Académique Documents
Professionnel Documents
Culture Documents
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp
.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.
Sage Publications, Ltd. is collaborating with JSTOR to digitize, preserve and extend access to Journal of Peace
Research.
http://www.jstor.org
GUDRUN 0STBY
Department ofPolitical Science, University ofOslo; and Centre for the Study of
Civil War (CSCW), International Peace Research Institute, Oslo (PRIO)
Recent war conclude that inequality does not increase the risk of violent conflict.
large-N studies of civil
This article argues that such conclusions may be premature because these studies, which usually test the
conflict potential of Vertical inequality' (i.e. income inequality between individuals), tend to neglect the
group aspect of inequality. Case studies suggest thatwhat matters for conflict is a concept closely linked
to both economic and ethnic polarization: 'horizontal inequalities', or inequalities that coincide with
identity-based cleavages. Horizontal inequalities may enhance both grievances and group cohesion
among the relatively deprived and thus facilitate mobilization for conflict. This article provides a quan
titative test of this argument, exploring whether various forms of polarization and horizontal inequali
ties affect the probability of civil conflict onset across 36 developing countries in the period 1986-2004.
National household data from the Demographic and Health Surveys (DHS) are used to construct
measures of ethnic, social and economic polarization, as well as vertical and horizontal inequalities along
two dimensions: social and economic. The article also introduces a combined measure of ethnic/socio
economic polarization as an alternative to the horizontal inequality measure. Robust results from panel
and cross-section analyses show that social polarization and horizontal social inequality are positively
related to conflict outbreak. Variables for purely ethnic polarization, inter-individual inequalities and
combined ethnic/socio-economic are not
polarization significant.
race, religion or language. So-called 'ethnic' riskof civilwar (e.g.Collier & Hoeffler, 2004;
occur between which are
conflicts groups Fearon & Laitin, 2003; Hegre, Gissinger &
*
This article is part of the Polarization and Conflict in those meetings. I am also grateful to Juha Auvinen,
Project CIT-2-CT-2004-506084 funded by the European Tanja Ellingsen, James Fearon, Scott Gates, Nils Petter
Commission-DG Research Sixth Framework Programme. Gleditsch, H?vard Hegre, Jo Thori Lind, Nicholas
This article reflects only the author's views and the Sambanis, Henrik Urdal, two anonymous referees and the
Community is not liable for any use thatmay be made of editors of this special issue forhelpful comments. A special
the information contained therein.An earlier version was thanks toH?vard Strand for useful insights and technical
presented to the PAC Winter Meeting inBarcelona, 10?11 assistance in generating several of the variables. The data,
December 2004 and to the 46th Annual Convention of the codebook and do-files used in the analysis, as well as the
International Studies Association, Honolulu, HI, 1?5 online appendix, are available at http://www.prio.no/jpr/
March 2005, and I thank the discussants and participants datasets. Correspondence: gudrun.ostby@ stv.uio.no.
143
Gleditsch, 2003). I argue in this article The analysis is based on a new dataset
that this conclusion is potentially false, constructed from microdata from the
because these studies largely neglect the Demographic and Health Surveys (DHS) for
group aspect of inequality. Civil wars are 39 developing countries in the period
? not
conflicts confrontations between 1986-2004. I construct macro
group aggregated
individuals randomly fighting each other indicators of polarization and horizontal
(Duelos, Esteban & Ray, 2004). Hence, the inequalities. The latter refers to inequality
focus should be on or between the two ethnic groups in each
polarization, inequal largest
not between two dimensions:
ity between groups, individuals. country along
economic
Inequalities that coincide with ethnic cleav (ownership of various household assets) and
ages may enhance both
grievances
and group social (educational opportunities).
cohesion among the relativelydeprived and The article is structured as follows: I first
thus facilitate mobilization for conflict discuss some problems with the inequality
(Gurr, 2000; Murshed & Gates, 2005; conflict literature. Next, I present a theoretical
itybetween individuals (see Stewart, 2000). break.This indicates that it is too early to reject
Moreover, the majority of
inequality-conflict
the inequality?conflict nexus and that future
studies concentrate
exclusively
on economic research should focus more on the interplay
income inequality, thus failing to capture the between ethnicity and socio-economic distrib
multidimensional nature of inequality. utions. terms for
The purely ethnic polariza
In a series of case studies, Stewart (2002) tion and
purely
socio-economic inter-individual
found that various horizontal inequalities inequalities
are not
significant
in the
analysis.
have provoked some kind of conflict, ranging Neither is the combined measure of ethnic/
from severe criminality inBrazil to civilwar in socio-economic
polarization.
Uganda and Sri Lanka. These studies provide
deep insight into the specific cases but are less Problems with the Inequality?Conflict
suited for to a universal relation
generalizing
Literature
ship between His and conflict. Furthermore,
Stewartmay be criticized for selecting on the A remarkablydiverse literature,theoretical as
(and independent) variable(s), well as has evolved as a response to
dependent empirical,
since she investigates only countries where theproposition thatpolitical violence isa func
His have led to some kind of conflict. In order tion of economic
inequality. There are, in
to assess the between His eco
general relationship theory, five possible relationships between
and conflict, one needs a
systematic, large-N
nomic inequality and political conflict: posi
research The main tive, convex (inverted
quantitative design. objec negative, U-shaped),
tive of this article is to provide such a study concave (U-shaped) or null. The empirical lit
and also to compare pure measures of ethnic erature of all.1 Ever since
provides examples
and socio-economic
polarization, respectively, the publication of a seminal paper byRussett
with HI measures and measures of two
dimensional polarization, which cover both 1
Lichbach (1989) located 43 quantitative studies of the
ethnic and socio-economic aspects. inequality?conflict nexus.
(1964), scholars have tried to statistically While a standard critique about missing
determine the between vertical control variables, poor data3 and statistical
relationship
1996; Auvinen & Nafziger, 1999; Collier & these studies take an individualistic approach
Hoeffler, 2004; Fearon & Laitin, 2003; Hegre, to
inequality, concentrating exclusively
on
Gissinger & Gleditsch, 2003; M?ller & inequality among individuals, even though
Seligson, 1987; Nagel, 1974; Parvin, 1973; what they are trying to explain is group
Sigelman & Simpson, 1977;Weede, 1981.) conflict. This may explain why
so many
Most traditional studies of inequality and researchers have failed to find a relationship
conflict relate to the theory of relative depri between inequality and civil war.4
vation (see Gurr, 1970). This theory argues The second problem with the inequal
that while absolute poverty may lead to ity-conflict literature is that most studies
and inactivity, with consider economic
apathy comparisons only inequality, usually
those in the same society who do better may measured as income inequality. This is also
result in violence. Recent work on civil war true for the most recent studies, almost all of
has distinguished between so-called greed which use the inequality data provided by
Deininger & Squire (1996) and UNU/
driven and rebellion: on
grievance-driven
the one hand, rationalist theories that focus WIDER & UNDP (2000). Sen (1992: 28)
on the to a rebellion asks a that is essential for the present
opportunities organize question
on the other hand, relative
(greed) and, inquiry: 'Equality of what?', stressing that
theories that focus on motives for can be much more than in
deprivation inequality just
groups to change their situation. Collier & come inequalitymeasured by theGini index.
Hoeffler (2004) discuss several grievances He focuses on three different categories of
(e.g. inequality, lack of political rights, and equality: equality of income or other finan
ethnic divisions) as well as opportunities for cial assets; equality of welfare; and equal
rights and liberties. Stewart (2002) concurs
a rebel access to
forming organization (e.g.
finance and natural resources), and that may occur in eco
inequal inequalities political,
nomic In sum,
ity is among the grievance factors that they and social dimensions. inequal
3A
problem with the income inequality data is the high
level of missing observations. Furthermore, this pattern is
2
Although political opportunities may moderate the rela non-random, that is,we have fewer inequality data for con
tionship between His (or polarization) and conflict flict-ridden countries. This problem is likely to be less
(Schock, 1996), such interaction effects are not reported severe here, since theHI data are generated from surveys of
herein, as my focus is the link between various forms of which the original intentionwas not to assess inequalities.
4
polarization, His and conflict. However, I also ran all Exceptions include Besan?on (2005) and Midlarsky
models with measures of political regime type and interac (1999), who do find an impact of (certain kinds of) verti
tionswith His, but the resultswere insignificant. cal inequality on (certain forms of) political violence.
imply some sort of clustering in the distribu sion and identity. Esteban & Ray (1994)
tion. The degree of polarization increases argue that traditionalmeasures of inequality
when there are few (equally) large groups are concerned with only interpersonal
alien
group, and differences in characteristics group identity. I concur with this critique,
between them.Unlike horizontal inequalities assuming that a
country's level of vertical
measured
along
one
single dimension, such
according toEsteban & Ray (1994), is rather
as ethnicity (Montalvo & Reynal-Querol, economic
polarization. They argue that a
2005b) or economic welfare (Esteban & Ray, society that is split into two well-defined
1994). However, questions
are often asked groups with substantial intragroup homo
about the dynamics of polarization between geneity and intergroup heterogeneity in
exogenously given groups, such as urban/rural, incomes is particularly likely to experience
regional, religious or, as in the current study, social unrest. Such a society would be
ethnic groups. Unfortunately, currently marked by both strong group identification
available
polarization
measures are not fit to and sharp divisions (alienation) between the
answer such questions (Fedorov, 2002). To groups. The DHS data allows me to measure
overcome this limitation, this intro economic in terms of household
study polarization
duces a measure of horizontal and assets versus social in terms of
inequality polarization
alternative measures of polarization educational Hence, I suggest
proposes opportunities.
that cover both socio-economic and ethnic the following hypotheses:
dimensions.5 Below, I elaborate on the
H la: The higher the level of economic
mechanisms between various types of
polar in a the
polarization country, higher
ization,HI and the outbreak of civil conflict.
the risk of civil conflict.
H lb: The higher the level of social polariza
Polarization, Horizontal Inequalities tion in a country, the higher the riskof
and Civil Conflict civil conflict.
religion, rather than class, as the dividing line largerwith few groups.My second hypothesis
of warring groups, stressing that one should derives from this logic:
strive to single out the specific dimension of
H2: The higher the level of ethnic polariza
group divisions that ismost relevant to each
tion in a country, the higher the risk of
specific conflict. In line with this, many
civil conflict.
researchers would argue that, although
perhaps
important for intergroup hostilities, economic Several conflicts today are fought between
welfare (or social, for that matter) is not nec ethnic but there are also ex
groups, many
the most important characteristic for of diverse countries where
essarily amples ethnically
group identityand that one should also focus ethnic groups live peacefully side by side.
on other factors (see Stewart, 2000). Tanzania and Ghana, for have
e.g. example,
Ethnic identity isparticularly conflict-prone, avoided the severe internal conflicts ofmany
since it is based on fundamental factors like of their
neighbours.
Part of the reason may
language, race and religion. Although there be that such countries have avoided socio
aremixed results
regarding differentforms of economic
inequalities between the groups
ethnicity and conflict (see e.g. Ellingsen, (Minority Rights Group International,
2000; Fearon & Laitin, 2003; Montalvo & 1997: 517, 419). Given that a shared ethnic
Reynal-Querol, 2005a,b), civil wars often group identity is not sufficient to explain
seem to have an ethnic dimension, with well mobilization, one should
group investigate
defined identity groups fighting each other whether, and under what circumstances,
groups. Ethnic
polarization,
on the other powerful mobilizing agent that can lead to a
hand, increaseswhen there are few (equally) range of political disturbances.
with characteris There are many cases in which various
large groups homogenous
tics within each group, and differences in a kinds of His seem to have conflict.
provoked
cluster of characteristics among groups. The civil conflict between the Senegalese
Indices of polarization are highest when government and the Mouvement des Forces
societies with few rival because few in the are the Diola,
groups, group region among
groups imply clearly defined ethnic divisions, whom separatist sentiments have existed
and because the recruitment becomes since colonial times. This stems from the
pool
against Tamils (Gurr, 2000: 109). have higher levelsofGDP per capita than the
Why and how do horizontal inequalities national average (Tadjoeddin, 2003). Either
breed violent conflict?A shared identitymay scenario leads to themain proposition of this
overcome collective action
problems, whereby article: that societies of high horizontal
inequalities have higher risksof civil conflict.
are unable to because of
people cooperate
mutual (Olson, 1965). Ethno
suspicions As with polarization, I propose hypotheses
political organizations often use symbols of which distinguish between two different
shared identity and
grievances
as a
strategy
to kinds ofHis:7
build a sense of common interest. If the cul
H3a: The higher the level of horizontal eco
tural group identity is weak, people can
nomic in a the
inequality country,
seldom be mobilized by leaders in response
to a threat (Gurr, 2000: 75). higher the risk of civil conflict.
H3b: The higher the level of horizontal
However, a shared is not a
group identity
social in a the
to inequality country,
sufficient factor violent conflict.
produce
As & Gates higher the risk of civil conflict.
argued by Murshed (2005),
some shared are
group-grievances required
for identity-based conflict. Horizontal in
a collective of
equalities capture aspect 7
Preferably, one should have data on political dimensions
relativedeprivation which can facilitatemobi of His as well, but theDHS do not include any questions
lization for conflict.When such collective which are suited for such a measure.
conflict are omitted (coded as missing inequality' and Vertical social inequality'.
observations). There are no hypotheses for the relationship
between vertical inequalities and conflict in
Independent Variables: Ethnicity and this article (because I assume that there is no
ER=
where tt? denotes the population share of
K??nrnj\y,-y\ (2)
ethnic group / in a country. The degree of
ethnic increases when there are where tt? and 77. denote the size of groups /
polarization
few (equally) large groups with homogenous andy,
and j/; and y: denote the share of house
characteristicswithin each group, and differ hold assets (or education years) possessed by
ences in a cluster of characteristics the groups. The formula summarizes all per
among
groups. mutations of n groups defined by the asset or
Since theDHS surveys lack information education variables.11 The constant A" is set
on income or
consumption expenditures,9
to 1/2 |jl to allow for comparability across
I use instead the information collected on populations, and a is set to 1. Note that this
other household characteristics.The analysis measure equals theGini coefficient ifa is set
uses two different indicators of socio-economic to 0. As an alternative to this, I also calculate
welfare to calculate measures of socio-economic measures for two-point distrib
polarization
vertical inequalities and polarization: a house utions (bipolarization), dividing the popula
hold asset index and educational attainment. tions in each into two groups, above
country
The household asset index is calculated on and below the country mean, jul.This is
given
the basis of the following variables from the by the formula below (see Esteban & Ray,
DHS surveys: vll9?vl25 (dummies for 1994: 842):
whether or not each household has electric
a a television, a a +
ity, radio, refrigerator, bicycle, />=[7rl+a(l_7r)
a and/or a car). second indi (3)
motorcycle My (l-Try+<*7T](y-x)
cator is based on the variable v 133 (total
years of education For both where tt denotes size and x and
completed).10 group y
household assets and education years, I generate denote the share of household assets (or edu
Gini coefficients to measure vertical economic cation the two
years) possessed by groups
Hence, I further apply Esteban, Gradin & owned bymembers of the richer group. The
measure potentially ranges from 0 (perfect
Ray's (2007: 6) proposed extension of the
bipolarization measure, which ismore sensi equality) to 1 (horizontal inequalitywith the
tive to the choice of cut-off point: richer group owning
all the assets). Since rel
best
13
Henceforth, I sometimes use the shorthand ethnic/socio
economic polarization for these two variables.
12 14 I include all the ethnic
Alternatively, one could compare the best and the
worst groups in theDHS which con
sisted of at least 1% of the national population.
group regardingwelfare, or the best vs. the rest.
than my
simple
HI ratio measures. Note that
polarization respectively)aremarkedly stronger.
these are not interaction terms for ethnic and This is not since both mea
very surprising,
economic polarization, but rather a hybrid sures are two-dimensional and hence con
measure of which combines related.16
polarization ceptually very closely
ethnic and socio-economic considerations.15
For this purpose, I use the ER polarization
Control Variables and Statistical
formula (2) again, but this time I let thegroups
Model
be exogenously given in terms of ethnicity as
defined byDHS. Ignoring intragrouphomo Since my sample consists of only developing
geneity, I assume that each individual scores countries, I restrict the number of control vari
the mean value of its ethnic group in terms ables to economic size
development, population
of household assets and education years. and conflicthistory}7The direct linkfrom eco
Unlike the HI measure, this alternative, com nomic to internal is one of
development peace
posite polarization
measure is sensitive to themost robust findings in the literature.To
group size and covers all groups in
society. proxy level of development, I use log
For all the variables that stem from the transformed GDP per capita measured
in con
DHS surveys, I interpolate values for inter stant 1996 US$. I also include a term for
vening years and
extrapolate
the value from the
laggedpopulation size.Both variables are lagged
survey nearest in time for
previous and subse by one year and stem fromGleditsch (2002).
quent yearswithin theperiod 1986-2004. For I employ a logistic regressionmodel with
countries with one I use that value standard errors, an exten
only survey, country-clustered
for all years within the period, in order to sion of robust variance estimation. As sug
15 See
Duelos, Esteban & Ray's (2004: 1760) discussion
on hybrid polarization measures which combine pure 16
Stewart, Brown & Mancini (2005) provide a compre
income polarization and pure social (ethnic) polarization, hensive comparison of polarization and HI measures.
and their suggested 'social polarization with income-medi 17 I reran the
analyses controlling for growth and regime
ated identification. My measure (socio-economic polariza type (single and squared), but, like other studies of similar
tionwith ethnically mediated group identification) can be samples (e.g. Miguel, Satyanath & Sergenti, 2004), these
seen as an alternative to this. standard controls did not yield significant effects.
Table I. Correlation Between Horizontal Inequalities and Polarization Measures in 39 Developing Countries,
1986-2004 (N = 717)
HSI SP SB ESP
***/>< 0.01.
All the models contain the controls for eco as or bipolarization). All
regularpolarization
nomic development, population
size and the four terms are significant at the 10%
conflict history. Models 1-7 in Table II level. These results substantiate Esteban &
report the effects of vertical inequalities, and Rays (1994) argument that societies with
economic, social and ethnic few well-defined or clusters, with
polarization, groups
The terms are introduced but consid
respectively. sepa strong intragroup homogeneity
rately, owing to high multicollinearity and erable divisions between the groups, are
par
verticaleconomic inequality (measured as house welfare may contribute to both the intra
hold asset ownership) is related to the outbreak group cohesion and the intergroup alienation
of internalarmed conflict inModel 1.This is in required in order to provoke conflict. The
linewith Collier & Hoeffler (2004), Fearon & log likelihood scores reveal that themodels
Laitin (2003) and others,who fail to establish a with social polarization (Models 4 and 6)
significant link between vertical income perform slightlybetter than themodels with
inequality and conflict. The alternative indica economic polarization (Models 3 and 5).
torof verticalsocial inequalitydoes not turnout As discussed in the theory section,
tobe significanteither (Model 2). In accordance although
the socio-economic
polarization
with the theoreticaldiscussion above, thismay measures seem to be
important, there may be
indicate that simple inequality between indi other and more salient characteristics than
viduals is not to spur violent conflict, socio-economic welfare that create group
enough
because, unlike polarization,
there is no clear identities and mark group boundaries, such
idea of group identification (apart from class) as ethnicity.To test this possibility, Hypo
associated with vertical inequality.
thesis2 posited a positive relationship between
What
happens, then, when I introduce the level of ethnic polarization and conflict,
terms of economic and social for two reasons. First, few sized
polarization? equally
Hypotheses la and lb, assuming higher risks ethnic groups imply clearly defined ethnic
of conflict with higher levels of economic divisions and group identities, and second,
and social receive support in it should be easier for ethnical entrepre
polarization,
Models 3?6, which demonstrate neurs and leaders to mobilize a rebellion
significant
effects of the terms for both economic when the recruitment
pool
increases
positive
inequality (0.74)
Vertical social 2.17
inequality (1.38)
Economic 9.46*
polarization (1.88)
Social polarization 4.80*
(1.87)
Economic 9.82*
bipolarization (1.93)
Social bipolarization 4.51*
(1.82)
Ethnic polarization 0.30
(0.27)
GDP per capita (In) -0.55 -0.33 -0.30 -0.36 -0.55 -0.35 -0.83**
(-1.37) (-0.82) (-0.72) (-0.83) (-1.52) (-0.81) (-2.20)
Population (In) -0.23 -0.15 -0.19 -0.19 -0.26 -0.19 -0.24
(-0.82) (-0.56) (-0.80) (-0.63) (-0.97) (-0.64) (-0.85)
Constant 1.06 -1.66 -1.44 -0.69 1.44 -0.72 4.00
2002). However, Model 7 suggests that the move on to test the impact of horizontal
itymay be very relevant for group identity the variables of vertical inequality and pure
formation, it is not obvious that ethnic group socio-economic and ethnic polarization,
differences, as such, are a source of alienation Models
respectively. 8-11 inTable III intro
or Furthermore, the duce these terms.
intergroup hostility.
finding reported here is not incompatible Hypotheses 3a and 3b, assuming a higher
with Esteban & Rays (2008) argument that riskof conflictwith higher levels of horizon
societies with sharp ethnic polarization do tal inequalities between ethnic groups, receive
not have a higher riskof civil conflict (due to mixed support inTable III. As
expected, the
term for economic a
horizontal inequality has
the cost of rebellion) but, rather, are asso
high
ciated with higher conflict intensity.18I now positive sign, as shown inModel 8, but the
relationship is far from statistically signifi
18 cant. The variable horizontal social
An empirical testof the latterwould require a largernumber measuring
of conflict years than is currently included inmy dataset. on the other hand, is
inequality, strong,
positive and significantly related to conflict access to the education system, civil war may
at the 10% level.The lattermight not seem result.
very impressive,but given the relatively small The difference in the results for economic
sample presented here, with hardly signifi and social His inModels 8 and 9 is some
cant (or non-significant) results for the usual what surprising, but may indicate that
suspects' GDP per capita ana population size, inequality
in outcomes in terms of educa
10% significance should be considered quite tional and
occupational opportunities ismore
equality from the 5 th to the 95th percentile ethnic groups tend to control state resources
(1.79)
Ethnic/economic polarization 20.18
(0.93)
Ethnic/social polarization 6.59
(0.62)
GDP per capita (In) -0.89* -0.82* -0.86* -0.78*
(-1.79) (-1.75) (-1.96) (-1.96)
Population (In) -0.41 -0.39 -0.36 -0.26
(-0.85) (-1.10) (-0.97) (-0.89)
Constant 5.66 4.51 4.98 3.79
(0.86) (0.87) (0.94) (0.85)
Log likelihood -81.03 -79.64 -80.99 -81.35
Pseudo/?2 0.11 0.13 0.11 0.11
N 519 519 519 519
Number of countries 36 36 36 36
Number of civilwars 22 22 22 22
on countries in parentheses. */><0.10;
Logit estimates with robust z-statistics clustered **/>^0.05;
***p<0.01. Estimates for peace-years and three natural cubic splines not reported.
of social opportunities associated with ethni The terms are tested inModels 10 and 11.
cally based unequal access to education The effects of ethnic/economicpolarization
indeed contributes to the risk of civilwar. and ethnic/social
polarization
are both positive
However, it is also
quite possible that the
regarding conflict onset, but not significantly
summary measure for economic HI so. Horizontal social seems to be a
relating inequality
to the of various household assets better of conflict.
ownership predictor
does not
capture the
general
structural eco As for the control variables, I find some
nomic differences between ethnic groups in evidence for a negative relationship between
a
country. On the one hand, goods
other GDP per capita and conflict, although not
-
than household assets land tenure, for so. This is not so since
robustly surprising,
-
example may be more important indicators my sample consists of only developing coun
of ethnically based economic distribution in tries.The population term is not significant
many countries. Moreover, the in any model.
developing
distribution of visible status symbols (cars,
motorcycles, etc.) may be a result of differ
Robustness Checks and Alternative
ences in cultural not
preferences, necessarily
actual differences in eco Specifications
reflecting general
nomic resources, such as income. Since the sample of countries included in this
My measures of horizontal inequalities study is limited to those countries that have
focus exclusively
on the ratio between the been surveyed by theDHS project and where
two
largest groups and do not account for ethnicitywas included as one of the ques
group size and tions, we cannot refer to this as a random
intragroup inequalities.
According to Stewart, Brown & Mancini sample. The selection mechanism is likely to
(2005: 22), group weighting by population cause a bias, questioning the
generalizability
would seem desirable in most cases. The of the resultspresented above. The sampling
purely socio-economic
polarization
measure mechanisms work on both the dependent
discussed earlier is sensitive to both group and
independent variables. War-ridden
sizes and However, countries are less to be chosen,
intragroup inequality. likely owing
a restrictionwith thismeasure is that identifi to
security and countries where
problems,
cation and alienation have to be driven by ethnic differences are salient should more
the same characteristic. In fact, Duelos, likely be censored with regard to the eth
Esteban & Ray (2004: 1759) themselves nicity question. The bias is therefore likely to
stress that it more com cases
they would find exclude that would support the theory
pelling to adopt amultidimensional approach presented
in this article, and, hence, the
to alienation to estimates and herein are
polarization, permitting findings reported
depend
on characteristics other than the one
likely to be robust, or perhaps even biased
that defines group identity'. This downwards.
corresponds
well to the logic of the horizontal inequality Second, since the number of onsets relative
conflict argument. In linewith this, I finally to the number of control cases is quite skewed,
test the alternative
operationalization of it isnot unlikely that thefindings reportedhere
horizontal inequality, which is, in fact, could be driven by some highly influential
a combination of the two pure measures of cases. However, the results inTables II and III
(ethnic and socio-economic). are robust to the removal of outliers. In fact,
polarization
Group identity is given by shared ethnicity, the results become much stronger when out
and group alienation is defined as socio liers are censored,
notably the coefficients for
economic
inequalities between the groups. economic and social His and the related
composite
term for ethnic/economic polariza Concluding Remarks
tion (all tests are available in the online
This study has demonstrated that one of the
Appendix D).
suggested indicators of horizontal inequalities
Finally, the analysis presented in this seems tomatter for the riskof conflict indevel
article has potential endogeneity problems,
since some of the inequality/polarization oping countries. Horizontal social inequality
has a robust effect on con
are based on that were carried positive significant
data
flict- both in the panel and the cross-section
surveys
out late in the sample period (e.g. HI could
mea
setup. The socio-economic polarization
result from former conflict instead of the
sures also turn out to be in the
significant panel
other way round). The rationale behind
but the effect of purely economic
a was to retain as analyses,
choosing panel design to insignificance when
much information as possible from theDHS polarization drops
tested The terms for vertical
In order to check robustness of cross-sectionally.
surveys. the
ethnic and the
results for the socio-economic inequalities, polarization
the
significant combined ethnic/socio-economic polarization
polarization terms and horizontal social in measures are not The main con
a significant.
equality, I also applied cross-sectional design clusion one can draw from the is that
one observation study
with only per country (see
other recent studies of the inequality-conflict
Appendix C for descriptive statistics). The
nexus (Collier & Hoeffler, 2004; Fearon &
variable counts the years of con
dependent
Laitin, 2003; Hegre, Gissinger & Gleditsch,
flict following the year of the first DHS
in each whereas the inde
2003) may be premature in concluding that
survey country,
to conflict.Even though
variables stem from the year of the inequality isunrelated
pendent data suffer from some and
my shortcomings
firstDHS survey in the country.The controls must be regarded as preliminary, this studyhas
are the same as in the apart from
design,
panel
demonstrated that it is too early to reject the
the peaceyears and which are
splines, replaced nexus!
a for whether or not the country inequality?conflict
by dummy
There is still room for improvement with
conflict in the five years
experienced preced to our of the relation
regard understanding
ing the firstDHS survey. Since conflicts do
not tend to end at a constant rate ship^) between polarization, inequality and
(Collier,
conflict. national surveys as a starting
2004), we
Hoeffler & should Using
S?derbom,
for generating various measures seems
some of over-dispersion, which point
expect degree a way forward. First, it is a way of
at a binominal promising
points negative regression are difficult to find
generating new data that
model. The spells observed vary in length elsewhere. for polarization and hori
and each was entered into Figures
country's exposure
zontal can be sensi
as an rather inequalities politically
the equation ancillary parameter
tive, and national are to
governments likely
than as a control variable (Long
& Freese,
2003: 264f). Table IV shows the results from report biased data. Biased information is less
data are from national
tests. likely when generated
the cross-sectional
surveys like theDHS, as the original inten
The term for economic
positive purely was not to assess socio
tion behind these
to insignificance, but the
polarization drops economic between ethnic
inequalities groups.
results for the variables social polarization and
Second, the aggregation of survey data en
horizontal social inequality stillhold. Not sur
sures rather than evaluative data.
a small N, the controls descriptive
prisingly,with such That is, researchers do not need to on
are not rely
for GDP per and population sig
capita as the sole source
their personal judgement
nificant, but the effectof the conflicthistory is scores.
for HI Third, survey
to be in most models. determining
shown positive
bipolarization (0.87)
Social 6.84**
bipolarization (2.55)
Horizontal social 2.39**
inequality (2.01)
GDP per 0.11 0.07 -0.33 0.07 -0.39
capita (In) (0.17)
(0.15) (-0.62)(0.17) (-0.96)
Population (In) 0.12
0.02 0.09 0.02 -0.07
(0.40) (0.06) (0.29) (0.08) (-0.24)
Conflict history 1.86***
0.89 0.82 1.84*** 0.96*
(1.59) (2.95) (1.32) (2.97) (1.78)
Constant -4.98 -3.99 -0.73 -4.10 0.74
(-0.75) (-0.91) (-0.13) (-0.94) (0.18)
Alpha 1.48
1.70 1.46
1.79 1.50
(1.34) (1.02) (1.49) (0.99) (0.99)
Log likelihood -70.27 -68.00 -70.83 -67.88 -69.25
Pseudo/?2 0.06 0.09 0.05 0.09 0.07
N 36
36 36
36 36
Negative binominal regression estimates with z-statistics in parentheses. */><0.10; **/>^0.05; ***/>^0.01.
data
provide
a source
for
calculating
not of socio-economic distribution and ethnicity
only inequalities between groups, but also when
studying
conflict. We need a measure
inequalities may reduce the potential for Duelos, Esteban & Ray (2004: 1760) and
intergroup conflict for any given degree of Zhang & Kanbur (2001) suggest some
horizontal inequalities, because it may be hybrid measures for this purpose. One
more difficult to obtain group cohesion variant of such a measure was
proposed
in
where there ishigh intragroup inequality.We this article, but did not yield significant
need to explore how to deal with different results in the analyses.
Future research
levels of
inequality
in the same
analysis. should further explore and apply such mea
Stewart, Brown & Mancini (2005: 9) argue sures of multidimensional polarization with
that although it would be interesting to survey data and test their effects on conflict.
include a measure of
within-group inequal
ity in the measure of each ethnic groups
socio-economic 'this is not References
performance,
desirable because the meaning of the Alberto & Roberto 1996. 'Income
Alesina, Perotti,
measure would no be clear'. Political Invest
longer Distribution, Instability, and
Most this analysis has shown ment', Economic Review 40(6):
important, European
that one should focus on combined measures 1203-1228.
Annan, Kofi, 1999. 'Peace and Development: Esteban, Joan & Debraj Ray, 2008. 'Polarization,
One Struggle,
Two Fronts', address to theWorld Fractionalization and Conflict', Journal of
Bank Staff, Washington, DC, 19 October Peace Research 45(2): 163-182.
Collier, Paul & Anke Hoeffler, 2004. 'Greed and ties at Risk in theNew Century. Washington,
Grievance in Civil War', Economic DC: United States Institute of Peace Press.
Oxford
Papers 56(4): 563-595. H?vard; & Nils Petter
Hegre, Ranveig Gissinger
Collier, Paul; Anke Hoeffler & Mans S?derbom, Gleditsch, 2003. 'Globalization and Internal
2004. 'On the Duration of Civil War', Journal Conflict', in Gerald Schneider, Katherine
Ray, 2007. 'An Extension of a Measure of Long, J. Scott & Jeremy Freese, 2003. Regression
Polarization, with an to the Income Models Dependent Variables
Application for Categorical
Distribution of Five OECD Countries',
Journal Stata: Revised Edition. College Station,
Using
TX: Stata Press.
ofEconomic 5(1): 1?19.
Inequality
Manus 1999. The Evolution
Esteban, Joan-Maria & Debraj Ray, 1994. 'On the Midlarsky, I., of
Measurement of Polarization, Econometrica Wan State Survival, and Democracy
Inequality:
62(4): 819-851. in Comparative Perspective. Stanford, CA:
Nagel, Jack, 1974. 'Inequality and Discontent: Study of Civil War, PRIO (http://www.prio.
A Nonlinear Hypothesis',WorldPolitics26(4): no/cscw/armedconflict).
453-472. Tadjoeddin, Mohammad Zulfan, 2003. 'Aspiration
Olson, Mancur, 1965. The Logic ofCollective Action. to and Centre
Inequality: Regional Disparity
Cambridge, MA: Harvard University Press. Conflicts in Indonesia', paper pre
Regional
Parvin, Manoucher, 1973. 'Economic Deter sented at the UNI/WIDER Project Conference
minants of Political Unrest: An Econometric on in Asia, Tokyo, 28-29
Spatial Inequality
Approach', Journal of Conflict Resolution March (http://www.wider.unu.edu/conference/
17(2): 271-296. conference-2003-1 /conference-2003-1 -papers/
Russett, Bruce M., 1964. 'Inequality and mohammad%20zulfan%20tadjoeddin.pdf).
The Relation of Land Tenure to Charles, 1978. From Mobilization to
Instability: Tilly,
Polities',World Politics 16(3): 442-454. Revolution. Reading, MA: Addison-Wesley.
Sahn, David E. & David Stifel, 2003. 'Exploring UNU/WIDER & UNDP, 2000. World Income
Alternative Measures of Welfare in the Database. User Guide and Data
Inequality
Absence of Expenditure Data', Review Sources. Helsinki: UNU/WIDER (http://www.
of
IncomeandWealth 49(4): 463-489. unu.edu/wiid/wiid.htm).
Sambanis, Nicholas, 2002. A Review of Recent Weede, Erich, 1981. 'Income Inequality, Average
Advances and Future Directions in the Income, and Domestic Violence', Journal of
Literature on Civil War',
Defence and Peace ConflictResolution24(4): 639-654.
Economics 13(3): 215-243. Xiaobo & Ravi Kanbur, 2001. 'What
Zhang,
Schock, Kurt, 1996. 'A Conjunctural Model of Difference Do Polarization Measures Make?
Political The
Conflict: of Political An Application to China', Journal
Impact ojDevelop
the on Berween ment Studies 37(3): 85-98.
Opportunities Relationship
Economic Inequality and Violent Political
Conflict',Journal ofConflictResolution40 ( 1) :
98-133. GUDRUN 0STBY, b. 1977, Cand.Polit.
Sen, Amartya, 1992. Inequality Reexamined. (Norwegian University of Science and
New York: Clarendon. 2003); Researcher, International
Technology,
Sigelman, Lee & Miles Simpson, 1977. A Cross PeaceResearch Institute,
Oslo (PRIO) (2005- );
National Test of the Linkage Between Economic PhD candidate in Political Science, University
Inequality and Political Violence', Journal of of Oslo (2005- ); current main interest: hori