World Development Vol. 59, pp.

275–297, 2014
Ó 2014 Elsevier Ltd. All rights reserved.
0305-750X/$ - see front matter
www.elsevier.com/locate/worlddev
http://dx.doi.org/10.1016/j.worlddev.2014.01.012

How Ethnic Diversity Affects Economic Growth

ERKAN GÖREN *
Carl von Ossietzky University Oldenburg, Campus Haarentor, Building A5, 26111 Oldenburg, Germany
Summary. — This paper investigates the empirical relationship between ethnic diversity, polarization, and economic growth. Ethnicity is
assumed to affect economic growth through a number of possible transmission channels that are generally included in cross-country
growth regressions. This paper provides an extensive empirical analysis shedding light on the various sources through which ethnic diver-
sity and polarization affects economic growth indirectly. It advances and empirically establishes the hypothesis that ethnic diversity has a
strong direct negative impact on economic growth, whereas ethnic polarization has non-negligible indirect economic effects through the
specified channel variables.
Ó 2014 Elsevier Ltd. All rights reserved.

Key words — ethnic diversity, ethnic polarization, transmission channels, economic growth

1. INTRODUCTION Kurlat, and Wacziarg (2003), providing new, on a highly

In recent years, there has been growing research interest in
the relationship between ethnic diversity and economic
growth. In a seminal paper, Easterly and Levine (1997)
showed that the growth rate of GDP per capita is inversely re-
lated to the degree of ethnic diversity within a country. They
argued that the poor economic performance of most of the
African countries is due partly to the large number of different
ethnic groups living in the same country and partly to the ab-
surd borders drawn by former colonial powers. However,
when controlling for other factors—namely, human capital,
and political instability—the effect of the ethnic diversity mea-
sure was weaker. This may have been due to the variables in-
cluded in a standard growth regression acting as transmission
channels for the indirect impact of ethnic diversity on eco-
nomic growth. 1 This point raises the question of whether
highly diverse societies suffer indirectly from poor government
performance. Although Easterly and Levine (1997) addressed
this important issue by means of ethnic diversity and govern-
ment performance indicators, they also provided a regression
of government performance indicators on the only explana-
tory variable, ethnic diversity. They concluded that ethnic
diversity is indeed accompanied by low school attainment,
financial depth, and infrastructure quality. Furthermore, they
showed that ethnic diversity also leads to higher market distor-
tions.
A more comprehensive analysis of ethnic diversity and the
quality of government was carried out by La Porta, Lopez-
de-Silanes, Shleifer, and Vishny (1999). In their article, they
investigated the influence of a broader set of possible determi-
nants of the quality of good government performance. They
argued that good economic institutions, especially those in
the public sector, promote per capita GDP growth—for exam-
ple, by limiting the private influence of the government and
establishing an uncorrupt bureaucracy and legal system that
protects property rights and enforces contracts. One of their
main conclusions is that ethnic diverse societies exhibit inferior
government performance. The findings in Collier (2001) also
suggest that ethnic diverse societies suffer from bad public sec-
tor performance, which in turn reduces economic perfor-
mance.
An updated analysis of the findings in Easterly and Levine
(1997) was performed by Alesina, Devleeschauwer, Easterly,
275
disaggregated level, measures of ethnic, linguistic, and reli-
gious diversity for about 190 countries. They criticized the
widely used ethnic measure from Atlas Narodov Mira for its
inability to distinguish between ethnic and linguistic differ-
ences. Such a distinction may well be unproblematic in African
and European countries, where people identify themselves by
both ethnic group and language: in these contexts, the two cri-
teria coincide. This is not the case in Latin American coun-
tries, where people are more or less homogeneous in terms
of language (e.g., Spanish or Portuguese) but distinct in terms
of ethnic membership. Alesina et al. (2003) reran the regres-
sions of Easterly and Levine (1997), but on their ethnic diver-
sity measure, which relies on ethnic distinctions rather than
linguistic distinctions. As they controlled for variables that
can be interpreted as channels through which ethnic diversity
affects growth (for example schooling, government consump-
tion, infrastructure quality, etc.), the magnitude of the ethnic
effect vanished and became statistically insignificant, suggest-
ing once again that ethnic diversity may affect economic per-
formance indirectly through these channels. Besides the
variable for market distortions, the authors found a statisti-
cally significant relationship of ethnic diversity with schooling,
political instability, financial depth, the fiscal surplus to GDP
ratio, and infrastructure quality using the same econometric
specification as Easterly and Levine (1997). However, the lack
of specified transmission channels leads to the conjecture that
the estimated effect of ethnic diversity in each of the loosely
specified transmission channels suffers from omitted variables
bias. It is not clear precisely what the diversity measure cap-
tures when the transmission channels are not well specified.
Furthermore, the authors failed to explain the importance of
each of the transmission channels for real per capita income
growth.
* I am grateful to Romain Wacziarg for sharing his data with me. Com-
ments from three anonymous referees improved the manuscript substan-
tially. I would like, also, to thank Jürgen Bitzer, the seminar participants
at the Carl von Ossietzky University Oldenburg 2011, the NIW workshop
in Applied Economics in Hanover 2012 and the Tuborg Research Center
2013, Aarhus University, for useful comments and suggestions. Final
revision accepted: January 19, 2014.
276 WORLD DEVELOPMENT
Another strand of literature investigates ethnic violence,
especially civil wars, and its detrimental effects on economic
performance. Garcia-Montalvo and Reynal-Querol (2005b)
stressed the importance of political instability on economic
development when countries exhibit a high degree of potential
ethnic conflict. The authors found that in ethnic diverse soci-
eties, the diffusion of ideas is impeded, especially when the dif-
ferent ethnic groups are in conflict. In such environments of
latent ethnic violence, business as usual is impossible because
all levels of economic activity are affected. Empirical studies
of this problem by Collier (2001) and Garcia-Montalvo and
Reynal-Querol (2005b) reported no positive relationship be-
tween ethnic diversity and the incidence of civil wars. Instead,
they found that high ethnic diversity makes societies safer be-
cause the coordination costs are higher and because no one
ethnic group is large enough to dominate the others. Garcia-
Montalvo and Reynal-Querol (2005b) found that not ethnic
diversity but rather polarization is one of the main factors
affecting the incidence of civil wars. Unfortunately, a measure
of ethnic diversity is unable to capture ethnic polarization
across countries. As the authors stressed, there is less violence
in highly homogeneous and highly diverse societies, and the
incidence of civil wars is the highest in societies where the rul-
ing ethnic group dominates a non-negligible minority. Such
cases require a measure of polarization, rather than diversity,
that captures the latent danger of ethnic conflicts. 2
Despite the fact that the ethnic composition of countries
may have strong indirect effects on economic growth, the
above articles deal primarily with its direct empirical quantifi-
cation. Campos and Kuzeyev (2007) confirmed that more re-
search is needed in order to identify the main transmission
channels through which ethnic diversity and polarization af-
fects economic performance. 3 A first attempt to assess the
quantitative importance of possible transmission channels by
which ethnic diversity and polarization indirectly affects
growth was made by Garcia-Montalvo and Reynal-Querol
(2005a). They specified and estimated a comprehensive system
of equations determining growth and possible transmission
channels through which ethnic diversity and polarization
may affect growth. They argued that ethnic polarization nega-
tively affects growth because it reduces the rate of investment
and increases public consumption and the incidence of civil
wars. By contrast, their results suggest that ethnic diversity
does not affect real per capita income growth indirectly
through these channels but rather directly, for example, by
reducing the diffusion of ideas across the economy. These re-
sults are in contrast to the estimates of Easterly and Levine
(1997) and Alesina et al. (2003), who find a negative associa-
tion between ethnic diversity on the one hand and fiscal stance
and political stability on the other.
The starting point of this paper are the empirical findings of
indirect effects of ethnic diversity and polarization on eco-
nomic growth reported in Garcia-Montalvo and Reynal-Quer-
ol (2005a). Studies on the relationship between ethnicity and
economic growth have shown the importance of focusing on
the effects of channel variables that are also important explan-
atory variables in reduced-form growth regressions. This issue
deserves closer examination, both with regard to the ethnic
measures used and the specification of the relevant channels.
Opening the black box on how both ethnic measures affects
economic growth can contribute a more thorough understand-
ing of its costs and allow its direct and indirect effects to be dis-
tinguished from its economic and political outcomes. The
narrow focus on direct effects of ethnicity on social or political
outcomes neglects important indirect effects, as discussed
above. As will be shown, ethnicity directly plays an important
role in the social and political institutions that influence eco-
nomic development. An analysis of the indirect economic ef-
fects must, in a first step, formulate sound hypotheses
explaining why ethnicity should affect economic growth
through an explicit channel. For instance, it has been argued
that ethnic polarized societies may breed ethnic hatred, and
in the worst case, may end up in civil war. But civil wars hin-
der the economic development of countries. So far, the eco-
nomics literature on the indirect effects of ethnicity has
focused primarily on the channels of investment, government
consumption, and civil war. Further research is needed on
the remaining channel variables in cross-country growth
regressions. For example, how do both ethnic measures affect
economic growth through schooling, political instability, mar-
ket distortions, foreign trade, and the fertility rate? All these
variables are generally known to affect long-run economic
growth and hence economic development of countries. This re-
search undertakes an elaborative attempt in trying to integrate
the two so far separate literatures on ethnic polarization as a
source of ethnic conflict and ethnic diversity as a factor for
the macroeconomic performance of countries. Therefore, a
closer examination of the involved transmission channels
seems necessary. The methodology chosen here is to specify
all relevant channel variables that appear on the right-hand
side of the growth equation. In this paper, each channel vari-
able affecting economic growth is specified on previous find-
ings in the political and economics science literature. This
will serve to assess the direct and indirect effects of both ethnic
measures on economic growth for each of the channel vari-
ables. The empirical analysis suggests that ethnic diversity is
a strong predictive measure for the direct effect of ethnicity
on economic growth, whereas ethnic polarization has strong
predictive power for the indirect effects. Furthermore, test sta-
tistics for the nonlinear combinations of estimators using the
delta method are provided. 4 Contrary to most economic
growth studies regarding ethnic diversity and polarization,
an updated data set for the period 1960–99 is used. One pos-
sible advantage of using an updated data set is that it allows
to test whether the relationship between ethnicity and growth
can be confirmed. Furthermore, the updated data set delivers
more observations for the extended econometric system and
hence will improve the efficiency of the parameter estimates.
The remaining part of the paper is organized as follows. Sec-
tion 2 deals with the measurement of ethnic diversity and
polarization and highlights some important conceptual differ-
ences. Section 3 describes possible transmission channels
through which both ethnic measures may indirectly influence
economic growth, and more importantly, discusses potential
reasons why these measures may be related to each of the
channel variables. Section 4 deals with the specification of
the relevant transmission channels from existing political
and economics science studies. Furthermore, Section 4 also
provides a brief discussion of the underlying data sources
and highlights the econometric methodology in quantifying
the economic effects of ethnicity on growth. The empirical re-
sults are given in Section 5. Section 6 checks the robustness of
the results and presents further insights into how ethnic diver-
sity and polarization are assumed to affect economic growth.
Finally, Section 7 concludes.
2. ON THE MEASUREMENT OF ETHNIC DIVERSITY
AND POLARIZATION
The empirical literature pertaining to the determinants of
socioeconomic outcomes generally relied on two different
 HOW ETHNIC DIVERSITY AFFECTS ECONOMIC GROWTH
ethnic measures. Because the underlying empirical analysis
clearly distinguishes between ethnic diversity and polarization,
this section describes both measures in some detail. As shown
below, both measures are conceptually different and the choice
of the most appropriate index for empirical work depends on
the purpose of the study.
The index of ethnic diversity is a Herfindahl-based measure
of the following form:
X
G
ETHFRAC ¼ 1  p2i ; ð1Þ
i¼1
where pi is the proportion of group i and G is the number of
ethnic groups in the particular country. This index has a sim-
ple interpretation, it represents the probability that two ran-
domly selected people do not belong to the same ethnic
group. The main notion of this measure is to capture the de-
gree of ethnic fragmentation in a society. Because this index
is bounded between 0 and 1, higher values corresponds to
more ethnic diversity. Furthermore, ethnic diversity increases
with the number of equally sized groups.
In contrast, the measure of ethnic polarization is primarily
intended to capture the potential of latent social conflicts in
the society. The main idea behind this measure is, that social
conflict is the highest when the society consists of a finite num-
ber of highly distinguishable groups. The concept of polariza-
tion was firstly introduced by Esteban and Ray (1994) as the
sum of interpersonal antagonism, where the distinguishable
feature lies predominantly on income differences through the
different groups. Esteban and Ray’s (1994) general class of
polarization measures, ER, can be written as:
X
G X
G
ERðy; KÞ ¼ K pi pj T ½Iðpi Þ; aðdðy i ; y j ÞÞ; ð2Þ
i¼1 j¼1
where y consists of a vector of intergroup differences and K is
an arbitrary constant. The term T ½Iðpi Þ; aðdðy i ; y j ÞÞ measures
the antagonism between group i and j. This antagonism de-
pends positively on the degree of self-identification toward
the own group, Iðpi Þ, and the alienation felt toward the other
groups, aðdðy i ; y j ÞÞ. The two authors narrow down the class of
allowable functions by imposing three important axioms: 5
X
G X
G equations model. Annett (2001) also used system estimates
ERða; D; KÞ ¼ K p1þa pj d ij ; ð3Þ
i
i¼1 j¼1
where d ij measures the cultural distance (alienation) between
group i and j. The antagonism between two particular groups
equals pai d ij . Because there is a proportion pj of group j the
effective antagonism between group i and j is pai pj d ij . Given
the size of group i the total effective antagonism between
group i and j equals pi1þa pj d ij . Therefore, Eqn. (3) refers to
the total effective antagonism between each pair of ethnic
groups within a country. The derived conditions under which
a polarization measure has to satisfy some crucial axioms,
bound a between ð0; a , where a ’ 1:6. The constant factor
K > 0 has no influence on the polarization order, but is used
for population normalization.
A particular special situation is obtained if each group feel
equally alien toward all groups other than their own. That
is, if d ij ¼ 0 for i ¼ j and d ij ¼ 1 for i – j. This measure of dis-
crete polarization was firstly introduced by Reynal-Querol
(2002) and has important impacts on the other parameters
of the polarization measure in Eqn. (3). First, their is only
one value for the polarization sensitivity ða ¼ 1Þ for which
the properties of a polarization measure are satisfied. Second,
277
only one value for the constant ðK ¼ 4Þ restricts this measure
between 0 and 1. Hence, the discrete polarization measure
constructed by Reynal-Querol (2002) can be simplified to:
G  2
X 0:5  pi
ETHPOL ¼ 1  pi : ð4Þ
i¼1
0:5
Ethnic polarization is maximized if there are two equally sized
groups. Furthermore, ethnic polarization decreases with the
number of equally sized groups.
The following Figure 1 shows the relationship between eth-
nic diversity and polarization for the 100 countries under
study. According to the Figure, higher values for ethnic diver-
sity up to 0.4 are associated with higher ethnic polarization. In
the intermediate part there is virtually no correlation between
ethnic diversity and polarization, whereas higher values of eth-
nic diversity are accompanied with falling ethnic polarization.
Therefore, ethnic polarization is the lowest for highly homog-
enous and highly diverse countries.
From a theoretical and empirical standpoint it seems rea-
sonable to distinguish these two ethnic concepts and its partic-
ular influence on economic growth. Therefore, this study
undertakes an elaborate attempt quantifying potential direct
and indirect effects of ethnic diversity and polarization on eco-
nomic growth.
3. ETHNIC DIVERSITY AND GROWTH:
TRANSMISSION CHANNELS
The empirical analysis is based on the findings of Garcia-
Montalvo and Reynal-Querol (2005a) regarding the indirect
effects of ethnic diversity and polarization on economic
growth through the channels of investment, government con-
sumption, and civil war. This kind of econometric methodol-
ogy is quite common in the economic and political science
literature. For instance, Alesina and Perotti (1996) studied
the effects of income distribution on investment, choosing
political instability as the channel that links these two vari-
ables. Tavares and Wacziarg (2001) analyzed the indirect ef-
fects of democracy on economic growth through several
transmission channels in the framework of a simultaneous
to study whether ethnic diversity leads to higher political insta-
bility, in turn increasing government consumption to appease
conflicts between ethnic groups. The starting point of this
analysis is therefore the assumption that ethnic diversity and
polarization may influence economic growth through a num-
ber of possible transmission channels, which are generally in-
cluded in a standard growth equation. Despite the progress
made in Garcia-Montalvo and Reynal-Querol (2005a), several
important questions remain unresolved. For instance, it has
been argued that ethnic diversity may have a detrimental effect
on the level of schooling and hence economic growth (see be-
low), although no attempt has been made to include a fully
specified channel for schooling testing this hypothesis. Fur-
thermore, theoretical and empirical studies on the fertility rate
have predicted that countries with higher fertility rates will
have lower per capita income growth. Hence, an open ques-
tion is whether ethnic diversity and polarization is linked to
economic growth through the fertility channel. The economic
and political science literature provides explicit specifications
for each of the transmission channels. Therefore, in specifying
each channel equation, it is possible to estimate a full econo-
metric system to determine the direct and indirect effects of
ethnic diversity and polarization on economic growth. All
278 WORLD DEVELOPMENT

1
TZA
UGA
IND LBR
KEN
CIV
MLIBEN
PHL
MOZGAB
CMR
SEN
.8 IDN COD
CAFSLE
ZMB
IRN CAN
TGO
GHA GMB
COGNER
SDN BOL
NPL MWIMYSCOL
PERTTO
ECU
BRA
Ethnic Diversity

PAK
.6

MEX USA
CHE FJI
ZWE VEN BEL
GTMKWT
JOR
NIC
BWA MUS MAR
ZAF
DOM LKA
ESP CHL
SGP
ARG
.4

IRQ
GBR
THAJAM CYP
PNG
MRT
AUS
DZA LUX
BDI ISR
URY
HND
EGY
CRI
RWA
.2

SYR
LSONZL
SWE
SWZTUR
PRY
FIN
FRA
SLV
AUT
NLD
HTI
GRC
TUN
ITA
IRL
BGD
SAU
DNK
NOR PAN
JPN
ISL
KOR
PRT
0

0 .2 .4 .6 .8 1
Ethnic Polarization

Sub−Sahara Africa Latin America East Asia

OECD Other

Figure 1. Relationship between ethnic diversity and ethnic polarization.

channels used in the analysis are affected by ethnic diversity
and polarization from a theoretical and empirical point of
view.
In the following section, the transmission channels are dis-
cussed and explanations why these channels may be broadly
influenced by ethnicity. The selection of the transmission chan-
nels relies heavily on economic or political indicators that have
a strong empirical relationship to long-run economic growth
and are widely used in cross-country growth regressions.
(a) Investment
The allocation of talent in an economy is disturbed, when a
large proportion of people pursue high rent-seeking behavior.
Because of the lack in protecting property rights, especially in
Sub-Saharan African countries, the most talented people are
engaged in rent-seeking occupations (e.g., joining the army
or becoming government officials). These activities generally
redistribute wealth from others without creating new wealth,
leading to lower productivity and income growth. By reallo-
cating people from entrepreneurial activities into rent-seeking
activities, the accumulation of physical capital is hampered
due to lower technological progress and higher rent-seeking
individuals. 6 Easterly and Levine (1997) have shown, that eth-
nic diverse countries are more prone to rent-seeking behavior
by different ethnic groups leading to difficulties agreeing on
public goods such as infrastructure, education, and good gov-
ernment policy. Furthermore, as emphasized by Garcia-Mont-
alvo and Reynal-Querol (2005a), when a society is divided
along ethnic lines, tensions and other cleavages may emerge.
This in turn would lead some individuals to devote their re-
sources (time, labor, capital, etc.) to achieving political influ-
ence. These social costs, along with rent-seeking behavior,
imply a non-productive use of inputs and may reduce invest-
ments in productive sectors and thus inhibit economic growth.
As noted by Annett (2001), ethnic diverse countries may be
characterized by uncertain political environments, to which
investment activities are sensitive. 7
(b) Civil war
The potential causes of ethnic civil wars have been analyzed
extensively in Collier (2001), Reynal-Querol (2002) and
Garcia-Montalvo and Reynal-Querol (2005b). Civil wars are
accompanied by human capital destruction when the repres-
sion and killing of subversives who do not share their govern-
ment’s ideology causes people to flee into exile. 8 If these
refugees are highly educated, the outflow of human capital will
be severe and will hamper the country’s economic develop-
ment. Furthermore, the traumatic events of civil wars may
have lasting repercussions, requiring decades for social life
to recover. Civil wars also lead to a destruction of physical
capital and to lower investment ratios. The prevailing atmo-
sphere of uncertainty and risk may discourage investment in
both physical and human capital accumulation, factors that
are crucial for economic development. As mentioned by Col-
lier (2001), ethnic polarized societies may lead to dysfunctional
politics—of which civil wars may be viewed as the extreme
manifestation—if the different groups do not find cooperative
agreements. Therefore, societies divided along ethnic lines are
therefore more prone to ethnic violence, riots, and civil wars. 9
However, the question remains why some countries have expe-
rienced civil wars and others have not. Kanbur, Rajaram, and
Varshney (2011) argued that the potential consequences of
ethnic cleavages are mediated by the kind of state, political,
and civil institutions. Ethnically based politics in polarized
societies may lead to riots, which, in the worst case, may
end in ethnic civil wars. Political and economic factors in poor
 HOW ETHNIC DIVERSITY AFFECTS ECONOMIC GROWTH
and nondemocratic countries, can shape the identity of a par-
ticular group and cause ethnic hatred. 10
(c) Human capital
The provision of public goods like education in ethnic di-
verse countries are often not neutral but may be strategically
designed to establish the state’s authority over a population. 11
However, the question arises whether these societies have low-
er schooling levels due to this non-neutral provision of educa-
tion. Easterly and Levine (1997) noted that public goods like
education may bring less satisfaction to everyone in a society
when the country is highly ethnically diverse because of dis-
agreements between ethnic groups on issues like the language
of instruction, the learning content, location, etc. This may
lead the society to invest less in human capital.
An anthropological perspective on lower school perfor-
mance in ethnic diverse societies is formulated in Ogbu’s cul-
tural-ecological theory of minority school performance. 12
The theory states that differences in school performance in eth-
nic diverse societies may be caused by the treatment of minor-
ity groups, both in society at large and in the school system, as
well as by the minorities’ own perceptions and responses to
school resulting from their treatment. Ogbu and Simons
(1998) argue that in the case of involuntary minorities, 13 their
long experience of discrimination, racism, and conflict leads
them to distrust institutions of the ruling ethnic group.
Schools in particular are treated with suspicion because the
minorities believe that the public schools will not educate their
children as they do children from the ruling ethnic group.
Societal discrimination against minority groups (e.g., denigra-
tion of the minority culture and language, social and residen-
tial segregation, etc.) discourage them from investing in
schooling because of the perceived lower rewards of employ-
ment and wages for educational accomplishments.
(d) Government consumption
La Porta et al. (1999) have argued that the adverse effects of
ethnic diversity on economic growth through government con-
sumption result from the fact that ethnic diversity captures to
some degree the predisposition of ethnic groups in power to
redistribute wealth. 14 Annett (2001) analyzed the relationship
of ethnic diversity with political instability and government
consumption using a neoclassical growth model. The main
hypothesis was that ethnic diversity inherently leads to higher
levels of political instability and that this imposes a political
cost on the government, which risks being overthrown and los-
ing any rents from holding power. The government is forced to
use government expenditures to appease the competing groups
and mitigate the latent danger of ethnic conflicts in order to
reduce political instability and the resulting danger of being
overthrown.
(e) Political instability
Ethnic politics are more likely in societies divided along eth-
nic lines. 15 Alesina and La Ferrara (2005) emphasized that in
ethnic diverse countries, dictators use ethnic hatred to mobi-
lize support for one group over others in repressing ethnic con-
flict. La Porta et al. (1999) argued that in ethnic diverse
societies, it is common for the groups that come to power to
create government policies that expropriate (or kill) the ethnic
losers, restrict their freedom to organize opposition, and limit
the rights of ethnic groups outside the ruling group to con-
sume public goods. Therefore, as ethnic diversity increases,
279
government policies become less politically free, which may
end in politically unstable situations. Furthermore, as argued
by Annett (2001), when countries are strongly divided along
ethnic lines, they are likely to experience situations of political
violence and are prone to frequent breakdowns of law and or-
der. Moreover, Easterly and Levine (1997) argued that the
politically unstable environment in ethnic diverse countries
provide evidence of strong rent-seeking behavior and the
inability to find cooperative agreements on the provision of
public goods.
(f) Market distortions
La Porta et al. (1999) stressed that ethnic diverse societies
are associated with less political freedom (e.g., less democracy)
with the consequence that autocratic regimes may use their
power more arbitrarily. Moreover, as stressed by Easterly
and Levine (1997), these societies may produce situations of
uncoordinated government ministries, each pursuing its own
rent-seeking strategy without taking into account the effect
of its actions on others’ rents. For example, an overvalued offi-
cial exchange rate and strict exchange rate controls benefit
those in power who resell foreign currency on the black mar-
ket. Furthermore, an overvalued official exchange rate creates
incentives to invest domestic capital abroad because of the fear
of devaluation.
(g) Trade openness
A series of studies by Alesina and Spolaore (1997), Alesina
and Wacziarg (1998), and Alesina, Spolaore, and Wacziarg
(2000) emphasize the determinants of the number and size of
countries. Their argument is that country size emerges from
a trade-off between economies of scale in supplying public
goods in large countries and the costs of ethnic and cultural
diversity, which may increase with the size of countries.
According to their findings, in a world of free international
trade, the size of a country does not determine the size of
the market. Therefore, small and homogeneous countries
can benefit from free trade without sharing public policies over
which they differ in opinion. 16 But, conditional on these deter-
minants, some ethnic diversity in a country may be beneficial
to international trade because of the existence of transnational
ethnic or cultural networks. 17 One reason may be that ethnic
diverse societies also have more diverse labor skills and are
therefore more engaged in open trade. A second reason would
be that ethnic diverse societies may also have diverse con-
sumption preferences resulting in higher trade openness. Fur-
thermore, Parrotta, Pozzoli, Pytlikova, and Sala (2013) have
shown that firm-level ethnic diversity increases the probability
to export using a large sample of Danish firms.
(h) Fertility
Recent studies provide evidence that a large proportion of
variation in cross-country fertility and female labor force par-
ticipation can be explained by cultural norms and beliefs. 18
Fernandez and Fogli (2009) showed that, even when control-
ling for various characteristics of women (age, education,
wealth, etc.), there remains a statistically significant effect of
culture on women’s work and fertility choices. 19 Nevertheless,
fertility rate choices in ethnic diverse societies may also bear
an important political dimension. More recently, Janus
(2013) argued in favor of political causes in ethnic diverse soci-
eties in explaining cross-country net fertility rate differences.
Based on existing work, the author elaborate that in the
280 WORLD DEVELOPMENT

present of weak political institutions fertility may be a strategic premium (Market Distortions). Furthermore, the log of
choice for ethnic groups engaged in redistributive conflict. (1 + average schooling years in the total population over 25)
Indeed, competing ethnic groups may advocate higher fertility is used as a proxy for human capital accumulation (Human
rates in achieving higher voting power or increasing their com- Capital) instead of the percentage of secondary and primary
bat strength in situations of ethnic conflict. The empirical school enrollment. In addition, a full specification for the
analysis in Janus (2013) supports this theoretical view that eth- transmission channel Political Instability is included, which
nic diverse countries on average tend to have higher fertility is measured as the number of revolutions and coups per
rates, especially when political institutions are weak. year 24 and a transmission channel for the net fertility rate
(Fertility). The vector for the transmission channels therefore
consists of the following eight variables TC it ¼
0
4. DATA AND ESTIMATION METHODOLOGY ðINV ; CW ; HC; GOV ; INS; DIST ; OPEN ; FERT Þit , where
INV ; CW ; HC; GOV ; INS; DIST ; OPEN ; and FERT refers
The data set used for the empirical study is an updated ver- to the transmission channels Investment, Civil War, Human
sion of the well-known Barro–Lee data set and consists of 100 Capital, Government Consumption, Political Instability, Market
countries with a total sum of 651 observations for the period Distortions, Trade Openness and the net Fertility rate. In every
1960–99. Because of the extended seemingly unrelated regres- equation, the initial value of the log of real GDP per capita is
sion (SUR) model, the updated data set will deliver more included to control how poor the countries are. Then a
observations and hence will improve the efficiency of the measure of ethnic diversity and polarization is included
parameter estimates. Table 7 gives a brief overview of the separately in each structural equation to assess the direct
countries included and their corresponding values for ethnic and indirect effects on growth. For completeness, the base
diversity and polarization taken from Garcia-Montalvo and specification used for each transmission channel with the
Reynal-Querol (2005b). Additional information regarding underlying source is given in the following: 25
the definitions and sources of the variables used are given in The transmission channel for Investment is similar to the
Table 14 in the Appendix. specification in Garcia-Montalvo and Reynal-Querol
The empirical analysis consists of the following Barro-type (2005a), where the variables for revolutions, coups, and polit-
growth equation 20: ical assassinations per year are replaced by a measure for
Political Instability introduced above:
GROWTH it ¼ aGR þ bGR GR
1 LNGDPC it þ b2 LNGDPCSQit
0 INV
þ TC0it cGR þ XGR GR
þ /GR GR INV it ¼ aINV þ bINV
1 LNGDPC it þ TCit c þ XINV
it h
INV
it h ETH ETH i þ uit ð5Þ
where GROWTH it ðGRÞ is the growth rate of real per capita þ /INV INV
ETH ETH i þ uit ; ð6Þ
 0
GDP of country i from 1960 to 1999 averaged in five-year peri- INV INV INV INV INV INV INV
where c ¼ 0; cCW ; cHC ; cGOV ; cINS ; cDIST ; cOPEN ; 0 . No-
ods (t = 1960–64, 1965–69, . . ., 1995–1999), LNGDPC is the
log of real GDP per capita at the beginning of every 5 year tice, that the form of cINV indicates that the investment channel
period and LNGDPCSQ is LNGDPC squared. 21 The vector does not incorporate the variable FERT as explanatory vari-
TC is a set of values for the transmission channels. These able. 26 To incorporate the potential effects of civil wars on
are, for example, the share of real investment 22 to GDP growth, the linear probability model in Garcia-Montalvo
(Investment), a measure for the political situation in countries and Reynal-Querol (2005a) for the incidence of civil wars is
proposed by the Barro–Lee data set (Instability), and the ratio used:
of real government consumption to real GDP (Government). 0 CW
CW it ¼ aCW þ bCW
1 LNGDPC it þ TCit c þ XCW
it h
CW
ETH refers to ethnic diversity and polarization, respectively,
and uGR is a country-specific error term. The parameter vector þ /CW CW
ETH ETH i þ uit ; ð7Þ
cGR refers to the coefficients of the transmission channels in the   0
growth equation. For example, the parameter cGR INV measures where cCW ¼ 0; 0; cCW CW CW
HC ; 0; cINS ; 0; 0; cFERT .
the direct effect of the investment ratio on the growth rate of The transmission channel for Human Capital is specified as
real GDP per capita. The row vector XGR incorporates other follows:
control variables that do not belong to the transmission chan-
nels. These are, e.g., regional dummies for Sub-Sahara Africa, HC it ¼ aHC þ bHC 0 HC
þ XHC HC
1 LNGDPC it þ TCit c it h
East Asia, and Latin America countries, the number of polit-
ical assassinations per million population per year, the per- þ /HC HC
ETH ETH i þ uit ; ð8Þ
centage of secondary and primary school enrollment, as well   0
as the absolute deviation of the PPP value for investment where cHC ¼ 0; 0; 0; cHC HC HC
GOV ; 0; 0; cOPEN ; cFERT . The trans-
goods relative to the United States (PPDEV). The latter is in- mission channel for Government Consumption is specified
tended to capture market distortions for capital goods to some according to Tavares and Wacziarg (2001), Annett (2001),
extent and therefore the consequences on growth. 23 and Garcia-Montalvo and Reynal-Querol (2005a):
Garcia-Montalvo and Reynal-Querol (2005a) add to this
basic growth regression the incidence of civil wars (Civil GOV it ¼ aGOV þ bGOV
1 LNGDPC it þ TC0it cGOV þ XGOV
it hGOV
War), ethnic diversity, and polarization. Note, that they only
þ /GOV GOV
ETH ETH i þ uit ; ð9Þ
incorporate the variables for Civil War, Investment, and Gov-
ernment as fully specified transmission channels into their  0
where cGOV ¼ 0; cGOV GOV GOV GOV
CW ; 0; 0; cINS ; cDIST ; cOPEN ; 0 . The
SUR system. Based on their empirical findings two more empirical specification for Political Instability comes from
transmission channels proposed by Tavares and Wacziarg Tavares and Wacziarg (2001):
(2001) are included in the system. These are the share of im-
ports and exports to GDP (Trade Openness) to measure the ef- INS it ¼ aINS þ bINS 0 INS
þ XINS INS
1 LNGDPC it þ TCit c it h
fect of trade intensity on growth and a measure of
government-induced distortions captured by the black market þ /INS INS
ETH ETH i þ uit ; ð10Þ
 HOW ETHNIC DIVERSITY AFFECTS ECONOMIC GROWTH
 0
where cINS ¼ 0; 0; 0; cINS
GOV ; 0; 0; 0; 0 . In addition, the
contribution in Tavares and Wacziarg (2001) also delivers an
empirical specification for the transmission channel Market
Distortions:
DIST it ¼ aDIST þ bDIST
1 LNGDPC it þ TC0it cDIST
þ XDIST
it hDIST þ /DIST DIST
ETH ETH i þ uit ; ð11Þ
 
0
where cDIST ¼ 0; 0; cDIST DIST DIST DIST
HC ; cGOV ; cINS ; 0; cOPEN ; 0 . The
transmission equation for Trade Openness incorporates mainly
gravity variables, for example, country land area, landlocked
country dummy, island dummy, and oil exporter dummy. 27
OPEN it ¼ aOPEN þ bOPEN
1 LNGDPC it þ TC0it cOPEN
þ XOPEN
it hOPEN þ /OPEN OPEN
ETH ETH i þ uit ; ð12Þ
  0
where cOPEN ¼ 0; 0; cOPEN OPEN OPEN OPEN
HC ; cGOV ; cINS ; cDIST ; 0; 0 . Fi-
nally, the specification for the transmission channel Fertility
comes from Barro (1991) with important determinants such
as per capita income and schooling level:
FERT it ¼ aFERT þ bFERT
1 LNGDPC it þ TC0it cFERT
þ XFERT hFERT þ /FERT FERT
ETH ETH i þ uit ; ð13Þ
it
 0
where cFERT ¼ 0; 0; cFERT FERT FERT
HC ; cGOV ; cINS ; 0; 0; 0 .
The econometric methodology used to highlight the trans-
mission effects of ethnic diversity and polarization on growth
is based on panel data and SUR estimation procedures. The
model consists of the nine structural equations discussed
above, that is, one cross-country growth equation and eight
structural transmission equations, each describing one of the
channel variables. Each of the M ¼ 9 relationships can be for-
mulated for each of T ¼ 8 time periods. The parameters of
interest are the coefficients that describe the effect of a mar-
ginal change in the ethnic diversity or polarization index,
where ethnic diversity or polarization affects growth through
the specified channel equations. To estimate the indirect effect
of ethnic diversity or polarization through the transmission
channels under consideration, the coefficient for ETHFRAC
or ETHPOL in each transmission channel is multiplied by
the coefficient of each channel in the growth equation. The
standard deviation for this nonlinear combination of estima-
tors is then based on the delta method. The SUR estimator
used is different from ordinary least squares, particularly if
each structural equation consists of a different set of exoge-
nous variables. 28 All equations are linked by their distur-
bances. This implies that for a particular country and time,
all structural equations regarding growth, investment, educa-
tion, etc. are linked by their disturbances. The intuition behind
SUR estimation is that the disturbances in each structural
equation include factors that are common to all other struc-
tural variables in a particular country: that is, worldwide
aggregate demand shocks, wars, the general health status of
the economy 29, etc. are factors that affect every structural
equation. Although the SUR estimator is superior to ordinary
least squares in efficiency, it neglects the possibility of simulta-
neous equations bias arising when some of the endogenous
variables appearing on the right-hand side of each structural
equation are correlated with the disturbance term. It therefore
seems reasonable to instrument the possible endogenous vari-
ables using adequate instrumental variables (IV) and to esti-
mate the full system using Three Stage Least Squares
(TSLS), which is an IV-GLS estimator. Despite the efficiency
and consistency gains that result from using TSLS, any
281
specification error in each structural equation of the model
will be transmitted throughout the system. As mentioned by
Garcia-Montalvo and Reynal-Querol (2005a), the SUR
estimator has the advantage that it is potentially less sensitive
to specification errors than the TSLS estimator.
5. ETHNIC DIVERSITY, CHANNEL EFFECTS, AND
ECONOMIC GROWTH
Table 8 in the Appendix shows the system estimate of the
base specification including the measure for ethnic diversity.
As can be seen, the estimated coefficients for each channel
have the expected signs. For example, a higher investment ra-
tio and schooling level is positively correlated with real per ca-
pita income growth, whereas higher government consumption,
political instability, and the net fertility rate show a negative
association with economic growth. The coefficients on log ini-
tial income and its squared value imply that the convergence
rate is the highest for countries with a per capita income level
of about $324. Africa’s average initial per capita income aver-
aged from 1960 to 1999 is $2,002. Therefore, Sub-Saharan
African countries on average have already experienced the
maximum catch-up effect. Notice that the effect of Civil War
is not statistically significant at the 10% level due to possible
collinearity problems in the proxy for Political Instability. 30
The empirical findings for the channel equations are in line
with existing results in the economics and political science lit-
erature. For example, the results for the Investment channel
suggest that countries with higher log initial income and
schooling levels also will have higher investment ratios. On
the other hand, factors like civil wars and market distortions
on capital goods are negatively associated with physical capi-
tal accumulation. Therefore, as stated above, countries with
an uncertain economic environment also have on average low-
er investment ratios and hence real per capita income growth.
In addition, trade openness is accompanied with higher invest-
ment rates. This may be due to the fact that countries inte-
grated into the world market, may also have better
international financial institutions, hence fostering domestic
investments through higher inflow of foreign direct invest-
ments. 31 Proceeding in the same way for the Fertility channel,
countries with higher schooling levels also have lower net fer-
tility rates and therefore, indirectly, higher real per capita in-
come growth. As predicted by cultural theories of fertility
rate differences, ethnically diverse countries seem to have on
average higher net fertility rates. One reason may be competi-
tion among the various ethnic groups resulting in an increased
net fertility rate in ethnic diverse countries. Furthermore, the
results for the Openness channel are in line with the findings
in Alesina and Wacziarg (1998), where country size is very sig-
nificantly related to the degree of trade openness. The negative
impact of the democracy dummy is in line with protectionist
policies in democratic regimes. For example, groups that ben-
efit from restricting open trade tend to mediate their concerns
through political engagement, lobbying and public cam-
paigns. 32
Notice, that the R2 in some specifications is considerably
lower than in comparable empirical studies. One reason is that
the current study is based on a panel data set opposed to cross-
section data. Therefore, in addition to the cross-section varia-
tion there exists also a within country variation. Furthermore,
because some of the dependent variables are measured either
as rates (e.g., the equations for Growth, Investment, and Gov-
ernment Consumption) or levels (e.g., the equations for Human
Capital and the Fertility rate), it should be clear, that because
282 WORLD DEVELOPMENT
of the higher variation in growth rates the R2 in these specifi-
cations is considerably lower than in specifications where the
dependent variable is measured in levels. 33
The empirical results reveal that ethnic diversity has a neg-
ative direct association with real per capita income growth.
This finding confirms the empirical results in Easterly and
Levine (1997), Alesina et al. (2003), Garcia-Montalvo and
Reynal-Querol (2005a), among others. The results of the growth
equation state that, ceteris paribus, going from perfect ethnic
homogeneity (an index of 0) to perfect ethnic diversity (an in-
dex of 1), a fully diverse country (e.g., Tanzania) has a 1.96%
lower real per capita income growth per year on average than
a fully homogeneous society (e.g., South Korea). The follow-
ing Table 1 shows the indirect effects as well as the total effect
going from full ethnic homogeneity to full ethnic diversity, for
example, by redrawing the borders. The estimates suggest that
a change in ethnic diversity from 0 to 1 is accompanied by a
0.10%, but statistically insignificant, reduction in real per capi-
ta income growth through the transmission channel Fertility
alone. The total indirect effect of ethnic diversity through the
transmission channels on growth is small (0.03%) compared
to the total direct effect of 1.96%. The remaining transmis-
sion channels are not important for the indirect effects of eth-
nic diversity on economic growth. In the case of the Investment
channel, this does not mean that physical capital accumulation
is not important for growth, but rather that this channel is not
significantly affected by ethnic diversity. Alternatively, a one
standard deviation change of ETHFRAC (0.28), would bring
about a 0.55% decrease in real per capita income growth per
year. The estimated effect is statistically significant at the 1%
level and economically notable. The empirical results state
that, all else equal, if Cameroon (0.82) had the same ethnic
diversity as Botswana 34 (0.48) its real per capita income
growth would be 0.68% higher. Taken literally, a 0.68% higher
real per capita income growth would result in about 7% higher
real per capita income level after ten years. Nevertheless, the
estimates suggest potential second-order indirect effects of eth-
nic diversity on economic growth through the Human Capital
channel. First, ethnic diverse countries on average have higher
net fertility rates. Second, the latter is negatively related with
education. Therefore, ethnic diversity, ceteris paribus, medi-
ates its negative impact on economic growth also through
higher net fertility rates in the channel for Human Capital.
Table 1. How ETHFRAC affects growth
Channel Effect of ETHFRAC on the
channel
Investment 1.6006 (1.1702)
Civil war 0.0633 (0.0619)
Human capital 0.0499 (0.0468)
Government 1.6731 (1.4557)
Instability 0.0097 (0.0620)
Distortions 0.0526 (0.0560)
Openness 14.2083** (7.1136)
Fertility 0.3330** (0.1555)
Total indirect effect – –
Total effect – –
N – – –
Notes: The second column shows the effect of ethnic diversity on the transmission channels. The third column corresponds to the channel effects on
growth. The fourth column presents the product of the two coefficients. Robust standard errors between parenthesis. Robust standard errors for the
products of the two coefficients in column four based on the delta method.
*
Significant at the 10% level.
**
Significant at the 5% level.
***
Significant at the 1% level.
Table 9 shows the empirical results including the measure
for ethnic polarization in the base specification. The results
are robust to the inclusion of ETHPOL, but the statistical sig-
nificance of the transmission coefficients in the growth equa-
tion increases. Ethnic polarization has virtually no direct
association with real per capita income growth. In contrast,
there seem to be substantial indirect effects through the chan-
nels Investment, Civil War, Government Consumption, Political
Instability, Trade Openness, and the net Fertility rate. Inter-
preting the channel equation Civil War, note first that the lin-
ear probability model is somewhat inappropriate because it
does not restrict the probabilities between zero and one. Espe-
cially the R2 cannot be interpreted in the usual form when the
dependent variable is binary and the explanatory variables are
measured continuously (as is the case here). Apart from this
shortcoming, a huge advantage of the linear probability model
is its direct interpretation of the estimated coefficients. Taken
literally, going from full to zero ethnic polarization decreases
the incidence of a civil war by about 15%. The remaining coef-
ficients are also of the expected signs: a higher population in-
creases the probability of a civil war because of the huge pool
available to rebel groups in such situations to recruit insur-
gents. Furthermore, a country with a high population is more
difficult to govern and control than a small one. In addition,
the coefficient on the democracy variable is positive and statis-
tical significant at the 1% level. This result is not surprising if
one takes into account that the dummy variable used for
democracy also incorporates democracies classified as inter-
mediate, which seems to increase the risk of ethnic conflicts.
For example, ethnically divided democracies in which ethnic
parties that fail to reach a given threshold in general elections
are likely to fall into violent ethnic conflicts outside of party
politics. 35 The following Table 2 shows the quantitative mag-
nitudes going from zero ethnic polarization (an index of 0) to
full ethnic polarization (an index of 1). The total effect of
ETHPOL on growth is 1.38% and statistically significant
at the 1% level. The estimated magnitude is somewhat smaller
to that of ETHFRAC, but there are qualitative differences.
Ethnic polarization primarily affects growth indirectly through
the channels Investment and Fertility. There is also weak evi-
dence that ETHPOL negatively affects growth through the
transmission channel Government Consumption. The total indi-
rect effect of ETHPOL is about 0.59% and significant at the
Effect of the channel on growth Effect of ETHFRAC on
growth
0.0900*** (0.0144) 0.1440 (0.1078)
0.3885 (0.3341) 0.0246 (0.0320)
0.5848* (0.3405) 0.0292 (0.0322)
0.0346*** (0.0131) 0.0579 (0.0549)
0.6004* (0.3164) 0.0058 (0.0374)
1.9340*** (0.3378) 0.1018 (0.1098)
0.0032 (0.0026) 0.0459 (0.0435)
0.3027** (0.1221) 0.1008 (0.0622)
– 0.0288 (0.1922)
– 1.9883*** (0.5161)
651
 HOW ETHNIC DIVERSITY AFFECTS ECONOMIC GROWTH 283

Table 2. How ETHPOL affects growth

Channel Effect of ETHPOL on the Effect of the channel on Effect of ETHPOL on growth
channel growth
Investment 3.7221*** (1.2221) 0.0891*** (0.0146) 0.3317*** (0.1217)
Civil war 0.1492** (0.0586) 0.4566 (0.3391) 0.0681 (0.0573)
Human capital 0.0022 (0.0487) 0.6772** (0.3449) 0.0015 (0.0330)
Government 2.6297* (1.4979) 0.0355*** (0.0133) 0.0935 (0.0637)
Instability 0.0459 (0.0635) 0.6263* (0.3195) 0.0287 (0.0424)
Distortions 0.0444 (0.0575) 1.9260*** (0.3414) 0.0854 (0.1118)
Openness 15.7303** (6.7847) 0.0056** (0.0026) 0.0881 (0.0562)
Fertility 0.7684*** (0.1553) 0.3093** (0.1240) 0.2377** (0.1067)
Total indirect effect – – – 0.5877*** (0.2224)
Total effect – – – 1.3751*** (0.5223)
N – – – 651
Notes: The second column shows the effect of ethnic polarization on the transmission channels. The third column corresponds to the channel effects on
growth. The fourth column presents the product of the two coefficients. Robust standard errors between parenthesis. Robust standard errors for the
products of the two coefficients in column four based on the delta method.
*
Significant at the 10% level.
**
Significant at the 5% level.
***
Significant at the 1% level.

1% level, whereas the direct effect of 0.79% is not statistically
significant. Alternatively, a one standard deviation increase of
ETHPOL (0.24) is accompanied by about 0.33% decrease of
real per capita income growth per year. The estimates suggest
that, ceteris paribus, if South Africa (0.72) had the same
polarization index as South Korea (0.03), its real per capita
income growth per year would be about 1% higher. This
growth difference is non-negligible and would result in a
10% higher real per capita income level after 10 years. The
overall negative indirect impact of ETHPOL on real per capita
income growth is not due to the inclusion of the two channels
Investment and Fertility. The following Table 3 restricts the
negative indirect effect of ETHPOL to the channels Civil
War, Human Capital, Government Consumption and Instabil-
ity. The estimates reveal that, although the individual indirect
effects through these channels seem to be statistically insignif-
icant, its total indirect effect of 0.19% is statistically signifi-
cant at the 10% level indicating that the primary results are
not driven by the inclusion of the channels Investment and Fer-
tility.
In summary, the comparison of both ethnic measures leads
to qualitatively different conclusions. Ethnic diversity primar-
ily affects economic growth directly, whereas ethnic polariza-
tion has substantial indirect effects. Therefore, as also
emphasized by Garcia-Montalvo and Reynal-Querol (2005a),
there seems to be no empirical evidence that ethnic diversity
affects growth primarily through the specified transmission
channels. The estimates suggest, that ethnic diverse countries
are not prone to worse public good performance than
ethnically homogeneous countries. Furthermore, the analysis
suggests that ethnic polarization has non-negligible effects on
the channels Investment, Civil War, Government Consumption,
Openness and Fertility.
6. SENSITIVITY ANALYSIS
The benchmark results regarding ethnic diversity and polar-
ization may be sensitive to time and region effects, to the elim-
ination of per capita income, to the specification of the
benchmark model and to the sample coverage. In this section,
the robustness of the results to several modifications of the
benchmark model is examined. For ease of comparison, only
the results in the growth equation and the coefficient for ETH-
FRAC or ETHPOL in each channel equation as shown in
Tables 10 and 11 are reported. 36 The second column
corresponds to the benchmark results shown above and is
given for comparison purposes. The third column shows the
estimated effect of the ethnic variable, when the intercepts of
each structural equation are allowed to vary over time. The

Table 3. How ETHPOL affects growth

Channel Effect of ETHPOL on the Effect of the channel on growth Effect of ETHPOL on growth
channel
Civil war 0.1492** (0.0586) 0.4566 (0.3391) 0.0681 (0.0573)
Human capital 0.0022 (0.0487) 0.6772** (0.3449) 0.0015 (0.0330)
Government 2.6297* (1.4979) 0.0355*** (0.0133) 0.0935 (0.0637)
Instability 0.0459 (0.0635) 0.6263* (0.3195) 0.0287 (0.0424)
Total indirect effect – – – 0.1918* (0.0979)
N – – – 651
Notes: The second column shows the effect of ethnic polarization on the transmission channels. The third column corresponds to the channel effects on
growth. The fourth column presents the product of the two coefficients. Robust standard errors between parenthesis. Robust standard errors for the
products of the two coefficients in column four based on the delta method.
*
Significant at the 10% level.
**
Significant at the 5% level.
***
Significant at the 1% level.
284 WORLD DEVELOPMENT
fourth column allows for regional dummies in each channel
equation, whereas column five allows for time and region ef-
fects in each equation. The direct effect of ETHFRAC is very
robust to these modifications. The magnitude of ETHFRAC
on real per capita income growth virtually remains unchanged.
On the other hand, ETHPOL retains its insignificant direct ef-
fect on real per capita income growth but shows a strong asso-
ciation with the channel variables.
The benchmark specification includes the log of per capita
income in each channel equation. Since ethnic diversity and
initial per capita income are negatively correlated (0.51), it
may be hard to disentangle the indirect effects of ETHFRAC
from the income effects. La Porta et al. (1999) argued that if
ethnic diversity undermines government performance, and
poor government performance reduces per capita income, then
including per capita income together with variables of ethnic-
ity in the same regression would reduce the estimated effect of
ethnic diversity on the dependent variable. Therefore, for illus-
trative purposes, initial per capita income is excluded from
every channel equation. The above indications would lead
one to expect an overall rise in the statistical significance of
ethnic diversity in each channel equation. But, the effect of
ETHFRAC through the specified channels does not change
substantially, indicating that concerns of collinearity are not
justified. However, it remains interesting to test whether the
exclusion of log initial income results in omitted variable bias.
If the results regarding ETHFRAC suffer from collinearity
problems with initial income, then it will indeed be difficult
to disentangle the income from the ethnic effect. Nevertheless,
the point estimates of ETHFRAC are still unbiased, also in the
case of collinearity, but the standard error of the estimator
may be too large, which translates into less accurate hypothe-
ses. Because of the lower negative correlation between ETH-
FRAC and initial income, collinearity should not be a
serious problem causing the results excluding initial income
to suffer from omitted variable bias. Consider, for example,
the Investment channel: because of the negative correlation be-
tween ETHFRAC and initial income and the positive effect of
initial income on investment, excluding initial income from the
Investment channel should result in a downward bias of ETH-
FRAC. Indeed, the estimate for ETHFRAC excluding initial
income decreases from 1.60 to 2.66. The remaining chan-
nels can be interpreted in a similar way. Furthermore, if these
sensitivity results are mainly driven by collinearity problems
between ETHFRAC and initial income, then one would expect
an overall increase in the significance level of ETHFRAC in
the Fertility channel, but this is not the case. Rather, excluding
initial income from the base specification leads to a statistically
insignificant effect of ETHFRAC on the Fertility channel.
Excluding initial income in the specification for ETHPOL re-
sults in insignificant channel effects for Government Consump-
tion. But, as mentioned above, initial income is a strong
explanatory variable in each channel variable, and excluding
it would result in omitted variable bias. Indeed, the estimated
effect of ETHPOL in the Investment, Human Capital and Fer-
tility channel changes substantially once initial income is
excluded. 37
The analysis for each of the ethnic measures is also restricted
to geographic coverage by excluding Sub-Saharan African,
Latin American, East-Asian, and OECD countries subse-
quently to highlight whether the effect of ethnicity depends
on the regional pattern. Restricting the analysis to non-Sub-
Saharan African countries, the overall magnitude of ETH-
FRAC in the growth equation decreases substantially. Exclud-
ing countries from the benchmark model which are highly
diverse reveals an interesting pattern: ETHFRAC seems to
have non-negligible indirect effects in the remaining less heter-
ogeneous countries through the channels Investment, Civil
War, Government Consumption, Openness, and Fertility. One
implication of this analysis may be the different effects of
ETHFRAC conditional on the level of ethnic diversity in the
particular country. 38 Furthermore, the channel effects of
ETHPOL remain important, also excluding Sub-Saharan
African countries.
An important consideration is whether the results are driven
primarily by the inclusion of developed countries, for example,
OECD members. Concentrating on less-developed countries is
of great interest from the standpoint of policy because these
are precisely the countries in which the economic effects of eth-
nicity are most debated. Restricting the sample to non-OECD
members therefore leads to interesting results: Ethnic diversity
in developing countries seems to have virtually no indirect ef-
fect on economic growth through the channels under consider-
ation. The indirect effect of ETHFRAC on the channels is not
statistically significant, whereas the direct effect slightly in-
creases. In addition, the results regarding ETHPOL state that
polarized societies still suffer from a higher incidence of civil
wars, lower schooling levels, higher government consumption,
and higher market distortions. The results lead to the conclu-
sion that not ethnic diversity but rather ethnic polarization is a
strong predictive measure for the indirect effects of ethnicity in
both developed and developing countries. This result is some-
what puzzling and contradicts the findings in Easterly and
Levine (1997) and Alesina et al. (2003) but is in line with the
findings reported by Garcia-Montalvo and Reynal-Querol
(2005a).
Political instability is difficult to define and measure in a way
that can be used for econometric work. For example, Alesina
and Perotti (1996) identifies factors of political instability that
capture the idea of adverse effects on property rights and
abrupt changes in executive power. Because of these difficulties
in constructing an adequate measure of political instability,
this measure is excluded from the benchmark model in order
to determine how the results change. The benchmark results
are not affected by the exclusion of the Political Instability
channel for both specifications, aside from the fact that the
channel effect for Civil War in the growth equation now be-
comes statistically significant. This result may be due to the
politically unstable environment that generally accompanies
civil wars, making it difficult to disentangle the effects when
both measures are included in the same regression.
Another interesting point is to what extent greater or lesser
variations in the measures within regions drive the results. 39
Table 6 in the Appendix shows the variation of both ethnic
measures across regions. The lowest variation of ETHFRAC
is found in Latin American and OECD countries followed
by Sub-Saharan African and East Asian countries. Neverthe-
less, Sub-Saharan African countries are among the highest
diverse countries in the world, while OECD members seem
to be among the least diverse countries in the world. Varia-
tional differences across countries are more pronounced
focussing on the measures for ETHPOL. While Sub-Saharan
African countries still have high values of ethnic polariza-
tion, variational differences in ETHPOL are more prevalent
among OECD member countries. Therefore, an interesting
robustness test would be to discover the effect of ethnic
diversity and polarization on socioeconomic performance
within regions. For example, since OECD members have
good installed political institutions it would be interesting
to detect whether the effect of ethnicity vary across regions.
Such an analysis would be of great political interest in miti-
gating potential detrimental effects of ethnicity on the course
 HOW ETHNIC DIVERSITY AFFECTS ECONOMIC GROWTH
of economic development. Table 12 shows the results for the
measure ETHFRAC. Restricting the sample on Sub-Saharan
African countries results on a strong negative impact of eth-
nic diversity on real per capita income growth. The magni-
tude of ETHFRAC in this specification outpaces that in
the base model. Interestingly, ethnic diversity has a positive,
but statistically insignificant effect, on real per capita income
growth in OECD countries. A possible explanation would be
that the quality of installed political institutions and the ab-
sence of ethnically based politics in highly developed OECD
members make potential detrimental effects of ethnic diver-
sity on socioeconomic performance of minor interest. Ana-
lyzing the effect of ETHFRAC on the transmission
channels indicate substantial nonlinearities. For example,
the negative impact of ETHFRAC on Government Consump-
tion is more pronounced in Sub-Saharan African than in
OECD countries, while countries with intermediate values
of ethnic diversity (e.g., Latin America and East Asia coun-
tries) have higher government consumption shares. Further-
more, while ethnic diversity seems to be unrelated with the
accumulation of physical capital in Sub-Saharan African
countries, it shows a negative effect in OECD countries indi-
cating once again the impact of ethnic diversity conditional
on the level of ethnic diversity. 40 Table 13 shows the regio-
nal results including the measure for ETHPOL. The direct
impact of ethnic polarization on real per capita income
growth retains its insignificant effect in all but one specifica-
tion. This result confirms the hypothesis that ethnic polariza-
tion do not affect real per capita income growth directly.
Interestingly, high ethnic polarization in Sub-Saharan Afri-
can countries is accompanied with higher investment rates,
lower government consumption and fertility rates. This
may be due to the fact that higher polarization measures
in these countries are inevitably associated with lower mea-
sures of ethnic diversity, 41 which in turn improves the mac-
roeconomic performance in these countries. Focusing on
OECD countries, high ethnic polarization decreases physical
capital investment and government consumption, while
increasing the fertility rate.
Overall, the sensitivity results underscore the general pattern
of ethnic diversity showing a direct negative association with
economic growth, whereas ethnic polarization seems to have
strong indirect effects. Furthermore, differential impacts of
ethnic diversity and polarization on economic growth and
the transmission channels across regions reveal possible non-
linearities. An interesting result is that the direct impact of eth-
nic diversity on economic growth is non-existent among highly
developed OECD countries indicating the possibility to miti-
NOTES
1. Nevertheless, Easterly and Levine (1997) used four additional
measures of ethnic diversity to check whether their results depend on
the particular choice of the ethnic measure. In contrast to the Soviet
ethnic measure from Atlas Narodov Mira, three of the other four
measures showed a strong direct link to growth, even when controlling
for the entire set of available explanatory variables, suggesting that
ethnic diversity may have a direct impact on growth, independent of
public policy decisions.
2. Easterly and Levine (1997) used a measure of ethnic diversity to
investigate this issue, despite asserting the potential consequences of ethnic
polarization on policies that affect economic growth.
285
gate the detrimental effects of ethnic diversity through good
political institutions.
7. CONCLUSIONS
This paper investigates the economic effects of ethnic diver-
sity and polarization and explores the relative importance of
these two concepts through an extended SUR model. It con-
tributes to the existing literature in several ways. First, it
provides an updated analysis for 100 developed and develop-
ing countries. Second, this paper deals with an extended
SUR model in which every transmission channel appearing
in the growth equation is specified on the grounds of existing
empirical work in the economic and political science litera-
ture. The SUR specification takes into account further equa-
tions that seem to be important transmission channels for
the effects of ethnicity on economic growth-namely, school-
ing, political instability, market distortions, trade openness,
and the fertility rate-but that have not been analyzed so
far in the existing literature. Furthermore, the full system is
jointly estimated using the SUR estimator and presents de-
tailed evidence of direct and indirect effects of ethnicity on
economic growth with the corresponding significance levels.
The results suggest, for instance, that ethnic diversity pro-
duces positive indirect effects through international trade
and leads to higher net fertility rates, which may be benefi-
cial in advanced economies through higher international
trade integration and the possibility to rejuvenate the popu-
lation. Third, this research undertakes an elaborative attempt
in trying to integrate the two so far separate literatures on
ethnic polarization as a source of ethnic conflict and ethnic
diversity as a factor for the macroeconomic performance of
countries. The distinction between ethnic diversity and polar-
ization seems to be empirically justified since it leads to qual-
itatively opposite conclusions: ethnic diversity primarily
affects economic growth directly, whereas ethnic polarization
seems to have substantial indirect effects. The detailed anal-
ysis on direct and indirect effects of ethnicity on socioeco-
nomic factors may support policy makers to undertake
concrete political steps in mitigating the potential detrimental
effects of ethnicity on economic growth. For example, as
shown in the sensitivity analysis, the impact of ethnicity on
economic growth is substantially lower in OECD than in
Sub-Saharan African countries. This result shows the impor-
tance of political institutions in shaping ethnic identity and
hence its role for socioeconomic performance. Further re-
search should therefore investigate this issue in more detail.
3. Campos and Kuzeyev (2007) analyzed for the first time the dynamic
nature of ethnic, linguistic, and religious diversity on economic growth for
26 former communist countries. The authors conclude, that exogenous
ethnic diversity seems to be unrelated with real per capita income growth,
whereas their new (time-variant) ethnic diversity measure shows a
statistically significant negative association with economic growth.
4. The delta method makes it possible to highlight the statistical
significance of the indirect effects of ethnic diversity and polarization
through the proposed transmission channels on growth, providing the
asymptotic distribution of the nonlinear combinations of the estimated
parameter values.
286 WORLD DEVELOPMENT
5. For a thorough discussion of these axioms the interested reader is
referred to Esteban and Ray (1994).
6. For a detailed discussion of the relationship of rent-seeking behavior,
the allocation of talented people, and the implications for technological
progress and economic growth, see Murphy, Shleifer, and Vishny (1991). I
thank an anonymous referee for suggesting this point.
7. See, also the findings in Barro (1991), Alesina and Perotti (1996), and
Tavares and Wacziarg (2001).
8. Collier (2001) argued that a large diaspora in Europe and America is
able to finance rebel groups in their home countries out of solidarity to
their people. Such outcomes can lead to long-term ethnic conflicts in the
particular country, producing adverse effects on a wide range of social
activities.
9. Vanhanen (1999) gave an overarching explanation of why so many
conflicts take place along ethnic lines: He argues that a society’s evolved
predisposition to ethnic nepotism will lead people to seek influence
allowing them to control the political situation for their own benefit. In
this sense, the author noted that “the members of an ethnic group tend to
favor their group members over non-members because they are more
related to their group than to the remainder of the population.” Such a
tendency toward ethnic nepotism eventually ends in ethnic struggle when
the members of an ethnic group tend to support one another in situations
of conflict.
10. Collier (2001) argued that outbreaks of secessionist movements in
polarized societies increase when natural resources are discovered.
11. Kanbur et al. (2011).
12. See Ogbu and Simons (1998) for an overview.
13. Involuntary minorities are people who have been conquered,
colonized, or enslaved. Unlike immigrant groups, involuntary minorities
are seen to be part of the ruling majority group against their will.
14. The main idea is that higher government consumption is inevitably
related with distorting tax policies and hence reduced per capita income
growth.
15. See, La Porta et al. (1999), Collier (2001), and Alesina and
Zhuravskaya (2011).
16. Alesina and La Ferrara (2005).
17. See Rauch (2001) for an overview.
18. See Fernandez and Fogli (2006), who find that variation in fertility
rates among US immigrants from different ethnic backgrounds can be
explained by the total fertility rate of the women’s country of origin as a
proxy for culture. See also Fernandez (2007) for evidence on female labor
force participation.
19. The measures of ethnic diversity and polarization capture, to some
extent, the degree of cultural diversity within a country.
20. The growth equation is very similar to that of Barro (1991). This
specification is the widely used form for cross-country growth analysis.
Furthermore, it follows the argument of Garcia-Montalvo and Reynal-
Querol (2005a) in choosing the most common specifications for the growth
equation and for each of the transmission channels to avoid “variables
fishing.”
21. The inclusion of the square of log real GDP per capita in the growth
equation incorporates the possibility that the rate of convergence depends
on the level of GDP per capita. A positive coefficient on LNGDPC and a
negative coefficient on LNGDPCSQ indicates that the convergence rate is
a concave function of real GDP per capita. Thus, initial real per capita
 GR 
b
income that maximizes the catch-up effect lies around exp 2b1 GR .
2
22. The investment share refers to the total economy investment share of
real GDP.
23. See Barro (1991).
24. This definition of political instability is adopted from the Barro–Lee
data set. In addition, the variable PPDEV in the growth equation is
excluded because it appears to be statistically insignificant. The growth
equation therefore consists (besides the intercept, regional dummies, and
income variables) only of the variables for the transmission channels.
25. I am grateful to two anonymous referees for their additional help in
specifying the transmissions channels.
26. In contrast to the study in Mauro (1995), this specification dispenses
from the inclusion of a corruption index. The reason behind this decision
is that corruption variables are only available for the 1980s onward, and in
most cases only as cross-section dimension. Nevertheless, using a
composite index of the quality of government (taken from The Quality
of Government Dataset, version 6Apr11. University of Gothenburg: The
Quality of Government Institute, http://www.qog.pol.gu.se), which is the
mean value of the International Country Risk Guide (ICRG) variables
Corruption, Law and Order and Bureaucracy Quality, does not change the
qualitative results. Overall, ETHPOL retains its significant indirect effect
on economic growth, while ETHFRAC, in most cases, is still negatively
related with economic growth. The results are available in the online
Appendix to this paper. Unfortunately, the number of observations due to
the inclusion of this single variable decreases substantially from 651 to
319. Because of serious data limitations, I have decided not to consider
corruption indices as an additional regressor.
27. The specification for the Openness equation is an extended version to
that in Tavares and Wacziarg (2001) with the exception that it dispenses
with the variable measuring the distance from major trading partners. This
variable is not available for many African countries and therefore would
lead to a reduction in sample size.
28. See Zellner (1962).
29. See Greene (2008, p. 252).
30. The argument behind the inclusion of a political instability measure
separately from a civil war channel is, that political instability captures the
notion to observe major government changes that may be detrimental to
property rights. Revolutions and coups d’etat may be seen as ways to
change the political order which in turn creates an environment of political
uncertainty discouraging potential investment decisions. On the other
hand, civil wars capture the notion of social unrest that may be unrelated
to government changes. For example, the struggle of an ethnic group
gaining regional independence. Indeed, both, civil wars and political
instability may be highly related. Section 6 will check the statistical
significance of the civil war channel by excluding Political Instability from
the base specification.
31. I thank an anonymous referee for suggesting this point.
32. See, for example, Tavares and Wacziarg (2001) and the references
therein.
 HOW ETHNIC DIVERSITY AFFECTS ECONOMIC GROWTH
33. I thank an anonymous referee for pointing out this issue.
34. Botswana is an African success story, with average growth rates in
per capita income of about 6.38% for the 1970–99 period and is, by
African standards, ethnically homogeneous. In contrast, Cameroon
experienced a low growth rate of about 0.85% during the same period.
35. It remains an interesting question whether democratic institutions
lower the incidence of violent ethnic conflicts or help to institutionalize
them in nonviolent form, in the sense that these conflicts take on political
importance (see also Vanhanen (1999) and the references therein).
36. The compounding results regarding the indirect channel effects of
ethnic diversity and polarization on economic growth are available from
the author upon request.
37. In general, the correlation matrix in Table 4 in the Appendix
indicates that all variables analyzed are moderately correlated, especially if
one focusses on both ethnic measures, indicating that multi-collinearity is
not a serious problem in the empirical study. Additionally, multi-
collinearity is not a problem for econometric inference as the coefficients
remain unbiased even if multi-collinearity is present. Finally, our ability to
estimate the coefficients with statistical significance indicates that the
variables contain enough variation for separating their influence.
REFERENCES
Alesina, A., Devleeschauwer, A., Easterly, W., Kurlat, S., & Wacziarg, R.
(2003). Fractionalization. Journal of Economic Growth, 8, 155–
194.
Alesina, A., & La Ferrara, E. (2005). Ethnic diversity and economic
performance. Journal of Economic Literature, 43, 762–800.
Alesina, A., & Perotti, R. (1996). Income distribution, political instability,
and investment. European Economic Review, 40, 1203–1228.
Alesina, A., & Spolaore, E. (1997). On the number and size of nations.
Quarterly Journal of Economics, 112(4), 1027–1056.
Alesina, A., Spolaore, E., & Wacziarg, R. (2000). Economic integration
and political disintegration. American Economic Review, 90(5),
1276–1296.
Alesina, A., & Wacziarg, R. (1998). Openness, country size and govern-
ment. Journal of Public Economics, 69(3), 305–321.
Alesina, A., & Zhuravskaya, E. (2011). Segregation and the quality of
government in a cross section of countries. American Economic Review,
101, 1872–1911.
Annett, A. (2001). Social fractionalization, political instability, and the
size of government. IMF Staff Papers, 48(3), 561–592.
Atlas Narodov Mira. (1964). Moscow: Miklukho-Maklai Ethnological
Institute at the Department of Geodesy and Cartography of the State
Geological Committee of the Soviet Union.
Banks, A. S (2011). Cross-national time-series data archive. Jerusalem,
Israel: Databanks International.
Barro, R. (1991). Economic growth in a cross section of countries.
Quarterly Journal of Economics, 106(2), 407–443.
Barro, R., & Lee, J. (1993). International comparisons of educational
attainment. NBER working paper No. 4349.
Barro, R., & Lee, J. (2010). A new data set of educational attainment in the
world, 1950–2010. NBER working paper 15902 (accompanying data
are available at www.barrolee.com).
Campos, F. C., & Kuzeyev, V. S. (2007). On the dynamics of ethnic
fractionalization. American Journal of Political Science, 51(3), 620–639.
Central Intelligence Agency (1996). The CIA World Factbook. Washing-
ton, DC.
Cerqueti, R., Coppier, R., & Piga, G. (2012). Corruption, growth and
ethnic fractionalization: A theoretical model. Journal of Economics,
106(2), 153–181.
Collier, P. (2001). Ethnic diversity – An economic analysis. Economic
Policy, 128–166.
287
38. Indeed, in a more theoretical study Cerqueti, Coppier, and Piga
(2012) have shown that the relationship between ethnic diversity,
corruption, and economic growth may be nonlinear. They argue, that
corruption is high in homogenous or very diverse countries, but low in
intermediate diverse countries. The intuition behind their findings is that,
despite its detrimental effect of ethnic diversity on growth, intermediate
diverse societies act as a balance check, limiting ethnically based
corruption. Their conclusion is that the effect of ethnic diversity on
economic growth should be nonlinear, affecting growth in a hump-shaped
fashion. I thank an anonymous referee for suggesting this point.
39. I thank an anonymous referee for suggesting this point.
40. An alternative approach investigating potential nonlinearities of
ethnic diversity on economic growth and the transmission channels
would be to include ETHFRAC squared into each equation. In addition,
as noted by an anonymous referee, whether the impact of ethnicity
depends on the development status of the country could be addressed by
the inclusion of an interaction term between ETHFRAC and the log of
GDP per capita. However, addressing this point would open a complete
new discussion on non-linearities and is therefore beyond the scope of
this paper.
41. See also Figure 1 above.
Doyle, M., & Sambanis, N. (2000). International peacebuilding: A
theoretical and quantitative analysis. The American Political Science
Review, 94(4), 779–801.
Easterly, W., & Levine, R. (1997). Africa’s growth tragedy: Policies and
ethnic divisions. The Quarterly Journal of Economics, 112(4),
1203–1250.
Easterly, W. (2001). The lost decades: Developing countries stagnation in
spite of policy reform 1980–1998. World Bank.
Encyclopedia Britannica. (2000). Chicago: Encyclopedia Britannica.
Esteban, J., & Ray, D. (1994). On the measurement of polarization.
Econometrica, 62(4), 819–851.
Fernandez, R. (2007). Women, work and culture. Journal of the European
Economic Association, 5(2–3), 305–332.
Fernandez, R., & Fogli, A. (2006). Fertility: The role of culture and family
experience. Journal of the European Economic Association, 4(2-3),
552–561.
Fernandez, R., & Fogli, A. (2009). Culture: An empirical investigation of
beliefs, work, and fertility. American Economic Journal: Macroeco-
nomics, 1(1), 146–177.
Garcia-Montalvo, J., & Reynal-Querol, M. (2005a). Ethnic diversity and
economic development. Journal of Development Economics, 76(3),
293–323.
Garcia-Montalvo, J., & Reynal-Querol, M. (2005b). Ethnic polarization,
potential conflict, and civil wars. American Economic Review, 95(3),
796–816.
Greene, W. (2008). Econometric analysis. New Jersey: Pearson Prentice
Hall.
Heston, A., Summers, R., & Aten, B. (2006). Penn World Table Version
6.2. Center for International Comparisons of Production, Income and
Prices, University of Pennsylvania.
Janus, T. (2013). The political economy of fertility. Public Choice, 155,
493–505.
Kanbur, R., Rajaram, P. K., & Varshney, A. (2011). Ethnic diversity and
ethnic strife. An interdisciplinary perspective. World Development,
39(2), 147–158.
La Porta, R., Lopez-de-Silanes, F., Shleifer, A., & Vishny, R. (1999). The
quality of government. The Journal of Law, Economics, & Organiza-
tion, 15(1), 222–279.
Mauro, P. (1995). Corruption and growth. Quarterly Journal of Econom-
ics, 110(3), 681–712.
288 WORLD DEVELOPMENT

Murphy, K. M., Shleifer, A., & Vishny, R. W. (1991). The allocation of
talent. Quarterly Journal of Economics, 106(2), 503–530.
Ogbu, J. U., & Simons, H. D. (1998). Voluntary and involuntary
minorities: A cultural-ecological theory of school performance with
some implications for education. Anthropology & Education Quarterly,
29(2), 155–188.
Parrotta, P., Pozzoli, D., Pytlikova, M., & Sala, D. (2013). Ethnic diversity
and firm’s exporting behavior. mimeo.
Rauch, J. E. (2001). Business and social networks in international trade.
Journal of Economic Literature, 39(4), 1177–1203.
Reynal-Querol, M. (2002). Ethnicity, political systems, and civil wars.
Journal of Conflict Resolution, 46(1), 29–54.
Tavares, J., & Wacziarg, R. (2001). How democracy affects growth.
European Economic Review, 45, 1341–1378.
Vanhanen, T. (1999). Domestic ethnic conflict and ethnic nepotism: A
.comparative analysis. Journal of Peace Research, 36(1), 55–73.
Zellner, A. (1962). An efficient method of estimating seemingly unrelated
regressions and tests for aggregation bias. Journal of the American
Statistical Association, 57(298), 348–368
APPENDIX A
See Tables 4–14.

Table 4. Correlation matrix for ETHFRAC, ETHPOL, growth and its determinantsa,b
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)
(1) ETHFRAC 1.0000
(2) ETHPOL 0.5775 1.0000
(0.0000)
(3) Growth 0.2311 0.1343 1.0000
(0.0000) (0.0006)
(4) Log initial income 0.4967 0.1424 0.0306 1.0000
(0.0000) (0.0003) (0.4362)
(5) Investment 0.3050 0.1434 0.3283 0.4690 1.0000
(0.0000) (0.0002) (0.0000) (0.0000)
(6) Civil war 0.1355 0.1219 0.1141 0.1686 0.1848 1.0000
(0.0005) (0.0018) (0.0036) (0.0000) (0.0000)
(7) Human capital 0.4237 0.1795 0.0872 0.7914 0.4595 0.1134 1.0000
(0.0000) (0.0000) (0.0261) (0.0000) (0.0000) (0.0038)
(8) Government 0.0767 0.1302 0.0759 0.2096 0.0677 0.0504 0.1104 1.0000
(0.0503) (0.0009) (0.0530) (0.0000) (0.0843) (0.1987) (0.0048)
(9) Instability 0.1268 0.0937 0.1209 0.2314 0.1783 0.3761 0.1661 0.0790 1.0000
(0.0012) (0.0167) (0.0020) (0.0000) (0.0000) (0.0000) (0.0000) (0.0439)
(10) Distortions 0.1396 0.0333 0.2307 0.3145 0.2088 0.1725 0.2389 0.1303 0.1654 1.0000
(0.0004) (0.3960) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0009) (0.0000)
(11) Openness 0.0505 0.1322 0.0969 0.0995 0.1753 0.1248 0.0510 0.2116 0.0760 0.0361 1.0000
(0.1980) (0.0007) (0.0134) (0.0111) (0.0000) (0.0014) (0.1934) (0.0000) (0.0526) (0.3578)
(12) Fertility 0.4279 0.2751 0.1674 0.7378 0.4590 0.1337 0.7510 0.2177 0.1707 0. 2934 0.1346 1.0000
(0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0006) (0.0000) (0.0000) (0.0000) (0.0000) (0.0006)
a
Number of observations: 651.
b
P-value of pairwise significance shown in parenthesis.

Table 5. Descriptive statistics for the main variablesa

Mean Std. Dev. Min Max
ETHFRACb 0.4630 0.2781 0.0100 0.9586
ETHPOLb 0.5246 0.2367 0.0199 0.9824
Growth 1.9079 3.1954 13.8274 15.5342
Log initial income 8.2983 1.0642 5.8054 11.0808
Investment 15.8787 9.1378 1.0249 91.9660
Civil war 0.1490 0.3564 0 1
Human capital 1.5507 0.5896 0.0764 2.5995
Government 20.3434 8.8690 5.6543 67.4266
Instability 0.1999 0.3743 0 2.800
Distortions 0.1778 0.3402 0.3013 2.6545
Openness 60.2787 46.7204 5.5812 559.1631
Fertility 4.0270 1.5366 1.1584 7.1145
a
Number of observations: 651.
b
Number of countries: 100.
 HOW ETHNIC DIVERSITY AFFECTS ECONOMIC GROWTH 289

Table 6. Descriptive statistics for ETHFRAC and ETHPOL by region

Region Na Mean Std. Dev. Min Max
ETHFRAC
Unrestricted sample 100 0.4630 0.2781 0.0100 0.9586
Sub-Sahara Africa 31 0.6755 0.2314 0.1777 0.9586
Latin America 21 0.4293 0.2108 0.0476 0.7084
East Asia 8 0.5050 0.2747 0.0139 0.8429
OECD 23 0.2348 0.2128 0.0100 0.7668
ETHPOL
Unrestricted sample 100 0.5246 0.2367 0.0199 0.9824
Sub-Sahara Africa 31 0.5465 0.1468 0.2710 0.8031
Latin America 21 0.6265 0.2080 0.2070 0.9547
East Asia 8 0.5827 0.2632 0.0278 0.9298
OECD 23 0.3568 0.2551 0.0199 0.8707
a
Number of countries.

Table 7. Measures for ETHFRAC and ETHPOL

Country ETHFRAC ETHPOL Sub-Sahara Africa Latin America East Asia OECD Frequency
Algeria 0.2986 0.5139 0 0 0 0 7
Argentina 0.4079 0.5788 0 1 0 0 8
Australia 0.3154 0.4918 0 0 0 1 8
Austria 0.1281 0.2398 0 0 0 1 8
Bangladesh 0.0684 0.1318 0 0 0 0 5
Belgium 0.5439 0.8707 0 0 0 1 8
Benin 0.8683 0.4364 1 0 0 0 5
Bolivia 0.7084 0.7666 0 1 0 0 8
Botswana 0.4845 0.6502 1 0 0 0 6
Brazil 0.6441 0.7732 0 1 0 0 8
Burundi 0.2859 0.5123 1 0 0 0 5
Cameroon 0.8166 0.5756 1 0 0 0 7
Canada 0.7668 0.6724 0 0 0 1 8
Central African Republic 0.7870 0.5778 1 0 0 0 6
Chile 0.4316 0.7226 0 1 0 0 8
Colombia 0.6747 0.7889 0 1 0 0 8
Congo, Dem. Rep. 0.7993 0.5859 1 0 0 0 5
Congo, Rep. 0.7214 0.6737 1 0 0 0 8
Costa Rica 0.2408 0.4204 0 1 0 0 8
Cote d’Ivoire 0.8743 0.4319 1 0 0 0 7
Cyprus 0.3565 0.6522 0 0 0 0 1
Denmark 0.0492 0.0967 0 0 0 1 8
Dominican Republic 0.4598 0.7254 0 1 0 0 7
Ecuador 0.6566 0.8372 0 1 0 0 8
Egypt, Arab Rep. 0.2468 0.4270 0 0 0 0 8
El Salvador 0.1453 0.2791 0 1 0 0 7
Fiji 0.5590 0.9298 0 0 1 0 1
Finland 0.1481 0.2941 0 0 0 1 8
France 0.1472 0.2944 0 0 0 1 8
Gabon 0.8338 0.5188 1 0 0 0 3
Gambia, The 0.7279 0.6893 1 0 0 0 5
Ghana 0.7310 0.6610 1 0 0 0 8
Greece 0.0988 0.1861 0 0 0 1 8
Guatemala 0.5201 0.9547 0 1 0 0 7
Haiti 0.1045 0.2070 0 1 0 0 6
Honduras 0.2541 0.4296 0 1 0 0 7
Iceland 0.0276 0.0552 0 0 0 1 1
India 0.9012 0.3482 0 0 0 0 8
Indonesia 0.7934 0.5288 0 0 1 0 6
Iran, Islamic Rep. 0.7563 0.5984 0 0 0 0 5
Iraq 0.3905 0.6649 0 0 0 0 4
Ireland 0.0723 0.1406 0 0 0 1 8
Israel 0.2856 0.5477 0 0 0 0 5
Italy 0.0798 0.1540 0 0 0 1 8
(continued on next page)
290 WORLD DEVELOPMENT

Table 7 (continued)
Country ETHFRAC ETHPOL Sub-Sahara Africa Latin America East Asia OECD Frequency
Jamaica 0.3538 0.6002 0 1 0 0 7
Japan 0.0336 0.0672 0 0 0 1 8
Jordan 0.5152 0.9824 0 0 0 0 8
Kenya 0.8902 0.3813 1 0 0 0 7
Korea, Rep. 0.0139 0.0278 0 0 1 0 8
Kuwait 0.5130 0.9798 0 0 0 0 4
Lesotho 0.1850 0.3428 1 0 0 0 6
Liberia 0.8905 0.3904 1 0 0 0 4
Luxembourg 0.2978 0.5955 0 0 0 1 1
Malawi 0.6844 0.7359 1 0 0 0 7
Malaysia 0.6950 0.7616 0 0 1 0 8
Mali 0.8620 0.4199 1 0 0 0 6
Mauritania 0.3339 0.5361 1 0 0 0 5
Mauritius 0.4822 0.8031 1 0 0 0 5
Mexico 0.5765 0.6536 0 1 0 0 8
Morocco 0.4747 0.8974 0 0 0 0 8
Mozambique 0.8379 0.4986 1 0 0 0 3
Nepal 0.6822 0.6518 0 0 0 0 8
Netherlands 0.1133 0.2137 0 0 0 1 8
New Zealand 0.1959 0.3658 0 0 0 1 8
Nicaragua 0.4962 0.6809 0 1 0 0 6
Niger 0.7185 0.6977 1 0 0 0 6
Norway 0.0454 0.0902 0 0 0 1 8
Pakistan 0.6084 0.6976 0 0 0 0 6
Panama 0.0476 0.5862 0 1 0 0 7
Papua New Guinea 0.3537 0.6687 0 0 1 0 5
Paraguay 0.1739 0.3096 0 1 0 0 8
Peru 0.6581 0.8170 0 1 0 0 8
Philippines 0.8429 0.4965 0 0 1 0 8
Portugal 0.0100 0.0199 0 0 0 1 7
Rwanda 0.2213 0.4013 1 0 0 0 7
Saudi Arabia 0.0589 0.1139 0 0 0 0 3
Senegal 0.8093 0.5596 1 0 0 0 8
Sierra Leone 0.7927 0.6002 1 0 0 0 5
Singapore 0.4214 0.6663 0 0 1 0 1
South Africa 0.4693 0.7178 1 0 0 0 5
Spain 0.4359 0.6933 0 0 0 1 7
Sri Lanka 0.4519 0.7493 0 0 0 0 8
Sudan 0.7114 0.6994 1 0 0 0 4
Swaziland 0.1777 0.3184 1 0 0 0 5
Sweden 0.1891 0.3368 0 0 0 1 8
Switzerland 0.5604 0.7242 0 0 0 1 8
Syrian Arab Republic 0.2066 0.3726 0 0 0 0 7
Tanzania 0.9586 0.2710 1 0 0 0 7
Thailand 0.3608 0.5823 0 0 1 0 8
Togo 0.7325 0.6733 1 0 0 0 8
Trinidad and Tobago 0.6625 0.8417 0 1 0 0 5
Tunisia 0.0872 0.1673 0 0 0 0 7
Turkey 0.1851 0.3424 0 0 0 1 7
Uganda 0.9319 0.2786 1 0 0 0 6
United Kingdom 0.3730 0.5706 0 0 0 1 8
United States 0.5828 0.6913 0 0 0 1 8
Uruguay 0.2595 0.4264 0 1 0 0 8
Venezuela 0.5394 0.7579 0 1 0 0 8
Zambia 0.7874 0.6063 1 0 0 0 7
Zimbabwe 0.5338 0.6978 1 0 0 0 6
P
31 21 8 23 651
 Table 8. System estimate of the base specification for ETHFRAC. Dependent variable is per capita GDP growth (five-year averages)
Growth Investment Civil war Human capital Government Instability Distortions Openness Fertility
***
Sub-Sahara Africa dummy 1.6451 (0.3841) 0.0144 (0.0512)
*** ***
Latin America dummy 0.8935 (0.3357) 0.1388 (0.0491)
**
East Asia dummy 0.2348 (0.4998) 0.1367 (0.0573)
** *** *** ** *** *** *** ***
Log initial income 4.0801 (1.9362) 2.2160 (0.4825) 0.0187 (0.0251) 0.2163 (0.0161) 0.9827 (0.4660) 0.0690 (0.0201) 0.0758 (0.0238) 8.4257 (3.0142) 0.1962 (0.0631)
***
Log initial income sq. 0.3529 (0.1159)
***
Investment 0.0900 (0.0144)
***
Civil war 0.3885 (0.3341) 2.9776 (0.8523) 0.3684 (1.0253)
* *** *** ***
Human capital 0.5848 (0.3405) 4.2662 (0.7875) 0.0207 (0.0409) 0.0251 (0.0421) 24.6579 (4.8150) 1.8389 (0.1321)

HOW ETHNIC DIVERSITY AFFECTS ECONOMIC GROWTH

*** *** *** *** *** ***
Government 0.0346 (0.0131) 0.0278 (0.0337) 0.0074 (0.0013) 0.0044 (0.0016) 0.0060 (0.0015) 1.3563 (0.1785) 0.0209 (0.0043)
* *** ** ** *
Instability 0.6004 (0.3164) 1.3453 (0.8184) 0.3190 (0.0347) 2.3381 (0.9778) 0.0807 (0.0348) 7.1033 (4.2638) 0.0499 (0.0948)
*** *** ***
Distortions 1.9340 (0.3378) 0.5836 (0.8924) 4.1974 (1.0369) 22.2740 (4.8464)
*** * *** ***
Openness 0.0032 (0.0026) 0.0261 (0.0061) 0.0004 (0.0002) 0.0632 (0.0080) 0.0014 (0.0003)
** * ***
Fertility 0.3027 (0.1221) 0.0288 (0.0151) 0.2033 (0.0115)
***
PISH 5.6945 (0.5786)
***
PPDEV 5.7577 (0.6214) 0.0167 (0.0231)
*** *** ***
Log population 0.0776 (0.0146) 0.1698 (0.2814) 0.0380 (0.0114) 10.6929 (1.4779)
*** *** **
Democracy dummy 0.0875 (0.0328) 0.1221 (0.0259) 0.9572 (0.7883) 0.0388 (0.0344) 0.0484 (0.0323) 9.6361 (3.9888)
Island dummy 0.0720v (0.0419) 0.0499 (0.0419) 0.4868 (4.6837)
**
Landlocked dummy 0.0030 (0.0398) 0.0365 (0.0426) 11.4466 (4.5121)
Mountains 0.0001 (0.0007)
*** ***
Log area 0.0472 (0.0116) 4.8357 (1.2970)
*** *** ***
Ever a colony dummy 0.2290 (0.0541) 0.1819 (0.0304) 0.0475 (0.9955) 0.1452 (0.0431) 0.0351 (0.0385)
*** ***
Postwar dummy 0.0761 (0.0507) 1.1301 (0.9151) 0.1184 (0.0383) 0.0403 (0.0384) 15.8119 (4.1828)
***
Oil producer dummy 33.7947 (7.0458)
Muslim dummy 0.0659 (0.0438) 0.0312 (0.0533)
***
Confucian dummy 0.4644 (0.0599) 0.0686 (0.0776)
***
Catholic dummy 0.2580 (0.0404) 0.0074 (0.0529)
***
Other Christian dummy 0.3538 (0.0432) 0.0335 (0.0556)
Mortality rate 0.3798 (0.9896)
* *** *** *** *** ***
Constant 5.7808 (8.0101) 6.9886 (3.7019) 0.7747 (0.2832) 0.0568 (0.1706) 21.5742 (5.1200) 0.2898 (0.2348) 0.7128 (0.1949) 185.8909 (26.6580) 7.9482 (0.6143)
*** ** **
ETHFRAC 1.9595 (0.4783) 1.6006 (1.1702) 0.0633 (0.0619) 0.0499 (0.0468) 1.6731 (1.4557) 0.0097 (0.0620) 0.0526 (0.0560) 14.2083 (7.1136) 0.3330 (0.1555)
N 651 651 651 651 651 651 651 651 651
Adjusted R2 0.28 0.38 0.23 0.77 0.07 0.09 0.12 0.28 0.61

Notes: The sample includes 100 countries for the period 1960–99. Investment, government consumption, political instability, market distortions, trade openness, and fertility are entered as five-year
averages (i.e., 1960–64, 1965–69, . . ., 1995–99). Log initial income, Human capital and log population are measured at the beginning of each five-year period. Civil war dummy takes value 1 if the
country experienced a civil war during the five-year periods. PISH and PPDEV refers to the price level of investment (PPP I/Xrate relative to US) and the magnitude of the absolute deviation of PISH
from the sample mean, respectively. The democracy dummy takes value 1 if the Polity IV democracy score (0 = low; 10 = high) is equal to or higher than 4. Island dummy takes value 1 if country is an
island (Australia is not an island). Landlocked dummy takes value 1 if country has no coastline. Mountains is the percent of mountainous terrain. Log area is the log of area in 1,000s of square
kilometers. Ever a colony dummy takes value 1 if the country was ever a colony since 1776. Postwar dummy takes value 1 if country gained independence after the Second World War. Oil producer
dummy takes value 1 if country is oil exporter. Muslim dummy takes value 1 if majoritarian religion is Muslim. Confucian dummy takes value 1 if majoritarian religion is Buddhism, Xintoism,
Confucianism, etc., excluding Hindus. Catholic dummy takes value 1 if majoritarian religion is Catholicism. Other Christian dummy takes value 1 if majoritarian religion is Christian, but not
Catholicism. Mortality rate is the children mortality rate (ages 0–4). The method of estimation is seemingly unrelated regression. Robust standard errors between parenthesis.
*
Significant at the 10% level.
**
Significant at the 5% level.
***
Significant at the 1% level.

291
 292
Table 9. System estimate of the base specification for ETHPOL. Dependent variable is per capita GDP growth (five-year averages)
Growth Investment Civil war Human capital Government Instability Distortions Openness Fertility
***
Sub-Sahara Africa dummy 2.1404 (0.3693) 0.0035 (0.0474)
*** ***
Latin America dummy 1.0121 (0.3462) 0.1589 (0.0488)
**
East Asia dummy 0.1108 (0.4972) 0.1439 (0.0569)
*** *** *** ** *** *** *** ***
Log initial income 5.3889 (1.9292) 2.5847 (0.4596) 0.0084 (0.0253) 0.2148 (0.0156) 0.9012 (0.4498) 0.0704 (0.0197) 0.0702 (0.0236) 10.0608 (2.9415) 0.2516 (0.0607)
***
Log initial income sq. 0.4277 (0.1156)
***
Investment 0.0891 (0.0146)
***
Civil war 0.4566 (0.3391) 2.7133 (0.8511) 0.3335 (1.0270)
** *** *** ***
Human capital 0.6772 (0.3449) 3.8359 (0.7884) 0.0145 (0.0408) 0.0225 (0.0424) 26.1492 (4.8260) 1.7166 (0.1298)
*** *** ** *** *** ***
Government 0.0355 (0.0133) 0.0463 (0.0335) 0.0072 (0.0013) 0.0042 (0.0016) 0.0063 (0.0015) 1.2566 (0.1784) 0.0169 (0.0043)
* *** ** ** *
Instability 0.6263 (0.3195) 1.1729 (0.8135) 0.3179 (0.0346) 2.2051 (0.9759) 0.0800 (0.0348) 7.8527 (4.2622) 0.0010 (0.0934)
*** *** ***
Distortions 1.9260 (0.3414) 0.5601 (0.8882) 4.4304 (1.0334) 22.1197 (4.8144)
** *** ** *** ***
Openness 0.0056 (0.0026) 0.0287 (0.0061) 0.0005 (0.0002) 0.0601 (0.0080) 0.0014 (0.0003)
** * ***
Fertility 0.3093 (0.1240) 0.0268 (0.0151) 0.2001 (0.0116)
***
PISH 5.9260 (0.5772)
***
PPDEV 5.9807 (0.6189) 0.0203 (0.0227)
*** *** ***
Log population 0.0808 (0.0145) 0.0705 (0.2732) 0.0374 (0.0110) 10.3541 (1.4910)
*** *** **
Democracy dummy 0.0844 (0.0327) 0.1271 (0.0260) 0.9663 (0.7867) 0.0388 (0.0344) 0.0461 (0.0324) 9.7972 (3.9793)

WORLD DEVELOPMENT
*
Island dummy 0.0735 (0.0415) 0.0508 (0.0417) 0.0717 (4.6891)
***
Landlocked dummy 0.0061 (0.0396) 0.0324 (0.0427) 11.7203 (4.5097)
Mountains 0.0004 (0.0007)
*** ***
Log area 0.0477 (0.0113) 4.3982 (1.2385)
*** *** ***
Ever a colony dummy 0.2174 (0.0539) 0.1969 (0.0308) 0.9657 (1.0228) 0.1283 (0.0445) 0.0431 (0.0399)
* *** ***
Postwar dummy 0.0921 (0.0503) 1.1025 (0.9016) 0.1172 (0.0376) 0.0295 (0.0377) 17.9236 (4.0139)
***
Oil producer dummy 34.2923 (7.0255)
Muslim dummy 0.0616 (0.0428) 0.0469 (0.0520)
***
Confucian dummy 0.4554 (0.0577) 0.0468 (0.0749)
***
Catholic dummy 0.2574 (0.0395) 0.0001 (0.0524)
***
Other Christian dummy 0.3522 (0.0425) 0.0420 (0.0548)
Mortality rate 0.0526 (0.9525)
*** *** *** *** *** ***
Constant 11.9158 (7.9436) 8.6698 (3.2751) 0.7433 (0.2819) 0.0432 (0.1593) 20.6263 (5.0533) 0.2959 (0.2328) 0.6492 (0.1834) 191.2786 (26.0828) 8.0162 (0.5931)
*** ** * ** ***
ETHPOL 0.7874 (0.5023) 3.7221 (1.2221) 0.1492 (0.0586) 0.0022 (0.0487) 2.6297 (1.4979) 0.0459 (0.0635) 0.0444 (0.0575) 15.7303 (6.7847) 0.7684 (0.1553)
N 651 651 651 651 651 651 651 651 651
Adjusted R2 0.27 0.39 0.23 0.77 0.08 0.09 0.12 0.28 0.63

Notes: The sample includes 100 countries for the period 1960–99. Investment, government consumption, political instability, market distortions, trade openness, and fertility are entered as five-year
averages (i.e., 1960–64, 1965–69, . . ., 1995–99). Log initial income, Human capital, and log population are measured at the beginning of each five-year period. Civil war dummy takes value 1 if the
country experienced a civil war during the five-year periods. PISH and PPDEV refers to the price level of investment (PPP I/Xrate relative to US) and the magnitude of the absolute deviation of PISH
from the sample mean, respectively. The democracy dummy takes value 1 if the Polity IV democracy score (0 = low; 10 = high) is equal to or higher than 4. Island dummy takes value 1 if country is an
island (Australia is not an island). Landlocked dummy takes value 1 if country has no coastline. Mountains is the percent of mountainous terrain. Log area is the log of area in 1,000s of square
kilometers. Ever a colony dummy takes value 1 if the country was ever a colony since 1776. Postwar dummy takes value 1 if country gained independence after the Second World War. Oil producer
dummy takes value 1 if country is oil exporter. Muslim dummy takes value 1 if majoritarian religion is Muslim. Confucian dummy takes value 1 if majoritarian religion is Buddhism, Xintoism,
Confucianism, etc., excluding Hindus. Catholic dummy takes value 1 if majoritarian religion is Catholicism. Other Christian dummy takes value 1 if majoritarian religion is Christian, but not
Catholicism. Mortality rate is the children mortality rate (ages 0–4). The method of estimation is seemingly unrelated regression. Robust standard errors between parenthesis.
*
Significant at the 10% level.
**
Significant at the 5% level.
***
Significant at the 1% level.
 Table 10. Sensitivity analysis for ETHFRAC. Dependent variable is per capita GDP growth (five-year averages)
Variables Base model Time effects Region effects Region/Times effects excl. initial income excl. Sub-S. Africa excl. Latin America excl. East Asia excl. OECD No Instability equ.
*** *** *** *** *** *** *** *** ***
Sub-Sahara Africa dummy 1.6451 (0.3841) 1.0413 (0.3769) 1.6217 (0.3851) 0.9988 (0.3776) 1.6074 (0.3840) – – 1.6967 (0.4119) 1.6392 (0.3973) 1.2976 (0.4446) 1.6671 (0.3853)
*** ** *** *** *** *** *** ***
Latin America dummy 0.8935 (0.3357) 0.8016 (0.3203) 0.9333 (0.3363) 0.8612 (0.3206) 0.8747 (0.3356) 1.1846 (0.2991) – – 0.9123 (0.3446) 0.3347 (0.4496) 0.9920 (0.3335)
East Asia dummy 0.2348 (0.4998) 0.4781 (0.4780) 0.2337 (0.5010) 0.5276 (0.4789) 0.2836 (0.4995) 0.0014 (0.4213) 0.1758 (0.5330) – – 0.7869 (0.6025) 0.1471 (0.4999)
*** *** *** *** *** *** *** *** *** ***
Investment 0.0900 (0.0144) 0.0610 (0.0147) 0.0902 (0.0144) 0.0608 (0.0147) 0.0909 (0.0144) 0.0915 (0.0171) 0.0748 (0.0165) 0.0899 (0.0153) 0.0821 (0.0174) 0.0906 (0.0145)

HOW ETHNIC DIVERSITY AFFECTS ECONOMIC GROWTH

*
Civil war 0.3885 (0.3341) 0.0167 (0.3238) 0.3931 (0.3340) 0.0002 (0.3235) 0.3898 (0.3340) 0.3896 (0.3170) 0.2257 (0.4146) 0.3187 (0.3590) 0.0791 (0.3811) 0.5986 (0.3177)
* ** * *** ** ** *
Human capital 0.5848 (0.3405) 0.8606 (0.3411) 0.6075 (0.3405) 0.8991 (0.3411) 0.4185 (0.3403) 0.3685 (0.3651) 0.8587 (0.3791) 0.7154 (0.3612) 0.3451 (0.4087) 0.6150 (0.3415)
*** * *** ** *** *** ** *** * ***
Government 0.0346 (0.0131) 0.0230 (0.0126) 0.0359 (0.0131) 0.0248 (0.0126) 0.0339 (0.0131) 0.0395 (0.0138) 0.0368 (0.0147) 0.0365 (0.0135) 0.0267 (0.0152) 0.0361 (0.0131)
* ** ** ** * * ** * *
Instability 0.6004 (0.3164) 0.6410 (0.3053) 0.6247 (0.3163) 0.6546 (0.3053) 0.5778 (0.3164) 0.5277 (0.2859) 0.9145 (0.4564) 0.5592 (0.3389) 0.6736 (0.3694) – –
*** *** *** *** *** *** *** *** *** ***
Distortions 1.9340 (0.3378) 1.5086 (0.3295) 1.9311 (0.3377) 1.4883 (0.3294) 1.8913 (0.3376) 2.7156 (0.3814) 1.8508 (0.3803) 1.9484 (0.3431) 1.8061 (0.3779) 1.9803 (0.3382)
** * ***
Openness 0.0032 (0.0026) 0.0053 (0.0026) 0.0030 (0.0026) 0.0047 (0.0026) 0.0033 (0.0026) 0.0026 (0.0029) 0.0077 (0.0030) 0.0036 (0.0028) 0.0051 (0.0032) 0.0035 (0.0026)
** *** ** *** ** *** *** ** * **
Fertility 0.3027 (0.1221) 0.5058 (0.1208) 0.2942 (0.1221) 0.4874 (0.1207) 0.2922 (0.1220) 0.3091 (0.1179) 0.3936 (0.1446) 0.3040 (0.1290) 0.2488 (0.1495) 0.2883 (0.1223)
*** *** *** *** *** ** *** *** *** ***
ETHFRAC 1.9595 (0.4783) 1.8243 (0.4565) 1.9720 (0.4784) 1.8511 (0.4566) 1.8800 (0.4782) 1.0813 (0.4511) 2.3388 (0.5614) 1.8331 (0.5162) 2.3469 (0.6139) 1.9502 (0.4798)
**
Constant 5.7808 (8.0101) 6.5114 (7.6566) 5.8293 (8.0068) 6.6707 (7.6499) 6.7928 (8.0053) 20.2771 (10.1634) 5.4242 (8.4201) 6.6655 (8.2425) 6.7328 (10.2342) 6.1203 (8.0338)
N 651 651 651 651 651 469 496 606 484 651
Adjusted R2 0.282 0.350 0.281 0.349 0.283 0.361 0.311 0.258 0.257 0.277
Effect of ETHFRAC on the channel
** ** ***
Investment 1.6006 (1.1702) 0.7504 (1.1288) 0.7136 (1.2086) 0.3119 (1.1577) 2.6607 (1.1521) 2.6763 (1.1761) 4.0076 (1.3677) 0.5098 (1.2248) 1.3991 (1.3888) 1.5969 (1.1715)
***
Civil war 0.0633 (0.0619) 0.0709 (0.0609) 0.0671 (0.0620) 0.0707 (0.0609) 0.0602 (0.0620) 0.2188 (0.0775) 0.0486 (0.0647) 0.0037 (0.0625) 0.0357 (0.0859) 0.0594 (0.0658)
Human capital 0.0499 (0.0468) 0.0177 (0.0444) 0.0411 (0.0492) 0.0315 (0.0459) 0.0566 (0.0530) 0.0803 (0.0501) 0.0888 (0.0567) 0.0242 (0.0468) 0.0110 (0.0605) 0.0498 (0.0468)
*** **
Government 1.6731 (1.4557) 1.7886 (1.4482) 0.9377 (1.5438) 1.0232 (1.5350) 1.0888 (1.4135) 4.2267 (1.6316) 3.4379 (1.7302) 1.8057 (1.5647) 0.7095 (1.8758) 1.6912 (1.4573)
Instability 0.0097 (0.0620) 0.0093 (0.0613) 0.0318 (0.0666) 0.0214 (0.0660) 0.0298 (0.0611) 0.0938 (0.0845) 0.0610 (0.0582) 0.0150 (0.0641) 0.0010 (0.0775) – –
*
Distortions 0.0526 (0.0560) 0.0537 (0.0551) 0.0073 (0.0578) 0.0022 (0.0565) 0.0314 (0.0556) 0.0803 (0.0559) 0.0534 (0.0674) 0.0298 (0.0604) 0.1391 (0.0749) 0.0501 (0.0562)
** * ** ** *** ** *** *
Openness 14.2083 (7.1136) 13.8239 (7.0645) 16.2104 (7.5952) 17.1135 (7.4741) 18.6619 (6.9759) 15.1814 (7.0159) 24.9147 (8.2565) 5.2816 (7.1754) 6.9405 (9.1774) 13.8824 (7.1247)
** *** ** *** **
Fertility 0.3330 (0.1555) 0.4420 (0.1547) 0.0946 (0.1565) 0.1833 (0.1538) 0.1681 (0.1545) 0.3565 (0.1808) 0.5629 (0.1729) 0.2417 (0.1648) 0.0617 (0.1783) 0.3346 (0.1555)

Notes: The sample includes 100 countries for the period 1960–99. Investment, government consumption, political instability, market distortions, trade openness, and fertility are entered as five-year
averages (i.e., 1960–64, 1965–69, . . ., 1995–99). Log initial income, Human capital, and log population are measured at the beginning of each five-year period. Civil war dummy takes value 1 if the
country experienced a civil war during the five-year periods. PISH and PPDEV refers to the price level of investment (PPP I/Xrate relative to US) and the magnitude of the absolute deviation of PISH
from the sample mean, respectively. The democracy dummy takes value 1 if the Polity IV democracy score (0 = low; 10 = high) is equal to or higher than 4. Island dummy takes value 1 if country is an
island (Australia is not an island). Landlocked dummy takes value 1 if country has no coastline. Mountains is the percent of mountainous terrain. Log area is the log of area in 1,000s of square
kilometers. Ever a colony dummy takes value 1 if the country was ever a colony since 1776. Postwar dummy takes value 1 if country gained independence after the Second World War. Oil producer
dummy takes value 1 if country is oil exporter. Muslim dummy takes value 1 if majoritarian religion is Muslim. Confucian dummy takes value 1 if majoritarian religion is Buddhism, Xintoism,
Confucianism, etc., excluding Hindus. Catholic dummy takes value 1 if majoritarian religion is Catholicism. Other Christian dummy takes value 1 if majoritarian religion is Christian, but not
Catholicism. Mortality rate is the children mortality rate (ages 0–4). The method of estimation is seemingly unrelated regression. Robust standard errors between parenthesis.
*
Significant at the 10% level.
**
Significant at the 5% level.
***
Significant at the 1% level.

293
 294
Table 11. Sensitivity analysis for ETHPOL. Dependent variable is per capita GDP growth (five-year averages)
Variables Base model Time effects Region effects Region/Times effects excl. initial income excl. Sub-S. Africa excl. Latin America excl. East Asia excl. OECD No Instability equ.
*** *** *** *** *** *** *** *** ***
Sub-Sahara Africa dummy 2.1404 (0.3693) 1.5139 (0.3642) 2.1176 (0.3703) 1.4554 (0.3648) 2.0865 (0.3689) – – 2.2094 (0.4010) 2.1240 (0.3769) 1.8563 (0.4273) 2.1586 (0.3704)
*** *** *** *** *** *** *** ***
Latin America dummy 1.0121 (0.3462) 0.9325 (0.3303) 1.0413 (0.3467) 0.9799 (0.3307) 0.9797 (0.3459) 1.2280 (0.3056) – – 1.0480 (0.3533) 0.4339 (0.4589) 1.1114 (0.3438)
East Asia dummy 0.1108 (0.4972) 0.1443 (0.4758) 0.1400 (0.4986) 0.1694 (0.4766) 0.0636 (0.4966) 0.1811 (0.4135) 0.1929 (0.5324) – – 0.4492 (0.6066) 0.1944 (0.4972)
*** *** *** *** *** *** *** *** *** ***
Investment 0.0891 (0.0146) 0.0590 (0.0149) 0.0898 (0.0146) 0.0593 (0.0149) 0.0908 (0.0146) 0.0883 (0.0172) 0.0789 (0.0167) 0.0872 (0.0154) 0.0790 (0.0177) 0.0897 (0.0146)
**
Civil war 0.4566 (0.3391) 0.0441 (0.3287) 0.4546 (0.3391) 0.0568 (0.3286) 0.4460 (0.3387) 0.4201 (0.3186) 0.3445 (0.4217) 0.3709 (0.3640) 0.1755 (0.3876) 0.6702 (0.3225)
** *** ** *** ** ** **
Human capital 0.6772 (0.3449) 0.9869 (0.3457) 0.6837 (0.3449) 1.0147 (0.3457) 0.4648 (0.3446) 0.3944 (0.3675) 0.9651 (0.3854) 0.7934 (0.3642) 0.5170 (0.4149) 0.7068 (0.3458)
*** * *** ** *** *** ** *** * ***
Government 0.0355 (0.0133) 0.0248 (0.0128) 0.0362 (0.0133) 0.0255 (0.0128) 0.0345 (0.0133) 0.0406 (0.0140) 0.0345 (0.0150) 0.0390 (0.0136) 0.0292 (0.0156) 0.0372 (0.0133)
* ** ** ** * ** * *
Instability 0.6263 (0.3195) 0.6559 (0.3083) 0.6467 (0.3195) 0.6641 (0.3083) 0.6013 (0.3196) 0.5690 (0.2866) 0.8832 (0.4624) 0.5519 (0.3417) 0.6901 (0.3741) – –
*** *** *** *** *** *** *** *** *** ***
Distortions 1.9260 (0.3414) 1.4999 (0.3329) 1.9173 (0.3414) 1.4704 (0.3329) 1.8786 (0.3409) 2.7254 (0.3832) 1.8227 (0.3857) 1.9393 (0.3462) 1.7729 (0.3840) 1.9699 (0.3417)
** *** ** *** ** *** ** ** **
Openness 0.0056 (0.0026) 0.0078 (0.0026) 0.0056 (0.0026) 0.0073 (0.0026) 0.0057 (0.0026) 0.0008 (0.0029) 0.0095 (0.0031) 0.0064 (0.0028) 0.0077 (0.0032) 0.0059 (0.0026)
** *** ** *** ** ** *** ** **
Fertility 0.3093 (0.1240) 0.5259 (0.1228) 0.3031 (0.1240) 0.5068 (0.1227) 0.2989 (0.1239) 0.3058 (0.1190) 0.4018 (0.1475) 0.3090 (0.1309) 0.2260 (0.1516) 0.2952 (0.1242)
ETHPOL 0.7874 (0.5023) 0.6671 (0.4793) 0.7736 (0.5024) 0.6418 (0.4794) 0.8178 (0.5022) 0.6342 (0.4392) 0.9595 (0.5935) 0.6982 (0.5222) 0.7319 (0.7111) 0.7793 (0.5038)
* **
Constant 11.9158 (7.9436) 12.1125 (7.5985) 12.0505 (7.9435) 12.5433 (7.5962) 12.9900 (7.9329) 24.0008 (10.0128) 12.4719 (8.3423) 12.9501 (8.1133) 12.9312 (10.2401) 12.2317 (7.9659)

WORLD DEVELOPMENT
N 651 651 651 651 651 469 496 606 484 651
Adjusted R2 0.267 0.337 0.267 0.337 0.268 0.357 0.291 0.246 0.237 0.262
Effect of ETHPOL on the channel
*** *** ** *** *** ** ***
Investment 3.7221 (1.2221) 3.2876 (1.1729) 1.1220 (1.2649) 0.7465 (1.2121) 2.9174 (1.2662) 3.8437 (1.0798) 4.6124 (1.4832) 2.9332 (1.2565) 1.3983 (1.6705) 3.7800 (1.2225)
** *** ** ** *** *** ** ** **
Civil war 0.1492 (0.0586) 0.1551 (0.0577) 0.1406 (0.0587) 0.1453 (0.0577) 0.1500 (0.0578) 0.1878 (0.0645) 0.1225 (0.0621) 0.0759 (0.0591) 0.1902 (0.0865) 0.1588 (0.0622)
** ***
Human capital 0.0022 (0.0487) 0.0291 (0.0459) 0.0293 (0.0497) 0.0392 (0.0463) 0.1366 (0.0583) 0.0123 (0.0465) 0.0212 (0.0591) 0.0024 (0.0485) 0.1309 (0.0696) 0.0021 (0.0487)
*** *** ** ** *** *** ***
Government 2.6297 (1.4979) 2.5970 (1.4904) 3.0369 (1.4944) 3.0866 (1.4830) 2.2825 (1.5036) 5.2767 (1.4315) 2.5048 (1.8603) 2.1427 (1.6057) 6.7531 (2.1531) 2.7796 (1.4992)
Instability 0.0459 (0.0635) 0.0480 (0.0628) 0.0088 (0.0645) 0.0132 (0.0638) 0.0170 (0.0641) 0.0623 (0.0715) 0.0306 (0.0625) 0.0675 (0.0651) 0.0122 (0.0910) – –
**
Distortions 0.0444 (0.0575) 0.0389 (0.0564) 0.0536 (0.0584) 0.0449 (0.0571) 0.0674 (0.0577) 0.0815 (0.0514) 0.0578 (0.0703) 0.0372 (0.0616) 0.1900 (0.0847) 0.0380 (0.0577)
** ** *** *** ** *** *** **
Openness 15.7303 (6.7847) 14.2001 (6.7211) 26.2663 (7.0824) 26.0182 (6.9835) 15.6149 (6.8161) 18.9247 (5.7864) 26.4089 (8.5330) 7.1528 (6.7129) 1.0898 (10.2477) 14.7109 (6.7798)
*** *** ** *** *** *** *** *** ***
Fertility 0.7684 (0.1553) 0.8309 (0.1527) 0.3444 (0.1585) 0.4237 (0.1554) 0.6202 (0.1625) 0.7679 (0.1624) 0.6081 (0.1824) 0.8019 (0.1609) 0.0196 (0.2093) 0.7680 (0.1550)

Notes: The sample includes 100 countries for the period 1960–99. Investment, government consumption, political instability, market distortions, trade openness, and fertility are entered as five-year
averages (i.e., 1960–64, 1965–69, . . ., 1995–99). Log initial income, Human capital, and log population are measured at the beginning of each five-year period. Civil war dummy takes value 1 if the
country experienced a civil war during the five-year periods. PISH and PPDEV refers to the price level of investment (PPP I/Xrate relative to US) and the magnitude of the absolute deviation of PISH
from the sample mean, respectively. The democracy dummy takes value 1 if the Polity IV democracy score (0 = low; 10 = high) is equal to or higher than 4. Island dummy takes value 1 if country is an
island (Australia is not an island). Landlocked dummy takes value 1 if country has no coastline. Mountains is the percent of mountainous terrain. Log area is the log of area in 1,000s of square
kilometers. Ever a colony dummy takes value 1 if the country was ever a colony since 1776. Postwar dummy takes value 1 if country gained independence after the Second World War. Oil producer
dummy takes value 1 if country is oil exporter. Muslim dummy takes value 1 if majoritarian religion is Muslim. Confucian dummy takes value 1 if majoritarian religion is Buddhism, Xintoism,
Confucianism, etc., excluding Hindus. Catholic dummy takes value 1 if majoritarian religion is Catholicism. Other Christian dummy takes value 1 if majoritarian religion is Christian, but not
Catholicism. Mortality rate is the children mortality rate (ages 0–4). The method of estimation is seemingly unrelated regression. Robust standard errors between parenthesis.
*
Significant at the 1% level.
**
Significant at the 5% level.
***
Significant at the 10% level.
 Table 12. Sensitivity analysis for ETHFRAC. Dependent variable is per capita GDP growth (five-year averages)
Variables Base model Sub-S. Africa Latin America East Asia OECD
*** ** *** ***
Investment 0.0900 (0.0144) 0.0641 (0.0289) 0.1107 (0.0370) 0.0940 (0.0738) 0.1235 (0.0320)
Civil war 0.3885 (0.3341) 0.4646 (0.8471) 0.8172 (0.5719) 1.5576* (0.9383) 1.4264 (1.5488)
Human capital 0.5848* (0.3405) 0.1194 (0.6928) 0.6983 (0.8838) 1.3478 (1.3687) 0.0298 (0.6305)
Government 0.0346*** (0.0131) 0.0329 (0.0282) 0.0338 (0.0379) 0.0588 (0.1318) 0.0344 (0.0382)

HOW ETHNIC DIVERSITY AFFECTS ECONOMIC GROWTH

Instability 0.6004* (0.3164) 1.6518* (0.9297) 0.3493 (0.3830) 0.2679 (0.9269) 0.7758 (0.5722)
Distortions 1.9340*** (0.3378) 0.6089 (0.6456) 2.8831*** (0.8251) 4.4531 (5.0598) 13.5862*** (3.7970)
Openness 0.0032 (0.0026) 0.0076 (0.0055) 0.0105 (0.0064) 0.0042 (0.0136) 0.0063 (0.0044)
Fertility 0.3027** (0.1221) 0.0750 (0.3727) 0.1330 (0.2870) 0.4789 (0.4548) 0.5382** (0.2623)
ETHFRAC 1.9595*** (0.4783) 4.6347*** (1.2648) 1.8165* (1.0915) 2.5359 (2.0079) 0.1945 (0.6597)
Constant 5.7808 (8.0101) 55.3536** (23.8431) 66.4576 (52.8518) 22.7851 (58.4827) 62.2949 (38.7281)
N 651 182 155 45 167
Adjusted R2 0.282 0.240 0.216 0.367 0.401
Effect of ETHFRAC on the channel
Investment 1.6006 (1.1702) 4.3231 (3.1297) 11.3334*** (2.1456) 14.9391*** (2.3829) 7.8662*** (1.5840)
Civil war 0.0633 (0.0619) 0.0106 (0.1621) 0.3773** (0.1826) 2.6428* (1.4588) 0.0204 (0.0357)
Human capital 0.0499 (0.0468) 0.0112 (0.1219) 0.1894** (0.0867) 0.5482 (0.8395) 0.0867* (0.0471)
Government 1.6731 (1.4557) 11.3205*** (3.4242) 5.0522** (2.5355) 6.7521** (2.7612) 6.7134*** (1.5659)
Instability 0.0097 (0.0620) 0.2046 (0.1406) 0.3649 (0.2361) 0.5661 (1.0095) 0.0615 (0.1046)
Distortions 0.0526 (0.0560) 0.2021 (0.1453) 0.3931*** (0.0989) 0.0496 (0.2086) 0.0134 (0.0125)
Openness 14.2083** (7.1136) 9.6010 (24.8154) 36.6986*** (12.2899) 210.5391*** (78.7544) 58.0732*** (8.8796)
Fertility 0.3330** (0.1555) 0.0329 (0.2622) 0.2005 (0.2827) 1.2633*** (0.3294) 0.3090* (0.1701)
Notes: The sample includes 100 countries for the period 1960–99. Investment, government consumption, political instability, market distortions, trade openness, and fertility are entered as five-year
averages (i.e., 1960–64, 1965–69, . . ., 1995–99). Log initial income, Human capital, and log population are measured at the beginning of each five-year period. Civil war dummy takes value 1 if the
country experienced a civil war during the five-year periods. PISH and PPDEV refers to the price level of investment (PPP I/Xrate relative to US) and the magnitude of the absolute deviation of PISH
from the sample mean, respectively. The democracy dummy takes value 1 if the Polity IV democracy score (0 = low; 10 = high) is equal to or higher than 4. Island dummy takes value 1 if country is an
island (Australia is not an island). Landlocked dummy takes value 1 if country has no coastline. Mountains is the percent of mountainous terrain. Log area is the log of area in 1,000s of square
kilometers. Ever a colony dummy takes value 1 if the country was ever a colony since 1776. Postwar dummy takes value 1 if country gained independence after the Second World War. Oil producer
dummy takes value 1 if country is oil exporter. Muslim dummy takes value 1 if majoritarian religion is Muslim. Confucian dummy takes value 1 if majoritarian religion is Buddhism, Xintoism,
Confucianism, etc., excluding Hindus. Catholic dummy takes value 1 if majoritarian religion is Catholicism. Other Christian dummy takes value 1 if majoritarian religion is Christian, but not
Catholicism. Mortality rate is the children mortality rate (ages 0–4). The method of estimation is seemingly unrelated regression. Robust standard errors between parenthesis.
*
Significant at the 10% level.
**
Significant at the 5% level.
***
Significant at the 1% level.

295
 296
Table 13. Sensitivity analysis for ETHPOL. Dependent variable is per capita GDP growth (five-year averages)
Variables Base model Sub-S. Africa Latin America East Asia OECD
Investment 0.0891*** (0.0146) 0.0764** (0.0306) 0.0957*** (0.0370) 0.1696** (0.0654) 0.1184*** (0.0319)
Civil war 0.4566 (0.3391) 0.5422 (0.8767) 0.9491 (0.5827) 2.3475*** (0.8983) 1.5132 (1.5478)
Human capital 0.6772** (0.3449) 0.5587 (0.7140) -0.9596 (0.8860) 2.4390* (1.3541) 0.0050 (0.6336)
Government 0.0355*** (0.0133) 0.0149 (0.0291) 0.0426 (0.0382) 0.1405 (0.1217) 0.0411 (0.0374)
Instability 0.6263* (0.3195) 1.2931 (0.9567) 0.4066 (0.3854) 0.0980 (0.8908) 0.7950 (0.5720)
Distortions 1.9260*** (0.3414) 0.6655 (0.6663) 2.7688*** (0.8298) 12.8893** (5.8758) 13.9466*** (3.7909)
Openness 0.0056** (0.0026) 0.0086 (0.0057) 0.0048 (0.0058) 0.0230 (0.0170) 0.0060 (0.0044)
Fertility 0.3093** (0.1240) 0.1695 (0.4068) 0.1958 (0.2887) 0.0618 (0.4871) 0.5291** (0.2625)
ETHPOL 0.7874 (0.5023) 0.0375 (2.0752) 0.5184 (1.0483) 7.6393** (3.3175) 0.0071 (0.5534)
Constant 11.9158 (7.9436) 47.7260* (24.6574) 71.8594 (53.8346) 46.5847 (58.8804) 62.0112 (38.7871)
N 651 182 155 45 167

WORLD DEVELOPMENT
Adjusted R2 0.267 0.186 0.206 0.408 0.400
Effect of ETHPOL on the channel
Investment 3.7221*** (1.2221) 11.2641** (4.4451) 10.8254*** (2.0563) 6.2951 (5.7632) 6.2617*** (1.3411)
Civil war 0.1492** (0.0586) 0.0576 (0.1754) 0.3888** (0.1624) 0.7953 (1.9418) 0.0079 (0.0273)
Human capital 0.0022 (0.0487) 0.2201 (0.1944) 0.0810 (0.0845) 2.2703*** (0.5183) 0.0400 (0.0389)
Government 2.6297* (1.4979) 10.2618** (5.1215) 3.3248 (2.3016) 18.0190*** (2.9875) 4.7025*** (1.2996)
Instability 0.0459 (0.0635) 0.0891 (0.1857) 0.2225 (0.2146) 3.2898*** (1.0133) 0.0645 (0.0803)
Distortions 0.0444 (0.0575) 0.0710 (0.2262) 0.2216** (0.0922) 0.3246** (0.1313) 0.0132 (0.0102)
Openness 15.7303** (6.7847) 36.5795 (29.3167) 15.6689 (11.3656) 181.3010*** (53.9450) 41.9367*** (7.0156)
Fertility 0.7684*** (0.1553) 1.4904*** (0.3675) 0.3962 (0.2804) 1.0446** (0.4664) 0.2892** (0.1425)
Notes: The sample includes 100 countries for the period 1960–99. Investment, government consumption, political instability, market distortions, trade openness, and fertility are entered as five-year
averages (i.e., 1960–64, 1965–69,. . ., 1995–99). Log initial income, Human capital, and log population are measured at the beginning of each five-year period. Civil war dummy takes value 1 if the
country experienced a civil war during the five-year periods. PISH and PPDEV refers to the price level of investment (PPP I/Xrate relative to US) and the magnitude of the absolute deviation of PISH
from the sample mean, respectively. The democracy dummy takes value 1 if the Polity IV democracy score (0 = low; 10 = high) is equal to or higher than 4. Island dummy takes value 1 if country is an
island (Australia is not an island). Landlocked dummy takes value 1 if country has no coastline. Mountains is the percent of mountainous terrain. Log area is the log of area in 1000s of square
kilometers. Ever a colony dummy takes value 1 if the country was ever a colony since 1776. Postwar dummy takes value 1 if country gained independence after the Second World War. Oil producer
dummy takes value 1 if country is oil exporter. Muslim dummy takes value 1 if majoritarian religion is Muslim. Confucian dummy takes value 1 if majoritarian religion is Buddhism, Xintoism,
Confucianism, etc., excluding Hindus. Catholic dummy takes value 1 if majoritarian religion is Catholicism. Other Christian dummy takes value 1 if majoritarian religion is Christian, but not
Catholicism. Mortality rate is the children mortality rate (ages 0–4). The method of estimation is seemingly unrelated regression. Robust standard errors between parenthesis
*
Significant at the 10% level.
**
Significant at the 5% level.
***
Significant at the 1% level.
 HOW ETHNIC DIVERSITY AFFECTS ECONOMIC GROWTH 297

Table 14. Data description

Growth Growth of real GDP per capita (Chain) in %: rgdpch [Base year: 2000] for 1960–99, constant price
entries. Source: Heston et al. (2006), Penn World Table Version 6.2 (SH v. 6.2)
Sub-Sahara Africa dummy Dummy for Sub-Saharan African countries
Latin America dummy Dummy for Latin-American countries
East Asia dummy Dummy for East Asian countries
OECD dummy Dummy for OECD countries
Log initial income Log of per capita GDP at the beginning of each five-year period. Source: SH v. 6.2
Log initial income squared Log of per capita GDP squared at the beginning of each five-year period. Source: SH v. 6.2
Investment Investment share of real GDP per capita (Laspeyres): ki [Base year: 2000], constant price entries.
Source: SH v. 6.2
Civil war Takes value 1 if the country experienced a civil war during the five-year periods. Source: Doyle and
Sambanis (2000)/compiled by Garcia-Montalvo and Reynal-Querol (2005b)
Human capital Log of (1 + avg. schooling years in the total population over age 25). Source: Barro and Lee (2010)
Government Government share of real GDP per capita (Laspeyres) in %: kg [Base year: 2000], constant price
entries. Source: SH v. 6.2
Instability Measure of political instability. Unit: Number of revolutions and coups per year. Source: Banks
(2011)
Distortions Log of (1 + black market premium). Source: Easterly (2001)
Openness Exports plus imports divided by GDP (rgdpl) [Base year: 2000] in %, constant price entries. Source:
SH v. 6.2
Fertility Net fertility rate calculated as TFR(1-MORT04)
PISH Price level of investment (PPP I/Xrate relative to US) (US = 1.0). Source: SH v. 6.2
PPDEV Magnitude of the absolute deviation of the Price Level of Investment (PISH) from the sample mean.
Source: SH v. 6.2
Log poulation Log of the population at the beginning of each five-year period. Source: SH v. 6.2
Democracy dummy Democracy score (0 = low; 10 = high). The democracy score is transformed into a dummy variable
that takes the value 1 if the democracy score is equal to or higher than 4. Source: Polity IV data set/
compiled by Garcia-Montalvo and Reynal-Querol (2005b)
Island dummy Takes value 1 if country is an island (Australia is not an island). Source: Tavares and Wacziarg
(2001)
Landlocked dummy Takes value 1 if country has no coastline. Source: CIA World FactBook (1996)/compiled by Tavares
and Wacziarg (2001)
Mountains Percent mountainous terrain. Source: Garcia-Montalvo and Reynal-Querol (2005b)
Log area Log of area in 1,000s of square kilometers. Source: Barro and Lee (1993)/completed by Tavares and
Wacziarg (2001)
Ever a colony dummy Takes value 1 if the country was ever a colony since 1776. Source: Barro and Lee (1993)/completed
by Tavares and Wacziarg (2001)
Postwar dummy Takes value 1 if country gained independence after the Second World War. Source: Barro and Lee
(1993)/completed by Tavares and Wacziarg (2001)
Oil producer dummy Takes value 1 if country is oil exporter. Source: Barro and Lee (1993)/completed by Tavares and
Wacziarg (2001)
Muslim dummy Takes value 1 if majoritarian religion is Muslim. Source: Encyclopedia Britannica/compiled by
Tavares and Wacziarg (2001)
Confucian dummy Takes value 1 if majoritarian religion is Buddhism, Xintoism, Confucianism, etc. (excluding Hindus).
Source: Encyclopedia Britannica/compiled by Tavares and Wacziarg (2001)
Catholic dummy Takes value 1 if majoritarian religion is Catholicism. Source: Encyclopedia Britannica/compiled by
Tavares and Wacziarg (2001)
Other Christian dummy Takes value 1 if majoritarian religion is Christian, but not Catholicism. Source: Encyclopedia
Britannica/compiled by Tavares and Wacziarg (2001)
Mortality rate Children mortality rate (ages 0–4). Source: World Bank
Total fertility rate Total fertility rate (children per woman). Source: World Bank
ETHFRAC Index of ethnic fractionalization calculated using the data from the World Christian Encyclopedia
(WCE). Source: Garcia-Montalvo and Reynal-Querol (2005b)
ETHPOL Index of ethnic polarization calculated using the data of the WCE. Source: Garcia-Montalvo and
Reynal-Querol (2005b)

APPENDIX B. SUPPLEMENTARY DATA

Supplementary data associated with this article can be

found, in the online version, at http://dx.doi.org/10.1016/
j.worlddev.2014.01.012.

Available online at www.sciencedirect.com

ScienceDirect