
Share of the total population that are
Protestant
La Porta R., et al., (1999).
ENGLISH
Dummy variable indicating that the legal
origin of the country is English
La Porta R., et al., (1999).
42
Appendix 3. Determinants of Corruption
Variable
(1)
COR-ICRG
Two-Bound
Tobit Estimates
(2)
COR-TI
OLS
Estimates
(3)
COR-ICRG
Two-Bound
Tobit Estimates
(4)
COR-TI
OLS
Estimates
GDPPC
0.369**
(0.077)
0.844**
(0.076)
0.365**
(0.087)
0.896**
(0.085)
DEMOCRACY
0.039*
(0.023)
-0.046**
(0.021)
0.039
(0.025)
-0.056**
(0.022)
INTCONFLICT
0.174**
(0.029)
0.032
(0.027)
0.189**
(0.033)
0.024
(0.030)
PROTESTANT
0.003
(0.008)
0.026**
(0.008)
0.002
(0.008)
0.024**
(0.009)
ENGLISH
0.822**
(0.323)
1.320**
(0.284)
0.787**
(0.345)
1.318**
(0.307)
AFRICA
0.061
(0.472)
-0.612
(0.499)
0.177
(0.525)
-0.669
(0.563)
ASIA
-0.533**
(0.182)
0.156
(0.154)
-0.556**
(0.191)
0.187
(0.163)
EUROPE
-0.579**
(0.217)
-0.785**
(0.204)
-0.556**
(0.191)
-0.902**
(0.227)
FREQUENCY
-0.130
(0.104)
0.083
(0.090)
-0.171
(0.245)
0.043
(0.098)
DISTANCE
0.097
(0.085)
-0.067
(0.071)
0.074
(0.092)
-0.073
(0.077)
MAGNITUDE
-1.745**
(0.881)
-0.513
(0.753)
-1.740*
(0.982)
-0.916
(0.836)
DENSITY
-0.038
(0.050)
-0.012
(0.045)
-0.035
(0.053)
-0.011
(0.048)
POPULATION
0.024
(0.055)
-0.226**
(0.056)
0.010
(0.578)
-0.236**
(0.059)
Intercept
1.898
(1.796)
0.426
(1.576)
1.995
(1.946)
0.828
(1.692)
Number of Obs.
344 305 310 279
LR Chi-Square FM
216.08** 198.44**
F-Statistic FM
62.92** 58.44**
Pseudo R-Square
0.16 0.17
R-Square
0.74 0.74
Note: * denotes significance beyond the 0.10 level while ** denotes significance beyond the 0.05 level.